Indoor Model

Czech Technical University in Prague
Faculty of Electrical Engineering Department of Electromagnetic Field
Dissertation Thesis
Novel Approach to Indoor Electromagnetic Wave Propagation Modelling
Martin Klepal
July 2003 Prague, Czech Republic
Czech Technical University in Prague

Faculty of Electrical Engineering Department of Electromagnetic Field
Dissertation Thesis
Novel Approach to Indoor Electromagnetic Wave Propagation Modelling

Martin Klepal
Doctoral Study Program: Electrical and Informatics Engineering Study Field: Radioelectronics Supervisor: Doc. Ing. Pavel Pecha Ph.D.
Dissertation submitted in partial fulfilment of the requirements for the degree of Doctor of Philosophy in Electrical and Informatics Engineering
July 2003 Prague, Czech Republic
Preface and Genesis of the Motif Model

In the 1980s, Prof. Novotn and Dr. Pecha put heads together at the Faculty of Electrical engineering at the Czech Technical University in Prague, and thought up first fundaments of easy and rapid method of ray propagation. After graduating, I started enthusiastically working on indoor propagation modelling based on semi-empirical models and on full ray tracing prediction within an anechoic shielding chamber [81-93]. Semi-empirical models are easily understandable and applicable, which is their big advantage is. However, they did not satisfy my expectation that they would provide a correct and clear prediction of signal coverage. Full ray tracing is very potent, but it is too rigorous and too complicated for a complex indoor environment. Neither semi-empirical approach nor full ray-tracing approach seemed to me to be well suited for indoor propagation. For this reason I started to work on something new, which seemed to be more promising for indoor propagation prediction. First, I began to implement a ray tracing model. This was a really complicated task, and it led me to reflect on the special features of an indoor environment. New ideas began to take root. This resulted in the suggestion of ray propagation within a grid and precalculated local environment behaviour. During a consultation with my supervisor Dr. Pecha, he retrieved old ideas of a method of easy and rapid ray propagation that had never been checked out. He also recommended implementing them into new model. This was a crucial suggestion, for which I am grateful. The idea was, firstly, to make a probabilistic description of a local environment including a probabilistic radiation pattern of the obstacles and, secondly, to represent the signal level by simple ray density. These two thoughts, together with grid propagation, remain the principal elements of the current model. These new, stimulating recommendations led a well-working implementation of a new model into a software package and to the first publications of description if the new model, at that time known as the XY-model [94-96]. The first success was followed by applications of the model not only for signal coverage prediction, but also for prediction of some wideband channel parameters. This was accompanied by several publications [101-103, 108, and 109] and the name of the model was changed to the Motif Model. A further great stimulus while working on the new model was provided during my fellowship in Ireland by Dr. Pesch, who was also captivated by the new model and supported my work enthusiastically. At that time I began to work on optimising the parameters of the model, and he suggested the use of evolution strategies. Implementation of evolution strategies led to a powerful optimization routine [112].
Various successful Motif Model applications [113, 119-121] during evaluation of the model affirm the reasonability of the concept of the new model algorithm, and underline its advantages over conventional models, which are usually based on ray-optical principles. I would like to apologise in advance to readers for the shortcomings in my English. However, most of the researchers interested in the Motif Model come from outside my home country. I have therefore done my best to make the text easily comprehensible. I also apologise in advance for any errors which may occurred in this text. I will be grateful to receive any comments or suggestions that will help in further development of the Motif Model.
Acknowledgements
First of all I would like to express my sincere appreciation to my supervisor, Dr. Pavel Pecha, Vice Head of Department of Electromagnetic Field, Czech Technical University, for his entire encouragement, without which the Motif Model would have never come into the world. I would like to acknowledge to Prof. Milo Maznek, Head of Department of Electromagnetic Field, Czech Technical University, for his support and patience with me. It is my pleasure to thank also to Prof. Karel Novotn, Assistant Professor at Department of Electromagnetic Field, Czech Technical University, for inspiring and invaluable consultation. Special thanks are to Dr. Dirk Pesch, Department of Electronic Engineering, Cork Institute of technology, Ireland, for his sharing of my enthusiasm. I am particularly indebted also to Arthur Robin Healey, Department of International Relations, Rectorate of Czech Technical University, who tried hard to make my pidgin English more readable. Finally, my acknowledgements are due to my dearest wife and son for their patience and constant furtherance.
Martin Klepal Prague, July 2003
ii
Contents
1. Introduction ........................................................................................................ 1 2. Aims of the Dissertation ................................................................................... 3 3. Current Approaches to Indoor Propagation Modelling .............................. 4
3.1 3.2 3.3 General Outline of Electromagnetic Wave Propagation inside Building ..... 4 Models for Indoor Propagation Prediction ................................................... 8 Summary and Motivation for Developing a New Model ............................. 17
4. Motif Model ....................................................................................................... 19 - Novel Approach to Electromagnetic Wave Propagation Modelling

4.1 4.2 Delineation of New Model Ideas .................................................................. 19 Environment Description .............................................................................. 24
4.2.1 A Novel Method for Describing an Environment with the Use of Motifs ................................................................................................. 24 4.2.2 Definition of Motif Parameters............................................................... 30 4.2.3 Optimization of Motif Parameters ......................................................... 38 4.3 Rays Launching and Propagation ................................................................. 46
4.3.1 Description of Ray Launching and Propagation Technique .................. 46 4.3.2 Necessary Number of Launched Rays and Ray Coverage Acceleration ........................................................................................... 49 4.4 Prediction ...................................................................................................... 53
4.4.1 Mean Signal Level Prediction by Ray Density ..................................... 53 4.4.2 Prediction of other Channel Parameters ................................................. 62 4.5 Model Setting & Evaluation ......................................................................... 64
4.5.1 Reduction of Fast Fading ....................................................................... 64 4.5.2 Measuring Signal Coverage and Setting Motif Parameters.................... 71 4.5.3 Impulse Response Measurement and Prediction ................................... 96 4.6 4.7 4.8 Further Applications of the Motif Model ..................................................... 104 Probabilistic 2D Model Based on Ray Density Distribution in Periodic Building Structures Supplement of 2D Motif Model for 3D Prediction ... 111 Ray Expansion .............................................................................................. 120
iii
5. Comparison of Models ............................................................................... 127 6. Summary and Future Research Directions ................................................. 137 7. Conclusion .................................................................................................. 140 8. References .................................................................................................. 141 Appendix A Description of the Basic Motifs ..................................................... 151
iv
Abbreviations and Acronyms

AoA BUT CDF CIT - Angle of Arrival - Brno University of Technology - Cumulative Distribution Function - Cork Institute of Technology
COST231 COperation europenne dans le domaine de la recherche Scientifique et Technique - Action 231 CTU DPM ECT EIRP ES FD FSL GA GSM GTD IEE IEEE IR ISM LOS MoM - Czech Technical University - Dominant Path Model - Evolutionary Computation Tester - Effective Isotropic Radiated Power - Evolution Strategies - Finite Difference Method - Free Space Loss - Genetic Algorithms - Global System for Mobile - Geometry Theory of Diffraction - Institute of Electrical Engineers - Institute of Electrical and Electronics Engineers - Impulse Response - Industrial, Science and Medical Band - Line of Sight - Method of Moment
FDTD - Finite Different Time Domain
ITU-R - International Telecommunication Union - Recommendation MIMO - Multiple Input Multiple Output MWM - Multi Wall Model NLOS - Non Line of Sight OSM PDF PDP RF - One Slope Model - Probability Density Function - Power Delay Profile - Radio Frequency
RMS SNR TLM UHF URSI UTD Wi-Fi
- Root Mean Square - Signal to Noise Ration - Transmission Line Matrix - Ultrahigh Frequency - Union Radio-Scienfifique Internationale - Uniform Theory of Diffraction - Commercial Trademark of IEEEs 802.11b Wireless Local Network Standard
TETRA - Trans European Trunked Radio System
WLAN - Wireless Local Area Network
vi
List of Figures
Fig. 3.1. Fig. 3.2. Fig. 3.3. Waveguiding mechanism of a corridor.......................................................... 7 Principal indoor models ................................................................................. 9 Examples of first Fresnel zones at a frequency of 2 GHz for different distances of ray vertices those are common for a) outdoor and b) indoor environments .................................................................................... 12 Natural influence of obstacle dimension on ray propagation ...................... 13 First Fresnel zone interaction with obstacle................................................. 14 Principles of ray-optical models .................................................................. 14 Example of room-oriented dominant paths.................................................. 16 Basic approaches.......................................................................................... 18 Example of site-specific description of a building, based on grid elements ....................................................................................................... 22 Overall probability radiation pattern of a cluster of obstacles (Motif) ........ 22 Floor plan filling up with an increasing number of launched rays .............. 23 Example of the ray propagation process...................................................... 23 A typical floor plan arrangement a) divided into grid elements b).............. 25 A typical situation - wall surrounded by many vague obstacles.................. 25 A wall fraction with randomly distributed obstacles and the overall probability radiation pattern of a reradiated incident wave ......................... 26 Example of possible ray power reradiation by a plane wall ........................ 26 Basic motifs ................................................................................................. 27
Fig. 3.4. Fig. 3.5. Fig. 3.6. Fig. 3.7. Fig. 3.8. Fig. 4.1. Fig. 4.2. Fig. 4.3. Fig. 4.4. Fig. 4.5. Fig. 4.6. Fig. 4.7. Fig. 4.8. Fig. 4.9.
Fig. 4.10. Examples of the material-specific variation of motifs, plus one universal motif ............................................................................................. 27 Fig. 4.11. Example of an environment spatial description........................................... 28 Fig. 4.12. Structure of an implemented motif database ............................................... 28 Fig. 4.13. Example of describing a building by a nonuniform grid ............................. 29 Fig. 4.14. Distribution of emitted power into quadrants of T-Wall motif.................... 32 Fig. 4.15. Dependency of distribution of power into quadrants on probability of reflection of the motif .................................................................................. 32 Fig. 4.16. Example of T-Wall motif radiation pattern (PrDS = 0.73, PrRT = 0.5).......... 32
vii
Fig. 4.17. Example of motif radiation pattern formation for the Plain-Wall motif ..... 33 Fig. 4.18. Comparison of 1/ PrDS, DPW and directivity Dcos of nth powered cosine lobe as a function of probability of diffuse scattering ................................. 34 Fig. 4.19. Variation of cosine power with the probability of diffusion scattering ....... 34 Fig. 4.20. Example of Plane-Wall motif radiation pattern formation .......................... 35 Fig. 4.21. Example of the dependency of the reflection coefficient on the impact angle............................................................................................................. 36 Fig. 4.22. Dependency of probability of reflection PrRT on material constants ........... 36 Fig. 4.23. Dependency of PrRT on material constants for the linear part of the previous chart (Linear part ~ part independent of the / variable) ......... 36 Fig. 4.24. Dependency of probability of absorption PrA on material constants and frequency (material constants are assumed to be frequently invariant)....... 37 Fig. 4.25. Charts corresponding to (4.18) and (4.19), respectively.............................. 38 Fig. 4.26. Our evolutionary computation tester ........................................................... 39 Fig. 4.27. Example of a fitness landscape with layout of individuals ......................... 42 Fig. 4.28. Motif Model software tool together with the implemented parameters in the optimisation window.......................................................................... 43 Fig. 4.29. Noise level of fitness function ..................................................................... 44 Fig. 4.30. Example of approximation process for one parameter ................................ 44 Fig. 4.31. Emitter launching rays a-d ........................................................................... 46 Fig. 4.32. Isotropic radiation pattern and relevant distribution function...................... 47 Fig. 4.33. Refining the transmitter radiation pattern .................................................... 47 Fig. 4.34. Power radiation pattern and relevant distribution function.......................... 48 Fig. 4.35. Formed radiation pattern ............................................................................. 48 Fig. 4.36. The same radiation pattern turned by 45 degree.......................................... 48 Fig. 4.37. Example of a ray propagation process ......................................................... 49 Fig. 4.38. Example of a launched beam consisting of 100 rays................................... 50 Fib. 4.39. Mean number of beam interactions, I B as a function of the number of rays incorporated within a beam and the average motif probability of absorption PrA ............................................................................................. 51 Fig. 4.40. Average standard deviation and mean value between signal coverage predictions .................................................................................... 52
viii
Fig. 4.41. Filling up the floor plan with an increasing number of launched rays ........ 52 Fig. 4.42. Representation of a squared grid by a circular grid ..................................... 53 Fig. 4.43. Surface of an element in a squared grid and in a circular grid .................... 54 Fig. 4.44. Effective length of element.......................................................................... 55 Fig. 4.45. Conversion constant c2D ............................................................................... 57 Fig. 4.46. 3D elements in a cubic grid and in a spherical grid..................................... 57 Fig. 4.47. Conversion constant c3D ............................................................................... 60 Fig. 4.48. Orthogonal planes passing transmitter and receiver .................................... 60 Fig. 4.49. Geometry of measured signal level simulation............................................ 65 Fig. 4.50. Residual fast fading fluctuation for variable averaging window dimension dw and the number of applied averaging .................................... 66 Fig. 4.51. Residual fast fading fluctuation for variable w and the number of measured samples per wavelength............................................................... 66 Fig. 4.52. Spreading window averaging technique ...................................................... 68 Fig. 4.53. Non-spreading averaging window technique............................................... 68 Fig. 4.54. Examples of measurement data averaging by different modifications of the averaging window technique in the trial floor plan. .......................... 70 Fig. 4.55. Measurement system configuration ............................................................. 72 Fig. 4.56. Transmitter antenna radiation pattern .......................................................... 73 Fig. 4.57. Receiver antenna radiation pattern............................................................... 73 Fig. 4.58. Example of signal coverage measurement................................................... 74 Fig. 4.59. Example of measured data ........................................................................... 74 Fig. 4.60. Example of a floor plan with marked measured tracks................................ 75 Fig. 4.61. Measured samples projected into the floor plan grid................................... 75 Fig. 4.62. Enlarged measured coverage with reduced fast fading fluctuation ............. 75 Fig. 4.63. Photo of CTU building................................................................................. 77 Fig. 4.64. Input floor plan ............................................................................................ 77 Fig. 4.65. Measured signal coverage............................................................................ 78 Fig. 4.66. Predicted signal coverage (3 seconds, 100 000 beams)............................... 78 Fig. 4.67. Difference between measurement and prediction........................................ 78 Fig. 4.68. PDF of difference between measurement and prediction ........................... 78 Fig. 4.69. Photo of BUT building ................................................................................ 79 ix
Fig. 4.70. Input floor plan ............................................................................................ 79 Fig. 4.71. Measured signal coverage ........................................................................... 79 Fig. 4.72. Predicted signal coverage (5 seconds) ......................................................... 80 Fig. 4.73. Difference between measurement and prediction........................................ 80 Fig. 4.74. PDF of difference between measurement and prediction ............................ 80 Fig. 4.75. Comparison of measurement and prediction along the section ................... 80 Fig. 4.76. Photo of Koospol building ........................................................................... 81 Fig. 4.77. Input floor plan ............................................................................................ 81 Fig. 4.78. Measured signal coverage............................................................................ 81 Fig. 4.79. Predicted signal coverage (20 seconds) ....................................................... 81 Fig. 4.80. Difference between measurement and prediction........................................ 81 Fig. 4.81. PDF of difference between measurement and prediction ............................ 82 Fig. 4.82. Comparison of measurement and prediction along the section ................... 82 Fig. 4.83. Photo of the Diplomat hotel building........................................................... 82 Fig. 4.84. Input floor plan ............................................................................................ 83 Fig. 4.85. Measured signal coverage............................................................................ 83 Fig. 4.86. Predicted signal coverage (10 seconds) ....................................................... 83 Fig. 4.87. Difference between measurement and prediction........................................ 83 Fig. 4.88. PDF of difference between measurement and prediction ............................ 83 Fig. 4.89. Comparison of measurement and prediction along the section ................... 84 Fig. 4.90. Ground floor plan of the CIT building complex ......................................... 86 Fig. 4.91. First floor plan ............................................................................................. 87 Fig. 4.92. Measured signal coverage on ground floor.................................................. 87 Fig. 4.93. Predicted signal coverage on ground floor .................................................. 87 Fig. 4.94. Difference between measurement and prediction........................................ 87 Fig. 4.95. PDF of difference between measurement and prediction on ground floor.............................................................................................................. 88 Fig. 4.96. Measured signal level on first floor ............................................................. 88 Fig. 4.97. Predicted signal level on first floor.............................................................. 88 Fig. 4.98. Difference between measurement and prediction on first floor................... 88 Fig. 4.99. PDF of difference between measurement and prediction on first floor....... 89
Fig. 4.100. Signal coverage prediction throughout the CIT building complex............. 90 Fig. 4.101. Trial floor plan composed only of universal motifs..................................... 92 Fig. 4.102. Measured signal coverage............................................................................ 92 Fig. 4.103. Predicted signal coverage ............................................................................ 92 Fig. 4.104. Difference between measurement and prediction ....................................... 92 Fig. 4.105. PDF of difference between measurement and prediction ............................ 93 Fig. 4.106. Difference between prediction with four basic sets of motif parameters and prediction with only one set of universal parameters............................ 93 Fig. 4.107. PDF of previous difference from Fig.4.106................................................. 93 Fig. 4.108. Influence of grid spacing on prediction ....................................................... 94 Fig. 4.109. Launched wave front i.e., Dirac impulse propagation and interaction simulated by the Motif Model ................................................... 96 Fig. 4.110. Agilent E8358A vector analyzer.................................................................. 97 Fig. 4.111. Wide band double ridge horn antenna and its return loss............................ 97 Fig. 4.112. Antenna radiation pattern............................................................................. 97 Fig. 4.113. Example of omni-directional IR composing................................................ 98 Fig. 4.114. Trial floor plan with indicated measurement locations in brackets ............. 98 Fig. 4.115. Comparison of measured and predicted impulse responses in five locations ....................................................................................................... 99 Fig. 4.116. Directional impulse response prediction at point [3] in the corridor ......... 100 Fig. 4.117. Directional impulse response prediction at point [2] within a room ......... 101 Fig. 4.118. Example of IR prediction by a ray-tracing model taken from [35] ........... 102 Fig. 4.119. RMS delay spread measurement and Motif Model prediction .................. 102 Fig. 4.120. Coverage prediction from a leaky coaxial cable antenna .......................... 104 Fig. 4.121. Instantaneous power prediction along sections A and B marked in Fig.5.1 with resolution 0.2 (~3 cm) ......................................................... 105 Fig. 4.122. Instantaneous power prediction in a grid 1x1 .......................................... 105 Fig. 4.123. Site-specific prediction of Rices k factor.................................................. 106 Fig. 4.124. Fast fading PDF in different locations and the appropriate Rice distribution functions ................................................................................. 106 Fig. 4.125. Site-specific prediction of signal mean delay and RMS delay spread prediction and their PDF on the trial floor................................................. 107
xi
Fig. 4.126. Site-specific prediction of MIMO channel capacity ................................. 108 Fig. 4.127. Site-specific prediction of MIMO channel capacity for a different arrangement of antenna elements .............................................................. 109 Fig. 4.128. Floor excitation by a launched Dirac impulse ........................................... 110 Fig. 4.129. Electrical intensity field distribution of a radiated continuous wave at 100 MHz .................................................................................................... 110 Fig. 4.130. Ray streams (a) and an example of possible ray probability distribution as a function of PrRT (b) .......................................................... 112 Fig. 4.131. Ray probability distribution simulation for PrRT = 0.9 and PrA = 0 with its three sections......................................................................................... 113 Fig. 4.132. Ray probability distribution simulation for PrRT = 0.5 and PrA = 0 with its three sections......................................................................................... 113 Fig. 4.133. Ray probability distribution simulation for PrRT = 0.1 and PrA = 0 with its three sections......................................................................................... 113 Fig. 4.134. Ray probability distribution simulated by the Motif Model with PrA = 0 ........................................................................................................ 114 Fig. 4.135. Sections of ray probability distribution simulated by the Motif Model..... 114 Fig. 4.136. Ray probability distribution for PrRT = 0.5 and PrA = 0.3 .......................... 114 Fig. 4.137. Sections of ray probability distribution for PrRT = 0.5 and variable PrA indicated in the chart.................................................................................. 115 Fig. 4.138. Problem geometry ..................................................................................... 115 Fig. 4.139. Ray probability along an empty floor with transmitter as a function of floor reflection ........................................................................................... 117 Fig. 4.140. Ray probability along an empty floor with transmitter as a function of floor absorption PrA for an isotropic antenna............................................. 117 Fig. 4.141. Ray probability as a function of the antenna directivity and floor reflection at normed distance r0 = 50 m from a transmitter....................... 117 Fig. 4.142. Maximum directivity of the antenna pattern formed by the n-times powered cosine........................................................................................... 118 Fig. 4.143. Example of signal level prediction at 100 MHz for different corridor cross-sections and radiation power of 1 W ................................................ 122 Fig. 4.144. Example of a common ray and an expanded ray propagating between transmitter and receiver at 100 MHz ......................................................... 122 Fig. 4.145. Power flow between transmitter and receiver at different frequencies...... 124 Fig. 4.146. Indoor propagation at different frequencies for an isotropic transmitter... 125 xii
Fig. 4.147. Example of power flow around a small obstacle at 20 MHz ( = 15 m)... 125 Fig. 4.148. Example of power flow through and around a corridor at 100 MHz......... 125 Fig. 4.149. Determining the angle of arrival of a ray................................................... 126 Fig. 5.1. Fig. 5.2. Fig. 5.3. Fig. 5.4. Fig. 5.5. Fig. 5.6. Fig. 5.7. Fig. 5.8. Fig. 5.9. Floor plan of CTU building ....................................................................... 127 Measured and predicted coverage by different models ............................. 128 Differences between measurement and predictions................................... 128 Comparison of signal coverage predictions along four sections ............... 129 Stuttgart University building...................................................................... 130 A detail from the floor plan of a Stuttgart University building ................. 130 Comparison of the measured and predicted signal level from transmitter A along section A ................................................................... 131 Comparison of the measured and predicted signal level from transmitter B along section B .................................................................... 132 Measured and predicted signal coverage by three different models for transmitter position B................................................................................. 132
Fig. 5.10. Other Motif Model predictions in a building of the University of Stuttgart ..................................................................................................... 136
xiii
List of Tables
Tab. 4.1. Tab. 4.2. Tab. 4.3. Tab. 4.4. Tab. 4.5. Tab. 4.6. Tab. 4.7. Tab. 4.8. Tab. 4.9. Example of optimised motif parameter values and their associated quality measures obtained with ( 1 , 5 ) -ES- -SA ........................................ 44 Transmitter and receiver parameters............................................................ 71 General motif parameters at 1.9 GHz .......................................................... 76 Resultant prediction accuracy in four tested building.................................. 84 Motif parameters for CIT building complex at 0.5 GHz ............................. 85 Universal motif parameters at 1.9 GHz ....................................................... 90 Motif parameters for two discrete sets......................................................... 90 Motif parameters for seven discrete sets...................................................... 91 Dependency of prediction accuracy on the number of discrete motif parameter sets............................................................................................... 91
~~
Tab. 4.10. Specific Rices k factor prediction............................................................. 106 Tab. 5.1. Tab. 5.2. Tab. 5.3. OSM and MWM parameters by COST 231 .............................................. 127 Time complexity of models ...................................................................... 134 Models frequency bounds .......................................................................... 135
xiv
1. Introduction
1.
Introduction
The slowdown in the growth of the communications industry in the last two years has led to greater concentration on attracting costumers by new, more alluring services, for which the technology was already available. However, the new services of the third generation (3G) mobile communication systems tend to demand increased network capacity. This has stimulated the motivates towards the development of algorithms that exploit adaptive antennas [1], and increasing attention has also been paid to spatial selectivity by multiple-input multiple-output (MIMO) [2, 3] broadband wireless communication systems, where multiple antennas at both the transmitter and the receiver open up effective multiple parallel spatial data pipes, or add diversity. Besides 3G systems, there is also IEEE 802.11 wireless local area network (WLAN) technology [4], also known by the commercial trademark Wi-Fi. These networks have quietly evolved, especially in the United States, into the fastest-growing wireless data application in the world. Wi-Fi networks have been springing up, not only in businesses and self-contained organisations but also in public places like waiting rooms and coffee shops as well. Yet Wi-Fi systems still proves more complementary to 3G systems than competitive. When cellular-based data services gives users seamless coverage and mobility, while local area services will give them high speed in selected areas. Wi-Fi also stands to benefit from the fact that cellular systems are found everywhere in Europe, which create opportunity for their synergies. The design and deployment of both Wi-Fi and 3G wireless networks employing new technologies has become all the more crucial, and requires careful sate-specific planning and prediction of all appropriate mobile channel parameters [5]. There is huge demand in the wireless industry for the development of accurate propagation prediction techniques. In this field, the current trend in indoor prediction is shifting from empirical models to complex deterministic models, due to their decreasing computation cost and the increasing requirement for wideband prediction, which is by empirical models reduced. State-of-the-art models are usually based on a deterministic algorithm with many empirical features [6]. However, current concepts have still shortcomings and actually are not able to fully satisfy increasing demands for complete prediction of mobile channel parameters, as required by the rising third and fourth generation of wireless mobile systems incorporating adaptive antennas or MIMO systems. A combination of deterministic models with statistics is believed to yield significant benefits [7]. It opens up a challenging opportunity for a new approach with cutting-edge features, providing wireless engineers with any number of accurate and site-specific channel parameter predictions, including delay spread, angle-of-arrival and departure, fading characteristics and the complete channel impulse response with s large dynamic
1. Introduction range, which is needful, e.g., for the system level simulations of MIMO technologies. The new model should be more easily applied and more robust than purely deterministic models. The goals of this thesis are, firstly to summarise the research results for an innovative model with cutting-edge features known as the Motif Model, which aims to be really suitable for indoor propagation prediction. Though many papers have dealt with particular aspects of the Motif Model development, no full, unabridged description of the Motif Model has been published before. The comprehensive description of the Motif Model algorithm, together with the presentation of new concept for describing an indoor environment, should offer a totally new perspective on propagation prediction in a generally indescribable environment. This dissertation begins by laying out the main aims of the thesis, in Chap.2. Chap.3 addresses issues associated with indoor electromagnetic wave behaviour, and presents an overview of current indoor electromagnetic wave propagation models. Chap.4 presents the key contribution of the thesis, describing the authors research and results so far. Chap. 4 also validates the Motif Model performances and flexibility by evaluation measurements and a variety of model applications. Chap.5 compares the model with other state-of-the-art models from various points of view. Chap.6 then summarizes the results and offers proposals for further research. The thesis ends with a brief conclusion in Chap.7 and a list of references and authors publications in Chap.8.
2. Aims of the Dissertation
2.
Aims of the Dissertation
The main aims of this dissertation can be summarized as follows: 1. Summarize current approaches to indoor electromagnetic wave propagation modelling and their capability to satisfy the demands of mobile systems of new generations, in order to explain the motivation for developing the novel model. Evolve a new indoor electromagnetic wave propagation model, which outperforms current models, and which achieves features really suitable for specific and complex indoor environments. Implement the new approach in order to verify its performance. Carry out evaluation measurement campaigns at the frequencies of current mobile systems, and compare the new approach with current models.
2.
3. 4.
3. Current Approaches to Indoor Propagation Modelling
3.
Current Approaches to Indoor Propagation Modelling
This chapter describes common electromagnetic wave indoor propagation behaviour and the main electromagnetic phenomena that occur inside buildings (Chap.3.1). Further it brings an overview of various models for indoor propagation prediction (Chap.3.2) together with a summary of their advantages and disadvantages (Chap.3.3), which provide the motivation for developing the novel model.
3.1
General Outline of Electromagnetic Wave Propagation Inside Buildings
This chapter gives a general outline of the electromagnetic wave behaviour inside buildings. However, it does not provide a rigorous description of electromagnetic wave propagation and interaction, because an application of a rigorous description is very difficult in indoor propagation due to an impossible detail building structures description, anyway. Moreover, a rigorous description of electromagnetic wave propagation and its interaction with even idealized obstacles is a quite complex problem, since all electromagnetic phenomena would have to be taken into account, and as the obstacle becomes more complex (inhomogeneous) an overall mathematical description becomes less and less feasible.
Electromagnetic Wave Indoor Propagation Behaviour

Electromagnetic wave propagation within the highly obstructed environment that is usually found inside buildings is brought about mostly by a combination of many mechanisms. First of all, for the indoor case the wave can simply propagate at a relatively short distance between obstacles, just through the air. The impact of the wave on a conducting obstacle induces currents in the obstacle, which are successively reradiated by means of further electromagnetic waves in relevant directions. If the obstacle is at least partly conducting, the current can be guided along the whole obstacle in the form of bound waves [8] even though of the impinging wave interacts only with a fraction of the obstacle. This effect is encouraged especially by wide-spread reinforced metal rods within building skeletons, metal drains and water piping, central heating system pipes, metal window frames, metallized window panes and so on. These can all very effectively guide the bound waves over surprisingly long distances and reradiate parts of the bound waves at the obstacle discontinuities and finally at the end of the obstacle. Even non-conducting dielectric obstacles can guide waves as dielectric waveguides. It will be discussed in detail together with corridor waveguiding mechanisms. 4
An indoor environment is usually very changeable, due to moving people, doors, windows, lifts, furniture and equipment. People have a particularly big influence on propagation waves at frequencies of mobile systems, because the human body is highly conductive. For example, a group of people filling a significant part of the cross section of a corridor can reduce the signal level behind them by a few tens of decibels. Every place in a building passes still an infinite number of waves from various directions, which interfere with each other and further contribute to extensive signal level variation throughout the building. Many further effects can occur within buildings, making indoor propagation a very complex and fascinating process, which is not feasible to describe exactly in rigorous mathematic formulations. Indoor signal measurement and prediction is still therefore a kind of a ghost story. The main factors influencing the character of electromagnetic wave propagation are the type of building structure and the location of the antenna. There are two main types of principally different buildings from a wave propagation point of view. On the one hand, there are buildings with large open spaces; buildings with open offices and low plasterboard partitions, airport and railways terminals, shopping malls and large industrial assembly halls, for example. Wave propagation in such open spaces is naturally similar to propagation in free space, so that simple prediction models can be satisfactorily accurate. On the other hand, there are old office and residential buildings, usually with many small rooms and full of various pieces of equipment and furniture. Here, the propagation is close-knit with a specific building configuration, and any preliminary estimation of wave propagation can be misguided. More sophisticated models should be used instead. The waveguiding effects of obstacles and their configurations (e.g., corridor, lift or air shaft) and obstacle shielding effects are naturally the dominant factors in indoor wave propagation. If the corridor waveguiding effect is moreover enhanced by locating an antenna in the corridor, the dominant radiated wave will be along the corridor almost regardless of the antenna radiation pattern. The wave propagation attenuation along a corridor can in this case be even less than in free space. The corridor wave distribution effect into adjacent rooms can be very beneficial for attaining the greatest possible coverage from a single antenna. If the coverage area is to be limited however, it is preferable to locate the antenna inside a room, due to the shielding effects of the walls. Wave propagation prediction can sometimes be simplified in multi-storey buildings by assuming wave propagation only in the horizontal plane i.e., only on they floor where the transmitter is located. This assumption is sometimes satisfactorily fulfilled, because the transmission loss of the floors is usually much higher than the loss of partitions on the floor. Indoor propagation can be strongly influenced not only by the structure of the building, but also by the surrounding environment. For example, adjacent buildings can be effective reflectors, which may encourage propagation between floors [9]. 5
3. Current Approaches to Indoor Propagation Modelling Indoor propagation is a site-specific task and the choice of a suitable modelling approach should take this into account.
Waveguiding Effect of Corridors

Although this chapter only summarizes the problems, I would like to describe the corridor waveguiding mechanism in more detail. This description also illustrates the high complexity of indoor propagation. The mean signal attenuation can be characterised by a signal decay factor [5], which equal to two corresponds to the wave power attenuation in free space, which is proportional to the second power of the distance from a transmitter. As the decay factor decreases, so does the signal attenuation, and vice versa. The signal decay factor along the corridor can usually be reduced to the range of 2 1.3 and lower. There are in fact two mechanisms that support the waveguiding effect of a corridor (Fig.3.1). Dominant Corridor Waveguiding Mechanism At the beginning of the corridor close to the transmitter, the signal level is given by the common waveguiding mechanism of corridor, due to wave multi-reflection within the corridor. However, the mainly non-conducting materials of the surrounding walls, ceiling and floor can only partly encourage this waveguiding mechanism. In long corridors with relatively homogeneous walls, however, an even lower signal decay factor can be found, due to the existence of a second mechanism bound with the waveguiding effect of surrounding dielectric walls, ceiling and floor. Minor Corridor Waveguiding Mechanism The waveguiding effect of the dielectric waveguide is generally bound with possible total reflection of the guiding waves from a surrounding boundary between core and cover. However, the propagating wave through the core can be totally reflected from the surrounding boundary only if the refractive index of the waveguide core is higher than the refractive index of the cover [10, 11]. Hence the walls surrounding the corridor can be more effective waveguides than the corridor itself. The propagating waves within the walls are called guided waves, due to their total reflection [10]. If the wall material was lossless, the guided waves would be losslessly propagating. Along with the guided waves, there must also exist a around dielectric waveguide evanescent (surface) waves [10] owing to the waveguide boundary conditions, which must conform with electromagnetic principles. The evanescent waves propagate in parallel along the waveguide (wall), and their amplitude diminishes exponentially with the perpendicular distance from the waveguide boundary (Fig.3.1). The evanescent wave decays along the waveguide (wall) only as the guided wave within the wall, so it can be attenuated with the distance from the transmitter slower than the direct and reflected waves propagating within the corridor. The existence of evanescent waves is
3. Current Approaches to Indoor Propagation Modelling also supported by the fact that, far from a transmitter, the signal level close to a wall is usually higher than the level in the centre of a corridor.
Guided Wave Wall Coupling with Walls Trans. Evanescent Wave
Direct and Reflected Waves
Waveguiding Effect of Corridor Contribution of Evanescent Waves
Fig. 3.1. Waveguiding mechanism of a corridor The complete corridor waveguiding mechanism is shown in Fig.3.1. The transmitter on the right side radiates a spherical wave at the beginning of the corridor. At first the radiated wave is coupled with the surrounding walls, which behave as dielectric waveguides to excite within them guided waves, and outside them evanescent waves. Then with increasing distance from the transmitter, the level of the direct and reflected waves is bound with the above mentioned common waveguiding mechanism due to their reflection, whereas the evanescent waves can diminish much more slowly along the corridor. The contributions of the two mechanisms depend on many factors, e.g., as the homogeneity and material parameters of the walls. Whilst homogeneity supports the second mechanism, increasing material conductivity supports the first mechanism and reduces the second one. In addition to these two main mechanisms, even other mechanisms can exist. For example, the bound waves can be guided along reinforced rods, water and heating pipes, etc., which may be inside the floor and ceiling of the corridor. This example of the most frequent wave propagation mechanism inside buildings shows the high complexity of indoor propagation phenomena. Therefore only rigorous numerical methods should be used, but the necessary rigorous building description is totally unavailable. Thus many simplifying assumptions and approximations are necessary in order to make the problem of indoor propagation to some extent feasible.
3.2
Models for Indoor Propagation Prediction
Indoor electromagnetic wave propagation modelling is a quite new and still rapidly developing discipline. It has become essential with the installation of WLAN and picocell mobile systems installation inside buildings. Many companies spent a great deal of their resources on satisfactory automating their indoor wireless system design supported by indoor propagation modelling [12, 13, and 14] and others. Most of them have simply adapted their outdoor prediction models on the basis of either an empirical or a deterministic approach. This has led to a variety of different models originally unsuited for application to indoor environments. The first exception was the Motley-Keenan model [15], later adopted for example by COST231 [16] and know as the Multi-Wall model. The most general division of the models can be on the basis of fundamental model approaches into deterministic and empirical models. Whilst deterministic models try to follow electromagnetic wave propagation by a rigorous description, empirical models take advantage of opportunity to estimate the signal level in a simple way by single formulas with empirical parameters based on measurements. In recant last years some interesting models primarily aiming at indoor environments have been put forward. The most interesting models are the Dominant Path model [17] and the ParFlow approach [18]. However, the indoor environment is so erratic that indoor propagation modelling is still a largely open problem. None of the present day-models generally outperforms the other. Thus many various models are still being used today from the pure empirical One-Slope model via the Multi-Wall model to various deterministic models. The principal models arranged from the deterministic to the empirical are shown in Fig.3.2. On the right-hand side there is an indication of the relations used in the modelling. However the model arrangement is problematic, because most of the successful models combine deterministic and empirical features. There follows a brief description of the models, beginning with the most using empirical and ray-tracing models. The description itself is primarily based on literature references.
Model Category
Model
Used Relations
Deterministic
Transmission Line Matrix Hybrid
ParFlow Maxwell Equations
Method of Moment + Ray-Optical Models Ray Optical Models Ray-Tracing Ray-Launching Fresnel Equations + Uniform/Geometry Theory of Diffraction Probabilistic Relations
Semi Det.
Motif Model ~ Wave Oriented Ray Model Semi Emp. Empirical Hybrid Ray Optical Models + Multi-Wall Model Dominant Path Model Multi-Wall Model
Empirical Relations One-Slope Model
Fig. 3.2. Principal indoor models
COST231 One-Slope Model

Empirical models describe the signal level loss by empirical formulas with empirical parameters optimized by measurement campaigns in various buildings to make the empirical parameters of the model as universal as possible. The COST231 One-Slope model (OSM) [16] is the simplest approach to signal loss prediction, because it is based only on the distance between the transmitter and the receiver. This simplest prediction model does not take into account the position of obstacles, the influence of influence which is respected only by the power decay factor (3.1). Factor n and the signal loss at a distance d0 from the transmitter L0(d0) in equation (3.1) increase for a more lossy environment, but they are constant for the whole building.
d LOSM = L0 ( d 0 ) + n10 log d0

where: LOSM...........Predicted signal loss (dB) L0(d0).........Signal loss at distance d0 from transmitter (dB) n ................Power decay factor (-) d ................Distance between antennas (m) d0 ...............Reference distance between antennas (usually 1 m) (m)
(3.1)
Path loss computation with the OSM is very fast, and the influence of obstacle database accuracy can be neglected, because the obstacles are in any case not site-specifically 9
3. Current Approaches to Indoor Propagation Modelling considered in the prediction. The model can only be calibrated by optimizing factor n and L0(d0) from measurements. However, their values are also recommended and tabled for different kinds of buildings [16] or the signal loss at distance d0 from transmitter L0(d0) can be considered to be equal to the signal loss in free space (3.2).
4d0 L0 ( d0 ) = FSL ( d0 ) = 20 log

where:
(3.2)
FSL(d0) .....Free space loss at reference distance d0 from transmitter (dB) ................Wavelength (m) The OSM prediction is beneficial only for an unknown building description, where other models can not be applied, or for a truly homogeneous environment, which, however, is not found inside buildings. Despite of its great simplification of signal propagation prediction, the OSM is still often used for sketching WLAN access point locations, or for estimating the necessary number of access points.
Multi-Wall Model
The OSM is insufficiently accurate for most applications, due to the usually inhomogeneous structure of building with long waveguiding corridors or large open spaces on one side and small complex rooms with many obstacles on the other side. For such cases, the more accurate, but still partly empirical, Multi Wall model (MWM) employing a site-specific building structure description can be used. The Multi-Wall model [16] takes into account wall and floor penetration loss factors in addition to the free space loss (3.3). The transmission loss factors of the walls or floors passed by the straight-line joining the two antennas are cumulated into the total penetration loss LWalls (3.4) or LFloor (3.5), respectively. Depending on the model, either homogenous wall or floor transmission loss factors [15] or individual transmission loss factors [19] can be used. The more detailed the description of the walls and floors, the better the prediction accuracy. The penetration losses are optimised as other empirical parameters from measurements, so they are not equal to the real obstacle transmission losses, but only correspond to the appropriate empirical attenuation factors of the obstacles.
LMWM = L1 + 20 log(d ) + LWalls + LFloors

where:
(3.3)
LWalls = awi k wi
i =1
(3.4) (3.5)
LFloors = a f k f
LMWM .........Predicted signal loss (dB) L1 ...............Free space loss at a distance of 1 m from transmitter (dB)
10
3. Current Approaches to Indoor Propagation Modelling LWalls ..........Contribution of walls to total signal loss (dB) LFloors .........Contribution of floors to total signal loss (dB) awi ..............Transmission loss factor of one wall of i-th kind (dB) kwi ..............Number of walls of i-th kind (-) I.................Number of wall kinds (-) af ...............Transmission loss factor of one floor (dB) kf ................Number of floors (-) Since the MWM considers the positions and specific transmission loss factor of walls, its results are more accurate than those of OSM. However, the shadowing effect of more closely adjacent walls are often overestimated, because their cumulated transmission loss factors lead to very small values of predicted signal level behind these elements. In other words the real signal may not follow a straight-line between antennas, but it can go around the walls. The computation time of the MWM is also quite short, and the sensitivity of the model to the accuracy of the description of the building is limited due to the simple consideration of only the number of obstacles passed by a straight line.
Ray-Optical Models
As mentioned above, a mathematical description is not feasible in an indoor environment, due to its complexity and the requirement for an exact site-specific building structure description. To overcome the description complexity and separate out the main electromagnetic phenomena, some simplifications must therefore be applied. The basic simplification used by most present-day deterministic models is based on a wave approximation by ray-optic principles. This makes the wave description much easier, and, for example, only two cases of a ray impinging upon the obstacle can be discerned. The first case is the ray impinging upon a plane boundary when the Fresnel equations [11] are used to calculate the ray specular reflection or direct ray transmission. In the second case, if the ray strikes just an obstacle edge the Uniform / Geometry Theory of Diffraction (UTD/GTD) [20, 21 22, and 23] is applied instead. Use of the Fresnel equations is very straightforward, if the boundary is expected to be sufficiently large, plain, smooth and homogeneous (with respect to the wavelength), and if the two materials forming a boundary have known electromagnetic parameters. Otherwise, if any assumption cannot be satisfied, usually due to a more complex obstacle structure, the Fresnel equations should not be used. However, ordinary deterministic ray models used them. Use of UTD or GTD is much more difficult than the Fresnel equations, because only a few very easily idealised obstacle configurations can be described by feasible mathematical relations. The easiest case is ray diffraction on an edge, when the edge is infinitely conductive, thin and long [23]. However, a solution of the Fresnel integrals is required even for this easiest diffraction calculation. A more precise approximation in the absence of any of these conditions makes the problem more and more complicated [24]. 11
3. Current Approaches to Indoor Propagation Modelling A description of different diffractions on more realistic obstacles is still a big challenge, and many authors have published various solutions for different approximations and conditions [25, 26, 27, 28, and 29]. However, these are mostly so complicated that they cannot be directly applied within the complex indoor environment. Moreover, their accuracy is heavily degraded by inaccurate obstacle descriptions, which are common in the indoor environment. Thus, these the more and more precise, rigorous diffraction characterizations, do not usually lead to much higher accuracy of indoor propagation prediction. Ray-optical models are the most widespread used models, but they were originally developed for an outdoor environment and only then subsequently adopted for indoor signal coverage prediction. However, ray-tracing models are not very suitable for an indoor environment, because the outdoor model application takes into account large obstacles compared to the wavelength, e.g., as buildings, terrain profile and occasionally trees. Indoor applications on the other hand, also need to take into account much smaller obstacles in other to make reasonable characterization of the complex indoor environment. The principles of the ray-model assume that all relayed power between ray vertices is focused into the ray axis. This assumption came from the fact that the relayed power between antennas is predominantly cumulated within the first Fresnel zone [23], which is very narrow for long distances between outdoor antennas at the frequencies of mobile systems, so that its approximation by the ray is a reasonable simplification (Fig.3.3a). The consequential adoption of ray-optical models for the much more refined indoor environment is not very reasonable, because the distances between the ray vertices are much smaller, while the first Fresnel zone cross section dimension remains almost same (Fig.3.3b). Thus the Fresnel zone is now much thicker relatively to the distance of the ray vertices, and the ray approximation becomes questionable. This implies that for the relatively thick first Fresnel zone in an indoor environment, is a much higher probability, than for outdoor case, that some obstacle may interact with only a part of Fresnel zone, while the ray axis is not obstructed. In such cases, ray models neglect the ray interaction altogether.
a) Outdoor environment
b) Indoor environment
Fig. 3.3. Examples of first Fresnel zones at a frequency of 2 GHz for different distances of ray vertices those are common for a) outdoor and b) indoor environments
12
3. Current Approaches to Indoor Propagation Modelling The problem of ray approximation in an indoor environment can be illustrated by the following examples. The first example is shown in Fig.3.4, where the two configurations are the same, apart from the dimension of the obstacle. Different obstacle dimensions naturally cause different influences on the impinging ray propagation. In case a) most of the ray power really interacts with the obstacle, whilst in case b) only a fraction of the ray power interacts and the rest just passes around the obstacle. The important behaviour b) cannot be properly characterised by ray-optical models, because the total ray power is assumed to be focused in the ray axis. Hence in both case a) and case b) the ray-optical models assume the total ray refraction.
Impinging Ray Obstacle Impinging Ray Obstacle
Transmitter First Fresnel Zone Cross Section Almost Total Reflection and Diffraction
Transmitter First Fresnel Zone Cross Section Fractional Reflection and Diffraction
a) Most of the ray power interacts
b) Only a fraction of the ray power interacts
Fig. 3.4. Natural influence of obstacle dimension on ray propagation This unwanted effect can introduce quite a large error into the resultant signal ray-optical prediction, especially in the case of an indoor environment densely obstructed by relatively small obstacles. This error could be reduced by a proper ray refraction characterisation instead of just full ray power refraction. However, such a rigorous description of all possible cases of ray interaction is in practice infeasible. From this point of view, ray approximation in itself is valid only within environments consisting only of big obstacles, at least 10 times bigger than the wavelength [8]. Neglecting small obstacles is therefore practical in ray-optical models, since it usually causes a smaller error in predictions than their inaccurate inclusion would lead to. The second example deals with a similar problem, when a ray passes around a nearby obstacle. This problem is raised by the fact that ray-optical models usually discern only two kinds of ray interactions with obstacles (Fig.3.5). In the first case, ray models assume that an obstacle crosses the whole first Fresnel zone a), hence only the reflection and transmission are taken into account. The other possible interaction assumes that the ray strikes just the obstacle edge b), taking into account only ray diffraction. The third common case, especially in any indoor environment is when the ray passes around an edge c), and the obstacle crosses only a part of the first Fresnel zone, but a considerable part of the ray power is still scattered. However, ray-optical models do not take into account any interaction in this case.
13
Incoming Ray Transmitter
Obstacle Transmitting Ray
Incoming Ray Transmitter
Obstacle Edge
Obstacle Edge Transmitter
Passing Ray
First Fresnel Zone Cross Section
Reflected Ray
Power Diffraction
Partial Diffraction
a)
b) Fig. 3.5. First Fresnel zone interaction with obstacle
c)
a) Obstacle crosses the whole of first Fresnel zone Fresnel relations. b) Fresnel zone axis (ray) strikes only the obstacle edge GTD/UTD relations. c) Obstacle crosses only a fraction of the first Fresnel zone part of the ray power is diffracted, but ray-optical models totally neglect it. The above examples are the simplest cases of ray interaction in an indoor environment, where much more complex obstacles are common, and differentiating between simple diffraction and reflection becomes quite impossible. Moreover, other kinds of ray scattering, known as ray diffusion, due to rough surfaces should also be taken into account. Unfortunately, ray diffusion is very hard to incorporate into the common ray-optical models, due to the grate increase in time complexity. Thus ray-optical models can be used only for UHF and higher frequencies. Neglecting smaller obstacles greatly simplifies the description of the building, but even these small obstacles have a certain influence on real signal propagation. Neglecting them means that empirical corrections of ray models by various empirical approximations [30] are essential, if an accurate prediction inside buildings is to be obtained. Two different principles can be utilised for ray-optical models: ray-tracing or ray-launching (Fig.3.6).
T mirrors
R1
new T
R2
a) Ray-tracing
b) Ray-launching (shooting)
Fig. 3.6. Principles of ray-optical models Ray-Tracing The ray-tracing algorithm determines all relevant rays for each receiver point independently of the other points by successive transmitter mirroring over the obstacles and obstacle visibility verification. The computation time increases in comparison to the ray-launching, but on the other hand constant resolution and accuracy can be obtained. 14
3. Current Approaches to Indoor Propagation Modelling The computation of the signal level is done through the GTD/UTD and Fresnel equations, as the ray-launching. Ray-Launching (Shooting) The ray launching algorithm launches rays in discrete angle increments from the transmitter and determines their path through a building. It there is an intersection between a ray and an obstacle, the specular reflection angle is computed and the penetrated and reflected rays are launched independently from each other. If the ray passes an edge, all rays on a diffraction cone [23] must be considered. Therefore an angle increment is defined and a discrete number of rays are launched from the diffraction point. If the ray intersects a prediction plane, the signal level of this ray is added to the already computed signal level of a receiver point. Ray propagation is ended if the number of ray interactions is higher than a predefined number or if the signal level at the end of the ray is smaller than a predefined threshold. Among the main disadvantages of deterministic prediction models is their very long computation time [31, 32]. Different papers present some ideas on acceleration the prediction [33, 34] and some of them lead to acceleration factors beyond 1000. However, these algorithms need preprocessing of the building database, which is itself also very time-consuming [35]. Other papers approach to the acceleration for example by a parallel ray-tracing calculation on multiprocessors systems [36]. A further disadvantage of deterministic models is their marked dependence on the accuracy of the database. Small error in the position and material of the obstacles influences the predicted results [37], but the accuracy of building databases is not generally very high. These effects are very important, and they limit the wider spread of ray optical models for indoor propagation prediction.
Dominant Path Model

The main ideas at the centre of the Dominant Path model (DPM) [17] came from observations of traced rays, which frequently pass through similar rooms between a transmitter and a receiver. The relayed power is then propagating mainly through the same sequence of rooms. Such similarly propagating rays can therefore be grouped into a dominant path (Fig.3.7). The model then traces only the different dominant paths by means of a room constellation and their neighbourhood description by a preprocessed room oriented database.
15
R1
R2
Fig. 3.7. Example of room-oriented dominant paths The signal loss of each dominant path can be predicted by a neural network [38], in which different neurons are bound by semi-empirical relations. The neural network must be trained by measurements, similarly as other empirical models. According a DPM description [17] the model is insensitive to the accuracy of wall location owing to the principles of the model, when only the positions of the whole room with respect to adjacent rooms are important for room-oriented dominant-path tracing. Nevertheless, it is obvious that the accuracy of DPM depends mostly on the authors experience, because the model prediction is based on the suggested neural network with semi-empirical equations trained by measurements. The absence of physical fundaments within the model algorithm may cause model instability or poor capability of the model to characterise real power flow. However, this model is a very novel and progressive approach using different principles of signal level prediction than other models. For this reason, the prediction of DPM is included for model comparison in Chap.5 together with the other models.
ParFlow Approach
The original Parflow approach was proposed by Choppard et al. [39] in the context of GSM base station planning. This technique is a time-domain discreet approach, which accurately reflects the behaviour of wave propagation but in turn requires high computation and time resources. The new resolution scheme (Frequency Domain ParFlow) solves the discrete ParFlow equations in the Fourier domain. The problem can thus be solved in two steps, taking advantage of a multi-resolution approach to accelerate prediction [18]. The time domain ParFlow approach simulates the field radiated by a source located somewhere on a 2D discrete grid. In this method the electrical field is divided into 5 components and the flows driven by the local transition matrices are derived from Maxwells equations. The algorithm is similar to the well-known transmission line matrix (TLM) proposed by [40]. The application of the method for indoor modelling is described in [18].
16
Ray-Optical Method of Moment Hybrid Model

The method presented in [41] performs an electromagnetic simulation using a combination of the Method of Moment (MoM) and UTD. The MoM method requires the problem to be discretized using wire segments, often in the form of a wire grid, when the surfaces are modelled. Segment and grid sizes of around 0.1 wavelengths are required. The problem solution time is then proportional to the third power of the grid vertex number. Such time complexity thus limits the size of the problems that can by solved only by using MoM. A combination of MoM with UTD can provide suitable features for making a prediction within an indoor environment. UTD is an electromagnetic high-frequency approximation theory for solving problems where the elements making up problems are large in terms of the wavelength. The UTD method complements the MoM method. Electrically large problems may be analysed using the UTD theory, whilst smaller problems may be analysed using MoM. The similar hybrid technique yielding the combination of Ray-optical approach with Finite-Element Time Domain method (FDTD) is introduced in [42].
Ray-Optical Multi-Wall Hybrid Model

The limited number of ray interactions with obstacles in ray-optical models leads to underestimated or totally unpredictable signal level in areas that are far from a transmitter. Signal prediction in such areas is much easier and above all faster by the Multi-Wall model. The transition between the two models should be smooth, so a suitable transition function is defined [43]. The combination of the two models is a compromise between the accuracy of ray-optical models and the speed of empirical models.
3.3
Summary and Motivation for Developing a New Model
Because all further steps in the design of a mobile communication network are based on the prediction, the accuracy attained with the models has an important impact on the final performance of the communication networks. New mobile wideband communication networks have been established in recent years. The bit rates of these networks can be about 100 MBit/s. Even higher data rates, e.g., for online video transmission are possible with new wideband networks. Unfortunately, these networks are still being planned with prediction models which are mostly able to compute the mean signal level only. If the mean level is higher than a user-defined threshold, high quality of the signal is assumed. However, due to the high data rates and complex indoor environment, multi-path propagation, which causes fast fading fluctuation, and also the power delay profile, should also be taken into account in
17
3. Current Approaches to Indoor Propagation Modelling the prediction models as an integral part of mobile channel performance prediction. Complete wave propagation prediction is therefore essential.
T
a) Empirical models
b) Deterministic models Fig. 3.8. Basic approaches
Fig.3.8 shows the two basic approaches to the prediction of the signal level inside buildings. On the one hand there are empirical models which analyse the direct path between a transmitter and a receiver. On the other hand there are ray-optical (deterministic) models, which determine a few main rays between the transmitter and the receiver [31]. Empirical models are the simplest approach to field level prediction. They are usually very fast and easy-to-apply. They do not need an exact building description, but their prediction accuracy is limited. Their straightforward basis makes the prediction of any other channel parameters impossible. However, there are few empirical models of some channel parameters in existence [16, 44]. Deterministic models are theoretically able to predict any channel parameter due to their deterministic principles. In practice, however, common ray approaches has been actually failing so far in complete signal propagation prediction. The other disadvantage of ray-tracing models is their high sensitivity to exact obstacle description and location. The way in which the models trace exact rays by specular reflections and direct transition means that a small difference in obstacle position can lead to totally different ray propagation and hence signal level. Moreover, the common ray-tracing models are based on rigorous equations (e.g., Fresnel equations, UTD), which need as input the exact material properties of all obstacles. However, it is not feasible to collect the exact properties of all obstacles in an indoor environment. Therefore a different simplification and approximation based on calibration measurement is applied. Ray-tracing then turns into something more like a semi-deterministic model [45, 46]. The other models do not overcome the main constraints of both principal approaches, and mostly try to yield the advantages of both concepts by combination them. All those does not really provide reasonable model for indoor propagation prediction. Thus the model suggested in this thesis will try to achieve features really suitable for indoor propagation prediction and convenient to prediction of any mobile channel parameters.
18
4. Motif Model
4. Motif Model Novel Approach to Electromagnetic Wave Propagation Modelling

This chapter forms the main part of the thesis, dealing with the description of a novel approach to electromagnetic wave propagation modelling. First, the ideas behind the new model will be delineated (Chap.4.1) and a novel approach to environment description will be explained (Chap.4.2). Then the principles of ray launching and propagation (Chap.4.3) and channel parameters perdition (Chap.4.4) are formulated. Chap.4.5 puts great emphasis on model setting and especially on model evaluation by various measurements. Examples of the great capabilities of the model are illustrated by several other model applications (Chap.4.6). Finally, a useful probabilistic model based on ray density distribution is suggested (Chap.4.7), and an interesting modification of the new model by ray expansion is proposed (Chap.4.8).
4.1
Delineation of New Model Ideas
After analysing the advantages and mainly disadvantages of current models (Chap.3), the following principal statements, which the new model should satisfy, can be put forward. The new model should not only predict dominant signal paths or rays, but should characterise real electromagnetic wave propagation in order to offer a realistic prediction of any mobile channel parameters. Besides, the new model should offer an option of much faster simple mean signal level prediction. The model should be as robust as possible, in order to be not dependant on a carefulness of the description of buildings. That is, the model should not require an accurate and complete description of obstacles. The model should represent the electromagnetic wave propagation as simply as possible, in order to be easy-to-apply, effective, and to save time and computation resources. The Monte-Carlo character of the new model should be especially conductive due to the possibility of having an adjustable level of prediction accuracy. The above statements contain some contradictions, so a reasonable compromise must be found. The two pivotal assumptions will help to find this compromise. The first statement, concerning real wave electromagnetic propagation prediction, can be achieved by applying the first assumption.
19
4. Motif Model
First Assumption:
Each general wave can be represented through the principle of superposition by generally an infinite number of individual plane waves. The individual plane waves can then be approximated by rays propagating in the directions of the individual plane waves, similarly as in wavefront-oriented ray-tracing [47]. Such wave approximation, theoretically by the infinite number of rays, partially overcomes the disadvantages of common ray-optical approximation (Chap.3), even within the indoor environment. Nevertheless, the partial ray features of such approximation remains, so the frequency limitation remains too (Chap.3.2). For example, a spherical wave can be approximated be many uniformly angularly spaced rays propagating from the centre of the sphere. The number of different ray directions (the fineness of the ray angular spacing) depends on the accuracy of real wave approximation. This kind of desired wave tracing can in fact be based still on rays, but the number of rays should be almost infinite. Therefore the common ray-tracing models cannot be applied, because they are able to trace only a few major rays between antennas, while other rays are omitted altogether.
Second Assumption:
Fast fading signal fluctuation is caused by a different phase shift of incoming waves to a receiver, but prediction of fast fading fluctuation is not usually the main aim of prediction. The influence of fast fading fluctuation, and thus wave phase shifts interference, on a mean signal level prediction is actually usually undesired. The prediction of wave phase shifts can therefore by completely omitted from mean signal level prediction. However, if the channel parameters connected with fast fading fluctuation are to be predicted, the phase of the incoming waves must naturally also be estimated. A combination of these two assumptions leads to further assertions Each individual ray can carry same unified unit of power, which is equal to the total radiated power divided by the total number of launched rays. The level of radiated power for different directions from a transmitter can therefore be represented just by ray density. The number of rays (ray density), propagating for the same direction and thus approximating a particular individual plane wave, must correspond to the power of the wave fraction that the individual plane wave represents. In the case of an isotropically radiated spherical wave, the ray density will have uniform distribution over the whole sphere. If the radiation has a general pattern the ray density for different directions should correspond to it. The receiving local mean signal level can therefore be given simply by the number of rays coming into the receiver over the total number of launched rays. These very straightforward ideas provide the true local mean signal level, and they also offer the potential of full electromagnetic characterization if the ray phase is also
20
4. Motif Model predicted. These ideas also enable the new model to have a desired Monte-Carlo character, when the punctuality of prediction of propagating waves is determined by the total number of launched rays.
Wave-Tracing Technique Ideas

In order to satisfy the demand for such wave-tracing with ray basis, an extremely rapid type of ray-launching in discrete angle steps can be used instead common ray-optical models. The most time consuming item in common ray-optical models is the ray scattering calculation and obstacle visibility verification. Both of these must therefore be avoided. The first problem is more complicated, because the interaction of the wave with the environment must comprise all electromagnetic phenomena including considerable diffuse scattering, in order to appropriately form complete subsequently scattered waves in all directions. Moreover, the characterisation of the obstacles or environment should offer the potential of universal application for any obstacle or cluster of obstacles. The complex solution of the first problem is described in detail in Chap.4.2. The practical solution of the second problem offers a modification of the simple line-drawing technique from the computer graphic utilizing the division of a picture into a grid [48]. The adaptation of this for the ray propagation prediction is described, among others, in Chap.4.3.
Conclusion
Easy correspondence between ray density and signal level, together with novel way of ray propagation and scattering within complex environment, as described below, can bring amazing features to the new model. Our model can also be called the Wave Front Oriented Ray Model, which more accurately characterises the behaviour of the model, but the name Motif Model has already been widely spread. The features required of this model are very challenging, and the rest of this thesis will show how successful we have been.
21
4. Motif Model
Outline of the Principles of the Motif Model

Chap.4 addresses each aspect of the Motif Model in detail. The following very brief outline of the principles of the Motif Model is presented at this stage in order to clarify the further description, and especially the cohesion of the individual models components. Model Fundaments: The model is based on a modified ray-launching technique. Ray propagation takes advantages of the modified simple line-drawing technique utilizing environment division into a grid. 2D or 3D environments are composed of square or cubic grid elements (Fig.4.1). Each grid element characterises the local mean environment behaviour by its own specific probabilistic parameters. The grid elements in which obstacles or a cluster of obstacles occur are called motifs, and are represented by probabilistic motif parameters, including the probability of absorption of a ray and the overall probability radiation pattern that probabilistically characterises the next direction of an impinging ray (Fig.4.2).
Overall Probability Radiation Pattern
Incident Wave
Fig. 4.1. Example of site-specific description of a building, based on grid elements
Fig. 4.2. Overall probability radiation pattern of a cluster of obstacles (Motif)
The prediction of the mobile channel parameters is based on an almost infinite number of launched rays. The mean local signal level in each grid element is then proportional simply to the number of rays passing the element over the total number of launched rays (Fig.4.3). In order to predict any other channel parameters the ray delay, phase and angle of arrival are predicted besides. Any model aspects, especially (Fig.4.4): ray launching and propagation ray interaction with motifs (clusters of obstacles)
22
4. Motif Model are driven by a random number generator to satisfy pure probabilistic model behaviour. Thus the model has a Monte Carlo character, and the resultant accuracy of the model prediction is dependent on the number of launched rays.
Fig. 4.3. Floor plan filling up with an increasing number of launched rays
Fig. 4.4. Example of the ray propagation process
23
4. Motif Model
4.2
Environment Description
The Motif Model was primarily developed for indoor propagation prediction of electromagnetic waves. A typical indoor environment consists partly of a solid building structure of walls and floors that have an evident influence on wave propagation, and partly of a vast variety of other obstacles, such as furniture, equipment, and people. These have an influence on wave propagation that may be even stronger than the influence of the solid building structure. Moreover, these various obstacles can be shortterm variants, such as people moving around, doors and lifts, or long-term variants, e.g., positions of furniture and equipment. All this makes the indoor environment hard-to-describable, or even totally indescribable. As a consequence, conventional purely deterministic models fail to provide an accurate prediction of wave propagation within such an environment. Nevertheless, the physical behaviour of the new model is important, and thus a kind of a site-specific environment description is unavoidable. Therefore the new approach should localize the main, easily-definable obstacles, forming mostly the building skeleton, in the input building plan. The other indefinable obstacles should be characterised by the mean local statistical behaviour of the space having an influence on wave propagation. Such a combination of deterministic and statistic building characterisation will yield significant benefit. In order to meat these needs, a new approach to environment characterisation is suggested.
4.2.1 A Novel Method for Describing an Environment with the Use of Motifs
All parts of a building can be characterised by their site-specific local mean statistical behaviour. A building plan divided into a square 2D grid or a cubical 3D grid is therefore useful, because each grid element (pixel) can have its own assigned attributes, which statistically characterise the mean local environment behaviour. This seems to be hard task. However, the easily localized solid building structure can form the skeleton of the environment description, as shown in example (Fig.4.5). The other vague obstacles make local changes to the statistical behaviour of the solid building structure.
24
4. Motif Model
a) b) Fig. 4.5. Typical floor plan arrangement a) divided into grid elements b) Fortunately, it was found that the number of sets of various grid elements with different statistical behaviour can be greatly reduced but can still precisely characterise any standard building. Such unified grid elements are called motifs. When the new motif concept is used, employing the statistical description of the environment, all the electromagnetic effects, including diffuse scattering, can be taken into account. If the behaviour of each motif were characterised by a straightforward calculation using for example the Uniform Theory of Diffraction (UTD) and the parameters of real obstacles, the accuracy of the Motif Model would achieved only the accuracy of deterministic models, which decreases rapidly if some significant obstacles are not included in the input database. For example, let us assume a typical wall inside a building (Fig.4.6), which is illuminated by an incident plane wave. We usually do not just find a bare wall, but we also typically find furniture or people moving near the wall. All of these have a significant influence on electromagnetic wave propagation.
Drain Cabinet Door Shelf Table Cabinet
Cabinet
Person
Incident Plane Wave
Fig. 4.6. A typical situation - wall surrounded by many vague obstacles A precise prediction of the incident wave scattering using deterministic models requires every obstacle (at least those bigger than the wavelength) and their electrical parameters to be included in the database. Our approach, however, describes walls and other vague obstacles in a semi-deterministic way, which considers walls together with other nearby smaller obstacles as a cluster of obstacles, or as a motif. Such a general motif then can be described by motif parameters (Chap.4.2.2), which characterise, among others, the
25
4. Motif Model overall probability radiation pattern of the motif (Fig.4.7). An example of a 3D motif probability radiation pattern is shown in Fig.4.8.
Overall Probability Radiation Pattern
Incident Plane Wave
Fig. 4.7. A wall fraction with randomly distributed obstacles and the overall probability radiation pattern of a reradiated incident wave
Fig. 4.8. Example of possible ray power reradiation by a plane wall The statistical motif parameters can be found either by optimization from measurements or by calculating the incident wave scattering on randomly distributed obstacles surrounding a wall (Fig.4.7). The input parameters of the later approach should be the real parameters of obstacles, but their definition is often unknown. The output is statistical motif parameters, which are obtained after analyzing a sufficient number of randomly generated clusters. The first method of gathering statistical motif parameters, through optimization (Chap.4.2.3) from measurements, was found to be the preferable way.
Number of Motif Parameter Sets

A typical building structure can be roughly contextured from basic motifs (Fig.4.9), when each motif at least approximately characterises the local shape of the main building structure. Motifs can also characterise any other obstacles, e.g., dominant furniture, equipment or plants.
26
4. Motif Model
Plane Wall motif
Corner motif
T-Wall motif
End of Wall motif
Pillar motif Fig. 4.9. Basic motifs
Cross Wall motif
Generally, every motif used for the description of the building can have its own unique set of motif parameters, which precisely characterise the local mean statistical behaviour. In practice, the number of different motif parameters to be used must be extensively reduced in order to make the description of a building feasible. It was found during research into the Motif Model that it is sufficient to distinguish up to four kinds of materials, and therefore up to four different sets of motif parameters. The practical evaluation of this statement appears in Chap.4.5.2. The different parameter sets are denoted as Light Wall, Heavy Wall, Window and Metal. The equivalent colour notation (Fig.4.10) to each set is kept in every floor plan throughout the work.
Light Wall
Heavy Wall
Window
Metal
Universal
Fig. 4.10. Examples of the material-specific variation of motifs, plus one universal motif In practice, it can still be hard to reasonably assign a set of motif parameters to each motif in the building description, even from four possible sets. For this reason, a universal set of motif parameters is also defined, but it will not be mixed with the others in one building plan. The example of the final floor plan, which distinguishes different motif parameter sets, is shown in Fig.4.11a), and example of the same floor plan forming only by the universal motif is shown in Fig.4.11b).
a) Trial floor plan compound of different motif parameter sets 27
4. Motif Model
b) Trial floor plan compound of the universal motif parameter set Fig. 4.11. Example of an environment spatial description Describing the environment by the basic motifs with the same universal motif parameter set can greatly simplify the necessary description of the building by a user, which is usually the biggest problem in signal coverage prediction, as far as the user is concerned. Thus, the utilization of a single universal set of motif parameters is highly beneficial. Moreover, just a black-and-white bitmap blueprint of the floor plan could then totally replace the typical unhelpful handmade building description by the user. Such a blueprint should be properly clarified in order to avoid confusion. Research into this interesting way of applying the Motif Model is left for a further investigation. When the environment description had been made and a propagation prediction is to run, a complete database of motif probability radiation patterns is pre-processed to accelerate the prediction. The scheme of an implemented three dimensional motif database is shown in Fig.4.12, where one dimension corresponds to different angles of ray arrival, while the second corresponds to different materials, and the last corresponds to different motifs.
Fig. 4.12. Structure of an implemented motif database 28
4. Motif Model
Grid Spacing
The question of reasonable grid spacing is quite a complex problem. There are many factors that influence it. On the one hand, the grid element spacing must ensure an appropriate environment characterisation and detailed coverage prediction with respect to the wavelength. On the other hand, as the element dimension decreases, environment specification by a reduced number of motifs and the assignment of motif parameters become more difficult. Moreover, the prediction time complexity also increases. The authors suggestion, resulting from the experimental investigation of indoor environments, is grid spacing from 1 at 0.5 GHz (60 cm) via 2 at 2 GHz (30 cm) to 4 at 5 GHz (24 cm). The experimental verification is discussed in Chap.4.5.2. The two main factors (appropriate building characterisation and time complexity), which counteract each other, can be partly overcome by applying a nonuniform grid (Fig.4.13). An environment with fast changes in statistical behaviour can be divided by a fine grid to deal with the first factor, whereas the large open spaces are tiled by larger grid elements to deal with the second factor. This kind of nonuniform grid is a common technique in many numerical methods [40, 49, 50, and 51]. It has not yet been completely implemented into the Motif Model software package and tested. Future research should investigate this question.
Fig. 4.13. Example of describing a building by a nonuniform grid
29
4. Motif Model
4.2.2 Definition of Motif Parameters

This chapter presents the description of a motif (cluster of obstacles) by three motif parameters. The first two parameters result from a suggested power balance, while the third involves forming the resultant shape of the motif radiation pattern. Finally, the relation of the motif parameters to material constants in some idealized cases is treated.
Power Balance
The suggested power balance of incident power on a cluster of obstacles refers to the following concept. First, the incident power strikes a cluster of obstacles and excites it. The excitation itself burns part of the incident power, and the remainder is only then further emitted, partly omnidirectionally and partly within the main lobes (4.1) and (4.2). PwI = Pw A + PwE (4.1) where: PwI ......... Incident power on the motif (W) PwA ......... Absorbed (burned) power in the motif (W) PwE ......... Power emitted by the motif (W)
The total emitted power consists of reflected and transmitted parts

PwE = PwR + PwT
(4.2)
where:
PwR ......... Power reflected by the motif (W) PwT ......... Power transmitted through the motif (W)
The suggested power balance is universal for general clusters of obstacles, which consist not only of homogeneous walls, but also of many other nearby obstacles.
Definition of Motif Parameters

In terms of the suggested power balance, we can define three overall motif parameters. The first motif parameter is the probability of absorption of the incident power. This is defined as
Pr A = Pw A PwI
(4.3)
then the absorbed and emitted powers are

Pw A = Pr A PwI
and
PwE = (1 Pr A ) PwI .
(4.4), (4.5)
The probability of absorption is angle invariant, and as such is applied whenever a ray strikes a motif. 30
4. Motif Model The second motif parameter is the probability of reflection of the emitted power.
PrRT = PwR PwE
(4.6)
Thus the reflected and transmitted powers are

PwR = PrRT PwE
and
PwT = (1 PrRT ) PwE .
(4.7), (4.8)
Distribution of the emitted power into a reflected and transmitted part is straightforward by previous definition in case of the Plane-Wall motif. For other motifs somewhat more calculation is necessary. The example of the T-Wall motif follows. To simplify the derivation the following assumptions are made. The incident power can be divided into two identical independent halves around the axis of the incident ray, and treated independently. Same probability of absorption is applied on both incident power halves to get the total emitted power PwE. The motif walls are infinitely thin, thus avoiding multireflections within a wall. Distribution of the emitted power into a reflected and a transmitted part should depend only on the probability of reflection and the incident angle. To satisfy this rule the next power distribution equations can be written for the T-Wall motif and incident angle of 0 90 degrees (ray incoming from the first quadrant). Distribution of the emitted powers into quadrants No. 1, 2, 3, 4 according (Fig.4.14) follows:
Pw1 =0 PwE
(4.9)
Pw2 1 = (1 PrRT ) PrRT PwE 2
(4.10)
Pw3 1 1 2 = (1 PrRT ) + (1 PrRT ) (4.11) PwE 2 2
Pw4 1 1 = PrRT + PrRT = PrRT PwE 2 2
(4.12)
31
4. Motif Model
1
PwI 2 PwI 2
PwE (1 PrRT ) PrRT 2
PwE PrRT 2 PwE PrRT 2
PwE 2 (1 PrRT ) 2 PwE (1 PrRT ) 2

3
Fig. 4.14. Distribution of the emitted power into quadrants of the T-Wall motif
1 2
PwE
Pw2
Pw4
4
Pw3
3
Fig. 4.15. Dependency of the distribution of power into quadrants on probability of reflection of the motif
Fig. 4.16. Example of T-Wall motif radiation pattern (PrDS = 0.73, PrRT = 0.5) Complete descriptions of analogous power distribution for the T-Wall motif and the other basic motifs are given in Appendix A.
32
4. Motif Model The third motif parameter is the probability of diffuse scattering of the emitted power. The probability of diffuse scattering characterizes the shape of the radiation pattern. The radiation pattern itself combines two subsidiary effects: Directional radiation formed by lobes approximated by the nth powered cosine function as in Phong's law [52] Omnidirectional radiation Phong's law has been introduced from computer graphics models, which state that a reflection should not be described by a simple cosine law as in Lambert's law [53]. Phong's law states that a specular component will be significant if the surface is considered smooth enough when compared with the wavelength, but the diffuse component must also be taken into account. The final motif radiation pattern is the simple sum of these two subsidiary effects, as shown in Fig.4.17 for the Plane-Wall motif.
Fig. 4.17. Example of motif radiation pattern formation for the Plain-Wall motif (PrDS = 0.73, PrRT = 0.6) These two subsidiary radiation effects should be characterised by one motif parameter PrDS. This specifies the probability that the emitted power is spread in all directions in order not to be consolidated within the lobes. PrDS is also roughly indirectly proportional to the directivity of the radiation pattern of the Plane-Wall motif (4.13) (Fig.4.18).
PrDS =
Pwomni 1 PwE DPW
(4.13)
where:
PrDS ........... Probability of diffuse scattering of emitted power (-) Pwomni ....... Omnidirectionally emitted power (W) DPW ........... Directivity of the radiation pattern of the Pane-Wall motif (-)
The directivity of the radiation pattern is given by (4.14) assuming the symmetrical motif radiation pattern formation for the Plane-Wall motif and PrRT = 0.5.
33
4. Motif Model DPW (PrDS ) = PrDS + (1 PrDS )
cos
(4.14)
n (PrDS )
( )d
For the nth power of the cosine, the approximative indirect empirical dependence on PrDS was found (4.15) (Fig.4.19).
1 PrDS PrDS 1 n(PrDS ) = 100 2
(4.15)
where:
n ............. Power of cosine (-)
Fig. 4.18. Comparison of 1/ PrDS, DPW and directivity Dcos of nth powered cosine lobe as a function of probability of diffuse scattering
Fig. 4.19. Variation of cosine power with the probability of diffusion scattering
34
4. Motif Model
PrDS = 0.0 1.0; PrRT = 0.6; Angle of arrival of the ray 50 deg
PrDS = 0.73; PrRT = 0.6; Angle of arrival of the ray 0 90 deg
Fig. 4.20. Example of Plane-Wall motif radiation pattern formation
Relation of Motif Parameters to Material Constants

As was explained above, the motif parameters are statistical parameters. They came from average statistical behaviour of the relevant motifs, and are thus not an exact equivalent to any common quantities. Nevertheless, it may be useful to find their approximative relations to some idealized obstacles and their material constants. Here we will be first deal with the relation of PrRT to the reflection coefficient of an ideal boundary, then with the relation between PrA and attenuation of the transit ray, due to its absorption within an ideal plane wall, and finally the relation of the measured wall transit attenuation to the motif parameters.
Relation of PrRT to the Reflection Coefficient of in Ideal Boundary

Let us assume that the power of a ray which impacts on a boundary is equally divided into both polarizations. The exact reflection coefficients of each polarization are given by the Fresnel equations [11]. Their relation to PrRT can be found, assuming that PrRT is equivalent to the mean value of the both reflection coefficients over the whole range of the impact angles. This leads to (4.16).
PrRT 1 2 2 R ( ) + R ( ) d 2 0
(4.16)
where:
R ............ Complex reflection coefficient of the parallel component of

electrical intensity to the impact plane (-) R ............ Complex reflection coefficient of the perpendicular component of electrical intensity to the impact plane (-) ............... Impact angle (rad) 35
4. Motif Model An example of the dependency of the power reflection coefficients on the impact angle for a typical wall at 2 GHz is shown in Fig.4.21, together with the value of the relevant motif parameter PrRT.
Fig. 4.21. Example of the dependency of the reflection coefficient on the impact angle (r = 3, / = 8 1013 ) The following charts of the dependency of PrRT on the material constants (Fig.4.22) and (Fig.4.23), where a horizontal variable / makes the chart invariant with frequency, may be useful.
Fig. 4.22. Dependency of probability of reflection PrRT on material constants
Fig. 4.23. Dependency of PrRT on material constants for the linear part of the previous chart, (Linear part ~ part independent of the / variable)
36
4. Motif Model
Relation of PrA to Transit Ray Attenuation Due to its Absorption within an Ideal Plane Wall
The relation between PrA and the attenuation of a ray propagating perpendicularly through a plain wall of thickness d gives (4.17).
PrA = 1 e d
where:
(4.17)
............... Specific attenuation [11] (m-1)

d ............... Wall thickness (m)
The specific attenuation depends on the parameters of the wall material and the wave frequency. Fig.4.24 may give a useful view of typical values of PrA for a wall 0.3 m in thickness.
Fig. 4.24. Dependency of probability of absorption PrA on material constants and frequency (material constants are assumed to be frequently invariant)
Relation of PrA and PrRT to Measured Plane Wall Attenuation

If we want to compare the measured transit ray attenuation through a plane wall or floor, respectively, the following relations (4.18), (4.19) and Fig.4.25 may be helpful.
Lt = 10 log ( (1 PrA )(1 PrRT ) )
(4.18)
PrRT
10 10 = 1 1 PrA
Lt
(4.19)
where:
Lt .............. Relative transit attenuation (dB)
37
4. Motif Model
Fig. 4.25. Charts corresponding to (4.18) and (4.19), respectively
4.2.3 Optimization of Motif Parameters

There are motif parameters in the Motif Model that are used to define local mean probability ray behaviour in a grid element. Motif parameters can be unique for different clusters of obstacles or can be partly or totally unified to simplify environment description. If four different kinds of motif parameter sets plus floor parameters are assumed in the general motif model configuration, 14 parameters are necessary to define them. The best way to identify appropriate values for such a number of parameters from measured signal strength distributions is by means of optimisation.
Optimisation Technique
Applying the probabilistic local environment description presents many advantages. However, it is necessary to carry out many measurement campaigns with different settings and environments to produce a table of suitable motif parameters that generalize a wide range of building environment behaviours. A similar calibration process is often used in other deterministic models, where exact calculations are also not feasible because exact data is lacking. A probabilistic approach avoids the virtually impossible characterization of all obstacles, but optimization of 14 parameters is also quite intricate, and it is crucial to select an appropriate optimization technique. The standard deviation and mean error of the difference between measurement and prediction is commonly used as an indicator of the signal strength prediction accuracy. However, these indicators fluctuate slightly due to the Monte Carlo principle of the Motif Model. Alteration of the motif parameters results in a change of the and of prediction to produce a fitness landscape containing many sub-optimal solutions with accuracy similar to the optimal global solution. In such cases conventional optimization approaches such as the basic Down
38
4. Motif Model Hill Simplex method or the exact Conjugate Gradient method are incapable of finding the optimal solution. The optimization problem is in fact such that evolutionary computation techniques are best suited to finding the optimum parameter setting for the Motif Model.
Description of Evolution Strategies and Genetic Algorithm Testing

The two best-known evolutionary computation methods are Genetic Algorithms (GA) [54] and Evolution Strategies (ES) [55]. In order to compare these two techniques, including their many variants, a software tool called Evolutionary Computation Tester (ECT) was developed (Fig.4.26).
Fig. 4.26. Our evolutionary computation tester The optimisation of semi-empirical models (i.e. the Multi Wall Model) for indoor propagation prediction was also implemented in the ECT with additive noise to simulate behaviour of the fitness landscape similar to that expected from optimisation of the Motif Model. The following summarises the main points relating to application of ES and GA in semi-empirical or Motif Model optimisation. The fitness function (cost function) is given by the sum of the standard deviation and the scaled mean value of the difference between prediction and measurement. Vectors of real numbers are used as chromosomes. Both crossover and mutation techniques are used in various forms.
39
4. Motif Model Various techniques to control mutation strength and reduce noise disturbance reduction in the case of ES were tested.
We found that ES were better suited and especially more effective than GA to accelerate the convergence towards and to find an optimal solution. An in-depth description and comparison of ES and GA performance is not the object of this work; therefore we will only focus on Evolution Strategies in the following.
Evolution Strategies
The emphasis when applying ES will be on those variants which proved most useful for the Motif Model optimisation problem. A method for estimating the optimisation limit will be described. Finally, effective ways of reducing of noise in the fitness are debated. In the following, we comply with the conventions and symbols that are used in [55]. The most important conventions in ES are as follows:
.......Number of children (offspring) .......Number of parents

A tilde ~ above a symbol represents an ES disturbed by noise. An asterisk * above a symbol represents parameters normalised by N/R, where N is the number of optimised parameters and R is the absolute distance to the optimum. Such normalisation becomes crucial if a spherical landscape model is used as in [55].
Basic ES with Disturbed Fitness Value

ES can be roughly divided into four groups according to the number of individuals and the selection technique. (~ + ~ ) ES ES with one child replacing a parent if it has better fitness. 1 1 (1 + ) ES The same selection technique as the previous ES, but with a higher number of children. (1, ) ES The parent does not take part in the selection as in the previous ES.
( , ) ES The same as the previous ES, but with a higher number of parents. This ES is less efficient than the previous one. Modifications of this ES by the intermediate ( /I, ) ES and the dominant ( /d, ) ES were also implemented and tested.
Control Mechanisms of Mutation Strength

In contrast to GA, it is important for ES to control the mutation strength such that is driven to zero as the optimum is approached in order to achieve a maximum progress rate. Two mechanisms have been suggested for this [55] 40
4. Motif Model 1/5th Rule This control mechanism keeps the Success Probability Ps (~0.2) of the optimisation process constant by directly altering . For a noisy landscape the rule can be modified to 1/10th.
-Self-Adaptation
The principal idea lies in coupling individuals with their set of strategy parameters. This means that when an offspring is selected, its strategy parameter vector also survives. During the optimisation process not only individuals are mutated but also their mutation strengths. Of the many different mutation operators that have been proposed [55], the Log-Normal operator , was implemented as follows:
~ =
= eN ( 0,1)
(4.20)
where N(1,0) is a normally distributed learning parameter.
Final Evolution Strategy Used for Motif Parameters Optimization

We carefully selected appropriate ES with respect to their computational efficiency (the number of necessary fitness evolutions) and reliability in order to find the global optimum in a noisy fitness landscape. We evaluated many variants of ES with appropriate parameters and -control that were able to find a global minimum in many test landscape. The ES that was eventually selected despite its lower efficiency ~ ~ compared to the ( 1 + 1 ) ES, was:
~~ ( 1, 5 ) -ES with:
-Self-Adaptation mechanism coordinate-dependent strengths n learning parameter from interval <0.1, 0.15> Making a distinction of N independent mutation strengths was significant because each parameter has a different influence on the fitness landscape. Fig.4.27 shows an example of a fitness landscape in which two parameters have a different influence. The position of individuals in the figure reflects different for the two dimensions. Such behaviour is highly desirable as it provides protection against pre-maturity of an optimisation process. The ( 1, 5 ) -ES is the most efficient variant of (1, ) -ES.
~~
41
4. Motif Model
(n+1)th Parameter
Individuals
nth Parameter
Fig. 4.27. Example of a fitness landscape with layout of individuals Comments on other methods: (~ + ~ ) and ( 1 + ) ES showed poor behaviour with -Self-Adaptation. 1 1
~ ( , ) ES and its variants yield high stability of the global optimum location, but ~ ~ ~
their computational efficiency is reduced by the high number of offspring. ESs with the 1/5th Rule tended to pre-maturity in our particular case despite many modifications and parameters combinations. The ECT contains many other recommended test fitness functions for which the 1/5th Rule had similar or better performance than -Self-Adaptation.
Finite Limit of ES Convergence

Due to the intrinsic noise of all Monte Carlo based methods such as the Motif Model the convergence of parameter optimisation is reduced. If the fitness landscape Q(R) within the vicinity of the optimum is approximated by a power function of distance R from the optimum (4.21),
Q( R) = cR
(4.21)
the finite limit of convergence for an infinite number of generations can be estimated by the following equation (4.22) [55]:
R =
Q N 1 2 c 4c1, *2
(4.22)
where: q ......Noise level *.......Normalised mutation strength N.........Number of dimensions c1, ......Progress coefficient
42
4. Motif Model
Implementation of ES in Motif Model Software Package

We added a new optimisation window to the existing Motif Model software package (Fig.4.28) to allow smart setting of the optimisation algorithm in conjunction with an actual floor plan. The optimisation kernel of the ECT tool was implemented into the Motif Model package in order to test different combinations of ESs.
Fig. 4.28. Motif Model software tool together with the implemented parameters in the optimisation window
Optimisation of Motif Model Parameters by ES

All parameters of motifs within a particular floor plan were optimised together to find the limit of convergence for each of them. To estimate the finite limits of convergence of each parameter it is first necessary to compute the noise level q and then to approximate the fitness function. The noise level was estimated as the standard deviation of a sufficiently large set of fitness values calculated by the Motif Model with constant optimal motif parameters. Due to the Monte Carlo behaviour of the Motif Model the noise level is reduced with the number of launched rays, as shown in Fig.4.29.
43
4. Motif Model
Fig. 4.29. Noise level of fitness function
Fig. 4.30. Example of approximation process for one parameter
For the approximation of the fitness landscape by equation (4.21) it is essential to estimate an optimum within the noisy fitness landscape. For this purpose the following method is suggested. The fitness values of particular continuously changing parameters were calculated with other fixed parameters. Then by regression the optimum was estimated and approximation (4.21) carried out. An example of such a process is shown in Fig.4.30. In order to evaluate the optimisation process from the aspect of convergence, the standard deviation par of the parameter values during the last n iterations can be compared with the standard deviation calculated from R according to equation (4.23). The convergence of par towards indicates the level of available accuracy of the optimisation process that cannot be exceeded. Therefore par can serve as a good indicator of progress achieved.
= R 0.56
(4.23)
The constant 0.56 represent a standard deviation of uniformly distributed samples in the interval <-1,1>. An example of some optimised parameter values and their associated quality measures obtained through ES is shown in Tab.4.1.
Light Wall PrRT Optimal Value (%) R (%) 90 3.80 1.98 2.13 PrA 64 1.85 3.38 1.04 PrDS 27 3.15 3.77 1.77 PrRT 98 9.76 5.15 5.47 Heavy Wall PrA 50 5.23 4.87 2.93 PrDS 60 13.10 5.85 7.35
par (-)
(-)
Tab. 4.1. Example of optimised motif parameter values and their associated quality ~~ measures obtained with ( 1, 5 ) -ES--SA 44
4. Motif Model The obtained values for motif parameters satisfy physical rules. By comparing par with we can observe slightly different values of par to . The reason for this is the behaviour of the -SA process, which tries to achieve an optimum of , but due to noisy fitness and the evolution strategy itself the actual mutation strength is altered only in the vicinity of the optimum mutation strength.
Improving Convergence Property at Noisy Fitness

If more accurate values are demanded, it is essential to reduce R is. From (4.22) we can see that R is influenced by various factors. R increases with N, q, and and decreases only with c1, ; i.e. by increasing the number of offspring. (c, and N are problem specific and therefore cannot be altered). There are two methods for reducing R : Fitness Averaging Fitness averaging aims at reducing q. The m independent fitness measurements per offspring are carried out and their mean values are entered into the original (~, ~) -ES. 1 Because the measurements are statistically independent, q is reduced by the square root of m. The m times averaged ( 1 , ) -ES should be preferred to the ( 1 , m) -ES, as m fulfils the next criterion: mc1, < c1,m (4.24)
~~ ~ ~
~~ for > 3 it is not possible to satisfy this criterion for any m. Therefore the ( 1 , 5 ) -ES ~~ with m measurement repetitions and subsequent fitness averaging is better than ( 1 , 5 m) -
ES. Inheriting Scaled Mutations This technique imaginarily decreases q by a factor , because the best offspring does not become the parent of the next generation, but rather the mutation that produced it is inherited after downscaling by . The necessary conditions for this are described in [55]. To theoretically realise the same progress rate, the mutation strength can be multiplied by factor , if the following holds: *2 N << 1 .
Conclusion
Motif parameter optimisation based on broad measurement data using evolutionary computation techniques to create the table of general motif parameters has been presented. The most suitable ESs were identified and their assessment was discussed. Improving the prediction accuracy of the Motif Model by implementing optimisation techniques proves the high performance and robustness of the Motif Model, as well the suitability of the implemented ES. 45
4. Motif Model
4.3
Ray Launching and Propagation within Grid
The main part of the Motif Model algorithm emerges from ray propagation within a grid. Rays are traced during propagation in a simple pixel by pixel way, which takes advantage of scenario description by a uniform or non-uniform grid. The empty grid elements represent free space, while the filled elements represent one of the predefined basic motifs (Chap.4.2) Propagation within a grid has many benefits. First of all, the obstacle visibility calculation is avoided when compared to other ray-optic models and thus the propagation becomes extremely fast. However, the satisfactory representation of a scenario by a grid can sometimes be tricky.
4.3.1 Description of Ray Launching and Propagation Technique

Let us assume an emitter which occupies one pixel. It does not matter if it is a real transmitter or an emitting cluster of obstacles (motif) characterized by its probability radiation pattern. An example is given in Fig.4.31, where the emitter has launched rays a-d into a uniform square grid. To obtain the same propagation loss (ray density) in every direction each ray has to fill the same number of pixels per distance. Fig.4.31 illustrates this.
Fig. 4.31. Emitter launching rays a-d Ray a (~0) propagating parallel to the grid orientation fills pixels steadily in a row. The other non-parallel oriented rays c (~45), b (~30) and d (~60) fill not only the chain of pixels corresponding to their axis, but also other neighbouring pixels or they occasionally fill some chain pixels twice. The rays uniformly launched from an emitter into a grid claiming to form an isotropic radiation pattern (Fig.4.32) must truly form it. The angle of the launched rays is determined by a generated random number and a relevant distribution function (Fig.4.32). Such a distribution function is transformed from the relevant radiation pattern, which represents the probability of the direction of the launched ray. In the case
46
4. Motif Model of an isotropic radiation pattern the distribution function is just a straight line with constant steepness. As the number of launched rays increases the radiation pattern is refined (Fig.4.33).
Fig. 4.32. Isotropic radiation pattern and relevant distribution function In the implemented form of the radiation pattern formation a discrete angle step of ray propagation is employed. This is because; there are rows of pixels precalculated for each discrete ray direction. The rows of pixels serve for locating the pixels of in the grid that are being filled by the ray during its propagation. On the one hand, this leads to further acceleration of Motif Model prediction, but on the other hand, the refined radiation pattern is not smooth enough and additional window averaging needs to be applied. Fortunately, the volume of space averaging can be very small, so it is sufficient to perform window averaging twice only over neighbour elements. I believe if angle discreteness was not applied the formed radiation patterns would be smooth. In the case of real propagation among obstacles the number of emitters increases and additional smoothing is usually unnecessary.
100 rays
50 000 rays
50 000 rays Averaged
Fig. 4.33. Refining the transmitter radiation pattern Another example of the radiation pattern (Fig.4.34) formed by a ray density follows. The ray density represents a power density (Chap.4.4), therefore the pattern concurs to a power level. From the radiation pattern a distribution function is generated, which ensures that the uniformly generated random numbers are assigned to preferable angles of launched rays. The steepness of the distribution function corresponds to then the 47
4. Motif Model probability of angle assignment. The flattest part of the distribution function corresponds to the most probable range of angles (peak of the main lobe) and, conversely, the steepest part corresponds to the least probable angles.
Fig. 4.34. Power radiation pattern and relevant distribution function The following example of a radiation pattern was formed by 200 000 rays (Fig.4.35). The same radiation pattern turned by 45 degrees is also shown (Fig.4.36) to demonstrate the algorithm orientation indifference.
200 000 rays
Fig. 4.35. Formed radiation pattern
200 000 rays
Fig. 4.36. The same radiation pattern turned by 45 degree
48
4. Motif Model The described way of radiation pattern formation is consistent with the general Motif Model concept, when complete ray behaviour is driven by a random number. An example of a complete ray propagation process within a building is given in Fig.4.37.
Fig. 4.37. Example of a ray propagation process
4.3.2 Necessary Number of Launched Rays and Ray Coverage Acceleration

A signal level, as described in (Chap.4.4), is predicted in the Motif Model by ray density. The basic Motif Model algorithm defines the volume of a ray to be equal to one unit of power. This assumption calls for a minimum number of launched rays bT, which should be estimated according to the predictable highest signal attenuation Lmax (4.25). bT = 10
Lmax 10
(4.25)
where: bT .......... Total number of necessary rays (-) Lmax....... Predictable highest signal attenuation (dB) Unfortunately, the expected highest signal attenuation is usually hard to estimate if any measurement has been omitted. Our advice for estimating Lmax is expressed in (4.26) by the transmitted power and system threshold. There are extra 20 dB added for potential higher model prediction dynamics.
49
4. Motif Model Lmax = PT Pthreshold + 20 where: PT ......... Transmitted power (dB) Pmax ...... System threshold of minimal received power (dB) This quantity of launched rays greatly exceeds the demand for smooth coverage arising from the Monte Carlo principle of the Motif Model, so it is useful to employ and the following suggested prediction acceleration. (4.26)
Acceleration of Rays Coverage

As rays are launched from a transmitter, they are more likely to cover areas close to the transmitter than other areas. Nearby areas are therefore less influenced by the statistical imprecision that arises from the Monte Carlo character of the Motif Model. Disproportional precision is a waste of computation time. If we want to obtain the same imprecision over the whole scenario, the rays close to the transmitter should not be treated individually, but can be incorporated into beams of rays and traced together instead. In this way the beam represents the same rays to avoid recurrent operations during the prediction. In practice a combination of the beam approach with the common ray approach is used to yield the best result. A combination of the two approaches lies in incorporating the rays into a beam only till the ray number is higher than the predefined level of beam splitting SB. Then the rays are treated in the normal way. There is one difference between the beam and ray propagation. A beam striking a motif is not terminated with a motif probability PrA as an individual ray is, but the number of rays in the beam is reduced by a relevant PrA multiplication instead (Fig.4.38).
Fig. 4.38. Example of a launched beam consisting of 100 rays (All motif parameters were unified to have PrA = 0.5) The number of rays launched in one beam is then:
b= bT bB
(4.27)
where: b ........... Number of rays launched in one beam (-) bB.......... Total number of launched beams (-) 50
4. Motif Model In order to keep the benefit of statistical smooth coverage, beam splitting was introduced. If the last SB rays are broken up and the average motif attenuation parameter
PrA of the whole building is estimated, then the mean number of beam interactions with
obstacles before beam splitting can be evaluated (4.28).

S log B b IB = log 1 PrA
(4.28)
where:
I B ........ Mean number of beam interactions with obstacles (-)
SB.......... Level of beam splitting the highest number of rays, when a beam is broken up (-)
PrA ....... Average probability of motif attenuations in the building (-)
If SB is assumed to equal one, the mean number of beam interactions without splitting can be estimated by (4.28) as well as by (Fig.4.39).
Fib. 4.39. Mean number of beam interactions, I B as a function of the number of rays incorporated within a beam and the average motif probability of absorption PrA The PrA value for a typical building structure varies in the range of 0.4 0.7 and the number of rays in a beam is usually about 1M. This leads according to (4.28) to the mean number of interactions in the range of 27 11. This is in substantial contrast to typical ray-optical models, where the maximum number of ray interactions with obstacles is usually limited to 6 interactions (Chap.5). Such a huge number of interactions is one of the main reasons way the Motif Model is so efficient and why it is more capable than ray-optical models. Ray incorporation is a step toward conventional ray-tracing techniques. However, if the total number of launched beams is kept in a reasonable range the prediction acceleration is enormous, while the introduced uncertainty can still be neglected. Our experience suggests 30 500 thousands rays if simple coverage is processed. For precise wideband parameters prediction even more rays are preferable (0.3 2 millions and more). Our advice for the level of beam splitting is about 1 10 increasing with the extent of the scenario. 51
4. Motif Model A stochastic simulation of the value of the introduced uncertainty in terms of average standard deviation and mean value between signal coverage predictions is shown in Fig.4.40. There are also measures m and m calculated from the same predictions, which are additionally twice window averaged over neighbouring pixels (recommended averaging technique for a prediction).
Fig. 4.40. Average standard deviation and mean value between signal coverage predictions (The simulation was carried out in the trial floor plan of CTU building, Lmax = 110 dB, SB = 1)
Conclusion
A insufficient number of launched beams could introduce significant prediction imprecision due to the Monte Carlo character of the Motif Model. On the other hand the high number of launched beams is very time consuming, so an appropriate balance between these two factors should be estimated before making a prediction. I believe the given recommendations may be a good guideline to find a desired balance. An example of filling up the floor plan with an increasing number of launched rays is given in Fig.4.41.
Fig. 4.41. Filling up the floor plan with an increasing number of launched rays 52
4. Motif Model
4.4
Prediction
The very fast ray propagation used in the Motif model algorithm allows the tracing of a huge number of rays in a fraction of time required for common ray-tracing models. This is primarily given by obstacle visibility and strenuous mathematical computation exclusion from the ray tracing, while a probabilistic environment is employed. However, raw ray coverage depends on many factors, e.g., grid spacing, 2D or 3D grid application, and the number of launched rays. Moreover, the prediction should naturally also take into account the antenna properties and carrier frequency. This chapter therefore fixes the fundamental relations of all of these, in order to achieve highly accurate signal level prediction just by ray density without the need for any secondary forced correction by measurement, as is often required in other models. Chap.4.4.2 also describes a method of basic algorithm modification to make it able to predict any other mobile channel parameters.
4.4.1 Mean Signal Level Prediction by Ray Density

This chapter deals with predicting mean signal level attenuation by ray propagation within a squared or cubic grid. First, the 2D propagation within a squared grid is described and then by analogy the 3D propagation within a cubic grid. Then, a way of representing 3D space by the 2D propagation is suggested, which leads finally to a proposal for a new probabilistic model (Chap.4.7).
2D Propagation within a Squared Grid

Ray propagation within a squared grid, as described above (Chap.4.3.1), ensures uniform ray thinning and power decay around a transmitter with increasing distance from the transmitter, and thus signal attenuation. The described angle dependent ray propagation (Chap.4.2.1) is in fact a form of transformation of a squared grid to a circular grid (Fig.4.42). However, the circle grid is only a fiction, but it greatly simplifies the subsequent mathematical description, and can therefore be useful.
~ Transmitter d
d d n= 1 2 3 n= 1 2 3
Fig. 4.42. Representation of a squared grid by a circular grid 53
4. Motif Model The ring spacing d in the circular grid corresponds to the squared grid spacing. It is the same for the element (pixel) areas. However, the ring areas need not be an integral multiple of the element areas. Let us now introduce some variables.
Surface of the Motif Model element
S = d2
where: S ...................Element surface (m2)
(4.29)
d ....................Element dimension (m)

S d S d
d d
Fig. 4.43. Surface of an element in a squared grid and in a circular grid

Mean diameter of the nth ring
r = nd
(4.30)
where: r ......................Mean diameter of ring (m) n .....................Index of ring (-)

Number of elements in the nth circle
This means that the surface of the outer circle of the nth ring over the element surface (4.29).
d 2 ( n + 0.5 ) d 2 ( n + 0.5 ) 2 N c ( n) = = = ( n + 0.5 ) 2 S d

2 2
(4.31)
where: N c (n) ...........Number of elements in nth circle (-) n ....................Index of circle (-) S ...................Element surface (m2) Considering (4.31) the number of elements in the nth ring can be determined as the difference of the number of elements between the nth and (n-1)th circles.
N r (n) = N c (nd ) N c ((n 1)d ) =
54
4. Motif Model
= ( n + 0.5 ) ( n 0.5 ) = ( n 2 + n + 0.25 ) ( n 2 n + 0.25 ) = 2n
2 2
N r (n) = 2n
(4.32)
As expected, the result corresponds to the well-known relation of the increasing circumference to radius n. Now, we can calculate the ray thinning with a distance from the transmitter. If 2D free space is presumed, the ray density or number of rays passing every ring must be same. Thus, the number of rays passing each element of the nth ring in the case of 2D free space propagation reads
b (n) = bT
1 N r ( n)
(4.33)
where: b (n) ..............Number of rays passing element (-)
bT ....................Total number of launched rays by transmitter (-) N r .................Number of elements in nth ring (-)
Considering that ray density represents a power, we can find the signal level in the elements of the nth ring predicted by the Motif-Model in 2D free space, taking into account (4.30), (4.32) and (4.33).
PR b (r ) = = PT bT 1 d = N r ( r d ) 2r (4.34)
n=
r d
where: PR ...................Signal power at a distance r from transmitter (W) PT....................Output power of transmitter (W) r......................Distance from transmitter (m) In the circular grid, element side d is still perpendicular to the rays. Therefore, it corresponds to the effective antenna length Lef, which is given [56] by:
Lef =
DR 2
(4.35)
where: Lef ...................Effective antenna length (m) DR ...................Receiver antenna directivity (-) .....................Wave length (m)
d ~ Lef
Fig. 4.44. Effective length of element
55
4. Motif Model If the effective length of a real receiving antenna was really equal to the element dimension, then it would yield the signal level as
PR d = = PT 2r Lef
d Lef
2r
(4.36)
and predicted signal loss
L2 D =
PT 2r = = PR d
d Lef
2r . Lef
(4.37)
This signal attenuation is predicted by the 2D Motif Model in an empty grid. It is consistent with the attenuation of a cylindrical wave propagating in free space. However, the real effective antenna length of a received antenna is usually different from the element size, but this can be taken into account by an additional conversion.
L2 D = 2r d d LefR
(4.38)
where: LefR ............Real effective length of receiving antenna (m) To simplify the notation, the conversion constant incorporating the effective length of the real antenna and grid dimension are introduced and taking into account (4.35) this leads to
c2 D = d d 2 = LefR GR
(4.39)
Another useful parameter is w, which is the rate of grid dimension d over wavelength . w= d
(4.40)
Naturally, the effective antenna length is dependent on the angle of ray arrival, so the conversion constant is also angle dependent. When an isotropic antenna is employed as a probe and inserting (4.40) into (4.39) we obtain. c2 D = d 2 2w = GR GR
GR =1
= 2w
(4.41)
Now, a final equation for signal loss in 2D space, inserting (4.39) into (4.38), can be constructed. L2 D = 2r c2 D d (4.42)
The dependency of conversion constant c2D on w is shown in Fig.4.45.
56
4. Motif Model
Fig. 4.45. Conversion constant c2D
3D Propagation within a Cubic Grid

Ray propagation within a cubic grid is analogous with 2D propagation, so the following derivations can easily be described.
Representation of spherical wave propagation by a cube grid
For better understanding, imagine the squares in Fig.4.42 as cubes, and imagine the circles as spheres for a 3D Motif Model. In the following text, rings also become shells.
d S ~ Aef d d d
Volume of the grid element:
V d
Fig. 4.46. 3D elements in a cubic grid and in a spherical grid
V = d3
(4.43)
where: V ..................Element volume (m3)

d ....................Element dimension (m)
Mean diameters of the nth shell
r = nd
(4.44)
where: r ...................... Diameter of shells (m)
57
4. Motif Model n .....................Index of shells (-)

Number of elements in the nth sphere
- The volume of the outer sphere of the nth shell over the element volume (4.43). 4 3 4 3 3 3 d ( n + 0.5 ) d ( n + 0.5 ) 4 3 N S ( n) = 3 =3 = ( n + 0.5 ) 3 V d 3 where: NS (n) ...........Number of elements in the nth sphere (-)
n ....................Index of sphere (m) V ...................Element volume (m3)
Number of elements in the nth shell
(4.45)
- Difference of elements within adjacent sphere (4.45) after applying (4.44). 4 4 3 3 ( n + 0.5 ) ( n 0.5 ) = 3 3 4 4 = ( n3 + 1.5n 2 + 0.75n + 0.125 ) ( n3 1.5n 2 + 0.75n 0.125 ) = 3 3 4 1 = ( 3n 2 + 0.25 ) = 4n 2 + 3 3
N r (n) = NS (nd ) NS ((n 1)d ) =
(4.46)
(2.4) can be approximated for n >> 1 as N r (n) = 4n 2 .

Number of rays in each element of the nth shell:
(4.47)
b (n) = bT
1 N r ( n)
(4.48)
where: b (n) ..............Number of rays passing element (-)

bT ...................Total number of launched rays by the transmitter (-) N r .................Number of elements in the nth shell (-)
Signal Level in the elements of the nth shell for free space predicted by the 3D Motif Model, with considering of (4.44), (4.47) and (4.48).
PR b (n) = = PT bT
r n= d
1 d2 = N r ( r d ) 4r 2
(4.49)
where: PR ...................Signal power at distance r from transmitter (W)
58
4. Motif Model PT....................Output power of transmitter (W) r ......................Distance from transmitter (m)
The element surface S (Fig.4.46) in a spherical grid is still perpendicular to the rays. Therefore it can correspond to an effective antenna area Aef, which is given [56] by
DR 2 Aef = 4
(4.50)
where: Aef ...................Effective antenna area (m2) DR ...................Receiver antenna directivity (-) .....................Wave length (m) If the effective area of the real receiving antenna was really equal to the element area, then it would yield the signal level as
PR d2 = = PT 4r 2 Aef
d 2 = S = Aef
4r 2
(4.51)
and finally signal loss

L3 D = PT 4r 2 = 2 = PR d
d 2 = S = Aef
4r 2 . Aef
(4.52)
This signal attenuation is predicted by the 3D Motif Model in an empty grid. It is consistent with the attenuation of a spherical wave propagating in free space.
Like in the 2D case the incorporation of the real antenna will now be performed by an additional conversion.
L3 D 4r 2 d 2 = 2 d AefR (4.53)
where: AefR ............Real effective area of receiving antenna (m2) And by analogy with c2D the 3D conversion constant incorporating the effective area and grid dimension of the real antenna are introduced and, taking into account (4.50) and (4.40), this leads to
c3 D d2 d 2 4 4w2 = = = AefR GR 2 GR
GR =1
= 4w2
(4.54)
Now, a final equation for signal loss in 3D space, inserting (4.54) into (4.53), can be constructed. L3 D = 4r 2 c3 D d2 (4.55)
59
4. Motif Model
Fig. 4.47. Conversion constant c3D
3D Space Representation by Two 2D Predictions

Simple ray propagation within a 2D horizontal grid agrees with real cylindrical wave propagation in 3D space, when the axis of the cylinder is perpendicular to the 2D grid. This also means that all obstacles included in a 2D grid are in fact unlimited in the dimension orthogonal to the 2D grid. This is a natural penalty of a 2D approximation, but still there is a way to make a satisfactory approximation of 3D space by amended 2D propagation. A suggested amendment involves independent signal level prediction within two orthogonal planes. Such an amendment is not exactly correct from electromagnetic theory point of view, but it seems to offer a very good approximation of reality. The two orthogonal planes have to pass a straight line between transmitter and receiver (Fig.4.48). The signal level can then be predicted in each plane by widely-available models such as: Empirical or semi-empirical models (Chap.3.2) 2D Motif Model Probabilistic model (Chap.4.7) Other deterministic models (Chap.3.2)
Vertical Plane
Horizontal Plane T
Fig. 4.48. Orthogonal planes passing transmitter and receiver
60
4. Motif Model The overall 3D signal level loss L3D can be obtained from the independently predicted signal level losses in the horizontal LH and the vertical LV plane by the suggested formula (4.56). L3D
L2DH L2DV (4.56) , otherwise In practical cases, the product of partial losses product is much higher than , so formula (4.56) can be reduced to the last term. L3 D = L2 DH L2 DV (4.57)
L2 DH L2 DV ,
L2 DH L2 DV <
A justification for the suggested transformation follows. Signal losses in free space have to be same regardless of how they are predicted. Attenuation in 2D free space (a cylindrical wave) is
L2D = LH = LV = 2r Lef
(4.58)
Attenuation in 3D free space (a spherical wave) is

L3 D 2r 1 4r 2 = = L Aef ef
2
(4.59)
The confrontation of (4.58) and (4.59) employing transformation (4.57) yields a proof (4.60).
L2 DH L2 DV 2r 1 4r 2 = = = L3 D L Aef ef
2
(4.60)
The same transformation as (4.57) should be applied for the conversion constants c2D and c3D, which take into account real antenna parameters.
c3 D =
c2 D H c3 DV
(4.61)
For an isotropic antenna the 2D conversion constants for the horizontal and vertical planes are the same, so it reads
c3 D =
c2 D H c2 DV =
2 c2 D . iso.ant .
(4.62)
where: c2DH ...........2D conversion constant for horizontal plane (-) c2DV............2D conversion constant for vertical plane (-) We can easy verify the validity by inserting (4.41) and (4.54) into (4.62), then 4w = c3 D
2
( 2w ) = 4w2 c2 = 2D =
2
(4.63)
61
4. Motif Model
4.4.2 Prediction of other Channel Parameters

Signal level prediction by ray density is apt to provide very fast and reasonable results. But it does not offer the prediction of any other mobile channel parameters of the first and second order, such as fast fading fluctuation, the angle of ray arrival, and the signal power delay profile (narrowband and wideband parameters). However the ray based algorithm of the Motif Model, which deals with an enormous number of individual rays provides a promising opportunity to make an accurate prediction of any channel parameter besides the signal level. The required Motif Model modification lies only in the assessing the phase, direction and delay of each ray. To meet these requirements the following quite easy and straightforward modification is employed.
Ray Angle of Arrival Assessment

Prediction this involves nothing more than storing the ray direction in specific prediction points (grid elements) of the building plan. The ray direction itself is naturally also provided in the basic algorithm. So no extra modification except storage is needed.
Ray Delay Assessment

Ray delay and phase prediction must be extra cared besides the number of rays within a beam. The influence of the interaction of the ray with the obstacle need to be taken into account by an extra delay and phase shift, which can be directionally dependent. However, the character and disposition of the clusters of obstacles (motifs) is such that the appropriate real delay has a probabilistic behaviour. It is therefore hard to define it exactly similarly as in the case of the probability of attenuation of the motif. Two techniques for extra delay of ray interaction are suggested.
The motif delay can be optimised as a further motif parameter, or, as implemented in the current Motif Model, we can estimate a zero delay for reflected rays and a twice longer delay than in free space for transmitting rays.
Ray Phase Shift Assessment

The ray phase shift in the current Motif Model implementation is managed in similar manner to ray delay assessment. Such a simple approach occasions inaccurate locations of fades in a fast fading prediction. Fortunately, the exact location of fast fades is not the goal of any mobile channel parameters prediction, and only statistical fading fluctuation is estimated instead. More sophisticated ray delay assessment based on optimisation by measured impulse responses and phase shift assessment remains for further research. Nevertheless the
62
4. Motif Model
promising accuracy of the impulse response predictions (Chap.4.5.3) supports the adequacy of the current implementation.
Conclusion
This chapter fixed the fundamental relations between factors, e.g., grid spacing, 2D or 3D grid applications, and the number of launched rays, antenna properties and carrier frequency, in order to achieve highly accurate signal level prediction just by ray density, without the need for any secondary forced correction by measurement. The suggested adaptation involved 3D prediction by two 2D independent signal level predictions within orthogonal planes. Such an adaptation is not exactly correct from the point of view of electromagnetic theory, but it seems to offer a very good approximation of reality at least for typical indoor configurations. Chap.4.4.2 also described a method for modifying the basic algorithm to enable it to predict any other mobile channel parameters.
63
4. Motif Model
4.5 Model Setting & Evaluation

The best way to set the motif parameters is by optimization them from measured signal coverage in various buildings. This was discussed in Chap.4.2. To meet the needs of model setting, measurement campaigns were carried out to provide a sufficient amount of various signal samples for setting the motif parameters and consequently for evaluating the Motif Model. However, in the measured coverages it is essential to reduce fast fading, so an optimal technique for doing this will be investigated first. Finally, a challenging comparison between impulse response prediction and experimental measurements is described, along with a similar ray-tracing prediction.
4.5.1 Reduction of Fast Fading

The fluctuation in level of the received signal from a communication channel point of view is caused by two influences: namely, fast fading and slow fading (local mean signal value fluctuation). While prediction of the exact location of signal level dips caused by fast fading fluctuation is not usually the object of signal coverage prediction, the local mean signal value free of any fast fading fluctuations is an appropriate prediction objective. For this reason, the measured signal coverage used for model evaluation and for optimizing the model parameters should be properly processed by window averaging to eliminate the fast fading fluctuation and retain the slow fading information due to the configuration of the obstacles. Unfortunately, this is not a simple task and the best way depends on the specific environment. This hot topic has often been omitted in many references. Exceptions are [57, 58, 59, and 60]. The way for derivation of optimal averaging is mentioned only in [57], but in a rather obscured manner. In effect, only the final recommendation is suggested. The recommended optimal length of an averaging window for mobile systems was given as 20 40. This corresponds for example to a length of 3 12 m or 1.2 4.8 m at 1 GHz or 5 GHz, respectively. It is obvious that such an extensive averaging window for the indoor case is mostly inadmissible, at least at low frequencies. In the two further references the used averaging lengths are just stated without any explanation. The window averaging lengths used by [58] and [59] are 10 and 12, respectively. Three of the papers used averaging over a power signal level in watts. The exception is averaging by [60], which operates with a relative signal level in dB, which is incorrect from the point of view of electromagnetic theory.
Stochastic Simulation of Reduction of Fast Fading

The problem of averaging is so crucial for proper tuning and evaluation of the Motif Model that stochastic simulations of fast fading reduction in the framework of Motif Model applications were carried out together with a survey of averaging results in real
64
4. Motif Model
scenarios to find the reasonable best averaging technique. At first the level of the fast fading reduction was simulated in a stochastic way, when a vast number of samples of received signal level along an imaginary measured track were calculated to cover major aspects of indoor propagation configurations. Each point of the signal level along the track (Fig.4.49) was obtained by a complex sum of individual received signals, which were generated by randomly localized emitters with an isotropic radiation pattern. The number of emitters varied from 5 to 50. The geometry of the simulation is shown in Fig.4.49. All dimensions are normed to a wavelength in the simulations to make the resultant charts invariant on the frequency.
Area of Randomly Located Transmitters
200
Path Segment Length
300
500
Fig. 4.49. Geometry of measured signal level simulation The imaginary track of the measured signal level is segmented to correspond to the grid used for a floor plan description in the Motif Model. If the calculated signal points spacing dm along the track is smaller than the track segment length, the signal points falling in one segment are averaged. This agrees with the projection of real measured data into a grid in the Motif Model. Conversely, bigger spacing of signal points spacing causes some track segments to be unfilled. A restriction on the minimum and maximum distance between the transmitters and the track ensures that the overall distribution of signal level fluctuation follows Rices distribution with small factor k. Thus the mean signal level of the samples can be approximated by a value of one. If the contributions from the transmitters were distance independent the distribution of signal level fluctuation along the track would follow the Rayleigh distribution (Rices distribution with zero factor k). To test the ability of various modifications of window averaging techniques, each sample of signal level fluctuation along the track underwent the relevant averaging. Then a stochastic mean value of the remaining signal level deviation due to fast fading as a function of the window dimension dw and the number of applied averagings is shown in Fig.4.50. In this simulation the spacing dm between signal points along the track corresponds to the wavelength ( d m = 1 ), and the track segments are twice as long as the wavelength (w = 2).
125
250
Imaginary Track of Measured Signal Level
65
4. Motif Model
Fig. 4.50. Residual fast fading fluctuation for variable averaging window dimension dw and the number of applied averaging The reduction of fast fading would be complete if the signal level along the track was constant. Then the signal level standard deviation dB would reach zero value. As can be seen (Fig.4.50), the reduction of fast fading is not linearly proportional to the window dimension or to the number of averagings. Enlarging the dimensions of a small window has a significant influence on the reduction of fast fading. This influence diminishes later. A similar dependency can also be observed for the number of applied averagings. The other stochastic simulation tests the influence of signal point spacing and grid dimension (track segment length) on the reduction of fast fading (Fig.4.51). Both dimensions are normed by a wavelength, which makes the chart invariant on the frequency.
Fig. 4.51. Residual fast fading fluctuation for variable w and the number of measured samples per wavelength (Red curve corresponds to red curve in previous chart.) This simulation gives a recommendation for the measured sample spacing, which should be at least twice smaller than the wavelength for w = 2 (an interesting correspondence with Nyquist's criterion). Even so, a bigger grid dimension (four times the wavelength) is preferable. If either the signal point or the grid spacing is too small
66
4. Motif Model
the consequent averaging should be all the more powerful. The consequent averaging used in this chart is the weakest version, with a window dimension dw = 2 pixels. However, it remains difficult to recommend an optimum averaging technique, and further observations have to be processed. Next, however, our window averaging techniques for fast fading reduction in a real scenario are described.
Window Averaging Technique

One dimensional window averaging was used in previous simulations of the signal level along the track as suggested in [57, 58, 59, and 60]. The one dimensional technique is unsuitable for a 2D building plan and its 2D modification should be used instead. Similarly, a 3D averaging modification is desirable for a 3D building plan. However, only 2D averaging is considered in the next sections. 2D averaging operates above a 2D grid, in which all tracks of measured data are projected. If measured signal samples fill in the same grid element, their pre-averaging is applied during their projection. This corresponds to the zero number of averagings in Fig.4.51. Examples of the final projection of the same measured data into grids with spacing w = 1, 2 or 4 are shown in Fig.4.54 a), k) and m), respectively (w = wavelength / grid spacing). After such a projection the consequent fast fading reduction can be processed. There are two proposed ways of window averaging, namely the spreading way and the non-spreading way.
Spreading Averaging Window
The spreading averaging window determines the mean local signal level in such a grid element which has a measured signal level or at least within its circular averaging window are line-of-sight elements with a measured signal level. The signal level of such a central grid element is then calculated as the mean signal level of all line-of-sight grid elements with a measured signal level. The central element itself need not contain a measured signal level in contrast to the non-spreading averaging technique. When all elements within the window have line-of-sight with the centre element, the window shape will be circular. Otherwise the elements without line-of-sight are not taken into account. Fig.4.52 a) d) illustrates this for different window dimensions.
a) d w = 2 pixels
b) d w = 4 pixels
c) d w = 6 pixels
67
4. Motif Model
dw
Centre Element of Averaging Window Line-of-Sight Element within Window Cluster of Obstacles Other Element Element with Measured Signal Level
d) d w = 12 pixels Fig. 4.52. Spreading window averaging technique

Non-spreading Averaging Window
Signal samples spreading around measured tracks could introduce confusion or unreal parts into signal coverage. However this effect is reduced already by the fact that only line-of-sight elements are taken into account during averaging. However, the next useful averaging technique can be used instead (Fig.4.53). Here, the mean local signal levels of elements without a measured signal level are not calculated, and such elements are left empty. Unfortunately, this deprives us of larger coverage area, which is beneficial for an optimization routine or just for a more attractive measurement depiction.
Centre Element of Averaging Window Line-of-Sight Element within Window Cluster of Obstacles Other Element Element with Measured Data
Fig. 4.53. Non-spreading averaging window technique
Window Averaging of Measured Data

The stochastic simulations that have been performed give us a first useful appraisal of the potential of different averaging windows. However, visual surveys are also very fruitful. The following set of signal coverages of the trial floor plan (Fig.4.54) is therefore given here. The notion underlying each coverage is related to the way in which the window averaging process is applied to the projected measured data. The first number (e.g., 1x) indicates the number of applied averages. The second number (e.g., over 2) states the size of the averaging window. The other optional note is non-Spread. If this is not stated, spreading averaging is meant. Each averaging technique more or less successfully converges to the desire coverage of the local mean signal value, which is unknown. Therefore it is still hard to make an exact classification of the fast fading reduction ability of each technique. The only way would be to optimize the Motif Model by each averaged coverage assuming that the optimized Motif Model predicts the exact signal local mean value. The resultant Motif 68
4. Motif Model
Model accuracy may then be the quantity of the fast fading reduction. Such an optimization campaign has not yet been carried out. In any event, a visual survey of the presented set of coverages can also give us a fruitful guideline. Our expectation is that the local mean value of the signal power will correspond to the power flow. Thus on the one hand, the variation within a single room cannot be sharp (Fig.4.54 b-d, g and h). On the other hand, an almost constant signal level over the whole room (Fig.4.54 f) is also strange, and it implies excessively strong averaging. Strong averaging also introduces the confusion or unreal parts into the signal coverage. The following best averaging technique of the fast fading reduction for Motif Model evaluation and optimization is therefore recommended as follows.
1st step non-spreading averaging with window dimension dw = 12 2nd step spreading averaging with window dimension dw = 4
The primary goal of the first step is fast fading reduction, and the goal of the second step is a sensible enlargement of signal coverage area. The resultant signal coverages obtained by this recommendation are shown in Fig.4.54 j), l) or n) for w = 1, 2 or 4, respectively. The first step of the recommendation is also supported by Fig.4.50, where the balance between the window dimension and its efficiency at dw = 12 can also be estimated (red mark in the chart).
a) Unaveraged data projection; w = 1
b) 1x over 2
c) 1x over 4
d) 1x over 6
e) 1x over 12
f) 1x over 24
69
4. Motif Model
g) 6x over 2
h) 12x over 2
i) 1x over 12 (non-Spread)
j) 1x over 12 (non-Spread) + 1x over 4
k) Unaveraged data projection; w = 2
l) 1x over 12 (non-Spread) + 1x over 4
m) Unaveraged data projection; w = 4
n) 1x over 12 (non-Spread)
Fig. 4.54. Examples of measurement data averaging by different modifications of the averaging window technique in the trial floor plan. The scales of all coverages are same (-30 -90 dBm), and are therefore not shown.
Conclusion
At least a modest level of fast fading reduction is essential for successful optimization of motif parameters and for objective evaluation of the accuracy of signal coverage prediction. Our recommendation efficiently reduces the fast fading signal fluctuation and it also introduces only limited confusion into the coverage. It enlarges the coverage area, which is beneficial for optimization and evaluation of the model. The presented results of the expected level of fast fading residual fluctuation in Fig.4.50 also provide a good estimation at 1 dB of one of the cumulative factors defined in Chap.4.5.2, the sum of which indicates the level of the highest possible Motif Model prediction accuracy.
70
4. Motif Model
4.5.2 Measuring Signal Coverage and Setting Motif Parameters

This chapter begins by describing a measurement system and measurement campaigns within five different buildings. Further general motif parameters derived from the measured coverages at 1.9 GHz are tabulated and the Motif Model prediction is compared with the measurements in every building to estimate the model flexibility. This is repeated for a 0.5 GHz measurement campaign within a building at the Cork Institute of Technology. Then a reduction to a set of two or one universal motif parameters is used in the trial floor plan to test an easier building description. Also, conversely, a higher number of motif parameter sets is used to obtain potential higher model prediction accuracy. Finally, the influence of grid spacing is tested in the trial floor plan to evaluate the independence of the Motif Model from it.
Measurement at 1.9 GHz and Motif Parameters Setting

The use of an indoor prediction tool usually involves the design and installation of WLAN or pico-cells systems indoors. Both systems operate mainly within an office building environment. The measurement campaign therefore concentrated on this environment.
Measurement System
A special measurement device set was designed for the measurement campaign. The measurement system consists of a measurement transmitter [61] and a measurement receiver [62] with automated data collection. Tab.4.2 presents the basic parameters, and Fig.4.55 shows the configuration of the measurement system. A 1900 MHz signal AM modulated by 1 kHz was transmitted by a patch antenna [63] (Fig.4.56) at a height of 1.8 2 m above floor level.
Frequency band Transmitter output power Modulation Receiver dynamic range Receiver resolution Measurement bandwidth 1850 - 1950 MHz, 1 MHz step 25 dBm +/- 0.5 dB 100 % On-Off keying (1 kHz) -47 -107 dBm 0.1 dB 0.3 MHz
Tab. 4.2. Transmitter and receiver parameters
71
4. Motif Model
GT
Transmitting Patch Antenna
GR - LRP
Receiving Wire Dipole Antenna Receiver
Generator
PG
AccT
Cable & Connectors Attenuation
AccR
Cable & Connectors Attenuation
Att
Additional Attenuator
PR
Palmtop
Fig. 4.55. Measurement system configuration where: PG..........Generated power AccT ........Attenuation of cable and connectors at transmitter site GT..........Transmitter antenna gain GR .........Receiver antenna gain LRP ........Polarization loss AccR........Attenuation of cable and connectors at receiver site Att ..........Additional attenuator PR ..........Received signal level Comments: (25 dBm) (5 dB) (5 dB) (2 dB) (3 dB) (2 dB) (dB) (dB)
The polarization loss of the power balance was caused by inability of the cross polarized power reception of the receiving wire antenna [64] assuming an equally divided incoming power into both polarizations. The additional attenuator pushed the upper bound of the dynamic range of the measured system to a higher signal level. Otherwise the upper bound of the system would be given by the receiver, which becomes overloaded in areas close to the transmitter. It may be useful to estimate the potential received signal level at a distance of 1 m from the transmitter in free space (4.64). PR = PG AccT + GT FSL0 + GR LRP AccR Att
(= -16 - Att dBm) (4.64)
Where Free Space Loss at a distance of 1 m from the transmitter (4.65) is:
4 FSL0 = 20 log
(at f = 1.9 GHz; FSLo = 38 dB)
(4.65)
72
4. Motif Model
Fig. 4.56. Transmitter antenna radiation pattern
Fig. 4.57. Receiver antenna radiation pattern The transmitting patch antenna radiated a linearly polarized wave directly (at 0 deg.) with the polarization plane slanted below an angle of 45 deg. Thus the radiated power was roughly divided into vertically and horizontally polarised halves (Fig.4.56). The wire dipole - the mobile receiver antenna - received only a vertical polarization. This antenna had an almost omni-directional radiation pattern, see Fig.4.57, at least in the horizontal section of the vertical polarization. There are also average radiation patterns in both sections of the transmitter radiation pattern (Fig.4.56), which serve for Motif Model prediction. Conversely, the mobile antenna is treated in the prediction as an isotropic probe, because the exact radiation pattern orientation of the mobile antenna during the measurements is in fact unknown. It is believed it has a negligible influence on the results. However, an ideal isotropic probe in both polarisations would naturally be preferable for the measurements. Let us take an example: a transmitted purely polarized wave in one room has a totally mishmash polarization even in the adjacent room. Therefore, it is believed that only one polarization reception has no significant influence on the measured coverage due to the polarization mishmash of a propagating electromagnetic wave through a typical indoor environment. Single polarization reception is taken into account in the receiver antenna
73
4. Motif Model
gain by subtracting a polarization loss of 3 dB (half power reception). However, the influence may increase for a line-of-sight case of the transmitter and receiver position.
Measurement Technique
The mobile antenna moved at a constant walking speed at a height of 1.5 m above the floor, while a portable palmtop connected to the receiver continuously collected a measured signal level with a set time step. The mobile antenna was fixed at the end of a wooden stick (Fig.4.58) to eliminate the body of the researcher. Moreover, the researcher attempted not to shield the dominant incoming signal with his body. An example of the final form of the measured data in a measurement tool window is shown in Fig.4.59.
Fig. 4.58. Example of signal coverage measurement
Fig. 4.59. Example of measured data
The position of each measured track was carefully noted in a floor plan (Fig.4.60) and finally projected into the floor plan grid (Fig.4.61). The projected coverage was then window averaged and enlarged (Fig.4.62). The data projection together with the fast fading reduction is described in Chap.4.5.1.
74
4. Motif Model
Fig. 4.60. Example of a floor plan with marked measured tracks
Fig. 4.61. Measured samples projected into the floor plan grid
Fig. 4.62. Enlarged measured coverage with reduced fast fading fluctuation This sample signal coverage measurement in the trial floor of a building at the Czech Technical University in Prague was not obstructed by denying access to any place and time restriction in contrast to other buildings. The measurement produced an enormous number (~140) of measured tracks with a total of more than 5 thousand signal samples. It is obvious even from a visual survey that such a density of signal samples in every room is redundant. Therefore the number of tracks was reduced in other measurement campaigns. This laborious complete coverage measurement is unique and gives us a better view of the power flow through the floor. Last but not least, it gave us valuable experience with a coverage measurement. Coverage measurement is a tricky task. People moving people in offices and mainly in corridors obstructing the main propagating paths had a powerful influence on the measured signal level. However, they are a natural feature of indoor environments. We
75
4. Motif Model
reduced their influence just by waiting for a non-blocked out main direct path at least in the case of the line-of-sight configuration. In other cases, the influence of people moving around was not reduced in any way.
Building Characterization
Indoor propagation measurement campaigns in the frequency range of 1.9 GHz were carried out in four office buildings. The first two were a building of the Czech Technical University building (CTU) in Prague and a building of the Brno University of Technology building (BUT). These are modern multi-storey buildings made with a concrete skeleton and large windows. There are usually waveguiding long corridors along the floors, surrounded by various rooms and halls. While a grid representation and motif parameters assigned to clusters of obstacles could be quite accurate for the CTU building and for the BUT building, where we have detailed construction plans, the characterization of other two buildings (the Koospol office building and the Diplomat Hotel building) were based only on roughly estimated building plans. Motif parameters were therefore assigned to the walls in the last two buildings following simple rules; the external walls were expected to be Heavy Walls and the internal partitions Light Walls. Exceptions were the bearing walls, which surrounded the lifts for example. The window positions and metal parts were not defined in these buildings at all. In effect, two sets of motif parameters were used for a coverage prediction in these last two buildings. For this reason, the first two buildings (CTU and BUT) were given a higher weighting than the others in the optimisation due to their accurate and reasonable description. There was another measurement campaign carried out at 0.5 GHz to evaluate the performance of the Motif Model at such a low frequency range. The campaign is described separately below.
Setting the Model Parameters

The key aspect of the Motif Model evaluation involves finding general motif parameters that will make a Motif Model prediction broadly valid and accurate over various buildings. The measured coverages at 1.9 GHz served for optimising the following four different sets of generally valid motif parameters and floor parameters tabulated in Tab.4.3. The motif or floor parameters definition is described in detail in Chap.4.2.2 or Chap.4.7, respectively.
Light Wall PrRT (-) PrA (-) PrDS (-) 0.90 0.64 0.27 Heavy Wall 0.98 0.50 0.60 Windows 0.24 0.37 0.24 Metal 0.98 0.60 0.07 Floor 0.93 0.25
Tab. 4.3. General motif parameters at 1.9 GHz 76
4. Motif Model
Comments on our general motif parameters: Surprisingly there is no significant difference between the motif parameter sets, except as regards those for the windows. This is a welcome feature offering a potential for reducing the number of motif parameter sets. It also proves the important influence of the surrounding vague obstacles, whose weighting in the motif parameters suppresses differences between the parameters of bare building walls. The exception is the window motif parameters, where on the one hand the values are lower than the other motif parameters sets, probably owing to the lower incidence of vague obstacles surrounding the windows. On the other hand, the effect of the window motif parameters on wave propagation is more significant than the effect of bare glass windows alone would have been. Another very exciting finding is the high value of the diffuse scattering probability (except metal). This is in essential contrast to ray tracing models, which usually take into account only specular reflection and direct transition of rays (PrDS = 0). The following prediction in four buildings compared with our measurements uses the motif parameters from Tab.4.3 to demonstrate the prediction robustness and universality of the Motif Model.
CTU Building Trial Floor Plan

Floor height....................3.3 m Transmitter height..........2.0 m Receiver height ..............1.5 m
Fig. 4.63. Photo of CTU building
Fig. 4.64. Input floor plan
77
4. Motif Model
Fig. 4.65. Measured signal coverage
Fig. 4.66. Predicted signal coverage (3 seconds, 100 000 beams)
Fig. 4.67. Difference between measurement and prediction
Optimistic Prediction
Pessimistic Prediction
Fig. 4.68. PDF of difference between measurement and prediction Note: The prediction in the trial floor compared in sections with our measurements is given in Chap.5 dealing with prediction by different models.
78
4. Motif Model
BUT Building
Floor height ................... 3.6 m Transmitter height ......... 2.0 m Receiver height.............. 1.5 m
Fig. 4.69. Photo of BUT building
79
4. Motif Model
Fig. 4.72. Predicted signal coverage (5 seconds)
Fig. 4.74. PDF of difference between measurement and prediction
Fig. 4.75. Comparison of measurement and prediction along the section 80
4. Motif Model
Koospol Building
Floor height ................... 2.8 m Transmitter height ......... 1.8 m Receiver height.............. 1.5 m
Fig. 4.76. Photo of Koospol building
81
4. Motif Model
Fig. 4.82. Comparison of measurement and prediction along the section
Diplomat Hotel Building

Floor height.................... 3.0 m Transmitter height.......... 1.9 m Receiver height .............. 1.5 m
Fig. 4.83. Photo of the Diplomat Hotel building
82
4. Motif Model
83
4. Motif Model
Fig. 4.89. Comparison of measurement and prediction along the section
Comments on evaluation results

Widely-used gauges of prediction accuracy are the mean error and standard deviation of the difference between the measurements and the prediction. Tab.4.4 summarises the prediction accuracy attained in each building. There are two pairs of columns in the table. The first pair recapitulates the accuracy gauges from Fig. 4.68, 4.74, 4.81 and 4.88 for each building for a prediction with the general motif parameters. The second pair of columns states the prediction accuracy (CTU and BUT building only) achieved for specific motif parameters, which were optimised for each building separately. Their mean error is thus zero.
Prediction with General Motif Parameters (dB) Mean Error CTU BUT Koospol Hotel Diplomat -0.2 2.0 1.3 0.7 Standard Deviation 3.3 4.0 4.8 4.9 Prediction with Optimised Motif Parameters (dB) Mean Error 0.0 0.0 Standard Deviation 3.1 3.2
Building
Tab. 4.4. Resultant prediction accuracy in four tested building The accuracy attained in the CTU and BUT building, if the general motif parameters are applied, is very high owing to the fact that the general parameters were tuned mainly by these two buildings. What is more surprising is the prediction accuracy in the Koospol and Diplomat buildings. In spite of a standard deviation worse by about 1 dB, the prediction is still extremely accurate and realistically reflects the power flow within the buildings. The results demonstrate the potential of the Motif Model, and especially its robustness and flexibility.
84
4. Motif Model
The motif parameters tuned for a specific building environment naturally provide higher prediction accuracy. However the standard deviation cannot in principle be lower than a certain value given by following cumulative factors:
the of residual fast fading in a measured coverage is estimated at (Chap.4.5.1)
~1 dB
standard deviation of signal coverage predictions due to the Monte-Carlo character of the Motif Model is roughly in the range of 2.0 0.2 dB or 0.2 0.02 dB for plain or averaged signal coverage (Chap.4.3.2) the considerable approximation of the environment by a reduced number of motifs and their parameters measurement inaccuracy
Taking these considerations into account, it can be stated that prediction accuracy about 3 dB for the investigated buildings is the absolute achievable minimum.
Measurement at 0.5 GHz and Setting the Motif Parameters

The Motif Model algorithm represents a propagating wave by a clump of rays, which are probabilistically dispersed by an environment. Such an approach brings fruitful features such as model robustness. Nevertheless, the model still has partly ray features, so it is limited at lower frequencies (Chap.4.1). In order to evaluate the model frequency limitation a similar measurement campaign to those at 1.9 GHz was also carried out at 0.5 GHz.
CIT Building
The measured building was the two-storey building complex of the Cork Institute of Technology (CIT). The environment of the building complex was sized up as an office environment but partly also as a light industrial environment with heavy metal machinery in the garages (Fig.4.90). The building structure consists of a reinforced concrete skeleton and long corridors surrounded by various offices, laboratories, lecture rooms and halls. There were mostly unmetallised large windows in the external walls and large metal cases along both sides of the corridors. There were rooms containing large quantities of equipment and devices. There were also garages with cars and heavy metal machinery in the measured area. The measurement system and measurement technique were similar to those used at 1.9 GHz, as described above. The measurement was carried out for two configurations. The first was located on the ground floor, with the transmitter in a room (Fig.4.90). The other measurement was moved to the first floor and the transmitter was placed at the
85
4. Motif Model
end of the corridor (Fig.4.91) to make a similar configuration to the measurement in the CTU building at 1.9 GHz and to investigate the waveguiding corridor effect at such low frequencies. Similar rules were used for obstacle classification as in the previous two buildings, where the external and internal walls were assigned to be heavy and light walls, respectively. The optimised general motif parameters at 0.5 GHz for this measurement campaign for the four different basic parameters sets are tabulated in Tab.4.5.
Light Wall PrRT (-) PrA (-) PrDS (-) 0.17 0.88 0.90 Heavy Wall 0.33 0.53 0.25 Windows 0.27 0.58 0.42 Metal 0.70 0.17 0.01 Floor 0.31 0.68
Tab. 4.5. Motif parameters for CIT building complex at 0.5 GHz The most noticeable difference in comparison with the parameters at 1.9 GHz (Tab.4.3) is generally the much lower probability of reflection of emitted power PrRT. This is caused mostly by the low frequency and often by the fanlights between rooms. The probability of absorption PrA is, on the contrary, higher than expected at 0.5 GHz. This is probably due to the denser obstacle environment than in the previous buildings. The denser environment also increased the probability of ray diffusion PrDS. However, high PrDS was expected due to the longer wavelength, for which the wall corners appear comparatively sharper than for the shorter wavelength, for example.
Floor height .............. 3.4 m Transmitter height .... 2.0 m Receiver height......... 1.5 m Fig. 4.90. Ground floor plan of the CIT building complex 86
4. Motif Model
Fig. 4.91. First floor plan
Fig. 4.92. Measured signal coverage on ground floor
Fig. 4.93. Predicted signal coverage on ground floor
Fig. 4.94. Difference between measurement and prediction 87
4. Motif Model
Fig. 4.95. PDF of difference between measurement and prediction on ground floor
Fig. 4.96. Measured signal level on first floor
Fig. 4.97. Predicted signal level on first floor
Fig. 4.98. Difference between measurement and prediction on first floor
88
4. Motif Model
Fig. 4.99. PDF of difference between measurement and prediction on first floor The measured coverages with reduced fast fading fluctuation are surprisingly very smooth throughout the floors, as one can be observed in (Fig.4.92) and (Fig.4.94). This is caused primarily by the longer wave, which makes an indoor environment relatively more homogeneously obstructed. Moreover the measured tracks were quite short (~1.5 m) and thus the applied window averaging (Chap.4.5.1) results in fact in an almost constant signal value along each whole track. A further increasing wavelength would bring similar results to those predicted with the Motif Model modified by ray expansion (Chap.4.8), where separation of antennas becomes a dominant factor of the signal level dependency. Application of the Motif Model ray expansion modification is intended to benefit such coverage prediction, but research into this it is still in progress. Nevertheless, the prediction accuracy obtained for the basic Motif Model algorithm at 0.5 GHz is very satisfactory and confirms the robustness and adaptability of the Motif Model even at such low frequencies. This section ends with an example (Fig.4.100) of signal coverage prediction throughout the CIT building complex at 0.5 GHz with prediction particulars.
89
4. Motif Model
Fig. 4.100. Signal coverage prediction throughout the CIT building complex (30 seconds; 2M beams)
Influence of the Number of Motif Parameter Sets on Prediction Accuracy

We also tested the achievable prediction accuracy on the trial floor, if the number of discerning parameters sets is reduced, or increased, compared to the four basic parameter sets. For each case the appropriate motif parameters were separately optimised in order to determine their values, which are tabulated in Tab.4.6, 4.7 and 4.8. An overview of the final prediction accuracies is given in Tab.4.9.
Universal PrRT (-) PrA (-) PrDS (-) 0.67 0.73 0.90 Floor 0.935 0.25
Tab. 4.6. Universal motif parameters at 1.9 GHz

Universal Wall PrRT (-) PrA (-) PrDS (-) 0.92 0.62 0.43 Window 0.10 0.37 0.47 Floor 0.935 0.25
Tab. 4.7. Motif parameters for two discrete sets 90
4. Motif Model
Motif Set PrRT (-) PrA (-) PrDS (-) Light Wall 0.94 0.64 0.30 Plane Wall Heavy Wall 0.97 0.40 0.60 Windo ws 0.24 0.37 0.24 Metal 0.87 0.55 0.20 Light Wall 0.55 0.42 0.20 Other Motifs Heavy Wall 0.98 0.61 0.20 Windows Floor Metal 0.82 0.34 0.64 0.93 0.25
Tab. 4.8. Motif parameters for seven discrete sets

Optimised Prediction with Basic 4 Sets 1 Universal Set 2 Sets 7 Sets Mean Error (dB) 0.0 0.0 0.0 0.0 Standard Deviation (dB) 3.1 4.3 3.5 3.0
Tab. 4.9. Dependency of prediction accuracy on the number of discrete motif parameter sets First, a description of the environment using one universal motif parameter set was tested. Such a greatly simplified environment description avoids uncertain motif assigning, and the achieved was only about 1.2 dB worse. The similarities between the general motif parameters offer the potential of reducing the motif parameter sets to two sets, namely Universal Wall and Window. The fact that the prediction remains accurate with this environment description supports this potentiality. Finally, an increased number of motif parameter sets were established in order to get a more precise prediction. However, the attained improvement is slight compared to the almost doubled number of necessary parameter sets. This result also supports the limited achievable minimum (see above) of prediction accuracy to 3 dB. The persisting Motif Model accuracy adds to the evidence that the environment description might be simplified to a single universal motif parameter set in the most extreme case. However, the test was based on the measured coverage on the trial floor plan only. To fully generalize the results, a greater variety of building environments and configurations must be tested.
Prediction with Universal Motif Parameters

The prediction on the trial floor plan composed only of the universal motifs (Fig.4.101) with universal motif parameters is so interesting that it is compared in detail with the measurements (Fig.4.102 - 4.105) and with the prediction for the four basic motif parameter sets (Fig.4.106, 4.107).
91
4. Motif Model
Fig. 4.101. Trial floor plan composed only of universal motifs
Fig. 4.103. Predicted signal coverage
92
4. Motif Model
Fig. 4.106. Difference between prediction with four basic sets of motif parameters and prediction with only one set of universal parameters
Fig. 4.107. PDF of previous difference from Fig.4.106 A prediction based on a simplified environment description can be still quite accurate, as the comparison of the two predictions proves. The problematic areas are those behind the window and the glassed-in gate at the end of the corridor. Likewise, the prediction of the metal chamber attenuation becomes inaccurate and the waveguiding effect of the corridor is reduced. Nevertheless, the achieved accuracy is still satisfactory and the advantages of such a reduced environment description outweigh the slight accuracy degradation.
93
4. Motif Model
Influence of Grid Spacing on Prediction

The other factor that remains to be examined is the influence of grid spacing on the prediction results. Until now we have mostly used a grid spacing of 2 at 1.9 GHz in this thesis, which seems to be reasonable (Chap.4.2.1). Grid spacing in the range of 0.2 1 m offers adequately detailed space description, on the one hand, and fast prediction on the other. Moreover, the obstacles specifically localized in a building plan are mostly walls. Describing a plane wall by a row of grid elements with wall motif parameters is very easy, and therefore the preferable way of localizing a wall. Thus a grid spacing of 0.2 1 m is also practical. If such a wall description is applied, the grid spacing modification also in fact changes the wall thickness. So let us examine the following predictions, which were made on the trial floor plan divided into a grid with spacing of 4 (~0.63 m) (Fig.4.108 a), 1 (~0.16 m) (Fig.4.108 b) and also 2 (~0.31 m) (Fig.4.65 - 4.68). The motif parameters were those from Tab.4.3.
Measurement
Prediction
Difference
PDF of differences a) Grid spacing = 4 (w = 4) b) Grid spacing = 1 (w = 1) Fig. 4.108. Influence of grid spacing on prediction 94
4. Motif Model
It is obvious that all the prediction results are very similar, regardless of the grid spacing. The exceptions are the areas where detailed localization of an obstacle leads to a more site-specific and therefore usually a more accurate prediction. This certainly demonstrates the independency of the Motif Model algorithm from alterations to the grid spacing, at least in the given range.
Conclusion
This chapter presents a measurement campaign that investigated the flexibility of the model. It also provides a valuable description of our unique measurement system and technique. The measurements were carried out inside four different buildings at 1.9 GHz and inside one building at 0.5 GHz. The measurements at the two frequencies served for optimizing the generally valid motif parameters, see Tab.4.3 and Tab.4.5. There was no significant difference between the motif parameter sets, except as regards those for the windows. This is a welcome feature offering the potential to reduce the number of motif parameter sets. Another very exciting finding is the high value of the diffuse scattering probability (except for metal). This is in essential contrast to ray tracing models, which usually take into account only specular reflection and direct transition of rays. The accuracy attained in the CTU and BUT buildings, when applying the general motif parameters, is very high owing to the fact that the general parameters were tuned mainly by these two buildings. What is more surprising is the prediction accuracy in the Koospol and Diplomat Hotel buildings. Although the standard deviation is worse by about 1 dB, the prediction is still extremely accurate and realistically reflects the power flow inside the buildings. The results demonstrate the potential of the Motif Model, and especially its robustness and flexibility. The prediction accuracy obtained for the basic Motif Model algorithm at 0.5 GHz is less good, but still very satisfactory and confirms the robustness and adaptability of the Motif Model even at such low frequencies We also tested the achievable prediction accuracy on the trial floor, if the number of discerning parameter sets is greater, or smaller, than the four basic parameter sets. The persisting Motif Model accuracy adds to the evidence that the environment description might be simplified to a single universal motif parameter set in the most extreme case. Increased complexity of the site-specific description brought only a slight improvement in prediction accuracy, supporting the defined achievable minimum. The other examined factor was the influence of grid spacing on the accuracy of the prediction. The results certainly demonstrated the independence of the Motif Model algorithm from alterations to the grid spacing, at least in the given range.
95
4. Motif Model
4.5.3 Impulse Response Measurement and Prediction

The Motif Model algorithm has been designed to deal with the real character of electromagnetic wave propagation in a generally indescribable environment, e.g., as an indoor environment. It is captured by the introduction of a novel method, which enables an enormous number of rays (108 1012), to be launched and traced individually and very rapidly. Such a quantity of rays is highly beneficial, because they do not simulate only the dominant power flows (common ray-tracing), but as the rays are probabilistically treated, they simulate the entire spherical wave front propagation. For demonstration purposes, Fig.4.109 shows only the density of ray-fronts in specific time delays after the launch of the rays. The launched rays form a wave front, which is gradually recast by wave fronts interacting with the environment to produce plentiful other diffused, reflected or diffracted sub-wave fronts until all waves escape from the building, or their power diminishes below a predefined threshold. The wave front propagation is equivalent to a transmitted Dirac impulse propagation.
a) after 60 ns
b) after 170 ns
Fig. 4.109. Launched wave front, i.e., Dirac impulse propagation and interaction simulated by the Motif Model This approach simulates the real full signal propagation. Thus all other parameters of the communication channel, apart from the signal level, can be predicted with a high accuracy. However, it is difficult to evaluate the prediction results, due to the insufficient dynamic range and impreciseness of commercial measurement systems, which cannot attain the performance of the Motif Model. Nevertheless, five samples of impulse responses were measured and compared with the Motif Model prediction on the trial floor plan.
Measurement System
Impulse response (IR) measurement can be provided directly in a time domain by a channel sounder [65, 66, 67] or indirectly in a frequency domain transformed to a time domain by the Fourier transformation. The crucial problem of IR measurement is the achieved dynamic range. This is limited by the available maximum transmitting power at the upper bound and the noise measurement threshold at the lower bound. The frequency domain measurement actively integrates the transmitting power over all the relayed frequency range.
96
4. Motif Model
In order to meet these requirements a wide-band measurement is necessary. For this reason, our IR measurement was based on the Agilent E8358A (300 kHz 9 GHz) vector analyzer (Fig.4.110) with an indirect time domain measurement. We used the identical double ridge horn antennas operating in a frequency range of 1 9 GHz, in which their return loss is under -10 dB (Fig.4.111). The attained measurement dynamic was around 60 dB and the time resolution was about 0.11 ns.
Fig. 4.110. Agilent E8358A vector analyzer
Fig. 4.111. Wide band double ridge horn antenna and its return loss
Fig. 4.112. Antenna radiation pattern We intended measuring the omni-directional IRs of the mobile communication channel only. The receiving antenna directivity (Fig.4.112) was overcome by the following technique. The antenna located in the positions of IR measurement was turned around in 97
4. Motif Model
a constant step of 45 deg. The resultant eight IRs were then compound by a maximum principle to obtain the desired omni-directional IR (Fig.4.113)
Fig. 4.113. Example of omni-directional IR composing The IR included all signal components of the measured frequency range. Therefore the frequency range was reduced over the Fourier transformation to 1.5 2.3 GHz, in which the Motif Model parameters are valid. The IRs are still influenced by impulse response of both antennas. Their influence can be reduced by deconvolution to obtain the desired pure mobile channel IRs which are also predicted by the Motif Model.
Comparison of Measurements and Prediction

The IR measurements were carried out on the trial floor plan with the antennas location including LOS and NLOS cases (Fig.4.114). The distance between the transmitter and the receiver was limited by the required dynamic range. A comparison of the IR measurements and the prediction is shown at five points in Fig.4.115. Thanks to the high speed of the Motif Model algorithm the prediction in every element of the floor plan took only a few seconds. The time needed depends on the desired level of accuracy, due to the Monte Carlo character of the algorithm.
Fig. 4.114. Trial floor plan with indicated measurement locations in brackets
98
4. Motif Model
Fig. 4.115. Comparison of measured and predicted impulse responses in five locations
99
4. Motif Model
The measured IRs are limited by 180 ns in time and by a level of about -90 dBm of background noise, unlike the prediction. We can find quite good agreement between the predicted and the measured power delay profile above the level of the noise (-90 dBm). The differences are caused by the fact that the real indescribable indoor scenario is only approximately described by unified clusters of obstacles. However, the character of the responses is preserved well. The IRs of a real mobile system are in fact convolutions of all IRs of all parts of the transmission chain. Therefore in order to obtain such IRs the predicted IRs should be additionally convoluted with an impulse shaped according to the system IR and adjusted to the system frequency range. This was not done here, because only pure channel IRs were desired.
Directional Impulse Response Prediction

To fully characterise the outgoing and incoming signal paths on the transmitter and receiver side, respectively, directional impulse responses are essential [68, 69]. Examples of directional response prediction by the Motif Model are shown in Fig.4.116 and 4.117. The first response is located in the corridor (Fig.4.114, point [3]) and illustrates the different signal paths forming a fascinating complex impulse response for LOS case. The dominant direct paths form a leading edge and the multi-reflected paths at the end of the corridor are also marked as secondary edges.
Fig. 4.116. Directional impulse response prediction at point [3] in the corridor The second predicted directional impulse response is located within a room at point [2], and represents an NLOS case, where the multi-path phenomenon fully signifies.
100
4. Motif Model
Fig. 4.117. Directional impulse response prediction at point [2] within a room However, the prediction of these directional impulse responses is more time-consuming. It took about 15 minutes to predict them in all elements of the floor plan (more than 12 000 elements). The time complexity depends on the predefine resolution of the time delay and the angle of arrival, which were 0.5 ns and 1 deg., respectively. It also depends on the definable desired level of accuracy, due to the Monte Carlo character of the algorithm. However, the prediction of omni-directional high quality IRs is possible only by the Motif Model, and, disputably, by ray-tracing models. Our prediction of high resolution directional IRs is absolutely unique, and no any other models can make such a prediction within an acceptable time.
An Example of IR Prediction by a Ray-Tracing Model

Let us compare the predicted IRs to an up-to-date example of omni-directional IR prediction by a ray-tracing model (Fig.4.118) [35]. The prediction was made in the building of the University of Stuttgart (Fig.5.5) for a transmitter and receiver separated by about 5 meters, and a NLOS case. Such a short distance was chosen for the prediction in order to be able to trace at least a few rays (24). It is obvious that in the prediction (Fig.4.118) are totally missing weak rays, due to a limited number of ray interactions in the ray-tracing algorithm.
101
4. Motif Model
Fig. 4.118. Example of IR prediction by a ray-tracing model taken from [35]
Site-Specific RMS Delay Spread Prediction

When high quality impulse response prediction in all floor plan elements is available, then other channel parameters such as, the signal root mean square (RMS) delay spread can be predicted. A site-specific RMS delay spread prediction by the Motif Model is therefore compared with a measurement. Continuous delay spread measurements require a channel sounder, which was not available at the authors department. For that reason, the RMS delay spread measurement referred to [59] was used to evaluate the Motif Model prediction.
a) Measurement taken from [59]
b) Motif Model prediction (3M beams; 120s)
Fig. 4.119. RMS delay spread Despite the rough sketch of the floor plan and the general motif parameters used for the Motif Model prediction, the results correspond very well to the measurement. However, the prediction took longer (120 s) and more beams were launched (3 Millions) than for the prediction of signal coverage only (Chap.5). The higher number of beams is beneficial for forming an appropriate power delay profile in every floor plan pixel (100 000 pixels).
102
4. Motif Model
Conclusion
In this section, the impulse response prediction of the Motif Model was verified by measurement. In addition, the directional impulse response prediction was shown and the site-specific RMS delay-spread prediction was compared with our measurement. We found quite good agreement between the predicted and the measured power delay profile and even the Motif Model prediction of the site-specific RMS delay spread corresponded very well with the measurement. The complete and accurate characterisation of the wide-band communication channel by directional impulse responses becomes crucial along with the design and implementation of the MIMO systems. In comparison with typical ray tracing, the Motif Model can give a much higher dynamic range of prediction (~100 dB) from the level of dominant signal paths to the level of the weakest paths, which behave almost as background noise. While ray-tracing traces only the dominant paths, such a dynamic prediction range is a natural feature for the Motif Model algorithm.
103
4. Motif Model
4.6
Further Applications of the Motif Model
The Motif Model algorithm very accurately simulates real electromagnetic wave propagation. This is further demonstrated by the high quality directional impulse response prediction (Chap.4.5.3). The high Motif Model performance allows any further wireless communication channel parameters to be predicted, and enables a variety of different applications. The chapter does not aim to provide a detailed description of our results, but it briefly introduces further results achieved by the Motif Model in order to demonstrate the flexibility and serviceability of the model for a variety of different applications, which would mostly by hard to predict by any other model. Descriptions of each result will be made mostly by references, because a detailed description would exceed the object of this work.
Prediction with a Leaky Coaxial Cable Antenna

Leaky coaxial cables have been used for many years as wireless feeders in coal mines and vehicular tunnels [70]. More recently, the leaky coax has been employed as a basestation antenna for indoor wireless systems [71]. Here we present an application of the Motif Model application for radio coverage prediction from a leaky coaxial cable antenna (Fig.4.120).
Fig. 4.120. Coverage prediction from a leaky coaxial cable antenna A similar prediction by a ray-tracing model can be found for purposes of comparison in [72].
Instantaneous Power Prediction

Motif Model predictions of the local mean signal level are based only on ray density (Chap.4.4.1), in order to avoid undesired the fast fading signal fluctuation caused by the interference of incoming waves with different phase shifts. However, the channel parameters that characterize very fast fading in some cases also need to be predicted.
104
4. Motif Model
Hence phase incorporation to each ray in Motif Model prediction is essential (Chap.4.4.2). The instantaneous signal fluctuation along sections A and B, whose position is marked in (Fig.5.1), is shown in Fig.4.121 with a step resolution of 0.2 (~3 cm). The complete trial floor coverage by the instantaneous signal level in a grid with spacing of 1 is shown in Fig.4.122. The exact location of the fades may not be exactly valid, but the character of the local signal fluctuation should be well expressed.
a) along section A
b) along section B
Fig. 4.121. Instantaneous power prediction along sections A and B marked in Fig.5.1 with resolution 0.2 (~3 cm)
Fig. 4.122. Instantaneous power prediction in a grid 1x1
Site-Specific Rices k Factor Prediction

The algorithm also enables prediction of the instantaneous power level at points in a detailed grid with a distance of 0.02 x 0.02. These detailed grids are distributed in every element of the main grid of the floor plan. The site-specific Rices k factor can therefore be assessed in each element of the main grid separately (Fig.4.123). The adjacent main grid neighbour elements also enter into the assessment of the k factor in the median element.
105
4. Motif Model
Fig. 4.123. Site-specific prediction of Rices k factor The probability density of the instantaneous power fluctuation normed by the local mean power is shown in Fig.4.124. The distributions in different areas of the floor together with the fitting overall Rices distributions with appropriate k factors are summarised in Tab.4.10. The assessed values of k factor vary roughly in same range as it was observed in [73] e.g.
a) Whole Floor
b) Corridor
c) Corridor Centre
d) Room 621
Fig. 4.124. Fast fading PDF in different locations and the appropriate Rice distribution functions
Location Rices k Whole Floor 1.8 Corridor 4.2 Corridor Centre 0.9 Room 621 0.8
Tab. 4.10. Specific Rices k factor prediction Characterisation of fast fading by Rices k factor corresponds to the extent of the multi-path character of propagation over the dominant path, and can therefore give a first guess of a MIMO system asset that exploits the multi-path propagation.
106
4. Motif Model
Predicting the Site-Specific Signal Mean Delay and RMS Delay Spread
The signal mean delay and RMS delay spread are statistical parameters characterising the channel impulse response, also referred to as the delay profile [5]. Chap.4.5.3 deals with impulse response measurement and Motif Model prediction along with site-specific RMS delay spread prediction compared with a measurement. Therefore only the site-specific prediction results of these channel parameters are stated here, together with their probability density functions in the trial floor (Fig.4.125).
a) Signal mean delay
b) RMS delay spread of signal
Fig. 4.125. Site-specific prediction of signal mean delay and RMS delay spread prediction and their PDF on the trial floor The mean delay corresponds to the centre of gravity of the impulse response. The RMS delay spread is the second moment, or spread, of the impulse response. This takes into account the relative powers of the incoming paths as well as their delays, making it a better indicator of system performance than the other parameters. Therefore the maximum of both exists, where variously delayed waves with similar power contribute to the signal level, e.g., as behind the metal chamber (Fig.4.125). The RMS delay spread is roughly indirectly proportional to the relative signal level, as one can observe in Fig.4.125. This is utilized by empirical models for estimating the RMS delay spread [74].
Site-Specific MIMO Channel Capacity Prediction

The most challenging Motif Model application is site-specific wide-band MIMO channel capacity prediction. This prediction is based on the directional impulse responses predicted in Chap.4.5.3, which serve for MIMO channel matrix estimation in each floor element. The capacity itself is then calculated by the Raleigh algorithm [75].
107
4. Motif Model
On the basis of a prediction of such impulse responses, the MIMO channel matrix can be estimated e.g. and a site-specific wide-band capacity of MIMO channel can be calculated, Raleigh [75]. The transmitter and receiver are the same antenna array consisting of four antenna elements forming a horizontal square of dimension . Unfortunately, no wide-band MIMO capacity calculation has been implemented before completion of this thesis, so a narrow-band calculation is used instead. The signal to noise ration (SNR) is assumed to be constant over the whole floor. However, for a fixed background noise level the SNR would increase with the mean signal level. Then the MIMO channel capacity would also increase with the signal level. These site-specific predictions point first of all to the influence of the scenario configuration, while the SNR effect is undesired in single capacity coverage. The probability and cumulative distribution functions of the predicted capacities over the whole floor plan are also shown in Fig.4.126. Site-Specific Prediction
Probability Density Function
Cumulative Distribution Function
a) SNR = 10 dB
b) SNR = 20 dB
Fig. 4.126. Site-specific prediction of MIMO channel capacity Finally, Fig.4.127 shows the same prediction to that in Fig.4.126a, but with a different antenna configuration for purposes of comparison. Both antennas were formed by a row of four elements. The horizontal array of the transmitter was oriented parallel to the corridor, while the horizontal array of the receiver was oriented perpendicularly. 108
4. Motif Model
Fig. 4.127. Site-specific prediction of MIMO channel capacity for a different arrangement of antenna elements We can trace down the MIMO channel capacity in correspondence with the Rices k factor behaviour over the floor. Along the corridor, where a direct signal path is dominant, the MIMO capacity improvement is low. This is contrast to the rest of floor, where rich multi-path propagation exists, which is why the MIMO capacity is much higher. The achieved capacity corresponds with the theoretical expectation [76]. The site-specific prediction has not been published so far (apart from the Motif Model prediction). A deeper discussion about the results should be based on a theoretical understanding of MIMO systems, which goes beyond the scope of this thesis. Hence only the interesting predictions are introduced here.
Prediction of Launched Dirac Impulse and Continuous Wave Propagation in Time

Despite the ray basis of the Motif Model algorithm the enormous number of launched rays (108 1012) means that not only single rays are traced to find the dominant power flows as in a ray-tracing algorithm, but complete waves are traced in fact interacting with the environment to produce various consequential sub-waves until their power diminishes below predefined value. This is also supported by the unlimited number of interactions between rays and obstacles. In order to clearly illustrate the predicted wave interaction with obstacles, only the ray-head density is depicted on the floor in certain time delays after rays launch (Fig.4.128). This corresponds with a wave-front and launched Dirac impulse propagation.
4 ns
140 ns
109
4. Motif Model
30 ns
165 ns
55 ns
240 ns Fig. 4.128. Floor excitation by a launched Dirac impulse
If not only the ray-heads are shown, but complete rays with appropriately assigned phases and in electrical intensity units, we can obtain an interesting animation of real wave propagation in time through a building. Shots from the animation are shown in Fig.4.129. To provide a lucid illustration, a 3 metre long wavelength was chosen, while the motif parameters are still those optimised for 1.9 GHz. The animations of the impulse and wave propagation are based on impulse responses predicted in each floor element, which are subsequently processed to obtain these animations.
after 40 ns
after 300 ns Fig. 4.129. Electrical intensity field distribution of a radiated continuous wave at 100 MHz The complete movies of the animation of the launched Dirac impulse and electrical intensity field distribution of a radiated continuous wave can be found on webpages dedicated to the Motif model development [122]. 110
4. Motif Model
4.7
Probabilistic 2D Model Based on Ray Density Distribution in Periodic Building Structures

Supplement of 2D Motif Model for 3D Prediction
Motivation
The aim of this section is to introduce and describe a new probabilistic model based on a ray density distribution in periodic building structures. The model especially aims at a signal level prediction within the vertical plane of buildings with uniformly spaced parallel identical floors. The prime reason for developing the probabilistic vertical 2D model is to supplement the 2D Motif Model, which is employed mostly for signal level loss prediction in the horizontal plane. Although the Motif Model can also be used for the predictions in the vertical plain, a straightforward probabilistic description of ray distribution across storeys is beneficial for a typical multi-storey building structure. The structure of typical buildings is usually regular in the vertical plane to the extent that employing a deterministic approach is not proportionally beneficial. A combination of both models (the Motif Model for the horizontal plane and the probabilistic model for the vertical plane) also has the feature of simultaneous, which is useful. This feature became many times important in our software implementation. Let us now state reasons for developing a new probabilistic model instead of using existing models. The useable existing models are the One-Slope model (OSM) and the Multi-Wall model (MWM) (Chap.3.3). OSM prediction is based only on the signal decay factor. In the case of propagation along a floor, OSM predicts the same signal decay regardless of floor parameters such as height, reflection and attenuation. Moreover, only a narrow part of the transmitter radiation pattern participates in the prediction along the floor, and the rest of it is simply ignored. The same is true for MWM, which in the case of an empty floor is reduced to free space propagation prediction, thus to OSM with the power decay constant equal to two. Thus existing models do not offer desirable features and it is essential to develop a new probabilistic model. The new model should appropriately take into account the parameters of the floor and the whole transmitter radiation pattern. First, a probabilistic description of ray distribution will be suggested, and then the new probabilistic model for signal loss prediction will be introduced.
Probabilistic Ray Distribution across Floors

Basically, a floor behaves as a row of clusters of obstacles (motifs) in the vertical profile, like a plane wall in the horizontal floor plan. Therefore, with respect to the motif description principles (Chap.4.2.2) the floor reflection coefficient is independent 111
4. Motif Model
of the angle of ray arrival and is described by the motif parameter PrRT. Similarly, the floor transit attenuation is described by the motif parameter PrA. The third motif parameter PrDS is assumed to be zero in this model description. The problem geometry is illustrated in Fig.4.130a. A transmitter on the left side launches rays at angle . The ray stream interactions with the periodic building structure excluding diffusion result in it being split into many sub-streams with values that depend on the floor reflection and attenuation. The consequent ray distribution across the floors (Fig.4.130 b) can be approximated by the Gaussian (Normal) distribution function (4.66), regardless of the floor attenuation, which does not affect the resultant ray distribution shape across the floors.
1 N ( x, , ) = e 2 2
( x )2
2 2
(4.66) (4.67)
with:
mean value
= 1 PrRT 2 = ( I (1 PrRT ) )
and squared standard deviation where:
(1 PrRT )
RRT
(4.68)
I ....... Number of ray stream interactions with floors (-)
As the number of ray stream interactions with floors approaches infinity, the Gaussian approximation becomes exact. The approximation is even very good for small I and PrRT > 0.5, as Fig.4.131 and Fig.4.132 demonstrate. The approximation deteriorates for small I and PrRT (Fig.4.133), where the rays are still partly cumulated in one dominant ray stream. However, even for such a case the suggested approximation is satisfactory.
1- PrRT
PrRT Transmitter
Ray Sub-Streams
a) b) Fig. 4.130. Ray streams (a) and an example of possible ray probability distribution as a function of PrRT (b) 112
4. Motif Model
Fig. 4.131. Ray probability distribution simulation for PrRT = 0.9 and PrA = 0 with its three sections
Fig. 4.132. Ray probability distribution simulation for PrRT = 0.5 and PrA = 0 with its three sections
Fig. 4.133. Ray probability distribution simulation for PrRT = 0.1 and PrA = 0 with its three sections 113
4. Motif Model
The previous ray probability distributions are simulated by an exact mathematical formulation in mathematic software. The same Monte Carlo simulations produced by the 2D Motif Model follow below (Fig.4.134 and 4.135). The angle of the launched rays is 45 deg, the floor height is 3 m and the number of interactions on the horizontal axis is replaced by the real distance from the transmitter.
a) PrR = 0.9 b) PrR = 0.5 c) PrR = 0.1 Fig. 4.134. Ray probability distribution simulated by the Motif Model with PrA = 0
a) Sections of chart the from Fig.4.134b b) Section of charts from Fig.4.134a-c at different distances at a distance of 15 m Fig. 4.135. Sections of ray probability distribution simulated by the Motif Model Another example of simulated ray density for nonzero floor attenuation PrA is shown in Fig.4.136. The sections along the floor where the transmitter was located are shown in Fig.4.137 for different PrA.
a) Simulation by mathematical software b) Simulation by Motif Model Fig. 4.136. Ray probability distribution for PrRT = 0.5 and PrA = 0.3 114
4. Motif Model
Fig. 4.137. Sections of ray probability distribution for PrRT = 0.5 and variable PrA indicated in the chart These simulations prove the reasonability of the suggested probabilistic description of ray density distribution in a cross section of floors and along the floors.
Probabilistic model
Our probabilistic model is based on the ray probability distribution described above. The model is especially suitable for predicting the signal level in vertical cuts of a multi-storey building. The model takes into account the probability of ray absorption PrA on a floor and the probability of ray reflection PrRT from the floor. In addition, the model incorporates the full transmitter radiation pattern and the floor height. The problem geometry is depicted in Fig.4.138.
Adjacent Parallel Floors
y x
Adjacent Parallel Floors
Fig. 4.138. Problem geometry To further simplify the notation, the distance from the transmitter is normed to the floor height. r (4.69) r0 = d where:
r0 ...... Normed distance from transmitter (-) r ....... Distance from transmitter (m) d....... Floor height (m)
Then the predicted ray probability at distance r0 corresponds to the ray probability at real distance r. In order to obtain the ray probability at distance r, the prediction should be processed for a normed distance r0. 115
4. Motif Model
The total number of ray interactions with the floors depends on the angle of the launched ray and the normed distance. This leads to
I (r0 , ) = r0 tan( ) .
(4.70)
The contribution of a ray launched at angle to the ray probability at distance r0 on floor FN considering (4.66 4.68) is
p( r0 , ) =
(1 PrA )
1
Pr N z , PrTR + FN , ( I (r0 , ) PrTR ) TR PrRT 0.5

0.5
dz
I (r0 , ) > 0.5 FN > 0
elsewhere
(4.71)
where the following substitution was applied

PrTR = (1 PrRT ) ...... Probability of ray transition of emitted ray (-)
(4.72)
and
FN ............................ Number of floors where the ray probability is calculated (-) for a floor with transmitter is FN = 0
A number of ray interactions with floors smaller than 0.5 means that the rays reach a distance of r0/2 from the transmitter without any interaction, and for this reason the ray stream does not split. The rays then fully contribute to the signal level on the floor with the transmitter (FN = 0). The overall ray probability along a floor is obtained by integrating (4.73) over all ray launching angles that contribute by (4.71) to the ray probability on the floor at distance r0. Moreover, the contributions are weighted by the transmitting antenna directivity radiation pattern.
P (r0 ) =
p(r , ) D( )d
0 2
(4.73)
D( )d
Examples of ray probability prediction by the suggested model for different configurations are given in Fig.4.139 4.141.
116
4. Motif Model
a) for an isotropic antenna
b) for a directional antenna DT = 6 dBi
Fig. 4.139. Ray probability along an empty floor with transmitter as a function of floor reflection
a) PrRT = 1
a) PrRT = 0.5
Fig. 4.140. Ray probability along an empty floor with transmitter as a function of floor absorption PrA for an isotropic antenna
Fig. 4.141. Ray probability as a function of the antenna directivity and floor reflection at normed distance r0 = 50 m from a transmitter We can trace from the ray probability behaviour that the limit of the lowest ray probability corresponds to the 2D free space propagation (red dashed line). The limit of the highest ray probability level also corresponds to the lossless propagation
117
4. Motif Model
(1D propagation). The highest probability level of -3 dB came from the rays splitting into equal halves propagating in the opposite directions from the transmitter along the floor. The antenna power radiation pattern used for the simulations was formed by the n-times powered cosine in the vertical plane (4.74), and was independent of the horizontal plane angle.
F ( ) = cos n ( )
(4.74)
The resultant antenna pattern therefore has two equal lobes in opposite directions and the resultant antenna directivity D() is
D( ) =
4F ( )
2 0 0
2 F ( )
0
(4.75)
F ( ) sin( )d d F ( ) sin( )d
Fig. 4.142. Maximum directivity of the antenna pattern formed by the n-times powered cosine
Conversion of Predicted Ray Probability to Signal Level

The predicted ray probability distribution along the floor assumes an uniform ray distribution in the cross sections of each floor. However the receiving antenna does not receive all rays passing through the vertical cross section of the floor, but only a fraction of them proportional to the ratio between the floor height and the effective antenna length for the vertical case. Due to the same ray base of our probabilistic model and the Motif Model, the final conversion of predicted ray probability to a reasonable received signal level should be based on the 2D technique described in Chap.4.4.1 by equations (4.34 4.42) assuming that the floor height is equal to the element dimension of the 2D Motif Model grid.
118
4. Motif Model
Conclusion
The model setting and evaluation is given through a Motif Model evaluation (Chap.4.5), because all presented 2D Motif Model predictions in the horizontal plane are supported by the predictions of the introduced probabilistic model in the vertical plane to get reasonable signal level propagation prediction in real 3D space. Our probabilistic model is satisfactorily valid for a large range of different buildings. Even for low-storied or single-storied buildings, the model can provide good performance, if the parameters are appropriately chosen. The parameters can be estimated by optimizing from the measurements, as int he case of the other motif parameters (Chap.4.2.3). However, the floor reflection and absorption can be common for a large range of kinds of buildings, and the signal level predicted by our probabilistic model can be still satisfactorily formed by the specific floor height and the real radiation pattern. This is one of the main advantages of our probabilistic model in comparison to OSM and MWM. The probabilistic character of our model, together with the combination of probabilistic and specific input parameters of the floor description, classifies the model somewhere in the group of semi-empirical models. The advantage of our new probabilistic model is that; in spite of the simple input parameters, the model provides a reasonably correct prediction. Thus it avoids the need for a complete site-specific environment description, unlike, for example, MWM. The advantages of our model are especially prominent along a corridor or for propagation within floors. These advantages and also the level of accuracy are reduced as the number of floors between the transmitter and the receiver increases. Nevertheless, our new probabilistic model is used mostly together with the 2D Motif Model for signal level prediction on the floor where both terminals are located. A disadvantage of the model is its inherent assumption that the building structure is periodic. This probabilistic model was implemented in the prediction software package along with the 2D Motif Model, demonstrating the complementarity of the two models.
119
4. Motif Model
4.8
Ray Expansion
This section will introduce a very interesting novel method known as ray expansion. The method is in fact a modification of the common Motif Model algorithm. It differs just in the way of the ray treatment. The results demonstrate the potentiality of significantly shifting the bottom frequency limit of any ray-based prediction technique. Although research on ray expansion is still in progress and it is not the main aim of the thesis, the first results are presented here. The author appeals to the reader to take an interest in this new idea even though the assertions have not been yet evaluated by measurements.
Motivation
Indoor electromagnetic wave prediction in a frequency range from tens to hundreds of megahertz remains an outstanding problem, despite many applications working in this frequency range. Let us mention some of the most important applications (TETRA, fire fighting rescue systems, Radio and TV Broadcasting, etc.) The behaviour of the basic Motif Model is very natural. It provides a precise characterization of various scenarios. In spite of the imposing robustness of the Motif Model, there is still an inherit limit arising from the ray-based algorithm. If a ray is used to approximate a propagating wave, then the obstacles included in a scenario should at least be bigger than the wavelength. The interaction of the ray with large obstacles (bigger than 10) can be dealt with by splitting a ray into transmitting and reflecting rays. However, as the dimension of the obstacles diminishes below, let say, 10 times those of the wavelength, it is hard to assume that the ray is still only reflected or transmitted, and a full characterization of the scattered field around must be taken into account. This is a major complication for any ray tracing algorithm. In the case of the Motif Model such small obstacles are grouped into a cluster of obstacles, as described above (Chap.4.2). This could be done, because usually there is no need for signal level prediction within such clusters. What problems arise, if the wavelength is longer than a room dimension or even longer than whole building? Ray approximation becomes altogether faulty, and a full electromagnetic numerical calculation needs to be processed instead. In practically terms, such a calculation is not feasible, mostly due to the impossibility of electromagnetic characterisation of all obstacles inside buildings. Hence, there are similar shortcomings as, for example, in ray-tracing prediction, so it is unusable. The suggested ray expansion should partly overcome this problem of ray approximation, and should enable us to push the warrantable ray approximation to a lower ratio of obstacle dimension to wavelength. It is believed in the case of indoor propagation that the frequency limit of the ray model modified by the ray expansion method could be as low as 10 MHz. 120
4. Motif Model
Example of Signal Propagation that is Indescribable by the Common Ray Based Model
Let us now look at the propagation problem from the point of view of attenuation behaviour in various environments and with various obstacle arrangements. In the case of common ray model applications for indoor propagation, the main consideration is accurate characterization of propagation along the boundaries (walls) and wave guiding propagation. The problem is illustrated by the next example of signal attenuation in the scenario according to (Fig.4.143 a). There is a transmitter with a highly directional radiation pattern on the left side, two parallel common walls forming a corridor, and perfectly absorbing walls, which make the problem more lucid. This is a critical scenario, where ordinary ray approximation is altogether faulty, if the wavelength becomes comparable with or smaller than the corridor width. Nonetheless the expanded ray can be employed to offer a reasonable prediction of the signal level along the whole section. The section of signal level coverage goes along the axis of the scenario (Fig.4.143 a). Parts A and C of the section correspond to the free space propagation, while the part B corresponds to propagation within a corridor. In the two cases the character of the signal attenuation is different. The first and last parts can be precisely characterized simply by ray thinning, as, for example, in the basic Motif Model algorithm, but the second part is not properly describable for common ray models (Fig.4.143 b) due to the lack of interactions between the rays passing through the straight corridor and the walls. The suggested solution (Fig.4.143 c-f) can therefore be understood as a forced increase in ray interactions with surrounding obstacles.
a) Scenario
b) Common ray (d = 2.2 m)
c) Expanded ray (d = 1.6 m)
d) Expanded ray (d = 2.2 m)
e) Expanded ray (d = 2.8 m)
f) Expanded ray (d = 4 m)
121
4. Motif Model
Fig. 4.143. Example of signal level prediction at 100 MHz for different corridor cross-sections and radiation power of 1 W A directional antenna was used in the simulations, and the rays around the corridor mouth on the right end of the section A were absorbed by the absorbing walls. The final power flow is then nicely formed as it passes through the corridor and spreads around.
Ray Expansion Description

A ray launched from a transmitter does not in itself have a natural counterpart. It is therefore hard to find a natural equivalent of ray expansion. Nevertheless, it is obvious that the expanded ray is much closer to reality than simple an infinitely thin ray, as shown in Fig.4.144.
Fig. 4.144. Example of a common ray and an expanded ray propagating between transmitter and receiver at 100 MHz The closest natural counterpart may be emitted power approximated by a Gaussian beam [77]. However, a Gaussian beam describes the emitting power from a highly directional antenna, and the power of a Gaussian beam is attenuated in free space as a spherical wave. Such behaviour already simulates ray thinning in the basic Motif Model. Nevertheless, a beam cross section with a Gaussian shape is suggested. The Gaussian cross section of the expanded ray will be formed by individual rays, which are randomly deviated from the beam axis in the cross section of beam propagation. In order to retain the Motif Model principles of fully statistical ray behaviour, the Gaussian distribution of rays around the beam axis is achieved by the following method. The distance of each individual ray from the beam axis is deviated by a random number with Gaussian distribution with standard deviation 0. Let this be
122
4. Motif Model
referred to as the deviation strength. Such a deviation is carry out in every step of ray propagation within a grid. It was found that the standard deviation of the actual ray distance from the beam axis after n deviations is given by (4.76)
2 = n 02
where: ........... Standard deviation of ray distance from beam axis (-) n ........... Number of deviations performed (-) 0.......... Deviation strength (-)
(4.76)
Deviation Strength
The behaviour of the deviation strength may have the same character as the other Motif Model parameters. For this reason, it could be treated as a further unknown parameter, which can be optimized from a measurement in the same way as the other parameters are. If the wave propagation is very complex, as it is inside buildings, there is no way to find a single value of the ray deviation strength even at a single frequency. Nevertheless, I believe that an average value, which will be satisfactorily valid over many building structures can be estimated. The suggested estimation, which seems to agree with the indoor behaviour of the expected ray over the frequency range of 10 MHz 0.5 GHz, is
0 = 0.06 per wavelength of propagating ray.
(4.77)
This means that the ray is deviated to have a standard deviation of 1 from the beam axis at a distance around 280 from the transmitter. If the scenario is described by a grid with spacing w, the deviation strength per grid element or per ray step, eventually, is:
n =
0
w
(4.78)
Let the reasonability of this assessment be judged by means of the following examples of expanded ray simulations at different frequencies on a trial floor (Fig.4.145). Only rays between the transmitter on the left side and the receiver on the right side are depicted.
a) 10 MHz
b) 20 MHz
c) 50 MHz
d) 100 MHz 123
4. Motif Model
e) 150 MHz
f) 200 MHz
g) 2 GHz Fig. 4.145. Power flow between transmitter and receiver at different frequencies; Radiation power of 1 W; Isotropic antennas; 10M Rays) At high frequencies, when the ray approximation is correct, the power flow occurs mostly in the corridor. As the wavelength reaches the corridor width (f ~ 136 MHz) the power flow begins to spread widely into the adjacent rooms. As the wavelength increases further, the signal level in the adjacent rooms increases steadily. If the complete signal level over the whole floor plan is predicted, the following results are obtained (Fig.4.146).
a) 10 MHz
b) 20 MHz
c) 50 MHz
d) 100 MHz
e) 150 MHz
f) 200 MHz
124
4. Motif Model
g) 2 GHz Fig. 4.146. Indoor propagation at different frequencies for an isotropic transmitter; Radiation power of 1 W; 1M Rays These signal distributions (Fig.4.146) clearly demonstrate the suitability of our suggested ray expansion. As the wavelength increases, the individual obstacles lose their direct influence on the signal distribution, and their influence is more bound simply with the density of obstacle occurrence in the floor plan. Therefore, the wave guiding effect of the corridor diminishes for low frequencies. The resultant signal coverage at low frequencies (10 20 MHz) is mainly dependent on the distance from the transmitter. The signal level can then be predicted as in a macro-homogeneous environment by the One-Slope model (Chap.3.3), where the signal level decreases monotonically with just the distance from the transmitter, forming concentric circles with the transmitter in the middle. Some other comparisons of signal propagation predicted by the basic ray algorithm and the expanded ray algorithm are shown in Fig.4.147 and Fig.4.148. As can again be seen, the expanded ray application disregards the direct influence of small separate obstacles, in contrast to the basic ray algorithm (Fig.4.147). Moreover, an expanded ray within a corridor can truly characterize the real signal attenuation much more realistically than the basic ray algorithm (Fig.4.148).
a) Prediction by an expanded ray
b) Prediction by an ordinary ray
Fig. 4.147. Example of power flow around a small obstacle at 20 MHz ( = 15 m); Antenna directivity of 6 dB; Radiation power of 1 W; Cross section of obstacle of 2.6 m
a) Prediction by an expanded ray
b) Prediction by an ordinary ray
Fig. 4.148. Example of power flow through and around a corridor at 100 MHz; Antenna directivity of 6 dB; Radiation power of 1 W; Corridor cross section of 2.9 m
125
4. Motif Model
Practical Problem in Application of Ray Expansion

Finally, there is one practical aspect of ray expansion application that needs to be mentioned.
Fig. 4.149. Determining the angle of arrival of a ray The treatment of individual rays forming a single expanded ray is same as in the basic Motif Model, except for the direction from which the ray reaches the cluster of obstacles. The individual ray is launched from an emitter (Fig.4.149) in a direction corresponding to the direction of the axis of the expanded ray. As the individual ray is deviated during its propagation the actual direction of the ray also changes. Therefore its direction of arrival at the moment when the ray strikes is calculated from the actual position of the struck cluster of obstacles and the emitter (direction a) in Fig.4.149. In cases similar to Fig.4.149, for example, this calculation could give a false direction (the ray did not strike the Plane-Wall from the bottom side). In such cases the direction of ray arrival is therefore calculated from the three elements last filled by the ray, as illustrated in Fig.4.149 (direction b).
Conclusion
This section introduced a novel method known as ray expansion, and demonstrated its capability to push the warrantable ray approximation to a lower ratio of obstacle dimension to wavelength, thus significantly shifting the bottom frequency limit of any ray-based prediction technique. It is believed in the case of indoor propagation that the frequency limit of the ray model modified by the ray expansion method could be as low as 10 MHz. This method is in fact a modification of the common Motif Model algorithm. It differs only in the way in which the ray is treated. The described expanded ray behaviour sufficiently supports the reasonability of our suggested method. However, there no evaluating measurement campaign has not been carried out at really low frequencies (10 500 MHz), and this remains for future research.
126
5. Comparison of Models
5.
Comparison of Models
Different approaches to indoor electromagnetic wave propagation prediction need to be implemented in a single software package for an appropriate comparison. For this purpose, the empirical One-Slope model (OSM) and the semi-empirical Multi-Wall model (MWM) were implemented in our software package together with the Motif Model. Other fundamental models with which the Motif Model should be compared are the ray-tracing model [78, 17] and the Dominant Path model (DPM) [17]. We believe that the ray-tracing and DPM implementations [17, 35] are among the best, and they have been presented in detail in many papers. The Stuttgart implementations were therefore chosen for evaluating the Motif Model. While the Dominant Path model is an interesting model from the category of advanced semi-empirical models, ray-tracing is a representative of conventional deterministic approaches based on ray-optical propagation and the Uniform Theory of Diffraction or the Geometrical Theory of Diffraction [21]. Their implementation is beyond the scope of this work, so only their predicted coverages are compared with the Motif Model prediction and the measurements at 1.8 GHz in the same building of the University of Stuttgart [79]. The OSM and MWM parameters used in the chapter were set according the recommendation of COST 231 [16] (Tab.5.1).
Power Decay n () OSM MWM 3.0 3.4 6.9 =aLW =aHW Light Wall aLW (dB) Heavy Wall aHW (dB) Windows aW (dB) Metal aM (dB)
Tab. 5.1. OSM and MWM parameters by COST 231 [16]
Comparison of Prediction Coverages in CTU Building

First, the prediction models implemented in the software package will be compared with the measurements at 1.9 GHz and with each another in the trial floor plan. Fig.5.1 shows the location of the transmitter and four sections. The model predictions and their differences from the measurements are given in Fig.5.2 and Fig.5.3, respectively. A comparison of the models in four sections of coverage is shown in Fig.5.4.
Fig. 5.1. Floor plan of CTU building 127
a) Measurement
b) One-Slope model
c) Multi-Wall model
d) Motif Model
Fig. 5.2. Measured and predicted coverage by different models
a) Difference between measurement and OSM prediction
b) Difference between measurement and MWM prediction
c) Difference between measurement and Motif Model prediction Fig. 5.3. Differences between measurement and predictions
128
a) Section A
b) Section B
c) Section C
d) Section D
Fig. 5.4. Comparison of signal coverage predictions along four sections OSM prediction (Fig.5.2b) depends on the power decay factor (Chap.3.2) and the transmitter radiation pattern. It does not take into account the position of obstacles, that influence of which is respected by the power decay constant only. Therefore waveguiding and shadowing effects are not predictable at all (Fig.5.2b). For an isotropic transmitter radiation pattern, the predicted coverage will consist of concentric circles with a constant signal level and with the transmitter in the centre. The signal loss will then depend only on the distance between the transmitter and the receiver. OSM prediction is exact only for a truly homogeneous environment, which is never found inside a building, however. The MWM prediction (Fig.5.2c) takes into account the position and attenuation of the walls. Therefore the prediction is mostly more accurate than an OSM prediction. However, shadowing elements like heavy walls and metal shields are often overestimated, because their transition loss leads to an extensively small signal level behind them, and any diffraction or diffusion is not taken into account in MWM. This is notable especially behind the metal chamber and in the entrance hall (Section D Fig.5.3d). On the other hand, MWM usually predicts line-of-sight propagation accurately. The trial floor presents difficulties for widely used empirical and semiempirical prediction models, because it contains a long corridor with a strong waveguiding effect and a shielding metal chamber. Moreover, on the opposite site of the corridor from the transmitter is the entrance hall, where, surprisingly, a high signal level was measured. 129
5. Comparison of Models Unlike the OSM and MWM prediction, the Motif Model successfully characterises the influence of all obstacles and tough features of the floor plan to predict the signal level with very high accuracy over the whole floor. It is also supported by the resultant standard deviations and mean errors of each prediction from the measurement (Fig.5.3).
Comparison of Prediction Coverages in the University of Stuttgart Building

The second model comparison involves a comparison of predictions with measurements carried out in a University of Stuttgart building (Fig.5.5). First, we will show the measurements at 1.8 GHz and predictions by various models referred to [17], and then the equivalent Motif Model and MWM predictions for purpose of comparison (Fig.5.75.9). However, the MWM is not discussed here, because its basic effects were shown in the previous comparison. The Motif Model prediction was based on the general motif parameters from Tab.4.3 (Chap.4.5.2) and a roughly sketched floor plan (Fig.5.6) according to [79]. The transmitting power was 10 dBm and the antennas had isotropic radiation patterns. The detailed measurement description is given in [35]. For such a configuration, 100 dBV/m (used in the referred coverages) corresponds to -57 dBm (used in the Motif Model prediction).
Fig. 5.5. Stuttgart University building
Fig. 5.6. A detail from the floor plan of a Stuttgart University building 130
a) Measurement and prediction by different models, taken from [17]
b) Equivalent Motif Model and Multi-Wall model prediction Fig. 5.7. Comparison of the measured and predicted signal level from transmitter A along section A The ray-tracing model is very pessimistic in the distance range of 8 20 m and > 40 m, because it takes into account in the prediction [17] only up to four interactions (including a maximum of two diffractions), which is not enough to described the complete scenario. DPM gives a more accurate prediction, but the shielding effect of the heavy walls (8 15 m) is not sufficiently taken into account, and in the distance range of 17 30 m the signal level is, on the contrary, underestimated. Even the signal level predicted by the DPM is suspiciously smooth. In such a complex inhomogeneous scenario, where the dominant incoming power flow is oriented perpendicular to the corridor and section A, a more variant local mean signal value is characteristic. The measurements also indicate this.
a) Measurement and prediction by different models, taken from [17]
131
b) Equivalent Motif Model and Multi-Wall model prediction Fig. 5.8. Comparison of the measured and predicted signal level from transmitter B along section B This section (Fig.5.8) emphasizes the capability of the models to simulate the waveguiding effect of a corridor and the power flow around a corner. It is notable that, although all models render the waveguiding effect well, all of them more or less underestimated the signal level in the middle of the corridor end (around the 7th metre of the section distance). In this configuration, the ray-tracing model and the Motif Model outperform DPM which demonstrates its poor ability to predict the power flow around a corner. It is also evident from Fig.5.9b where the dominant power flow is strongly spread inside the room behind the end of the corridor, while insufficient power flow turns around the corner into the perpendicular corridor.
a) Measurement taken from [17]
b) Dominant Path Model prediction taken from [17]
c) Ray-Tracing prediction taken from [17]
d) Motif Model prediction (300t beams, 7s)
Fig. 5.9. Measured and predicted signal coverage by three different models for transmitter position B 132
5. Comparison of Models The measurements and predictions in Fig.5.9 refer to [17] except the Motif Model prediction, which is depicted in an equivalent colour scale in dBm to make the comparison easier. As is obvious from a visual survey, all coverages are quite similar. However, the DPM behaviour (Fig.5.9b) in the room behind the right end of the corridor is strange, even from a power balance point of view. The power impinging on the whole right side of the room seems to be higher than the power penetrating through the left side of the room. The ray-tracing prediction (Fig.5.9c) behaves more realistically within this room, but no measurements have been made there. In addition, ray-tracing models the power flow around the corner well. However, in comparison with the measurement, ray-tracing is pessimistic in areas where incoming rays have to interact with many obstacles. This is mainly due to the omitting the diffuse scattering of rays, and due to the limited number of ray interactions allowed in the model in order to reduce the prediction time. The question is, whether a grater number of allowed interactions will make the prediction more accurate [80]. The Motif Model prediction (Fig.5.9d) behaviour is more similar to the ray-tracing prediction than to the others, but due to taking into account a diffuse scattering and the unlimited number of ray interactions the Motif Model is accurate everywhere. Moreover it appears that ray-tracing and DPM do not appropriately simulate the power flow outside the building. Especially in this building, this can cause a significant error in rooms on the left side of the perpendicular corridor, which are illuminated from outside according to the Motif Model prediction (Fig.5.9d).
133
Time Complexity of Models Prediction

The time complexity remains an interesting factor in the model performance, but its importance decreases as personal computer power increases. However, for mobile system access point design when an automatic optimization process is used and many access point positions have to be evaluated, the prediction speed of the model remains crucial. The most popular models for this purpose are therefore OSM and MWM, which can be very fast. The computation time for OSM or MWM in an ordinary building is usually no longer than a few seconds. The exact prediction time of the compared models is tabulated in (Tab.5.2) for the University of Stuttgart building.
Grid (m) Model preprocessing ??? 12 prediction 0.31 0.31 1 1 0.31 Time (s) preprocessing ??? 7200 400 PC prediction <1 3 < 10 110 67 7 P.IV 1.9 GHz P.IV 1.9 GHz P.II 350 MHz P.II 350 MHz P.IV 1.9 GHz [17] [35] Ref.
OSM MWM DPM Ray-Tracing Motif Model
Tab. 5.2. Time complexity of models DPM advanced semi-empirical predictions based on a neural network are fast, but the room oriented database [17] also needs to be pre-processed. Unfortunately, a more specific computation time has not been published. In comparison to DPM classical ray-tracing models are very slow due to the tracing of possible rays between the transmitter and the receiver by very time consuming ray reflection and obstacle visibility evaluation (many hours). The need to accelerate the model can be resolved by obstacle visibility pre-processing to avoid recurrent operations during the prediction. This pre-processing can be done only once for a building, and then predictions based on the pre-processed obstacle database are much faster for any transmitter and receiver position (Tab.5.2). Despite its much finer prediction grid (10 times more pixels), the Motif Model prediction is still much faster than the ray-tracing model. Moreover, the Motif Model does not need any obstacle visibility pre-processing.
134
Frequency Bounds of Models

The model frequency limitation is primarily a matter of the principles underlying their propagation wave characterisation. Empirical and semi-empirical models scarcely concern themselves with real wave propagation. This makes them almost frequency independent, although MWM indoor prediction becomes similar to OSM prediction at low frequencies. The frequency limitation of the advanced semi-empirical DPM is hard to estimate, because no references for it have been found. However, its semi-empirical principles may by frequency independent. On the other hand, there are deterministic and semi-deterministic models, which are mostly based on the ray approximation of a real wave. However, the approximation is only valid for those obstacles interacting with the wave that are at least ten times bigger in every dimension than the wavelength. This is fulfilled only for the main building structure of walls and floors, which then serve for a site-specific building description. All other obstacles must be incorporated in different ways. For example, usable indoor ray-tracing models modify the UTD based ray description by approximative equations, which can be tuned by a measurement [45, 35]. The models then become more semideterministic. Another possible method is to combine the general statistical and ray approach, which brings a significant benefit (the Motif Model). The dimensions of the building structure elements then give the lower frequency bounds of the ray model validity, estimated at about 0.5 GHz (~0.6 m long wavelength). This means that ray based models are unable to predict indoor signal propagation or penetration through a building for many important applications, e.g., as TETRA, TV, Radio etc. For this reason, modification of the Motif Model by ray expansion was suggested, pushing the lower frequency bounds as low as 10 MHz (Chap.4.8).
Frequency Bounds (GHz) Lower none none none* ~ 0.5 ~ 0.5 Upper none none none* none none
Model OSM MWM DPM Ray-Tracing Motif Model Basic
with Ray ~ 0.01 none Expansion * No reference to frequency restriction was found
Tab. 5.3. Models frequency bounds
135
Conclusion
This chapter presents a detailed comparison of the most widespread models for indoor propagation prediction. The performance of these models refers to particular implementations. The Motif Model was compared with measurements and with the simple empirical OSM and the semi-empirical MWM, and also with the advanced semiempirical DPM and deterministic prediction of the ray-tracing model. The results demonstrated that OSM and MWM performed poorly from the point view of prediction accuracy, as described in Chap.3. The predictions of the other models are more accurate, but DPM demonstrated strange behaviour in some cases. The ray-tracing prediction was quite accurate, but showed imprecision in places where diffuse scattered rays or rays with a high number of interactions became dominant, because these rays are usually neglected in ray-tracing models. The Motif Model prediction behaviour is more similar to the ray-tracing prediction than to the others, but, because it takes diffuse scattering and an unlimited number of ray interactions into account, the Motif Model is more flexible and more accurate. The models were also compared from the points of view of time-complexity and frequency limitation. The OSM and MWM models, which utilise straightforward calculations, are naturally much faster than the other models. However, Motif Models and DPM prediction are also fast, especially in comparison with the very time-consuming ray-tracing model. Moreover, the Motif Model does not need any obstacle visibility pre-processing. The empirical and semi-empirical models scarcely concern themselves with real wave propagation, which makes them frequency independent. The ray basis of the ray-tracing models leads to their frequency limitation. For this reason, modification of the Motif Model by ray expansion was suggested, pushing the lower frequency bounds (Chap.4.8) Finally, there are examples of a signal coverage prediction over the whole Stuttgart University building at 1.8 GHz (Fig.5.10) with prediction particulars.
Fig. 5.10. Other Motif Model predictions in a building of the University of Stuttgart (500t beams, ~20s) 136
6. Summary and Future Research Directions
6.
Summary and Future Research Directions
The main goals of this dissertation were set out in Chap.2, as follow:
1. Summarize current approaches to indoor electromagnetic wave propagation modelling and their capability to satisfy the demands of mobile systems of new generations, in order to explain the motivation for developing the novel model. Evolve a new indoor electromagnetic wave propagation model, which outperforms current models, and which achieves features really suitable for specific and complex indoor environments. Implement the new approach in order to verify its performance. Carry out evaluation measurement campaigns at the frequencies of current mobile systems, and compare the new approach with current models.
2.
3. 4.
Results
1. The behaviour of electromagnetic wave propagation was described (Chap.3.1) and current approaches to indoor propagation prediction were summarized (Chap.3.2). Their advantages and, above all, disadvantages (Chap.3.3) led to the delineation the desired features of a new approach, taking into account the increasing demands of new generation of mobile wireless systems (Chap.4.1). A very novel approach to the description of an indoor environment (Chap.4.2), ray launching and propagation (Chap.4.3), and the channel parameter prediction (Chap.4.4) known as the Motif Model was evolved and described in detail. The Motif Model algorithm very accurately simulates the real electromagnetic waves propagation (Chap.4.5.2). This is demonstrated by the most challenging prediction of high quality directional impulse responses (Chap.4.5.3) and by the site-specific prediction of the MIMO channel capacity. The high generality and flexibility of the Motif Model enables a variety of different applications, and allows any further wireless communication channel parameters to be predicted (Chap.4.6). 3. The model algorithm was tested, and was implemented into the software prediction package for the design of indoor mobile systems. In addition, webpages dedicated to the principles and results of the Motif Model have been administered since the beginning of the development of the Motif Model [122].
2.
137
6. Summary and Future Research Directions A reduced version of our software prediction package is used in the lab lessons at our department to demonstrate the principles of indoor propagation and wireless system design. An adjustment of the software package for the purposes of laboratory demonstrations was supported by a grant from the Ministry of Education of the Czech Republic. 4. To meet the needs of model setting, measurement campaigns were carried out (Chap.4.5.2) to provide a sufficient amount of various signal samples for setting the motif parameters and consequently for evaluating the Motif Model at a frequency of 1.9 GHz to approximately cover the GSM1800 frequency bands and partly also the UMTS frequency bands. A second campaign at a frequency of 0.5 GHz focused on algorithm validation at low frequencies. In addition, channel impulse response measurements were performed to validate the ability of the Motif Model to predict wideband channel parameters (Chap.4.5.3). The final comparison of the new approach with other state-of-the-art models underlined the high performance of the Motif Model (Chap.5).
Future Research Directions

Although much time was spent on developing the Motif Model further possible Motif Model improvements and applications remain to be investigated. The most challenging future research directions are specified below:
The current implementation of the Motif Model algorithm is broadly valid for many applications. However, it was reduced to 2D space and supplemented by developed probabilistic model (Chap.4.7). This seems to be adequate for evaluation the model, and we can believe that full 3D implementation would lead to even better model performance. Describing the environment by a fixed set of basic motifs (Chap.4.2) with the same universal motif parameters can greatly simplify the description of the building that the user must provide. This description is usually the biggest problem in signal coverage prediction, as far as the user is concerned. Moreover, just a black-andwhite bitmap blueprint of the floor plan could then totally replace the typical unhelpful handmade building description supplied by a user. Such a blueprint should be properly clarified in order to avoid confusion. Research into this interesting way of applying the Motif Model remains for further investigation. Our evaluation measurements and obtained motif parameters at 1.9 GHz roughly cover the frequency bands of the most frequently-used mobile systems (GSM1800 and UMTS). However, the motif parameters are generally dependent on the frequency, so further research should be directed at this dependency, in order to
138
6. Summary and Future Research Directions cover, in particular, the frequency of WLAN systems at 2.4 GHz (IEEE 802.11b(g)) [4] and 5 GHz. (IEEE 802.a) [4].
Uniform grid spacing should ensure a reasonable building characterization, but should have a limited number of elements in order to accelerate the prediction (Chap.4.2) The two main factors, which counteract each other, can be partly overcome by applying a nonuniform grid (Fig.4.13). An environment with rapid changes in statistical behaviour can be divided by a fine grid to deal with the first factor, whereas the large open spaces are tiled by larger grid elements to deal with the second factor. This kind of nonuniform grid is a common technique in many numerical methods [40, 49, 50, and 51]. It has not yet been completely implemented into the Motif Model software tool and tested. Future research should investigate this question. The promising accuracy of the impulse response predictions (Chap.4.5.3) supports the adequacy of the current implementation (Chap.4.4.2). However, more sophisticated ray delay assessment based on optimisation by measured impulse responses and phase shift assessment remains for further research. Our description of expanded ray behaviour sufficiently supports the reasonability of our suggested method (Chap.4.8). However, no evaluating measurement campaign has yet been carried out at really low frequencies (10 500 MHz), and this remains for future research. The generality of the Motif Model offers the possibility of applying the Motif Model for any further purpose, when appropriate motifs and their parameters are specified. Further possible Motif Model applications might include:
outdoor prediction for microcell mobile systems within a dense urban environment, where application of the 2D Motif Model should be sufficient full 3D Motif Model prediction in rural and hilly environments the principles of the Motif Model can in fact be applied for predicting complex wave propagation within complicated structures or simply within waveguides. (however, the ray features of the model must still be kept in mind)
139
7. Conclusion
7.
Conclusion
The research conducted in this thesis deals with the prediction of electromagnetic wave propagation inside buildings, in order to predict any parameter of mobile communication channels. The main motivation for this work was to develop and introduce a novel model which will outperform current models, with features really suitable for indoor propagation prediction and convenient for predicting any mobile channel parameters. Such demands have resulted from the emerging third and fourth generation of wireless mobile systems that incorporate adaptive antennas or MIMO systems. A brief outline was presented of electromagnetic wave propagation inside a building, and of various different approaches to indoor propagation modelling. It was shown that the available methods cannot reasonably satisfy the requirements of the third and fourth generation of wireless mobile systems. In order to overcome the shortcomings of the current models, a new approach with cutting-edge features called the Motif Model has been developed. Summarising the results of the research done on the novel model was the main goal of this thesis. The proposed Motif Model algorithm was tested from the beginning, and implemented into a software prediction package. Many evaluation measurements were made, including wide-band impulse responses, and comparisons with other models. According the results presented in this thesis, our Motif Model can provide an accurate narrowband and wideband characterization of the radio channel. This fact, in conjunction with the computation-time reduction achievable with the new algorithm, makes the model an efficient option for the planning and design of new wireless systems. Everything indicates that the performance of the new Motif Model is more than satisfactory. It comprehensively outperforms other models in many aspects. While raytracing models fail to predict wide-band parameters with a large dynamic range, the Motif Model algorithm has this as an inherent feature. The Motif Model therefore aspires to be a new standard for characterizing the mobile channel of communication systems of the new generations based on MIMO systems, due to its more realistic behaviour. The ideas behind the Motif Model algorithm seem to be very general. If a full electromagnetic description of the motifs were used, the Motif Model could also be applied in various other electromagnetic areas.
140
8. References
8.
[1] [2]
References
M. Cooper, R. Roy, SDMA Technology Overview and Development Status, ArrayComm-ID-010, Inc., Mountain View, California, www.arraycomm.com G. J. Foschini, Layered Space-Time Architecture for Wireless Communication in a Fading Environment When Using Multiple Antennas, Bell Labs Technical Journal, vol. 1, no. 2, Autumn 1996, pp 41-59 G. J. Foschini, M. J. Gans, On Limits of Wireless communication in a fading Environment when using Multiple Antennas, Wireless Personal Communication, vol. 6, no. 3, pp. 311-335, March 1998 IEEE Standards Association, http://standards.ieee.org T. S. Rappaport, Wireless Communication: Principles and Practice, New Jersey, Prentice Hall Inc., 2002, ISBN: 0-13-0422320 H. L. Bertoni, Radio Propagation for Modern Wireless Systems, New Jersey, Prentice Hall, Inc., 2000, ISBN: 0-13-0263737 S. R. Sounders, Antenna and propagation for Wireless communication Systems, Chichester, John Wiley & Sons Ltd, 1999, ISBN: 0-471-98609-7 B. E. A. Saleh, M. C. Teich, Fundamentals of Photonics, John Wiley & Sons, Inc., ISBN: 0-471-83965-5, September 1991 W. Honcharenko, H. L. Bertoni, J. Dailing, Mechanisms governing propagation between different floors in buildings, IEEE Trans. Antennas Propagation, vol. 41, pp. 787-790, June 1993
[3]
[4] [5] [6] [7] [8] [9]
[10] D. K. Cheng, Field and Wave Electromagnetics, Addison-Wesley Publishing Corporation, ISBN: 0201128195, January 1989 [11] W. H. Hayt, Engineering Electromagnetics, ISBN: 0070273952, January 1981 [12] Ericsson - Planning and Optimization Solutions, www.ericsson.com/tems [13] AVE Communications, www.awe-communications.com [14] StrixSystems, www.strixsystems.com [15] J. M. Keenan, A. J. Motley, Radio Coverage in Buildings, BTSJ, vol. 8, pp. 1924, January 1990 [16] COST231 Final Report, Digital Mobile Radio: COST231 View on the Evolution towards 3rd Generation Systems, European Commission / COST Telecommunications, Brussels, 1998 [17] G. Wolfle, F. M. Landstorfer, Prediction of the field strength inside buildings with empirical, neural and ray-optical prediction models, COST 259 TD(98)008 141
8. References [18] J-M. Georce, E. Jullo, K. Runser, An adaptive Multi-Resolution Algorithm for 2D Simulations of Indoor Propagation, International Conference on Antennas and Propagation - ICAP 2003, Exeter UK, April 2003 [19] G. C. Angelo, I. Neto, L. M. Correia, Health and Penetration Issues in Buildings with GSM Base-Station Antennas in Top, 48th IEEE Conference on Vehicular Technology - VTC1998, pp. 450-454, Ottawa, May 1998 [20] J. B. Keller, Geometry Theory of Diffraction, J. Opt. Soc. Amer., vol. 52, pp. 116-130, 1962 [21] V. A. Borovikov, B. Y. Kinber, Geometrical Theory of Diffraction, IEE, London ISBN: 0852968302, 1994 [22] R. G. Kouyoumjian, P. H. Pathak, A uniform geometrical theory of diffraction for an edge in a perfectly conducting surface, in Proceedings of the IEEE, vol. 62, pp. 1448-1461, November 1974 [23] S. Ramo, Fields and waves in communication electronics, John Wiley, New York 1997 [24] R. J. Luebbers, "Finite conductivity uniform GTD versus knife-edge diffraction in prediction of propagation path loss, IEEE Trans. Antennas Propagation, vol. AP32, pp. 70-76, Jan. 1984. [25] J. F. Rouviere, N. Douchin, P. F. Combes, Diffraction by Lossy Dielectric Wedges Using Both Heuristic UTD Formulations and FDTD, IEEE Transaction on Antennas and Propagation, vol. 47, November 1999, pp.1702-1708 [26] A. J. Booysen, C. W. I. Pistorius, Electromagnetic Scattering by a two-dimensional wedge composed of conductor and lossless dielectric, IEEE Transaction on Antennas and Propagation, vol. 40, April 1992, pp.383-390. [27] P. L. E. Uslenghi, Geometrical Optics exact solution for two classes of dielectric wedge structures, IEEE Transaction on Antennas and Propagation, vol. 44, no. 2, January 1996, p. 129 [28] P. Bernardi, R. Cicchetti, O. Tesla, A Three-Dimensional UTD Heuristic Diffraction Coefficient for Complex Penetrable Wedges, IEEE Transaction on Antennas and Propagation, vol. 50, no. 2, February 2002, pp.217-224 [29] S. Selleri, G. Pelosi, Diffraction by a Transparent Wedge with Resistive Faces with the Parabolic Equation Method, IEEE Transaction on Antennas and Propagation, vol. 50, no. 7, July 2002, pp.1030-1031 [30] G. Wlfle, R. Hoppe, F. M. Landstorfer, R. R. Collmann: Vergleich deterministischer und empirischer Ausbreitungsmodelle fr die Planung von Mikrozellen, ITG-Workshop Wellenausbreitung bei Funksystemen, Weling, May 1998
142
8. References [31] T. Huschka, Ray tracing models for indoor environments and their computational complexity, IEEE 5th International Symposium on Personal, Indoor and Mobile Radio Communications, pp. 486490, 1994 [32] H. Zhu, Verification of a two-dimensional/three-dimensional hybrid ray-tracing method for spatiotemporal channel modelling, Radio Science, vol. 36, no 1, pp. 53-66, January 2001 [33] C. Carciofi, A. Cortina, C. Passerini, S. Salvietti, Fast Field Prediction Techniques for Indoor Communication Systems, 2nd EPMCC, pp. 37-42, Bonn, November 1997 [34] Z. Ji, B. H. Li, H. X. Wang, H. Y. Chen, T. K. Sarkar, Efficient Ray-Tracing Methods for Propagation Prediction for Indoor Wireless Communications, IEEE Antennas and Propagation Magazine, vol. 43, no. 2, April 2001 [35] R. Hoppe, G. Wolfle, P. Wertz, F. M. Landstorfer, Advanced ray-optical wave propagation modelling for indoor environment including wideband properties, European Wireless 2002, Florence, Italy, February 2002 [36] T. Frach, W. Fischer, B. Rembold, Fast Parallel Ray-Tracing Simulator, Millennium Conference on Antennas & Propagation in Davos, April 2000, vol. 2P7 (CD-ROM), ISBN 92-9092-776-3 [37] P. Danielle, V. Degli-Esposti, F. Falciasecca, M. Frullone, G.J. Riva, Evaluation of the reliability of a ray tracing microcellular field prediction model, Progress in Electromagnetic Research Symposium 1996, p.513, Innsbruck, July 1996 [38] M. A. Arbib, Handbook of Brain Theory and Neural Networks, ISBN: 0262011972, MIT Press, November 2002 [39] B. Chopard, P. O. Luthi, J. F. Wagen, A lattice boltzmann method for wave propagation in urban microcells, IEEE Proceedings of Microwaves, Antennas and Propagation, vol. 144, pp. 251-255, 1997 [40] W. J. R. Hoefer, The Transmission Line Matrix Method Theory and Applications, IEEE Trans. on Microwave Theory Tech. 33(4) 882-893, 1985 [41] A. Fourie, D. Nitch, SuperNEC: Antenna and Indoor-Propagation Simulation Program, IEEE Antennas and Propagation Magazine, vol. 42, no. 3, June 2000 [42] Y. Wang, S. Safavi-Naeini, S. K. Chaudhuri, A Hybrid Technique Based on Combining Ray Tracing and FDTD Methods for Site-Specific Modeling of Indoor Radio Wave Propagation, IEEE Transaction on Antennas and Propagation, Vol. 48, No. 5, May 2000, pp.743-754. [43] G. Wlfle, R. Hoppe, F. Landstorfer, A Fast and Enhanced Ray Optical Propagation Model for Indoor and Urban Scenarios, Based on an Intelligent Preprocessing of the Database, 10th IEEE International Symposium an Personal, Indoor and Mobile Radio Communications (PIMRC), Japan, September 1999 143
8. References [44] S. Y. Seidel, T. S. Rappaport, A Ray Tracing Technique to Predict Path Loss and Delay Spread Inside Buildings, Proc. 1992 IEEE GlobeCom, pp. 649-653 [45] H. L. Bertoni, P. Pongsalamanee, Z. Chen, Defining and Computing the Diffuse Scattering Coefficient for Building Surfaces XXVIIth URSI General Assembly, Maastricht, August 2002, 1407.pdf [46] G. Wlfle, R. Hoppe, T. Binzer, F. Landstorfer, Radio Network Planning and Propagation Models for Urban and Indoor Wireless Communication Networks, Millennium Conference on Antennas & Propagation (AP2000), Davos, 14 April 2000 [47] R. Coman, D. Gajewski, Wavefront-Oriented Ray Tracing with Optimal Ray Density EAGE 64th Conference & Technical Exhibition Florence, Italy, May 2002 [48] F. P. Preparata, M. I. Shamos, Computational Geometry: An Introduction, Springer-Verlag, 1988. [49] A. Taflove, S. C.Hagness, Computational Electrodynamics: The FiniteDifference Time-Domain Method, Artech House, ISBN: 1580530761, 2000 [50] K. S. Yee, Numerical Solution of Initial Boundary Value Problems Involving Maxwells Equations in Isotropic Media, IEEE transaction on Antennas and Propagation, vol. 14, no. 5, 1966, pp.302-307 [51] R. Mittra, Numerical and Asymptotic Techniques in Electromagnetics, Springer-Verlag, New York, ISBN: 0387070729, 1975 [52] T. B. Phong, Illumination for Computer Generated Images'', Communications of the ACM, vol. 18, no. 6, pp. 311-317, June 1975. [53] F. R. Gfeller and U. Bapst, Wireless In-House Data Communication via Diffuse Infrared Radiation'', Proceedings of the IEEE, vol. 67, no. 11, pp. 1474-1486, November 1979. [54] K. F. Man, K.S. Tang, and S. Kwong, Genetic Algorithms, 2nd ed., SpringerVerlag, ISBN: 1852330724, 1999 [55] H. Beyer, The Theory of Evolution Strategies, Springer-Verlag, Berlin, ISBN: 3540672974, 2001 [56] K. Fujimoto, J. R. James, Mobile Antenna Systems Handbook, Artech House, ISBN: 1580530079, 2001 [57] W. C. Y. Lee, Mobile Cellular Telecommunications: Analog and Digital Systems, McGraw-Hill Book Company, ISBN: 0070380899, 1990 [58] R. Valenzuela, O. Landron, D. Jacobs, Estimating Local Mean Signal Strength of Indoor Multipath Propagation, IEEE Transactions on Vehicular Technology, vol. 46, no. 1., Feb. 1997 144
8. References [59] G. Wolfle, F. M. Landstorfer, Prediction and measurement of delay spread, fading statistics and receiving quality in indoor wireless networks The Third European Personal Mobile Communications Conference - EPMCC'99, Paris, France, March 1999 [60] S. Zvnovec, Methods of Indoor Mobil Signal Measurement, Technical Report, CTU Prague, 2001 [61] E. Mdle, Transmitter at 1900 MHz, Diploma Thesis, CTU Prague, 2000 [62] M. Nisznansk, Receiver 1900 MHz, Diploma Thesis, CTU Prague, 2000 [63] M. Polvka, Design and Properties of Base Station Antenna Element for 1800 MHz Band, 10th International Czech-Slovak Scientific Conference Radioelektronika, Bratislava, June 2000, pp. IV-38-IV-39, ISBN 80-227-1389-9 [64] P. Piksa, Probe for mobile measurement at 1900 MHz, Technical Report, CTU Prague, 2001 [65] S. Salous, V. Hinostroza, Indoor and between building Measurement with high Resolution Channel Sounder, IEE International Conference on Antenna and Propagation, Exeter, in CD-ROM, 123.pdf, 2003 [66] P. Vainikainen, J. Kivinen, P. Suvikunnas, K. Sulonen: Experimental characterisation of MIMO propagation channels, Proc. of the XXVIIth General Assembly of the International Union of Radio Science, Maastricht, August 2002, p. 2071 [67] M. Gans et. al., Multielement Antenna Systems Capacity Measurements at 2.44GHz in Suburban Outdoor Environment, IEEE Vehicular Technology Conference, Spring 2001 [68] J. Laurila, K. Kalliola, M. Toeltsch, K. Hugl, P. Vainikainen, E. Bonek: Wideband 3-D characterization of mobile radio channels in urban environment, IEEE Transaction on Antennas and Propagation, vol. 50, no. 2, February 2002, pp. 233243
[69] A. Kuchar, J. P. Rossi, E. Bonek, Directional Macro-Cell Channel Characterization from Urban Measurement, IEEE Transaction on Antennas and Propagation, vol. 48, no. 2, February 2000, pp.137-146 [70] P. Delogne, Leaky Feeders and Subsurface Radio Communications, Inspec, ISBN: 090604877X, 1982 [71] A. J. Motley, D. A. Palmer, Directed Radio Coverage within Buildings, IEE Conference on Conservation of Radio Spectrum II, no. 224, September 1983 [72] S. P. Morgan, Prediction of Indoor Wireless Coverage by Leaky Coaxial Cable using Ray Tracing, Bell Laboratories, Lucent Technologies, http://www.belllabs.com/org/wireless/ 145
8. References [73] S. Loredo, L. Valle, R. P. Torres, Accuracy Analysis of GO/UTD Radio-Channel Modeling Scenarios at 1.8 and 2.5 GHz, IEEE Antennas and Propagation Magazine, vol. 43, no. 5, October 2001 [74] L. J. Greenstein, V. Erceg, Y. S. Yeh, M. V. Clerk, A new path-gain/delay-spread propagation model for digital Cellular Channels, IEEE Trans. On Vehicular Technology, vol. 46, no. 2, May 1997 [75] G. Raleigh, M. Cioffi, Spatio-Temporal Coding for Wireless Communication, IEEE Transaction on Communications, vol. 46, no. 3, March 1998 [76] B. Holter, On the Capacity of the MIMO Channel A tutorial introduction, Norwegian University of Science and technology, Department of Telecommunication, 2000 [77] C.-L. Chen, Elements of Optoelectronics and Fibre Optics, Irwin Publishing, Inc., 1996 [78] ITU-R Recommendation P.1238: Propagation Data and Prediction Models for the Planning of Indoor Radiocommunication Systems and Radio Local Area Networks in the Frequency Range 900 MHz to 100 GHz, Geneva, 1997 [79] R. Hoppe, G. Wlfle, F. M. Landstorfer, Messung der Gebudepenetration und Entwicklung von Ausbreitungsmodellen zur Berechnung der Funkversorgung in Gebuden Projekttag des FMS, Forschungsverbund Medientechnik Sdwest, Stuttgart, December 1998 [80] R. A. Valenzuela, S. Fortune, J. Ling, "Indoor Propagation Prediction Accuracy and Speed Versus Number of Reflections in Image-Based 3-D Ray-Tracing", IEEE 48th Annual International Vehicular Technology Conference, Ottawa, May 1998, pp. 539-543
146
8. References
List of Publications
Note: The publications are in chronological order. [81] P. Pecha, M. Klepal, 900 MHz Signal Propagation in Multistorey Buildings, Proceedings of Workshop 99, Prague, February 1999, pp. 172 [82] P. Pecha, M. Klepal, M. Maznek, Modelling and Testing PMC Systems Inside Buildings, 9th International Czech-Slovak Scientific Conference Radioelektronika 99, Brno, May 1999, pp. 439-441, ISBN 80-214-1327-1 [83] M. Maznek, M. Klepal, Anechoic and EMC Chamber Modelling, 9th International Czech-Slovak Scientific Conference Radioelektronika 99, Brno, May 1999, pp. 237-240, ISBN 80-214-1327-1 [84] M. Klepal, EMC Design Rules for Mechanical Enclosures, Technical Report, Alcatel Bell in Belgium, Hoboken, June 1999, 48 p [85] P. Pecha, M. Klepal, Empirical Modelling of Multi-Floor Signal Propagation, Proceedings of Workshop 2000, CTU Prague, Fabruary 2000, vol. A, pp. 215, ISBN 80-01-02229-3 [86] M. Polvka, M. Maznek, M. Klepal, EMC (EMI) Standards in RF Frequency Bands, Proceedings of Workshop 2000, CTU Prague, February 2000, vol. A, p. 172, ISBN 80-01-02229-3 [87] M. Maznek, H. Brtk, P. Pecha, M. Klepal, Measurement Methods in Antenna Laboratory at the Department of Electromagnetic Field, Proceedings of Workshop 2000, CTU Prague, March 2000, vol. A, pp. 184, ISBN 80-01-02229-3 [88] P. Pecha, M. Klepal, M. Maznek, Indoor Propagation Modelling in MultiStorey Buildings in Prague, Millennium Conference on Antennas & Propagation in Davos, April 2000, p1580.pdf (CD-ROM), ISBN 92-9092-776-3 [89] M. Maznek, M. Klepal, P. Pecha, M. Polvka, Anechoic and EMC Chambers Modelling, Design, Testing, Millennium Conference on Antennas & Propagation in Davos, April 2000, vol. p0725.pdf (CD-ROM), ISBN 92-9092-776-3 [90] M. Maznek, M. Klepal, P. Pecha, Electromagnetic Field in Anechoic and EMC Chambers - Part 1 Modeling, Radioengineering, vol.9, no 1, April 2000, pp. 21-26, ISSN 1210-2512 [91] P. Pecha, M. Klepal, Empirical Models for Indoor Propagation in CTU Prague Buildings, Radioengineering, vol.9, no 1, April 2000, pp. 31-36, ISSN 12102512 [92] M. Maznek, M. Klepal, P. Pecha, EMI Standards and Measurement above 30 MHz, 10th International Czech-Slovak Scientific Conference Radioelektronika, Brno, May 2000, pp. IV-34-IV-37, ISBN 80-227-1389-9 147
8. References [93] M. Klepal, Empirick modely en elektromag. vln uvnit vcepodlanch budov, Graduation Thesis, Prague, June 2000, 55 p [94] P. Pecha, M. Klepal, K. Novotn, Novel Approach for Indoor Propagation Modelling, Radioengineering, vol. 9, no. 3, pp. 12-16, ISSN 1210-2512 [95] M. Klepal, P. Pecha, Novel Approach for Outdoor and Indoor Propagation, 10th International Czech-Slovak Scientific Conference Radioelektronika, Bratislava, June 2000, pp. IV.54 57m, ISBN 80-227-1389-9 [96] P. Pecha, M. Klepal, Novel Approach for Indoor Propagation, IEEE APS International Symposium USNC/URSI National Radio Science Meeting, Utah, August 2000, pp. 50-51, ISBN 0-7803-6372-8 [97] P. Pecha, M. Klepal, Software pro nvrh bezdrtovch systm uvnit vcepodlanch budov, Czechoslovakia Section IEEE Workshop CAD&CAE 2000, Prague, September 2000, pp. 1-4, ISBN 80-902417-6-3 [98] M. Klepal, P. Pecha, Modely pro nvrh pikobukovch st uvnit budov, Sdlovac technika 2001, vol. 49, no. 4, pp. 16-19, ISSN 0036-9942 [99] P. Pecha, Z. Hradeck, M. Klepal, Students Projects in Modern Antennas, Proceedings of Workshop 2001, CTU Prague, February 2001, vol. A, pp. 401, ISBN 80-01-02335-4 [100] M. Klepal, P. Pecha, M. Maznek, New Fast Approach to Wide-Band Propagation Prediction in Picocells, 11th IEE International Conference on Antennas and Propagation, Manchester, April 2001, vol. 1, pp. 216-217, ISBN 085296-733-0 [101] M. Klepal, P. Pecha, Signal Strength and Impulse Response Prediction by Motif Model, 11th International Czech-Slovak Scientific Conference Radioelektronika, Brno, May 2001, pp. 274-277, ISBN 80-214-1861-3 [102] M. Maznek, Z. Hradeck, M. Klepal, Antenna testing at 884 and 894 MHz, Technical Report, CTU Prague, April 2001, 5 p, 4/K317/21044 [103] M. Klepal, Effective Indoor Propagation Predictions, 5th International Students Conference on Electrical Engineering POSTER 2001, CTU, Prague, May 2001, pp. C12 [104] Z. Hradeck, M. Klepal, Studie rozmstn antn na micm voze, Technical Report, Prague, CTU, May 2001, 7/41/21091/317, 42 p [105] P. Pecha, M. Klepal, S. Zvnovec, Results of Indoor Propagation Measurement Campaign at 1900 MHz, Radioengineering, 2001, vol. 10, no. 4, pp. 2-4, ISSN 1210-2512
148
8. References [106] P. Pecha, M. Klepal, I-PROP 1.3 - software pro plnovn pikobukovch st, Czechoslovakia Section IEEE Workshop CAD&CAE' 2001, Praha, September 2001, pp. 18-19, ISBN 80-86582-04-3 [107] M. Klepal, Z. Hradeck, Study and Design of Antennas Layout on Car Roof, 11th Conference on Microwave Technique - COMITE 2001, Pardubice, September 2001, pp. 229-232 [108] M. Klepal, P. Pecha, Wide-band indoor propagation prediction by Motif model, 10th Aachen Symposium on Signal Theory - Algorithms and Software for Mobile Communications, Aachen, September 2001, VDE Verlag, pp. 405-409, ISBN 3-8007-2610-6 [109] P. Pecha, M. Klepal, Effective Indoor Propagation Predictions, 54th IEEE Vehicular Technology Conference VTC Fall, Atlantic City, October 2001, pp.1-4, ISBN 0-7803-7007-4 [110] M. Klepal, Z. Hradeck, M. Maznek, P. Pecha, Theory and Modelling of Electromagnetic Wave Propagation and Design Leading to Reduction of Information Leakage due to Electromagnetic Radiation, Technical Report, CTU, Prague, January 2002, pp. 53, K317/1/VZ/02 [111] Z. Hradeck, M. Klepal, Study and Design of Antennas Layout for System Qvoice, POSTER 2002, Book of Extended Abstracts, CTU Prague, May 2002, pp. C12 [112] M. Klepal, D. Pesch, Z. Hradeck, Optimising Motif Models for Indoor Radio Propagation Prediction using Evolutionary Computation, The Irish Signals and Systems Conference 2002, Cork, June 2002, pp. 301-306. ISBN 0-9542973-0-X [113] M. Klepal, P. Pechac, Prediction of Wide-Band Parameters of Mobile Propagation Channel, XXVIIth URSI General Assembly the International Union of Radio Science [CD-ROM], Maastricht, August 2002, pp. 459 [114] P. Pecha, M. Klepal, P. Ldl, F. Mikas, S. Zvnovec, en elektromagnetickch vln, Workshop Education Forum Preposition, CTU. Prague, September 2002, pp. 47, ISBN 80-86582-07-8 [115] M. Klepal, Predikce en signlu uvnit budov pro systmy novch generac, Workshop - Grant Forum, CTU Prague, September 2002, pp. 4-9, ISBN 8086582-06-X [116] M. Klepal, M. Maznek, Vpoet a rozbor innosti stnn, Technical Report, CTU Prague, November 2002, 24/41/02090/317, 6 p [117] Z. Hradeck, H. Brtk, M. Klepal, V. Tysl, Men stncch vlastnost materil, Workshop - Grant Forum, CTU, Prague, December 2002, pp. 65-67, ISBN 80-86582-06-X
149
8. References [118] M. Klepal, Z. Hradeck, P. Pecha, Support of Tutorial by Up-to-date Methods of Electromagnetic Wave Propagation Prediction, Proceedings of Workshop 2003, CTU Prague, February 2003, vol. A, pp. 496-497, ISBN 80-01-02708-2 [119] M. Klepal, P. Pecha, Large Dynamic Range Prediction of AOA, AOD and PDP for MIMO Systems, IEE 12th International Conference on Antennas & Propagation - ICAP 2003, Manchester, March 2003, pp. 775-779, ISBN 0-852967527 [120] M. Klepal, P. Pecha, Rices k Factor and Channel Capacity Distribution Prediction for MIMO Systems, 13th International Czech-Slovak Scientific Conference Radioelektronika, Brno, May 2003, pp. 209-212, ISBN: 80-2142383-8 [121] M. Klepal, Site-Specific Wide-Band Channel Capacity Prediction for MIMO Systems 7th International Students Conference on Electrical Engineering POSTER 2003, CTU Prague, May 2003, pp.C22 [122] M. Klepal, Motif Model - New Approach to Wideband Propagation Prediction in Picocells, Internet Presentation, web.cvut.cz/en/fee/k317/personnel/klepal/motifmodel/
150
Appendix A Description of the Basic Motifs

Plane-Wall Motif
Angle of Arrival 0180 Degrees
R T
PwI
Relative Reflected Power PwR = PrRT PwE Relative Transmitted Power PwT = (1 PrRT ) PwE
PwE PrRT
PwE (1 PrRT )
(Example for PrDS = 0.73, PrRT = 0.5)
End-Wall Motif
R
PwI 2
PwI 2
Relative Reflected Power PwR 1 = PrRT PwE 2

PwE 2
Relative Transmitted Power PwT 1 1 = (1 PrRT ) + PwE 2 2
PwE PrRT 2
PwE (1 PrRT ) 2
151
T-Wall Motif
1
PwI 2 PwI 2
Emitted Powers Distribution into Quadrants: Pw1 =0 PwE

PwE PrRT 2 PwE PrRT 2 PwE 2 (1 PrRT ) 2 PwE (1 PrRT ) 2
3
Pw3 1 1 2 = (1 PrRT ) + (1 PrRT ) PwE 2 2 Pw4 1 1 = PrRT + PrRT = PrRT PwE 2 2
152

1
PwI 2 PwE 2 PrRT 2
PwE PrRT (1 PrRT ) 2
PwI 2 PwE 2 PrRT 2
Emitted Powers Distribution into Quadrants: Pw1 1 = PrRT (1 PrRT ) PwE 2
Pw2 1 2 1 2 2 = PrRT + PrRT = PrRT PwE 2 2 Pw3 1 1 = PrRT (1 PrRT ) + (1 PrRT ) PrRT PwE 2 2
= (1 PrRT ) PrRT
PwE (1 PrRT ) 2
PwE PrRT (1 PrRT ) 2 PwE (1 PrRT ) PrRT 2
PwE 2 (1 PrRT ) 2
Pw4 1 1 2 = (1 PrRT ) + (1 PrRT ) PwE 2 2
Corner Motif
1
PwI 2 PwE PrRT 2
PwI 2
Emitted Powers Distribution into Quadrants: Pw1 =0 PwE
Pw2 1 = PrRT PwE 2 Pw3 1 1 = (1 PrRT ) + (1 PrRT ) = PwE 2 2

= (1 PrRT )
PwE PrRT 2
4
PwE (1 PrRT ) 2
3
PwE (1 PrRT ) 2
Pw4 1 Pw2 = PrRT = PwE 2 PwE

153

1
PwE PrRT 2
PwI 2
PwI 2
Emitted Powers Distribution into Quadrants: Pw1 1 = PrRT PwE 2
Pw2 =0 PwE
PwE 2

PwE 2 (1 PrRT ) 2 PwE (1 PrRT ) PrRT 2
Pw4 1 2 = (1 PrRT ) PwE 2
154

1
PwE (1 PrRT ) 2
PwE (1 PrRT ) 2
Emitted Powers Distribution into Quadrants: Pw1 1 1 = (1 PrRT ) + (1 PrRT ) = PwE 2 2
= (1 PrRT )
PwE 2 PrRT 2 PwE PrRT (1 PrRT ) 2

4
Pw2 1 = PrRT (1 PrRT ) PwE 2 Pw3 1 2 1 2 2 = PrRT + PrRT = PrRT PwE 2 2 Pw4 1 Pw2 = PrRT (1 PrRT ) = PwE 2 PwE
PwE 2 PrRT 2 PwI 2
PwI 2
Cross-Wall Motif
1
PwE 2 PrRT 2 PwI 2
PwE 2 PrRT 2 PwI 2
Emitted Powers Distribution into Quadrants: Pw1 1 2 1 2 2 = PrRT + PrRT = PrRT PwE 2 2

4
PwE 2 (1 PrRT ) 2 PwE 2 (1 PrRT ) 2

3
Pw2 1 1 = PrRT (1 PrRT ) + (1 PrRT )PrRT = PwE 2 2 = PrRT (1 PrRT ) Pw3 1 1 2 2 = (1 PrRT ) + (1 PrRT ) = PwE 2 2
= (1 PrRT )
2
Pw4 Pw2 = PwE PwE

155
Pillar Motif
R T
PwI
Pw E
Relative Transmitted Power PwT =1 PwE
156
Martin Klepal was born in Koln, Czech Republic. He received his bachelor and masters degrees in Radioelectronics at the Department of Electromagnetic Field, Faculty of Electrical Engineering, Czech Technical University in Prague in 1998 and 2000, respectively. Since 1998, he has been a member of the anechoic and EMC antenna chamber research group at Czech Technical University, working toward his Ph.D. degree. He took up the research fellowships at Alcatel Bell in Belgium in 1999 and at the Cork Institute of Technology in Ireland in 2002. His research focuses on radio wave propagation and RF channel modelling. His recent work has been on indoor propagation. He is the author or co-author of about 30 journal and conference publications. He was received an award from the Czech Republic Branch of IEE in 2001, and the URSI Young Scientist Award in 2002.
EFFICIENT IMPLEMENTATION OF DETERMINISTIC 3 -D RAY TRACING MODEL TO PREDICT PROPAGATION LOSSES IN INDOOR ENVIRONMENTS
Abdullah E Shaikh, Faisal Majeed, Muhammad Zeeshan, Tahir Rabbani, Dr. Imran Sheikh University of Engineering & Technology, Lahore, Pakistan, ashaikh@iee.org
Abstract This paper discusses an efficient implementation of Deterministic Ray Tracing model for indoor wireless propagation. It is important to characterize the indoor propagation channel to ensure satisfactory performance of a wireless communication system. Ray tracing method is extremely effective in indoor environments for estimating the average propagation loss caused by severe multi path reflections. In this model the dominant rays are calculated using Geometrical Optics implemented by the method of images. Deterministic models are computationally expensive but provide accurate results, which become essential in Indoor Wireless site -specific propagation modeling. A Software package has been developed which performs 3-D ray tracing and calculates signal strength variations in a given wireless environment. To validate software results, experimentation at 836 MHz & 1 GHz was carried out in the building of Electrical Engineering Department at UET Lahore. Measured results were mostly found to be in close agreement with the calculated values, proving the validity of the implemented model. Keywords Indoor Propagation Modeling, Ray Tracing Technique. I. INTRODUCTION Wireless has recently become an increasingly viable option for indoor communication systems. The availability of higher frequency bands in the 900 MHz and 2.4 GHz range has made wireless an attractive option for high bandwidth digital communication applications such as IEEE 802.11, Blue Tooth, Palm Pilot and temporary communication setups. Wireless networks can be particularly advantageous for applications, which require portability, or where installation of wiring is undesirable or impractical. Multipath interference, or interference due to the reception of multiple copies of a signal due to reflections, is known to be a problem in all types of indoor communication channels. At UHF and microwave frequencies, the presence of walls and large objects in rooms makes the indoor multipath environment quite different from most outdoor scenarios. As a result, the study of indoor propagation characteristics has become an area of increased importance. II. CHANNEL MODELING In order to evaluate the effectiveness of a given channel coding and processing technique before implementation, some model of the channel must be developed that adequately describes the environment. Such analysis reduces the cost of developing a complex system by reducing the amount of hardware that has to be developed for evaluation of performance. Theoretical models have an added advantage in their ability to reproduce a channel for comparison between various communication strategies, resulting in an accurate measure of relative performance. Indoor channels are highly dependent upon the placement of walls and partitions within the building. In such cases, a model of the environment is a useful design tool in constructing a layout that leads to efficient communication strategies. To achieve this aim, a channel model of an indoor environment must be applied to various layout plans of offices, which will lead to the characterization of design methodologies. Much work has been carried out in measuring the channel in buildings of various layouts, and structural compositions. Theoretically the solution of Maxwells equations with boundary conditions derived from the building geometry can be used to calculate EM wave propagation characteristics. Practically, this approach is not feasible due to its mathematical complexities, which is beyond the power of todays computers. This opens the door for methods that can be used to calculate approximate numerical solutions, ray-tracing being one of these. Ray tracing is successfully used in Optics to model characteristics of various optical systems. The concept of ray-tracing modelling is based on the fact that high-frequency radio waves behave in a ray-like fashion. Therefore signal propagation can be modelled as ray propagation. By using the concept of ray-tracing, rays may be launched from a transmitter location and the interaction of the rays with the partitions within a building modelled using wellknown reflection and transmission theory. Two types of ray -tracing methods the image method and the brute-force ray-launching method are being used in the characterization of indoor electromagnetic-wave propagation. For scatters bounded by plane faces it is
0-7803-7589-0/02/$17.00 2002 IEEE
PIMRC 2002
convenient to employ the image method to mirror the radio wave source at a particular face. The point where the mirror face intersects the line connecting the transmitter image and the receiver is the point at which specular reflection occurs. This method is well suited to radio propagation analysis in the case of geometries of low complexity and where a low number of reflections are considered. The brute-force ray-tracing method considers a bundle of transmitted rays that may or may not reach the receiver. The number of rays employed and the distance from the transmitter to the receiver location determine the available spatial resolution and hence the accuracy of the model. Ray tracing can be much less demanding of computations than methods based on Maxwells equations. With the computing powers currently available on personal computers and workstations, the ray tracing approach provides a challenging but feasible method of propagation modeling 1- 2. Reliable site-specific ray-tracing propagation prediction models for each building based on its detailed geometry and construction can be very effective tools in designing indoor communication system. When a signal is transmitted through or reflected off a wall or a partition, the degree of signal attenuation and the amount of phase change depend on the complex transmission and reflection coefficients, respectively. These are computed from the complex permittivities of the materials that signal rays encounter. Other factors affecting the transmission and reflection of the signal are the angle of incidence and the relative polarization. The complex transmission coefficient is defined as the ratio of the transmitted to the incident electric field strengths and the complex reflection coefficient is defined as the ratio of the reflected to the incident electric field strengths. Due to the complexities of modeling a dynamic environment using a ray tracing technique, it is only practical to simulate the fading channel of a mobile system by moving either the transmitter or the receiver around the environment. The results of this process may be presented in terms of the signal power that would be experienced by a receiver moving through this environment, or in a more detailed form showing the multi path nature of the environment, and the changing channel impulse response as the receiver moves around the environment. In ray tracing methods, the location of transmitters and receivers are assigned to points referenced by three-dimensional coordinates. The walls, partitions, ceilings and floors in an indoor environment are modeled as plane surfaces of given thickness and permittivity. For simplification, curved surfaces can be modeled by piece-wise planar surfaces. The rays from the transmitter antenna are reflected off walls, partitions, ceilings, floors and tables etc. to arrive at the receiver.
It is evident that in case of indoor wireless propagation the proximity and number of walls, which act as good reflector of electromagnetic energy and also introduce a signal loss and phase change when a ray is transmitted through them, only ray tracing based modeling methods can give a true picture. This certainly requires much work because for any successful ray tracing based software requires the following as minimums inputs: A detailed geometry of the building under consideration. This geometry must be in three-dimensional coordinates and for image theory based method non-planar surfaces must be broken down in planar segments before they can be inputted to the software. Fortunately, the information about building geometry is mostly available, sometimes in electronic CAD formats and this for almost all practical purposes consists of planar surfaces . Each wall and partition must be accompanied by its real and complex permitivitivities, because these are used to calculate the complex reflection and transmission coefficients given in [3]. Extensive tables of these values for common building walls and partitions are available in the literature. In case the exact values are not available typical values can give a much better picture compared to classical statistical techniques. In case of outdoor open environment one obstacle that is always present is the ground; this is true for indoor scenarios too. Ground acts as a good reflector and this must be modeled as an infinite plane in case of indoor scenarios. Any energy from the radiating element directed toward the ground undergoes reflection. The amount of reflected energy and its direction are controlled by the geometry and constitutive parameters of the ground. In general the ground is a lossy medium whose effective conductivity increases with frequency. Therefore it should be expected to act as a very good conductor above a certain frequency, depending primarily upon its moisture content. To simplify the analysis we will assume that the ground is a perfect electric conductor, flat, and infinite in extent. The same procedure can also be used to investigate the characteristics of any radiating element near any other infinite, flat, perfect electric conductor. In practice, it is impossible to have infinite dimensions but we can simulate (electrically) very large obstacles.
III. SOFTWARE IMPLEMENTATION The software provides a Graphical User Interface (GUI) for editing and defining 3D building database. For simplicity and efficiency only planar partitions are allowed. Any polygonal shape with n-edges can be entered with the coordinates of its vertices in a CCW direction. Each
partition must be defined with a thickness and complex permittivity, which is then used to calculate reflection and transmission coefficients. Two types of studies are supported. The Point Study calculates field strength for only one location of transmitter. The Line Study can be used to simulate a mobile receiver traveling along a given linear path, which is sampled at a user-defined interval. For each study the user is asked to define the number of reflections that need to be considered. T algorithm starts by prehe computing all images and reflection and transmission coefficients of all partitions. This pre-processing is found to give a significant boost to software performance. Ray tracing is done by an exhaustive search of this image tree taking into account decomposition of ray at each planar intersection. First of all contribution of line of sight ray is calculated. This ray is always possible but the number of transmissions it has to go through varies from case to case. It is found from the profiling studies of the software that the maximum time is consumed in calculating intersections, because if we take the brute force approach, for each ray we need to check each partition as a potential transmission candidate. Fortunately, a well known optimization technique from Computer Graphics comes to our rescue. Each partition is accompanied by a cubic bounding box, and actual intersection calculations are done only if the ray passes though the bounding box. This optimization technique enables us to discard most of false targets without incurring the overhead of intersection calculations. The partitions, which pass the transmission test, are then used to calculate the resultant complex electric field value at the reception point. The phasor sum of individual contribution of each successful ray is then used to compute final value of Electric field vector which is then used to calculate received power. In case of line study this procedure is repeated for each point on the specified linear path. The results are then graphically presented to the user. The software also provides the option to export the results to MATLAB for further processing if desired. A flow chart of the software is presented in Fig.1.
Figure 1: Flow chart of the implemented software IV. RESULTS Following are the sample results obtained from our software package named Ray Tracer 2000. The scenario consists of Communication Lab at UET Lahore with dimensions of 8.23 x 24.7 meters. The number of reflections to be accounted for was set equal to 4, and the number of points (for line study) was set as 100. Average execution time was less than a minute on an AMD K6-II 550 MHz processor, with 256 MB RAM. Simulation Results Results of two simulations are shown in Fig 2. The two scenario s represent real-life indoor scenarios at UET Lahore. In both cases the results are shown for a line study with 100 points. The fast fading nature of indoor channel is quite evident. Also in case of both scenarios the 0-th order line-of-sight contribution is smaller than the contribution of higher order reflections, making it evident that line-of-sight approximation which is universally used for outdoor environment is no longer valid in indoor case.
Simulations:
Room Layout (S cenario 1)
Room Layout (Scenario 2)
E Field for Line study (Scenario 1)

0.14 Tx : (2,2,2) Rx: (7,7,7) moved to (9,9,9) 0.12 Sampling points: 100 points Frequency: 1 Ghz 0.1 Number of reflections: 4
0.08
0.06
0.04
Total E Field 0th Order 1st Order 2nd Order 3rd Order 4th Order
0.02
Calculation Points
E Field for Line study (Scenario 2)

0.14 Tx : (7,7,2) Rx: (15,15,2) moved to (18,18,6) Sampling points: 100 points Frequency: 2 Ghz Number of reflections: 4
0.12
Absolute E Field (V / m sq.)
0.1
0.08
0.06
Total E Field 0th Order 1st Order 2nd Order 3rd Order 4th Order
0.04
0.02
0
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51 53 55 57 59 61 63 65 67 69 71 73 75 77 79 81 83 85 87 89 91 93 95 97 99
Calculation Points
Figure 2, Simulation Results
V. FURTHER WORK We plan to work on following missing pieces in the future: Inclusion of diffraction effects in ray tracing calculations There must be a provision to input 3D directional characteristics of antenna, which should be used a direction dependent gain adjustment factor. For digital systems the delay spread is very important, the software will be enhanced for its calculation. Software should provide a 3D CAD like environment to replace the current laborious process of manual entry. Also there should be a provision to import building database, if one already exists in electronic CAD format. Conclusion The high degree of correlation between the experimental results and calculated values indicate the accuracy and utility of the 3D Ray tracing in predicting the propagation channel characteristics of indoor wireless environments. We feel that there is lot of work to be done in extending the functionality of the software developed to include the RMS Delay spread, wide band channel characterization, area study and modulation cum coding effects.
VI. REFERENCES
[1] M. C. Lawton and J. P. McGeehan,: The application of GTD and ray launching techniques to channel modelling for cordless radio systems, in IEEE Proc. Vehicular Technology Conference, VTC92, Denver, CO, May 1992, pp. 125-130 [2] K. R. Schauback, N.J. Davis, and T.S Rappaport,: A . ray tracing method for predicting path loss and delay in microcellular environments, in IEEE Proc. Vehicular Technology Conference, VTC92, Denver, CO, May 1992, pp. 932-935 [14] W. K. Tam & V.N. Tran: Propagation modelling for indoor wireless communication, IEE Journal of Electronics & Communications October 1995, Vol. 7 Number 5 pp. 224-226 About the Authors Abdullah E Shaikh, Faisal Majeed Hayat, Muhammad Zeeshan and Tahir Rabbani Shah are the students of final year Electrical Engineering at the University of Engineering & Technology (UET) Lahore, Pakistan. The project was supervised by Dr. Imran Sheikh, who has significant contributions to the field of wireless propagation modelling both in indoor and outdoor environments.
398
Z. LAI, G. DE LA ROCHE, N. BESSIS, P. KUONEN, G. CLAPWORTHY, D. ZHOU, J. ZHANG, INTELLIGENT RAY LAUNCHING. . .
Intelligent Ray Launching Algorithm for Indoor Scenarios

Zhihua LAI1 , Guillaume De La ROCHE2 , Nik BESSIS2,3 Pierre KUONEN4 , Gordon CLAPWORTHY2 , Dibin ZHOU5 , Jie ZHANG6
Wireless Network Design Ltd., Suite 25, The Business Competitiveness Centre, Kimpton Road, Luton, Bedfordshire, United Kingdom, LU2 0SX 2 Institute for Research in Applicable Computing, University of Bedfordshire, Park Street, Luton, Bedfordshire, United Kingdom, LU1 3JU 3 University of Derby, Kedleston Road, Derby, Derbyshire, United Kingdom, DE22 1GB 4 GRID and Ubiquitous Computing Group, University of Applied Sciences of Fribourg, UCH-1705 Fribourg - Switzerland 5 College of Information Science and Engineering, Hangzhou Normal University, China, 310036 6 Dept. of Electronic and Electrical Engineering, University of Shefeld, Mappin Street, Shefeld, United Kingdom, S1 3JD zhihua.lai@ranplan.co.uk, guillaume.delaroche@beds.ac.uk, n.bessis@derby.ac.uk, pierre.kuonen@hefr.ch, gordon.clapworthy@beds.ac.uk, dibinz@zju.edu.cn, jie.zhang@shefeld.ac.uk
1 Ranplan
Abstract. This article describes the indoor IRLA (Intelligent Ray Launching Algorithm), which originates from an efcient outdoor propagation prediction model. Implementation and validation are given in detail. An indoor ofce scenario is selected and simulations via the IRLA model and two other reference models have been performed. Predictions are analyzed and recommendations are given. Results show that the indoor IRLA model is suitable for indoor wireless network planning and optimization process.
small-scale propagation prediction deals with fast fading (i.e. the variation of signal strength over a short period of time such as one wavelength). For example, Rice fading distribution has been used to model the LOS (Line-of-Sight) case while Rayleigh fading distribution is being used widely in NLOS (None-Line-Of-Sight) [3]. On the other hand, the large-scale propagation prediction computes the average signal strength over a longer period of time. In this case, the predictions will give path loss based on (1): PL = PRx PTx (1)
Keywords
Intelligent ray launching algorithm, radio wave propagation prediction, indoor scenarios, path loss models, in-building network planning.
where PRx represents the signal strength (dBm) at the location of the receiver and PTx represents the signal strength (dBm) at the location of the transmitter. The modeling of large-scale radio wave propagation in indoor environments plays a crucial role in the investigation of 3G/4G network planning applications (such as localisation). In indoor environments, there are usually more irregular objects and material types, which make modeling much more complex, compared to outdoor environments. Many outdoor large-scale propagation models such as [4] are accelerated based on the simplication that outdoor buildings are 2.5-D polygons with at roofs. However, as objects in indoor environments can be of any shape and in any position, such as lamps hanging at different heights, etc. Indoor radio wave propagation predictions are usually more challenging because these irregular objects impact greatly the indoor propagation characteristics (such as fast fading). In general, large-scale propagation models fall into two kinds: empirical and deterministic. Empirical models are mainly based on empirical factors such as distance or frequency. They are computational fast but they do not consider much environmental information so their accuracy is limited. For example, the ITU (International Telecommunication Union) Model for Indoor Attenuation [5] is based on a single equation and the path loss prediction is valid only for
1. Introduction
Radio wave propagation prediction modeling has become increasingly important in wireless network planning and optimization [1] [2] since the emergence of 3G networks. The propagation predictions serve as a fundamental output for the advanced analysis and optimization such as capacity and link budgets etc. The identication of QoS (Quality of Services) or weak signal spots are based on the estimation of propagation prediction. At present, in order to build an efcient indoor DAS (Distributed Antenna System), the candidate antennas need to be tested and the possible combination is evaluated, which will give an optimal solution for indoor antenna placement. The solution is based on propagation prediction modeling because the minimal coverage ratio has to be considered. Currently, radio wave propagation models consist of two kinds: small-scale and large-scale. On the one hand,
RADIOENGINEERING, VOL. 20, NO. 2, JUNE 2011
399
frequency ranging from 900 MHz to 5200 MHz and oors from 1 to 3. Similarly, LAM (Linear Attenuation Model) [6] relies on measurement data, based on which a linear equation can be built. On the other hand, deterministic approaches take into account the environmental information such as object positions and the corresponding materials. Generally speaking, these approaches are more time-consuming compared to empirical models but a higher level of accuracy can be obtained. For example, in [7], the authors propose an accelerated dominant ray-based method for indoor scenarios. Despite many acceleration techniques such as [8], [9] and [10], the use of accurate propagation modeling for indoor scenarios remains limited due to the complex indoor propagation environment. Apart from these two categories, some propagation models consider both empirical and deterministic factors, which are categorized as semi-empirical (or semideterministic) approaches. For example, MOTIF [11] can be considered as a semi-deterministic approach that includes stochastic factors and deterministic computation. Such models usually perform faster than deterministic approaches such as ray tracing and their accuracy is high in some scenarios. For example, MOTIF is limited in 2-D scenarios. Ray-based methods can be categorized as deterministic approaches. They are widely used in propagation prediction. Compared to FDTD (Finite Difference Time Domain)-like methods [12], they consume less memory and are far more efcient. These ray-based methods compute the possible rays between the emitter and receivers in complex environments and they need to search the rays to compute reections and diffractions, based on Descartes laws. Hence, they tend still to be very time-consuming if the environment is complex, i.e., if there are a large number of obstacles. Usually, the accuracy of ray-based methods is limited by the number of rays that can be computed within a reasonable time. Different methods to accelerate ray-based methods are proposed in the literature. For example, in [8], a preprocessing stage is required to compute the visibility tree between obstacles and in [7], the authors propose a dominant path model which only computes the few rays that comprise of the ones that give the most contribution. Depending on the ways of computing the rays, ray-based methods can be further categorised into ray tracing and ray launching. On the one hand, ray tracing computes the rays backwards from the receivers. For example, the reection ray can be computed by: (a) mirroring the receiver at the targeted facet, (b) computing the intersection point between line segment from transmitter to the mirrored point and the facet, (c) launching the reection from the intersection point to the receiver direction. Ray tracing gives precise rays between a transmitter and a receiver and is thus suitable in point-to-point scenario where there are only a few receiver locations of interests. However, if the number of prediction locations is considerably large, ray tracing will suffer from a long running time, especially in scenarios where there are many complex obstacles which will incur more data-intensive operations such as in-
tersection tests. Besides, ray tracing treats the calculation of neighbor pixels equally the same i.e., the computational time is roughly linearly proportional to the number of receiver locations. In contrast, ray launching computes the rays from the emitter. Ray launching is an image-sampling method which uses discrete rays by an angle. Inevitably, gaps will be created gradually after the rays undergo reections and diffractions. In order to solve this problem, a reception sphere [13] can be used to capture the missing rays. More rays can be launched to improve the accuracy but this will slow down the computation. In general, ray launching walks through the rays and computes the reections, transmissions and diffractions iteratively. The pixels gain experiences from its previous pixels along the same path, which is faster than ray tracing. Ray launching is suitable in point-to-many scenarios, such as coverage prediction. The ray launching may be more suitable for wireless network planning and optimisation in indoor scenarios because generally it is computationally more efcient than ray tracing and it provides a relevantly acceptable level of accuracy.
1.1 Related Work and Contribution

The authors have originally developed the IRLA model for outdoor scenario [2]. In [14] and [15], the authors proposed and implemented the parallel IRLA based on POPC++ (Parallel Object-oriented Programming in C++) [16] and a performance speed up was observed. In [17], the IRLA model was improved by smart algorithms to solve angular dispersion of ray launching and thus the accuracy is improved. In [18], the authors extended the IRLA model to indoor scenarios and it has been validated by the measurement campaign proposed in [19]. The results show that the indoor IRLA model is promising because it is capable of providing accurate results within a short amount of time. In [20], the authors further combine the IRLA model with a FDTDlike method: MR-FDPF (Multi Resolution Frequency Domain ParFlow) [21] for an indoor to outdoor scenario where the accuracy was validated by the measurement campaign. In [22] and [23], the authors combine the IRLA model with MR-FDPF for outdoor to indoor scenarios. Based on the aforementioned work on the IRLA model [24], This article will contribute by providing: Details on the calibration of the indoor materials. A simplied 2.5-D indoor IRLA model. Simulation results that are compared with two referenced models: MR-FDPF [21] and COST231-Multi Wall [25]. The rest of the article is organised as follows. First, the outdoor model IRLA [2] [14] will be briey described. Next the details will be investigated to make this model suitable for indoor scenarios. The modied 2.5-D IRLA model for indoor scenarios will also be presented. The calibration of materials will be described. Next an indoor measurement
400
campaign will be described, which is used to validate the model. Performance (such as speed, accuracy) will be analysed by contrasting the results obtained via MR-FDPF and COST231-Multi Wall models. The comparison between the 3-D and 2.5-D IRLA models are also discussed with recommendations given and nally, the future prospectives are described, which concludes this article.
2. IRLA Model
The idea of the IRLA model is based on the discrete ray launching algorithm. The input to this model relies on the creation of a 3-D discrete data set which contains the vector building data, material items associated to each obstacle, and the discrete data set made given a dened resolution. Finally, the basic unit of the discrete data set contains the property values (such as edge or corner) and index to the object list. Each object list stores the polygon coordinates and index to the material table. The IRLA model traces the rays from the emitter and by adopting the techniques proposed in [17], the angular dispersion of discrete ray launching is eliminated. The outdoor IRLA model [2] comprises three main components: LOS (Line-of-Sight), VD (Vertical Diffraction) and HRD (Horizontal Reection and Diffraction). The LOS component deals with visibility pixels, collecting direct paths from emitter and most importantly the secondary pixels for the use of VD and HRD. Mathematically and reversely, the VD component calculates the dominant multiple roof-top diffractions by a fast pixel checking principle that draws the shortest edges between the emitter and receivers. The HRD component performs the actual 3-D ray launching. The rays are abandoned when they hit the roofs due to the fact that there seldom exist dominant rays which are a combination of vertical and horizontal planes [4]. The IRLA model for outdoors has been tested to show suitability (in the aspects of both speed and accuracy) in use for wireless network planning applications [24] and the inherent principle of IRLA is easily parallelizable. In [14], a parallel implementation of IRLA via POP-C++ (Parallel Object Oriented Programming in C++) has been presented and performance is evaluated.
approach proposed in [17] to solve the angular dispersion of ray launching. Rays are collected and the multipaths are obtained and hence channel characteristic such as PDP (Power Delay Prole) can be simulated. The process of IRLA prediction for indoor scenarios starts with launching rays in all 3-D directions. Based on the discrete data set, the resolution and the number of cubes along each dimension (X, Y and Z) are known. Therefore the number of discrete rays required can be obtained by connecting the emitter to all the cubes at the fringe of the scenarios [17], which can be found in (2): N = 2Nx Ny + 2(Nz 2)(Nx + Ny 2) (2)
where N is the number of discrete rays and Nx , Ny , Nz are the number of cubes in dimension X, Y and Z respectively. This ensures that no pixels are missing due to angular dispersion of ray launching [17] from the LOS component. The principle is useful in distribution of rays, e.g. in parallel. The secondary cubes collected in component LOS serve as input to the HRD component, which iteratively follows the discrete rays. Rays disperse as they propagate, which causes coverage gaps. To solve this, an intelligent procedure is proposed in [17], which dynamically accounts for rays that ll the gaps. Material indices are recorded within each cubic entry and applied to discrete rays that are being followed. Based on a few measurement locations, the material values are calibrated once and applied to predictions.
2.2 2.5-D Modication of IRLA

In-building deployment usually involves planning and optimization of a multi-oor building in indoor DAS. Thus, the predictions obtained from the indoor IRLA model need to account for the propagation characteristics such as transmission between oors, and the use of a discrete data set may not be so efcient if the scenario is large (such as a skyscraper) and the resolution is ne-grain. It has been observed that the radio wave signal strengths drop dramatically when they undergo a few transmissions through different oors. The trade-off between speed and accuracy thus may be based on the assumption that the signal strength affects a maximum of n (n >= 1) oors, including the oor on which the emitter resides. Thus, discrete data sets for each oor can be created individually and combined during runtime in each prediction, which consumes less memory and improves the efciency. In order to develop a propagation model suitable for indoor DAS planning and optimisation process, the 2.5-D IRLA model for indoor scenarios can thus be proposed based on the following. The LOS component accounts for 3-D LOS rays and corresponding antenna pattern values are estimated and added. The HRD component computes the NLOS rays in the horizontal plane in order to improve the efciency.
2.1 3-D IRLA Model

In [18], the IRLA model was rst extended to the indoor scenario. Modications of the outdoor IRLA model were made. First the component VD has to be eliminated from the indoor IRLA model because it is not applicable to the calculation of vertical diffractions in indoor scenarios. Instead, the indoor IRLA component of HRD is enhanced by also calculating vertical diffractions. The LOS and HRD components are kept as two fundamental components with slight modications. It is known that ray launching suffers from the angular dispersion problem because ray launching is a sampling method which launches the rays that are separated by an angle. Both components are optimized via a new
401
Similarly to the outdoor VD component, the 2.5-D indoor IRLA model employs a VD component that cuts vertically to the indoor scenario which accounts for the vertical diffraction rays. In order to validate the 2.5-D indoor IRLA model, the following section will describe an indoor scenario. The measurement was conducted and compared to four models: the 3-D IRLA model, the 2.5-D IRLA model, the MR-FDPF model and the COST231 Multiwall model, respectively.
3. Experiments
3.1 Scenario
Propagation models need to be validated by measurements for the accuracy. The comparison analysis can aid the improvement of the propagation models. In order to validate the IRLA model, an indoor ofce (Fig. 1) has been selected as the indoor testbed. The ofce has three rooms and is located on the rst oor. There are 255 polygons and more than 1000 vertices all together. The dimension for this scenario is 16 9 4 (m3 ). The materials found in the scenario include: Glass, Wood, Metal, Plastic, and Concrete. As shown in Fig. 1, there are cubicles in the middle, which is a challenge for 2-D propagation models because in reality the radio wave signals travel in vertical directions (such as diffractions on the edges of desk).
indoor base stations. The emitter is located on the table (1.35 meter height) in the meeting room (see Fig. 1) and measurement locations (0.98 meter height) are positioned by the grid pattern on the oor (see Fig. 2). This helps recording positions without an indoor GPS (Global Position System). To avoid as much signal variation (e.g. due to noise disruption) as possible, a measurement campaign is carried out when there are few people in the room. To avoid the interference of human bodies when manually triggering the spectrum analyzer, the measurement data of the rst few and last few points are removed. Several measurement snapshots are taken to average the nal signal strength. Around 200 measurement locations gridded by 0.5-meter-square [19] at ground-level 1.5 meter height are chosen. The measurement techniques and the removal of human body inuence are detailed in [19].
Fig. 2. Measurement locations, positioned by grid pattern on the oor.
3.3 Calibration
It is not possible to know exact properties of materials in every scenario. Therefore, a calibration process is needed to adjust the properties of materials (such as conductivity). The IRLA model provides the calibration process to make the simulations t into reality. A rst run with defaulted parameters of materials is performed and the multipaths associated to each measurement locations are obtained. Based on the multipaths, the calibration process can be performed.
20 18 16 14 12 RMSE 10 8 6 4 2 0 0 2000 4000 6000 8000 Iterations 10000 12000 14000 Current State Global Temperature
Fig. 1. Indoor ofce; X: emitter; A: LOS, B: N-LOS.
3.2 Measurements
The emitter is a 3.525 GHz signal generator (power 6 dBm) with an omni-directional antenna (gain 2.8 dBi, EIRP Equivalent Isotropically Radiated Power 8.8 dBm). This frequency has been selected in order to study WiMax
Fig. 3. Calibration of the IRLA model based on simulated annealing.
The 3-D path loss matrix can also be obtained by the rst run of IRLA. To make it even more efcient, only selected layers (locations) can be considered. Multipaths information for selected locations are computed. With this in-
402
formation, channel characteristics can be investigated. However, since the IRLA model computes the rays from the emitter, the requirement for multipath data does not incur extra overheads since this can be easily recorded together with the path loss. To improve the accuracy of the model, a calibration of the parameters, based on an SA (Simulated Annealing) approach, was implemented in [2] where the RMSE (Root Mean Square Error) between the simulation and the measurement is minimised. The calibration is based on the multipaths i.e., one single IRLA simulation has to be performed and all the rays are stored in memory. The material parameters to be tuned can be considered as a vector v. At each iteration of SA, v is adjusted and the tness is evaluated. Based on a probability, the v is accepted. The SA approach converges to an sub-optimal v nally (Fig. 3). The use of multipaths avoids rerunning the simulation at each iteration of SA, because the tness value can be computed based on the multipath reached at each measurement location.
40 35 30 25 Frequency 20 15 10 5 0 25
20
15
10 5 Prediction error (dB)
10
Fig. 4. Error in dB before calibration: RMSE 6 dB.
4. Performance Evaluation
The indoor scenario described above (Sec. 3) is used to validate the IRLA model. This section introduces the calibration process of the IRLA model and experimental results. Based on the prediction, comparisons can be investigated and recommendations are given.
4.1 IRLA Validation

A single run using a standard PC (2.5 GHz CPU, 4 G RAM) with this scenario takes around 1 minute for the computation of the 3-D path loss and multipaths information. The SA calibration takes around 2 minutes to complete and v is obtained. The default parameters yield around 6 dB RMSE (Figs. 4 and 5) on the rst run. It can be observed that there are some prediction points that are of large differences. This could be caused by the following: The materials and other network parameters (such as path loss coefcients) are not calibrated. The measurement data itself may be affected by many factors such as the variation of environment (such as moving vehicles). The IRLA model may terminate the ray computations at the early stage if the rays carry weak signal strengths due to incorrect summation from uncalibrated materials. After proper calibration, the prediction results compared to measurements show an agreement (Tab. 1), with the 3.5 dB RMSE and a mean error of 0.01 dB.
Fig. 5. Prediction errors (dB).
RMSE (dB) Mean (dB)
Uncalibrated 6 3.2
Calibrated 3.5 0.01
Tab. 1. Calibration of 3-D IRLA.
It can be observed that most predictions are accurate within the ranges of [10, +10] dB difference. There are few points that prediction tends to be either too optimistic or pessimistic. From Fig. 5, the prediction errors can be visualised geographically. It can be seen that the most optimistic predictions are distributed within a short distance range from the emitter and receivers (such as the locations near by the refrigerator or behind the door). The pessimistic prediction points are located far from the emitter. This may be used as important evidence to further optimize the model.
4.2 Comparison with Reference Tools

In this section, the 2.5-D and 3-D IRLA model will be compared with two reference models: MR-FDPF and COST231 Multiwall.
403
The advantage of MR-FDPF is that, due to the multiresolution approach, the computational phase of the propagation of one source is very low compared to a time domain implementation. However, when moving to 3-D implementations, the number of ows to compute increases and in such case the frequency domain implementation has no obvious advantages in term of complexity. Therefore this model is usually restricted to 2-D. In [28] MR-FDPF was successfully used for indoor network planning and it was shown that, when considering at environments where the main propagation effects are in the horizontal plane, it was possible to reach very high accuracy. Moreover, a calibration of the method was also proposed to compensate for these 3-D effects by changing the parameters of the materials. The method was also extended to simulate larger bandwidth, more details can be found in [29]. Due to its accuracy, MR-FDPF is included in a Wi network optimization tool [30]. MR-FDPF has lower complexity than FDTD because of its pre-processing and it directly solves the nal Maxwell equations without time information. At this stage, 2-D MRFDPF is usually tested due to much larger computational complexity requirement when this model is applied in the 3-D cases. In order to use MR-FDPF to predict this indoor scenario, some assumptions have to be made. First a cut on receiver locations from full 3-D data is required. However, a 2-D scenario does not fully reect the 3-D characteristics by approximation of one cut. For example, a table not blocking rays may be a reected source, which is difcult to model in 2-D. Fortunately, by calibration, similarly to the IRLA model, 2-D MR-FDPF can adjust the material properties so that the accuracy can be improved. For example, rays transmitted by a window do not attenuate much but this will be treated as a heavy-thick wall in this 2-D scenario used by MR-FDPF. By calibrating with measurements, this material is adjusted. 4.2.2 COST-231 Multiwall Model COST-231 Multi Wall [25], is a semi-empirical indoor model, which only accounts for the attenuations for walls and oors. Therefore, it does not compute reection or diffraction rays. The only output from this model is the path loss/power prediction, which is an estimation based on the material properties and the number of transmitted walls/oors. This model is computationally efcient and it does not require pre-processing. The running time of the Multi Wall model is usually less than a few seconds. The prediction errors tend to increase with the number of transmitted walls or oors. e.g. the COST231 Multi Wall model generally produces pessimistic results when the receiver locations are far from the emitter.
Fig. 6. Coverage prediction after calibration.
4.2.1 MR-FDPF Model MR-FDPF [21] is a FDTD-like method but in the frequency domain. MR-FDPF (Multi-Resolution Frequency Domain Parow) is based on the ParFlow model derived in [26]. It is a nite difference approach similar to nite difference time domain method (FDTD) which has the advantage of being able to compute all the reections and diffractions without limitations since it solves the Maxwells equations [27]. The formulation of ParFlow is based on the transmission line matrix (TLM) method. In this approach (in 2-D) the eld is modeled by four ows corresponding to the four cardinal directions. In each pixel, also referred to as a node, scattering matrix is associated which efciently models reection and diffraction effects. The advantage of ParFlow compared to FDTD is that the four elds are scalar, thus reducing the number of variables (no E and H elds). In [26] a frequency domain implementation of ParFlow was proposed. The advantage of this formulation is that the steady state of the source can be computed using a recursive formalism, instead of solving the equations for the whole environments. Therefore, a multiresolution approach is used where the nodes are gathered into multi resolution nodes (MR-nodes) and where the problem is divided into sub-problems, thus highly reducing the overall complexity (mainly due to the need for inversion of large matrices). The MR-FDPF algorithm works into two steps. First, a pre-processing phase where the environment is divided into MR-nodes and where the scattering matrices are computed. This phase does not depend on the sources to simulate but only on the scenario. Therefore it only has to be performed once. The second step is the propagation phase which works on the boundary conditions: A source is recursively included in larger space blocks up to the full space, and the backward propagation is done by propagating incoming boundary ows toward the separation line and down to the unitary cells.
4.3 Comparison and Recommendations

Tab. 2 compares the prediction performance obtained via the IRLA (2.5-D and 3-D) model, the 2-D MR-FDPF
404
model and the COST231 Multi Wall model. All simulation results are obtained after calibration. It can be observed that these models generally give a high agreement (cf. Fig. 10) between prediction and measurement. In a full 3-D scenario, at least in this indoor scenario, 2-D MR-FDPF relies heavily on the calibration without which, this model tends to give large prediction errors due to inaccurate modeling of materials. The accuracy before calibration for MR-FDPF is around 8 dB whilst this is dramatically improved to around 3.5 dB due to calibration of the materials. For example, the emitter is placed on a table and the table should be removed from 2-D cut, otherwise it will be treated as an obstacle in MRFDPF model. On the standard PC (AMD 64+ Dual, 4 GB), the preprocessing for MR-FDPF takes around 3 seconds and the computation time is less than 1 second, which is fast in a small 2-D scenario. However, due to its 2-D characteristics, some important ray phenomenal in 3-D are not efciently captured. For example, MR-FDPF treats the ows in only 2-D, as they only propagate in the 2-D plane. Rays bouncing by reecting on the ceiling or oor are ignored. The accuracy obtained though MR-FDPF is 3.5 dB RMSE (0 mean error after calibration). The prediction via 2-D MRFDPF is designed for power level/path loss only, which does not compute the delay information. The 3-D IRLA model for indoors, as presented in this article, is fully applicable in 3-D scenarios in which the model is capable of capturing important 3-D dominant rays. Compared to MR-FDPF, the IRLA models (3-D and 2.5-D) do not require a preprocessing stage. However, since this is a full 3-D model, all levels of receiver locations are computed which requires longer computation time than 2-D MRFDPF. The timing for the 3-D IRLA model, at least for this indoor scenario, is still within an acceptable range (less than 3 minutes) where it can be used to fully predict 3-D propagation mechanism such as PDP, DS. The accuracy before calibration via the 3-D IRLA model is around 6 dB by using standard parameters and this can be improved so that a similarly high accuracy can be obtained (3.5 dB RMSE). The 3-D IRLA model is not overly reliant on exact materials, whereas this is of critical importance in ensuring high accuracy for MR-FDPF models. The 2.5-D IRLA model, computes the N-LOS rays in horizontal plane, which is less time-consuming than the 3D IRLA model. The running time, compared to the full 3-D IRLA model, is shortened to less 10 than seconds in this scenario. The accuracy provided with the 2.5-D IRLA model is still acceptable, with the RMSE equal to 4.9 dB. Therefore, the 2.5-D IRLA model may be used as a compromise between speed and accuracy in indoor wireless network planning and optimisation. Figs. 7 and 8 plot the rays generated by the 2.5-D and 3-D IRLA model. As 2.5-D IRLA only computes the NLOS rays in the horizontal plane, it is faster but does not consider diffractions rays or reections rays in the vertical plane (such as ground reection rays).
Fig. 7. rays generated in the 3-D IRLA model.
Fig. 8. rays generated in the 2.5-D IRLA model. 3-D IRLA 3.5 < 60 MR-FDPF 3.5 <5 Multi Wall 5.6 <1 2.5-D IRLA 4.9 < 10
RMSE (dB) Time (s)
Tab. 2. Performance comparison.
Fig. 9 plots the tting curves of PDF (Probability Density Function) of prediction errors in dB. It can be seen that the 3-D IRLA model and MR-FDPF gives the highest accuracy (higher probability with small errors) while COST231 Multi-wall and 2.5-D IRLA models yield the similar accuracy in this indoor scenario. The COST231-Multi Wall model [25], is extremely computational efcient and this model also does not require preprocessing. In this scenario, this semi-empirical model obtains high accuracy, which is mainly because there are few walls to penetrate. It is easy to calibrate with the losses for each wall and oors. Therefore, an agreement can be observed. However, the performance of this model is limited due to the absence of capturing reection and diffraction rays. For example, in a corridor where diffractions dominate, COST231-Multi Wall model will fail. The running time for this model is usually less than 1 second, and the accuracy obtained generally depends on the scenarios. By comparing these four indoor models, a recommendation for their use can be given. COST231-Multi Wall is efcient and is suitable for use when an estimation of indoor coverage is required on an less complex building structure such as the scenario presented in this article. MR-FDPF should have high accuracy because it incorporates radio wave propagation physics (a differential solver of Maxwell equations) but, as 3-D MR-FDPF is time and memory con-
405
0.11 0.1 0.09 0.08 0.07 Density 0.06 0.05 0.04 0.03 0.02 0.01
MRFDPF 2.5D IRLA COST231 MultiWall 3D IRLA
20
15
10 5 Prediction error (dB)
10
Fig. 9. Fitting curves of PDF of errors.
suming and still under investigation, the 2-D MR-FDPF is suitable only for indoor structures in which most propagation phenomena take place horizontally in the 2-D plane. Thus, it is not suitable in multi-oor propagation simulation, where a full 3-D model is required. However, 2-D MR-FDPF is capable of providing high accuracy on a oor after the calibration from measurements to correctly model the material properties. 3-D IRLA does not rely on calibration and is useful in prediction for multi-oor indoor structures or complex, large indoor areas. If there are no measurements, IRLA is preferred because it can be used to nd coverage gaps which may not be practically feasible for 2-D MR-FDPF and COST231 Multi Wall models. The 2.5-D IRLA generally gives a high level of accuracy and multipaths (2.5-D NLOS rays and 3-D LOS rays). The advantage of using 2.5-D IRLA model is that it provides an acceptable level of accuracy but within a much shorter time than the full 3-D IRLA model. Therefore, the 2.5-D IRLA model is suitable in some indoor applications, such as DAS planning and optimisation.
The full 3-D rays/prediction matrix are computed. It does not require preprocessing. It is fast compared to standard ray tracing methods. Compared with the 3-D IRLA model, the advantages of the 2.5-D IRLA model are: The 2.5-D IRLA model is even faster, providing a similar execution speed to empirical models. Therefore it is possible to test many indoor network congurations within a short amount of time. The oor separation and horizontal NLOS rays computation further improve the efciency and it may improve the efciency of parallelism. For example, further work includes the investigation of prediction errors in NLOS cases for some locations. It is also useful in validating the delay spread prediction via the IRLA model through measurements.
5. Conclusion and Perspectives

This article describes an extended ray launching model, IRLA, which was originally designed for outdoor scenarios. A full indoor scenario (a typical ofce) is chosen to validate the performance of this model. Comparisons with several recommendations were made. Compared with other models, the advantages of the 3D IRLA model are: It offers an accuracy similar to existing deterministic tools.
Acknowledgements
This work is supported by the EU-FP7 iPLAN and FP6 GAWIND under grant number MTKD-CT-2006-042783 (Marie Curie Fellowship for Transfer of Knowledge). The authors would like to thank Malcom Foster for his valuable corrections of this article. Also, acknowledgements have to be extended to the iBuildNet tool from Ranplan Wireless Network Design Ltd., UK.
406
Measurement 2.5D IRLA Power (dBm) 40 60 80
20
40
60
80 100 120 Measurement index
140
160
180
200
Power (dBm)
20 40 60 80 0 20 40 60 80 100 120 Measurement index 140 160
Measurement 3D IRLA
180
200
Measurement MRFDPF Power (dBm) 20 40 60 80 0 20 40 60 80 100 120 Measurement index 140 160 180 200
Measurement COST231MultiWall Power (dBm) 20 40 60 80 0 20 40 60 80 100 120 Measurement index 140 160 180 200
Fig. 10. Accuracy comparison.
References
[1] CORRE, Y., LOSTANLEN, Y. Three-dimensional urban EM wave propagation model for radio network planning and optimization over large areas. IEEE Transactions on Vehicular Technology, 2009, vol. 58, no. 7, p. 3112 - 3123. [2] LAI, Z., BESSIS, N., De La ROCHE, G., SONG, H., ZHANG, J., CLAPWORTHY, G. An intelligent ray launching for urban propagation prediction. In The Third European Conference on Antennas and Propagation EuCAP 2009. Berlin (Germany), 2009, p. 2867 - 2871. [3] HASLETT, C. Essentials of Radio Wave Propagation. Cambridge (UK): Cambridge University Press, 2008. [4] MATHAR, R., REYER, M., SCHMEINK, M. A cube oriented ray launching algorithm for 3D urban eld strength prediction. In IEEE International Conference on Communications ICC 2007. Glasgow (UK), 2007. [5] ITU-R P.1238-6 Propagation Data and Prediction Methods for the Planning of Indoor Radiocommunication Systems and Radio Lo-
cal Area Networks in the Frequency Range 900 MHz to 100 GHz. Geneva: ITU, 2009. [6] SEYBOLD, J. Introduction to RF Propagation. New York: Wiley, 2000. [7] WOLFLE, G., WAHL, R., WERTZ, P., WILDBOLZ, P., LANDSTORFER, F. Dominant path prediction model for indoor scenarios. In German Microwave Conference (GeMIC) 2005. Ulm (Germany), 2005, p. 176 - 179. [8] WOLFLE, G., GSCHWENDTNER, B., LANDSTORFER, F. Intelligent ray tracing - a new approach for the eld strength prediction in microcells. In IEEE Vehicular Technology Conference. Phoenix (AZ, USA), 1997, p. 790 - 794. [9] DEGLI-ESPOSTI, V., FUSCHINI, F., VITUCCI, E., FALCIASECCA, G. Speed-up techniques for ray tracing eld prediction models. IEEE Transactions on Antennas and Propagation, 2009, vol. 57, no. 5, p. 1469 - 1480. [10] ZAKHAROV, P., DUDOV, R., MIKHAILOV, E., KOROLEV, A., SUKHORUKOV, A. Finite integration technique capabilities for in-
407
door propagation prediction. In 2009 Loughborough Antennas & Propagation Conference. Loughborough (UK), 2009, p. 369 - 372. [11] KLEPAL, M. Novel Approach To Indoor Electromagnetic Wave Propagation Modeling. PhD thesis. Prague: Czech Technical University In Prague, 2003. [12] NAGY, L., DADY, R., FARKASVOLGYI, A. Algorithmic complexity of FDTD and ray tracing method for indoor propagation modelling. In The Third European Conference on Antennas and Propagation EuCAP 2009. Berlin (Germany), 2009. [13] GSCHWENDTNER, B. E., WOLFLE, G., BURK, B., LANDSTORFER, F. Ray tracing vs. ray launching in 3-d microcell modelling. In First European Personal and Mobile Communications Conference (EPMCC). Bologna (Italy), 1995, p. 74 - 79. [14] LAI, Z., BESSIS, N., KUONEN, P., De La ROCHE, G., ZHANG, J., CLAPWORTHY, G. A performance evaluation of a grid-enabled object-oriented parallel outdoor ray launching for wireless network coverage prediction. In The Fifth International Conference on Wireless and Mobile Communications. Cannes/La Bocca (France), 2009, p. 38 - 43. [15] LAI, Z., BESSIS, N., De La ROCHE, G., KUONEN, P., ZHANG, J., CLAPWORTHY, G. The development of a parallel ray launching algorithm for wireless network planning. International Journal of Distributed Systems and Technologies, IGI, 2010, vol. 2, no, 2, p. 1 - 2. [16] NGUYEN, T., KUONEN, P. Programming the grid with POP-C++. Future Generation Computer Systems, 2007, vol. 23, no. 1, p. 23 30. [17] LAI, Z., BESSIS, N., De La ROCHE, G., KUONEN, P., ZHANG, J., CLAPWORTHY, G. A new approach to solve angular dispersion of discrete ray launching for urban scenarios. In 2009 Loughborough Antennas & Propagation Conference. Loughborough (UK), 2009, p. 133 - 136. [18] LAI, Z., BESSIS, N., De La ROCHE, G., KUONEN, P., ZHANG, J., CLAPWORTHY, G. On the use of an intelligent ray launching for indoor scenarios. In The Fourth European Conference on Antennas and Propagation EuCAP 2010. Barcelona (Spain), 2010, p. 1 - 5. [19] LAI, Z., BESSIS, N., De La ROCHE, G., KUONEN, P., ZHANG, J., CLAPWORTHY, G. The characterisation of human-body inuence on indoor 3.5 GHz path loss measurement. In Second International Workshop on Planning and Optimization of Wireless Communication Networks. Sydney (Australia), 2010. [20] UMANSKY, D., De La ROCHE, G., LAI, Z., VILLEMAUD, G., GORCE, J., ZHANG, J. A new deterministic hybrid model for indoor-to-outdoor radio coverage prediction. In The Fifth European Conference on Antennas and Propagation EuCAP 2011. Rome (Italy), 2011, p. 3771 - 3774. [21] De La ROCHE, G., GORCE, J., ZHANG, J. Optimized implementation of the 3D MR-FDPF method for indoor radio propagation predictions. In The Third European Conference on Antennas and Propagation EuCAP 2009. Berlin (Germany), 2009. [22] De La ROCHE, G., FLIPO, P., LAI, Z., VILLEMAUD, G., ZHANG, J., GORCE, J. Combination of geometric and nite difference models for radio wave propagation in outdoor to indoor scenarios. In The Fourth European Conference on Antennas and Propagation EuCAP 2010. Barcelona (Spain), 2010. [23] De La Roche, G., FLIPO, P., LAI, Z., VILLEMAUD, G., ZHANG, J., GORCE, J. Combined model for outdoor to indoor radio propagation. In 10th COST2100 Management Meeting, TD(10)10045. Athens (Greece), 2010.
[24] LAI, Z. The Development of An Intelligent Ray Launching Algorithm for Wireless Network Planning. PhD thesis. University of Bedfordshire, 2010. [25] European Cooperation in the Field of Scientic and Technical Research. Digital Mobile Radio Towards Future Generation Systems, COST231 nal report. [Online] Available at: http://www.lx.it.pt/cost231/. [26] GORCE, J.-M., JAFFRES-RUNSER, K., ROCHE, G. D. L. Deterministic approach for fast simulations of indoor radio wave propagation. IEEE Transactions on Antennas and Propagation, 2007, vol. 55, no. 3, p. 938 - 942. [27] VALCARCE, A., ROCHE, G. D. L., NAGY, L., WAGEN, J.-F., GORCE, J.-M. Finite difference methods: A new trend in propagation prediction. IEEE Vehicular Technology Magazine, Special Issue on Trends in Mobile Radio Channels, 2011. [28] ROCHE, G. D. L., JAFFRES-RUNSER, K., GORCE, J.-M. On predicting in-building WiFi coverage with a fast discrete approach. International Journal of Mobile Network Design and Innovation, 2007, vol. 2, no. 1, p. 3 - 12. [29] GORCE, J.-M., VILLEMAUD, G., FLIPO, P. On Simulating Propagation for OFDM/MIMO Systems with the MR-FDPF Model. In Proceedings of the Fourth European Conference on Antennas and Propagation EuCAP 2010. Barcelona (Spain), 2010, p. 1 - 5. [30] Wiplan Propagation Tool. [Online] INRIA ARES/CITI Laboratory, Lyon (France). Available at: http://wiplan.citi.insa-lyon.fr .
About Authors. . .
Zhihua LAI is a research fellow (since 2010) at Ranplan Wireless Network Design Ltd., United Kingdom. He completed a B.Sc. (Honours, First Class) and Ph.D. at the University of Bedfordshire in 2006 and 2010, respectively. He was also a visiting scholar at GRID and Ubiquitous Computing Group, University of Applied Sciences of Fribourg, Switzerland in 2009 when he developed parallel distributed radiowave propagation models. His main research interests include radiowave propagation modelling and distrbuted/parallel algorithms. He has published over 10 papers and has been involved in a number of funded European projects in these areas. Guillaume DE LA ROCHE is a senior research fellow (since 2007) at the Centre for Wireless Network Design (CWiND), United Kingdom. Earlier he was with Inneon (2001-2002, Germany), Sygmum (2003-2004, France) and CITI Laboratory (2004-2007, France). He was also a visiting researcher at DOCOMO-Labs (2010, USA) and AxisTeknologies (2011, USA). He holds a Dipl-Ing from CPE Lyon, and a M.Sc and Ph.D. from INSA Lyon. He is the principal investigator of European FP7 project CWNetPlan on combined indoor/outdoor wireless network planning. Nik BESSIS is the Head of Distributed and Intelligent Systems research group and a full Professor of Computer Science in the School of Computer Science and Maths, University of Derby, UK. Nik is also associated with the Department of Computer Science and Technology at the University
408
of Bedfordshire, UK. His research interests include grids, clouds, crowds and collective intelligence. He is involved in and leading a number of funded projects in these areas. He has published 100 papers and 3 books, served as a committee member and a reviewer, a conference and/or workshop chair and the Editor-in-Chief of the International Journal of Distributed Systems and Technologies. Pierre KUONEN obtained a Master degree in electrical engineering from the Swiss Federal Institute of Technology (EPFL) in 1982. After six year of experience in industry he joined the Computer Science Theory Laboratory at EPFL in 1988 and started working in the eld of parallel computing. He received his Ph.D degree in 1993. Since 1994 he steadily worked in the eld of parallel and distributed computing. First, at EPFL where he founded and managed the GRIP (Parallel Computing Research Group), then at the University of Applied Sciences Western Switzerland (HES-SO). Since 2003 he is a full professor at the HES-SO of Fribourg at the institute of Information and Communication Technologies (TIC) where he is leading the GRID & Cloud Computing Group. Gordon CLAPWORTHY received a BSc (Honours, First Class) in Mathematics and a Ph.D. in Aeronautical Engineering from the University of London, and an MSc, with Distinction, in Computer Science from The City University, London. He is currently Professor of Computer Graphics in the Department of Computer Science & Technology and Head of the Centre for Computer Graphics and Visualization (CCGV) at the University of Bedfordshire, UK. He has
nearly 200 publications, and his interests are medical visualization, computer animation, biomechanics, virtual reality, surface modelling and fundamental computer graphics algorithms. He is a member of ACM, ACM SIGGRAPH and Eurographics. Dibin ZHOU is a research fellow (since 2010) in University of Bedfordshire, United Kingdom. He nished BSc and Ph.D. in Zhejiang University, China in 2002 and 2008 respectively. He has been a lecturer at Hangzhou Normal University since 2008. His main research interests include scientic data visualization and analysis, distrbuted/parallel algorithms, GPGPU programming. Jie ZHANG is a Professor at the Communications Group, the Department of Electronic and Electrical Engineering, University of Shefeld (www.shefeld.ac.uk). Before taking the Chair in Wireless Systems at Shefeld in Jan. 2011, from 1997 to 2010, he had been with University of Bedfordshire, Oxford University, Imperial College London and University College London etc. His research interests are focused on radio network planning and optimisation. Since 2003, as the Principal Investigator, he has been awarded 17 projects worth over 4.0 million (his share) by the EPSRC, the European Commission (FP6/FP7) and the industry etc. He was/is a Co-Investigator of two EPSRC-funded projects on femtocell (B)4G mobile communications. Since 2006, he has published over 100 papers in referred journals and conferences, over 10 of which have been widely cited. He is an author of the book - Femtocells: Technologies and Deployment (Wiley, Jan. 2010).
Comparisons of Ray Tracing Predictions and Field Trial Results for Broadband Fixed Wireless Access Scenarios.
G.E. ATHANASIADOU*, I.J. WASSELL+ *University of Peloponnese, Department of Telecommunications Science & Technology, Terma Karaiskaki, Tripoli 22100, GREECE +University of Cambridge, Laboratory for Communication Engineering, Dept. of Engineering, William Gates Building, 15 JJ Thomson Ave., Cambridge, CB3 0FD, UNITED KINGDOM
Abstract: In this paper, results from a three-dimensional (3D) ray-tracing propagation model are compared with field trial measurements. The ray tracing algorithm is optimised for Broadband Fixed Wireless Access (BFWA) operational scenarios. The model works with raster terrain as well as 3D vector building and foliage databases. It considers reflections off building walls, off-axis diffractions on roof tops, terrain and tree tops, as well as foliage attenuation. Outputs include the received power and the impulse response of the wireless channel. The field trial measurements carried out for different scenarios (propagation conditions, antenna heights, distance from the access point, etc.) in the city of Cambridge (UK) with a commercial BFWA operating system (3.5 GHz). The analysis studies the propagation characteristics of the BFWA radio channel and the accuracy of the presented deterministic propagation modelling approach, and it shows that the simulated results follow the measured with a mean error of 2.8 dB and a standard deviation of 6.7 dB. Key-words: Propagation modeling, Ray-tracing, Power Measurements, Broadband Fixed Wireless Access.
Radio spectrum is a scarce and valuable resource which must be used as efficiently as possible in order to deliver widespread wealth-creating services. This research project [1] has sought to investigate the wireless channel and hence the factors which govern how a FWA system can be optimally configured and operated in order to maximise its efficient use of bandwidth. Ray-tracing techniques have been used extensively for the modelling of the propagation characteristics in small cells [2], [3] and/or in conjunction with large cell models which operate with raster databases [4]. In microcellular environments the dominant propagation mechanisms are reflection and corner diffraction. Ray-tracing models suitable for medium sized cells [5] still only support roof diffractions after the reflection points. In this model, emphasis is given to off-axis diffractions on roof tops, terrain [6] and also, tree tops. These diffractions are fully supported both before and after reflections, since this is the dominant propagation mechanism in the situation where antennas are positioned on building tops or high on the outside walls.
I. INTRODUCTION Fixed Wireless Access (FWA) systems have the potential to provide widespread high performance broadband wireless access and are of particular interest in regions not covered by cabled broadband. FWA spectrum at 3.5 GHz represents a useful compromise between the superior range associated with the use of lower frequency bands and the wide bandwidths available at higher frequencies. In the last few years a combination of the alignment of international spectrum allocations in these bands and the availability of a new generation of high performance FWA equipment has created significant opportunities for the provision of broadband services with extensive geographical coverage. In developing countries FWA can provide an immediately available alternative to cabled networking, while in developed countries it holds the possibility of providing broadband services in rural locations.
AP image
Reflecting wall Roof ridge
AP
Vegetation (possible foliage attenuation)
A three-dimensional (3D) ray-tracing propagation model has been developed. The algorithm is optimised for Fixed Wireless Access (FWA) operational scenarios, i.e. the Access Point (AP) antenna is well above the height of the building clutter and the Customer Premises Equipment (CPE) antenna is at roof top height. The model works with raster terrain as well as 3D vector building and foliage databases. It considers reflections off building walls and off-axis diffractions from roof tops and terrain, while it traces the rays that are diffracted on top of the trees, as well as those that pass through foliage. Outputs include the received power and the impulse response of the wireless channel and as such the tool permits a detailed and realistic assessment of the performance of a broadband system, including network planning and deployment issues. Results from field trial measurements carried out in the City of Cambridge using the Cambridge Broadband Ltd network equipment are compared with simulation results produced with the ray-tracing propagation model. The field trials were carried out under the project A Study on the Efficient Dimensioning of Broadband Wireless Access Networks, which was funded by the Office Communication, UK [1].
Building structure which is a possible obstruction CPE
(a)
Possible obstruction Reflection point Vegetation
II. DESCRIPTION OF THE RAY-TRACING ALGORITHM The ray-tracing algorithm is written in the C++ programming language, while MATLAB code is used to drive the model and also display the output results. The main steps of the program are: Loading and processing of the building, foliage and terrain databases
AP
CPE
(b)
This ray is diffracted on top of a building
AP antenna
Valid reflections (since reflection heights within the limits of the reflecting wall)
Reflecting wall
This ray is diffracted on top of the tree
CPE antenna
Terrain height along ray path
Possible diffraction point for lower CPE antenna
Foliage attenuation
A database containing all terrain, foliage and building data are stored as a file in .DXF format. The first time the tool reads the DXF file it extracts all useful information and extensive pre-processing takes place so that the model can use the data in an efficient way. Consequently, the map is divided in sub-areas, and all buildings and trees in each subarea are classified by their map position to permit fast access. The terrain is processed as 10 m by 10 m tiles rather than points, and the representative vectors (i.e., a vertical vector in the middle of each tile) are saved. This enhances the speed of the calculation of the terrain points along a particular path. Creation of the AP image map For a given AP position, the image tree is created, i.e., the mirror images of the AP with respect to the building walls. In a similar way, the AP images are used to create their own images, which will eventually yield paths owing to second and higher orders of reflection. During this process the heights of the reflection walls and the AP are taken into account, as well as the actual position, orientation and size of each wall in order that only images that can generate feasible reflections are considered, and hence limit significantly the size of the image tree. Also, the reflections from the inner side of the walls are not considered. The
(c)
Figure 1: The ray tracer: (a) The image map of the AP is generated for all possible reflectors despite possible building/terrain/vegetation obstructions. (b), (c) When an image is illuminating the examined CPE, the vertical profile along the path is produced. The reflected paths are valid if the height of the reflection point is within the limits of the reflecting wall.
the path from the AP to the CPE antenna are calculated. All buildings and trees in the database along this path are also found, and the x-y coordinates and the heights of the exact points of intersection are evaluated (Figure 1). Hence the vertical profile along the path is generated; this is used to find whether there is a LOS path between the two antennas, or in the case of obstructions, the exact coordinates of the diffraction points on buildings, terrain and/or trees. Rays that pass through foliage, as well as those diffracted on top of the trees are considered. The points that the ray path intersects with foliage, and from that, the path length inside foliage, are also calculated. After the direct rays, the reflected rays are traced. The algorithm goes through the image tree to find those AP images that illuminate the specific CPE antenna under consideration. Each such image is examined in order to determine whether it can produce valid paths. First, the reflection points are found and then, the vertical profile along the whole path is generated, i.e., the terrain, building and foliage points along this ray path. From the vertical profile, the algorithm calculates the exact points of diffractions and the heights of the reflections (Figure 2). As with the direct ray, both the rays that pass through foliage, and those diffracted on top of the trees are considered. (a) Field calculations When the exact geometry of a ray is found, the calculation of its field strength starts. The angles of arrival (azimuth and elevation) at the AP and CPE antennas are calculated and the respective antenna gains are considered during the field calculations. From the analysis for the direct ray, the orientation of the CPE antenna is determined, so that the CPE antenna is pointing towards the serving AP (the direction of maximum power in the vast majority of cases). The model computes the diffraction and reflection coefficients along each path as a function of the incident and departing angles, using Uniform Theory of Diffraction and Geometrical Optics. The algorithm performs full calculation of the vector electromagnetic field and the depolarization of the power is taken into account. Hence, both linear and circular polarization can be modelled, as well as the reception between antennas with different polarization. The foliage attenuation as a function of the path length inside vegetation is also considered. The foliage attenuation models incorporated are the ones recommended by COST-235 [7] for the cases of trees in and out of leaf, as well as models which suggest that the foliage attenuation is directly proportional to the path length in trees [8]. Channel characterisation When the algorithm has gone through all the images in the image tree, found the valid paths for the specific CPE antenna position, and computed their power and time delay, it then has the complete power delay profile of the radio channel. From that it calculates the total received power, RMS delay spread, mean and excess delay, and the K-factor. In order to avoid misleading results from
(b)
Figure 2: 2D, (a) and 3D, (b) ray tracing visualisation
obstructions from other buildings or terrain result in rooftop and terrain diffractions. Path tracing After the image tree of the AP is created (or loaded from a previous study), the path tracing can begin. For each CPE position, the heights of the terrain points along
relatively weak but long delayed rays, all delay parameters are calculated within a 30 dB power window relative to the strongest ray. COMPARISON OF PRELIMINARY MEASURED AND SIMULATED
RESULTS
threshold equal to the measurement noise floor (-98dBm) was employed for the simulated power results. The digital geographical database of the simulated area included ~180000 building vectors and 160000 vegetation vectors for a total area of 31 km2. Rays with up to two reflections and unlimited roof diffractions were considered for the simulation analysis. Also, the foliage attenuation was calculated with the COST model for vegetation in leaf. Results: A total of 375 measurements were considered at the CPE positions shown in Figure 3 at a number of different antenna heights.
Set-up of measurement campaign [1]: Access Point position: On top of tall buildings and hills (as marked in Figure 3) 4x1 patch antenna, 90o horizontal beamwidth 16 dBi 3x3 patch antenna pointing towards the AP, 23o horizontal beamwidth 13 dBi on a telescopic mast on top of a van, pointing towards the APantenna Right Hand Circular 29 dBm 3 dB 3.5 GHz
Access Point Antenna: AP antenna directivity: CPE Antenna:
CPE antenna directivity: CPE antenna:
Polarisation: Transmitted Power: Losses: Frequency:
(a)
Anglia Polytechnic University
Addenbrooks Hospital
Lime Kiln Hill
Figure 3: APs and measurement positions.
(b)
Figure 4: (a) Measured vs. simulated power results. (b) Field trial (red line with x) and ray-tracing results (blue line with dots) as a function of distance from AP.
Measurements outside or at the edge of the simulated building database were not considered. Simulation results were averaged over five positions (the GPS location and four positions, 2m away and at the same height). Also, a
Figure 4 (a) shows the measured versus the simulated power results for the various CPE positions and (b) shows both of these results as a function of distance from the AP. The simulated results follow the trend of the measured. The calculated mean error and standard deviation are 2.8 dB and 6.7 dB, respectively. The field trial results show a higher variance due to the limited spatial averaging. Unlike measurements for mobile communications where results are usually averaged over distances of many wavelengths [2], for these measurements the antenna is positioned on a telescopic mast on top of a van which cannot be moved during the measurement procedure. Hence, the measured results are only time averaged. This confined averaging is a limiting factor in order to achieve better agreement between the measured and simulated results. CONCLUSIONS- FUTURE WORK In this paper, results from a 3D ray-tracing propagation model for Broadband Fixed Wireless Access systems were compared with field trial measurements. The measurements carried out for different scenarios (propagation conditions, antenna heights, distance from the access point, etc.) in the city of Cambridge (UK) with a commercial BFWA operating system at 3.5GHz. The analysis showed that the simulated results followed the measured with a mean error of 2.8 dB and a standard deviation of 6.7 dB. In the longer term, the ray-trace approach will permit realistic assessment of system performance, network planning and deployment issues. ACKNOWLEDGEMENTS The field trials were funded by the Office Communication, UK [1]. Georgia Athanasiadou would like to thank Dr Malcolm Sellars from Cambridge Broadband Ltd for his valuable help and support throughout this project. REFERENCES [1] A Study on the Efficient Dimensioning of Broadband Wireless Access Networks, Cotares Ltd., Cambridge University Engineering Dept., Cambridge Broadband Ltd, Ofcom Ref: AY4463, http://www.ofcom.org.uk/research/industry_market_res earch/technology_research/ses/ses2003-04/ay4463a/ [2] G.E.Athanasiadou, A.R.Nix, J.P.McGeehan, A Microcellular Ray-Tracing Propagation Model and Evaluation of its Narrowband and Wideband Predictions, IEEE Journal on Selected Areas in Communications, Wireless Communications series, vol 18, no 3, pp. 322-335, March 2000. [3] G.E.Athanasiadou, A.R.Nix, A novel 3D Indoor RayTracing Propagation Model: The Path Generator and Evaluation of Narrowband and Wideband Predictions,
IEEE Transactions on Vehicular Technology, vol. 49, No 4, July 2000, pp. 1152-1168. [4] G.E.Athanasiadou, E.K.Tameh, A.R.Nix, Channel Impulse Three-dimensional Rural-Urban Simulator (CITRUS): An integrated micro-macro ray-based model which employs raster and vector building databases simultaneously, IEEE PIMRC 99, Osaka, Japan, September 1999. [5] G.Liang, H.L.Bertoni, A New Approach to 3D Ray Tracing for Propagation Prediction in Cities, IEEE Trans. Antennas and Propagation, vol.AP-46, pp. 853863, 1998. [6] G.E.Athanasiadou, I.J. Wassell, C.L. Hong, Deterministic propagation modelling and measurements for the broadband fixed wireless access channel IEEE VTCF 2004, Los Angeles, USA, 26-29 September 2004. [7] COST235, Radiowave Propagation Effects on Next Generation Terrestrial Telecommunications Services, Final Report, 1996. [8] G.Durgin, T.S.Rappaport, H.Xu, Measurements and Models for Radio Path Loss and Penetration Loss In and Around Homes and Trees at 5.85GHz, IEEE Transactions on Communications, vol.46, no 11, November 1998.
112
IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 1, NO. 1, JANUARY 2002
A New Statistical Model for Site-Specific Indoor Radio Propagation Prediction Based on Geometric Optics and Geometric Probability
Mudhafar Hassan-Ali, Member, IEEE and Kaveh Pahlavan, Fellow, IEEE
AbstractThe ray-tracing (RT) algorithm has been used for accurately predicting the site-specific radio propagation characteristics, in spite of its computational intensity. Statistical models, on the other hand, offers computational simplicity but low accuracy. In this paper, a new model is proposed for predicting the indoor radio propagation to achieve computational simplicity over the RT method and better accuracy than the statistical models. The new model is based on the statistical derivation of the ray-tracing operation, whose results are a number of paths between the transmitter and receiver, each path comprises a number of rays. The pattern and length of the rays in these paths are related to statistical parameters of the site-specific features of indoor environment, such as the floor plan geometry. A key equation is derived to relate the average path power to the site-specific parameters, which are: 1) mean free distance; 2) transmission coefficient; and 3) reflection coefficient. The equation of the average path power is then used to predict the received power in a typical indoor environment. To evaluate the accuracy of the new model in predicting the received power in a typical indoor environment, a comparison with RT results and with measurement data shows an error bound of less than 5 dB. Index TermsPower coverage, power delay profile, probabilistic geometry, rat tracing, site-specific channel model, statistical indoor radio propagation, wireless deployment tool.
I. INTRODUCTION
E ARE living with ever increasing demand on telecommunications speed and ubiquity. The advent of the Internet and data networks has escalated this demand. The mobility and ease of installation make wireless communication networks one of the most important communication systems to deploy. Personal communications systems (PCS), wireless local area networks (WLANs), wireless private branch exchanger (WPBXs), and Home Phoneline Network Alliance (HomePNA) are the services that are being deployed in indoor areas on an increasing scale. The latter application is proving to have a large market since it will be integrated to the emerging Digital Subscriber Loop technologies (ADSL, VDSL, etc.). The market of these services will try to reach out to offices, schools, hospitals, and factories [10], [12]. Because the indoor radio
Manuscript received December 1, 1999; revised February 1, 2001; accepted March 7, 2001. The editor coordinating the review of this paper and approving it for publication is R. Valenzuela. M. Hassan-Ali is with the Systems Engineering, Alcatel USA, Petaluma, CA 94954 USA. K. Pahlavan is with the CWINS, Worcester Polytechnic Institute, MA 01609 USA. Publisher Item Identifier S 1536-1276(02)00185-X.
channel has a tremendous amount of impairment and variability [1], [5], [6], large-scale deployment of these services provides a major challenge to the network designers. For this reason, it is imperative to develop deployment tools, where efficient but accurate radio channel models are required. The efficiency of a model is measured by the computational complexity, whereas accuracy is measured by the estimation error. Ray-tracing (RT) [1], [14], [20] is one of the most popular techniques for predicting radio channels used in the deployment tools. The main characteristic of the RT is the computational intensity, which is the main reason for the prediction tools to be slow in spite of its accuracy compared to the tools based on the statistical model. This has motivated a significant research effort to pursue alternative methods including the so-called Fast RT [2], [21] in an attempt to expedite the computation time. Still these alternative methods require more complex floor-plan databases and the need to trace all rays regards of their significance to the received power. The purpose of this paper is to introduce a new model for statistically predicting the indoor radio propagation in order to contrive a more computationally efficient method for predicting the received power within a building. The paper is organized as follows. Section II states the theory behind the new model and presents a key equation for estimating path power. Section III shows a method whereby the total received power can be estimated. In Section IV, the prediction of indoor radio power using the new model is compared to the prediction of RT software and data collected from measurements for a typical office environment. II. POWER OF A PATH WITH A GIVEN LENGTH RT approximates the radio propagation in a finite number of rays originated from the transmitter. Each ray encounters reflection and transmission upon intersecting with an obstacle (such as walls, doors, windows, etc.) The pattern of ray propagation is dictated by the geometry of the floor layout and the materials from which these obstacles are made. Hence, as an alternative, the statistical characterization of radio propagation can be related explicitly to the statistics of these patterns [4]. The statistical features of the propagation can be deduced directly from the layout and the materials of the floor under consideration. The purpose of this section is to relate the path power to the key site-specific propagation parameters. The path power relationship will be used in Section III to predict the received power.
15361276/02$17.00 2002 IEEE
HASSAN-ALI AND PAHLAVAN: STATISTICAL MODEL FOR SITE-SPECIFIC INDOOR RADIO PROPAGATION PREDICTION
113
A. Path Power and the Number of Reflections and Transmissions When a path arrives at a point, it has already gone through many reflections and transmissions (object-intersections). Consequently, the path power tends to decay rapidly with distance more than the inverse-square distance law for the free-space. Each path is traced throughout its entire trip from the transmitter to the receiver. Each time there is an object-intersection the ray loses a certain amount of power while the propagation loss in between intersections will maintain the free-space rate, i.e., inverse-square distance law. The intersection loss is either due to reflection or transmission, since other mechanism, such as diffraction and diffused scatter, can be ignored in indoor propagation [9]. Each loss can be expressed in power formulation as a multiplication by a loss coefficient. Hence, after traveling meters from the transmitter (Tx) and undergoing intersections transmissions), the path power is ex( reflections and pressed (1) where and are the mean voltage reflection and transmisis the free-space power at dission coefficients, respectively, tance 1 meter, which is expressed by
Fig. 1. The rectangular model used to find PDF of q and p.
Where and ( for isotropic antenna) are gain of is the speed transmit and receive antennas, respectively, of light in free space, and is the frequency of the radio signal, which is 900 MHz in this paper. For rest of this paper, the assumption is that the transmit and receive antennas are isotropic; i.e., omni-directional propagation. The mean path power can be expressed as follows: (2)
is the mean free distance between two intersections, where which depends on the floor layout Mean Free Distance. It is defined as the mean distance a ray can travel before it intersects with an object. This parameter is estimated within a given shape, which is assumed to be rectangular due to the adoption of the rectilinear model. In Section III-C, a method for estimating this parameter will be presented using probabilistic techniques. The from knowing the width and length of the method estimates , rectangles of the floor plans. The second function on the other hand, gives the probability of having exactly reflections and transmissions in path length . As mentioned earlier, these are independent and exclusive, hence binomial PDF fits these conditions [18]. Then (5)
is the PDF of a path that intersects objects where transafter traveling distance with reflections and missions. In the following section, this PDF will be discussed in detail. B. Calculation of One can think of the process of hitting obstacles as a combination of reflections and transmissions. These two events are independent and exclusive in one path at one instance. Hence, can be decomposed as a multiplication of two functions (3) is the PDF for a path that has undergone inwhere tersections after traveling distance . In [13], it has been demonstrated through a Monte Carlo simulation that this function is a Poisson distribution for the indoor environment. Hence (4)
and are the probabilities of reflection and transwhere mission, respectively, for a path of length . Note that . After a few manipulations on (2) we obtain the following results (see Appendix A for derivation): (6) This equation gives an explicit relationship between the average power of a path with site-specific details and the building layout via , and the floor materials via ( and ). By estimating the values of these parameters based on the location of both transmitter and receiver, (6) can be applied to predict the power of a path versus distance. C. Calculation of and
To use (6) for predicting the power of a multipath arrival knowing the location of the transmitter and the receiver, it is important to know how and change with path distance . In order to do that, let denote the TransmitterReceiver distance, therefore, , where is the difference between
114
Fig. 2.
The probability distribution of reflection (p).
the total path length and TransmitterReceiver distance. Hence , because LOS can not undergo any reflections. For large excess path lengths, reflection and transmission events are taken to be equally likely; i.e., approaches 0.5. Therefore, one can conjecture that this behavior can be exponential, i.e.,
D. General Formula for Path Power Substituting (7) in (6) yields (8) Note serves as the time delay of the profile since where is the speed of light. Hence ,
(7) A Monte Carlo simulation has supported this conjecture where the rectangular shape model is employed. The simulation can be summarized as follows: The rectangular model of a floor plan is taken to be 10 5 with 50% uniform randomness in both where length and width. This means that the width is is a uniform random variable in the range of , where is a uniform random variable and the length is . A numerous number of rays that have in the range of undergone through intersections are generated. For each ray, the type of intersection (reflection or transmission) is recorded at each intersection as seen in Fig. 1. The reach and length of each path are then computed, where the reach of a path is the direct distance between Tx and Rx , while the path is equivalent to . Hence, and are is estimated. assigned for each from which the PDF of Fig. 2 shows the result of this simulation. In this figure, both the PDFs derived from simulation and the best exponential fit estimated is very close are plotted together. The value of to the mean free distance of a rectangle with the dimension of 10 5 as can be calculated using the formulas presented in Section III-C. (9) This equation represents an average power delay profile for indoor radio channel. To visualize the significance of the parameters to the shape of the profile, Fig. 3(a)(d) show profiles where one parameter is made variable while the others are held constant. The most influential parameter is the TxRx distance, whereas ranks second. and have a roughly similar effect. , then (9) will give the expected value of Note that when the power for the LOS ray (10) Clearly, LOS power is inversely proportional to squared TxRx distance (free-space component), and exponentially is the to transmission loss in this distance. Note that average transmission loss (no Reflection coefficient exists is the mean number of transmission since it is LOS), and occurrences within . Since the LOS ray is nothing more than the line drawn between Tx and Rx, this portion of signal power can be replace by the deterministic power calculation. If there
115
Fig. 3.
The model parameters and their effect of predicted channel profiles.
are object-intersections along this line, then the correction to (8) and, therefore, to (9) will be as
No close form could be found for this integral, thus, it has to be computed numerically as (12b)
(11)
III. THE MULTIPATH RECEIVED POWER In this section, the total power received from multiple paths will be estimated based on the key equation derived in the previous section. For a wideband receiver, the multipath power is simply defined as the sum of the their individual power regardless of the phase of the individual paths [19]; i.e., (12a)
where is the bin time unit. In (12b), it is assumed that a path exists in each bin, which is 5 ns in this case, since the bandwidth is 200 MHz. The result from (12b) will be compared to the power estimation using RT results as well as measurement data. In the following two subsections, methods for determining the three parameters ( , , and ) are presented. A. The Area Where , and to be Estimated
For a given pair of (Tx, Rx), we need to identify the neighborhood; i.e., obstacles surrounding Tx and Rx, that influences the estimation of the received power by determining the mean value of , , and . To identify these obstacles, maximum path
116
length can serve as an indication of how far the inclusion of the obstacles should be. The path power is assumed to reach a threshold (say 10 dB below the strongest ray arrives at Rx) under which the ray will be neglected. The shape of area that the maximum path length traverses before its power drops below the threshold is naturally ellipsoid. Within this ellipsoid shape, the path is expected to have the lowest power when it undergoes only one reflection out of intersections. This is true on a statistical basis, since in (assuming that ). general As depicted in Fig. 4, the locations of Tx and Rx serve as the foci of the ellipse whose boundary acts as the farthest reflector on which rays bounce with the same length. The idea of confining the area of interest inside an ellipsoid shape has been presented in the literature primarily for studying the mulitpath scattering [15]. Rectilinear partitioning of the floor plan simplifies the issue of deciding which spaces that the ellipse overlaps have to be included in the estimation of the three model parameters. During rectilinear partitioning process, fictitious exten(no transmission loss) sions will be drawn and assigned (no reflection). These extensions will be included and a during the estimation of the average model parameters ( and ), see Fig. 4. If the maximum path length is denoted by , then the threshold is computed as (13a) where is the maximum power of a ray that travels from dB, Tx to Rx, which can be derived from (8). For then . One method for determining is to use the following equations: for LOS and for OLOS, where is the distance between Rx are deand Tx as seen in Fig. 4. These two equations for rived from numerous evaluations of (8) for various values of its parameters. Another alternative is to use the ad hoc model derived from measurements for the received power; such as JTC, or wall-dependent [1], as follows: (13b) where the decay parameters [1]. B. Estimation of and
Fig. 4. An illustrative example for maximum path length relative to the ellipsoid shape.
floor layout.) and expressions [1]:
can be calculated through the following
Horizontal polarization Vertical polarization (15a) (15b) where is the complex permittivity, is the relative normalized dielectric constant, is the conductivity, and is a coefficient that accounts for the transmission loss and it is usually taken to be 0.5 [9]. Note that (15a) is a function of incidence angle; which is a uniform random variable . Therefore, for a given material, one can find the over mean value of the reflection and transmission coefficient by av, i.e., and eraging (15) over . C. Estimation of The same shape used in the previous section will be used to estimate . In order to estimate the mean value of a random variable , the probability distribution function (PDF) of the ray length is required. Appendix B shows the geometric probability distributions of a ray within the rectangular shape, which was shown to have three different types of rays. The mean free path can be estimated as follows: (16)
Once the ellipsoid shape is determined, the reflection and transmission coefficients for all objects (say objects) enclosed in it will be collected. Then and are estimated as follows: and
(14) are reflection, transmission, and size cowhere , , and efficients, respectively, for object . In a two-dimensional (2-D) case, is the length of the object (such as wall length in the
where , , and denote the mean length of rays between adjacent sides (four cases), opposite width sides, and opposite
117
in proportional to its area overlapping with the ellipse, as indicated in Fig. 4. Hence [16] (18) where is the overlapping area of rectangle , whose area . The assumption is that the ellipse confines is complete rectangles. A part from the having only rectangular shapes, there could be, within the confining ellipse, parallel lines along or axis, such as a portion of a hallway. In this case, the or , mean free distance is computed as respectively. IV. VALIDITY OF THE PROPOSED PREDICTION MODEL In this section, the results of power prediction using the new model are compared to the power predicted by the RT software and the measured power in a typical office environment. The second and the third floors in Atwater Kent (AK2 and AK3, respectively) Laboratories are taken as case study to check the validity of the new model. Throughout this work, we maintained the following parameters for both cases (AK2 and AK3) [7], [8]. The center frequency of the channel is 1 GHz, and the bandwidth of 200 MHz. The number of profiles is 620 taken from different locations in the second floor at the AK Labs building. A. Comparison With the Results of RT In this section, the results of the previous subsections will be used for estimating the power in AK3 and then compared with the results obtained from the RT software. In order to do that, the three zones LOS, OLOS1, and OLOS2 are treated individually. On this floor, walls, doors, and windows are considered to be highly dielectric materials, nearly perfect conductors, and low dielectric materials, respectively. Walls are assumed to have a 10 dielectric constant and a 0.001 conductivity, therefore, and coefficients are (0.75,0.48), which are the average over the incidence angle range of . These coefficients are assumed to be (0.95,0.01) and (0.1,0.9) for doors and windows, respectively [3]. 1) LOS: In this case, all Rxs are located in the same room (number 317) where Tx is as seen in Fig. 5. This room is about m . The distance between Tx and each Rx is in the range from 0.2 to 6 m implying the ellipse for each TxRx combination embracing this room as seen in Fig. 5 and portion of the surrounding hallway (OLOS1) and Room 318. The first step is to estimate and using (14) using the reflection and transmission coefficients given above. The size coefficients are determined for walls as follows: , doors , and windows . Hence, , and . The second step is to estimate from the dimensions of Room 317 and its adjacent vicinities. OLOS1 is a hallway; which is about 2 meters wide. The ellipse only includes about 5 meters of the two parallel walls of OLOS1. Therefore, the dimensions of OLOS1 portion are about m .
Fig. 5.
AK3 floor layout.
length side. They are computed, with assistance of Appendix II, as follows:
(17a)
(17b)
(17c) is comFor rectangle in the floor, a mean free distance and width . Any rectangle inputed based on its length cluded inside the ellipse will be used to estimate the mean value
118
Notice that in this case equation (17b) or (17c) is used to find . Using (12) in conjunction with (13) yields the estimation of as seen in Table I. 2) OLOS1: Similar analysis can be done for this case. As can be seen in Fig. 5 there are left, right, top, and bottom areas whose parameters are unique because their neighbors are different. The most influential neighbor is Room 317, which possesses about 70% of the ellipse area for this case. Hence is in the range of 0.2 to 0.25 for left, right, and bottom areas. On the other side, the bottom area is shadowed deeply by Room 318. The ellipse in this area embraces parallel walls of Room 318 along the width, with size about 4 3. In this case, is about 0.3 using (17b). The upper boundary of the top area has small rooms (Room 319, Room 320, and the entrance of Room 320-CWINS.) Roughly, their sizes are in the order of 3 2; which cause to be in the range of 0.5 to 0.6. Furthermore, the ellipse does not confine Room 217 entirely, so that is higher than 0.14 as found in LOS. It is estimated to be 0.22 by using (16). Therefore, the average for the top area is in the range 0.4 to 0.6. 3) OLOS2: It consists of a row of offices, most of them of size 3 2 except Rooms 311 and 301. By inspecting Fig. 5, Rooms 301310 seem to have comparable parameter values. Room 311, on the other hand, is deeply shadowed by Room 318. If we assume that 50% of the ellipse resides inside Room 317, 25% inside LOS1, and 25% inside OLOS2. Then for the Rooms 301310, is in the range 0.25 to 0.3. Room 312, however, is in a deep shadow due to Room 318. Also, the receivers in this room are the farthest from Tx. The ellipse at the most two adjacent walls (top and right ones) causes to be calculated using (17a) when the two adjacent walls are included. Table II summarizes the parameters that will be used in (11) for all Rx locations in the AK3 experiment to estimate the received power. 4) Results of Power Prediction From the new Model and RT: Fig. 6 shows the scatter plot of power estimated using the new technique versus RT results. The similarity is apparent between the two cases indicating that the new model is a valid technique for power prediction. LOS case shows agreement to all points: i.e., the pattern of power change is very similar. OLOS1 has the same trend except some points located at the intersection of the top with both left and right area of this zone. In the case of OLOS2, the periodical decay is not as deep as the RT results. This is due to the fact that parameter is assumed equal throughout Rooms 301310. The reality is that the Rooms 308310 should have a gradual increase to this parameter to account for the gradual increase in the effect of Room 318 shadow. Generally, however, the standard deviation of the prediction with respect to RT estimation is about 5 dB over all zones. However, the standard deviation for the individual zone is as follows: 1.2 dB in LOS, 5.9 dB in OLOS1, 5.5 dB in OLOS2. This is a remarkable achievement considering the fact that the piecewise-linear statistical power modeling [1] had a standard deviation of more than 10 dB. Fig. 7 shows a scatter plot of the predicted power for the three zones. It is apparent that LOS case shows the highest match, whereas results of OLOS-2 show the lowest match.
TABLE I PARAMETER ESTIMATION FOR LOS ZONE
TABLE II MODEL PARAMETERS FOR AK3 IN THREE ZONES
Fig. 6. Power prediction using the new model and RT data versus location index at AK3 (operating frequency is 900 MHz).
When power-distance relationship is drawn with both axes are logarithmic, as seen in Fig. 8, the shape of the relation is anticipated to be slowly decaying approximately in the first 10 meters and the decay becomes much steeper [9], [11] . In [11], empirically this relationship was fit to an exponentially decaying function; i.e., , which is very close to our theoretical derivation as showed in (6).
B. Comparison Between the Results From the New Model and Measurements To compare the prediction of the received power with those obtained from measurements, the frequency-domain measure-
119
Fig. 7.
Comparison of power prediction using the new model and RT data at the three zones of AK3 (operating frequency is 900 MHZ).
Fig. 8. Power prediction using the new model and RT data versus TxRx distance at AK3 (operating frequency is 900 MHz).
ments for AK2 used in [9] will be employed here also. A similar analysis is carried out to the locations (see Fig. 9) to find the three parameters in each location. The materials of walls, windows, and doors are similar to AK3 floor mention above. The estimation of the three parameters as done in Section IV-A is repeated here, as seen in Table III. Compared to
AK3 floor, the value of is the same, whereas is slightly smaller due to the fact that this floor has more metallic doors. The receivers in Room-1, where the transmitter is located, are associated to the four surrounding spaces according to their closeness to these spaces. The fifth space consists of the receivers in Room-4.
120
TABLE III MODEL PARAMETERS FOR AK2
rectangular shape in the floor plan. This operation is performed once for each floor plan under study. 3) Locate the Transmitter (Tx) and Receiver (Rx) in the floor plan. 4) Draw an ellipse, whose foci are the locations of Tx and Rx as explained in Section III-A. 5) Find the overlapping area between the ellipse and the floor plan. This step identifies the inclusion of all rectangular shapes (rooms) that will be use in the next step. 6) Compute the average values of , , and for the overlapping area identified in the step 5) using(14) and (18). 7) Use the parameters computed in the step 6) in (11) and (12) to estimate the multipath received power. Note that steps 1) and 2) are considered as preprocessing operations and performed once for the floor plan. Steps 4) and 5) are computationally more involved than the rest of the steps in this algorithm. Specifically, step 5) requires answering the query of knowing the rectangles that the ellipse overlaps in a floor plan. queries for checking the The brute-force method results in overlap with all the rectangles in the floor plan for each TxRx pair. This complexity can be improved by using a spatial data structure [22][24] for relating the rectangles in a floor plan with each other. This data structure reduces the query time from to , which is the number of the rectangles that overlap with the ellipse in the floor plan. In practice, is determined by the TxRx distance and the size of the rectangles. The rest of the steps are straightforward and are computationally simple, since computation of , , and is performed only once for each rectangle as indicated in step 2. Furthermore, , and are the , for the average value computed over the angle range rectilinear wall model that is assumed in this paper as explained in Section III-B. There are two methods to implement RT; image technique and ray shooting [23]. The RT tool starts with shooting rays from the transmitter to all direction around. Each ray will be traced under it reaches the receiver after undergoing through wall-intersections ( reflections and transmissions). Upon each intersection, the ray splits into two child rays, a reflected ray and a transmitted ray. Hence, the number of operations in the brute force ray shooting RT is proportional to
Fig. 9.
AK3 floor layout.
The result of the power prediction is depicted in Fig. 10. The standard deviation between the prediction and measurement is about 2.87 dB and the mean error is 2.77 dB, compared to a standard deviation of 2.4 dB when using RT [3]. Fig. 11 shows the power levels obtained from three methods; i.e., measurement, RT [8], and the new model at AK2.
V. THE COMPUTATIONAL COMPLEXITY OF THE NEW METHOD COMPARED TO RT According to the new method, the multipath-received power is estimated using the following algorithm. A. Algorithm 1) Perform the rectilinear partitioning (rectangulation) of the floor plan. The result is that the floor plan is partirectangles, where is tioned approximately in to the number of walls. 2) Calculate the propagation parameters ( , , and ) as explained in Sections III-B and C. and represent the avfor each erage over the incidence angle range of wall, door, and window using (12). is computed for each
121
Fig. 10.
Power prediction using the new model and measurement data at AK2 (operating frequency is 900 MHz).
Fig. 11. Comparison between power prediction using measurement data, ray-tracing, and the new model at AK2 (operating frequency is 900 MHz).
for distinct paths and without ray splitting [23]. Using triangulation as a data structure, the ray shooting was expedited such that the query of ray-wall intersec1 1 factor accounts for that the process is repeated times for each shooting 1) factor accounts for the number of rays angle around the transmitter, (2 due to intersections (each intersection spawns 2 rays), and factor accounts searches to find the ray-wall intersection in the for walls; i.e., we require brute-force tracing.
tion can be performed in much less than operations. On the other hand, the brute force implementation of the image technique results in a computational complexity proportional to [23]. Beam tracing [24], which is a variation of image technique, is reported recently to have a complexity proportional to . In practice, the parameters and are usually assumed to be 180 and 3, respectively. These parameters, for each TxRx pair, will entail 2700 rays, each one requires power cal-
122
culation at the intersection with a wall. The new model, however, requires a number of computations [using (14) and (18)] equivalent to the number of the rectangles that overlap with the ellipse. In the two floor plans (AK2 and AK3) that this study is based on, the total number of rectangles is less than 50; which implies that the computation ratio is better than 2700 : 50. Furthermore, the new method predicts the average received power, thus, the prediction is less sensitive to the sampling artifact [24] compared to the RT method. We expect that the new model can be useful in optimal placement tools [2]. VI. CONCLUSION In this paper, a new model for indoor radio channels is presented. The model relies on the geometric probability of the layout of the indoor environment from which a simple equation for power delay profile was derived. This equation has three key parameters, which are directly related to the geometry of the floor layout and the materials of its walls, doors, and windows through simple equations. The model was, then, used to predict the power received in two office floors, AK2 and AK3 at WPI, and compared with the results obtained from running the RT software and measurement data to check the validity of the new model. It was found that the new prediction had an error bound of 5 dB respect to RT and measurement data. This model can be accepted to surrogate the use of the brute-force RT technique for prediction of radio propagation in indoor environment. The advantage will be in terms of computational simplicity when compared to RT. As a future work, this model can be extended further to include diffracted paths, which can be applied to microcellular environment where reflection and diffraction are the dominant propagation mechanisms. Furthermore, the analysis can be extended to three-dimensional geometry, which enables us to predict power coverage in multistory buildings. APPENDIX I Substituting (4) and (5) in (2) yields
Fig. 12. A rectangle with different rays between two sides.
Substituting this equation in (A-2) yields (A-4) This summation is Taylor expansion of rewritten as follows: , hence, it can be
(A-5)
APPENDIX II In this appendix, the Geometric Probability of a ray inside a rectangle will be presented. A number of PDFs will be derived and to be used for estimation of the mean free distance . Traveling inside a rectangle , a ray can be envisaged geometrically as its hypotenuse between two intersection points. Originally, this ray either came from a reflected parent ray (in), or from a transmitted parent ray coming from an side adjacent rectangle, such as . Therefore, a path from Tx to Rx can be decomposed into a set of these consecutively linked rays. From the basic literature of Geometric Probability [16][18], the probability distribution function of the ray length can be derived. The derivation of the ray length probability distribution is based on the assumption that the two terminal points of a ray are independent random variables (RV). Moreover, the coordinates of any point in a rectangle have a uniform PDF with respect to length and width of that rectangle [16][18]. Fig. 12 are typshows the possible types of rays, i.e., , , and ical examples for these rays. The probability density of the ray is split into three cases. length 1) Rays between adjacent walls, , , , and in Fig. 12: In this case the ray length is expressed as , note that , and . Then the probability distributions is expressed as [16], [17]
(A-1) By taking out all the factors that are not function of gives and
(A-2) The inner summation of (A-2) is the binomial expansion of simplified as follows:
(A-3)
(B.1)
123
2) Rays between opposite sides (width case), In this case, the ray distance is and has a PDF as follows:
in Fig. 12:
3) Rays between opposite sides (length case), This case, the ray distance is and has a PDF as follows:
(B.2) in Fig. 12.
(B.3) After knowing all the possible types of rays in a rectangle, the PDF of the length of a ray irrespective to its relation to the rectangle sides is determined. The probability of having a specific ray type for rays between adja, whereas for rays between opcent sides is . posite sides (either case) the probability is Hence, the PDF of ray length is
(B.4)
ACKNOWLEDGMENT The authors wish to thank Dr. J. Beneat and Mr. R. Tingley for reviewing the paper and their valuable remarks. Many thanks go to the IEEE JSACs reviewers for the constructive comments.
[9] H. Bertoni, W. Honcharenko, L. R. Maciel, and H. Xia, UHF propagation prediction for wireless personal communications, Proc. IEEE, vol. 82, pp. 13331359, Sept. 1994. [10] J. Keenan and A. Motley, Radio coverage in buildings, Br. Telecom Technol. J., vol. 8, pp. 1924, Jan. 1990. [11] D. Devasirvatham, C. Banerjee, M. Krain, and D. Rappaport, Multifrequency radiowave propagation measurements in the portable radio environment, in Proc. IEEE ICC90, 1990, pp. 13341340. [12] D. Molkdar, Review on radio propagation into and within buildings, Proc. Inst. Elect. Eng.-H, vol. 138, pp. 6173, Feb. 1991. [13] M. Hassan-Ali, Using Ray-Tracing Techniques in Site-Specific Statistical Modeling of Indoor Radio Channels, Ph.D. dissertation, Worcester Polytechnic Institute, Worcester, MA, 1998. [14] M. Hassan-Ali and K. Pahlavan, Site-specific wideband and narrowband modeling of indoor radio channel using ray-tracing, in PMIRC98, Boston, MA, Sept. 811, 1998. [15] R. Ertel, P. Cardieri, K. Sowerby, T. Rappaport, and J. Reed, Overview of spatial channel models for antenna array communication systems, IEEE Pers. Commun., vol. 5, pp. 1022, Feb. 1998. [16] L. Santalo, Integral Geometry and Geometric Probability. Reading, MA: Addison-Wesley, 1976. [17] B. Ghosh, Random distances within a rectangle and between two rectangles, Bull. Calcutta Math. Soc., vol. 43, pp. 1724, 1951. [18] A. Sveshnikov, Problems in Probability Theory, Mathematical Statistics and Theory of Random Functions. New York: Dover, 1968. [19] R. Valenzuela, O. Landron, and D. Jacob, Estimating local mean signal strength of indoor multipath propagation, IEEE Trans. Veh. Technol., vol. 46, pp. 203212, Feb. 1997. [20] J. McKown and R. Hamilton, Ray tracing as a design tool for radio networks, IEEE Network Mag., pp. 2730, Nov. 1991. [21] D. Ullmo and H. Baranger, Wireless propagation in buildings: A statistical scattering approach, IEEE Trans. Veh. Technol., vol. 48, pp. 947955, May 1999. [22] J. ORourke, Computational Geometry in C. Cambridge, U.K.: Cambridge Univ. Press, 1993. [23] S. Fortune, A beam-tracing algorithm for prediction of indoor radio propagation, in Proc. First ACM Workshop on Applied Computational Geometry, 1996, pp. 7681. , Algorithms for the Prediction of Indoor Radio Propagation,, [24] http://cm.bell-labs.com/cm/cs/who/sjf/pubs.html, 1998.
REFERENCES
[1] K. Pahlavan and A. Levesque, Wireless Information Networks. New York: Wiley, 1995. [2] S. Fortune, D. Gay, B. Kernighan, O. Landron, R. Valenzuela, and M. Wright, WISE design of indoor wireless systems: Practical computation and optimization, IEEE Comput. Sci. Eng., vol. 2, pp. 5869, Spring 1995. [3] G. Yang, Performance evaluation of high speed wireless data systems using a 3D ray tracing algorithm, Ph.D. dissertation, Worcester Polytechnic Inst., Worcester, MA, 1994. [4] P. Bello, A troposcatter channel model, IEEE Trans. Commun. Technol., vol. COM-17, pp. 130137, Apr. 1969. [5] A. Saleh and R. Valenzuela, A statistical model for indoor multipath propagation, IEEE J. Select. Areas Commun., vol. SAC-5, pp. 128137, Feb. 1987. [6] G. Yang, K. Pahlavan, and J. F. Lee, A 3D propagation model with ploarization characteristics in indoor radio channels, in Proc. IEEE Globcom, 1993, pp. 12521256. [7] R. Ganesh, Time domain measurements modeling and simulation of the indoor radio channel, Ph.D. dissertation, Elect. Eng. Dept., Worcester Polytechnic Inst., Worcester, MA, 1991. [8] S. Howard, Frequency domain characteristics and autoregressive modeling of the indoor radio channel, Ph.D. dissertation, Elect. Eng. Dept., Worcester Polytechnic Inst., Worcester, MA, 1991. Mudhafar Hassan-Ali (M99) was born in Baquba, Iraq, in 1962. He received the B.Sc. and M.S.c. degrees in electrical engineering from University of Baghdad (with highest honors), Bagdad, Iraq, in 1984 and 1989, respectively, and the Ph.D. degree in electrical engineering from Worcester Polytechnic Institute (WPI), Worcester, MA, in 1998. From 1984 to 1987, he was a teaching and research assistant at University Baghdad. From 1989 to 1991, he was an instructor at University of Baghdad, teaching in Electrical Engineering and Computer Science Departments. From 1992 to 1993 he was a lecturer and head of Electronics Engineering Department, University of Seven of April, Libya, From 1993 to 1996, he was with WPI, Electrical Engineering Department and CWINS, where he was involved in Radio Propagation for wireless system design research. In 1996, he joined the Wireline Access System of Alcatel USA, where at present he is a Senior System Architect and has been involved in developing Broadband system access systems based on SONET, xDSL, and IP/ATM technologies. Currently he is leading the effort in building a novel ATM switch fabric for the next generation access system. His Interests are in Communications protocols, VLSI, system architecture, and wireless design. Dr. Hassan-Ali is a member of Eta Kappa Nu and Sigma Xi.
124
Kaveh Pahlavan (M79SM88F96) is a Professor of ECE, a Professor of CS, and Director of the Center for Wireless Information Network Studies, Worcester Polytechnic Intitute, Worcester, MA. He is also a Visiting Professor of Telecommunication Laboratory and CWC, University of Oulu, Finland. His area of research is broadband wireless indoor networks. He has contributed to numerous seminal technical publications in this field. He is the principal author of Wireless Information Networks (New York: Wiley, 1995). He has been a consultant to a number companies including CNR, Inc, GTE Laboratories, Steinbrecher Corp., Simplex, Mercurry Computers, WINDATA , SieraComm, 3COM, and Codex/Motorola in Massachusetts; JPL, Savi Technologies, RadioLAN in California, Airnoet in Ohio, United Technology Research Center in Connecticut, Honeywell in Arizona; Nokia, LK-Products, Elektrobit, TEKES, and Finnish Academy in Finland, and NTT in Japan. Before joining WPI, he was the director of advanced development at Infinite Inc., Andover, MA, working on data communications. He started his career as an Assistant Professor at Northeastern University, Boston, MA. He is the Editor-in-Chief of the International Journal on Wireless Information Networks. He was the founder, the program chairman and organizer of the IEEE Wireless LAN Workshop, Worcester, MA, in 1991 and 1996, and the organizer and technical program chairman of the IEEE International Symposium on Personal, Indoor, and Mobile Radio Communications, Boston, MA, in 1992 and 1998. He has also been selected as a member of the Committee on Evolution of Untethered Communication, U.S. National Research Council in 1997, and has lead the U.S. review team for the Finnish R&D Programs in Electronic and Telecommunication in 1999. For his contributions to the wireless networks he was the Westin Hadden Professor of Electrical and Computer Engineering at WPI during 19931996, became a fellow of Nokia in 1999. From May of December of 2000, he was the first Fulbright-Nokia scholar at the University of Oulu, Finland. Because of his inspiring visionary publications and his international conference activities for the growth of the wireless LAN industry, he is referred to as one of the founding fathers of the wireless LAN industry. Details of his contributions to this field are available at http://www.cwins.wpi.edu.
UNIVERSITY OF WITWATERSRAND JOHANNESBURG
A heuristic tool for indoor radio-wave propagation prediction

Brian Whitaker
A dissertation submitted to the Faculty of Engineering and the Built Environment, University of the Witwatersrand, Johannesburg, in fulfilment of the requirements for the degree of Master of Science in Engineering. Johannesburg, January 2005
ii
Declaration
I declare that this is my own, unaided work, except where otherwise acknowledged. It is being submitted for the degree of Master of Science in Engineering at the University of the Witwatersrand, Johannesburg. It has not been submitted for any degree or examination at any other university.
Signed this ________ day of ______________ 2005
Brian Whitaker
iii
Abstract
For the effective implementation of a wireless local area network (WLAN) within a building, a complete understanding of indoor signal propagation is required. This paper compares three empirical propagation prediction models with regards to efficiency and accuracy. To achieve this, a software prediction tool was developed using C++ which allows an end user to quickly draw a building floor plan using user specified drawing materials. It also has the ability to calculate the required empirical parameters from entered measurements but this was found to produce results similar to that when theoretical empirical parameters were used. The accuracy of the prediction tool was gauged by comparing its outputs, using the different empirical models, to measurements. In doing so it was determined that two of the models produced functional levels of accuracy in which 93% and 82% of the simulated results were within 15 and 10 dB of the measured results respectively for the most accurate of the models used. All three empirical models were found to have computational times low enough, less than 5 minutes for an average building, as to allow for interactive WLAN design.
iv
Acknowledgements
I would like to thank the following people for their much needed contributions to this project: Prof. Alan Clark, School of Electrical and Information Engineering, Electromagnetics Department, University of the Witwatersrand, Johannesburg. For his insight and patience during the course of the work. Armscor, The Armaments Corporation of South Africa. For providing the financial support and resources that made this MSc possible. My family and loved ones for their support.
Foreword
This dissertation is presented to the University of the Witwatersrand, Johannesburg, South Africa for the degree of Master of Science in Engineering. Entitled A heuristic tool for indoor radio-wave propagation prediction, this dissertation investigates the feasibility of using three well know empirical propagation prediction models for wireless local area network (WLAN) planning. To achieve this, a C++ propagation predication tool was designed and developed with the ability to implement each of the models on user specified building floor plans. Comparisons of the tools outputs with measurements of the actual signal strengths were made and conclusions drawn as to the effectiveness of each model with regards to its efficiency and accuracy. The format of this document is such that it complies with the universitys paper format, with the main essence of the research contained within the actual paper. The accompanying appendices present supporting information not covered in detail by the main document. Appendix A contains a detailed description of both the developed tools functional and non-functional requirements. Appendix B presents and discusses the design decisions taken during the development of the tools user interfaces. Appendix C documents the tools various classes and their interrelationships. Appendix D describes in detail each of the tools various methods by class. Appendix E presents some measurement data and describes a sample tool output. Appendix F contains a compact disc with the tools source and compiled program files.
vi
Table of Contents
Abstract Acknowledgements Forward Table of Contents List of Figures List of Tables iii iv v vi viii ix 1 1 1 2 3 4 4 5 6 6 8 8 8 9 9 10 11 11 11 11 18 18
Paper: A heuristic tool for indoor radio-wave propagation prediction 1. Introduction 2. Background 3. The Empirical Models Used 4. The Simulation Mechanism 5. Measurement Technique and Results 5.1 Application and model efficiencies 5.2 Application and model accuracies 6. Conclusion References
Appendix A: Application Attributes A-1 Introduction A-2 Functional Attributes A-3 Non-functional Attributes A-4 A Basic Operational Scenario A-5 Conclusion Appendix B: User Interface Design B-1 Introduction B-2 User Interface Requirements B-3 User Interfaces Developed B-4 Conclusion B-5 Reference List
vii Appendix C: Application Class Overview C-1 Introduction C-2 Class Descriptions C-3 Class Diagram C-4 Conclusion Appendix D: Application Method Overview D-1 Introduction D-2 Application Classes D-2.1 AddMaterialDlg D-2.2 CAgent D-2.3 Cantenna D-2.4 Ccommand D-2.5 ChangeTransCoeffDlg D-2.6 CHIPDoc D-2.7 CHIPView D-2.8 Cline D-2.9 CmainFrame D-2.10 CmeasurementResult D-2.11 CpointResult D-2.12 DrawingSettingsDLG D-2.13 MaterialDlg D-2.14 PlaceExtraDlg D-2.15 SimulationSettingDlg D-3 Conclusion Appendix E: Measurement Data and Example Output E-1 Introduction E-2 Object attenuation factor as a function of the angle of arrival E-3 Example Output E-4 Conclusion Appendix F: Source Code 19 19 19 20 21 22 22 22 22 24 25 26 26 27 40 68 69 78 79 80 80 84 85 87 88 88 88 89 89 91
viii
List of Figures
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 Obliquely incident waves (B) transmit less power than waves that occur at more normal incidences (A). Tinted glass attenuation as a function of the angle . The approximation represented is OAF/cos(). Graphical illustration of the L parameter used to determine the diameter of the first Fresnel zone. The agent decision tree. Simulation time vs. agent count for various AISPs. Simulated signal strength distribution for a single floor section of a building in dB. Measured signal strength distribution for a single floor section of a building in dB. Left hand side of the applications main user interface. Right hand side of the applications main user interface. The Material Menu interface. The Add Material interface. The Drawing Settings interface. The Simulation Setting interface. The Select Item interface. A simplified class diagram for the developed application. An Example application output. An Example application output.
2 3 3 4 5 5 6 13 14 15 15 16 16 17 20 89 90
ix
List of Tables
1 A comparison of the coinciding simulated and measured points with error less than 5, 10 and 15 dB for the three empirical models used. 2 The non-functional attributes of the developed tool. 3 The functionality of the applications menu options. 4 Measured loss due to a piece of tinted glass at various angles between two antennas.
5 9 12 88
A heuristic tool for indoor radio-wave propagation prediction

B. H. Whitaker
School of Electrical & Information Engineering, University of the Witwatersrand, Private Bag 3, 2050, Johannesburg, South Africa
Abstract: For the eective implementation of a wireless local area network (WLAN) within a building, a complete understanding of indoor signal propagation is required. This paper compares three empirical propagation prediction models with regards to eciency and accuracy. To achieve this, a software prediction tool was developed using C++ which allows an end user to quickly draw a building oor plan using user specied drawing materials. It also has the ability to calculate the required empirical parameters from entered measurements but this was found to produce results similar to that when theoretical empirical parameters were used. The accuracy of the prediction tool was gauged by comparing its outputs, using the dierent empirical models, to measurements. In doing so it was determined that two of the models produced functional levels of accuracy in which 93% and 82% of the simulated results were within 15 and 10 dB of the measured results respectively for the most accurate of the models used. All three empirical models were found to have computational times low enough, less than 5 minutes for an average building, as to allow for interactive WLAN design. Key words: Indoor signal propagation, Empirical prediction models, Interactive planning, Accuracy levels, Computation times
1.
INTRODUCTION
Many individuals and corporations have already, or are becoming, heavily reliant on the Internet and public/private data networks for their daily operation. The increased mobility and ease of installation make wireless communication networks the preferred choice in most applications, with increasing demand for wireless local area networks (WLANs) which when combined with other technologies such as voice over IP (VoIP), wireless CCTV cameras and control systems can supply a complete private/corporate operational package utilising a single network. In most WLAN deployment environments the overall strength and quality of the received signal at any given location is dominated by the many scattering processes due to layout and nature of various obstacles within the deployment environment [1]. Thus, the reliability and costeectiveness of WLAN technology relies heavily on the ability to predict the eects obstacles have in the deployment environment and to this end, numerous propagation models and methodologies have been developed, each having their associated advantages and disadvantages. In this paper three dierent empirical propagation prediction models are compared with regards to prediction accuracy and eciency using both theoretical and measured empirical coecients. Each empirical model was chosen for its straightforwardness and ability to produce accurate results [2] given fairly limited and inaccurate building information and computational resources. For the comparison of the three chosen models a propagation prediction tool, using a unique simulation mechanism, was developed and found to produce good predicted signal strength accuracy levels at high eciency rates. The format of the paper is as follows. Section 2 presents and examines literature in the eld of indoor radio wave propagation prediction. This is done to il-
lustrate the reasons behind the adoption of the chosen prediction models. The chosen empirical prediction models are then discussed in Section 3 whilst Section 4 describes the developed prediction tool used to implement the adopted prediction models. The measurement techniques used to gauge the accuracy of each selected model along with its resulting accuracy and eciency are presented in Section 5. Conclusions are included in Section 6. 2. BACKGROUND
Most if not all of the currently available propagation models can be categorised either as being empirical/statistical or deterministic so grouped as they are usually implemented in the same way. In the empirical/statistical approach, on-site measurement data is used to t either an empirical [2] or statistical [1] propagation model to the building which is then used to determine the signal strength values at unmeasured points of interest. Initially time consuming, since measurement data is required, the resulting models are straight forward to apply, fairly accurate and computationally ecient [2], [1]. Deterministic models, on the other hand, can be divided into both small and large scale subcategories. Small scale deterministic models are generally concerned with the propagation eects of individual objects and rely heavily on the numerically exact solutions to electromagnetic scattering related equations. These can include method of moments (MoM) using both surface and volume integrals [3], [4] as well as various hybrid techniques which combine the method of moments approach with that of geometrical optics (GO) [5]. Of these small scale numerically exact methods, none can be eectively implemented at todays communication frequencies and/or over a meaningful area of interest (with regards to WLAN planning) due to the excessive processing and memory capacity they would require [5]. This is a consequence of the fact that as frequency increases, the size of an object capable of
scattering an electromagnetic wave becomes smaller which results in a drastic increase in complexity. By considering the 802.11a standard which operates at around 5.8 GHz, all in-building obstacles larger than the operating wavelength will need modelling. This is approximately 5 cm and the resulting complexity from having to model obstacles such as coee cups and telephones make numerically exact solutions impractical. Ray-tracing techniques form the majority of the large scale, communication frequency, deterministic techniques [6] and often rely on electromagnetic theory such as the uniform theory of diraction (UTD) [2]. Therefore they require detailed building information pertaining to the layout and electromagnetic properties of its various construction materials and inbuilding obstacles which is not always readily available [7]. Propagation models based on ray-tracing can predict propagation phenomena such as diraction and reection [2] and as a consequence they tend to be the most accurate of all the currently available models, but are computationally complex and result in large computation times which prevent interactive WLAN planning. In light of the above, it can be seen that by using empirically based prediction models will satisfy the requirements for a propagation prediction tool that is not heavily reliant on the availability and accuracy of building information whilst being simple to use, ecient and accurate. 3. THE EMPIRICAL MODELS USED
surement and transmitter respectively. Equation (2) in decibel notation is as follows [2]:
P athLossB (d) = 10 log10
d dref
+
i=1
OAF(i) (3)
henceforth referred to as the Basic log-distance model, the empirical parameters n and OAF are found by best tting the model to measurement data. To extend this model, the OAF parameters dependence on the angle of incidence of the incoming electromagnetic wave needs consideration. This stems from the commonly observed phenomenon in which obliquely incident waves transmit less power than those that occur a normal incidences, see Figure 1.
Normal Incident angle
Normal Incident angle
Incident wave Obstacle Transmitted wave (A)
Incident wave Obstacle Transmitted wave (B)
By denition, the path loss at a distance d from a transmitter is given by: P athLoss = Pr (dref ) Pr (d) (1)
Figure 1 : Obliquely incident waves (B) transmit less power than waves that occur at more normal incidences (A). The results of the work by [2] established the following OAF dependance on the incident angle with respect to the normal, , and forms the basis of the Intermediate log-distance model:
Where dref is an arbitrary reference distance and Pr (d) is the received power at a distance d from the transmitting antenna. From [8] it is stated that by setting dref to 1m, only propagation eects are included in equation (1) by normalising the path loss at any distance to that which occurs at dref , eectively removing any antenna related parameters such as gain and transmitting power. By using equation (1) as a starting point and with the knowledge that it predicts free space path loss, it can be intuitively extended as follows:
P athLossI (d) = 10 log10
d dref
+
i=1
OAF(i) (4) cos()
This was veried by conducting measurements on a piece of metal-tinted glass and results [9] can be seen in Figure 2 . To further improve the accuracy of the Intermediate log-distance model, the empirical parameter ns dependance on the distance from the transmitting antenna was investigated. From [7] it was found that n as a function of distance has two distinct values, one
P athLoss = Pr (dref ) Pr (d)
OAF(i)
i=1
(2)
Where OAF and m are the obstacle attenuation factor and number of obstacles between the point of mea-
10
Approximated Measured
tting on-site measurement data to equation 5.
20
Ceiling
30
Obstacles L
Tinted glass loss (dB)
40
50
60
70
80
Floor
0 10 20 30 40 50 60 70 80 Incident angle (Degrees)
90
Figure 2 : Tinted glass attenuation as a function of the angle . The approximation represented is OAF/cos().
Figure 3 : Graphical illustration of the L parameter used to determine the diameter of the rst Fresnel zone. 4. THE SIMULATION MECHANISM
value when within the rst Fresnel zone and another once the rst Fresnel zone has become obstructed. This can be understood by considering [2] in which it is stated that at distances close to the transmitting antenna (within the rst Fresnel zone), obstacles do not signicantly eect propagation therefore it can be considered as free-space propagation. In light of this, equation (4) can be rewritten to constitute the Advanced log-distance model as follows:
To implement the empirical models presented in Section 3 a software application was developed using C++, chosen due to the speed at which it is able to carry out repetitive algorithms. Listed below is some of the softwares functionality. For a full listing, please see [10].
P athLossA (d) = 10 log10
d dref
n(d)
+
i=1
OAF(i) cos() (5)
Where: n(d) n(d) = 2 if within the rst Fresnel zone. = 2-5 if rst Fresnel zone is obstructed.
The distance Zf at which the rst Fresnel zone becomes obstructed is given by the following equation [7]: Zf L (6)
Is backed by an Access r materials database that allows for the editing, addition and deletion of user specied materials. Allows the end-user to draw a building plan as well as position various in-building obstacles using either a CAD like or mouse pointer driven drawing interface [11]. A simulation settings dialog that allows for the specication of the required simulation constants as well as the various empirical parameters outlined in Section 3 [11]. A drawing settings dialog that allows for the specication of the required drawing area size [11]. This automatically scales the representative pixel size. Has two dierent simulation mechanisms. The rst (o-site) implements the propagation models using user dened empirical parameters whilst the second (on-site) calculates the empirical parameters from user-entered measurement results on the drawing plan.
Where the parameter L is the maximum diameter, in meters, of an unobstructed ellipse drawn in the vertical plane of a building oor as shown in Figure 3. Thus for any given building, the diameter of the rst fresnel zone can be anywhere from 5-25 m depending on the height of the ceiling above the oor as well as the nature, layout and density of the various obstacles within the building environment. The value of n(d) , once the rst Fresnel zone has become obstructed, usually varies between 2 and 5 and can be found by best
A raster as opposed to a vector approach was used to ensure the lowest possible simulation computation times. This stemmed from the fact that intersection testing, using vector graphics, requires the solution of numerous modelling equations for objects that might not even be in the direction in question. In the raster methodology used, intersection testing was simplied by representing the required building oor plans, inbuilding obstacles and the transmitting antenna as corresponding values in a matrix. Agents (the data
structures responsible for the propagation prediction) then move through the matrix in all directions, having originated at the location of the transmitting antenna, setting the corresponding matrix entries to the required signal strength values. To move the Agents, Bresenhams line drawing algorithm was used as it only requires integer addition, subtraction and multiplication by 2 which computers can perform eciently [12]. On each move, an Agent is faced with the decisions outlined in Figure 4.
Move Agent to next position in matrix Update agent data Calculate signal strength Is current position free space? No Find intercepted material
The OAFs used in all the models were found from [13] and o-site measurements.
Yes What model was selected? (3)
(3) or (4)
What model was selected?
(4) or (5)
(5) Calculate the angle of interception
Yes
Is current position in first Fresnel zone?
Change n(d)
No
Figure 4 : The agent decision tree. To improve the eciency of the simulation mechanism, agents are preprogrammed to stop functioning at a user specied signal strength level or when they leave a user specied simulation boundary. Furthermore, a weighted averaging algorithm is also employed to calculate the signal strength at regions where no agent activity was present. 5. MEASUREMENT TECHNIQUE AND RESULTS
This process was dissimilar to that used in the determination of the empirical parameters for the on-site simulation mechanism. For models (3) and (4) n was set to 2 whilst for model (5) n(d) was set to 2 when within the rst Fresnel zone and 2.35 once the rst Fresnel zone became obstructed. These values were arrived at by best tting on-site measurement data to the respective model equation. ZL was found using equation (6) as in the o-site mechanism and the OAFs for the various building materials were found by the application using user entered measurement data on the building oor plan. For the actual measurements, a client card, attached to a laptop, was used in combination with an access point (AP). The client card had a maximum gain of 0 dBi and due to the non-uniformity of its radiation pattern in the plane of interest, the orientation of the laptop (with respect to the AP) whilst taking measurements was kept constant. The accuracy of the measurement system was gauged by comparing its reported path loss over a set distance in an anechoic chamber, to that of a known system. In doing so it was determined that the measurement system was fairly accurate with its reported path loss values being within 2 dB of the known values. For the comparison between the model predictions and measurements, more than 200 measurements of the actual signal strength were taken at random positions within the two buildings. Fast fading eects were initially considered by taking the average of a selection of measurements within a 20 area around the selected random points [8], but was found to produce a similar results to a single measurement in most to all instances. Measurements were also taken at a distance to prevent the measurer from aecting their integrity. 5.1 Application and model eciencies
Propagation predictions were made using the software for two dierent buildings using models (3), (4) and (5). The buildings in question were dissimilar in the fact that the one was primarily constructed from bricks and mortar whilst the other from reinforced concrete and dry walling. From visual inspection, rough sketches of the buildings were entered into the application. For the o-site simulation mechanism the required empirical parameters were determined as follows:
For models (3) and (4), n was set at 2 (free space approximation). For model (5), n(d) within the rst Fresnel zone was set at 2 and 2.5 once the rst Fresnel zone became obstructed [2]. ZL was found using equation (6).
By recording the simulation times for various agent counts and oor plan areas in square pixels (AISP), the curve in Figure 5 showing the relationship was sketched. In doing so it was also determined that the approximate run-times of the dierent models were similar and thus can be represented by the same set of curves. This is a result of the fact that free space agent movement calculations greatly dominated the overall simulation run-time in the tested building environments. Also shown in Figure 5 is that at lower agent counts (less than 50 per octant), the weighted average algorithm dominates the overall simulation run-time i.e. lower agent counts result in less calculated signal strength values and hence more averaging is required is achieve an approximately continuous signal strength distribution. As agent counts rise, the time taken to move them around the raster increases and coupled with the fact that they produce more calculated signal strength values causes their movement algorithm
to dominate over the overall simulation run-time.

80
70
AISP: 160000
60
50 Time (s)
AISP: 90000
40 AISP: 40000 30
doing so, the eects of high attenuation factor objects are underplayed, resulting in more consistent results. The accuracy of the models as a function of agent count was considered and it was determined that agent counts above 200 per octant produced consistently accurate results. This is a consequence of the fact that at lower agent counts, obstacles far away from the transmitting antenna are often not intersected as agents are programmed to move radially outwards from the transmitting antennas location. The eects of diraction and reection were not included in any of the models to eliminate the need for vector additions which require a high degree of accuracy and precision with regards to electromagnetic properties of the building materials used and the building dimensions respectively. Both of which are time consuming and hence expensive to obtain. As an example, the work done by [13] at 914 MHz, required building blueprint levels of precision to achieve accurate results. In light of this, the results in Table 1 show that even with fairly inaccurate building oor plans (maximum error in each dimension no greater than 0.2 m) usable levels of accuracy were obtained for both the Intermediate and Advanced log-distance models at interactive computational times.
20
AISP: 10000
10
50
100
150
200 250 Agent count per octant
300
350
400
450
Figure 5 : Simulation time vs. agent count for various AISPs. Tests conducted on a 600m2 section of building produced average run-times of approximately 3 minutes using a P4 2.8 GHz personal computer with 1 GB of memory, illustrating the eciency of the empirical models used. 5.2 Application and model accuracies
-70
By comparing the obtained measurement results to that of the simulated results for the dierent empirical models, the results in Table 1 were obtained.
9m -60
-80
-70
-70 -80
Table 1 : A comparison of the coinciding simulated and measured points with error less than 5, 10 and 15 dB for the three empirical models used. <5 dB Basic log-distance (3) 41% Intermediate log-distance (4) 47% Advanced log-distance (5) 51% <10 dB 61% 77% 82% <15 dB 76% 89% 93%
6m -80 3m -60 -70
-50 0m -60
-70
Shielded Chamber -40 -50 -40 -50 -60 -60 -60
-60
-70
These results were similar to those found by [2] for the Advance log-distance model. It was also found that the choice of using either the on-site or o-site simulation mechanism, did not signicantly alter the general accuracy of the model used. However, under the o-site simulation mechanism a larger standard deviation in the results with error greater than 15 dB was observed. The reason for this is that the objects with high attenuation factors such as steel cabinets, under the o-site mechanism, exaggerate their eects on the signal strength behind them as none of the models account for diraction and reection. On the other hand, the on-site simulation mechanism determines an objects attenuation factor from measurements which are aected by indirect signal paths. In
-70 -60 -70 -70 -70
-50 -60 -50 -60 -70 -60
Figure 6 : Simulated signal strength distribution for a single oor section of a building in dB. This assumption was made as it was found that the vast majority of the simulated results were within 15 dB of the actual signal strengths. Further more, it is backed up by the fact that WLANs should never be designed down to the sensitivity levels of the receiving antennas. A margin of at least 20 dB should be
-80
-70 -70
of the conducted simulations, the use of fairly inaccurate building oor plans were found to not signicantly aect model accuracies.
-70
9m
REFERENCES [1]
-70
6m -60 3m
-50 0m -60
[2]
Shielded Chamber -40
-40
-50
-60
[3]
-50 -60 -70 -50 -60 -50 -70 -70 -60 -60 -70 -60
[4]
Figure 7 : Measured signal strength distribution for a single oor section of a building in dB.
[5]
ensured due to the highly dynamic nature of the resulting channel between the transmitter and receiver. A example of this is found in [14] who found that losses from 6 to 11 dB can result from human bodyshadowing eects. Figures 6 and 7 show the similarity in contour plots between the simulated and measured signal strength distributions (using the Advanced logdistance model and the o-site simulation mechanism) for one of the measured buildings. The contour lines for the simulated results were generated by the software and Matlab r was used to obtain contour lines from the measurement data. Of interest in Figure 6 are the regions of rapid signal strength decline caused by the extremely oblique intersection angles with obstacles between the transmitting antenna and region in question. This is not reected in Figure 7 and thus constitutes a software anomaly that does not drastically alter the softwares overall accuracy. 6. CONCLUSION
[6]
[7]
[8]
[9]
[10]
By developing an indoor propagation prediction tool for WLANs, the accuracy and eciency of three different empirical propagation prediction models were tested. This was done by comparing the tool outputs to measurements and it was found that two of the models produced usable levels of accuracy, as the vast majority of their predicted results were within 15 dB of the actual measured signal strength values. All three of the empirical models were found to be computationally non-time consuming and thus could be used in interactive WLAN planning or in conjunction with access point optimisation algorithms that would require a vast number of simulations to be run. In all
[11]
[12]
[13]
T. S. Rappaport. Characterization of UHF Multipath Radio Channels in Factory Buildings. IEEE Trans. Antennas Propagat., vol. 37, no. 8, pp. 10581069, 1989. K. Cheung, J. H. Sau, and R. D. Murch. A New Empirical Model for Indoor Propagation Prediction. IEEE Trans. Veh Technol., vol. 47, no. 1, pp. 9961001, 1998. S. Govind, D. R. Wilton, and A. W. Glisson. Scattering from inhomogeneous penetrable bodies of revolution. IEEE Trans. Antennas Propagat., vol. 32, pp. 11631173, 1992. J. H. Richmond. Scattering by a dielectriccylinder of arbitrary cross section shape. IEEE Trans. Antennas Propagat., vol. 13, pp. 334341, May 1965. R. Bhalla and H. Ling. Three-dimensional scattering centre extraction using the shooting and bouncing ray technique. IEEE Trans. Antennas Propagat., vol. 44, no. 11, pp. 14451453, 1996. C. Yang, B. Wu, and C. Ko. A Ray-Tracing Method for Modeling Indoor Wave Propagation and Penetration. IEEE Trans. Antennas Propagat., vol. 46, no. 6, pp. 907919, 1998. W. Honcharenko, H. L. Bertoni, J. L. Dailing, J.Qian, and H. D. Yee. Mechanisms governing UHF propagation on single oors in modern oce buildings. IEEE Trans. Veh Technol., vol. 41, no. 4, pp. 496504, 1992. D. M. J. Devasirvatham. A comparison of time delay spread and signal level measurements within two dissimilar oce buildings. IEEE Trans. Antennas Propagat., vol. 35, no. 3, pp. 319324, 1987. B. H. Whitaker. A heuristic tool for indoor radiowave propagation prediction. Masters thesis, Appendix E: Measurement data and example output, University of the Witwatersrand, 2005. B. H. Whitaker. A heuristic tool for indoor radiowave propagation prediction. Masters thesis, Appendix A: Application Attributes, University of the Witwatersrand, 2005. B. H. Whitaker. A heuristic tool for indoor radiowave propagation prediction. Masters thesis, Appendix B: User interface design, University of the Witwatersrand, 2005. J. E. Bresenham. Algorithm for computer control of a digital plotter. IBM Syst. Jour., vol. 4, no. 1, pp. 2530, 1965. S. Y. Seidel and T. S. Rappaport. 914 MHz Path Loss Prediction Models for Indoor Wireless Communications in Multioored Buildings. IEEE Trans. Antennas Propagat., vol. 40, no. 2, pp. 207217, 1992.
[14] S. Obayashi and J. Zander. A Body-Shadowing Model for Indoor Radio Communication Environments. IEEE Trans. Antennas Propagat., vol. 46, no. 6, pp. 920926, 1998.
Appendix A: Application Attributes

1. Introduction The following document describes the developed WLAN propagation prediction tool by presenting both its functional and non-functional related attributes. In viewing this document it is important to remember that during the development of the tool, proof of concept was a priority and the need to produce a fully functional tool less important. 2. Functional Attributes In order for proof of concept, the tool has the following functionality: 1. Allows the end user to draw a building plan and any in-building obstacles such as book shelves, filing cabinets etc. The drawing can be done using either a text based drawing interface in which the commands MoveTo(x,y) or LineTo(x,y) are used where x and y are in meters or a mouse pointer driven interface. 2. Is backed by an Access materials data that allows for the editing, addition and deleting of user specified materials. Each material in the materials database is given a name, frequency of operation (as material attenuation is dependent on frequency), transmission coefficient and a unique colour. 3. Allows the end user to control all aspects of the simulation. This is achieved by having a simulation settings dialog that is used to specify the required simulation mechanism, frequency, cut-off level, initial agent count, first and post Fresnel Zone exponents, diameter of first Fresnel Zone and the measured signal strength at a distance of one meter from the transmitting antenna. 4. A drawing settings dialog that allows the end user to specify the required drawing area size. This is done to ensure the maximum level of accuracy is achieved in a simulation, as a raster approach was used in which each pixel represents a physical area of the floor plan. The allowable drawing size range is from 100 m2 to 25 000000 m2. 5. Has the ability to zoom in and out to aid in the drawing of a building plan. 6. Allows simulation results to be saved to a Device Independent Bitmap or DIB to allow for easy distribution. 7. Allows work in progress to be fully saved in a .HIP file which is the applications unique file format. 8. Allows the specification of a simulation boundary to reduce the simulation times. 9. Has two different simulation mechanisms. The first (off-site) implements the propagation models using user-defined empirical parameters whilst the second (on-site) calculates the empirical parameters from user-entered measurement results on the drawing plan.
9 3. Non-functional Attributes. The main non-functional attributes of the developed tool are that of reliability, maintainability, upgradeability and compatibility. Descriptions and means of achievement for these attributes can be found in Table 2.
Table 2: The non-functional attributes of the developed tool.
Non-functional requirement Reliability
Description of requirement The tool must perform to the same degree of accuracy and efficiency each and every tine it is used. It must be possible and easy to repair any errors/bugs that arise in the application during testing and in its use.
Why it is needed
How it was achieved
A reliable tool will increase user confidence in the results it produces. A maintainable tool will continue to be applicable in the event of finding bugs/errors.
Sound programming techniques and extensive testing procedures. By adopting an object orientated programming technique, errors and sources of inaccuracy are easily found and eliminated. By designing the program such that it does not require large amounts of system resources and by ensuring that no platform depending technology was used.
Maintainability
Compatibility
The tool must run on most personal and laptop computers running a Microsoft operating system.
A compatible tool can be used by a larger and more diverse user group.
Upgradeability
It must be possible and easy to add new functionality to the application.
An upgradeable tool will be applicable over a larger time span.
By adopting an object orientated programming technique, additional functionality can be easily added.
4. A Basic Operational Scenario On running the application, the vast majority of its end user will do the following: 1) Browse the well established materials database to ensure that the required building materials are present. If not, a new material will be entered and specified.
10 2) Using either the text or mouse based drawing facilities, draw the required building plan. 3) Place an antenna on the building plan at the desired location and specify a simulation boundary. 4) Save the building plan to a .HIP file. 5) Run the simulation using either the Basic, Intermediate or Advanced Log Distance simulation mechanism. 6) View the simulation results and repeat steps 3 5 until the desired signal strength distribution is reached. 7) Save the simulation results to a DIB. 8) Print a copy and send it to a WLAN installer to indicate the positions of the desired Access Points. 5. Conclusion From the discussion of the developed applications functional and non-functional attributes it is evident that the only limitation to the proof of concept is whether the chosen empirical propagation prediction models have the required accuracy and efficiency. This results from the fact that the developed application attributes allow the prediction models to be fully and unlimitedly implemented.
11
Appendix B: User Interface Design

1. Introduction A survey carried out in 1992 found that approximately 48% of an applications code was devoted to the user interface and that 50% of the development time was spend implementing the user interface [1]. This is a result of the fact that well designed interfaces have become extremely important in ensuring the overall success of any given software application. In light of this, the following document presents the design decisions taken in the construction of the various user interfaces for the developed WLAN propagation prediction tool. The user interfaces are presented and their functionality discussed. Conclusions on the degree to which the developed interfaces meet both their functional and non-functional requirements are also given. 2. User Interface Requirements A good user interface is one in which the balance between user-friendliness and transparent functionality is such that in affords its targeted users a high degree of productivity. As the developed software is targeted at a semi-technical to technical end user, the following user interface requirements were important:
Be user friendly to the extent that it is easy to use, but must not limit advanced users. Promote learning and be consistent. Provide all of the applications required functionality.
3. User Interfaces Developed Using the above outlined requirements, the applications interfaces were designed and developed. On running the application, the end user initially sees the main application interface. The main interface was designed in such a way that all the required functionality to draw a building floor plan is easily available. This can be seen in Figures 8 and 9 which are the left and right hand side of the main interface respectively. The key for Figure 8 is as follows: A. The applications drawing rulers. B. C. D. E. F.
They are auto scaling rulers that change depending on the user specified drawing size, see Figure 12. Ruler units are in meters. A simple tool bar that allows easy access to the New, Open, Save, About and Print drawing functions found in the File drop down menu option. Section of the tool bar used to provide feedback during a simulation. Possible feedback includes Generating agents, Moving agents around raster and Averaging in progress. Menu bar containing the applications various drop down menus, see Table 3 for a complete description. Applications drawing area. Having selected a material, a line is drawn by depressing the right mouse button at the start of the line and releasing it at its end. Horizontal scroll bar used to bring unseen areas of a large drawing into view.
12 G. Materials list box containing the ID and name of all the materials entered in the
materials database. A material is selected by single left-clicking its ID. H. Antennas list box containing the ID and name of all the antennas in the antennas database. I. Drawing input bar which responds to the commands MoveTo(x,y) and LineTo(x,y) where x and y are in meters. J. Status bar used to provide feed back on the state of the application.
And the key for Figure 9 is as follows:

A. The current process indicator bar. B. Vertical scroll bar used to bring unseen areas of large drawing into view. C. Command history list box which contains the carried out drawing commands. Commands and their resulting action are deleted by double left-clicking the command number. D. The current length in meters of a line being drawn. E. The currently selected drawing material. F. The current x-position in meters of the mouse curser. G. The current y-position in meters of the mouse curser.
Table 3 below describes the functionality of the applications menu options:

Table 3: The functionality of the applications menu options. Menu Menu option New Open Save Save As Print Print Preview Print Setup Recent File List Exit Zoom In Zoom Zoom Out Zoom Normal Toolbar Status Bar View Input Bar Display Bar Databases Materials Simulation Settings Settings Drawing Settings Calculate Coefficients Simulate Basic Log Distance Intermediate Log Distance Advanced Log Distance Save Results Help About HIP Function Clears existing work and opens a new .HIP file. Opens an existing .HIP file. Saves the open building plan .HIP file. Saves a new building plan .HIP file Prints the currently open building plan .HIP file. Opens a standard Print Preview interface showing the open building plan .HIP file. Opens a standard Print Setup interface. Shows the last five worked-on building plan .HIP files. Exits the application. Zooms in on the current building plan by a factor of 2. Zooms out on the current building plan by a factor of 2. Returns the zoom factor to unity. Shows/Hides the Toolbar. Shows/Hides the Status Bar. Shows/Hides the Input Bar. Shows/Hides the Display Bar. Displays the Materials interface, see Figure 10. Displays the Simulation Settings interface, see Figure 13. Displays the Drawing Settings interface, see Figure 12. Calculates material attenuation factors from entered measurement results Initiates the Basic Log Distance simulation mechanism. Initiates the Intermediate Log Distance simulation mechanism. Initiates the Advanced Log Distance simulation mechanism. Saves the simulation result in a standard, easily distributable, DIB. Displays a standard About dialog box.
File
13
Figure 8: Left hand side of the applications main user interface.
14
Figure 9: Right hand side of the applications main user interface.
By selecting the Materials menu option from the Databases menu, the interface presented in Figure 10 is displayed. This interface is used to maintain the applications material database.
15
Figure 10: The Material Menu interface. A. Add Material button. This displays the Add Material interface, see Figure 11, used to add a new material to the materials database. B. This button removes a material from the materials database. The material to remove is selected by single left-clicking its ID. C. This button displays a dialog box which prompts the user to enter the selected materials new transmission coefficient. The material is selected by single leftclicking its ID. D. This button closes the interface.
To add new material, the Add Material interface is used, Figure 11. In this interface, the new material name, transmission coefficient, representative colour and frequency of operation is selected.
Figure 11: The Add Material interface.
16 To setup both the drawing and simulation environments, the Drawing Settings and Simulation Settings interfaces are used. These can be seen in Figures 12 and 13 respectively.
Figure 12: The Drawing Settings interface.
Figure 13: The Simulation Setting interface.
The Drawing Settings interface in used to specify the required drawing size. This is done to maximize the accuracy of the simulated results, by ensuring that the specified drawing size is no bigger than the actual drawn building. The Simulation Settings interface is used to specify the required simulation constants. These include the initial agent count, first and post Fresnel Zone exponents, diameter of the first Fresnel Zone, simulation frequency and cut-off level as well as the antenna lost at one meter. The simulation mechanism can also be set to use either object attenuation factors derived from the database of materials or user entered measurement. User measurements and antennas are entered onto a building plan by double right-clicking the floor plan at the desired position. This displays the Select Item interface, Figure 14, in which either an antenna or specified measurement result is selected. Having successfully run a simulation, the predicted signal strength will be represented by means of a colour gradient on the actual floor plan. The colour gradient ranges from red (areas of high signal strength) to blue (areas of low signal strength). A more accurate signal strength value can be obtained by pressing Shift and single rightclicking at the desired position. By pressing Shift and single left-clicking at two different positions, a simulation cut-off boundary can also be specified if the user does not want to run the simulation over the entire building plan. This can often save considerable amounts of time.
17
Figure 14: The Select Item interface.
4. Evaluation of the user interfaces developed The developed interfaces were found to have the following attributes: Provide positive feedback to the user The interfaces constantly keep the end user informed of any ongoing processes and the initiation of new ones. This is accomplished by using dynamic buttons which move when pushed and on-screen displays which are constantly updated. Immediate feedback is important to preserve the users trust in and satisfaction with the application whilst ensuring that the user stays attentive to the process he/she is involved in. Provide good error messages From a users point of view an error message is useless if it is not in an understandable format. With this in mind the interfaces report errors in generally used language, containing no jargon, and suggest possible solutions to the errors that occur. Are consistent Research has shown [1] that consistency with regards to related functions and general procedures within a software application reduces the time taken for an end user to become familiar with the application thus increasing their confidence. Therefore, all the buttons in the interface are the same shape and colour, and related menu items are grouped together. Use colour advantageously The use of colour by all means can both add and detract from the overall quality of an interface. In order to gain the beneficial use of colour within the application, the following was adhered to: 1. Colour was not used to convey a meaning as different colours are perceived to mean different things by different users. 2. Colours that clash were not used as this distracts the end user and reduce his/her productivity. Are transparent The general presentation of information within the application is concise and contains no irrelevant information. System flow with regards to accomplishing a given task within the application is
18 intuitive thus allowing users of varying computer skill levels to utilize the application accurately and efficiently. 4. Conclusion From the introduction of this report, it became apparent of the importance quality interfaces play in the success of an application. It was found that a good quality user interface will increase user efficiency and productivity whilst reducing user error and time spent learning how to use the application. This will undoubtedly increase user acceptance, resulting in a more widely used application. The developed interfaces for a WLAN propagation prediction tool were presented and it was determined that they meet both their functional and non-functional requirements with regards the good interface design. 5. Reference List [1] Van Vliet H, Software engineering: Principles and Practice Second edition, John Wiley & Sons, New York, 2002.
19
Appendix C: Application Class Overview

1. Introduction Contained within this document is a listing and description of the developed WLAN propagation prediction tools main classes. A class diagram showing the relationship between the various classes is also given. From the presented class diagram, it can be seen that good object orientated programming techniques were carried out throughout the tools development phase resulting in a robust, maintainable and fully upgradeable application. 2. Class Descriptions The following main classes were used to implement the WLAN propagation prediction tool:
1. AddMaterialDlg A descendent of the CDialog class, this class is responsible for the displaying and functionality of the Add Material interface. In this interface, the materials name, frequency of operation, transmission coefficient and colour is specified. The classes function is to take in the user-entered material specifications, validate them and then add them to the materials database. 2. CAgent A descendent of the CObject class, this class encapsulates the entire attributes of the utilized agent data types. These data types move around the raster containing the user drawn building plan and determine the resulting signal strength distribution. To achieve this, they move a pixel at a time, check for an intersection and using the established simulation parameters determine the resulting signal at their current point using the specified simulation mechanism. 3. CAntenna A descendent of the CObject class, this class encapsulates a specified antenna attributes. These attributes include radiation pattern and gain characteristics. 4. CAntennaSet A descendent of the CRecordSet class, this class provides the functionality to read and write to the antennas database. 5. CCommand A descendent of the CObject class, this class encapsulates the attributes of a command object. Command objects are generated each time a line is drawn and are used loading, saving and manipulation building plan drawings. 6. ChangeTransCoeffDlg A descendent of the CDialog class, this class loads and controls the Change Transmission Coefficient interface. This interface is responsible for changing an entered materials transmission coefficient and simple takes in the materials new transmission coefficient, validates it and updates the materials database. 7. CHIPApp A core application class, his class retrieves all the applications event messages and passes them on to the CHIPView class. 8. CHIPDoc A core application class, this class is responsible for housing and manipulation the applications various data structures. These include all the objects derived from the CObject class. It also receives inputs from the CHIPView class, processors them and passes display information back to the CHIPView class. 9. CHIPView A core application class, this class displays information to the screen and contains the majority of the functionality that acts on the data structures contained within the CHIPDoc class. 10. Cline A descendent of the CObject class, this class encapsulates the attributes of a line object. A line object specifies a drawn line and hence contains the lines start and end points along with the material used to draw it.
20
11. CMainFrame A core application class, this class holds the applications various visible objects such as its menus, scrollbars and toolbars. 12. CMaterialSet - A descendent of the CRecordSet class, this class provides the functionality to read and write to the materials database. 13. CMeasurementResult A descendent of the CObject class, this class encapsulates the attributes of a measurement object. A measurement object is generated each time a user specified measurement is entered on a building floor plan and stores the measurements value and position. 14. CPointResult A descendent of the CObject class, this class encapsulates the attributes of a point result object. Point result objects are generated by the agent objects after them have moved from one pixel to another. They contain the calculated signal strength by the agent as well as the position in raster at which the agent did the calculation. 15. DrawingSettingsDLG A descendent of the CDialog class, this class displays the Drawing Settings interface and takes in the user-entered drawing size. 16. MaterialDlg A descendent of the CDialog class, this class displays the Materials Database interface. In the interface, the contents of the material database is displayed and the functionality. 17. PlaceExtraDlg A descendent of the CDialog class, this class allow the end user to place either an antenna or measurement result on the drawn building floor plan. 18. SimulationSettingDlg A descendent of the CDialog class, this class displays the Simulation Settings interface, takes in the user specified simulation parameters and validates them.
3. Class Diagram A simplified class diagram showing the relationships that exist between the various classes can be seen in Figure 15.
Figure 15: A simplified class diagram for the developed application.
21 From Figure 1 it can be seen that an object oriented approach was adopted for the design. Cohesion which is a measure of the mutual affinity between the classes is strong and coupling which indicates the dependency between classes/objects is low. This will ultimately lead to a robust, maintainable and upgradeable design as the occurrence of a problem in any given class will not cause the entire application to stop functioning. Maintainability and upgradeability was ensured by both producing well commented code, see Appendix D and by good code reuse. 4. Conclusion By utilizing an object oriented approach during the development of the prediction tool, a highly modular design resulted. Cohesion and coupling between the various modules was found to be strong and low respectively, resulting in a robust, maintainable and updateable application.
22
Appendix D: Application Method Overview

1. Introduction To ensure maintainability and updatability, well commented code is essential. In light of this, contained within this document are detailed descriptions of the developed WLAN propagation prediction tools various methods by class. Each description describes in detail what each method does along with its various dependencies, objects and variables. The applications classes can be found in appendix G which contains a compact disc of its source code. 2. Application Classes 2.1 AddMaterialDlg void AddMaterialDlg::AddRecord (int ID, CString Name, CString Frequency)
Description: This method adds a new record to the materials record set. Dependencies: AddNew() - Sets up record set in add mode. Update() - Required to complete AddNew() function. Requery() - Rebuilds the record set. GetRValue(~) - Returns (int) the red component of a COLORREF variable. GetGValue(~) - Returns (int) the Green component of a COLORREF variable. GetBValue(~) - Returns (int) the Blue component of a COLORREF variable. atof(~) - Does a CString to double conversion. Variables and Objects: CMaterialSet set - Materials record set. Int ID - New materials ID. CString Name - New material name. CString Frequency - New material frequency. COLORREF Colour - The selected material colour in RGB space. Bugs: To do:
void AddMaterialDlg::OnCancel () [protected]

Description: Closes the window, returning the focus to the calling method. Dependencies: OnOK() - Simply closes the dialog window. Variables and Objects: Bugs: To do:
void AddMaterialDlg::OnColourSelect () [protected]

Description: This method displays the CColorDialog which is a standard MFC dialog that allows for the selection of a colour. On closing the CColorDialog using its Ok button,
23
the ColourSelected variable is set to true and the selected colour is stored in Colour variable which is of type COLORREF. Dependencies: DoModal() - This method invokes a modal dialog box and returns the dialog-box result when done. GetColor() - Returns the selected COLORREF variable from a CColorDialog. Variables and Objects: BOOL ColourSelected - TRUE if a colour was selected. COLORREF Colour - The selected material colour in RGB space. CColorDialog dialog - The colour selection dialog window. Bugs: To do:
BOOL AddMaterialDlg::OnInitDialog () [protected, virtual]

Description: Called when the dialog is initialized, this method sets the current selection in the frequency combo box to its first entry. Dependencies: SetCurSel(~) - Sets the focus on the specified combo box item. Variables and Objects: CComboBox m_Frequency - The frequency selection combo box. Bugs: To do:
void AddMaterialDlg::OnOK () [protected, virtual]

Description: This method is called when the Add Material dialog is dismissed with Ok button and does the following: 1. Opens the materials record set. 2. Gets the name and frequency of the specified material and gets them into the required format. 3. CHECK 1: Was a material name specified. 4. CHECK 2: Was a material colour specified. 5. CHECK 3: Was the specified colour white (free space). 6. CHECK 4: Was the specified colour black (antennas). 7. CHECK 5: Was a valid transmission coefficient entered. 8. Having done and passed the above checks, the newly specified material is then added in one of two ways. If the materials record set is empty, the material is simply added using the AddRecord(~) function and the material set closed. Or else if the record set is not empty the following is done: 1) The ID of the last record in the set is obtained. 2) CHECK 6: The selected material colour is not already in use. 3) CHECK 7: The selected material name/frequency combination is not already in use. If both of the above checks pass, the newly specified material is then added to the record set. Dependencies: TrimLeft() - Removes any white space preceding a CString. Open() - Opens the specified database to allow access to its records.
24
Close() - Closes the specified database. GetRecordCount() - Returns the number of records in a record set. MoveNext() - Moves to the next record in the database. MoveFirst() - Moves to the first record in the database. MoveLast () - Moves to the last record of a record set. UpdateData(~) - By passing in TRUE, the entered variables are copied to their respective member variables. GetLBText(~) - Returns the CString at the specified index of a list box. OnOK() - Simply closes the dialog. GetLength() - Returns the number of characters in a CString. AfxMessageBox(~) - Simply displays the specified text in a message box. RGB(~) - Specifies a COLORREF variable. IsEOF() - TRUE until the end of the record set is reached. GetRValue(~) - Returns (int) the red component of a COLORREF variable. GetGValue(~) - Returns (int) the Green component of a COLORREF variable. GetBValue(~) - Returns (int) the Blue component of a COLORREF variable. CompareNoCase(~) - Compares two specified CStrings, ignoring case. AddRecord() - Adds the specified material to the material record set. Variables and Objects: CMaterialSet set - Materials record set. CString Name - Specified material name. int index - The index of the selected material frequency. CComboBox m_Frequency - The frequency selection combo box. CString Frequency - The selected material frequency. BOOL ColourSelected - TRUE if a material colour selected. double m_TransCoeff - The specified material transmission - coefficient. int LastID - The ID of the last material in the material - record set. MaterialDlg temp - An object of type MaterialDlg used to gain access to its DoubleToCString(~) method. int Check1 - 1 if material name is already in use. int Check2 - 1 if material frequency is already in use. Bugs: To do:
2.2 CAgent CAgent::CAgent (CPoint CurrPoint, double dY, double dX, int Octant, double MagnitudeOffset, int SourceAnt)
Description: Being a class constructor, this method is called each time an object of this class is created. This method is simply used to initialize some required variables. Dependencies: Variables and Objects: CPoint m_CurrPoint - The agents current point in the drawing plane. int m_iSourceAnt - The agents source antenna. int m_iOctant - The agents octant of propagation.
25
double m_dY - Variable used in Bresenhams Algorithm. double m_dX - Variable used in Bresenhams Algorithm. double m_d - Variable used in Bresenhams Algorithm. double m_dD - Variable used in Bresenhams Algorithm. double m_dU - Variable used in Bresenhams Algorithm. double m_dMagnitudeOffset - The agents initial magnitude offset. CPoint m_origin - The agents origin. Bugs: To do:
2.3 CAntenna
CAntenna::CAntenna (LPCTSTR Type, CPoint Location, int ID, int Orientation)
Description: Being a class constructor, this method is called each time an object of this class is created. This method is simply used to initialize some required variables. Dependencies: Format(~) - This function writes formatted data to a CString. In this method it is used to get a CString representation of an integer. Variables and Objects: CString m_Type - The name of the antenna. CPoint m_Location - The location of the antenna in the drawing plane. int m_Orientation - The antennas location. CString m_csID - The antennas ID. Bugs: To do:
void CAntenna::Draw (CDC * pDC)
Description: When called, this method draws the antennas representation to the screen at the required co-ordinates through the active device context. The antennas ID, name and a picture are drawn. Dependencies: SelectObject(~) - Selects the specified object into a device context. MoveTo(~) - Moves to the specified point in a device context. LineTo(~) - Draws a line to the specified point in a device context using the selected drawing object. SetTextColor(~) - Sets the required colour of the text to be output. SetTextAlign(~) - Sets the alignment of the text to be output. TextOut(~) - Outputs (draws through device context) the specified text at the specified point in the raster. Variables and Objects: CPen lpen - The newly created drawing object. CPen* pOldPen - The old drawing object. CDC* pDC - A pointer to the active device context. Bugs: To do:
26
void CAntenna::Serialize (CArchive & ar)
Description: Called when loading and saving a document, this function saves and loads an objects specified variables from an archive (the specified file). Dependencies: IsStoring() - TRUE: Archive is storing data. Variables and Objects: CString m_Type - The name of the antenna. CPoint m_Location - The location of the antenna in the drawing plane. int m_Orientation - The antennas location. CString m_csID - The antennas ID. CArchive &ar - The archive (file) being read or written to. Bugs: To do:
2.4 CCommand
CCommand::CCommand (LPCTSTR in)
Description: Being a class constructor, this method is called each time an object of this class is created. This method is simply used to initialize some required variables. Dependencies: Variables and Objects: CString- m_Command - The user specified or application generated command. Bugs: To do:
void CCommand::Serialize (CArchive & ar)

Description: Called when loading and saving a document, this function saves and loads an objects specified variables from an archive (the specified file). Dependencies: IsStoring() - TRUE: Archive is storing data. Variables and Objects: CString m_Command - The user specified or application generated command. CArchive &ar - The archive (file) being read or written to. Bugs: To do:
2.5 ChangeTransCoeffDlg void ChangeTransCoeffDlg::OnOK () [protected, virtual]

Description: Called when the user closes the dialog using the Ok button. All that this method does is update its member variables with the user inputted ones. It also checks that the new Transmission Coefficient is in the range of 0 to 1. Dependencies: UpdateData(~) - By passing in TRUE, the entered variables are copied to their respective member variables. OnOK() - Simply closes the dialog. Variables and Objects:
27
Double m_TransCoeff - User inputted Transmission Coefficient. Bugs: To do:
2.6 CHIPDoc CHIPDoc::CHIPDoc () [protected]

Description: This is the document classes default constructor. It is used to initialize the required variables as well as determining the initially specified pixel size. Dependencies: RGB(~) - Returns the specified COLORREF variable. pow(~) - Raises a specified double to another specified double. Variables and Objects: COLORREF m_Colour - The current drawing colour (white). CPoint m_ptCurrent - The current drawing point (0m,0m). int m_InitialAgentCount - The initial agent count (10). double m_dFirst_Zone_Exp - The path loss exponent of the first Fresnel zone (2). double m_dOther_Zone_Exp - The path loss exponent of the Fresnel zones other than the first (2.5). double m_dDiameter_First_Zone - The diameter of the first fresnel zone (10m). double m_dSimCutOffLevel - The simulation cut-off level (-100 dB). double m_dAntLoss - The initial antenna loss at 1m (0 db). BOOL m_bMeasurement - TRUE: Simulation by measurement. BOOL m_bDataBase - TRUE: Simulation by database coefficients. CString m_csDrawingSize - The initial drawing size (250m,250m). double m_dPixelSize - The initial pixel size. double m_dSimWavelenght - The initial simulation wave length. double m_dDistanceMoved - The distance moved from the source antenna. BOOL m_bTracerLinesValid - TRUE: Current tracer lines are valid. Bugs: To do:
CAgent * CHIPDoc::AddAgent (CPoint CurrPoint, double dY, double dX, int Quadrant, double MagnitudeOffset, int SourceAnt)
Description: This function is called to add a new agent to the agent object array. In more detail it does the following: 1. Creates a new agent object with the required attributes. 2. Next it checks that there is enough memory and adds the agent if there is. Note: If there was not enough memory to add the agent, an error message will be displayed and the created agent object removed. This method also returns the newly created agent. Dependencies: Add(~) - Adds a new object to a CObjectArray. AfxMessageBox(~) - Displays the specified text in a simple dialog box. Delete() - Used to delete an object of type CMemoryException. Variables and Objects:
28
CAgent* pAgent - Newly created agent object. CObjectArray m_oaAgents - Agent object array. CMemoryException* perr - Memory exception object. Bugs: To do:
CAntenna * CHIPDoc::AddAntenna (LPCTSTR in, CPoint pt, int ID, int ori)
Description: This function is called to add a new antenna to the antenna object array. In more detail it does the following: 1. Creates a new antenna object with the required attributes. 2. Next it checks that there is enough memory and adds the antenna if there is. 3. When a new antenna is added, the document is marked as dirty, which will prompt the user to save the document on exiting the application. Note: If there was not enough memory to add the antenna, an error message will be displayed and the created antenna object removed. This method also returns the newly created antenna. Dependencies: Add(~) - Adds a new object to a CObjectArray. SetModifiedFlag() - Marks the document as dirty. AfxMessageBox(~) - Displays the specified text in a simple dialog box. Delete() - Used to delete an object of type CMemoryException. Variables and Objects: CAntenna* pAntenna - Newly created antenna object. CObjectArray m_oaAntennas - Antenna object array. CMemoryException* perr - Memory exception object. Bugs: To do:
CCommand * CHIPDoc::AddCommand (LPCTSTR in)

Description: This function is called to add a new command to the command object array. In more detail it does the following: 1. Creates a new command object with the required attributes. 2. Next it checks that there is enough memory and adds the command if there is. 3. When a new command is added, the document is marked as dirty, which will prompt the user to save the document on exiting the application. Note: If there was not enough memory to add the command, an error message will be displayed and the created command object removed. This method also returns the newly created command. Dependencies: Add(~) - Adds a new object to a CObjectArray. SetModifiedFlag() - Marks the document as dirty. AfxMessageBox(~) - Displays the specified text in a simple dialog box. Delete() - Used to delete an object of type CMemoryException. Variables and Objects: CCommand* pCommand - Newly created command object. CObjectArray m_oaCommands - Command object array. CMemoryException* perr - Memory exception object.
29
Bugs: To do:
CLine * CHIPDoc::AddLine (CPoint ptFrom, CPoint ptTo)

Description: This function is called to add a new line to the line object array. In more detail it does the following: 1. Creates a new line object with the required attributes and using these attributes, generates other needed attributes. 2. Next it checks that there is enough memory and adds the line if there is. 3. When a new line is added, the document is marked as dirty, which will prompt the user to save the document on exiting the application. Note: If there was not enough memory to add the line, an error message will be displayed and the created line object removed. This method also returns the newly created line and invalidates and current tracer lines. Dependencies: Add(~) - Adds a new object to a CObjectArray. SetModifiedFlag() - Marks the document as dirty. AfxMessageBox(~) - Displays the specified text in a simple dialog box. Delete() - Used to delete an object of type CMemoryException. SetupLineVariables() - This method takes the lines start and end points and determines the max/min x/y co-ordinates. Variables and Objects: Cline* pLine - Newly created line object. CObjectArray m_oaLines - Line object array. CMemoryException* perr - Memory exception object. BOOL m_bTracerLinesValid - FALSE: Current tracer lines are invalid. Bugs: To do:
CMeasurementResult * CHIPDoc::AddMeasurementResult (CPoint Position, double Magnitude, int ID)

Description: This function is called to add a new measurement result to the point result object array. In more detail it does the following: 1. Creates a new measurement result object with the required attributes. 2. Next it checks that there is enough memory and adds the measurement result if there is. 3. When a new measurement result is added, the document is marked as dirty, which will prompt the user to save the document on exiting the application. Note: If there was not enough memory to add the measurement result, an error message will be displayed and the created measurement result object removed. This method also returns the newly created measurement result. Dependencies: Add(~) - Adds a new object to a CObjectArray. SetModifiedFlag() - Marks the document as dirty. AfxMessageBox(~) - Displays the specified text in a simple dialog box. Delete() - Used to delete an object of type CMemoryException. Variables and Objects:
30
CMeasurementResult* pMeasurementResult - Newly created measurement result object. CObjectArray m_oaMeasurementResults - measurement result object array. CMemoryException* - perr - Memory exception object. Bugs: To do:
CPointResult * CHIPDoc::AddPointResult (CPoint Position, double Magnitude, BOOL Origional)

Description: This function is called to add a new point result to the point result object array. In more detail it does the following: 1. Creates a new point result object with the required attributes. 2. Next it checks that there is enough memory and adds the point result if there is. 3. When a new point result is added, the document is marked as dirty, which will prompt the user to save the document on exiting the application. Note: If there was not enough memory to add the point result, an error message will be displayed and the created point result object removed. This method also returns the newly created point result. Dependencies: Add(~) - Adds a new object to a CObjectArray. SetModifiedFlag() - Marks the document as dirty. AfxMessageBox(~) - Displays the specified text in a simple dialog box. Delete() - Used to delete an object of type CMemoryException. Variables and Objects: CPointResult* pPointResult - Newly created point result object. CObjectArray m_oaPointResults - Point result object array. CMemoryException* perr - Memory exception object. Bugs: To do:
void CHIPDoc::ClearBackGround ()
Description: This method clears the DIB section by reloading the backing DIB bitmap. Dependencies: Load(~) - Belongs to the DIBSectionLite class and loads a bitmap into the required DIB section. Variables and Objects: DIBSectionLite m_dibsection - Oject encapsulating DIB section. Bugs: To do:
void CHIPDoc::ClearPointResults ()
Description: When called, this method deletes the CPointResult objects contained within the m_oaPointResults object array. It also removes the pointers to the deleted point result objects. Dependencies: RemoveAll() - This method retrieves all the pointers from this array and frees all memory used for pointer storage. GetSize() - Returns (int) the number of objects in a CObjectArray.
31
Variables and Objects: int PRNum - Number of point results. CObjectArray m_oaPointResults - Applications point result object array. Bugs: To do:
void CHIPDoc::DeleteAgent (int nIndex)

Description: This method deletes the specified agent (nIndex) from the agent object array. Dependencies: RemoveAt(~) - Removes the specified object from a CObjectArray. Variables and Objects: int nIndex - The position of the object in the CObjectArray. CObjectArray m_oaAgents - Agent object array. Bugs: To do:
void CHIPDoc::DeleteAntenna (int nIndex)

Description: This method deletes the specified antenna (nIndex) from the antenna object array and invalidates any current tracer lines. Dependencies: RemoveAt(~) - Removes the specified object from a CObjectArray. Variables and Objects: int nIndex - The position of the object in the CObjectArray. CObjectArray m_oaAntenas - Antenna object array. BOOL m_bTracerLinesValid - FALSE: Current tracer lines are - invalid. Bugs: To do:
void CHIPDoc::DeleteCommand (int nIndex)

Description: This method deletes the specified command (nIndex) from the command object array. Dependencies: RemoveAt(~) - Removes the specified object from a CObjectArray. Variables and Objects: int nIndex - The position of the object in the CObjectArray. CObjectArray m_oaCommands - Command object array. Bugs: To do:
void CHIPDoc::DeleteContents () [virtual]

Description: This method is called each time a document is closed and does the following: 1. Reloads a blank backing DIB. 2. Gets the number of objects in the lines, Commands, Antennas, Measurements and Point Results arrays.
32
3. Loops through each one of the above mentioned object arrays and deletes their contained objects. 4. Resets each of the above arrays as by deleting their objects, you don't remove the pointers to them. 5. Resets the current drawing colour to white and resets the current drawing position to (0m,0m). 6. Shows the above changes on the documents active view. Dependencies: Load(~) - Belongs to the DIBSectionLite class and creates a DIB section from the specified bitmap (in this case a blank one). GetSize() - Returns (int) the number of objects in a CObjectArray. RemoveAll() - This method retrieves all the pointers from this array and frees all memory used for pointer storage. UpdateView() - Updates the documents active view. Variables and Objects: int LiNum - Number of lines. int CoNum - Number of commands. int AnNum - Number of antennas. int PRNum - Number of point results. int MNum - Number of measurements. CObjectArray m_oaLines - Applications line object array. CObjectArray m_oaCommands - Applications command object array. CObjectArray m_oaAntennas - Applications antenna object array. CObjectArray m_oaMeasurements - Applications measurement object array. CObjectArray m_oaPointResults - Applications point result object array. COLORREF m_Colour - The current drawing colour (white). CPoint m_ptCurrent - The current drawing point (0m,0m). Bugs: To do:
void CHIPDoc::DeleteLine (int nIndex)

Description: This method deletes the specified line (nIndex) from the line object array and invalidates any current tracer lines. Dependencies: RemoveAt(~) - Removes the specified object from a CObjectArray. Variables and Objects: int nIndex - The position of the object in the CObjectArray. CObjectArray m_oaLines - Line object array. BOOL m_bTracerLinesValid - FALSE: Current tracer lines are invalid. Bugs: To do:
void CHIPDoc::DeleteMeasurement (int nIndex)

Description: This method deletes the specified measurement (nIndex) from the measurement object array. Dependencies:
33
RemoveAt(~) - Removes the specified object from a CObjectArray. Variables and Objects: int nIndex - The position of the object in the CObjectArray. CObjectArray m_oaMeasurements - Measurement object array. Bugs: To do:
void CHIPDoc::DeletePointResult (int nIndex)

Description: This method deletes the specified point result (nIndex) from the point result object array. Dependencies: RemoveAt(~) - Removes the specified object from a CObjectArray. Variables and Objects: int nIndex - The position of the object in the CObjectArray. CObjectArray m_oaPointResults - Point Result object array. Bugs: To do:
void CHIPDoc::DisplayAntennas ()
Description: This method redraws the antenna objects in the m_oaAntennas object array on the DIB section (m_dibsection). The way it does this is by creating a dummy memory device context which it passes into the GetMemoryDC(~) method to obtain a device context with the DIB section pre-selected in it. Having done this standard GDI drawing functions can be used through the created device context on the DIB section. To draw the antennas, the following process is followed: 1. Determine the number of antennas in the antenna object array. 2. If their are any antennas, loop through all the antennas and draw them. On completion, the method releases the obtained device context such that other methods can work with the DIB section. Dependencies: GetMemoryDC(~) - Belongs to the DIBSectionLite class and creates a device context with the specified DIB section selected. ReleaseMemoryDC() - Belongs to the DIBSectionLite class and releases the DIB section from the created device context. GetAntennaCount() - Belongs to the document class and returns the number of user entered antennas. Draw(~) - Belongs to the CAntenna class and draws the antenna object on the specified device context. GetAntenna(~) - Belongs to the document class and returns the specified - antenna object. Variables and Objects: int AntCount - Stores the number of entered antennas. CAntenna* pAntenna - Temp object used to retrieve CAntenna objects. CDC* pDC - Pointer to device context with DIB section selected. CDC* dc - Pointer to dummy device context DIBSectionLite m_dibsection - Oject encapsulating DIB section. Bugs: To do:
34
void CHIPDoc::DisplayLines ()
Description: This method redraws the line objects in the m_oaLines object array on the DIB section (m_dibsection). The way it does this is by creating a dummy memory device context which it passes into the GetMemoryDC(~) method to obtain a device context with the DIB section pre-selected in it. Having done this standard GDI drawing functions can be used through the created device context on the DIB section. To draw the lines, the following process is followed: 1. Determine the number of lines in the line object array. 2. If their are any lines, loop through all the lines and draw them. On completion, the method releases the obtained device context such that other methods can work with the DIB section. Dependencies: GetMemoryDC(~) - Belongs to the DIBSectionLite class and creates a device context with the specified DIB section selected. ReleaseMemoryDC()- Belongs to the DIBSectionLite class and releases the DIB section from the created device context. GetLineCount() - Belongs to the document class and returns the number of user entered lines. Draw(~) - Belongs to the CLine class and draws the line object on the specified device context. GetLine(~) - Belongs to the document class and returns the specified line object. Variables and Objects: int LineCount - Stores the number of entered lines. CLine* pLine - Temp object used to retrieve CLine objects. CDC* pDC - Pointer to device context with DIB section - selected. CDC* dc - Pointer to dummy device context DIBSectionLite m_dibsection - Object encapsulating DIB section. Bugs: To do:
void CHIPDoc::DisplayMeasurements ()
Description: This method redraws the measurements in the m_oaMeasurements object array on the DIB section (m_dibsection). The way it does this is by creating a dummy memory device context which it passes into the GetMemoryDC(~) method to obtain a device context with the DIB section pre-selected in it. Having done this standard GDI drawing functions can be used through the created device context on the DIB section. To draw the measurements, the following process is followed: 1. Determine the number of measurements in the measurement object array. 2. If their are any measurements, loop through all the measurements and draw them. On completion, the method releases the obtained device context such that other methods can work with the DIB section. Dependencies: GetMemoryDC(~) - Belongs to the DIBSectionLite class and creates a device context with the specified DIB section selected. ReleaseMemoryDC() - Belongs to the DIBSectionLite class and releases the DIB section from the created device context. GetMeasurementCount()- Belongs to the document class and returns the number of user entered measurements.
35
Draw(~) - Belongs to the CMeasurment class and draws the measurements on the specified device context. GetMeasurement(~) - Belongs to the document class and returns the specified measurement object. Variables and Objects: int MeasCount - Stores the number of entered measurements. CMeasurement* pMeasurement - Temp object used to retrieve CMeasurement objects. CDC* pDC - Pointer to device context with DIB section selected. CDC* dc - Pointer to dummy device context DIBSectionLite m_dibsection - Oject encapsulating DIB section. Bugs: To do:
CAgent * CHIPDoc::GetAgent (int nIndex)

Description: This method returns a pointer to the specified agent (nIndex) in the agent object array. Dependencies: Variables and Objects: int nIndex - The position of the object in the CObjectArray. CObjectArray m_oaAgents - Agent object array. Bugs: To do:
int CHIPDoc::GetAgentCount ()
Description: Returns (int) the number of agents in the agent object array. Dependencies: GetSize() - Returns (int) the number of objects in a CObjectArray. Variables and Objects: CObjectArray m_oaAgentss - Agent object array. Bugs: To do:
CAntenna * CHIPDoc::GetAntenna (int nIndex)

Description: This method returns a pointer to the specified antenna (nIndex) in the antenna object array. Dependencies: Variables and Objects: int nIndex - The position of the object in the CObjectArray. CObjectArray m_oaAntennas - Antenna object array. Bugs: To do:
int CHIPDoc::GetAntennaCount ()
Description: Returns (int) the number of antennas in the antenna object array. Dependencies: GetSize() - Returns (int) the number of objects in a CObjectArray.
36
Variables and Objects: CObjectArray m_oaAntennas - Antenna object array. Bugs: To do:
CCommand * CHIPDoc::GetCommand (int nIndex)

Description: This method returns a pointer to the specified command (nIndex) in the command object array. Dependencies: Variables and Objects: int nIndex - The position of the object in the CObjectArray. CObjectArray m_oaCommands - Command object array. Bugs: To do:
int CHIPDoc::GetCommandCount ()
Description: Returns (int) the number of commands in the command object array. Dependencies: GetSize() - Returns (int) the number of objects in a CObjectArray. Variables and Objects: CObjectArray m_oaCommands - Command object array. Bugs: To do:
CLine * CHIPDoc::GetLine (int nIndex)

Description: This method returns a pointer to the specified line (nIndex) in the line object array. Dependencies: Variables and Objects: int nIndex - The position of the object in the CObjectArray. CObjectArray m_oaLines - Line object array. Bugs: To do:
int CHIPDoc::GetLineCount ()
Description: Returns (int) the number of lines in the line object array. Dependencies: GetSize() - Returns (int) the number of objects in a CObjectArray. Variables and Objects: CObjectArray m_oaLines - Line object array. Bugs: To do:
CMeasurementResult * CHIPDoc::GetMeasurement (int nIndex)

Description: This method returns a pointer to the specified measurement (nIndex) in the measurement object array. Dependencies:
37
Variables and Objects: int nIndex - The position of the object in the CObjectArray. CObjectArray m_oaMeasurements - Measurement object array. Bugs: To do:
int CHIPDoc::GetMeasurementCount ()
Description: Returns (int) the number of measurements in the measurement object array. Dependencies: GetSize() - Returns (int) the number of objects in a CObjectArray. Variables and Objects: CObjectArray m_oaMeasurements - Measurement object array. Bugs: To do:
void CHIPDoc::GetNextPoint ()
Description: This method calculates the agents next point, given only the agents initial point and a line gradient. To do this an adapted Bresenhams Algorithm which is based on the mid point theorem is used. It was chosen as it requires only integer additions and subtractions which are very fast. In using this approach we need to consider each one of the octants separately. Note: We only ever consider the first agent in the agent object array as once it has been terminated, the next agent will take its place. Dependencies: GetAgent(~) - Returns the specified CAgent object from the agent object - array. Variables and Objects: CAgent* agent - The first agent object of the CAgent object array. Bugs: To do:
CPointResult * CHIPDoc::GetPointResult (int nIndex)

Description: This method returns a pointer to the specified point result (nIndex) in the point result object array. Dependencies: Variables and Objects: int nIndex - The position of the object in the - CObjectArray. CObjectArray m_oaPointResults - Point Result object array. Bugs: To do:
int CHIPDoc::GetPointResultCount ()
Description: Returns (int) the number of point results in the point result object array. Dependencies: GetSize() - Returns (int) the number of objects in a CObjectArray. Variables and Objects: CObjectArray m_oaPointResults - Point result object array. Bugs: To do:
38
BOOL CHIPDoc::OnOpenDocument (LPCTSTR lpszPathName) [virtual]

Description: Called each time a document is opened, this method displays the document lines, antenna and any made measurements. Dependencies: DisplayLines() - Displays the documents lines. DisplayAntennas() - Displays the documents antenna. DisplayMeasurements() - Displays the documents measurements. Variables and Objects: Bugs: To do:
void CHIPDoc::OnSettingsDrawingsettings () [protected]

Description: This method is called when the Drawing Settings button in the Settings menu option is pressed. It initially creates and object of type DrawingSettingDlg which it will later use to display the Drawing Settings dialog. It then initializes the Drawing Setting dialog variables and displays the Drawing Setting dialog. On the termination of the Drawing Settings dialog (with OK), the method finally updates the corresponding document class member variables with those altered in the Drawing Setting dialog. Note: The final two lines of this method, calculate and set the application pixel size in meters given the user specified drawing size. The larger dimension of the specified drawing size will be used in this calculation. Dependencies: DoModal() - Belongs to the CDialog class and displays the corresponding modal dialog and returns the dialog - result when done. pow(~) - Raises a specified double to another specified double. Variables and Objects: DrawingSettingDlg dlg - Dialog object of type DrawingSettingDlg. CSize m_csDrawingSize - CSize variable that contains the user specified drawing size. double m_dPixelSize - The calculated pixel size. Bugs: To do:
void CHIPDoc::OnSettingsSimulationsettings () [protected]

Description: This method is called when the Simulation Settings button in the Settings menu option is pressed. It initially creates and object of type SimulationSettingDlg which it will later use to display the Simulation Settings dialog. It then initializes the Simulation Setting dialog variables and displays the Simulation Setting dialog. On the termination of the Simulation Settings dialog (with OK), the method finally updates the corresponding document class member variables with those altered in the Simulation Setting dialog. Dependencies: DoModal() - Belongs to the CDialog class and displays the corresponding modal dialog and returns the dialog result when done. Variables and Objects: SimulationSettingDlg dlg - Dialog object of type SimulationSettingDlg. int m_InitialAgentCount - Both the dialog and document classes have this member variable which specifies the number of initial agents to use. Bugs: To do:
39
void CHIPDoc::OnUpdateDistanceMoved (CCmdUI * pCmdUI)

Description: This method is called each time the current position of the mouse pointer changes (OnMouseMove). It gets a CString representation of the mouse pointers displacement whilst the left mouse button is depressed, enables the status bar pane and then changes its text to that of the mouse pointers displacement. Note: This displacement is in meters. Dependencies: Format(~) - This function writes formatted data to a CString. In this method it is used to get CString representations of a double. Enable(~) - This method enables or disables the required user-interface item. SetText(~) - This function sets the text of the required user-interface item. Variables and Objects: CString str - A CString of the mouse pointers displacement. CCmdUI* pCmdUI - Used only within an ON_UPDATE_COMMAND_UI handler, an object of this type is used to govern the functioning of an interface item. Bugs: To do:
void CHIPDoc::OnUpdateIndicatorX (CCmdUI * pCmdUI)

Description: This method is called each time the current x-position of the mouse pointer needs updating (OnMouseMove). It gets a CString representation of the mouse pointers current x-position, enables the status bar pane and then changes its text to that of the mouse pointers current x-position. Dependencies: Format(~) - This function writes formatted data to a CString. In this method it is used to get CString representations of an integer. Enable(~) - This method enables or disables the required user-interface item. SetText(~) - This function sets the text of the required user-interface item. Variables and Objects: CString str - A CString of the mouse pointers current x-position. CCmdUI* pCmdUI - Used only within an ON_UPDATE_COMMAND_UI handler, an object of this type is used to govern the functioning of an interface item. Bugs: To do:
void CHIPDoc::OnUpdateIndicatorY (CCmdUI * pCmdUI)

Description: This method is called each time the current y-position of the mouse pointer needs updating (OnMouseMove). It gets a CString representation of the mouse pointers current y-position, enables the status bar pane and then changes its text to that of the mouse pointers current y-position. Dependencies: Format(~) - This function writes formatted data to a CString. In this method it is used to get CString representations of an integer. Enable(~) - This method enables or disables the required user-interface item. SetText(~) - This function sets the text of the required user-interface item. Variables and Objects: CString str - A CString of the mouse pointers current y-position.
40
CCmdUI* pCmdUI - Used only within an ON_UPDATE_COMMAND_UI handler, an object of this type is used to govern the functioning of an interface item. Bugs: To do:
void CHIPDoc::OnUpdateSelectedMaterial (CCmdUI * pCmdUI)

Description: This method is called each time the selected drawing material is changed. It simple enables the required status bar pane and replaces its text with that of the newly selected materials name. Dependencies: Enable(~) - This method enables or disables the required user-interface item. SetText(~) - This function sets the text of the required user-interface item. Variables and Objects: CCmdUI* pCmdUI - Used only within an ON_UPDATE_COMMAND_UI handler, an object of this type is used to govern the functioning of an interface item. Bugs: To do:
void CHIPDoc::Serialize (CArchive & ar) [virtual]

Description: This method reads or writes the required object arrays as well as the specified drawing size to or from an archive. Dependencies: IsStoring() - TRUE if the archive is storing data. Serialize(~) - Does the actual reading/writing from/to an archive. UpdateAllViews(~) - Used to reload the command history list box. Variables and Objects: CObjectArray m_oaLines - Applications line object array. CObjectArray m_oaCommands - Applications command object array. CObjectArray m_oaAntennas - Applications antenna object array. CObjectArray m_oaMeasurements - Applications measurement object array. CArchive& ar - Archive being read/written. Bugs: To do:
void CHIPDoc::UpdateView ()
Description: When called, this method updates all of the documents views. Dependencies: UpdateAllViews(~) - NULL: Update all views. Variables and Objects: Bugs: To do:
2.7 CHIPView CHIPView::CHIPView () [protected]

Description: This method is used to initialize variables that will be used in this class to the required values.
41
Dependencies: NullifyPointerArray() - This method sets the entire point result pointer array to NULL. Variables and Objects: CSize m_BackGroundSize - The size of the drawing area in pixels (5000 by - 5000). double m_ZoomFactor - The initial drawing zoom factor. double pi - PI. BOOL m_bBoundary - TRUE: A boundary has been specified. CPoint m_cpBoundaryTL - Top left corner of specified boundary. CPoint m_cpBoundaryBR - Bottom right corner of specified boundary. Bugs: To do:
BOOL CHIPView::AntMeasPosOk ()
Description: This methods checks that the user specified boundary encompasses all of the entered measurements and antenna. It does this by looping through all of the measurement objects and the antenna object, comparing their positions to that of the specified boundary. Dependencies: GetDocument() - Returns a pointer to the active document class. GetMeasurementCount() - Returns the number of user entered measurements. GetMeasurement(~) - Returns a pointer to the specified CMeasurementResult object. AfxMessageBox(~) - Simply displays the specified text in a message box. GetAntenna(~) - Returns a pointer to the specified CAntenna object. Variables and Objects: int m_iNumMeasResults - The number of entered measurements. CMeasurementResult* pMeasResult - A pointer to a CMeasurementResult object. CAntenna* pAntenna - A pointer to a CAntenna object. Bugs: To do:
BOOL CHIPView::AssessAgent (CAgent * pAgent)

Description: This method is called by the OnBasicLog() method. As its input parameter it takes in a pointer to a CAgent object. Using this pointer, it checks that the agents current position is within the raster. If the agent is found to not be within the raster or if it is found to be outside of the specified simulation boundary, it is deleted and TRUE returned. Dependencies: GetDocument() Gets a pointer to the active document. DeleteAgent() - Belongs to the document class and deletes the agent at the specified index. Variables and Objects: CAgent* pAgent - Pointer to object of type CAgent. Bugs: To do:
42
void CHIPView::AveragePointResults ()
Description: This method is used to approximate point results at points where no point results were calculated. It does this by taking a weighted average of the four closest known point results. In more detail, it does the following: 1. Loops through the point result pointer matrix until a null pointer (i.e. no point result calculated). 2. Using this null pointers position as a reference, it then moves in the -ve and +ve x and y directions until one of the following is found: a) An original point result (no an averaged one). b) A wall of and type (you don't want to average over walls as this causes inaccuracy). It does this by considering the corresponding pixel colour and any colour other than black or white will be considered a wall or obstacle. c) The specified simulation boundary is left. 3. Whilst doing this (2.) the number of moves is recorded to each valid point result as it will be needed in the weighted average calculation and on finding a valid point result, the variable m_bValidPR is set to TRUE prompting the following point (4.). 4. Having found a valid point result, the number of moves taken to the point result as well as the point results magnitude are stored in corresponding position in two different arrays. The total number of moves to all the valid point results is also stored. These values will be used later. 5. The final step in the method, calculates the magnitude of the new point result based on a weighted average of the above mentioned obtained valid point results and generates the required point result at the previously null position. Note: This method is very selective and will use the maximum number of obtainable, valid, point results (i.e. 1 being the minimum and 4 being the maximum). Dependencies: GetDocument() - Returns a pointer to the active document class. GetMemoryDC(~) - Belongs to the DIBSectionLite class and creates a device context with the specified DIB section selected. ReleaseMemoryDC() - Belongs to the DIBSectionLite class and releases the DIB section from the created device context. SetUpFeedBackDisplay(~) - Initializes the feedback display with the required heading and range. TidyUpFeedBackDisplay(~) - Removes the feedback heading and sets the current feedback process control position to 0. GetPixel(~) - Returns (COLORREF) the colour of the specified pixel. Format(~) - This method writes formatted string data into a string. In this method, it is used to convert a double variable to a CString. pow(~) - Raises the specified double to a specified power (in this case -1 or division). GetParentFrame() - Returns a pointer to the applications Frame class. Variables and Objects: CDC* dc - A pointer to a dummy device context. CDC* pDC - A pointer to a device context with the backing DIB pre-selected. CPoint m_cpBoundaryTL - The top left corner of the simulation boundary. CPoint m_cpBoundaryBR - The bottom right corner of the simulation boundary. int m_dADividend[4] - Stores the number of moves taken in each direction when finding an original point result. Double m_dMagnitude[4] - Stores the magnitudes of the found point results in each direction.
43
Double m_dTMagnitude - The calculated average of the surrounding point results. int m_dTDividend - The total number of moves taken in each direction. int m - Tracking variable which holds the current x value of the point result being calculated. int n - Tracking variable which holds the current y value of the point result being calculated. int o - Tracking variable which holds the number of moves taken to a valid point result. COLORREF m_crPixel The colour of the current pixel in the tracking algorithm. BOOL m_bValidPR - TRUE when a valid, original point result is found. CPoint m_cpPoint - The position of the newly averaged point result. CPointResult* pResult - The newly averaged point result. CPointResult* m_pPRMatrix[][] - The point result pointer array. Bugs: To do:
void CHIPView::ClearSimResults ()
Description: When this method is called, it clears any simulation results by reloading the DIB section with a blank bitmap and then by redisplaying the required walls, Measurements and antennas on the blank DIB section. Dependencies: GetDocument() - Returns a pointer to the active document. Load(~) - Belongs to the DIBSectionLite class and creates a DIB section from the specified bitmap (in this case a - blank one). DisplayLines() - From the document class and loops through all the lines in the line object array, redrawing them. DisplayAntennas() - From the document class and loops through all the antennas in the antenna object array, redrawing them. DisplayMeasurements() - From the document class and loops through all the measurement results in the measurement results object array, redrawing them. Variables and Objects: CDIBSectionLite m_dibsection - Created DIB section on which all application drawing takes place. Bugs: To do:
void CHIPView::DisplayPointResults ()
Description: This method is called when the generated point results need drawing. The method initially creates a memory device context with the backing DIB pre-selected in it. It them loops through the point result pointer matrix (m_pPRMatrix) and draws the corresponding point results if their pointers exist. The drawing is done using the Draw(~) method of the point result class which takes as its parameter, the above mentioned device context. The reason behind keeping a 2-D array of pointers to the point result objects it to allow for the direct accessing of point result objects using relative screen co-ordinate addressing. Note: this method also sets up the feedback display in the applications frame with the required heading and range (SetUpFeedBackDisplay(~)) and on completion clears and resets the feedback display (TidyUpFeedBackDisplay()). Dependencies: GetDocument() - Gets a pointer to the active document.
44
Draw(~) - Belongs to the CPointResult class and draws the Point result object on the specified device context. GetMemoryDC(~) - Belongs to the DIBSectionLite class and creates a device context with the specified DIB section - selected. ReleaseMemoryDC() - Belongs to the DIBSectionLite class and releases the DIB section from the created device context. SetUpFeedBackDisplay(~) - Initializes the feedback display with the required heading and range. TidyUpFeedBackDisplay(~) - Removes the feedback heading and sets the current feedback process control position to 0. Variables and Objects: CDC* dc - Pointer to dummy memory device context. CDC* pDC - Pointer to memory device context. int i - Used to loop throungh m_pPRMatrix[i][]. int j - Used to loop throungh m_pPRMatrix[][j]. DIBSectionLite m_dibsection - Object encapsulating DIB section. CPointResult* m_pPRMatrix[][] - Matrix of pointers to CPointResult objects used to both keep track of the generated CPointResult objects and allow them to be easily accessible via the corresponding screen co-ordinates. Bugs: To do:
CPoint * CHIPView::ExtractCoOrdinates (LPCTSTR m_In)

Description: This method takes as its input, the user entered command. By considering is structure with regards to the position of certain characters ('[',',', and ']') it extracts the required co-ordinates. It then validates these co-ordinates by initially converting them the pixel values (uses the user specified drawing size) and then by ensuring that they lie within the raster. It returns the validated co-ordinates in pixel values. In more detail: 1. Gets a CString representation of the LPCTSTR input argument. 2. Finds the position if the first bracket, comma and last bracket in the newly created CString. 3. CHECK 1: Ensures that the above extracted characters are in the correct sequence. 3. CHECK 2: Ensures that something exists between the first bracket and the comma as well as the comma and the last bracket. 4. Removes the sections of the CString between the first bracket and the comma and the comma and the second bracket. 5. Converts these values To doubles (will be in meters). 6. Finds out how many pixels the above double values are, given the user specified drawing size. 7. CHECK 3: Ensures that the calculated pixel values are still in the raster. 8. Returns a pointer to a CPoint object containing the required co-ordinates in pixel values. NOTE: If any of the above checks fail, a NULL pointer is returned. Dependencies: Find(~) - Searches a string for the first match of a substring and returns its index (indexed from 0). Mid(~) - Extracts a substring of specified length and starting index.
45
atof(~) - Converts a character string to a double-precision floating-point value. GetDocument() - Gets a pointer to the active document. pow(~) - Raises the specified double to a specified power (in this case -1 or division). Variables and Objects: CString m_csIn - CString representation of input argument. Int FirstB - Position of first bracket in input argument. Int Comma - Position of comma in input argument. Int LastB - Position of last bracket in input argument. CString m_X - Extracted CString between first bracket and comma. CString m_Y - Extracted CString between comma and last bracket. Double m_dX - Double representation of m_X. Double m_dY - Double representation of m_Y. Double m_dPixelSize - Pixel size, given user specified drawing size. Int m_iX - Pixel representation of m_dX. Int m_iY - Pixel representation of m_dY. CPoint m_Return - CPoint of above calculated pixel values. CPoint* cpReturn - Pointer to above CPoint object which gets returned. Bugs: To do:
CString CHIPView::ExtractLetters (LPCTSTR in)

Description: Returns the passed in CString without the following characters: 1,2,3,4,5,6,7,8,9,0,[,],(,) and . Dependencies: Remove(~) - Removes the specified character from a CString. Variables and Objects: CString m_csIn - CString representation of passed in LPCTSTR to allow for CString operations. Bugs: To do:
void CHIPView::GeneratePrimaryAgents ()
Description: This method generates a specified number of agents for each antenna placed. Its only required user input is that of the specified number of primary agents set throungh the simulation settings dialog. The detailed Description of this methods operation is as follows: 1. Gets the number of user specified antennas. 2. Gets the user specified initial agent count. 3. CHECK 1: Are there any antennas? NO: Exit method YES: Proceed 4. Create an object of type CAntenna to gain access to the user specified antenna objects. 5. Loop through the specified antennas and generate the required number of primary agents. NOTE: The line drawing algorithm is octant dependent so each primary agent will be generated for a specific octant i.e. 1-8. On the creation of an agent object, its origin, source antenna, magnitude offset and gradient information is also specified. The octancy dependency results in 8 times the number of user specified primary agents being drawn.
46
Dependencies: GetDocument() - Returns a pointer to the active document class. GetAntennaCount() - Returns (int) the number of user specified antennas. GetAntenna(~) - Returns a pointer to the specified antenna in the antenna object array found in the document class. AddAgent(~) - Adds the specified agent to the agent object array of the document class. Variables and Objects: int AntCount - The number of user specified antennas. int m_InitialAgentCount - The number of user specified primary agents. CAntenna *AntTemp - Pointer to object of type CAntennas to allow access to the user specified antennas. int AntPos - A for loop stepping variable used to iterate through the entered antennas. Bugs: To do:
void CHIPView::GetSimulationConstants ()
Description: This method, when called, updates the views simulation constants variables with the corresponding variables in the document class. This method is called before any of the simulation mechanisms are carried out. Dependencies: GetDocument() - Returns a pointer to the active document class. Variables and Objects: BOOL m_bMeasurement - TRUE: Intersection by measurement results. BOOL m_bDataBase - TRUE: Intersection by material coefficients found in database. Double m_dPixelSize - The user specified pixel size (i.e. depends on specified floor plan size). Double m_dSimCutOffLevel - The specified simulation cut-off level. Double m_dFirst_Zone_Exp - The specified first zone path loss exponent. Double m_dOther_Zone_Exp - The other than first zone path loss exponent. Double m_dDiameter_First_Zone - The specified diameter of the first zone. Double m_dAntLoss - The specified antenna loss at 1m. Bugs: To do:
void CHIPView::InhibitAgentActivity (CAgent * pAgent)

Description: Prevents an agent from generating a point result whilst it is within 1m from its source antenna. This is done as the utilized propagation equations were normalized at 1m to gain antenna independence. It does this by calculating the distance between the agents current point and it origin and compares it to unity. NOTE: The first line of code gives the agent a dummy magnitude value (not a possible value) such that the AssessAgent method can be used as it will cause an assert if the state of the agent magnitude is undefined. Dependencies: GetDocument() - Gets a pointer to the active document class. GetNextPoint() - Calculates the agents next point and moves the agent.
47
AssessAgent(~) - Checks to see whether the agent is still in the raster and deletes it if it is not. pow(~) - Belongs to the maths class and raises a specified number to another specified number. Variables and Objects: double m_dPixelSize - The user specified pixel size (determined from the user specified drawing size). double m_dDist - The calculated distance the agent has moved. CAgent* pAgent - Methods input argument. Bugs: To do:
void CHIPView::IntersectionByMeasurement (CAgent * pAgent, CDC * pDC, CMaterialSet * pSet)

Description: This method handles an intersection when the from measurement simulation mechanism is used. In more detail: 1. Gets the colour of the intersected wall. 2. Loops through the applications materials database to determine the transmission coefficient of the intersected wall. 3. Loops through the measurement result object array to determine if a measurement result exists for the intersected material. 4. If no measurement result was found, the materials name is added to the agents intersection history where as if a measurement result is found, the agents magnitude is reduced accordingly. Note: If the angle of arrival boolean variable is set and a and a measurement result is found, then the following is carried out: 1. A check is carried out to determine if the wall is vertical or horizontal. 2. Depending on the outcome of 1., two different approaches are used. One for octants 1, 8, 4 and 5 and the other for octants 2, 3, 6 and 7. In each of the above approaches the angle of arrival is calculated and used to alter the walls transmission coefficient (Coeff/cos( angle of arrival)). Dependencies: GetPixel(~) - Returns the colour of the specified pixel in a device context. MoveFirst() - Moves to the first record of a record set. IsEOF() TRUE until the end of the record set is reached. GetRValue(~) - Returns (int) the red component of a COLORREF variable. GetGValue(~) - Returns (int) the Green component of a COLORREF variable. GetBValue(~) - Returns (int) the Blue component of a COLORREF variable. MoveNext() - Moves to the next record in a record set. GetDocument() - Returns a pointer to the active document. log10(~) - Returns the log base 10 of the specified number. DeleteAgent(~) - Deletes an agent at the specified index from the agent object array. pow(~) - Raises a specified double to the power of another specified double. GetMeasurementCount() - Belongs to the document class and returns the number of user entered measurements. GetMeasurement(~) - Belongs to the document class and returns the specified measurement object. Format(~) - This function writes formatted data to a CString.
48
Empty() - Empties associated CString object. Variables and Objects: COLORREF m_colour - Variable which stores the intersected wallscolour. CAgent* pAgent - Input argument (the intersecting agent). CMeasurementResult* pMR - Pointer to measurement object. CDC* pDC - Input argument (a DC with backing DIB selected). CMaterialSet* pSet - Input argument (the materials record set). BOOL m_bNotEqual - TRUE if intersected wall found in database. BOOL M_bMeasResult - TRUE if valid measurement result is found. BOOL m_bAngleOfArrival - TRUE if angle of arrival is required in calculation. double m_dOffset - The dB representation if walls transmission coefficient. Bugs: To do:
void CHIPView::IntersectionBySimulation (CAgent * pAgent, CDC * pDC, CMaterialSet * pSet)

Description: This method handles an intersection when the from database simulation mechanism is used. In more detail: 1. Gets the colour of the intersected wall. 2. Loops through the applications materials database to determine the transmission coefficient of the intersected wall. 3. Calculates the dB value of the transmission coefficient and add it to the agents magnitude offset. Note: If the angle of arrival boolean variable is set, then the following is carried out: 1. A check is carried out to determine if the wall is vertical or horizontal. 2. Depending on the outcome of 1., two different approaches are used. One for octants 1, 8, 4 and 5 and the other for octants 2, 3, 6 and 7. In each of the above approaches the angle of arrival is calculated and used to alter the walls transmission coefficient (Coeff/cos( angle of arrival)). Dependencies: GetPixel(~) - Returns the colour of the specified pixel in a device context. MoveFirst() - Moves to the first record of a record set. IsEOF() - TRUE until the end of the record set is reached. GetRValue(~) - Returns (int) the red component of a COLORREF variable. GetGValue(~) - Returns (int) the Green component of a COLORREF variable. GetBValue(~) - Returns (int) the Blue component of a COLORREF variable. MoveNext() - Moves to the next record in a record set. GetDocument() - Returns a pointer to the active document. log10(~) - Returns the log base 10 of the specified number. DeleteAgent(~) - Deletes an agent at the specified index from the agent object array. pow(~) - Raises a specified double to the power of another specified double. Variables and Objects: COLORREF m_colour - Variable which stores the intersected walls colour. CAgent* pAgent - Input argument (the intersecting agent). CDC* pDC - Input argument (a DC with backing DIB selected). CMaterialSet* pSet - Input argument (the materials record set). BOOL m_bNotEqual - TRUE if intersected wall found in database.
49
double m_dOffset - The dB representation if walls transmission coefficient. BOOL m_bAngleOfArrival - TRUE if angle of arrival is required in calculation. Bugs: To do:
void CHIPView::IsIntersection (CAgent * pAgent, CDC * pDC, CPoint PrevPoint, CMaterialSet * pSet)
Description: This method takes as its input parameters the current agent, a DC loaded with the backing DIB, the agents previous point and the materials database record set. It does this to test for intersections between agents and walls by doing the following: 1. Gets the colour of the agents current point. 2. Checks to see if the colour is not black or white (white and black represent free space and antennas respectively.) 3. Checks if the agents previous point was white i.e. the intersection has just occurred. 4. If as a result of the above checks it is determined that an intersection has occurred, the simulation mechanism is then examined and the appropriate method called. Dependencies: IntersectionBySimulation(~) - Method that handles an intersection when the simulation mechanism uses predefined reflection and transmission coefficients. IntersectionByMeasurement(~)- Method that handles an intersection when the simulation mechanism determines the required reflection and transmission coefficients from measurement. RGB(~) - Specifies a COLORREF variable. GetPixel(~) - Returns the colour of the specified pixel in a device context. Variables and Objects: CAgent *pAgent - A pointer to the current agent object. CDC *pDC - A pointer to a device context with the backing DIB pre-selected. CPoint PrevPoint - The current agents previous point. CMaterialSet* pSet - A pointer to the materials record set. BOOL m_bDataBase - TRUE if simulation uses predefined coefficients. BOOL m_bMeasurement - TRUE if simulation generates coefficients from measurements. Bugs:
void CHIPView::IsNewPointResult (CAgent * pAgent, CDC * pDC)

Description: This method calculates a new point result if no point result currently exists at the agents current position. It does this as follows: 1. CHECK 1: Does a point result exist at the agent current position and is the agents 2. Calculates the distance the agent has moved based on the value of m_bFresnelZone. TRUE indicates that the Fresnel zones should be used. 3. Using this distance, it calculates the agents free loss (also included is losses due to any previous intersections). 4. Using 2-4, the existing point result is altered to reflect changes due to the impinging agent 5. A record is also kept of all the impinging agents intersections at the point result and the agents magnitude is updated.
50
Dependencies: GetDocument() - Gets a pointer to the active document class. pow(~) - Belongs to the maths class and raises a specified number to another specified number. AddPointResult(~)- Adds a specified point result to the point result object array of the document class. log10(~) - Returns the log base 10 of the specified number. GetPixel(~) - Returns the colour (RGB) of the specified pixel. Variables and Objects: double m_dPixelSize - The user specified pixel size (determined from the user specified drawing size). CAgent* pAgent - Methods input argument. double m_dDistance - Agents current displacement. double m_dMagnitude - Agents magnitude due to its displacement. double m_dMagnitudeNew - The new calculated point result magnitude value. CPointResult* pResult - A pointer to an object of type CPointResult used to alter an existing point result. double pi - The number 3.14 etc. BOOL m_bFresnelZone - TRUE: Use Fresnel Zones. Bugs: To do:
void CHIPView::LineTo (LPCTSTR in)

Description: This method is responsible for drawing a line in response to a user LineTo command. It does this as follows: 1. Extracts the required co-ordinates using the ExtractCoOrdinates method which returns a pointer to a CPoint object, containing the "to" co-ordinates in pixel values. This method also validates the input command. 2. Gets the current or "from" co-ordinate from the document class. 3. Adds the required line to the line object array of the document class if a material has been selected. 4. Changes the "from" co-ordinate to the "to" co-ordinate in the document class if a material has been selected. 5. Adds the required command to the command object array of the document class if a material has selected. 6. Draws the line and finally updates the command list in the frame class. Dependencies: ExtractCoOrdinates(~) - Gets a pointer to a CPoint object containing the extracted pixel value representation of the user inputted command. GetDocument() - Gets a pointer to the active document. AddLine(~) - Adds a line object to the line object array of the document class. AddCommand(~) - Adds a command to the command object array of the document class. DisplayLines(~) - Belongs to the document class and redraws all the lines in the line object array on the backing DIB. ReDrawCommands(~) - Reloads all the command objects in the command object array in the command history list in the application frame. RGB(~) - Returns the specified COLORREF variable.
51
Variables and Objects: CPoint* m_cpCoOrd - Pointer to CPoint object containing required "to" pixel values. CPoint ptFrom - "from" CPoint object. CPoint ptTo - CPoint object containing "to" pixel values. CLine* pLine - Pointer to newly added CLine object. CCommand* pCommand - Pointer to newly added CComand object. Bugs: To do:
BOOL CHIPView::MeasurementPositionsOk ()
Description: This method checks that all the entered measurement results were placed over valid simulation results. Dependencies: GetDocument() - Returns a pointer to the active document. GetMeasurement(~) - Returns the specified CMeasurementResult object from the measurement object array of the document class. GetMeasurementCount() - Returns the number of entered measurement results. Variables and Objects: int m_iNumMeasResults - The number of entered measurement results. CMeasurementResult* pMeasResult - A pointer to a measurement result object used to step through them. Bugs: To do:
void CHIPView::MoveAgents ()
Description: This method is one of the applications core methods and does the following: 1. Generates the specified number of primary agents. 2. Sets-up the feedback display with the required text. 3. Opens the materials record set. 4. Gets a pointer to a device context with the required backing DIB pre-selected. 5. Loops through all the agents in the agent object array and moves them. On each agent move, its previous point is stored (used in the intersection testing) and its current position and magnitude validated to ensure that it has not reached the specified cut-off magnitude or left the specified simulation boundary. 6. During the agent moving, the agents activity is inhibited whilst it is within 1m of the antenna (magnitude calculations were normalized at 1m). 7. On the completion of this method, the feedback display is reset, the DIB section released and the material record set closed. Dependencies: GeneratePrimaryAgents() - Generates the user specified primary agents. SetUpFeedBackDisplay(~) - Initializes the feedback display with the required heading and range. TidyUpFeedBackDisplay(~) - Removes the feedback heading and sets the current feedback process control position to 0. Open() - Opens the specified database to allow access to its records. Close() - Closes the database. GetDocument() - Returns a pointer to the active document class.
52
GetMemoryDC(~) - Belongs to the DIBSectionLite class and creates a device context with the specified DIB section - selected. ReleaseMemoryDC() - Belongs to the DIBSectionLite class and releases the DIB section from the created device context. GetAgentCount() - Returns the number of agents in the agent object array. GetAgent(~) - Returns the specified CAgent object. InhibitAgentActivity(~) - Prevents the agents from creating point results when they are within 1m from their source antenna. IsNewPointResult(~) - Generates a point result from an agents move. GetNextPoint() - Returns the agents next position (based on a line drawing algorithm). GetParentFrame() - Returns a pointer to the applications Frame class. SetPos(~) - Belongs to the CProcessControl class and sets the current process control position to that specified. IsIntersection(~) - Handles the case in which an agent has interacted with a wall. SetPos(~) - Belongs to the CProcessControl class and sets the current process control position to that specified. Variables and Objects: int i - Tracks the number of agents deleted and is used in the feedback display process bar. CMaterialSet Set - Material database record set object. CMaterialSet* pSet - Pointer to above object. CDC* dc - A dummy device context pointer. CDC* pDC - A pointer to a device context with the required backing DIB pre-selected. CAgent* pAgent - A CAgent object used to contain the various agents used within this method. Bugs: To do:
void CHIPView::MoveTo (LPCTSTR in)

Description: This method is responsible for moving the drawing cursor in response to a user MoveTo command. It does this as follows: 1. Extracts the required co-ordinates using the ExtractCoOrdinates method which returns a pointer to a CPoint object, containing the "to" co-ordinates in pixel values. This method also validates the input command. 2. Updates the current position variable in the document class. 3. Finally adds the required command to the command object array of the document class. Dependencies: ExtractCoOrdinates(~) - Gets a pointer to a CPoint object containing the extracted pixel value representation of the user inputted command. GetDocument() - Gets a pointer to the active document. ReDrawCommands(~) - Reloads all the command objects in the command object array in the command history list in the application frame. Variables and Objects: CPoint* m_cpCoOrd - Pointer to CPoint object containing required "to" pixel values. CPoint ptTo - CPoint object containing "to" pixel values. CCommand* pCommand Pointer to newly added CComand object.
53
Bugs: To do:
void CHIPView::NewCommand (LPCTSTR in)

Description: This is called from the frame class after a user command has been entered (the frame class waits for a ']' character). It then gets a CString representation of the above to allow for string operations on the entered command. Having done this it extracts the text part of the entered command and calls the respective method. Dependencies: GetDocument() - Gets a pointer to the active document. Find(~) - Finds the index if the specified character in a CString. Left(~) - Extracts the portion of CString left of the specified index. MoveTo(~) - Moves the drawing cursor to the specified point. LineTo(~) - Draws a line from the current drawing cursor point to a specified point. Variables and Objects: CString m_In - CString representation of the user command. Bugs: To do:
void CHIPView::NullifyPointerArray ()
Description: Sets each pointer in the m_pPRMatrix[5000][5000] which tracks the generated point result objects to NULL. Dependencies: Variables and Objects: int i - Goes from 0 to 4999 and used to step through rows of m_pPRMatrix. int j - Goes from 0 to 4999 and used to step through columns of m_pPRMatrix. Bugs: To do:
void CHIPView::OnAdvancedLog () [protected]

Description: This method carries out the Advanced log simulation mechanism which accounts of free space loss, intersections, angle of arrival and Fresnel Zones. It does this as follows: 1. Gets the required simulation constants and sets up the required simulation environment. 2. Checks that the specified antenna lies within the specified simulation boundary and returns if it does not. 3. Sets the required simulation governing variables to their required values (m_bAngleOfArrival=TRUE and m_bFresnelZone=TRUE). 4. It then does the required moving of the generated agents governed by the variables set in 3. 5. Having done this, the deposited point results are averaged. This involves initially setting up the required feedback and on completion, resetting the feedback display. 6. The method terminates with code to redraw the generated and averaged point results, the specified antenna and any drawn lines. Dependencies:
54
SetUpFeedBackDisplay(~) - Initializes the feedback display with the required heading and range. TidyUpFeedBackDisplay(~) - Removes the feedback heading and sets the current feedback process control position to 0. AveragePointResults() - Generates point results at regions where they were not calculated by taking the weighted average of up to four of the closest simulated point results. GetSimulationConstants() - Gets the requires simulation constants. SetupSimulationEnviroment() - Sets up the simulation environment. AntMeasPosOk() - Returns TRUE if the specified antenna is within the specified simulation boundary. DisplayPointResults() - Displays all the point results in the point result object array. GetDocument() - Returns a pointer to the active document class. DisplayLines() - From the document class and loops through all the lines in the line object array, - redrawing them. DisplayAntennas() - From the document class and loops through all the antennas in the antenna object array, redrawing them. ReDraw() - Invalidates and redraws the view to reflect changes. MoveAgents() - Moves the specified number of agents around the raster. Their movement is governed by the specified simulation mechanism and the specified simulation constants. Variables and Objects: BOOL m_bAngleOfArrival - TRUE: Use angle of arrival. BOOL m_bFresnelZone - TRUE: Use Fresnel Zones. Bugs: To do:
void CHIPView::OnBasicLog () [protected]

Description: This method carries out the Basic log simulation mechanism which accounts of free space loss and intersections. It does this as follows: 1. Gets the required simulation constants and sets up the required simulation environment. 2. Checks that the specified antenna lies within the specified simulation boundary and returns if it does not. 3. Sets the required simulation governing variables to their required values (m_bAngleOfArrival=FALSE and m_bFresnelZone=FALSE). 4. It then does the required moving of the generated agents governed by the variables set in 3. 5. Having done this, the deposited point results are averaged. This involves initially setting up the required feedback and on completion, resetting the feedback display. 6. The method terminates with code to redraw the generated and averaged point results, the specified antenna and any drawn lines. Dependencies: SetUpFeedBackDisplay(~) - Initializes the feedback display with the required heading and range. TidyUpFeedBackDisplay(~) - Removes the feedback heading and sets the current feedback process control position to 0.
55
AveragePointResults() - Generates point results at regions where they were not calculated by taking the weighted average of up to four of the closest simulated point results. GetSimulationConstants() - Gets the requires simulation constants. SetupSimulationEnviroment() - Sets up the simulation environment. AntMeasPosOk() - Returns TRUE if the specified antenna is within the specified simulation boundary. DisplayPointResults() - Displays all the point results in the point result object array. GetDocument() - Returns a pointer to the active document class. DisplayLines() - From the document class and loops through all the lines in the line object array, redrawing them. DisplayAntennas() - From the document class and loops through all the antennas in the antenna object array, redrawing them. ReDraw() - Invalidates and redraws the view to reflect changes. MoveAgents() - Moves the specified number of agents around the raster. Their movement is governed by the specified simulation mechanism and the specified simulation constants. Variables and Objects: BOOL m_bAngleOfArrival - TRUE: Use angle of arrival. BOOL m_bFresnelZone - TRUE: Use Fresnel Zones. Bugs: To do:
void CHIPView::OnDraw (CDC * pDC) [virtual]

Description: This method is called each time the applications view requires redrawing. It initially gets a device context with the backing DIB pre-selected into it. Following this it maps the backing DIB onto the screen (taking into account the specified zoom factor) and releases the backing DIB such that it can be selected into another device context when required. Dependencies: GetMemoryDC(~) - Belongs to the DIBSectionLite class and creates a device context with the specified DIB section selected. ReleaseMemoryDC()- Belongs to the DIBSectionLite class and releases the DIB section from the created device context. StretchBlt(~) - Does the mapping from one device context to another, taking into account any size differences (used for zooming in/out). GetDocument() - Returns a pointer to the active document class. Variables and Objects: CDC* pDC - Pointer to device context with DIB section selected. CDC* dc - Pointer to dummy device context DIBSectionLite m_dibsection - Object encapsulating DIB section. Bugs: To do:
BOOL CHIPView::OnEraseBkgnd (CDC * pDC) [protected]

Description: The framework calls this member function when the CWnd object background needs erasing. Changed to return FALSE to ensure flicker free drawing. Dependencies: Variables and Objects:
56
Bugs: To do:
void CHIPView::OnHScroll (UINT nSBCode, UINT nPos, CScrollBar * pScrollBar) [protected]

Description: Called every time the horizontal scroll bar is moved. This method passes the current scroll position to the UpdateRulerInfo method to exact changes in the horizontal ruler display. It also invalidates the window forcing a redraw which reflects the scrolled to portion of the display. Dependencies: Invalidate(~) - Marks the display as "dirty" forcing a redraw. GetScrollPosition(~) - Returns the current scroll position (int). UpdateRulerInfo(~) - Updates the rulers to reflect a scroll change. Variables and Objects: UINT nSBCode - Specifies a scroll-bar code that indicates a scrolling request by the user i.e. SB_BOTTOM = Scroll to bottom. UINT nPos - Contains the current scroll-box position. N.B. dependant on which nSBCode is specified. CScrollBar* pScrollBar - Pointer to object of type CScrollBar which encapsulates the applications horizontal scroll bar. Bugs: To do:
void CHIPView::OnInitialUpdate () [protected, virtual]

Description: This method is called by the framework after the view is first attached to the document, but before the view is initially displayed. It has been altered to show the applications required rulers and initialize certain variables. Dependencies: NullifyPointerArray() - This method sets the entire point result pointer array to NULL. ShowRulers(~) - When TRUE is passed in, the applications rulers are shown. GetParentFrame() - Returns a pointer to the applications Frame class. Variables and Objects: CSize m_BackGroundSize - The size of the drawing area in pixels (5000 by - 5000). double m_ZoomFactor - The initial drawing zoom factor. double pi - PI. BOOL m_bBoundary - TRUE: A boundary has been specified. CPoint m_cpBoundaryTL - Top left corner of specified boundary. CPoint m_cpBoundaryBR - Bottom right corner of specified boundary. Bugs: To do:
void CHIPView::OnIntermediateLog () [protected]

Description: This method carries out the Intermediate log simulation mechanism which accounts of free space loss, intersections, angle of arrival and Fresnel Zones. It does this as follows: 1. Gets the required simulation constants and sets up the required simulation environment.
57
2. Checks that the specified antenna lies within the specified simulation boundary and returns if it does not. 3. Sets the required simulation governing variables to their required values (m_bAngleOfArrival=TRUE and m_bFresnelZone=FALSE). 4. It then does the required moving of the generated agents governed by the variables set in 3. 5. Having done this, the deposited point results are averaged. This involves initially setting up the required feedback and on completion, resetting the feedback display. 6. The method terminates with code to redraw the generated and averaged point results, the specified antenna and any drawn lines. Dependencies: SetUpFeedBackDisplay(~) - Initializes the feedback display with the required heading and range. TidyUpFeedBackDisplay(~) - Removes the feedback heading and sets the current feedback process control position to 0. AveragePointResults() - Generates point results at regions where they were not calculated by taking the weighted average of up to four of the closest simulated point results. GetSimulationConstants() - Gets the requires simulation constants. SetupSimulationEnviroment() - Sets up the simulation environment. AntMeasPosOk() - Returns TRUE if the specified antenna is within the specified simulation boundary. DisplayPointResults()- Displays all the point results in the point result object array. GetDocument() - Returns a pointer to the active document class. DisplayLines() - From the document class and loops through all the lines in the line object array, - redrawing them. DisplayAntennas() - From the document class and loops through all the antennas in the antenna object array, redrawing them. ReDraw() - Invalidates and redraws the view to reflect changes. MoveAgents() - Moves the specified number of agents around the raster. Their movement is governed by the specified simulation mechanism and the specified simulation constants. Variables and Objects: BOOL m_bAngleOfArrival - TRUE: Use angle of arrival. BOOL m_bFresnelZone - TRUE: Use Fresnel Zones. Bugs: To do:
void CHIPView::OnLButtonDown (UINT nFlags, CPoint point) [protected]

Description: This method is called each time the left mouse button is depressed and responds in either of two ways: A. If the shift button is depressed. 1. Invalidates any current tracer lines 2. If it is the first time this combination has been depressed, a device context is obtained with the backing DIB pre-selected onto which a dot is drawn (at the current mouse position) to indicate a possible corner of the simulation boundary to be specified.
58
3. If it is the second time this combination is used, once again a device context is obtained with the backing DIB pre-selected onto which a rectangle is drawn having opposite corners corresponding to the selected point in 2 and this latest selected point. Having done this, the user specified lines, antennas and measurement results are redrawn. Following this the final stage in this case determines out of the two specified points, which one constitutes the top left and bottom right corners of the simulation boundary rectangle. B. If no key depressed The mouse cursor is captured and the starting point of a possible line (which is a function of both the current point and the current scroll bar positions) is set. Dependencies: GetKeyState(~) - Returns the state of the specified system key. If less than zero, indicates that key is depressed. GetDocument() - Returns a pointer to the active document class. GetMemoryDC(~) - Belongs to the DIBSectionLite class and creates a device context with the specified DIB section - selected. ReleaseMemoryDC() - Belongs to the DIBSectionLite class and releases the DIB section from the created device context. SelectObject(~) - Selects the specified object into a device context. GetScrollPosition() - Returns (int) the scroll position of the view. CreatePen() - Creates an object of type CPen which is used to draw on the screen through a device context. AfxMessageBox(~) - Simply displays the specified text in a message box. DisplayLines() - Displays the documents lines. DisplayAntennas() - Displays the documents antenna. DisplayMeasurements() - Displays the documents measurements. ReDraw() - Invalidates and redraws to display to show any changes resulting from the entered command. SetCapture() - Captures the mouse such that no other applications - can use it. Variables and Objects: BOOL m_bTracerLinesValid - FALSE: Current tracer lines are invalid. BOOL m_bBoundaryTL - TRUE if no boundary. CDC* dc - A dummy device context pointer. CDC* pDC - A pointer to a device context with the required backing DIB pre-selected. CDIBSectionLite m_dibsection - Created DIB section on which all application drawing takes place. CPen penBlack - New CPen drawing object. CPen* pOldPen - Old CPen drawing object. CPoint m_cpBoundaryBR - Bottom right corner of specified boundary. CPoint m_cpFrom - The starting point of a line. Bugs: To do:
void CHIPView::OnLButtonUp (UINT nFlags, CPoint point) [protected]

Description: This method is called each time the left mouse button is released. It initially checks that a material has been selected and that the shift key is not depressed and returns if it is i.e. if a boundary is being specified. It then releases the mouse such that it can be
59
used by another application and following this it retrieves the current mouse position which is a function of the windows current scroll position. Using the mouses current position as well as its previous position (obtained when the left mouse button was depressed) it generates the required MoveTo and LineTo commands of the specified line. Towards the end of the method, the generated commands and line objects are added to their respective object arrays. The method concludes displaying the entered line and corresponding commands. Dependencies: GetDocument() - Returns a pointer to the active document class. RGB(~) - Returns the specified COLORREF variable. GetCapture() - Returns the capture state of the mouse cursor. ReleaseCapture() - Releases the mouse cursor such that other applications can use it. GetScrollPosition() - Returns (int) the scroll position of the view. Format(~) - This function writes formatted data to a CString. AddCommand(~) - Adds a command to the command object array of the document class. AddLine(~) - Adds a line to the line object array of the document - class. DisplayLines() - Displays the lines contained in the line object array. ReDrawCommands() - Displays the commands contained in the command object array in the command history list box. ReDraw() - Invalidates and redraws the view to reflect changes. Variables and Objects: CPoint m_cpTo - The ending point of the specified line. CString m_csMoveTo - The required moveto command (starting point of line). CString m_csLineTo - The required lineto command (end point if line). CCommand* pCommand - The generated command object. CLine* pLine - The generated line object. Bugs: To do:
void CHIPView::OnMaterials () [protected]

Description: This method is called in response to the Databases->Materials menu item being selected. It displays the Materials Menu dialog and on closing the dialog with Ok button, clears and reloads the materials list box in the applications frame. Dependencies: DoModal() - Loads a modal dialog box. GetParentFrame() - Returns a pointer to the applications frame class. ClearMaterialList() - Clears the materials list box in the applications frame. LoadMaterialList() - Reloads the materials list box in the applications frame showing any changes. Variables and Objects: MaterialDlg dlg - An object of type MaterialDlg used to encapsulate the Materials Menu dialog. Bugs: To do:
60
void CHIPView::OnMouseMove (UINT nFlags, CPoint point) [protected]

Description: Called every time the mouse is moved, this function sets the variables m_X and m_Y in the document class to the current x and y mouse position taking into account the relative positions of the vertical and horizontal scroll sliders. It also updates both the vertical and horizontal ruler bars to reflect any changes governed by both the vertical and horizontal scroll sliders. All the feedback is in meters. Dependencies: GetDocument() - Gets a pointer to the current document. GetScrollPosition() - Returns the current scroll position (int). UpdateRulerInfo(~) - Updates the rulers to reflect a scroll change. Variables and Objects: int m_X - Current mouse position (x) in document class. int m_Y - Current mouse position (y) in document class. CPoint point - Current mouse position. Double m_dPixelSize - The specified pixel size. Bugs: To do:
void CHIPView::OnRButtonDblClk (UINT nFlags, CPoint point) [protected]

Description: This method is called each time the right mouse button is double clicked. Its primary function is to load the Place Extra dialog box and handle its events. In more detail, it does the following: 1. Sends the user specified simulation cut-off level as well as simulated signal magnitude at the selected point to the dialog before it is displayed. 2. On closing the dialog with its ok bottom it checks the state of the m_bMeasurement and m_bAntenna boolean variables. 3. If m_bMeasurement is true or a measurement was specified, the method gets the current mouse cursor and view scroll positions. Having done this, it then ensures that the current tracer lines are valid and that the specified measurement result position lies on a tracer line and that the respective point result has a single material intersection history. It then adds the required measurement result to the measurement result object array in the document class and also the required command in a similar way. After which it reloads the command history list box. 4. If m_bAntenna is true of an antenna was specified, an antenna will be added if no current antenna exists and the required commands added to the command history list box. Note: This method will only allow one antenna per simulation and on the addition of either a measurement or antenna the objects in the corresponding object array in the document class will be redrawn. The addition of either a measurement or antenna will also invalidate a previously specified simulation boundary. Dependencies: GetDocument() - Returns a pointer to the active document class. DoModal() - Displays a modal dialog box. AddMeasurementResult(~)- Adds a measurement result to the measurement result object array in the document class. AddCommand(~) - Adds a command to the command object array in the document class. ReDrawCommands() - Displays the commands contained in the command object array in the command history list box.
61
DisplayMeasurements() - Displays the measurement result object in the measurement result object array. AddAntenna(~) - Adds an antenna to the antenna object array in the document class. AfxMessageBox(~) - Simply displays the specified text in a message box. DisplayAntennas() - Displays the antenna objects in the antenna object array in the document class. ReDraw() - Invalidates and redraws the view to reflect changes. GetScrollPosition() - Returns (int) the scroll position of the view. Replace(~) - Replaces the specified character in a CString with another. NULL simply removes the specified character. Returns the number of characters replaced. Variables and Objects: PlaceExtraDlg dlg - An object of type PlaceExtraDlg which is displayed. CPoint Position - The current mouse cursor position. CMeasurementResult* pMR - The users specified measurement result. CCommand* pCommand - The required command object. CAntenna* pAntenna - The required antenna object. BOOL m_bBoundaryTL - TRUE if no boundary. Bugs: To do:
void CHIPView::OnRButtonDown (UINT nFlags, CPoint point) [protected]

Description: This method is called each time the right button is depressed and simply writes the point results magnitude (at the current mouse cursors position) to the screen. In more detail, it does the following: 1. Checks that a point result exists at the mouse cursors current point and that the shift key is depressed. 2. Gets the required point results magnitude and formats the required string. 3. Gets a pointer to a DC with the backing DIB pre-selected. 4. Does the displaying of the magnitude and redraws the view. Dependencies: GetScrollPosition() - Returns (int) the scroll position of the view. Format(~) - This function writes formatted data to a CString. GetDocument() - Returns a pointer to the active document. ReDraw() - Invalidates and redraws the view to reflect changes. SetTextColor(~) - Sets the DC's text colour. SetTextAlign(~) - Sets the DC's text alignment. TextOut(~) - Writes the specified text to the DC. GetKeyState(~) - Returns the state of the specified system key. If less than zero, indicates that key is depressed. Variables and Objects: CString m_csMagnitude - The CString representation of the required point results magnitude. CDC* dc - A dummy device context. CDC* pDC - The needed device context with backing DIB pre-selected. Bugs: To do:
62
void CHIPView::OnShowValidLines () [protected]

Description: This method is run before any measurement results can be entered. The reasons behind this are to prevent the user from entering a measurement result over an averaged point result. The basic operation of this method is as follows: 1. Gets the simulation constants. 2. Sets up the specified simulation environment. 3. Ensures that both angle of arrival as well as Fresnel zones will be used to establish the tracer lines. This will ensure maximum accuracy. It also sets the variable m_bMeasurement to TRUE thus ensuring that any predefined material coefficients will not be used (i.e. we want to calculate new ones). 4. It then moves the specified number of agents around the raster and on completion, displays their produced point results. 5. The final line of the method validates the generated tracer lines, allowing for measurement results to be entered. Note: Any changes to the drawing will invalidate the tracer lines. This method also checks to see whether the specified antenna lies within the specified simulation boundary. Dependencies: GetSimulationConstants() - Gets the requires simulation constants. SetupSimulationEnviroment() - Sets up the simulation environment. AntMeasPosOk() - Returns TRUE if the specified antenna is within the specified simulation boundary. DisplayPointResults() - Displays all the point results in the point result object array. GetDocument() - Returns a pointer to the active document class. DisplayLines() - From the document class and loops through all the lines in the line object array, redrawing them. DisplayAntennas() - From the document class and loops through all the antennas in the antenna object array, redrawing them. DisplayMeasurements() - From the document class and loops through all the measurement results in the measurement results object array, redrawing them. ReDraw() - Invalidates and redraws the view to reflect changes. MoveAgents() - Moves the specified number of agents around the raster. Their movement is governed by the specified simulation mechanism and the specified simulation constants. Variables and Objects: BOOL m_bAngleOfArrival - TRUE: Use angle of arrival. BOOL m_bFresnelZone - TRUE: Use Fresnel Zones. BOOL m_bMeasurement - TRUE: Use measurements. BOOL m_bDataBase - TRUE: Use coefficients in database. Bugs: To do:
void CHIPView::OnSimulateSaveresults () [protected]

Description: Saves the current screen contents to a user specified DIB such that the simulation results can be distributed and viewed on any picture editor. NOTE: The CFileDialog is initialized such that only *.bmp file extensions can be used and the overwriting of files will be prompted. Dependencies: DoModal() - Invokes the modal CFileDialog box.
63
Save(~) - Belongs to the CDIBSectionLite class and saves the contents of the applications backing DIB to a specified bitmap file. Variables and Objects: CString m_csPath - The path of the specified bitmap file. Bugs: To do:
void CHIPView::OnUpdate (CView * pSender, LPARAM lHint, CObject * pHint) [protected, virtual]
Description: Called by the framework after the views document has been modified. It has been altered to change both the vertical and horizontal scroll bars to reflect any zoom functions as well as update the required ruler information. Dependencies: SetScrollSizes(~) - Adjusts the scrolling characteristics of two scroll bars when the zoom factor is changed. UpdateRulersInfo(~) - Adjusts the applications rulers when either the zoom factor has changed or the drawing size is altered. Invalidate(~) - Marks the view for a redraw on the next UpdateWindow call. UpdateWindow() - Redraws the view if required. ReDrawCommands() - Redisplays the command history in the command history list box in the applications frame. GetScrollPosition() - Returns (int) the current scroll position of both the vertical and horizontal scroll bars. Variables and Objects: float m_ZoomFactor - The current view zoom factor. CSize m_BackGroundSize - The size of the backing DIB (5000x5000 pixels). Bugs: To do:
void CHIPView::OnVScroll (UINT nSBCode, UINT nPos, CScrollBar * pScrollBar) [protected]

Description: Called every time the vertical scroll bar is moved. This method passes the current scroll position to the UpdateRulerInfo method to exact changes in the vertical ruler display. It also invalidates the window forcing a redraw which reflects the scrolled to portion of the display. Dependencies: Invalidate(~) - Marks the display as "dirty" forcing a redraw. GetScrollPosition(~) - Returns the current scroll position (int). UpdateRulerInfo(~) - Updates the rulers to reflect a scroll change. Variables and Objects: UINT nSBCode - Specifies a scroll-bar code that indicates a scrolling request by the user i.e. SB_BOTTOM = Scroll to bottom. UINT nPos - Contains the current scroll-box position. N.B. dependant on which nSBCode is specified. CScrollBar* pScrollBar - Pointer to object of type CScrollBar which encapsulates the applications vertical scroll bar. Bugs: To do:
64
void CHIPView::OnZoomIn () [protected]

Description: This method is called in response to the Zoom->Zoom In menu item being selected. It initially shows the applications drawing rulers. It then multiplies the current drawing zoom factor by 2 and updates the applications rulers to reflect the required scale. Dependencies: GetParentFrame() - Returns a pointer to the applications frame class. ShowRulers(~) - TRUE: show rulers OnUpdate(~) - Called by the frame after the view has been altered (this does the scaling of the rulers see method). Variables and Objects: int m_ZoomFactor - The required zoom factor used in the OnDraw method by a DC StretchBlt function. Bugs: To do:
void CHIPView::OnZoomNormal () [protected]

Description: This method is called in response to the Zoom->Zoom Normal menu item being selected. It initially shows the applications drawing rulers. It then sets the current drawing zoom factor to one and updates the applications rulers to reflect the required scale. Dependencies: GetParentFrame() - Returns a pointer to the applications frame class. ShowRulers(~) - TRUE: show rulers OnUpdate(~) - Called by the frame after the view has been altered (this does the scaling of the rulers see method). Variables and Objects: int m_ZoomFactor - The required zoom factor used in the OnDraw method by a DC StretchBlt function. Bugs: To do:
void CHIPView::OnZoomOut () [protected]

Description: This method is called in response to the Zoom->Zoom Out menu item being selected. It initially checks that the current zoom factor is greater the 0.25 (i.e. you don't want to zoom out to much). It also ensures that the applications rulers are not shown if the zoom factor is less than one. If the above checks are passed, it divides the current drawing factor by 2 and updates the applications rulers to reflect the required scale. Dependencies: GetParentFrame() - Returns a pointer to the applications frame class. ShowRulers(~) - TRUE: show rulers OnUpdate(~) - Called by the frame after the view has been altered (this does the scaling of the rulers see method). Variables and Objects: int m_ZoomFactor - The required zoom factor used in the OnDraw method by a DC StretchBlt function. Bugs: To do:
65
void CHIPView::ReDraw ()
Description: Invalidates the view and forces a redraw. Dependencies: Invalidate(~) - Marks the view as "dirty" such that on the next update it will be redrawn. UpdateWindow() - Initiates an update. Variables and Objects: Bugs: To do:
void CHIPView::ReDrawCommands ()
Description: Called whenever the elements of the command history are altered i.e. whenever elements are added or removed. Dependencies: GetParentFrame() - Gets a pointer to a frame window if successful, otherwise NULL. ClearHistoryList() - Removes all the elements in the command history list. LoadHistoryList() Reloads the command history list with any alterations. Variables and Objects: Bugs: To do:
void CHIPView::ResolveMeasurementResult (CMeasurementResult * pMResult)

Description: This method takes a user specified measurement result and using the intersection history of the corresponding point result, determines the transmission coefficient of any intersected object. In more detail it does the following: 1. Before we begin it is important to realize that the corresponding point results intersection history will be in the following form (num)Name where num is the angle of arrival in radians and Name is the name of the intersected material. 2. With this in mind, the first stage of this method extracts the angle of arrival and gets a double representation of it. 3. Having done this, the material name is then extracted. 4. Having the above, the method then calculates the transmission coefficient of the intersected material by getting the difference between the simulated value at the specified point and the corresponding measured value. This difference is attributed to the material intersection and by using the cosine of the angle of intersection, the materials transmission coefficient is calculated. 5. The required variables of the passed in measurement result are then changed to contain the above calculated values. Dependencies: Replace(~) - Replaces the specified character in a CString with another. NULL simply removes the specified character. atof(~) - Converts a CString to a double. ExtractLetters(~) - Returns the letters contained within a CString. SpanExcluding(~) - Extracts all the characters of a CString that are within the specified set. cos(~) - Returns the cosine of the specified angle in radians.
66
Variables and Objects: CString m_csAngle - CString of extracted angle. double m_dAngle - Double of extracted angle. CMeasurementResult* pMResult - The passed in measurement result object. CString m_csMaterialName - CString of extracted material name. double m_dDifference - The difference between the measured value at a point and the simulated value. double m_dCoefficient - The calculated material transmission coefficient. Bugs: To do:
void CHIPView::SetUpFeedBackDisplay (CString heading, int range)

Description: This method sets up the feedback display which belongs to the Frame class. The feedback display consists of a read only edit box (m_eSimSection) and a process control bar (m_pcSimProcess). The setup process is as follows: 1. Select the first line of the edit box. 2. Change its text to the required text (heading). 3. Set the range of the process control to the required range (range). 4. Set the current process control position to 0 or the start position. Dependencies: SetSel(~) - Belongs to the CEdit class and sets the current edit focus on the specified line. N.B. -1 specifies entire line. ReplaceSel(~) - Belongs to the CEdit class and changes the selected text to the specified text. SetRange(~) - Belongs to the CProcessControl class the sets the range of the associated process control object. SetPos(~) - Belongs to the CProcessControl class and sets the current process control position to that specified. GetParentFrame() - Returns a pointer to the applications Frame class. Variables and Objects: CEdit m_eSimSection - CEdit control associated with the CEdit box in the applications frame. CProcessControl m_pcSimProcess - CProcessControl control associated with the CProcessControl bar in the applications frame. CString heading - Feedback heading (method argument). int range - Process control range (method argument). Bugs: To do:
void CHIPView::SetupSimulationEnviroment ()
Description: This method sets up the simulation environment by doing the following: 1. Clears the backing DIB and redisplays the specified lines and antennas. 2. Nullifies the point result pointer array as new point results are to be generated. 3. Removes any existing point results for the same reason. Dependencies: Load(~) - Loads a DIB into the utilized DIB section. AfxMessageBox(~) - Displays a simple message box with specified text.
67
GetDocument() - Returns a pointer to the active document class. DisplayLines() - From the document class and loops through all the lines in the line object array, redrawing them. DisplayAntennas() - From the document class and loops through all the antennas in the antenna object array, redrawing them. NullifyPointerArray() - Sets all the pointers in the point result pointer array to NULL. ClearPointResults() - From the document class and deletes all the point results in the point result object array. Variables and Objects: Bugs: To do:
void CHIPView::TidyUpFeedBackDisplay ()
Description: This method tidies up the feedback display by removing any specified text and by resetting the process control position back to zero. Dependencies: SetSel(~) - Belongs to the CEdit class and sets the current edit focus on the specified line. N.B. -1 specifies entire line. ReplaceSel(~) - Belongs to the CEdit class and changes the selected text to the specified text. SetPos(~) - Belongs to the CProcessControl class and sets the current process control position to that specified. GetParentFrame() - Returns a pointer to the applications Frame class. Variables and Objects: CEdit m_eSimSection - CEdit control associated with the CEdit box in the applications frame. CProcessControl m_pcSimProcess - CProcessControl control associated with the CProcessControl bar in the applications frame. Bugs: To do:
void CHIPView::UpdateRulersInfo (int nMessage, CPoint ScrollPos, CPoint Pos = CPoint(0, 0))
Description: This method is used to set the attributes of a stRULER_INFO object which when passed back to the MainFrame class will govern the appearance and functioning of both the vertical and horizontal rulers. Dependencies: GetDocument() - Returns a pointer to the applications document class. GetParentFrame() - Returns a pointer to the applications frame class. UpdateRulersInfo(~) - Adjusts the applications rulers when either the zoom factor has changed or the drawing size is altered. Variables and Objects: stRULER_INFO pRulerInfo - Object containing the applications rulers information. Double m_dPixelSize - The specified pixel size. Bugs: To do:
68
2.8 Cline CLine::CLine (CPoint ptFrom, CPoint ptTo, COLORREF Colour)

Description: Being a class constructor, this method is called each time an object of this class is created. This method simply initialized the required object variables. Dependencies: Variables and Objects: CPoint m_ptFrom - The starting point of the line. CPoint m_ptTo - The end point of the line. COLORREF m_Colour - The colour (or material) of the line. Bugs: To do:
void CLine::Draw (CDC * pDC)

Description: This method is responsible for drawing a CLine object. It does this by initially creating and selecting a pen into the active device context of the required colour. It then draws the line by moving to the lines starting point and then by drawing a line to the lines end point. The method ends by reselecting the old drawing pen into the active device context. Dependencies: SelectObject(~) - Selects the specified object into a device context. MoveTo(~) - Moves to the specified point in a device context. LineTo(~) - Draws a line to the specified point in a device context using the selected drawing object. Variables and Objects: CPen lpen - The newly created drawing object. CPen* pOldPen - The old drawing object. CDC* pDC - A pointer to the active device context. Bugs: To do:
void CLine::Serialize (CArchive & ar)

Description: Called when loading and saving a document, this function saves and loads an object specified variables from an archive (the specified file). Dependencies: IsStoring() - TRUE: Archive is storing data. Variables and Objects: CPoint m_ptFrom - The lines starting point. CPoint m_ptTo - The lines ending point. int m_MaxY - The lines maximum y-coordinate. int m_MaxX - The lines maximum x-coordinate. int m_MinY - The lines minimum y-coordinate. int m_MinX - The lines minimum x-coordinate. COLORREF m_Colour - The lines colour (or material). Bugs: To do:
69
void CLine::SetupLineVariables ()
Description: This method determines the maximum and minimum x and y coordinates of a lines starting and ending points. Dependencies: Variables and Objects: CPoint m_ptFrom - The lines starting point. CPoint m_ptTo - The lines ending point. int m_MaxY - The lines maximum y-coordinate. int m_MaxX - The lines maximum x-coordinate. int m_MinY - The lines minimum y-coordinate. int m_MinX - The lines minimum x-coordinate. Bugs: To do:
2.9 CMainFrame void CMainFrame::ClearHistoryList ()

Description: When called this method deletes the entire contents of the command history list box. Dependencies: DeleteAllItems() - Deletes the contents of a CListCtrl. Variables and Objects: CListCtrl m_CommandHistory - The command history list box control object. Bugs: To do:
void CMainFrame::ClearInputField ()
Description: Clears the input (for drawing commands) edit box by selecting the previously entered text and then deleting it. Dependencies: SetSel(~) - Selects the specified line of a CEdit control (-1 for entire line). Clear() - Deletes the current line of a CEdit control. Variables and Objects: CEdit m_Input - The input edit box control object. Bugs: To do:
void CMainFrame::ClearMaterialList ()
Description: When called this method deletes the entire contents of the materials list box. Dependencies: DeleteAllItems() - Deletes the contents of a CListCtrl. Variables and Objects: CListCtrl m_Materials - The materials list box control object. Bugs: To do:
70
BOOL CMainFrame::CreateAntennasListBox ()
Description: This methods adds a list control to the applications tool bar. It does this by initially configuring the list controls place holder (position and size) and then creating the required list control. The method concludes by inserting the required columns into the created list control and loading the list control with the required list items (antennas database). Dependencies: SetButtonInfo(~) - This function sets the separator's width in pixels. GetItemRect(~) - Gets the co-ordinates of a separator you want to replace with a control. Create(~) - Creates the required child list control. InsertColumn(~) - Inserts the specified column into a list control. Variables and Objects: int nWidth - The specified width of the list control. int nHeight - The specified heigth of the list control. CToolBar m_wndFeedBackBar - The tool bar object into which the list control is to be inserted. CRect rect - A rect object holding the final co-ordinates and size of the required list control. CEdit m_eSimProcess - The actual created list control object. Bugs: To do:
void CMainFrame::CreateEditInput ()
Description: This method adds an edit box to the applications Input tool bar. It does this by initially configuring the edit boxes place holder (position and size) and then creating the required edit box. NOTE: This edit box takes as its input, user drawing and placement commands. It's purpose is to implement a CAD like drawing interface for end users that are more comfortable with text based drawing. Dependencies: SetButtonInfo(~) - This function sets the separator's width in pixels. GetItemRect(~) - Gets the coordinates of a separator you want to replace with a control. Create(~) - Creates the required child edit box. Variables and Objects: int nWidth - The specified width of the edit box. int nHeight - The specified heigth of the edit box. CToolBar m_wndInputBar - The tool bar object into which the edit box is to be inserted. CRect rect - A rect object holding the final co-ordinates and size of the required edit box. CEdit m_Input - The actual created edit box object. Bugs: To do:
BOOL CMainFrame::CreateHistoryListBox ()
Description: This methods adds a list control to the applications tool bar. It does this by initially configuring the list controls place holder (position and size) and then creating the required list control. The method concludes by inserting the required columns into the created list control. Dependencies:
71
SetButtonInfo(~) - This function sets the separator's width in pixels. GetItemRect(~) - Gets the co-ordinates of a separator you want to replace with a control. Create(~) - Creates the required child list control. InsertColumn(~) - Inserts the specified column into a list control. Variables and Objects: int nWidth - The specified width of the list control. int nHeight - The specified heigth of the list control. CToolBar m_wndFeedBackBar - The tool bar object into which the list control is to be inserted. CRect rect - A rect object holding the final co-ordinates and size of the required list control. CEdit m_eSimProcess - The actual created list control object. Bugs: To do:
BOOL CMainFrame::CreateMaterialsListBox ()
Description: This methods adds a list control to the applications tool bar. It does this by initially configuring the list controls place holder (position and size) and then creating the required list control. The method concludes by inserting the required columns into the created list control and loading the list control with the required list items (materials database). Dependencies: SetButtonInfo(~) - This function sets the separator's width in pixels. GetItemRect(~) - Gets the co-ordinates of a separator you want to replace with a control. Create(~) - Creates the required child list control. InsertColumn(~) - Inserts the specified column into a list control. Variables and Objects: int nWidth - The specified width of the list control. int nHeight - The specified heigth of the list control. CToolBar m_wndFeedBackBar - The tool bar object into which the list control is to be inserted. CRect rect - A rect object holding the final co-ordinates and size of the required list control. CEdit m_eSimProcess - The actual created list control object. Bugs: To do:
void CMainFrame::CreateProcessPCFeedBack ()
Description: This method adds a process control to the applications tool bar. It does this by initially configuring the process controls place holder (position and size) and then creating the required process control. Dependencies: SetButtonInfo(~) - This function sets the separator's width in pixels. GetItemRect(~) - Gets the coordinates of a separator you want to replace with a control. Create(~) - Creates a smooth child progress control. Variables and Objects: int nWidth - The specified width of the process control.
72
int nHeight - The specified heigth of the process control. CToolBar m_wndToolBar - The tool bar object into which the process control is to be inserted. CRect rect - A rect object holding the final co-ordinates and size of the required process control. CProcessCtrl m_pcSimProcess - The actual created process control object. Bugs: To do:
void CMainFrame::CreateProcessTextFeedBack ()
Description: This method adds an edit box to the applications tool bar. It does this by initially configuring the edit boxes place holder (position and size) and then creating the required edit box. NOTE: The edit box is set to read-only as it is only required to report the different stages of simulation. Dependencies: SetButtonInfo(~) - This function sets the separator's width in pixels. GetItemRect(~) - Gets the coordinates of a separator you want to replace with a control. Create(~) - Creates the required child edit box. SetReadOnly(~) - TRUE: Associated edit box will be read-only. Variables and Objects: int nWidth - The specified width of the edit box. int nHeight - The specified heigth of the edit box. CToolBar m_wndToolBar - The tool bar object into which the edit box is to be inserted. CRect rect - A rect object holding the final co-ordinates and size of the required edit box. CEdit m_eSimProcess - The actual created edit box object. Bugs: To do:
void CMainFrame::LoadAntennasList ()
Description: When called, this method loads the antenna list box with the antennas found in the antenna record set. In more detail, it does the following: 1. Opens the antenna record set. 2. Gets the number of antennas in the antenna record set and returns if zero. 3. Loops through the antenna records in the antenna record set (last to first) and adds them (name and ID) to the antennas list box. Dependencies: Format(~) - This function writes formatted data to a CString. InsertItem(~) - Inserts an item into the list view control. SetItem(~) - Sets some or all of a list view items attributes. Open() - Opens the specified database to allow access to its records. Close() - Closes the database. GetRecordCount() - Returns the number of records in a record set. MovePrev() - Moves to the previous record in the database. MoveLast() - Moves to the last record in the database. IsBOF() - TRUE until the beginning of the record set is reached. Variables and Objects:
73
int SetRecordNo - The number of antennas in the antenna record set. LVITEM lvi - A list box item variable. CString strItem - CString of antennas ID. CAntennaSet set - The antennas record set. Bugs: To do:
void CMainFrame::LoadHistoryList ()
Description: When called, this method loads the command history list box with the commands found in the command object array. In more detail, it does the following: 1. Gets a pointer to the active document. 2. Gets the number of commands in the command object array and returns if zero. 3. Loops through the commands in the command object array (last to first) and adds them (name and position) to the command history list box. 4. Lastly it ensures that the last command entered is always visible. Dependencies: GetActiveDocument() - Returns a pointer to the active document. GetCommandCount() - Returns (int) the number of commands in the command object array. GetCommand(~) - Returns the specified command object from the command object array. Format(~) - This function writes formatted data to a CString. InsertItem(~) - Inserts an item into the list view control. SetItem(~) - Sets some or all of a list view items attributes. GetItemCount() - Returns the number of items in a list control. EnsureVisible(~) - Ensures that the specified item in a list control is visible. Variables and Objects: CHIPDoc* pDoc - A pointer to the active document class. int CommandNumber - The number of commands in the command object array. CCommand* pCommand - A pointer to a CCommand object used to retrieve command information. LVITEM lvi - A list box item variable. CString strItem - CString of commands position in command object array. int nCount - The number of inserted list box items Bugs: To do:
void CMainFrame::LoadMaterialsList ()
Description: When called, this method loads the materials list box with the materials found in the materials record set. In more detail, it does the following: 1. Opens the materials record set. 2. Gets the number of materials in the materials record set and returns if zero. 3. Loops through the material records in the material record set (last to first) and adds them (name and ID) to the materials list box. Dependencies: Format(~) - This function writes formatted data to a CString. InsertItem(~) - Inserts an item into the list view control.
74
SetItem(~) - Sets some or all of a list view items attributes. Open() - Opens the specified database to allow access to its records. Close() - Closes the database. GetRecordCount() - Returns the number of records in a record set. MovePrev() - Moves to the previous record in the database. MoveLast() - Moves to the last record in the database. IsBOF() - TRUE until the beginning of the record set is reached. Variables and Objects: int SetRecordNo - The number of materials in the material record set. LVITEM lvi - A list box item variable. CString strItem - CString of materials ID. CMaterialSet set - The materials record set. Bugs: To do:
BOOL CMainFrame::OnCreateClient (LPCREATESTRUCT lpcs, CCreateContext * pContext) [protected, virtual]

Description: This method is called each time the client area of the applications frame is created. It was changed to display both the required vertical and horizontal rulers. Dependencies: CreateRulers(~) - Creates the required rulers. Variables and Objects: Bugs: To do:
void CMainFrame::OnDisplayBar () [protected]

Description: This method toggles the visibility of the Display tool bar by doing the following: 1. Gets the state of the Display tool bar. 2. If it is visible, it hides it and vise versa. 3. Reshuffles the frame layout. Dependencies: GetStyle() - This method gets the styles currently applied to a toolbar control. ShowControlBar(~) - Method used to both show and hide a tool bar. RecalcLayout() - Usually called by frame work the reshuffle the fame window layout to maximize the available space when showing/hiding controls or resizing the entire frame. Variables and Objects: BOOL bVisiable - TRUE: Tool bar visible. Bugs: To do:
void CMainFrame::OnEnSelect ()
Description: This method is called each time the contents of the input edit field in the applications frame is changed. It does the following: 1. Gets a pointer to the active view. 2. Retrieves the CString from the input edit field.
75
Checks that the terminating character ']' has not yet been entered. Makes all string characters lowercase. Removes any white space before string. If the terminating character has been entered, the formatted input CString is passed to the NewCommand method in the view class and the edit input field is cleared before the view is redrawn. Dependencies: GetActiveView() - Returns a pointer to the active view. LineLength(~) - Retrieves a character index for a given line number. GetBuffer(~) - This method retrieves a pointer to the internal character buffer for the CString object. GetLine(~) - Retrieves the specified line of text from an edit control. Find(~) - Returns the index of the specified character in a CString. MakeLower() - Makes all the characters in the associated CString lower case. TrimLeft() - Removes any white space from the beginning of the associated CString. NewCommand(~) - Handles the input of a new command. ClearInputField() - Clears the input edit field. ReDraw() - Invalidates and redraws to display to show any changes resulting from the entered command. Variables and Objects: CString m_strInput - The current user inputted CString. CHIPView* pView - A Pointer to the active view. CEdit m_Input - The input edit control object. int iExit - -1: Terminating character not found in CString. Bugs: To do:
3. 4. 5. 6.
void CMainFrame::OnInputBar () [protected]

Description: This method toggles the visibility of the Input tool bar by doing the following: 1. Gets the state of the Input tool bar. 2. If it is visible, it hides it and vise versa. 3. Reshuffles the frame layout. Dependencies: GetStyle() - This method gets the styles currently applied to a toolbar control. ShowControlBar(~) - Method used to both show and hide a tool bar. RecalcLayout() - Usually called by frame work the reshuffle the fame window layout to maximize the available space when showing/hiding controls or resizing the entire frame. Variables and Objects: BOOL bVisiable - TRUE: Tool bar visible. Bugs: To do:
void CMainFrame::OnMaterialSelect ()
Description: This method is called each time the selection in the materials list box (in the applications frame) is changed. It initially gets a pointer to the active document class and a double representation of the selected list item (POSITION->CString-> Double). It then
76
opens the materials record set and returns if it is empty. If it is not empty, it loops through the record set until the position of the selected material is reached. It then updates the current selected material indicator in the application status bar and sets the current drawing colour in the document class to that of the selected material. Dependencies: GetActiveDocument() - Returns a pointer to the active document. GetFirstSelectedItemPosition() - Returns (POSITION) the position of the selected list box item. Format(~) - This function writes formatted data to a CString. In this method it is used to get a CString representation of a POSITION variable. atof(~) - Does a CString To double conversion. Open() - Opens the specified database to allow access to its records. Close() - Closes the database. GetRecordCount() - Returns the number of records in a record set. MoveNext() - Moves to the next record in the database. MoveFirst() - Moves to the first record in the database. RGB(~) - Returns the specified COLORREF variable. Variables and Objects: CHIPDoc* pDoc - Pointer to active document class. POSITION ListPosition - Value used to denote position of record in list view control. CString positionCS - CString representation of ListPosition. double positionD - Double representation of ListPosition. CMaterialSet set - Object of type MaterialSet (container class for database records). Bugs: To do:
void CMainFrame::OnRemoveCommand ()
Description: This method is responsible for removing a command from the command history list box and its associated action. It does this as follows: 1. Gets a pointer to the active document class. 2. Gets a pointer to the active view class. 3. Gets the number of commands in the command history list box and returns if zero. 4. Gets the position of the selected command and obtains a CString and then double representation of it. 5. It then prompts if the user is sure they want to delete the command and if they press Ok, the deletion process begins. 6. The next two steps involve looping through the commands in the command object array and determining how many "antenna" and "lineto" commands have been specified up to and including the specified command awaiting deletion. This is used to determine the index of the line/antenna object to delete if the specified command resulted in a line/antenna being drawn. 7. Determines if the selected command resulted in a line/ antenna being drawn and deletes the required line/ antenna from the respective object array. 8. Deletes the selected command from the command object array and clears/reloads the command history list box to show change. 9. The final stage involves clearing the backing DIB and redrawing the required information on the DIB.
77
Dependencies: GetActiveDocument() - Returns a pointer to the active document. GetActiveView() - Returns a pointer to the active view. GetCommandCount() - Returns (int) the number of commands in the command object array. GetFirstSelectedItemPosition() - Returns (POSITION) the position if the selected list box item. Format(~) - This function writes formatted data to a CString. In this method it is used to get a CString representation of a POSITION variable. atof(~) - Does a CString To double conversion. MessageBox(~) - Displays the specified text in a simple message box. GetCommand(~) - Returns the specified command object from the command object array. Find(~) - Returns the index of the specified character in a CString. Left(~) - Returns the CString to the left of the specified index in another CString. DeleteLine(~) - Deletes the specified line object from the line object array. DeleteCommand(~) - Deletes the specified command object from the command object array. DeleteAntenna(~) - Deletes the specified antenna object from the antenna object array. ClearHistoryList() - Deletes the contents of the command history list box. LoadHistoryList() - Reloads the command history list box with the command objects found in the command object array. ClearBackGround() - Selects a blank DIB into the active device context. DisplayLines() - Draws the lines in the line object array on the backing DIB. DisplayAntennas() - Draws the antennas in the antenna object array on the backing DIB. NullifyPointerArray() - Sets all the pointers in the point result pointer array to NULL. ReDraw() - Invalidates and redraws to display to show any changes resulting from the deletion. Variables and Objects: CHIPDoc* pDoc - Pointer to active document class. CHIPView* pView - Pointer to active view class. int ComCount - The number of commands in the command object array. POSITION ListPosition - Value used to denote position of record in list view control. CString positionCS - CString representation of ListPosition. double positionD - Double representation of ListPosition. int LinePos - The number of line related commands before and including the selected command. int AntPos - The number of antenna related commands before and including the selected command. Bugs: To do:
void CMainFrame::OnUpdateDisplayBar (CCmdUI * pCmdUI) [protected]

Description: This method is primarily responsible for checking the Display bar menu item when it is visible and un-checking the Display bar menu item when it is not. Dependencies: GetStyle() - This method gets the styles currently applied to a toolbar control.
78
SetCheck(~) - Used to set a user-interface item to the required check state. Variables and Objects: Bugs: To do:
void CMainFrame::OnUpdateInputBar (CCmdUI * pCmdUI) [protected]

Description: This method is primarily responsible for checking the Input bar menu item when it is visible and un-checking the Input bar menu item when it is not. Dependencies: GetStyle() - This method gets the styles currently applied to a toolbar control. SetCheck(~) - Used to set a user-interface item to the required check state. Variables and Objects: Bugs: To do:
void CMainFrame::ShowRulers (BOOL bShow)

Description: This method shows and hides the applications rulers. Dependencies: ShowRulers(~) - When TRUE is passed in, the applications rulers are shown. Variables and Objects: Bugs: To do:
void CMainFrame::UpdateRulersInfo (stRULER_INFO stRulerInfo)

Description: This method updates the current ruler information. Dependencies: UpdateRulersInfo(~) - Updates the applications rules with the information contained in the stRULER_INFO object. Variables and Objects: Bugs: To do:
2.10 CMeasurementResult CMeasurementResult::CMeasurementResult (CPoint Position, double Magnitude, int ID)

Description: Called as a constructor when an object of this type is created. This method simply initializes the objects required variables. Dependencies: Format(~) - This function writes formatted data to a CString. In this method it is used to get CString representations of both integer and double variables. Variables and Objects: CPoint m_cpPosition - The position of the specified measurement result in the raster. Double m_dMagnitude - The magnitude of the specified measurement result. CString m_csID - The CString of the measurement results ID. CString m_csMagnitude - The CString of the measurement results specified magnitude. BOOL m_bResolved - TRUE: Measurement result has been resolved.
79
Bugs: To do:
void CMeasurementResult::Draw (CDC * pDC)

Description: This method displays a measurement result on the screen. The measurement result is displayed (at its specified position) as text which contains both its ID and its specified magnitude. This method only displays unresolved measurement results. Dependencies: SetTextColor(~) - Sets the required colour of the text to be output. SetTextAlign(~) - Sets the alignment of the text to be output. TextOut(~) - Outputs (draws on device context) the specified text at the specified point in the raster. Variables and Objects: CDC* pDC - A pointer to the active device context. BOOL m_bResolved - TRUE: Measurement result has been resolved. Bugs: To do:
void CMeasurementResult::Serialize (CArchive & ar)

Description: Called when loading and saving a document, this function saves and loads an object specified variables from an archive (the specified file). Dependencies: IsStoring() - TRUE: Archive is storing data. Variables and Objects: CPoint m_cpPosition - The position of the specified measurement result in the raster. Double m_dMagnitude - The magnitude of the specified measurement result. CString m_csID - The CString of the measurement results ID. CString m_csMagnitude - The CString of the measurement results specified magnitude. BOOL m_bResolved - TRUE: Measurement result has been resolved. Bugs: To do:
2.11 CPointResult CPointResult::CPointResult (CPoint Position, double m_dmagnitude, BOOL m_borigional)

Description: Called as a constructor when an object of this type is created. This method simply initializes the objects required variables. Dependencies: Variables and Objects: CPoint m_Position - The position of the point result in the raster. Double m_dMagnitude - The magnitude of the point result. BOOL m_bOrigional - TRUE: The point result was created by simulation and not averaging. Bugs: To do:
80
void CPointResult::Draw (CDC * pDC)

Description: When this function is called, it changes the on-screen pixel colour at the point results position to a colour related to its magnitude. Dependencies: SetPixel(~) - Sets the specified pixel of a device context to the specified colour. GetPointResultColour() - Returns a colour (COLORREF) that corresponds to the point results magnitude. Variables and Objects: CDC* pDC - A pointer to the active device context. Bugs: To do:
COLORREF CPointResult::GetPointResultColor ()
Description: This method returns the colour representation of a point results magnitude. It does this by using the point results magnitude in a linear COLORREF scale which goes from pink to red to blue to white. Dependencies: RGB(~) - Specifies a COLORREF variable. Variables and Objects: Double m_dMagnitude - A positive representation of the point results magnitude. Bugs: To do:
2.12 DrawingSettingsDLG void DrawingSettingsDLG::OnOK () [protected, virtual]

Description: Called when the user closes the Drawing Setting dialog using the Ok button. All that this method does is update its member variables with the user inputted ones. Dependencies: UpdateData(~) - By passing in TRUE, the entered variables are copied to their respective member variables. OnOK() - Simply closes the dialog. Variables and Objects: Bugs: To do:
2.13 MaterialDlg CString MaterialDlg::DoubleToCString (double in)

Description: This method takes as its argument a double value and returns a CString representation. Dependencies: Format(~) - This method writes formatted string data into a string. In this method, it is used to convert a double variable to a CString. Variables and Objects: CString strFinalVal - CString variable to hold conversion. Double in - The methods input argument.
81
Bugs: To do:
void MaterialDlg::LoadList ()
Description: This method loads the materials list box with the required information from the materials database. It does this by initially opening the materials database and checking that it contains data. After it has done this, it moves to the last record in the materials record set and steps through them (starts at the end as list box is loaded bottom up) populating the required column of the list box with the required information from the materials database. The following information is extracted and displayed: 1. Material ID. 2. Material Name. 3. Material Reflection Coefficient. 4. Material Transmission Coefficient. 5. Material Frequency. Dependencies: Format(~) - This method writes formatted string data into a string. Open() - Opens the specified database to allow access to its records. MoveLast() - Moves to the last record of the database. MovePrev() - Moves to the previous record in the database. Close() - Closes the database. AfxMessageBox(~) - Simply displays the specified text in a message box. IsBOF() - TRUE until the beginning of the record set is reached. GetRecordCount() - Returns the number of records in a record set. InsertItem(~) - Inserts an item into the list view control. SetItem(~) - Sets some or all of a list view items attributes. DoubleToCString(~) - Returns the CString of the specified double. Variables and Objects: CMaterialSet set - Object of type MaterialSet (container class for database records). LVITEM lvi - List Box item object. CString strItem - Used to hold material attributes before they are inserted into the materials list box. Bugs: To do:
void MaterialDlg::OnAddMaterial () [protected]

Description: This method displays the Add Material dialog window and on closing the dialog, clears and reloads the materials list box. Dependencies: DoModal() - Displays a modal dialog box. DeleteAllItems() - This function deletes all the items from a list view control. LoadList() - This function reloads the list view control with the newly edited database records. Variables and Objects: AddMaterialDlg dialog - Add Material dialog object used to display the Add Material dialog. Bugs:
82
To do:
void MaterialDlg::OnChangeCoeff () [protected]

Description: This method is called in response to the user clicking the Change Transmission Coeff. in the Materials Menu. This method does the following: 1. Get the position of the selected item in the list. 2. Convert this position to a CString and then a double. 3. Check that an item in the list was selected. 4. Open the materials database. 5. Get the ID of the last record in the database. 6. Move to the selected record. 7. Open the Change Transmission Coefficient dialog and when dismissed with Ok, get the new Transmission coefficient, generate the required reflection coefficient, change the existing coefficients, close the database and tidy up any variables used. Dependencies: GetFirstSelectedItemPosition() - This method gets the position of the first selected item in the list view control. atof(~) - Convert specified character string to a double-precision floating-point number. Format(~) - This method writes formatted string data into a string. In this method, it is used to convert a POSITION variable to a CString. Open() - Opens the specified database to allow access to its records. MoveFirst() - Moves to the first record of the database. MoveNext() - Moves to the next record in the database. DoModal() - This method invokes a modal dialog box and returns the dialog-box result when done. Edit() - This function allows for changes to the current record. Update() - Required to complete Edit() function. Requery() - Rebuilds the record set. Close() - Closes the database. DeleteAllItems() - This function deletes all the items from the list view control. LoadList() - This function reloads the list view control with the newly edited database records. AfxMessageBox(~) - Simply displays the specified text in a - message box. Variables and Objects: POSITION ListPosition - Value used to denote position of record in list view control. CString positionCS - CString representation of ListPosition. double positionD - Double representation of ListPosition. CMaterialSet set - Object of type MaterialSet (container class for database records). ChangeTransCoeffDlg dlg - Object of type ChangeTransCoeffDlg used to interact with the Change Transmission Coefficient dialog. Bugs: To do:
83
BOOL MaterialDlg::OnInitDialog () [protected, virtual]

Description: This method is called each time the Materials Dialog is initialized. It basically sets up the required columns in the materials list box and loads the materials list box with the required data from the materials database. Dependencies: GetClientRect(~) - Returns (CRect) a rectangle which comprises the materials list box. Width() - Returns (int) the width of a CRect object. InsertColumn(~) - Inserts the specified column into a list box control. LoadList() - Loads the list control with the data from the materials database. Variables and Objects: CRect rect - Material list control rectangle. int nColInterval - The width of each added column in the material list. Bugs: To do:
void MaterialDlg::OnRemoveMaterial () [protected]

Description: Allows for the deletion of a material from the database. This is done by obtaining the user selected material ID from the list control. Having done this, the corresponding material and its data is removed from the material database. The subsequent materials in the material database will then get their ID values reduced by 1. The user is prompted if he/she is sure that they want to remove the material. This method also updates the list control to show any changes after a deletion. In detail: 1. Gets the position of the selected material 2. Gets a CString and then double representation of the selected items position. 3. CHECK 1: Was a material selected. 4. Opens the materials record set and confirms the deletion of a material. 5. Loops through the record set until it finds the selected record. When it finds the record, it deletes the record and then runs through the remainder of the records and decrements their ID values. 6. The method ends by setting the position variable to 0 or no material selected (else a double click will remove two materials), clears the material list box and then reloads it. Dependencies: GetFirstSelectedItemPosition() - This method gets the position of the first selected item in the list view control. atof(~) - Convert specified character string to a double-precision floating-point number. Format(~) - This method writes formatted string data into a string. In this method, it is used to convert a POSITION variable to a CString. Open() - Opens the specified database to allow access to its records. MoveFirst() - Moves to the first record of the database. MoveNext() - Moves to the next record in the database. DoModal() - This method invokes a modal dialog box and returns the dialog-box result when done. Edit() - This function allows for changes to the current record. Delete() - Deletes the current record of a record set. Update() - Required to complete Edit() function. Requery() - Rebuilds the record set.
84
Close() - Closes the database. DeleteAllItems() - This function deletes all the items from the list view control. LoadList() - This function reloads the list view control with the newly edited database records. AfxMessageBox(~) - Simply displays the specified text in a message box. IsEOF() - TRUE until the end of the record set is reached. Variables and Objects: POSITION ListPosition - Value used to denote position of record in list view control. CString positionCS - CString representation of ListPosition. double positionD - Double representation of ListPosition. CMaterialSet set - Object of type MaterialSet (container class for database records). Bugs: To do:
2.14 PlaceExtraDlg void PlaceExtraDlg::OnAntenna () [protected]

Description: This method is called each time the Antenna check box is checked and simple un-checks the Measurement check box. Dependencies: SetCheck(~) - Belongs to the CButton class and sets the state of the check box i.e. 0 = unchecked, 1 = checked, 2 = undetermined. Variables and Objects: Bugs: To do:
BOOL PlaceExtraDlg::OnInitDialog () [protected, virtual]

Description: This method is called each time the Place Extra dialog is initialized. It initially creates an object of type CAntennaSet to gain access to the Antennas database. It then moves to the first antenna record and then loops through all the antenna records, populating the antenna list box (m_Antenna_List) with the names of all the antennas currently in the applications Antennas Database. Dependencies: Open() - Opens the specified database to allow access to its records. MoveFirst() - Moves to the first record of the database. MoveNext() - Moves to the next record in the database. Close() - Closes the database. IsEOF() - TRUE until the end of the record set is reached. SetCurSel(~) - Sets the focus on the specified combo box item. AddString(~) - Adds the specified string to combo box. Variables and Objects: CAntennaSet set - Antenna database record set object. BOOL m_bMeasurementOK - TRUE: Valid measurement specified. BOOL m_bAntennaOK - TRUE: Valid antenna specified. Bugs: To do:
85
void PlaceExtraDlg::OnMeasurement () [protected]

Description: This method is called each time the Measurement check box is checked and simple un-checks the Antenna check box. Dependencies: SetCheck(~) - Belongs to the CButton class and sets the state of the check box i.e. 0 = unchecked, 1 = checked, 2 = undetermined. Variables and Objects: Bugs: To do:
void PlaceExtraDlg::OnOK () [protected, virtual]

Description: Called each time the Place Extra dialog is dismissed with Ok button. This method initially checks to see which check box was selected. If the Antenna check box was selected, the selected antenna name is found and the dialog closed. If the Measurement check box was checked, the specified measurement result is validated top ensure that it lies within the Dependencies: GetCheck() - Belongs to the CButton class and returns the state of button i.e. 0 = unchecked, 1 = checked, 2 = undetermined. GetCurSel() - Returns the zero-based index of the currently selected item, if any, in a single selection list box. GetLBText(~) - Return the CString at the specified index of a list box. UpdateData(~) - By passing in TRUE, the entered variables are copied to their respective member variables. OnOK() - Simply closes the dialog. AfxMessageBox(~) - Simply displays the specified text in a message box. Variables and Objects: BOOL m_bAntennaOK - TRUE: Antenna correctly selected. BOOL m_bMeasurementOK - TRUE: Measurement correctly selected and entered. Double m_dLowerMeasLimit - The simulated magnitude at the selected point. Double m_dSimCutOffLevel - The specified simulation cut-off magnitude level. Bugs: To do:
2.15 SimulationSettingDlg void SimulationSettingDlg::OnFromDatabase () [protected]

Description: Called when the user checks the from database check box. All this method does is uncheck the measurement box and set its corresponding variable to FALSE. Dependencies: SetCheck() - Belongs to the CButton class and sets the state of the check box i.e. 0 = unchecked, 1 = checked, 2 = undetermined. Variables and Objects: BOOL m_bMeasurement - TRUE: Use measurements in simulation mechanism. Bugs: To do:
86
BOOL SimulationSettingDlg::OnInitDialog () [protected, virtual]

Description: This method is called each time the Simulation Settings dialog is initialized. It initially checks the state of the two simulation mechanism governing variables and checks the respective check boxes. It also sets the focus of the frequency selection combo box to its first entry. Dependencies: SetCheck(~) - Belongs to the CButton class and sets the state of the check box i.e. 0 = unchecked, 1 = checked, 2 = undetermined. SetCurSel(~) - Sets the focus on the specified combo box item. Variables and Objects: BOOL m_bDataBase - TRUE: Use database values in simulation mechanism. BOOL m_bMeasurement - TRUE: Use measurements in simulation mechanism. Bugs: To do:
void SimulationSettingDlg::OnMeasurement () [protected]

Description: Called when the user checks the measurement check box. All this method does is uncheck the From Database box and sets its corresponding variable to FALSE; Dependencies: SetCheck() - Belongs to the CButton class and sets the state of the check box i.e. 0 = unchecked, 1 = checked, 2 = undetermined. Variables and Objects: BOOL m_bDataBase - TRUE: Use database values in simulation mechanism. Bugs: To do:
void SimulationSettingDlg::OnOK () [protected, virtual]

Description: Called when the user closes the Simulation Setting dialog using the Ok button. All that this method does is update its member variables with the user inputted ones. It also checks which check button (Calculation mechanism) is checked and sets the m_bMeasurement or m_bDataBase variables as required. This method also ensures that the entered simulation cut-off level is between 0 and 250 as well as ensuring that entered antenna loss at 1m is less than zero. Dependencies: UpdateData(~) - By passing in TRUE, the entered variables are copied to their respective member variables. GetCheck() - Belongs to the CButton class and returns the state of button i.e. 0 = unchecked, 1 = checked, 2 = undetermined. OnOK() - Simply closes the dialog. GetLBText(~) - Return the CString at the specified index of a list box. Variables and Objects: BOOL m_bMeasurement - TRUE: Measurement selected. FALSE: From Database selected. Bugs: To do:
87
3. Conclusion From Section 2 it is evident that the applications code is well documented. This will ultimately simplify any maintenance, updating or expanding the application might require by persons unfamiliar with it.
88
Appendix E: Measurement Data and Example Output

1. Introduction In this Document, the measured results for the test conducted to determine an objects attenuation factors dependence on the incoming waves angle of arrival are presented. Sample simulation outputs are also shown and discussed. 2. Object attenuation factor as a function of the angle of arrival To model the naturally occurring phenomenon in which obliquely incident waves transmit less power than waves that occurs at more normal incidences on objects, the following experiment was conducted: 1. Two antennas with know parameters were set-up a set distance apart in an anechoic chamber. 2. The path loss between the transmitting and receiving antenna was measured. 3. A large piece of tinted glass was then placed between the two antennas at various angles, and for each angle the corresponding path loss measured, see Table 4.
Table 4: Measured loss due to a piece of tinted glass at various angles between two antennas. Angle of incidence with respect to the normal (degrees) 0 5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 Loss due to tinted glass (dB) -14.8 -15.1 -14.7 -14.1 -15.4 -13.6 -15.8 -18.6 -19.0 -20.7 -22.1 -24.0 -28.3 -34.0 -39.4 -51.0 -69.3
These values were compared to the approximation (Loss at normal incidence)/cos()) where is the angle of incident with respect to the normal and found to be similar, see Figure 2. The measurement results deviated more from the approximation at large incidence angles as due to the limited size of the tinted glass and the resulting edge diffraction effects.
89 3. Example Output An example application output is shown in Figure 16 (EM Lab at the University of the Witwatersrand) which was generated using on-site measurement data. This is evident by the fact that signals are present inside the shielded chamber which result from the application down-playing the effects of high attenuation objects due to the presence of the indirect signal path contributions in the taken measurements.
Figure 16: An Example application output.
Figure 17 contains another example application output which was generated for a small office building using the advanced log distance model and on-site measurement data. Of interest is the overall low attenuation rate resulting from the use of low attenuation construction materials as well as building layout. In such buildings the placement of the required access point is of no consequence.
90
Figure 17: An Example application output
4. Conclusion Tests results for the verification of an objects attenuation factors dependence on the incoming waves angle of arrival were presented and discussed. Example application outputs when using on-site measurements to determine building material attenuation factors were also shown.
91
Appendix F: Source Code

Contained within the attached compact disc is the entire code listing for the developed WLAN propagation prediction tool, as well as an executable and the associated files (example materials database etc.) required to run it.
NAVAL POSTGRADUATE SCHOOL Monterey, California
THESIS
INDOOR PROPAGATION SIMULATION SOFTWARE by Juan Carlos Calle September 2000 Thesis Advisor: Jovan Lebaric
Approved for public release; distribution is unlimited.
20001117 028
REPORT DOCUMENTATION PAGE
Form Approved OMB No. 0704-0188
Public reporting burden for this collection of information is estimated to average 1 hour per response, including the time for reviewing instruction, searching existing data sources, gathering and maintaining the data needed, and completing and reviewing the collection of information. Send comments regarding this burden estimate or any other aspect of this collection of information, including suggestions for reducing this burden, to Washington headquarters Services, Directorate for Information Operations and Reports, 1215 Jefferson Davis Highway, Suite 1204, Arlington, VA 22202-4302, and to the Office of Management and Budget, Paperwork Reduction Project (0704-0188) Washington DC 20503. 1. AGENCY USE ONLY (Leave blank) 4. TITLE AND SUBTITLE : 2. REPORT DATE 3. REPORT TYPE AND DATES COVERED
September 2000 Indoor Propagation Simulation Software

6. AUTHOR(S)
Master's Thesis
5. FUNDING NUMBERS
Calle Juan C.
7. PERFORMING ORGANIZATION NAME(S) AND ADDRESSES)
Naval Postgraduate School Monterey, CA 93943-5000

9. SPONSORING / MONITORING AGENCY NAME(S) AND ADDRESSES)
8. PERFORMING ORGANIZATION REPORT NUMBER
N/A
10. SPONSORING/ MONITORING AGENCY REPORT NUMBER
11. SUPPLEMENTARY NOTES
The views expressed in this thesis are those of the author and do not reflect the official policy or position of the Department of Defense or the U.S. Government.
12a. DISTRIBUTION / AVAILABILITY STATEMENT 12b. DISTRIBUTION CODE
Approved for public release; distribution is unlimited.

13. ABSTRACT (maximum 200 words)
Computer simulation can be used to predict the signal strength in complex indoor environments. Signal propagation prediction is essential for determining the coverage of WLAN's (Wireless Local Area Networks). Increasing use of WLANs within infrastructures that have been built prior to the installation of the wireless networks requires measurements, semi-empirical models, or computer simulations to determine the number and location of access points for optimum coverage. In cases where the infrastructure has yet to be built (as in a new class of ships) the simulation may be the only option for WLAN coverage prediction. In such a case, blueprints may be used in conjunction with the indoor propagation simulation software in order to predict the best places to install the access points. The indoor propagation simulation software differs from the outdoor propagation software used for the cellular networks because of the differences in the characteristics of indoor and outdoor propagation channels. This thesis explains the characteristics and structure of a Wireless Local Area Network and presents the computer simulation results the 2.4 GHz wireless signal propagation inside an enclosed space. A building at the Naval Postgraduate School, where some previous physical measurements have been conducted, was selected for the indoor propagation simulation.
14. SUBJECT TERMS
Wireless Local Area Network, Simulation of Indoors propagation channel.
IS. NUMBER OF PAGES
ML.
16. PRICE CODE 17. SECURITY CLASSIFICATION OF REPORT 18. SECURITY CLASSIFICATION OF THIS PAGE 19. SECURITY CLASSmCATION OF ABSTRACT 20. LIMITATION OF ABSTRACT
Unclassified
NSN 7540-01-280-5500
Unclassified
Unclassified
UL
Standard Form 298 (Rev. 2-89) Prescribed by ANSI Std. 239-18
THIS PAGE INTENTIONALLY LEFT BLANK
11
Approved for public release; distribution is unlimited. INDOOR PROPAGATION SIMULATION SOFTWARE Juan Carlos Calle Lieutenant Junior Grade, Ecuadorian Navy B.S., Ecuadorian Naval Academy, 1992
Submitted in partial fulfillment of the requirements for the degree of MASTER OF SCIENCE IN ELECTRICAL ENGINEERING from the NAVAL POSTGRADUATE SCHOOL September 2000
Author:
Approved by: Jovan Lebaric, Thesis Advisor
Richard Adler, Thesis Co-Advisor
Jeffreys Knorr, Chairman Department of Electrical Engineering

ui
IV
ABSTRACT Computer simulation can be used to predict the signal strength in complex indoor environments. Signal propagation prediction is essential for determining the coverage of WLAN's (Wireless Local Area Networks). Increasing use of WLANs within infrastructures that have been built prior to the installation of the wireless networks requires measurements, semi-empirical models, or computer simulations to determine the number and location of access points for optimum coverage. option for WLAN coverage prediction. In cases where the infrastructure has yet to be built (as in a new class of ships) the simulation may be the only In such a case, blueprints may be used in conjunction with the indoor propagation simulation software in order to predict the best places to install the access points. The indoor propagation simulation software differs from the outdoor propagation software used for the cellular networks because of the differences in the characteristics of indoor and outdoor propagation channels. This thesis explains the characteristics and structure of a Wireless Local Area Network and presents the computer simulation results the 2.4 GHz wireless signal propagation inside an enclosed space. A building at the Naval Postgraduate School, where some previous physical measurements have been conducted, was selected for the indoor propagation simulation.
VI
TABLE OF CONTENTS
I. INTRODUCTION A.PURPOSE B. OBJECTIVES C. THESIS OUTLINE 1 1 1 2
n. BACKGROUND
A. THE MOBILE INDOOR PROPAGATION CHANNEL 1. Wireless Characteristics 2. Multipath B. WIRELESS LAN 1. Why Does the Navy Need Wireless Networks on Ships? 2. Wireless LAN Technology a. Narrowband Technology b. Spread Spectrum Technology d Infrared Technology 3. Wireless LAN Operation 4. Wireless LAN Configurations 5. Technological and Implementation Issues m. INDOOR PROPAGATION SOFTWARE A. SOFTWARE TOOLS FOR DESIGNING AND PREDICTING PERFORMANCE OFWLAN's 1. WiSE (Lucent Technologies) 2. CINDOOR (University of Cantabria, Spain) 3. WinProp (AWE) 4. SitePlanner (Wireless Valley Communications). '.
3
3 3 5 6 7 9 9 9 10 11 13 18 23
23 23 24 25 26
vii
5. PlaceBase IV. SIMULATION A. PHYSICAL MEASUREMENTS B. THE SIMULATION PROCESS V. CONCLUSIONS AND RECOMMENDATIONS A. CONCLUSIONS B. RECOMMENDATIONS LIST OF REFERENCES INITIAL DISTRIBUTION LIST
27 29 29 30 43 43 44 45 47
viu
LIST OF FIGURES Figure 1. Example of Multipath Fading Non-Line of Sight Case Figure 2. Example of Multipath Fading line of sight Case Figure 3. Access Point Installed at the Network Lab of the NPS ECE Department Figure 4. WLAN PC Card in a Notebook Computer Figure 5. A Wireless Peer-to-Peer Network Figure 6. Client and Access Point Figure 7. Multiple Access Points and Roaming Figure 8. Use of an Extension Point, Figure 9. Extension Point from the NPS ECE Network Lab Figure 10. The Use of Directional Antennas Figure 11. WiSE Output Screen Figure 12. Cindoor Output Screen Figure 13. Winpro Output Screen Figure 14. Predictor Coverage Map Output Screen Figure 15. SNR Measurements for the Second Floor of Bullard Hall Figure 16. PlaceBase Walls Editing Window Figure 17. PlaceBase Slab Editing Window Figure 18. Second Floor Relative Signal Strength (Coverage) Map Figure 20. Signal Strength for the First Floor with Staircase between Floors Figure 21. Signal Strength for the Second Floor with Two Access Points 6 6 12 13 14 14 15 16 17 18 24 25 26 27 29 32 33 35 39 41
IX
Figure 22. Signal Strength for the First Floor with Two Access Points in the Upper Floor 42
LIST OF TABLES
TABLE 1. COMPARISON BETWEEN THE PHYSICAL AND SIMULATION RESULTS TABLE 2. WALL PARAMETERS USED IN THE SIMULATION
30 31
XI
Xll
ACKNOWLEDGMENT I must take this opportunity to acknowledge the constant support of my wife, Patricia, who I had the luck to marry while I was attending NPS. I also want to recognize my parents, Carlos and Esperanza, for all the sacrifices they made in order for me to get an education which, led me to Monterey.
xni
xiv
I. INTRODUCTION Installing a wireless network system in a building is challenging with respect to the exact number and the best placement of the base stations. The placement is done either empirically or based upon measurements, resulting in considerable uncertainty regarding the minimum required number of base stations and the achievable coverage. In this thesis, accurate predictions of indoor signal propagation for a Wireless Local Area Network with the IEEE 802.11 standard are considered, using propagation simulation software, in order to determine the best place and proper number of access points to be installed for a particular environment.
A.
PURPOSE This research is based on the US Navy's interest to place Wireless Local Area
Networks (WLANs) inside future ships and submarines. Since these future ships will have more electronic equipment but limited personnel assets, it will be important to increase the productivity of every crewmember onboard. If we are able to give each crewmember appropriate information with minimal delay, the increase in efficiency will be considerable.
B.
OBJECTIVES The objective of this thesis is to evaluate software that can be used to determine
the best place to install an access point (transmitter/receiver) inside an enclosed space. Since we are interested in the next-generation ships, we must take into account that the infrastructure where the network will be installed has not been built yet. Consequently, 1
we will use the blueprints of a building located at the Naval Postgraduate School for the software evaluation, especially since there are results from another thesis where physical measurements of a WLAN were taken and documented. These measured results can be used for comparison with the simulation predictions.
C.
THESIS OUTLINE This thesis consists of five chapters. Chapter I presents the introduction. Chapter
II provides background information about the Indoor Wireless Channel and the LAN's EEEE 802.11 standard. Chapter III examines the different indoor propagation simulation software that are available. Chapter IV examines the simulation results using the software called PlaceBase. Chapter V concludes the thesis and provides recommendations.
H. BACKGROUND
THE MOBILE INDOOR PROPAGATION CHANNEL
1. Wireless Characteristics Most of the early research on the topic of indoor propagation channel dates back to 1959 [1] and most of this work focused on tunnels and mines. After the success of the cellular mobile radio systems, more measurement and modeling efforts started to appear. However, the models for outdoor cellular systems are not appropriate for indoor applications because of the difference in the channel characteristics. Although there is currently an intense effort to develop semi-empirical models for indoor (and indoor to outdoor and vice versa) propagation, that may lead to efficient tools for indoor propagation prediction (in a statistical sense), there is also a need for an accurate indoor simulation tool for accurate predictions in particular environments to which the "general" statistical models may not be applicable (such as shipboard compartments).
Accounting for the different state of progress in the wireless and wired worlds, we can identify the following major challenges faced by the wireless communication system designer:
Fading: Wireless channels are time varying. The range of channel operating characteristics for wireless modems is much wider than for typical wired
Channels. The nonstationary nature of the channel has a large impact on spectrally efficient transmitter-receiver design as well as channel capacity. Achieving low-bit-error-rate data transmission is particularly challenging in rapidly varying channels subject to a high Doppler Spread.
Multi-User: The wireless users are spatially separated (with a significant variation of mutual distances) and often uncoordinated. Spectral efficiency in multipoint-to-point and point-to-multipoint communication is significantly more challenging than in single-user communication, particularly when data rate requirements are heterogeneous. Although the rudiments of multi-user communication theory date back to the early 1970s, academic research in this area did not start until the 1980s, and mostly as a reaction to the wireless revolution.
Power Limitation: Since the majority of wireless terminals are battery operated, power efficiency, in addition to spectral efficiency, is crucial. This applies not only to transmitted power but also to circuit-dissipated power. As computational complexity translates into dissipated power, sophisticated designs required to approach capacity may not be advisable in a battery-operated terminal. Fortunately, the advances in low-power-dissipation sub-micron complementary metal oxide semiconductor (CMOS) circuit design
make increasingly complex designs not only feasible but also preferable from the standpoint of power efficiency.
2. Multipath Interference caused by signals bouncing off of walls and other barriers and arriving at the receiver at different times is called multipath interference. Multipath interference affects Infrared (IR), Radio frequency (RF), and Microwave (MW) systems. Frequency Hopping Spread Spectrum (FHSS) inherently solves the multipath problem by "hopping" between a large number of frequencies. Similarly, Direct Sequence Spread Spectrum
(DSSS) systems use large bandwidths for signal transmissions thus minimizing the probability of deep signal fading due to multipath. multipath algorithms to reduce the signal fading effects. Figure 1 shows the effect of Non-Line of Sight (there is no direct signal from the transmitter) multipath fading, where we can see that all energy is scattered from scatterers located around the wireless terminal. This causes either signal cancellation or Other systems use specific anti-
enhancement, with deep and often rapid variations in signal amplitude. Figure 2 shows the multipath effect when there is a Line of Sight transmission, adding a strong coherent component to the random scattered components. The signal fading is less pronounced than in the NLoS case and we have a distinct direction of arrival.
Transmitter
Figure 1. Example of Multipath Fading Non-Line of Sight Case (From Ref [2])
Transmitter
Figure 2. Example of Multipath Fading line of sight Case (From Ref [2])
B.
WIRELESS LAN A wireless local area network (LAN) is a flexible data communications system
implemented as an extension to, or as an alternative for, a wired LAN. Using radio frequency (RF) technology, wireless LANs transmit and receive data over the air,
minimizing the need for wired connections. Thus, wireless LANs combine data connectivity with user mobility. Wireless LANs have gained popularity in the commercial markets, including the health-care, manufacturing, warehousing, and academia. These have profited from the productivity gains of using hand-held terminals and notebook computers to transmit realtime information to centralized hosts for processing. The military is also beginning to
exploit the different advantages of wireless LANs. Today wireless LANs are becoming more widely recognized as a general-purpose connectivity alternative for a broad range of purposes.
1. Why Does the Navy Need Wireless Networks on Ships?
The widespread reliance on networking and the rapid growth of the Internet and online services are strong testimonies to the benefits of shared data and shared resources. With wireless LANs, crewmembers can access shared information without looking for a place to plug in, and network managers can set up or augment networks without installing or moving wires. Wireless LANs offer the following productivity, convenience, and cost advantages over traditional wired networks:
Mobility: Wireless LAN systems can provide LAN users with access to realtime information from anywhere on the ship. This mobility supports productivity and service opportunities not possible with wired networks.
Installation Speed and Simplicity: Installing a wireless LAN system can be fast and easy and can eliminate the need to pull cable through compartments and bulkheads.
Installation Flexibility: Wireless allows the network to go where wire cannot go-
Reduced Cost: While the initial investment required for wireless LAN hardware can be higher than the cost of wired LAN hardware, overall installation expenses and life-cycle costs can be significantly lower. Long-term cost benefits are greatest in dynamic environments requiring frequent moves and changes.
Scalability: Wireless LAN systems can be configured in a variety of topologies to meet the needs of specific applications and different type of ships. Configurations are easily changed and range from peer-to-peer networks suitable for a small number of users to full infrastructure networks of thousands of users. This enables roaming over a broad area.
2. Wireless LAN Technology Manufacturers of wireless LANs have a range of technologies, with its own set of advantages and limitations, to choose from when developing a wireless LAN solution.
a.
Narrowband Technology
A narrowband radio system transmits and receives user information on a specific radio frequency. Narrowband radio keeps the radio signal frequency range as narrow as possible, with just enough bandwidth to accommodate the required information rate. Undesirable cross talk between communications channels is avoided by carefully coordinating different users on different channel frequencies. In a radio system, privacy and noninterference are maintained by the use of separate radio frequencies. The receivers filter out all signals except the ones on their designated frequencies. One drawback of narrowband technology is that the end-user must obtain a FCC license for each site where it is employed.
b.
Spread Spectrum Technology
Most wireless LAN systems use spread-spectrum technology, a wideband radio frequency technique developed by the military for use in reliable, secure, mission-critical communications systems. Spread-spectrum is designed to trade off bandwidth efficiency for reliability, integrity, and security. In other words, more bandwidth is consumed than in the case of narrowband transmission, but the tradeoff produces a signal that is, in effect, louder and thus easier to detect, provided that the receiver knows the parameters of the spread-spectrum signal being broadcast.
If a receiver is not tuned to the right frequency, a spread-spectrum signal looks like background noise. There are two types of spread spectrum radio: frequency hopping and direct sequence.
Frequency Hopping Spread Spectrum Technology
A frequency-hopping spread-spectrum (FHSS) uses a carrier that changes frequency in a pattern known to both transmitter and receiver. When the terminals are properly synchronized, the message is maintained a single channel. To an unintended receiver, FHSS appears to be short-duration impulse noise [9].
Direct Sequence Spread Spectrum Technology
Direct-sequence spread-spectrum (DSSS) generates a redundant bit pattern for each information bit to be transmitted. This bit pattern is called a chip or chipping code. The longer the chip, the greater the probability that the original data can be recovered (at the expense of wider bandwidth). Even if one or more bits in the chip are lost or changed during transmission, statistical techniques embedded in the radio can recover the original data without the need for retransmission. To an unintended receiver, DSSS appears as low-power wideband noise and is rejected (ignored) by most narrowband receivers [9].
c.
Infrared Technology
A third technology, little used in commercial wireless LANs, is infrared. IR systems use very high frequencies, just below visible light in the
10
electromagnetic spectrum, to carry data. Like light, IR cannot penetrate opaque objects; it is either direct (line-of-sight) or diffuse technology. Inexpensive direct systems provide very limited range (3 feet) and typically are used for personal area. High performance directed IR is impractical for mobile users and is therefore used only to implement fixed sub-networks. Diffuse (or reflective) IR wireless LAN systems do not require line-ofsight, but cells are limited to individual rooms.
3. Wireless LAN Operation
Wireless LANs use electromagnetic waves of radio or infrared wavelengths to communicate information from one point to another without relying on any physical connection. EM waves are often referred to as carriers because they simply perform the function of delivering information in the form of modulation of the EM energy to a remote receiver. The data being transmitted is superimposed on a carrier for extraction at the receiving end. Once data is superimposed or modulated onto the carrier, the signal
occupies more than a single frequency, since the frequency or bit rate of the modulating information modifies the carrier. Multiple carriers can exist in the same space at the same time without interfering with each other if the waves are transmitted on different carrier radio frequencies. To extract data, a receiver is tuned to only one frequency while rejecting all other frequencies. In a typical wireless LAN configuration, an access point (AP), containing a transmitter/receiver as shown in Fig. 3, connects to a wired network from a fixed location, using standard cabling. The access point always receives, buffers, and transmits data 11
between the wireless LAN and the wired network infrastructure.
A single access point
can also support a small group of users and usually functions within a range of less than one hundred feet but can operate up to several hundred feet. The access point (or the antenna attached to the access point) is usually mounted at a high location for maximum coverage but may be mounted essentially anywhere that is practical as long as the desired coverage is obtained.
Figure 3. Access Point Installed at the Network Lab of the NPS ECE Department.
End users access the wireless LAN through wireless-LAN adapters, which are constructed on PC cards in notebook or palmtop computers, as PCI cards in desktop computers, or integrated within hand-held computers as shown in Fig. 4. Wireless LAN
12
adapters provide an interface between the client Network Operating System (NOS) and the antenna. The nature of the wireless connection is transparent to the NOS.
! ";p*;C/..fr W" *WP'"IIW ]
j'nyW.'fi,.'.1 J l"'l'"%fflW.t.u3iyi1.'*'-111*- i
111111111
Figure 4. WLAN PC Card in a Notebook Computer.
4. Wireless LAN Configurations Wireless LANs can be either simple or complex. At its most basic, two PCs equipped with wireless adapter cards can function as an independent network when they are within communication range of one another. This is called a peer-to-peer network, and is shown in Hg. 5. These on-demand networks, require no administration or
preconfiguration. In this case, each client would have access only to the resources of the other client and not to a central server.
13
Figure 5. A Wireless Peer-to-Peer Network (from Ref,[7])
Installing an access point can extend the range of an ad-hoc network, effectively doubling the range at which the devices can communicate. Since the access point is
connected to the wired network, each client would have access to server resources as well as to other clients. An example of this is shown in Fig. 6. Each access point can
accommodate many clients; the specific number depends on the number and nature of the transmissions involved. Many real-world applications exist where a single access point services from 15 to 50 client devices.
Figure 6. Client and Access Point (from Ref. [7])
Access points have a finite range, on the order of 500 feet indoors and 1,000 feet outdoors. On large Navy vessels such as Frigates, or Submarines, installing more than one
14
access point will probably be necessary. Access point positioning is accomplished by means of a site survey (physical measurements), or in the case of this thesis, by having a tool that will predict the signal strength at various locations via simulation. The goal is to blanket the coverage area with overlapping coverage "cells" such that clients might range throughout the area without ever losing network contact. More than one access point topology is shown in Fig. 7. The ability of the users to move seamlessly among a cluster of access points is called roaming. Access points handoff clients from one to another in a manner that is invisible to the client, while ensuring unbroken connectivity.
Figure 7. Multiple Access Points and Roaming (from Ref. [7])
To solve particular problems of topology, the network designer might choose to use Extension Points (EP) to augment the network of Access Points (AP). Extension Points look and function like access points, but they are not tethered to the wired network, as are APs. EPs function as their name implies: they extend the range of the network by relaying signals from a client to an AP or another EP. EPs may be strung
15
together in order to pass along messaging from an AP to far-flung clients, similar to the action of humans in a bucket brigade, passing pails of water, hand-to-hand from a water source to a fire. In Fig. 8 the WLAN with an extension point is shown and in Fig. 9 is an extension point that is being tested in the ECE network lab.
Figure 8. Use of an Extension Point, (from Ref. [7])
16
Figure 9. Extension Point from the NPS ECE Network Lab.
The last item of wireless LAN equipment to consider is the directional antenna. Assume a wireless LAN in building A is to be extended to include a 'leased" building B, one me away. One solution is to install a directional antenna on each
building, each antenna targeting the other. An example of this arrangement is shown in Fig. 10. The antenna on A is connected to the wired network via an access point The antenna on B is similarly connected to an access point in building B, extending wireless LAN connectivity in that facility.
17
Figure 10. The Use of Directional Antennas (from Ref [7])
5. Technological and Implementation Issues
Range and Coverage: The distance over which RF and IR communication
can be established is a function of transmitter power, receiver performance and the propagation path, especially in indoor environments. Interactions with typical building objects, including walls, metal, and even people, can affect how the electromagnetic energy propagates, directly affecting the range and coverage of a particular system. Solid objects block infrared signals, imposing additional
limitations on IR systems. Most wireless LAN systems use RF because radio waves can partially penetrate most indoor walls and obstacles. The range (or radius of coverage) for typical wireless LAN systems varies from under 100 feet to more than 300 feet. Coverage can be extended, and true freedom of mobility via roaming can be provided through microcells or micronets.
18
Throughput: Actual throughput performance in wireless LANs is equipment Factors that affect
and installation dependent, just as it is in wired LANs.
throughput include the number of users, propagation factors such as range and multipath, and the type of wireless LAN system used. Latency and bottlenecks on the wired portions of the LAN also affect throughput. Data rates for the most common commercial wireless LANs are in the 1.6 Mbps range. Users of
traditional Ethernet or Token Ring LANs generally experience little difference in performance when using a wireless LAN. Wireless LANs provide throughput sufficient for the most common LAN-based office applications, including electronic mail exchange, access to shared peripherals, Internet access, and access to multi-user databases and applications. By comparison, state-of-the-art V.90 modems transmit and receive at optimal data rates of 56.6 Kbps. In terms of throughput, a wireless LAN operating at 1.6 Mbps is almost thirty times faster.
Integrity and Reliability: Wireless data technologies have been proven
through more than fifty years of wireless application in both commercial and military systems. While radio interference can cause degradation in throughput, such interference is rare in the workplace. Robust designs of proven wireless LAN technology and the limited distance over which signals travel can produce connections that are far more robust than cellular phone connections and provide data integrity performance equal to or better than wired networking.
19
Compatibility with Existing Networks: Most wireless LANs provide
military standard interconnections with wired networks, such as Ethernet or Token Ring. Wireless LAN nodes are supported by network operating systems in the same fashion as any other LAN node: through the use of the appropriate drivers. Once installed, the network treats wireless nodes like any other network component.
Interoperability of Wireless Devices: Wireless LAN systems from different First, different
vendors might not be interoperable, for several reasons.
technologies do not interoperate. A system based on spread spectrum frequency hopping (FHSS) technology is not compatible with another system based on spread spectrum direct sequence (DSSS) technology. Second, systems using
different frequency bands will not interoperate even if they both employ the same technology. Third, systems from different vendors may not interoperate even if they both employ the same technology and frequency band, due to differences in implementation by each vendor.
Interference and Coexistence: The unlicensed nature of radio-based wireless
LANs make them susceptible to some measure of interference . Microwave ovens are a potential concern, but most wireless LAN designers have accounted for microwave interference. Another possible concern is the co-location of multiple wireless LANs. While wireless LANs from some manufacturers interfere with 20
other wireless LANs, others coexist without interference. Careful selection must be made by evaluating each vendor's specific system.
Licensing Issues: In the United States, the Federal Communications
Commission (FCC) regulates most radio transmissions, including those used by wireless LANs. Other nations have corresponding regulatory agencies. Wireless LANs are typically designed to operate in portions of the radio spectrum where the FCC does not require licensing. In the U.S. most wireless LANs broadcast over
one of the ISM (Instrumentation, Scientific, and Medical) bands. These include 902-928 MHz, 2.4-2.483 GHz, 5.15-5.35 GHz, and 5.725-5.875 GHz. For
wireless LANs to be sold in a particular country, the manufacturer of the wireless LAN must ensure its certification by the appropriate agency in that country.
Simplicity and Ease of Use: The wireless nature of a WLAN is transparent to
a user's NOS, and applications work the same as they do on wired LANs. Wireless LAN products incorporate a variety of diagnostic tools associated with the wireless elements of the system; however, products are designed so that most users rarely need these tools. Wireless LANs simplify many of the installation and configuration issues that plague network managers. Since access points are the only portion of wireless LANs that require cabling, cable pulling is minimized. This simplifies moves, adds, and changes simpler for WLANs. Finally, the
portable nature of wirelesses LANs lets network managers preconfigure and 21
troubleshoot the software aspects of entire networks before installing them at remote locations. Once configured, wireless LANs can be moved from place to place with little or no modification.
Security: Because wireless technology has roots in military applications,
security has long been a design criterion for wireless devices. Security provisions are typically built into wireless LANs, making them more secure than most wired LANs. It is extremely difficult for unintended receivers (eavesdroppers) to listen in on wireless LAN traffic. Complex encryption techniques make it impossible for all but the most sophisticated to gain unauthorized access to network traffic. In general, individual nodes must be security-enabled before they are allowed to participate in network traffic.
Radiation Safety: The output power of wireless LAN systems is very low and
is, much less than that of a hand-held cellular phone. Since radio waves attenuate rapidly with distance, personnel in the area of a wireless LAN system are exposed to very little RF energy.
22
m. A.
INDOOR PROPAGATION SOFTWARE
SOFTWARE TOOLS FOR DESIGNING AND PREDICTING PERFORMANCE OF INDOOR WffiELESS SYSTEMS This thesis research was initiated with a search for software that would meet the
following requirement: A propagation prediction software that would aid in designing indoor WLANs. There are several types of such software, and the following examines some of them. 1. WiSE (Lucent Technologies) WiSE is a system for designing indoor wireless systems. Given building data (wall locations and composition), WLAN system parameters, and desired AP locations, WiSE determines system performance (e.g., received power and delay spread) throughout the building or at specific points. For the WLAN specified and the required signal strength threshold, WiSE determines base-station locations which will maximize the fraction of the building over which the performance requirement is met. A sophisticated graphical user interface allows interactive design and optimization. A sample output screen is shown in Fig. 11.
23
-itrrni- iiliinniilini liini iiilrtt ri r-'i ITTIMIirrtt wtni BuSAnâ BJ fcx! Prct<>f Pjrinfir niaphy' Hfc>~; QuitjScat "" ThrWotd <dBi) -'"-","-" ----
SC*S*<35. -
M,j-,~rr*a -J*~~n
Figure 11. WiSE Output Screen (from Ref. [3])
2. CINDOOR (University of Cantabria, Spain) CINDOOR is a software engineering tool for use in the design, planning, and effective implementation of wireless systems in enclosed spaces. CINDOOR features a flexible propagation prediction process , which allows the analysis of indoor and outdoor environments and the interaction between them. The numerical modeling method is based on a full three-dimensional implementation of the Geometric Optics/Unified Theory of Diffraction (GO/UTD). Ray tracing is efficiently carried out by combining Image Theory with Binary Space Partitioning algorithms. The space-time distribution of the
electromagnetic field is processed to obtain a set of WLAN performance parameters: Coverage (mean power), fading statistics, power delay profile, and other associated parameters, such as rms delay spread and coherence bandwidth are produced. A typical Cindoor output screen is shown in Fig. 12.
24
[0,-8] dB [-8, -12] dB [-12.-181 dB [-18.-24] dB [-24.-30] dB &"'" [-30. -38] dB ' [-38,42] dB WM [-42,-48]dB f . [-48.-54] dB [54,-60] dB [-60. -88] dB [-66.-72] dB E -72L -78] dB ;-. [-78,-84] dB [-84,-90] dB [ <-9G]d8
Figure 12. Cindoor Output Screen (from Ref. [4])
3. WinProp (AWE) WinProp's unique Combined Network Planning(CNP) features are ideal for use in macro, micro and picocell network planning. It features terrain (macroceH) predictions using the parabolic equation method, predictions for urban ceDs, indoor receivers with penetration of buildings in mobile communication networks , indoor wireless networks, and planning of wireless LAN's (WLAN). Some output products of WinProp include prediction of field strength, received power, prediction of both delay spread and fast fading. A sample output screen from Winpro is shown in Fig. 13.
25
jplwtaProp wwn.lpfc
j0
OMPUUKJCI -..'gat'.
r*m sor J*; :Si*w. 5?"
(S-H:Sli B jo /- > , ^y jaffc """V'

1
'S'/*'
'<"' J -*
;
gsm?
: ; ;
?"""7""%^^
/ :
^
\! U
1 ! ill
:--r-rv"
i
;3:.aa
I
M5R
gtJMfc.
Figure 13. Winpro Output Screen (From Ref. [5])
4. SitePlanner (Wireless Valley Communications) SitePIanner is a fully-integrated design, measurement, optimization, and management engineering tool for in-building, campus-wide, and microcell wireless communications systems. The tool suite includes the Predictor, InFielder, and Optimatic modules that work together for all phases of deployment, maintenance, and optimization of a local wireless system. One of the output screens of the Predictor, module from SitePlanner, is shown in Fig. 14.
26
Figure 14. Predictor Coverage Map Output Screen (From Ref. [6])
5. PlaceBase PlaceBase was designed for two- and three-dimensional modeling of radio propagation in buildings, in the frequency range from 500 to 5200MHz. PlaceBase uses Geometric Optics and the Multi-Channel Coupling (MCQ algorithm, which reduces the computation load while providing reasonable prediction accuracy. PlaceBase considers the pattern and orientation of transmitter antennas and calculates multiple reflections and transmissions produced by structural elements of the building. Pkcebase was chosen for the simulation in this thesis. The process to select it was the following: We started testing a Demo version for Cindoor, the problems with this software was, that it was not user friendly, it was difficult to build the model and it did not have a printed output. Then we tested a Demo version of WinProp, the problem with this demo was that it did not allow us to build different types of indoor models. After this we tried to
27
test Site Planner but the company that produces this software does not provide Demo versions. There was a visit from one of their representatives but we were not able to test the software with our own model or frequency parameters. Finally we were able to get a Demo version of Placebase. This software was the one that gave us the flexibility to build our own model and that is why Placebase was chosen.
28
IV. SIMULATION
A.
PHYSICAL MEASUREMENTS As stated before, indoor signal strength measurements in the 2.4GHz ISM band
are documented in another thesis [8]. The measurements were made in Bullard Hall on the campus of the Naval Postgraduate School, using a wireless PCMCI card (IEEE 802.11 standard) and its accompanying software. The software presents Signal to Noise Ratio (SNR) on the computer screen while the user moves around the enclosed space, as shown in Fig. 15. Both signal and noise are shown on a linear scale from 0 to 100. Note that the drawing is not to scale.
Equipment/Obstruction
Door
Stairwy
Wireless Access Point
%signal/ %noisc
Figure 15. SNR Measurements for the Second Floor of Bullard Hall (From Ref. [8]) 29
To present an additional comparison of the relative signal levels predicted by the simulation and the measured signal levels. Table. 1 was created. In this table the results shown in Fig. 15 have been converted to dB and scaled so they can be compared with the relative signal levels predicted by the simulation wich are presented in Fig. 18.
if^HSiKiiiwi.
82/12 61/12 43/10 35/10 30/10 17/12 m î^^JJWRjP.-Vîtw^'] -40
-42.5
23
r^âBâ^
-40 -44 -45 -50 -61 -63
-44 -46 -47 -54
Table 1. Comparison Between the Physical and Simulation Results.
B.
THE SIMULATION PROCESS. To validate the simulation predictions qualitatively, the model of the second floor
of Bullard Hall (where previous measurements were taken) was developed, based on building blueprints. In order to use this model with PlaceBase software, electromagnetic characteristics of the walls were required and are listed in Table. 2.
30
Transparent Window Conductor Staircase Concrete Concrete Slab Plaster 2.3 7.4 0 0.1 0.02 0.05 0.05 0.03 0.05 0.1 0.4 0.25 0.40 0.1
To"
4.4
Table 2. Wall Parameters Used in the Simulation.
PlaceBase provides two working screens of input data: one screen to enter information on the location, size and composition of the walls and another screen for "slabs," which in this case are floors and ceilings, as shown in Figs. 16 and 17, respectively. This requires that the thickness of the walls, floors, and ceilings and thencomposition be known prior to model construction. Most of the building wall surfaces are plaster, glass windows, and concrete. The floor and ceiling are forty by forty meters concrete slabs. Most of this information was available from blue prints and from a physical inspection of the building. Different wall types are represented by the variety of colors on the Walls Editor Screen, shown in Fig. 16. A slab editor screen was used in defining two staircases and the thickness of the slab, as shown in Fig. 17. In addition, this screen can be used to define additional floors, for floor-to-floor propagation simulations.
31
Re
Edft
Parameters Show<'H'de "Zoom
look
Help
-lu\*\
Figure 16. PlaceBase Walls Editing Window
32
&*Smm*m
jjEgg&SEECSEiB Ffe EcR Parameter SbwwHide Zoom 1* Help
**m *:12?M
Figure 17. PlaceBase Slab Editing Window
33
After completing the editing of the walk and the slabs (floors/ceilings), it is necessary to define the transmitter: type, power, frequency (2.4 GHz) IEEE 802.11b, and the source coordinates (x, y, z). These parameters were matched with those of the physical measurements and are shown in Fig. 15. Fig. 18 shows the final simulation result for the structures of the second floor of Bullard Hall with all the walls, slabs, and transmitter parameters defined so that they best match the actual building and the conditions for the measurements. The cross (+) marks the transmitter access point location. The location and height of the AP are the same as those used for the measurements. The relative signal strength (in dB) at a height of 1.5 meters above the floor is shown in the color coverage map, with red for the highest signal strength, which is close to the transmitter.
34
00
o
CM
03 D O co
'S CO
fei
CS
o o
UL
2
"v> T5 JS ja
\*m
o p
WKsmBBs^Sm^^^lS^KA
Mj
CG CO
Figure 18. Second Floor Relative Signal Strength (Coverage) Map 35
Tabulated measured data for coordinates and the signal strength is not available, thus only a qualitative comparison between calculations and measurements can be made. This can be done by comparing Fig. 15 and Fig. 18. Please note that the floor plans in the two figures are "reversed." (Rotating either figure by 90 degrees aligns the simulated floor plan with the measured one.) A good agreement (the overlap of the gray area from the measurements plot and the orange/yellow from the simulation) is observed from the visual comparison of Figs. 15 and 18. One problem with PlaceBase is the complexity of including furniture in the simulation model. The measured floor signal strength includes the effects of furniture, but the simulated model does not. This fact, in addition to the possible inaccuracies in the model definition (exact thickness and electrical parameters for the walls, presence of metal meshes and pipes inside the walls, etc.) would most likely account for the differences between the simulated and measured values.
In addition to signal strength prediction for the floor which houses the transmitter, PlaceBase also includes floor-to-floor propagation prediction. In Fig. 19 and Fig. 20, the signal strength is shown for the following two cases: When we do not have a significant opening between the two floors, and When the model includes a staircase between the two floors.
There is a significant propagation from the second floor to the first floor, even without the staircase. This is due to floor-to-floor propagation through slabs/walls, windows, vent
36
ducts and elevator spaces, etc. The signal strength on the first floor increases, as expected, with the staircase included in the model.
37
g SS
Z3 CO TJ O
li-
in
o
CO CD
en
< _j
Z>
; 03
oa
TO
CO
o 00
mm
eu.,
Q,
TJ
[. ;
CQ XJ
o o
CQ TJ O
CO TJ O <N
Figure 19. Signal Strength for First Floor without Staircase between the Floors 38
CO
2
TJ O CO
03
"S
3 03 C it.
CO O
o o u.
3
CO
T3 O <0
t>
no -o o IV.
0" m o =0
4>
TJ
CO
Figure 20. Signal Strength for the First Floor with Staircase between Floors 39
From Chapter II, recall the objective in setting up a WLAN is to enable the clients to move throughout the WLAN area without losing network contact. This can be
achieved by having more than one access point. As shown in Figs. 21 and 22, adding one more access point increases the uniformity of the signal strength and thus the reliability of the WLAN connection for a mobile user. As an alternative to using multiple access points, the transmitted power may be split among several antennas, separated in space so that they provide improved reliability of the WLAN network connection [11], [14]. Another option is to implement a
"distributed" antenna in the form of a radiating or "leaky" coaxial cable, where the outer conductor is "slotted" to allow radiation to escape along the cable length. These antenna options however negate the advantage of easy access point relocation.
40
CD
!o
03
o
to
CD
"O XI
Hi
CQ
CO c o o
CO
in
ICQ
Io
I <0
"o
"o
1
m o
CO D
a.
O
CO
IIBo e&Sf o
m o o o
CO O
CO D
CO
TJ O
CM
Figure 21. Signal Strength for the Second Floor with Two Access Points 41
CD
CD
CD O
_CB
XI
co
O
(B
C
to
o o
CD
x> o
T3
as
CQ
"o
m xs
CO
CQ
o
o o
Figure 22. Signal Strength for the First Floor with two Access Points in the Upper Floor
42
V. CONCLUSIONS AND RECOMENDATIONS
A.
CONCLUSIONS
By comparing the coverage area displayed in Figures 15 and 18, we are able to see the similarity in the form and the distance of the signal coverage area. In addition and to confirm the results obtained with the simulation the SNR results displayed in Fig. 15 were compared to the ones obtained in the simulation, Fig 18, this is presented in Table 1. The results from the simulation have some differences from the physical results, because the model build for the simulation did not include the furniture and other type of obstacles. PlaceBase can produce a signal propagation coverage map. An accurate measurement is complicated, however, because during the building of the model, the furniture and other obstacles are very difficult to include. This is an important fact to consider when analyzing PlaceBase predictions inside ships or submarines where the enclosed spaces are crowded with many objects, which can scatter electromagnetic waves. The average price for typical WLAN test software is $10,000. This cost is too high for use on a single installation, but for a Navy project involving the installation of multiple WLAN's, it could be both useful and costeffective, especially if the vessel is not yet in the construction phase where physical measurements can be made. During the time frame of his thesis work, an explosion in the market of indoor wireless equipment has occurred. This is due to the mobility advantage that this technology offers. Different working groups, such as Bluetooth, IEEE 802.11b (high data rate), will be competing for the home and office WLAN market. This will reduce the cost of simulation software, the implementation of the equipment and the testing of these networks. 43
B.
RECOMMENDATIONS
Due to the growing demand of WLAN's, software for predicting signal propagation is becoming more accurate and user-friendly. Features like three-dimensional displays, and the capability to include furniture and fine geometrical details are becoming available. The software recommended here for further investigation is the SitePlanner (Wireless Valley Communications) software addressed in Chapter III, Ref. [6]. The Navy can reduce costs by correctly and efficiently planning the distribution of the access points, through the use of simulation and physical measurements during the initial stages of planning.
This thesis compared physical signal measurements with simulation, from a sample of available and affordable software. Software that provides additional information is needed. Examples of such factors include simulation with the receivers moving* around the enclosed space, an increased number of obstacles or people inside the space, channel interference, throughput, etc.
44
LIST OF REFERENCES 1. L.P. Rice, Radio Transmission into Buildings at 35 and 150 mc, Bell SystTechJ, vol. 38, no.l, pp. 197-210, Jan. 1959. 2. Lucent Technologies Ine, http:// www.plan9.bell-labs.com 3. Universidad de Cantabria, Cindoor project, http:// www.ccaix3.unican.es 4. Winprop, http://www.winprop.de 5. Wireless Valley Communications, Predictor, http://www.wvcomm.com
6. Wireless LAN's Working Group, http://ieee802.org/l 1/ 7. Steven M. Debus, Feasibility Analysis for a Submarine Wireless Computer Network Using Commercial Off-the-Shelf Components. Master's Thesis, Naval Postgraduate School, Monterey, CA, September 1998. 8. Lucent Technologies INC., WaveLAN Sales Bulletin: Differences between FHSS and DSSS Products, Parsippany, NJ. 9. Simon R. Saunders, Antennas and Propagation for Wireless Communication Systems, Baffis Lane, Chichester, England, 1999. 10. P. Chow, A. Karim, V. Fung and C. Dietrich, Performance Advantages of Distributed Antennas in Indoor Wireless Communication Systems, In Proc. IEEE Vehicular Technology Conference, Vol.3 1522-1526, 1994. H.H. Hashemi and D. The Indoor Radio Propagation Channel, Proc. IEEE, 81 (7), 943-67,1993. 12. A. A M. Saleh and R.A Valenzuela. A Statistical Model for Indoor Multipath Propagation, IEEEJ. Selected Areas in Commun, 5 (2), 28-137, 1987. 13. A AM. Saleh, A.J. Rustako, Jr and R.S. Roman, Distributed Antennas for Indoor Communications, IEEE Trans. Commun, 35 (11), 1245-51,1987. 14. Sergio Verdu. Wireless Bandwidth in the Making, IEEE Communications Magazine, Vol 38 No 7,53-58, July 2000. 15. Ray Tracing e.g. J.W. McKnow and R.W. Hamilton, "Ray Tracing as a Design Tool for Radio Networks," IEEE Network Magazine, Nov. 1991,27.
45
16. J.H. Tarng. T.R. Liu. Effective Models in Evaluating Radio Coverage on Single Floors of Multifloor Buildings, IEEE Trans. Vehicular Tech. Vol 48, No 3. May 1999. 17. W. Stallings. Data and Computer Communications, Prentice-Hall, Upper Saddle River, NJ, 1997. 18. T. Rappaport. Wireless Communications, Prentice-Hall, Upper Saddle River, NJ, 1999. 19. K. Feher. Wireless Digital Communications, Prentice-Hall, Upper Saddle River, NJ, 1995.
46
INITIAL DISTRIBUTION LIST
1.
Defense Technical Information Center. 8725 John Kirgman Rd, STE 0944 Ft. Belvoir, VA 22060-6218 Dudley Knox Library Naval Postgraduate School 411 Dyer Road Monterey, California 93943-5101 Chairman, Code EC Electrical Engineering Department Naval Postgraduate School Monterey, CA 93943-5121 Dr. Jovan Lebaric, Code EC/Lb Electrical Engineering Department Naval Postgraduate School Monterey, CA 93943-5121
2.
3.
5.
Dr. Richard Adler, Code EC/Ab. Electrical Engineering Naval Postgraduate School Monterey, CA 93943-5121 Direccion General del Material de la Armada. Base Naval Sur Ave. 25 de Julio Guayaquil-Ecuador Direccion General de Informatica de la Armada. Edificio de la Comandancia General de Marina Ministerio de Defensa Nacional Quito-Ecuador
8.
LTJG Juan Carlos Calle Cdla. El Bosque, Pasaje D, casa # 45 Quito-Ecuador
47
Throughput Measurements and Empirical Prediction Models for IEEE 802.11b Wireless LAN (WLAN) Installations
Benjamin E. Henty
Thesis submitted to the Faculty of the Virginia Polytechnic Institute and State University in partial fulllment of the requirements for the degree of
Master of Science in Electrical Engineering
Dr. Theodore S. Rappaport, Chair Dr. Ioannis M. Besieris Dr. Scott F. Midkiff Dr. William H. Tranter
August 9, 2001 Blacksburg, Virginia
Keywords: wireless LAN, IEEE 802.11b, throughput measurement, throughput prediction, RSSI measurement
Copyright 2001, Benjamin E. Henty
Throughput Measurements and Empirical Prediction Models for IEEE 802.11b Wireless LAN (WLAN) Installations
Benjamin E. Henty
(ABSTRACT)
Typically a wireless LAN infrastructure is designed and installed by Networking Professionals. These individuals are extremely familiar with wired networks, but are often unfamiliar with wireless networks. Thus, Wireless LAN installations are currently handicapped by the lack of an accurate, performance prediction model that is intuitive for use by non-wireless professionals. To provide a solution to this problem, this thesis presents a method of predicting the expected wireless LAN throughput using a site-specic model of an indoor environment. In order to develop this throughput prediction model, two wireless LAN throughput measurement products, LANFielder and SiteSpy, were created. These two products, which are patent pending, allow site-specic network performance measurements to be made. These two software packages were used to conduct an extensive measurement campaign to evaluate the performance of two IEEE 802.11b access points (APs) under ideal, multiuser, and interference scenarios. The data from this measurement campaign was then used to create empirically based throughput prediction models. The resulting models were rst developed using RSSI measurements and then conrmed using predicted signal strength parameters.
This work was supported through the Bradley Fellowship, the Bradley Industrial Fellowship, and through donations from ACDC, Inc. and Wireless Valley Communications, Inc.
The author would like to acknowledge the following people for assisting in making this work possible: Wireless Valley for the use of software and equipment. MPRG for the support, encouragement, equipment and supplies. Jiun Siew for his hard work and assistance in measurements and modeling. Max Roberts for conversations and reections about this work. Roger Skidmore for software and additional support. Charles Lepple for his advice and assistance. Ros Neplokh for sharing his near god-like competence in conguring networks. Sam Abtek for listening. And of course, my friends and family for their continued love and support.
iv
Contents
1 Introduction 1.1 1.2 1.3 2 Wireless Data Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Motivation for Throughput Prediction . . . . . . . . . . . . . . . . . . . . . . Thesis Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 1 1 2 5 5 5 8 9 9
Prior Research 2.1 Wireless LAN Propagation Research . . . . . . . . . . . . . . . . . . . . . . . 2.1.1 2.1.2 2.2 RF Coverage Measurement Techniques . . . . . . . . . . . . . . . . . Wireless LAN Related RF Propagation Research . . . . . . . . . . . .
Wireless LAN Network Performance Research . . . . . . . . . . . . . . . . . 2.2.1 2.2.2 2.2.3 2.2.4 Network Performance Statistics . . . . . . . . . . . . . . . . . . . . . .
Network Protocols . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11 Network Performance Measurement Techniques . . . . . . . . . . . . 12 Network Performance Measurement and Prediction Research . . . . 14
2.3 3
Summary of Prior Research . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17 19
Software Development 3.1
Motivation for a Wireless LAN Measurement Tool . . . . . . . . . . . . . . . 19 3.1.1 3.1.2 3.1.3 Difculties with Data Networks . . . . . . . . . . . . . . . . . . . . . 19 The Client/Server Architecture . . . . . . . . . . . . . . . . . . . . . . 20 Wireless Data Network Challenges . . . . . . . . . . . . . . . . . . . . 21
3.2
LANFielder and SiteSpy Details . . . . . . . . . . . . . . . . . . . . . . . . . . 22 v
3.2.1 3.2.2 3.2.3 3.2.4 3.2.5 3.2.6 3.2.7 3.3
Client/Server Architecture . . . . . . . . . . . . . . . . . . . . . . . . 22 Site-Specic Data Recording in LANFielder and SiteSpy . . . . . . . 23 Measured Network Performance Statistics . . . . . . . . . . . . . . . 24 Test Packet Format . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24 Additional Congurable Parameters of SiteSpy and LANFielder . . 25 Network Performance Calculations . . . . . . . . . . . . . . . . . . . 25 Rationale Behind Network Performance Measurement Technique . . 26
Custom RSSI Measurement System . . . . . . . . . . . . . . . . . . . . . . . . 27 3.3.1 3.3.2 3.3.3 3.3.4 Motivation of RSSI Measurement Technique . . . . . . . . . . . . . . 27 Justication for RSSI Measurement Technique . . . . . . . . . . . . . 28 RSSI Measurement Software . . . . . . . . . . . . . . . . . . . . . . . 28 RSSI Measurement Hardware . . . . . . . . . . . . . . . . . . . . . . . 29
3.4 4
Summary of Measurement Software . . . . . . . . . . . . . . . . . . . . . . . 29 31
Wireless LAN Measurement Campaign 4.1
Measurement Campaign Overview . . . . . . . . . . . . . . . . . . . . . . . . 31 4.1.1 4.1.2 4.1.3 Measurement Test-Bed . . . . . . . . . . . . . . . . . . . . . . . . . . . 31 Additional Equipment . . . . . . . . . . . . . . . . . . . . . . . . . . . 34 RSSI Measurement Congurations . . . . . . . . . . . . . . . . . . . . 34
4.2
Organization of Measurements . . . . . . . . . . . . . . . . . . . . . . . . . . 36 4.2.1 4.2.2 Measurement Cases . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37 Measurement Scenarios . . . . . . . . . . . . . . . . . . . . . . . . . . 41
4.3
Measurement Equipment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45 4.3.1 4.3.2 4.3.3 Physical Locations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45 Equipment Conguration . . . . . . . . . . . . . . . . . . . . . . . . . 45 Software Conguration . . . . . . . . . . . . . . . . . . . . . . . . . . 46
4.4 4.5
Measurement Precautions and Verication . . . . . . . . . . . . . . . . . . . 46 Measurement Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47 vi
Wireless LAN Measurement Results 5.1
49
Format Used to Present Throughput Data . . . . . . . . . . . . . . . . . . . . 49 5.1.1 5.1.2 Meaning of SNR . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49 Overview of the Data Measured . . . . . . . . . . . . . . . . . . . . . 50
5.2
Scenario A Data - One Client . . . . . . . . . . . . . . . . . . . . . . . . . . . 51 5.2.1 5.2.2 5.2.3 5.2.4 5.2.5 5.2.6 WaveLAN, Scenario A, All Data . . . . . . . . . . . . . . . . . . . . . 51 WaveLAN, Scenario A, Spatially Averaged Data . . . . . . . . . . . . 52 WaveLAN, Scenario A, Window Averaged Data . . . . . . . . . . . . 53 3Com, Scenario A, All Data . . . . . . . . . . . . . . . . . . . . . . . . 54 3Com, Scenario A, Spatially Averaged Data . . . . . . . . . . . . . . . 55 3Com, Scenario A, Window Averaged Data . . . . . . . . . . . . . . . 56
5.3
Scenario B Data - Two Clients . . . . . . . . . . . . . . . . . . . . . . . . . . . 57 5.3.1 5.3.2 5.3.3 5.3.4 5.3.5 5.3.6 WaveLAN, Scenario B, All Data . . . . . . . . . . . . . . . . . . . . . 57 WaveLAN, Scenario B, Spatially Averaged Data . . . . . . . . . . . . 58 WaveLAN, Scenario B, Window Averaged Data . . . . . . . . . . . . 59 3Com, Scenario B, All Data . . . . . . . . . . . . . . . . . . . . . . . . 60 3Com, Scenario B, Spatially Averaged Data . . . . . . . . . . . . . . . 61 3Com, Scenario B, Window Averaged Data . . . . . . . . . . . . . . . 62
5.4
Scenario C Data - One Client with Interference . . . . . . . . . . . . . . . . . 63 5.4.1 5.4.2 5.4.3 5.4.4 5.4.5 5.4.6 WaveLAN, Scenario C, All Data . . . . . . . . . . . . . . . . . . . . . 63 WaveLAN, Scenario C, Spatially Averaged Data . . . . . . . . . . . . 64 WaveLAN, Scenario C, Window Averaged Data . . . . . . . . . . . . 65 3Com, Scenario C, All Data . . . . . . . . . . . . . . . . . . . . . . . . 66 3Com, Scenario C, Spatially Averaged Data . . . . . . . . . . . . . . . 67 3Com, Scenario C, Window Averaged Data . . . . . . . . . . . . . . . 68
5.5
Scenario D Data - Two Clients with Interference . . . . . . . . . . . . . . . . 69 5.5.1 5.5.2 WaveLAN, Scenario D, All Data . . . . . . . . . . . . . . . . . . . . . 69 WaveLAN, Scenario D, Spatially Averaged Data . . . . . . . . . . . . 70 vii
5.5.3 5.5.4 5.5.5 5.5.6 5.6 5.7
WaveLAN, Scenario D, Window Averaged Data . . . . . . . . . . . . 71 3Com, Scenario D, All Data . . . . . . . . . . . . . . . . . . . . . . . . 72 3Com, Scenario D, Spatially Averaged Data . . . . . . . . . . . . . . . 73 3Com, Scenario D, Window Averaged Data . . . . . . . . . . . . . . . 74
Analysis of Throughput and SNR Measurements . . . . . . . . . . . . . . . . 75 Interference Signal Strength Measurements and Predictions . . . . . . . . . 75 5.7.1 5.7.2 5.7.3 5.7.4 5.7.5 Predicting the Interference at Signal Strength Locations . . . . . . . . 76 Computing Total Interference Signal Strength . . . . . . . . . . . . . 80 Advantages of SIR Calculation . . . . . . . . . . . . . . . . . . . . . . 81 Resulting Plots of Measured and Predicted SIR Data . . . . . . . . . 82 Analysis of Interference Measurements and Calculations . . . . . . . 91
5.8 6
Measurement Results Conclusions . . . . . . . . . . . . . . . . . . . . . . . . 91 93
Empirical Throughput Prediction Models 6.1 6.2 6.3
Development of Wireless LAN Performance Models . . . . . . . . . . . . . . 93 Optimizing Prediction Model Parameters . . . . . . . . . . . . . . . . . . . . 93 Prediction Models for Scenario A . . . . . . . . . . . . . . . . . . . . . . . . . 94 6.3.1 6.3.2 6.3.3 Scenario A, Linear Model . . . . . . . . . . . . . . . . . . . . . . . . . 94 Scenario A, Exponential Model . . . . . . . . . . . . . . . . . . . . . . 98 Foundations for Other Scenario Models . . . . . . . . . . . . . . . . . 102
6.4
Prediction Models for Scenario B . . . . . . . . . . . . . . . . . . . . . . . . . 106 6.4.1 6.4.2 6.4.3 6.4.4 6.4.5 Throughput Prediction Formula for Two Simultaneous Users . . . . 106 Calculating Parameters for the Two User Model . . . . . . . . . . . . 107 Constraining the Two User Throughput Measurement Model . . . . 108 Extending the Two User Model to N Users . . . . . . . . . . . . . . . 108 Analysis of the Two User Throughput Prediction Model . . . . . . . 110
6.5
Prediction Models for Scenario C . . . . . . . . . . . . . . . . . . . . . . . . . 113 6.5.1 6.5.2 Quantifying the Interference Power . . . . . . . . . . . . . . . . . . . 113 Formulation of the Scenario C Model . . . . . . . . . . . . . . . . . . 114 viii
6.5.3 6.5.4 6.5.5 6.5.6 6.6
Interference Model Parameters . . . . . . . . . . . . . . . . . . . . . . 114 Calculating the Interference Model Parameters. . . . . . . . . . . . . 115 Constraining the One User with Interference Model . . . . . . . . . . 116 Analysis of the One User with Interference Model . . . . . . . . . . . 117
Prediction Models for Scenario D . . . . . . . . . . . . . . . . . . . . . . . . . 121 6.6.1 6.6.2 Creating the Two Client with Interference Model . . . . . . . . . . . . 121 Analysis of the Two User with Interference Model . . . . . . . . . . . 122
6.7 6.8 7
Procedure for Optimizing Models in a Computer Simulation Tool . . . . . . 127 Summary of Prediction Models . . . . . . . . . . . . . . . . . . . . . . . . . . 128 129
Predicting Throughput from a Building Model 7.1 7.2
Motivation for Predicting Throughput from a Building Model . . . . . . . . 129 Prediction Technique . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129 7.2.1 7.2.2 Signal Strength Prediction . . . . . . . . . . . . . . . . . . . . . . . . . 130 Throughput Prediction from Predicted Signal Strength . . . . . . . . 130
7.3 8
Successful Prediction of Throughput . . . . . . . . . . . . . . . . . . . . . . . 130 141
Conclusions 8.1
Summary of Findings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141 8.1.1 8.1.2 8.1.3 Wireless LAN Measurement Software . . . . . . . . . . . . . . . . . . 141 Wireless LAN Measurement Campaign . . . . . . . . . . . . . . . . . 142 Throughput Prediction Models . . . . . . . . . . . . . . . . . . . . . . 143
8.2
Further Research . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143 145
A Overview of IEEE 802.11 and 802.11b
A.1 Organization of the Standard . . . . . . . . . . . . . . . . . . . . . . . . . . . 145 A.2 Physical Layer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145 A.2.1 DSSS Frequency Band and Channels . . . . . . . . . . . . . . . . . . . 146 A.2.2 DSSS Advantages . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146 A.2.3 Data Bandwidths and Modulation Techniques . . . . . . . . . . . . . 147 ix
A.2.4 1 and 2 Mbps Transmission Rates . . . . . . . . . . . . . . . . . . . . . 148 A.2.5 5.5 Mbps and 11 Mbps Transmission Rates . . . . . . . . . . . . . . . 149 A.2.6 CCK Encoding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150 A.2.7 PBCC Encoding for 11 and 5.5 Mbps Transmission Rates . . . . . . . 153 A.2.8 PHY Layer Packet Format . . . . . . . . . . . . . . . . . . . . . . . . . 154 A.3 MAC Layer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155 A.3.1 Basic Network Layout . . . . . . . . . . . . . . . . . . . . . . . . . . . 155 A.3.2 MAC Layer Packet Structure . . . . . . . . . . . . . . . . . . . . . . . 156 A.3.3 Multiple Access, DCF and CSMA/CA . . . . . . . . . . . . . . . . . . 156 A.3.4 RTS/CTS and the Hidden Terminal Problem . . . . . . . . . . . . . . 160 A.3.5 Additional Optional Provisions of the MAC Layer . . . . . . . . . . . 160 Vitae 167
List of Figures
3.1 3.2 3.3 3.4 4.1 4.2 4.3 4.4 4.5 4.6 4.7 4.8 4.9 An example data network. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20 A LANFielder Client in operation . . . . . . . . . . . . . . . . . . . . . . . . . 22 A LANFielder Server with several clients connected. . . . . . . . . . . . . . . 23 Format of LANFielder Test Packet . . . . . . . . . . . . . . . . . . . . . . . . 24 Logical layout of measurements and the software used. . . . . . . . . . . . . 33 Logical layout diagram of the RSSI measurement hardware congurations . 36 Physical Layout of Measurement Equipment in Case I . . . . . . . . . . . . . 38 A picture of the wireless LAN access point transmitter location for Case I . 38 A picture of the wireless LAN access point receiver location for Case I . . . 39 Physical Layout of Measurement Equipment in Case II . . . . . . . . . . . . 39 A picture of the wireless LAN access point transmitter location for Case II . 40 A picture of the wireless LAN access point receiver location for Case II . . . 41 Physical Layout of Measurement Equipment in Case III . . . . . . . . . . . . 41
4.10 A picture of the wireless LAN access point transmitter location for Case III . 42 4.11 A picture of the wireless LAN access point receiver location for Case III . . 43 5.1 5.2 5.3 5.4 All measurements for Scenario A, WaveLAN System . . . . . . . . . . . . . 51 All measurements averaged for all directions at each measurement location for Scenario A, WaveLAN System . . . . . . . . . . . . . . . . . . . . . . . . . 52 All measurements averaged over similar signal strength (i.e. RSSI) locations for Scenario A, WaveLAN System . . . . . . . . . . . . . . . . . . . . . 53 All measurements for Scenario A, 3Com System . . . . . . . . . . . . . . . . 54 xi
5.5 5.6 5.7 5.8 5.9
All measurements averaged for all directions at each measurement location for Scenario A, 3Com System . . . . . . . . . . . . . . . . . . . . . . . . . . . 55 All measurements averaged over similar signal strength (i.e. RSSI) locations for Scenario A, 3Com System . . . . . . . . . . . . . . . . . . . . . . . . 56 All measurements for Scenario B, WaveLAN System . . . . . . . . . . . . . . 57 All measurements averaged for all directions at each measurement location for Scenario B, WaveLAN System . . . . . . . . . . . . . . . . . . . . . . . . . 58 All measurements averaged over similar signal strength (i.e. RSSI) locations for Scenario B, WaveLAN System . . . . . . . . . . . . . . . . . . . . . . 59
5.10 All measurements for Scenario B, 3Com System . . . . . . . . . . . . . . . . 60 5.11 All measurements averaged for all directions at each measurement location for Scenario B, 3Com System . . . . . . . . . . . . . . . . . . . . . . . . . . . 61 5.12 All measurements averaged over similar signal strength (i.e. RSSI) locations for Scenario B, 3Com System . . . . . . . . . . . . . . . . . . . . . . . . 62 5.13 All measurements for Scenario C, WaveLAN System . . . . . . . . . . . . . . 63 5.14 All measurements averaged for all directions at each measurement location for Scenario C, WaveLAN System . . . . . . . . . . . . . . . . . . . . . . . . . 64 5.15 All measurements averaged over similar signal strength (i.e. RSSI) locations for Scenario C, WaveLAN System . . . . . . . . . . . . . . . . . . . . . 65 5.16 All measurements for Scenario C, 3Com System . . . . . . . . . . . . . . . . 66 5.17 All measurements averaged for all directions at each measurement location for Scenario C, 3Com System . . . . . . . . . . . . . . . . . . . . . . . . . . . 67 5.18 All measurements averaged over similar signal strength (i.e. RSSI) locations for Scenario C, 3Com System . . . . . . . . . . . . . . . . . . . . . . . . 68 5.19 All measurements for Scenario D, WaveLAN System . . . . . . . . . . . . . 69 5.20 All measurements averaged for all directions at each measurement location for Scenario D, WaveLAN System . . . . . . . . . . . . . . . . . . . . . . . . . 70 5.21 All measurements averaged over similar signal strength (i.e. RSSI) locations for Scenario D, WaveLAN System . . . . . . . . . . . . . . . . . . . . . 71 5.22 All measurements for Scenario D, 3Com System . . . . . . . . . . . . . . . . 72 5.23 All measurements averaged for all directions at each measurement location for Scenario D, 3Com System . . . . . . . . . . . . . . . . . . . . . . . . . . . 73 xii
5.24 All measurements averaged over similar signal strength (i.e. RSSI) locations for Scenario D, 3Com System . . . . . . . . . . . . . . . . . . . . . . . . 74 5.25 The entire measurement environment modeled in SitePlanner. . . . . . . . . 77 5.26 A diagram of the interference paths between the different wireless LAN users in Scenarios C and D. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80 5.27 Plots of the WaveLAN measured throughput versus the calculated net interference power averaged over all cardinal directions for each physical location . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83 5.28 Plots of the WaveLAN measured throughput versus the calculated SIR averaged over all cardinal directions for each physical location . . . . . . . . . 84 5.29 Plots of the 3Com measured throughput versus the calculated net interference power averaged over all cardinal directions for each physical location 85
5.30 Plots of the 3Com measured throughput versus the calculated SIR averaged over all cardinal directions for each physical location . . . . . . . . . . 86 5.31 Plots of the WaveLAN measured throughput versus the calculated net interference power averaged over all cardinal directions for each physical location. All data points for both clients are shown in the graph. . . . . . . . 87 5.32 Plots of the WaveLAN measured throughput versus the calculated SIR averaged over all cardinal directions for each physical location. All data points for both clients are shown in the graph. . . . . . . . . . . . . . . . . . 88 5.33 Plots of the 3Com measured throughput versus the calculated net interference power averaged over all cardinal directions for each physical location. All data points for both clients are shown in the graph. . . . . . . . . . . . . 89 5.34 Plots of the 3Com measured throughput versus the calculated SIR averaged over all cardinal directions for each physical location. All data points for both clients are shown in the graph. . . . . . . . . . . . . . . . . . . . . . 90 6.1 6.2 6.3 6.4 WaveLAN Scenario A, Location Averaged Data and Linear Throughput Prediction Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96 3Com Scenario A, Location Averaged Data and Linear Throughput Prediction Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97 WaveLAN Scenario A, Location Averaged Data and Linear Constrained Throughput Prediction Model . . . . . . . . . . . . . . . . . . . . . . . . . . . 98 3Com Scenario A, Location Averaged Data and Linear Constrained Throughput Prediction Model . . . . . . . . . . . . . . . . . . . . . . . . . . . 99 xiii
6.5 6.6 6.7 6.8 6.9
WaveLAN Scenario A, Location Averaged Data and Exponential Throughput Prediction Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101 3Com Scenario A, Location Averaged Data and Exponential Throughput Prediction Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102 WaveLAN Scenario A, Location Averaged Data and Exponential Constrained Throughput Prediction Model . . . . . . . . . . . . . . . . . . . . . . 103 3Com Scenario A, Location Averaged Data and Exponential Constrained Throughput Prediction Model . . . . . . . . . . . . . . . . . . . . . . . . . . . 104 WaveLAN Scenario B, 2 User Prediction Model. This gure shows the predicted versus actual throughput results and an ideal t line . . . . . . . . . . 109
6.10 3Com Scenario B, 2 User Prediction Model. This gure shows the predicted versus actual throughput results and an ideal t line . . . . . . . . . . . . . . 110 6.11 WaveLAN Scenario B, 2 User Constrained Prediction Model. This gure shows the predicted versus actual throughput results and an ideal t line . 111 6.12 3Com Scenario B, 2 User Constrained Prediction Model. This gure shows the predicted versus actual throughput results and an ideal t line . . . . . 112 6.13 WaveLAN Scenario C, One User with Interference, throughput prediction Model. This gure shows the predicted versus actual throughput results and an ideal t line . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116 6.14 3Com Scenario C, One User with Interference, throughput prediction model. This gure shows the predicted versus actual throughput results and an ideal t line . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117 6.15 Constrained WaveLAN Scenario C, One User with Interference, throughput prediction Model. This gure shows the predicted versus actual throughput results and an ideal t line . . . . . . . . . . . . . . . . . . . . . . 119 6.16 Constrained 3Com Scenario C, One User with Interference, throughput prediction model. This gure shows the predicted versus actual throughput results and an ideal t line . . . . . . . . . . . . . . . . . . . . . . . . . . 120 6.17 WaveLAN Scenario D, 2 User with Interference, throughput prediction Model. This gure shows the predicted versus actual throughput results and an ideal t line . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123 6.18 3Com Scenario D, 2 User with Interference, throughput prediction model. This gure shows the predicted versus actual throughput results and an ideal t line . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124 xiv
6.19 WaveLAN Scenario D, Constrained 2 User with Interference, throughput prediction Model. This gure shows the predicted versus actual throughput results and an ideal t line . . . . . . . . . . . . . . . . . . . . . . . . . . 125 6.20 3Com Scenario D, Constrained 2 User with Interference, throughput prediction model. This gure shows the predicted versus actual throughput results and an ideal t line . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126 7.1 7.2 7.3 7.4 7.5 7.6 7.7 7.8 Comparison of the t for Scenario A when throughput is predicted directly from a model of the building environment for the WaveLAN network. . . . 131 Comparison of the t for Scenario A when throughput is predicted directly from a model of the building environment for the 3Com network. . . . . . . 132 Comparison of the t for Scenario B when throughput is predicted directly from a model of the building environment for the WaveLAN network. . . . 133 Comparison of the t for Scenario B when throughput is predicted directly from a model of the building environment for the 3Com network. . . . . . . 134 Comparison of the t for Scenario C when throughput is predicted directly from a model of the building environment for the WaveLAN network. . . . 135 Comparison of the t for Scenario C when throughput is predicted directly from a model of the building environment for the 3Com network. . . . . . . 136 Comparison of the t for Scenario D when throughput is predicted directly from a model of the building environment for the WaveLAN network. . . . 137 Comparison of the t for Scenario D when throughput is predicted directly from a model of the building environment for the 3Com network. . . . . . . 138
A.1 A diagrammatic overview of the IEEE 802.11 standard as dened in [30] and [31]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146 A.2 Illustration to scale of IEEE 802.11 DSSS Channels. . . . . . . . . . . . . . . . 148 A.3 Illustration of the process of spreading data using a Barker Sequence. Adapted from [36]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149 A.4 Illustration of the basic IEEE 802.11 Spreading technique. Created using information from [30] and [31]. . . . . . . . . . . . . . . . . . . . . . . . . . . 150 A.5 Conceptual block diagram of the CCK encoding process carried out by the transmitter. Created using information from [31]. . . . . . . . . . . . . . . . . 153 A.6 A block diagram of the PBCC encoding technique. Created using information from [31]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154 xv
A.7 Basic structure of the IEEE 802.11b packet as its transmitted on the physical layer is shown here. Adapted from [31]. . . . . . . . . . . . . . . . . . . . . . 155 A.8 The structure of packet created at the MAC Layer. Adapted from [30]. . . . 156 A.9 Flow chart of the process of sending data packets under the CSMA/CA based Distributed Coordination Function (DCF). Created using information in [30]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158
xvi
List of Tables
2.1 2.2 Overview of Different RF Propagation Measurement Techniques Used in Wireless LAN Coverage Research . . . . . . . . . . . . . . . . . . . . . . . . . 6
Network Performance Measurement Techniques Used in Wireless LAN Research . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12 2.4 GHz Wireless LAN Equipment Used in Network Performance Measurements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32 Computers and Settings Used in Measurements. . . . . . . . . . . . . . . . . 35 Summary of All Congured Parameters Used in SitePlanner . . . . . . . . . 78 The Values of the Calculated Parameters and Error Statistics for the Scenario A Throughput Prediction Models, as dened in Equations 6.1 and 6.3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100 A Summary of the Accuracy of the Empirical Throughput Prediction Models.105 The Values of the Calculated Parameters and Error Statistics for the Scenario B Throughput Prediction Models, as given in Equation 6.4 . . . . . . . 108 The Calculated, Ideal Values and Error Statistics for the Scenario C, Throughput Prediction Model, given in Equations 6.8 and 6.9 for One User with WLAN Interference. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118 Table of Parameter Values Used for the Scenario D Prediction Model for Equation 6.10 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123 Comparison of Throughput Model Error Statistics for Starting from a Building Model or from RSSI Measurements . . . . . . . . . . . . . . . . . . 139
4.1 4.2 5.1 6.1
6.2 6.3 6.4
6.5
7.1
A.1 World Wide Spectrum Allocation for IEEE 802.11 and 802.11b use as dened in [30] and [29]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147 xvii
A.2 Mapping of dibits to Angles for CCK Modulation. Created using information from [31]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151 A.3 Mapping of Angles into Complex Chips for CCK Modulation. Created using information from [31]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152 A.4 Different Interframe Spacings (IFS) as dened in [30]. . . . . . . . . . . . . . 157
xviii
Chapter 1 Introduction
1.1 Wireless Data Networks
Wireless LANs (Local Area Networks) are typically installed and maintained by Information Technology (IT) staff members. These types of individuals do not generally possess detailed knowledge of wireless network performance or even a basic idea of the coverage area they can expect to achieve with a wireless LAN network. This is especially difcult because wireless networks can often experience an unexpected RF (radio frequency) environment when they are used indoors. For this reason this thesis has developed simple, accurate models of wireless LAN throughput based on indoor RF propagation. The use of these models can be used to easily predict the throughput experienced by a user at any location in the coverage area of a wireless LAN access point. This thesis also presents measurement campaign results of actual wireless LANs and the methods used to calculate the parameters of the empirical throughput prediction models.
1.2
Motivation for Throughput Prediction
There are several solutions for indoor wireless networking available today. A large number of these of products are based on the IEEE 802.11b wireless LAN standard [30]. These hardware devices are typically PC Card, PCI (Peripheral Component Interconnect), or USB (Universal Serial Bus) based based network interface cards (NICs) which can com1
CHAPTER 1. INTRODUCTION
municate with one another or through Access Points (APs) to a wired backbone network. Wireless LAN equipment based on the 802.11b standard operate between 2.4 and 2.497 GHz where they can be used in most countries worldwide. The APs are typically installed by network savvy IT staff who know little or nothing about wireless communications. Thus, the ability to simulate a wireless LAN design and predict throughput in a computer-aided design environment would be a valuable contribution to the wireless LAN industry. IT staff members are often interested in the network performance that will be experienced by a wireless LAN user. This performance varies with the type of wireless LAN card in use, the location of the wireless LAN user, as well as a number of other factors. For instance, the 802.11b products are capable of a shared, raw data bandwidth of 11 Mbps while the 802.11 products are capable of a shared, raw data bandwidth of 2 Mbps. However, this bandwidth is not usually realized because a single NIC is not capable of utilizing the full bandwidth. Further, the available data rate uctuates depending on the wireless channel characteristics. For this reason, it is important to be able to both accurately measure and reliably predict the actual throughput of IEEE 802.11 devices. This thesis shows that by combining measurements and prediction models, IT staff members can easily plan and deploy a wireless LAN network.
1.3
Thesis Overview
This thesis focuses on the difculties faced in designing and deploying a wireless LAN network through developing throughput prediction models to aid in designing a wireless LAN infrastructure. To create these models, the thesis rst presents the commercially available measurement products SiteSpy and LANFielder, which were created as part of the work for this thesis. These products are designed to measure wireless LAN throughput and other network performance criteria and record the results in a precise site-specic manner. This ability is key to accurate measurements of wireless LAN networks. These measurement tools are presented in Chapter 3. To verify the accurate workings of LANFielder and SiteSpy, and to collect data for the development of empirically based prediction models, an extensive measurement campaign was conducted using SiteSpy, LANFielder and some custom RSSI (Received Signal
1.3. THESIS OVERVIEW
Strength Intensity) measurement software, in conjunction with commercially available IEEE 802.11b wireless modems, in a real network setting. The details of the measurement campaign and the results of the campaign are presented in Chapters 4 and 5, respectively. Once the measurement campaigns were complete, models were developed using the measurements for two different test-bed IEEE 802.11b wireless LAN systems. The conceptual framework, development and accuracy of these models is presented in Chapter 6. The models are further validated in Chapter 7. The bulk of this thesis assumes some basic familiarity with the IEEE 802.11 and IEEE 802.11b standards. However, to aid the reader who is somewhat unfamiliar with the standards, a basic tutorial is presented in Appendix A. Further, a review of basic wireless networking terms and recent research is presented in Chapter 2. Together, all of the information presented in this thesis provides an important contribution to the design of IEEE 802.11b wireless LANs.
CHAPTER 1. INTRODUCTION
Chapter 2 Prior Research

2.1 Wireless LAN Propagation Research
2.1.1 RF Coverage Measurement Techniques

Several authors have performed research in the area of measuring and predicting the RF coverage area of indoor wireless LAN access points. Three of the more popular techniques described in the literature to measure the path loss of wireless LAN signals are the use of continuous wave (CW) transmitters with power meter receivers, Broadband Pulse Channel Sounding, and using the signal strength percentage reported by wireless LAN cards. Some of the research which uses each of these techniques is detailed in Table 2.1 and later in Section 2.1.2. The following sections detail the benets and shortcomings of each of these measurement techniques.
Continuous Wave (CW) Transmitter and Power Meter The continuous wave (CW) transmission technique has long been used to measure the basic path loss of a wireless channel. Many useful statistics, such as the average fade duration, level crossing rate and the probability density function of the received signal strength, can be quantied and used in the design of a wireless modem. This narrowband technique requires less measurement hardware and measurement campaigns can be carried out more quickly and easily because a minimum amount of equipment is needed. 5
CHAPTER 2. PRIOR RESEARCH
Table 2.1: Overview of Different RF Propagation Measurement Techniques Used in Wireless LAN Coverage Research
Measurement Technique CW Signal and Power Meter Prediction Technique Combination of direct path loss and partition losses with ray-tracing and refraction techniques Various exponential and partition loss based empirical models Exponential path loss exponent and partition losses along direct path None None Empirically t path loss to a quadratic function and used a neural network type analysis to model multipath effects Empirically t outdoor path loss measurements to Okumura model None None References [28]
CW Signal and Power Meter Pulse Transmission Network Analyzer Wireless LAN Card Signal Strength in dBm Wireless LAN Card Signal Strength Percentage Wireless LAN Card Signal Strength Percentage, converted to dBm Wireless LAN Card Signal Strength Percentage Wireless LAN Card Signal Strength Percentage, converted to dBm
[7] [11] [14] [23] [4]
[9] [5], [35] [17]
The main difculty with this technique is that it is a narrowband measurement technique. If the bandwidth of a wireless modem is wider than the coherence bandwidth of the wireless channel then the instantaneous continuous wave path loss measurements of the signal strength will not be valid, even if local area averaged. This is especially a problem for IEEE 802.11 wireless LAN cards which have an RF bandwidth of 22 MHz [30] but operate in RF environments with typical coherence bandwidths of between 5 and 46 MHz. [11].
Broadband Pulse Transmission Broadband Pulse Transmission overcomes many of the shortcomings of CW transmission at the expense of increased measurement cost and time. The technique uses a broadband antenna to transmit an impulse which is short in duration. By measuring the pulses and arrival times at receive locations, the exact channel impulse response can be measured. The bandwidth of the measured channel is determined by the duration of the original transmitted impulse. By measuring a bandwidth larger than the bandwidth of the wireless modem, the RF performance of a wireless modem can be exactly measured even if the channels coherence bandwidth is smaller than the bandwidth of the wireless mo-
2.1. WIRELESS LAN PROPAGATION RESEARCH
dem. This technique also allows the quantication of the same statistics as the CW measurement technique as well as the mean excess delay, delay spread, coherence time and coherence bandwidth of the measured channel. These statistics are extremely useful for designing wireless modems. The difculty is that full broadband pulse systems require a large amount of equipment and time to measure. It should also be noted that other more advanced channel sounding techniques are available with similar results, benets and limitations, including sliding correlator, network analyzers [14] and chirp channel sounders which provide broadband channel impulse response measurements. The use of this equipment has so far not been broadly reported in the literature.
Wireless LAN Card Reported Signal Strength
The third major wireless LAN channel measurement technique used in the literature and in practice is to use the reported signal strength percentage reported by software utilities which are provided with wireless LAN cards. In this technique a laptop or other computer is used to measure the received signal strength from another wireless LAN card or a wireless LAN access point. This technique is far simpler than the previously mentioned techniques and allows the comparison of different cards from different wireless LAN vendors. In addition, measurements can be taken quickly and easily, and the results are more easily applied to actual installations and coverage prediction. However, until recently, the signal strength reported by wireless LAN cards has just been a percentage value between 0 and 100. These numbers do not have any direct, known value or meaningful relationship to path loss. They allow relative comparisons of different locations, but are not practical for actual path loss measurement purposes. These magic numbers are hard to apply or compare to prediction models. As a result, research using this method is typically only applicable to a single type of wireless LAN card or must be calibrated to convert the values into actual units of power. To overcome these shortcomings, IEEE 802.11b vendors, such as WaveLAN and 3Com, have recently included software tools that report the actual dBm value of the received signal strength.
2.1.2 Wireless LAN Related RF Propagation Research

This section presents some of the progress that has been made in the research on wireless LAN propagation measurement, prediction and modeling. While each of these papers present important contributions to the eld, all are lacking a simple, accurate model of wireless LAN propagation based on a diverse array of measurements.
Indoor Propagation Measurements In [11] the authors have measured the channel impulse response characteristics at 2.4 GHz using a generated impulse in two different buildings. The authors used the measured impulse responses to nd the parameters of the Motley-Keenan/Seidel-Rappaport indoor propagation model ([25] and [18]). The authors used a path loss reference distance of 10 or 1.25 meters and calculated a path loss exponent of n=3.086 which resulted in a standard deviation of 5.84 dB. In addition, the authors measured and reported the mean excess delay, delay spread and coherence bandwidth for each of the measurement cases. Thus, the authors have provided a reliable prediction model for wireless LAN received signal strength based on measured data. However, the measured data was measured using pulsed transmission and laboratory equipment. It remains to be seen if the model extrapolates to RSSI measurements made directly with wireless LAN cards that use Direct Sequence Spread Spectrum (DSSS) techniques to spread the transmitted power over a broad bandwidth. In addition, the environments measured in [11] are limited in extent. Additional validation of the authors model is warranted.
Pre-IEEE 802.11 Wireless LAN Measurements The authors of [17] have published a survey of different measurements that demonstrate the performance of a wireless LAN test-bed. The modems used in the test-bed were an early precursor to IEEE 802.11 wireless LAN modems but operated at 900 MHz. It is this early paper many measurements that have been repeated for wireless LAN cards operating in the 2.4 GHz band for IEEE 802.11 networks are presented. For instance, the authors measured the throughput versus the number of simultaneous users, and the throughput versus signal strength. The authors have also performed a number of measurements which have not yet been performed for 802.11 hardware. For this reason, it is this early
2.2. WIRELESS LAN NETWORK PERFORMANCE RESEARCH
paper which has drawn a road map for measurements that can be made in future research.
2.2
Wireless LAN Network Performance Research
Several researchers in the wireless networking area have found that attempting to use traditional wired network protocols on wireless networks has resulted in unexpected difculties. The use of wireless packet data has opened large areas of new research areas. This section presents some of the recent work in wireless LANs and some of the terminology used.
2.2.1 Network Performance Statistics

The literature on wireless LAN network performance present measurements using various different statistics. The different network statistics used are often ill-dened. For this reason, this thesis denes the different statistics as they will be used throughout this document. Further, this thesis briey discusses the usefulness and usage of these statistics.
Delay: Latency and Round Trip Time One of the key statistics used in evaluating the performance of a network connection is the delay experienced by data which travels from one host to another. The term latency is used to describe this concept. However, care must be used in the use of the term because it is not always clear whether latency refers to the time to travel from one host to another, or the time required to transmit a packet and receive an acknowledgment, or some other delay. Unless otherwise stated, this document will always refer to latency as the time required to travel from one host to another. A related term to latency is the round trip time of a network connection. The RTT of a connection is the time required for a data packet to travel from one host to another and to return back to the original host. Latency and round trip times are typically measured in milliseconds for IP-based networks.
10 Throughput
Throughput is a measurement of the average rate that data (in bits) can be sent between a one user and another and is typically reported in kilobits per second or megabits per second, where kilo- is 103 and mega- is 106 (as opposed to 210 or 220 which are often used to dene data sizes as in kilobytes or megabytes). As with latency, care must be taken in the denition of this term. The throughput of the same network connection can vary greatly depending on the protocol used for transmission (e.g., UDP, TCP, etc.), the type of data trafc being sent (e.g., HTTP, FTP, VoIP or other trafc) as well as the quality and data bandwidth of a network connection. This is quite different from latency which generally does not vary for different protocols or trafc types. Throughput is measured at the highest protocol level possible to reect as accurately as possible the performance that will be experienced a user. Throughput is, thus, computed using the amount of data in the payload area of the highest protocol layer (e.g., the UDP payload size) of the transmitted packets. Overhead due to protocol headers and checksums are not included in the calculations of throughput as it is dened in this thesis.
Data Bandwidth The data bandwidth or channel data rate is the maximum available, raw rate at which data can be transmitted over a network connection. The data bandwidth of a connection is similar to the throughput of a connection except that the data bandwidth is the theoretical maximum rate at which data can be transmitted if all of the overhead and checksums of the protocols used is included and the multiple access protocol is completely efcient. Like throughput, data bandwidth is measured in units of bits per second or bps. However, the data bandwidth of a of a network connection is always larger than the measured throughput of the connection. For example, later in this thesis, wireless LAN connections with 11 Mbps data bandwidths have been measured to have throughputs of 2 Mbps.
Error Rates The error rate of a data connection, until recently, has not been a common metric of the network performance of a connection. Packets can be lost by routers over long trips across backbone networks due to collisions, but over short, LAN connections, typical wired or
11
ber optic transmission mediums have raw bit error rates on the order of 106 to as low as 1014 . With the addition of error checking in many packet transmission protocols, bit errors in wired and ber-based data transmissions are insignicant and, therefore, not often measured for local area networks. However, with the growing popularity of Voice over IP (VoIP), real-time video streaming and the use of wireless in data networks has meant that data packets are often dropped or excessively delayed with respect to the realtime requirements or shortcomings in wireless media. Thus, the error rate of a network connection has grown in importance. This has resulted in increased interest in bit error rates and packet error rates. The Bit Error Rate (BER) is the percentage of bits that are received in error or not received of those that are sent. The Packet Error Rate (PER) is the percentage of packets that are dropped or received incorrectly of those that are sent. Care must be exercised in the use the term BER because it can be used to imply the percentage of errors in only the payload data bits (e.g., excluding the header, footer and checksum bits) or to the raw data bits (e.g., including all overhead and non payload data bits). In this thesis the term BER refers to the error rate in just the payload data bits.
Delay Variation or Jitter Delay variation or jitter is an important metric for quantifying data network performance, especially for VoIP and video streaming applications in which the protocol relies on regular arrival rates of data packets. As a result, the delay of a packet sent from one host to another is extremely important, but so is the Delay Variation or Jitter. The delay variation or jitter of a packet is dened to be the average variation in the arrival time of a packet and is reported in milliseconds or other appropriate time scale [8].
2.2.2 Network Protocols

The overwhelming majority of network measurements are carried out using either Transmission Control Protocol (TCP) or User Datagram Protocol (UDP) packets. This is not surprising since most of the trafc on the Internet consists of mainly these two protocols. UDP, or User Datagram Protocol is a simple protocol. UDP does not guarantee whether or not packets will arrive, or if they do arrive they are not guaranteed to arrive in the order they were sent. In either case the sender will not receive an acknowledgment of
12
the transmission. These characteristics makes UDP ideal for video or voice streaming applications in which retransmissions and acknowledgments are a waste of bandwidth and the transmissions can make use of the low overhead associated with UDP. TCP, or Transmission Control Protocol guarantees that packets arrive and arrive in the order transmitted. TCP also attempts to avoid network congestion by sometimes delaying the transmission of packets. TCP however has a higher overhead than UDP. This overhead is not signicantly larger than for UDP though, so it is widely used in non-streaming applications.
2.2.3 Network Performance Measurement Techniques

There are many techniques and software tools available to measure the network performance of any data network which supports Internet Protocol (IP) transmissions, such as IEEE 802.11 wireless LANs. However, a relatively small number of methods have been applied to the actual measurement of wireless networks. This section outlines some of the popular techniques for wireless LAN measurement and the prior research that has gone into measuring the performance of wireless LAN connections. Additional information about how each of these techniques has been used in the literature is shown in Table 2.2 and Section 2.2.4. Table 2.2: Network Performance Measurement Techniques Used in Wireless LAN Research
Measurement Technique or Software A variant of ttcp, a UNIX, command line based program Test le transfer using FTP Test le transfer using FTP Chariot 3.1, Test Packet Based Software Harris WLAN Evaluation Software, Test Packet Based Software Prediction Technique None Reference [33] [27] [5] [10] [13]
None None Test Packet Based Software Predicted minimum adjacent channel interference power ratio for different channel settings
Due to the diverse array of techniques available for the measurement of network performance statistics, i.e., those parameters dened in Section 2.2.1), an overview of the popular techniques is presented here. The following network performance measurement techniques and software products are intended to quantify the network performance of the network connection between two hosts. In general, the measurement solutions will send some sort of test data from one host to another using a certain protocol and a cer-
13
tain test pattern. Measurements are repeatedly made of the test data and then averaged for better accuracy. Some of the measurement techniques and software programs used to achieve this process are outlined in this section.
UNIX Command Line Based Software There are several command line utilities that run under UNIX operating systems. These command line utilities are simple and intuitive to use (for individuals familiar with IP networks), and are often used to quickly test and diagnose problems with networks on a daily basis. Some examples of these programs include ping, and TTCP, but there are a great deal of others. Ping is a simple program that allows the user to send packets to a specic Internet host which acknowledges the packet. Ping allows the user to control the size of the data in a packet and to send repeated packets. However, the accuracy of Ping is low (on the order of 1 millisecond for the latency) and it is only capable of measuring the average latency for a packet to travel from one host to another, and the packet error rate. Ping has the advantage of not requiring a second software program to answer the test packets sent by the originating host since this is done by the operating system software. Alternatively, TTCP sends packets from one host to another and provides users with signicantly more control of the way packets are sent between two hosts. TTCP allows the user to transmit data using the TCP or the UDP protocol and measures the latency and throughput of a network connection.
File Transfers Using FTP A simple and inexpensive method for measuring the throughput of a connection is available using FTP software. In this technique, test les of a know size are transferred from one host to another host. The time required to transfer the le is then used to calculate the throughput. This technique, however, does not give any information about the latency of the connection. In addition, it is limited in that most of the data is sent in a single direction, and has a steady stream of maximally-sized packets sent continuously. Thus, the measured throughput using this technique tends to be optimistic because many network applications have vastly different network trafc characteristics, including packet size variations and fewer uni-directional transfers of data.
14
Test Packet Based Measurement Software and Chariot 3.1 Chariot 3.1 is a popular network measurement software tool. The from NetIQ Corporation, is representative of a class of network measurement software packages that precisely control the characteristics of test packets sent over a connection to measure a range of network performance statistics. Chariot 3.1 is capable of emulating a wide range of trafc types, such as Web trafc or Voice Over IP trafc, and a wide range of network protocols such as TCP, UDP and IPX [19]. For any of these protocols or trafc types, Chariot can measure the throughput, latency, jitter, and packet error rates [19]. The advantage of a software product such as Chariot is accurate control of the test trafc, combined with the ability to measure a wide range of network performance statistics. The difculty with these types of products is that they can often be too costly for research and, thus, are not necessarily practical to use as a standard for comparing network performance.
2.2.4 Network Performance Measurement and Prediction Research

This section presents some of the research on wireless LAN network performance measurement, prediction and modeling. None of these references present a simple means for predicting the throughput or other network performance statistic. However, the diverse array of research does show how measurements of the network performance of wireless data networks have been performed. Additionally, several papers indicate the plausibility of predicting throughput based on a small number of parameters such as the path loss at a receiver location.
BER and Throughput Correlation to Delay Spread Maeda, Takaya and Kuwabara published a measurement of wireless LAN performance and the validity of a ray-tracing technique to predict the performance of a wireless LAN [15]. The measurements were tracked in a small, highly-controlled radio frequency (RF) environment and indicated that the wireless LAN throughput and BER were correlated to the delay spread of the wireless channel. The researchers have not, however, presented any way to actually predict the bit error rate or throughput from the predicted delay spread prole output by a ray-tracing technique.
2.2. WIRELESS LAN NETWORK PERFORMANCE RESEARCH Early Wireless LAN Network Performance Measurements
15
Xylomenos and Polyzos explored the performance of UDP and TCP packets sent over several xed IEEE 802.11 wireless LAN network connections in [32] and [33]. The research focused on throughput limitations caused by software implementation issues and operating system shortcomings. The researchers used their own modied version of the command line utilities ttcp, tcpdump and nstat under Linux to perform UDP and TCP throughput tests. All measurements were taken between three xed locations and focused on varying the wireless LAN card types (PCMCIA or ISA) and the end-user computer hardware (i.e., Pentium 150 with 48 MB of RAM versus a Pentium 200 MMX with 64 MB of RAM). The researchers make recommendations for changes in the implementation of network protocols and Linux operating system enhancements. The measurements did not consider the effects of different physical locations, signal strength, or the effect of variations in the wireless communications channel on the network throughput.
IEEE 802.11 Throughput Measurements in a Hallway Duchamp and Reynolds presented packet throughput measurement results for varying distances for IEEE 802.11 wireless LANs [6]. These measurements were performed in a single hallway. Thus, these measurements, too, suffer from failing to measure a representative environment. The researchers do not present a model to predict their results nor do they attempt to validate any sort of computer prediction technique. Their work does not consider multiple users and was more focused on estimating the potential range of the wireless LAN in a nearly free space environment.
Idealized Wireless LAN Performance Measurements Bing presented measured results of the performance of an IEEE 802.11 Wireless LAN. In [2], Bing presents delay and throughput measurements as well as theoretically based throughput and delay estimations for various wireless LAN congurations. The results are given as optimal results and were measured on a lab bench rather than in an actual building environment. Therefore, the results presented are an upper bound on best possible results and do not extend into a site-specic wireless LAN performance prediction technique.
16
Causes of Throughput Variation in IEEE 802.11 Networks Demir, Komar, and Ersoy compared the effects of different system conguration factors on IEEE 802.11, 2-Mbps DSSS wireless LAN performance as measured by throughput [5]. The authors measured the throughput of an FTP-based le transfer and the signal strength percentage reported by the wireless LAN card hardware for 1, 2 and 3 simultaneous users. The authors used a 2k factorial design and sign table, which is a technique to determine what factors have the most signicant impact on the throughput experienced by a wireless LAN user. The authors considered the SNR, the number of simultaneous users, and the le size used in the data transfer. The authors concluded that the number of simultaneous users has the greatest effect on the throughput experienced by a single user, that the SNR level had a moderate effect on throughput, and that the le size had no effect on the throughput. The difculty with these results is that the authors have not used their data to present a model for predicting throughput; the SNR measurements are for an arbitrary, unknown vendor.
Wireless LAN Performance Issues Prasad, et. al., review many of the important factors in wireless LAN deployment, including a basic review of the IEEE 802.11 standard, standard indoor propagation models and interference and coexistence concerns with wireless LANs [22]. The authors also present some measurement results, although without any information about how the results were produced, in which an IEEE 802.11b wireless LAN users throughput is compared to the received signal strength. The results show that the user achieves a consistent throughput of about 4.8 Mbps until the received signal strength reaches about -85 dBm. From this point, the throughput in Mbps falls in an almost linear fashion relative to the dBm value of the received signal strength until it reaches zero at approximately -97 dBm. These results seem reasonable for a single user in a highly controlled environment.
Multiple-User Measurements of a Wireless LAN Chariot was used by Kamerman and Aben to measure the throughput for on a wireless LAN using TCP packet transfers for each of the four data bandwidths dened in the IEEE 802.11b standard [10]. The authors performed the measurements for 1, 3 and 5
2.3. SUMMARY OF PRIOR RESEARCH
17
simultaneous users and compared it to an analysis of the data bandwidth which is used by various overhead, collisions and unused bandwidth. The authors claim a maximum actual throughput of of 0.82, 1.52, 3.41 and 5.17 Mbps for the data bandwidths of 1, 2, 5.5 and 11 Mbps, respectively. It should be noted that these results are for ideal RF scenarios in which the data bandwidth has been forced to a desired setting.
Impact of Using Overlapping DSSS Channels in Close Proximity An exhaustive means of evaluating the interference between overlapping DSSS channels of IEEE 802.11 access points has been performed by Leskaroski and Mikael [13]. In this work, the authors used a simple network measurement tool called LANEval which was available from Harris/Intersil, the maker of several popular wireless LAN chipsets. The measurement tool sends test packets from a client to a server and quanties the throughput on the link between the two hosts. The authors have used this tool to measure the performance of a wireless LAN link operating on channel 1 in the presence of another interfering LAN operating on a varying channel between 1 and 6. The authors have used this data to calculate a minimum path loss between clients that allows the clients to operate without signicant impact on the throughput of one another. This is an extremely useful and practical result.
2.3
Summary of Prior Research
After extensive review of the current research into wireless LANs, it can be concluded that there is a need for simple, accurate throughput prediction models. While some basic information is available in the literature, no attempts have been made to predict wireless LAN throughput based on site-specic information. Further, the ability to predict the throughput of a wireless LAN in different locations can be critical to designing an efcient wireless network. For this reason, this thesis focuses on rst developing a means to measure realistic, non-optimistic, site-specic throughput, testing this measurement technique with an extensive measurement campaign, and creating throughput prediction models to allow wireless LAN design using throughput as a design parameter.
18
Chapter 3 Software Development for Throughput and RSSI Measurement

Several major software packages were used to perform the measurements that form this work. However, one of these packages was developed and released as a commercial product as part of the work for this thesis. The commercial wireless LAN measurement products LANFielder and SiteSpy were developed, marketed and sold at Wireless Valley Communications, Inc. The process of simultaneously developing a measurement product for research and commercial applications was a key contribution of this thesis. This chapter presents the key concepts and functioning of the LANFielder and SiteSpy products as well as some of the other custom software that was used in measurements conducted as part of this research. These measurements are presented further in Chapter 4.
3.1
Motivation for a Wireless LAN Measurement Tool
3.1.1 Difculties with Data Networks

Data networks have, since their inception, presented design and performance measurement issues. The key problem is that in order to design a well functioning distributed data network such as an IEEE 802.3, Ethernet network or an IEEE 802.11 or 802.11b wireless network, accurate information about the performance of the design is needed. To 19
20
CHAPTER 3. SOFTWARE DEVELOPMENT
Figure 3.1: An example data network. accurately measure the performance in a data network between different points or locations, data network measurement tools typically require two measurement agents. The reason for this is that by controlling the data ow between point A and point B, as in Figure 3.1, accurate information about the performance of the link between A and B can be known. If a single agent at A is used, it is impossible to know if the performance between point A and B or between point A and C or a combination of both is being measured due to the nature of distributed packet networks. If instead agents at A and B send test data back and forth over the single link between A and B, the performance of that link can be accurately determined.
3.1.2
The Client/Server Architecture
Data networks are often used as a way for many users to access a shared resource at the same time. For example, members of a company may access email from a single computer or people all over the world may access the same web page using the world wide web. The nature of this shared access has led to the development of the client/server architecture. This architecture, in its most basic form, allows many different clients (i.e. users at different network locations) to access a single shared resource on a server. The server
3.1. MOTIVATION FOR A WIRELESS LAN MEASUREMENT TOOL
21
allows clients to connect and access information on that server. Faced with the difculties of monitoring and quantifying the performance of data networks, the Client/Server architecture have often been used to accurately measure specic network connections. This architecture both mimics the way in which data networks are most often used and also allows the performance of specic network links to be identied. Referring to Figure 3.1, by installing a measurement server at network point A, and clients on points B and C, it is possible to reliably and accurately measure the performance of the links between point A and B as well as the network link between points A and C.
3.1.3 Wireless Data Network Challenges

Several techniques are available for the measurement of network performance as discussed in Section 2.2.3. The goal of this thesis is to predict a realistic throughput that reects the actual throughput that a user can expect to see during typical operation of a wireless LAN. For this reason, the optimistic, uni-directional throughput measured by a FTP measurement technique, while popular, was not selected. Although many different protocols can be easily implemented and certainly could be added at any time, this thesis focused on a test packet based solution using UDP, which could measure the throughput, packet error rate and latency of a wireless LAN connection, averaged over both directions, with was selected. It was found that the network measurement products which perform packet based network measurement are incapable of handling the site-specic nature of wireless LAN based data networks. To clarify, wireless LANs can have a different throughput at different locations. The throughput of the wireless LAN experience will vary as the user moves closer or further away from an access point. For this reason, a measurement tool is needed which can record the network performance of a wireless LAN at specic locations. LANFielder and SiteSpy were designed as add-on modules to SitePlanner, an indoor, site-specic wireless communications system design tool. Using LANFielder and SiteSpy, measurements of various network performance statistics in a site-specic manner could be taken. In addition, the software tools utilize a realistic measurement technique to ensure accurate measurement of wireless LAN networks.
22
3.2
LANFielder and SiteSpy Details
LANFielder is a plug-in software module to the Wireless Valley Communications, Inc. design and measurement tool, SitePlanner. SitePlanner 2000 was used with the LANFielder 1.1 plug-in module. SiteSpy is a stand-alone version of LANFielder that can operate without the use of SitePlanners precise site-specic modeling. Both products simply provide a different interface to the same network measurement core. LANFielder utilizes the full three-dimensional model provided by SitePlanner to record measurements sitespecically. SiteSpy utilizes a text based method of associating a location with a throughput, and is more useful as a stand-alone server or trafc generation tool. Patents were led for this product at Wireless Valley Communications. As will be seen in Chapter 4, this is exactly how LANFielder and SiteSpy were used for the measurements performed for this thesis.
Figure 3.2: A LANFielder Client in operation
3.2.1
Client/Server Architecture
LANFielder and SiteSpy use the Client/Server architecture to measure the performance of a wireless LAN link. When LANFielder or SiteSpy are started, the user is given the option of using the program as a client or a server. Measurements can be made using
3.2. LANFIELDER AND SITESPY DETAILS
23
a LANFielder client and server, a LANFielder client and a SiteSpy server or any other variation thereof. However, 1 client and 1 server are needed to perform measurements.
Figure 3.3: A LANFielder Server with several clients connected.
3.2.2
Site-Specic Data Recording in LANFielder and SiteSpy
SiteSpy does not support true three dimensional measurement recording. Rather, SiteSpy is typically used as a stand-alone server, and as a trafc generator to provide a source of interference. However, a SiteSpy client can save measured data by associating 4 textual tags with a measurement point. These 4 tags are intended to give an approximate location label to each recorded measurement point. The four tags give a oor number, receiver height, room number and general location to each measurement location. For convenience, these location tags can be loaded from a pre-dened le containing pre-set measurement locations or can be loaded on the y as measurements are performed. To save the data, a user simply clicks the Save Data button and the current settings and current measured statistics are saved to a logle along with a time stamp and the four text-based location tags. LANFielder works in much the same way as SiteSpy, except instead of assigning textual
24
tags to each measurement point, a precise X,Y,Z coordinate is used. This coordinate is derived from a mouse-click at a location in a three dimensional model of the environment and a predened receiver height or can automatically stored using a positioning system such as GPS. The current settings, measured statistics, time stamp and X,Y,Z coordinate are saved to a logle and also directly within the model of the environment.
3.2.3 Measured Network Performance Statistics

LANFielder and SiteSpy can both be used to measure the throughput, packet error rate and packet latency at each location. The denition of these terms are given in Section 2.2.1. To measure these statistics, each client is typically congured to transmit a test packet with the maximum amount of data possible. The format of a test packet is discussed further in Section 3.2.4.
3.2.4 Test Packet Format

The format of a test packet at the Ethernet layer is shown in Figure 3.4. This test packet is encapsulated by the MAC and then the PHY layer as is shown in Figures A.8 and A.7 respectively.
Figure 3.4: Format of LANFielder Test Packet
3.2. LANFIELDER AND SITESPY DETAILS
25
A single test packet is sent using UDP and consists of random data that is sent from the client to the server. The server echos an exact copy of the packet back to the client. This process is repeated for a timed measurement period. This period is called the averaging interval because after it is completed, the throughput, packet latency and packet error rates are calculated based on the average over all of the test packets transmitted.
3.2.5 Additional Congurable Parameters of SiteSpy and LANFielder

In addition to the packet size, SiteSpy and LANFielder allow the user to set the averaging interval and timeout period of a measurement period. The averaging interval is the period over which test packets are sent and the network performance statistics are calculated. Longer averaging intervals provide greater repeatability, but can result in time costly measurement campaigns. Since only one test packet is ever in transit for a single client/server pair, a timeout is needed to identify when a packet has been lost. The timeout interval set by the user is the maximum time after a packet is sent that the client will wait for a response from the server before sending a new packet.
3.2.6 Network Performance Calculations

The equations used to calculate the network performance statistics are shown in Equations 3.1, 3.2 and 3.3. Several important points need to be made about these calculations. First, note that throughput is based solely on the packets which are sent and received successfully. As shown in Equation 3.3, the latency is dened as the round trip time, from client to server and back, divided by two. The time required for this trip is measured by the client. Additionally, note that there is a direct relationship between throughput, packet error rate and latency. This relationship is given in Equation 3.4. This relationship brings up an important additional parameter. There is a timeout parameter that LANFielder uses to determine if a packet has been dropped or lost. During the averaging interval, if the client does not get a packet returned from the server for the duration of the timeout period, starting from when a test packet is sent, the packet is considered lost, and a new packet is transmitted. When a packet is dropped in this manner it is not counted in the payload bits sent successf ully parameter, or in the payload bits received successf ully parameter, but it is counted in the packets not received and total packets sen parameters.
26
T hroughput =
payload bits sent successf ully + payload bits received successf ully averaging interval duration (3.1) packets not received total packets sent
P acketErrorRate =
(3.2)
1 averaging interval duration P acketLatency = (mean round trip time) = 2 2 total packets sent (bits in packet payload) (1 P acket Error Rate) P acket Latency
(3.3)
T hroughput =
(3.4)
3.2.7
Rationale Behind Network Performance Measurement Technique
The measurement technique described above has its advantages and disadvantages for evaluating the latency, error rate and throughput of a connection. First of all, compared to FTP measurement methods and certain FTP emulating packet based measurement techniques, this measurement technique will often measure a smaller throughput but a more consistent packet latency. That is, the measurement technique used by LANFielder is a pessimistic measurement of throughput. There are, however, several good reasons for making measurements in this manner. 1. The FTP measurement technique, by nature, provides an optimistic measurement because it is continuously sending a large quantity of data in one direction. Actual network connections may not be so directional. Thus, the ability to submit large frames of data for transmission (i.e. multiple packets) at once is not typical of many types of network transmissions. LANFielder test trafc is split evenly between the up-link and down-link direction. In addition, only 1 packet is transmitted at a time. Thus, a more realistic, albeit pessimistic, throughput is measured. 2. LANFielder avoids the problem of the congestion avoidance algorithm found in TCP used on wireless LAN connections [34]. TCP makes the assumption that packet losses are caused by network congestion. Thus, when a packet is lost, TCP will de-
3.3. CUSTOM RSSI MEASUREMENT SYSTEM
27
crease the rate at which packets are sent. This can be a real problem in wireless LAN connections where packet losses can be caused by a signal fade or a bursty interfere in the wireless channel. Since fading and interference is a often much more transient than channel congestion, the TCP congestion avoidance algorithm can often be tricked into decreased transmission even when congestion is not present. This phenomenon can be measured using TCP packets as is done using the FTP and other techniques. The difculty with this technique, though, is that different TCP algorithms (e.g. TCP Tahoe, TCP Reno, etc) use different congestion avoidance methods (e.g. Fast Restart etc.) [21]. For this reason, LANFielder uses UDP packets. This means that no congestion avoidance will take place. However, the effect of different congestion avoidance methods can be evaluated analytically using statistical techniques and the measured packet error rate and packet latency statistics found using UDP transmissions. Due to the differences in the measurement technique from other published results (such as [33], [27], [5] [10] [13]), care should be used in comparing these measurements to other published results. Since throughput measurements are so sensitive to the measurement technique, it is often difcult to compare these measurements carefully. A possible solution to this is that measurements of throughput, made using different techniques, can rst be normalized to the maximum measured throughput. This has the potential to minimize the differences (i.e. normalize) caused by the particular throughput measurement technique.
3.3 Custom RSSI Measurement System

3.3.1 Motivation of RSSI Measurement Technique
The received power from DSSS signals such as IEEE 802.11b modems, is spread across a wide bandwidth and thus can be well below the noise oor. For this reason, measuring the RSSI (received signal strength intensity) using a spectrum analyzer can be quite difcult. In addition, a spectrum analyzer is unable to identify times when the wireless LAN is not transmitting. Thus, any measurement of power would need to be performed when the LAN is carrying as much trafc as possible, but would still slightly underestimate the
28
actual power received. For the same reason, a power meter would have difculty measuring the RSSI from a DSSS signal. Alternatively, measuring continuous wave signals neglects the effect of frequency selective aspects of the wireless channel. A full wideband channel sounder or network analyzer could have been used to measure the impulse response and thus the RSSI at various locations. However, a more practical and applicable method of RSSI measurement was desired.
3.3.2 Justication for RSSI Measurement Technique

The goal of this thesis is to allow installers of wireless LANs to easily predict the network performance of wireless LANs. If network performance models are to be based on measurements, then it is imperative that it be simple to measure the needed values. For this reason, this thesis has measured the RSSI of wireless LAN signals using a calibrated wireless LAN card which directly reports the RSSI of a wireless LAN signal in dBm. This value is the power which is spread across the 22 MHz bandwidth which is received by the wireless LAN card after despreading. Such data is sometimes provided by the software included with a wireless LAN card.
3.3.3 RSSI Measurement Software

The RSSI Measurement created for this thesis has a simple design. The software basically polls a wireless LAN card driver at a regular interval. The driver software, which is provided with every wireless LAN card, reports the signal and noise levels averaged since the last time the information was requested from the card. These averages are generally dB averages due to the logarithmic ampliers in the modem receiver. The signal and noise powers should, however, be averaged linearly, and not in dB values. Thus, software was written to poll the wireless LAN card with a high frequency (5 times per second) and to linearly average the data over a period of 5 seconds. Thus, each measured signal and noise power measurement is a linear average of 25 measurements. This is an important design decision. In order to more accurately measure the received signal strength, linear averaging of RSSI samples should be used. However, since the card drivers only perform dB averaging, the wireless LAN card driver should be polled as often as possible and the polled values can then be linearly averaged. However, there
3.4. SUMMARY OF MEASUREMENT SOFTWARE
29
is a limit to the rate at which the polling can take place, due to the nature of the operating system and the wireless LAN card drivers. In addition, requesting the received signal strength from the wireless LAN card driver will impede data transfers. For this reason, the RSSI was measured separately from the throughput.
3.3.4 RSSI Measurement Hardware

A wireless LAN, IEEE 802.11b wireless LAN card was used in to perform the RSSI measurements. The RSSI is measured by the wireless LAN card by measuring the strength of beacons transmitted by the wireless LAN access point. The noise is measured by measuring the received signal strength when no data transmission is detected. The signal strengths are averaged (in dB) repeatedly by the wireless LAN card until requested by the driver that interfaces the card to the operating system.
3.4
Summary of Measurement Software
In this Chapter, the details of software that was developed for Wireless LAN measurements are given. LANFielder and SiteSpy were two tools developed at Wireless Valley to allow the site-specic measurement of wireless LAN systems. In addition, RSSI measurement software was developed which interfaces to a CISCO Aironet API and allows the measurement of received signal strength by actual wireless LAN cards. The developed software allows the precise, location based measurement of throughput, packet error rates, packet latency and received signal strength and noise power for test-bed wireless LANs. The capabilities of this software was used in a measurement campaign detailed in the next chapter.
30
Chapter 4 Wireless LAN Measurement Campaign

4.1 Measurement Campaign Overview
In order to design a wireless LAN, it is often necessary to perform a measurement campaign to ensure the wireless LAN provides the needed coverage and capacity (i.e. throughput) for the intended users. With this need in mind, a full measurement campaign was performed in which multiple access points were setup in three different typical indoor environments. Measurements of the RF performance and network performance of the wireless LANs were measured for these test installations. A primary goal of the measurement campaign was to compare the performance of two IEEE 802.11b wireless LAN systems from different vendors under various different operating situations and to collect performance data to be used to develop empirical models of wireless LAN performance. The full details of the measurement campaign and some of the results are presented in this chapter. The measurement results were used to create empirical models of wireless LAN network performance, which is presented in Chapter 6.
4.1.1 Measurement Test-Bed

Two different IEEE 802.11b wireless LAN systems were compared using network performance measurements. The two systems used were made by WaveLAN (Lu31
32
CHAPTER 4. WIRELESS LAN MEASUREMENT CAMPAIGN
cent/ORiNOCO) and 3Com (AirConnect). Some of the specics of these two systems are presented in Table 4.1. These two systems were congured as an infrastructure network in which a single access point was placed in a certain location. For some of the measurements, a BayNetworks access point and a laptop with a BayNetworks DSSS IEEE 802.11, 2 Mbps wireless LAN card was setup in a xed location. For these measurements the BayNetworks equipment acted as interference sources. Measurements were then made in particular locations using laptop computers with wireless LAN cards corresponding to the vendor of the access point. The laptops are referred to as clients throughout this thesis. These two test-bed systems were setup at different times but all access points and clients were placed in the same exact locations so that the performance of the two systems could be compared in the identical environments. Table 4.1: 2.4 GHz Wireless LAN Equipment Used in Network Performance Measurements
System Type Maximum Throughput Access Point Type Access Point Transmit Power Access Point Antenna Gain Client Card Type Lucent (ORiNOCO) WaveLAN System IEEE 802.11b DSSS 11 Mbps AP-500 Access Point 3COM AirConnect System IEEE 802.11b DSSS 11 Mbps 11 Mbps Wireless LAN Access Point 18 dBm 2 dBi 11 Mbps Wireless LAN PC Card (PCMCIA) 15 dBm 2 dBi Bay Networks IEEE 802.11 DSSS 2 Mbps BayStack 660 Access Point 15 dBm 2 dBi BayStack 660 PC Card (PCMCIA) 15 dBm 2 dBi
15 dBm 2 dBi ORiNOCO Gold PC Card (PCMCIA) 15 dBm 2 dBi
Client Card Transmit Power Client Card Antenna Gain
A diagram of all of the equipment used in the test-bed setup is shown in Figure 4.1. This gure will be used throughout this chapter to explain how measurements were performed. Note that all of the equipment was not used for all measurements. The laptop clients used to make all of the actual measurements are shown in the gure and are named Laptop1 and Laptop2. All other equipment is present as part of the test-bed to allow measurements to be taken in different usage scenarios in which 1 or 2 users and with or without interference setups were considered. LANFielder and SiteSpy, the software developed for this thesis as described in Chapter 3, were used to measure the network performance and RSSI measurement of a wireless LAN user at different specic locations. Using the RSSI measurements, the path loss from
4.1. MEASUREMENT CAMPAIGN OVERVIEW
33
Figure 4.1: Logical layout of measurements and the software used.
the access point to the measurement user or client was known. The software was run on the client laptops. The software used on each individual computer is shown in Figure 4.1. Since LANFielder and SiteSpy use a client/server technique for measurements, SiteSpy was also run on desktop computers which were used as servers. These servers computers are also shown in Figure 4.1. It was necessary to connect the server computers to Ethernet hubs or switches as is shown in the gure. Ethernet hubs and switches were needed because the access points do not support being plugged directly into computers. For some of the measurement scenarios, an interfering IEEE 802.11 wireless LAN system was used to evaluate how the test-bed IEEE 802.11b system performed in the presence of interference. The interference system was a BayStack 660 manufactured by Bay Networks (since acquired by Nortel Networks). The equipment is no longer available for sale. How-
34
ever, the relevant specications for this equipment is listed in Table 4.1. In addition, this equipment functioned well as an interfering source and was setup using SiteSpy to generate data trafc that would interfere with the test-bed system. The interfering IEEE 802.11 equipment is shown in Figure 4.1. Throughout this thesis, the inferring equipment is referred to as the interfering access point and interfering client. The test-bed system is referred to as the desired access point and desired client. This terminology is used to distinguish the equipment.
4.1.2 Additional Equipment

Several computers, wireless LAN cards, access points and Ethernet hubs were used to perform network performance measurements. The conguration of the equipment is shown in Figure 4.1. The settings and specics of the computer equipment used is shown in Table 4.2. All of the computers used were chosen to a) be as nearly identical as possible and b) reect typical equipment for a wireless LAN-based network. With these two goals in mind, all of the computers used in the throughput measurements had the same operating system installed and had similar hardware congurations. To reect typical usage scenarios, all mobile clients were laptops with PCMCIA wireless LAN cards in use. The xed server computers were desktop computers with PCI 10/100 Ethernet cards installed. Additionally, the Ethernet hub (10 Mbit) and Ethernet switch (100 Mbit switched) are capable of supporting signicantly more trafc than the wireless LAN connections. Thus, the measured throughput and latency effects caused by these back end network devices should be minimal.
4.1.3 RSSI Measurement Congurations

The path loss between the desired access points and each client location was measured using the custom software from Wireless Valley, described in Section 3.3. This software does not require a server computer to operate. Rather, the RSSI measurements can be made using a single laptop and an access point. The layout used for the measurements is shown in Figure 4.2. The RSSI measurements were made at all client locations with the laptop oriented in each of the four cardinal directions (i.e. North, East, South . For each location and direction, four different measurements were performed. The rst two
4.1. MEASUREMENT CAMPAIGN OVERVIEW Table 4.2: Computers and Settings Used in Measurements.
Computer Name Laptop1 Operating System Windows 2000 Processor Celeron MHz 400 RAM 192 Megabytes 128 Megabytes 128 Megabytes 256 Megabytes Measurement Software SitePlanner with LANFielder Client SitePlanner with LANFielder Client SiteSpy Client
35
Laptop2
Windows 2000
Pentium II 266 MHz Pentium 700 MHz Pentium 850 MHz III
Laptop3
Windows 2000
Laptop4
Windows ME
III
RSSI Measurement Software
Desktop1 Desktop2
Windows 2000 Windows 2000
Pentium Pro 200 MHz Pentium Pro 200 MHz
128 Megabytes 128 Megabytes
SiteSpy Server SiteSpy Server
Network Card(s) Used 3Com, WaveLAN IEEE 802.11b cards 3Com, WaveLAN IEEE 802.11b cards BayNetworks IEEE 802.11 cards CISCO Aironet 340 IEEE 802.11b PCMCIA card 10/100 PCI Ethernet Card 10/100 PCI Ethernet Card
were of the received signal strength from the two different desired access points at the desired client locations. The third measurement was the received signal strength of the interfering access point at the desired client locations. Lastly, the fourth measurement was from the interfering access point, placed at the interfering client location and orientation, to the desired client locations. All four measurements were taken individually for each direction at each measurement point for a duration of 5 seconds. It was necessary to recongure the measurement client for each of these four measurements to talk to the necessary access point. The measurements of the signal strength from the desired access point was performed for both access point manufacturers, 3Com and WaveLAN. For these measurements the RSSI measurement client used Wireless LAN DSSS Channel 1, which has a center frequency of 2412 MHz [30]. The measurements of the interference signal strength were performed at the frequency of the interfering wireless LAN system. That is, the measurement client measured the received signal strength from the interference system while both the interfering access point and the measurement client were using DSSS Channel 3. The wireless LAN measurement card could not measure the signal strength from Channel 3 on Channel 1. To make these RSSI measurements, the software development laptop which had the custom software, Laptop4, was used with a commercial IEEE 802.11b wireless LAN card.
36
Figure 4.2: Logical layout diagram of the RSSI measurement hardware congurations The specications of the laptop used is given in Table 4.2. The wireless LAN card used was selected because it was capable of reporting the received signal strength from an access point more often than either of the 3Com or WaveLAN cards. In addition, the RSSI measurement card had a 2 dBi gain as do the other two client cards. It was assumed that the difference in the pickup pattern of the card was negligibly different from the patterns of the WaveLAN or 3Com cards. The measurements were averaged at each location and entered manually into SitePlanner for creating propagation models of the environment.
4.2
Organization of Measurements
The measurements were divided into 3 cases and 4 scenarios. Cases were the general environments used for measurements, Scenarios were how the equipment was congured. For each of the 12 combinations of cases and scenarios, 2 different IEEE 802.11b wireless LAN systems were measured. These two different wireless LAN systems were the Lucent/ORiNOCO WaveLAN and the 3Com AirConnect system. The conguration and specications of these products was discussed in Section 4.1.1 and shown in Table 4.1. This section now discusses the different cases and scenarios that were used to evaluate
4.2. ORGANIZATION OF MEASUREMENTS these two wireless LAN systems.
37
4.2.1 Measurement Cases

The three measurement cases reect different indoor environments that are likely to be used for wireless LAN networks. All three of these environments were selected to reect a range of typical installation locations. The Cases are identied as I, II and III and are, respectively, an Open/Large Ofce Environment, a Long Hallway Environment, and a Closed/Small Ofce Environment. Measurements were taken on a single oor at various locations throughout these environments with the wireless LAN access point located in a central location. Specic details of each of theses cases are given in the following sections. Figures 4.3, 4.6, and 4.9 show an overhead view of the blueprint of these different cases. Different colors are used to represent the different partition types. The numbered locations represent the locations where measurements were taken.
Case I - Large/Open Ofce Environment Case I was selected to reect the performance that could be expected in a large, mostly unobstructed ofce environment consisting mainly of movable cubicle type partitions, drywall, and glass doors. A layout of this area is shown in Figure 4.3. This gure is a screen capture from SitePlanner. The numbered dots represent the measurement locations while the Larger circles are the omnidirectional patterns of the antennas used in modeling the access points. The interfering client location is also shown with a *. As can be seen from the blueprint, the main room in this area measured 9.4 meters wide, 15.7 meters long, and 2.7 meters high. The walls are made of primarily drywall and are shown in blue in Figure 4.3. There was a drop ceiling made of particle board ceiling tiles and the cubicle partitions (shown in gray in Figure 4.3) did not extend to the ceiling but were rather 1.7 meters high. The oor was carpeted in the large center room and tiled in all other areas shown. Two elevators of different sizes can be seen in the picture near measurement point 2 and are represented in orange. The furniture in this case consisted of mainly primarily wooden desks which were built-in to the cubicle partitions, computers, rolling chairs, as well as some shelving, lab benches and a refrigerator. A picture of the desired Access Point location is shown in Figure 4.4 and a picture of a typical client receive
38
Figure 4.3: Physical Layout of Measurement Equipment in Case I
location is shown in 4.5.
Figure 4.4: A picture of the wireless LAN access point transmitter location for Case I
As can be seen in these pictures, stands and tripods were used to support the laptops and wireless LAN access points. These allowed the measurements to be taken without a user near the measurement platform.
4.2. ORGANIZATION OF MEASUREMENTS
39
Figure 4.5: A picture of the wireless LAN access point receiver location for Case I Case II - Hallway Environment Case II was selected to consider the performance of a wireless LAN in a hallway where waveguide effects could be have an effect on transmission. A blueprint layout of Case II is shown in Figure 4.6.
Figure 4.6: Physical Layout of Measurement Equipment in Case II The walls of the hallway were painted cinder-block (shown in blue in 4.6). A drop ceiling at 2.7 meters high was again present. The hallway is approximately 27.5 meters long and 2.1 meters wide and has a tile oor. Note the orange squares shown in the center of Figure 4.6 near the Access Point. These squares represent the elevators in this area. Note that these are the same elevators in Case II. The access point and measurement clients were setup in the same manner as in Case I as is illustrated in Figures 4.7 and 4.8.
40
Figure 4.7: A picture of the wireless LAN access point transmitter location for Case II
Case III - Closed/Small Ofce Environment
Case III is intended to reect a more closed ofce environment consisting of mainly closed ofces rather than cubicles. The area is shown in blueprint form in Figure 4.9. The rooms shown in the blueprint are almost entirely made up of ofces. The smaller rooms typically measure 3.75 meters long and 3 meters wide. The larger rooms typically measure 4.6 meters long and 3 meters wide. Other larger rooms are also present in the area. The area has a carpeted oor and a particle board ceiling tile drop ceiling. Doors are made of wood and glass (shown in pink in Figure 4.9), and walls are primarily drywall (shown in blue in Figure 4.9). Ofce furniture consisted of wooden desks an chairs and some bookshelves and computer and photocopying equipment. The measurement clients and access points were setup in the same manner as was done in the previous 2 cases and is illustrated in Figures 4.10 and 4.11.
41
Figure 4.8: A picture of the wireless LAN access point receiver location for Case II
Figure 4.9: Physical Layout of Measurement Equipment in Case III
4.2.2
Measurement Scenarios
The four measurement scenarios were chosen to reect the effects of different usage in which 1 or 2 users and 1 interfering client were combined in different ways. The four scenarios are identied as A, B, C and D. Scenarios A, B, C and D consist of 1 client, 2 clients, 1 client with interference and 2 clients with interference respectively. That is, referring to Figure 4.1, Scenario A used Desktop1, Laptop1, the Ethernet Switch and the IEEE 802.11b Access Point. Scenario B used all of the equipment of Scenario A and Laptop2. Scenario C used all of the equipment of Scenario A and added Desktop2, Laptop3, the Ethernet Hub and the IEEE 802.11 Access Point. Scenario D used all of the equipment shown. The actual locations of all of this equipment for each case is shown in Figures 4.3, 4.6, and 4.9.
42
Figure 4.10: A picture of the wireless LAN access point transmitter location for Case III
Scenario A One User
In this scenario, a single laptop was setup to transmit test packets over a wireless link to an access point, over a wired link, through an Ethernet switch and to a desktop computer. The single laptop was then moved to various different measurement locations while the access point remained xed for each case. Between 5 and 7 points were measured for each case. These points are identied by numbers in Figures 4.3, 4.6 and 4.9. For each measurement point point, four measurements were made in each of the four ordinal directions (North, East, South, West). That is the Ethernet cards antenna was positioned to face in each of the directions North, East, South and West in that order. This process was performed for both access points and the wireless LAN cards of the same manufacturer using Laptop1, the Ethernet Switch and Desktop1. At each location the throughput was measured using LANFielder as a client and SiteSpy as a server as shown in Figure 4.1.
43
Figure 4.11: A picture of the wireless LAN access point receiver location for Case III Scenario B - Two Users The two client scenario was performed much like Scenario A, except two laptop computers were moved to different locations. Both laptops transmitted test packets over the same access point to the same desktop computer. Again, the access point remained in a xed location for each case. However, for the 2 client scenarios, all different combinations of the different measurement points were considered. That is, each possible pair of measurement points was measured. Measurements were made with clients at points 1 and 2, 1 and 3..., 2 and 3, 2 and 4,..., etc. However, measurements were not made for all possible permutations of locations. That is, interchanged measurement locations were not additionally made at locations 2 and 1, or 5 and 3, for example. For each combination of measurement locations, the same direction was measured simultaneously. That is, the measurement clients were placed in two measurement locations with both wireless LAN cards facing in the same direction (e.g. North). Both clients then were used to measure their respective throughputs at each location simultaneously. The clients were then turned to the next direction and another measurement was taken. Thus, four measurements were taken at each pair of measurement points. Different combinations of directions were not measured. For these measurements, Laptop1, Laptop2, Desktop1 and the Ethernet switch were used, as shown in Figure 4.1. The laptops were running LANFielder while the desktop was running SiteSpy. The chat facility of LANFielder was used to ensure measurements were
44 taken in synchronization.
Scenario C One User with Interference Scenario C was performed in a similar manner to Scenario A except that another IEEE 802.11 system was operating in a static setup while measurements were performed. That is, a new laptop, Laptop3, was setup with an IEEE 802.11 (not an IEEE 802.11b) card to transmit to another computer, Desktop2. Both computers had SiteSpy installed and simply streamed the maximum amount of data possible to act as interference. The interfering client and access point remained in a xed location while the same measurements that were made in Scenario A were repeated. Referring to Figure 4.1, Laptop3 and Desktop2 exchanged test packets to create interference. Laptop3 sent test packets using SiteSpy over a BayNetworks, IEEE 802.11 card to a BayNetworks Access Point, from there to a 10 Mbps Ethernet hub to a Desktop computer. While this occurred, Laptop1 was moved to various different locations and transmitted test packets to Desktop1 over the original IEEE 802.11b wireless LAN connections. Measurements were taken in the exact same locations and manner as was done for Scenario A. The BayNetworks access point was congured using a different, incompatible Service Set Identier (SSID) and a different, but overlapping, channel as the IEEE 802.11b wireless LAN access points. This ensured that the interfering network cards never connected to the desired access points or vice versa. For these measurements the desired access point was set to channel 1, as it was for all cases, while the interfering access point was set to channel 3. The center frequencies of these channels are 2412 and 2422 MHz, respectively. The 22 MHz RF bandwidth of the channels [30] means the two channels overlapped by 12 MHz. These channels were chosen because it has been found that at this overlap some effect on throughput occurs, but not usually a catastrophic interference [13].
Scenario D - Two Users with Interference Scenario D was carried out in the same manner as scenario C, except that two clients were used, as was done in scenario B. That is, the two client measurements made in scenario B were repeated with the interfering wireless LAN system in use at the same static location as was used in scenario B. As a result, all of the equipment shown in Figure 4.1 was
4.3. MEASUREMENT EQUIPMENT
45
utilized for this scenario. Measurements were made using LANFielder on Laptop1 and Laptop2 which streamed data to Desktop1. These measurements were made with the interference system setup which consists of Laptop3, Desktop2 and the interfering Access Point.
4.3
Measurement Equipment
4.3.1 Physical Locations

For all cases the desired access points and the interfering access points were placed on metal tripods with wooden platforms. The tripods were extended so the base of the access point was 2.5 meters above the oor. The access points were oriented as they would be installed in a typical installation. The measurement client laptops were placed on wooden platforms on top of PVC and plastic stands. These stands were rotated in place to measure laptop throughput with North, East, South and West orientations. Measurements were recorded using LANFielder by clicking at the location in the model of the environment. Additionally, the software automatically saved all data to the Desktop1 computer as a backup.
4.3.2 Equipment Conguration

For all measurements the desired access points were congured to use channel 1. For the interference scenario measurements (scenarios C and D), the interfering access points were congured to use channel 3. The rational for this selection is given in Section 4.2.2. The access points were congured to only use the Distributed Coordinated Function (DCF) for all transmissions. That is, no polling or medium reservation using RTS/CTS packets was allowed. This was done by setting packet fragmentation thresholds at the maximum levels. All power-save features of the wireless LAN cards were fully disabled and no WEP encryption was used.
46
4.3.3 Software Conguration

The different software packages used in the measurement campaign are shown in Figures 4.1 and 4.2, and detailed in Table 4.2. All throughput measurements were made in LANFielder using a 15 second averaging interval. That is, when a measurement was performed, a full 15 second period was used to measure the network performance during which the measurement researcher(s) were motionless and distant from the client laptop. The value of the timeout duration used by LANFielder was set to 400 milliseconds. These values were selected through experimentation with various settings. It was found that these values provided a consistent, repeatable measurement at each location without causing overly excessive measurement times. The packet size was 1472 bytes (i.e. the maximum allowable payload). Identical settings were used for all SiteSpy clients. RSSI measurements were made using the custom software which recorded the signal strength reported by the wireless LAN card 5 times per second for a total of 5 seconds. This resulted in 25 samples of the received signal strength which were linearly averaged to yield a single RSSI measurement.
4.4
Measurement Precautions and Verication
A large number of precautions were taken to ensure the viability and repeatability of the measurements. Some of these precautions are listed here: All measurements were taken at night or during periods when little or no human presence would affect the measurements. The measurement software had a timer that allowed the researchers performing the measurements to step back several feet from the laptops, well out of the neareld of the antennas, and remain motionless while the measurements were performed. Also, the platforms used to support the wireless LANs were made of wood to minimize the neareld effects on radiation from the wireless LAN antennas. Care was also taken to ensure that measurements were not made while the elevator was on the same oor nor was it in motion when measurements were being taken.
4.5. MEASUREMENT CONCLUSIONS
47
Measurements were never made when any of the microwave ovens, photocopiers or other potential 2.4 GHz noise sources on the oor were in operation. The interfering access points were congured to use a different SSID than the desired access points. This prevented desired wireless LAN cards from connecting to the interfering access point and vice versa. Reciprocity conrmation measurements were made in several random locations to ensure that the path loss from the transmitter to the receiver was the same as the path loss from the receiver to the transmitter. All measurements were found to be within 3 dB or better of the expected values. This is well within the tolerances of the CISCO wireless LAN card measurement capabilities. Several random measurements were made at non-measurement point locations and compared to predicted values from SitePlanner. All measurements were found to be within 5 dB of the predicted value. RSSI and throughput measurements were made separately, but in the same locations. This allowed a reliably accurate measurement of the RSSI and prevented the polling of the wireless LAN card from lowering the measured throughput. All wired Ethernet connections were signicantly faster than the wireless links that they were attached to. Thus, for the IEEE 802.11b networks, all wired Ethernet links operated at 100 Mbps which is signicantly faster than the 11 Mbps wireless LAN link. The IEEE 802.11 links were used with 10 Mbps Ethernet links. This careful pairing of wired and wireless Ethernet data bandwidths was selected to ensure that the wired links were not the constraining factor of the measurement links.
4.5
Measurement Conclusions
This chapter has outlined the measurement campaign that was designed to measure a wide variety of IEEE 802.11b usage situations. Two IEEE 802.11b systems were measured in three different environments, and with 1 or 2 users in operation with or without the presence of an interfering DSSS IEEE 802.11 system operating. Measurements were made in 18 different points and in each of four directions at every point. This wide range of
48
measurement scenarios meant that 288 RSSI measurements and over 1600 throughput measurements were taken.
Chapter 5 Wireless LAN Measurement Results

5.1 Format Used to Present Throughput Data
This Chapter presents the measurement results that were obtained from the extensive measurement campaign described in Chapter 4. To present this data in an orderly format, MATLAB plots are used. The MATLAB plots presented throughout this chapter are arranged with the measured signal to noise ratio (SNR) for the desired link from the client to the access p on the x axis and the measured throughput on the y axis. These plots are created by pairing the throughput and SNR measured at each point. By combining the SNR, throughput pairs for all three cases, a wide range of throughputs and SNR levels considered.
5.1.1 Meaning of SNR

Note that the same SNR measurements were used for all scenarios. That is, the SNR between the access point and the client location was measured with 1 client operating. This value was seen to change slightly when two clients were operating, but not signicantly. The SNR will most certainly change when interference is present, but since it is difcult to predict this value the SNR was not measured in these circumstances, instead another parameter, the SIR, was measured. The SIR value is intended to be a measurement of the net interference power for a measurement location. Additional plots of the throughput 49
50
CHAPTER 5. WIRELESS LAN MEASUREMENT RESULTS
versus the SIR parameter as well as the computation of SIR itself are discussed further in Section 5.7.2. All plots presented in this section use the same SNR values and are labeled theoretical for scenarios B, C and D because the SNR used was actually only measured in Scenario A and extrapolated for use in Scenarios B, C and D.
5.1.2 Overview of the Data Measured

Each scenario has two sets of gures, one for each wireless LAN system (WaveLAN and 3Com). Each set has three gures. These three gures present the measured data for from all 3 Cases, but just a single Scenario in unaveraged, and in two different averaged formats. The rst of the three gures that make up a set is a plot in which each data point represents a different data point and cardinal direction. The second gure is a plot of the average of all four cardinal directions (i.e. North, East, South, West) forming a single, spatially averaged data point. In the second gure there is a single data point for each measurement point, rather than four data points for each measurement point as is the case for the rst gure. The third gure is a smoothed version of the rst gure. Each data point in the third gure represents the SNR of a single data point and a windowed average of the data points throughput and the throughput of the 4 data points closest in signal strength (i.e. RSSI) to the main data point and the data point itself. Stated another way, each of the points in the third plot of the set represents a (5 point) window average over data points with similar signal strengths. The actual number of 5 data points used in the window average to calculate the throughput was selected by trial and error to provide a degree of smoothing to the data while minimizing the smearing caused by selecting too large a number of data points.
5.2. SCENARIO A DATA - ONE CLIENT
51
5.2
Scenario A Data - One Client
5.2.1 WaveLAN, Scenario A, All Data

Figure 5.1 shows the single user, WaveLAN data. The data shown is for each cardinal direction. The general trend of a maximum throughput until the SNR reaches a critical level is clearly evident. Below this critical level the throughput becomes less stable and trends downward.
2000
throughput, kbps
1500
1000
500
0 0
10
20
30 40 SNR, dB
50
60
70
Figure 5.1: All measurements for Scenario A, WaveLAN System
52
5.2.2 WaveLAN, Scenario A, Spatially Averaged Data

In Figure 5.2 the positional averaging has smoothed out the throughput below the critical SNR level. This averaging has resulted in exposing the general trend of a dropping off throughput as the SNR decreases below the critical level.
locational average throughput, kbps
2000
1500
1000
500
0 0
10
20
30 40 SNR, dB
50
60
70
Figure 5.2: All measurements averaged for all directions at each measurement location for Scenario A, WaveLAN System
53
5.2.3 WaveLAN, Scenario A, Window Averaged Data

Figure 5.3, like Figure 5.2, also shows the general linear trend of the performance below the critical SNR level. The window averaging, however, has resulted in a clearer presentation of this trend.
windowed average throughput, kbps
2000
1500
1000
500
0 0
10
20
30 40 SNR, dB
50
60
70
Figure 5.3: All measurements averaged over similar signal strength (i.e. RSSI) locations for Scenario A, WaveLAN System
54
5.2.4 3Com, Scenario A, All Data

Figure 5.4 shows the single user, 3Com data. The data shown is for each cardinal direction. The general trend of a maximum throughput until the SNR reaches a critical level is evident. Below this critical level the throughput becomes less stable and trends downward.
2000
throughput, kbps
1500
1000
500
0 0
10
20
30 40 SNR, dB
50
60
70
Figure 5.4: All measurements for Scenario A, 3Com System
55
5.2.5 3Com, Scenario A, Spatially Averaged Data

Figure 5.5 presents a clearer view of the measured data than Figure 5.4. The trend of a steady drop-off in throughput as the SNR drops below a critical point is visible, but the maximal throughput is not always reached when the SNR is above the critical SNR value.
2000
1500
1000
500
0 0
10
20
30 40 SNR, dB
50
60
70
Figure 5.5: All measurements averaged for all directions at each measurement location for Scenario A, 3Com System
56
5.2.6 3Com, Scenario A, Window Averaged Data

The window averaging used in Figure 5.6 very clearly shows the validity of a exponential model of throughput. The 3Com data has a slight drop-off in throughput above the critical SNR value, but drops of more quickly below the critical value, making an exponential model of throughput a good choice for this data.
2000
1500
1000
500
0 0
10
20
30 40 SNR, dB
50
60
70
Figure 5.6: All measurements averaged over similar signal strength (i.e. RSSI) locations for Scenario A, 3Com System
5.3. SCENARIO B DATA - TWO CLIENTS
57
5.3
Scenario B Data - Two Clients
5.3.1 WaveLAN, Scenario B, All Data

Figure 5.7 shows all of the two client measurements made with the WaveLAN system. Both clients individual throughput is plotted in the graph. The gure shows a similar trend as the single client WaveLAN data. However, the two client data shows that a number of points show a lower than expected throughput and the overall throughput is lower than that of the single client data.
2000
throughput, kbps
1500
1000
500
0 0
10
20
30 40 theoretical SNR, dB
50
60
70
Figure 5.7: All measurements for Scenario B, WaveLAN System
58
5.3.2 WaveLAN, Scenario B, Spatially Averaged Data

Figure 5.8 shows a similar trend as Figure 5.7 but with a little less variation due to the spatial averaging.
2000
1500
1000
500
0 0
10
20
50
60
70
Figure 5.8: All measurements averaged for all directions at each measurement location for Scenario B, WaveLAN System
59
5.3.3 WaveLAN, Scenario B, Window Averaged Data

Figure 5.9 clearly shows the presence of a critical SNR level for the data. Additionally, the gure shows the same overall trend of a loss in throughput as was indicated in Figures 5.7 and 5.8.
2000
1500
1000
500
0 0
10
20
50
60
70
Figure 5.9: All measurements averaged over similar signal strength (i.e. RSSI) locations for Scenario B, WaveLAN System
60
5.3.4 3Com, Scenario B, All Data

Figure 5.10 shows all of the two client data points for both 3Com clients. In the gure, the drop-off in throughput below a certain critical SNR point is evident, however, there is a wide range in the throughput performance for all measured points.
2000
throughput, kbps
1500
1000
500
0 0
10
20
50
60
70
Figure 5.10: All measurements for Scenario B, 3Com System
61
5.3.5 3Com, Scenario B, Spatially Averaged Data

Figure 5.11 shows a smaller range in the variation of the throughput experienced than shown in Figure 5.10. The same overall trend found in the single client data is also present in Figure 5.11.
2000
1500
1000
500
0 0
10
20
50
60
70
Figure 5.11: All measurements averaged for all directions at each measurement location for Scenario B, 3Com System
62
5.3.6 3Com, Scenario B, Window Averaged Data

In Figure 5.12, the variation in throughput is still present. However, the exponential trend in the throughput versus the SNR seen in Figure 5.6 is also present.
2000
1500
1000
500
0 0
10
20
50
60
70
Figure 5.12: All measurements averaged over similar signal strength (i.e. RSSI) locations for Scenario B, 3Com System
5.4. SCENARIO C DATA - ONE CLIENT WITH INTERFERENCE
63
5.4
Scenario C Data - One Client with Interference
5.4.1 WaveLAN, Scenario C, All Data

Figure 5.13 presents all of the single client with interference data for the WaveLAN system. A denite loss in throughput can be seen in the gure. However, a new tendency to operate at a single throughput level can be seen in the data. This is probably due to the wireless LAN dropping to the 5.5 Mbps transmission setting. Other losses could be due to bit errors and transmission delays.
2000
throughput, kbps
1500
1000
500
0 0
10
20
50
60
70
Figure 5.13: All measurements for Scenario C, WaveLAN System
64
5.4.2 WaveLAN, Scenario C, Spatially Averaged Data

Figure 5.14 shows the overall trend in the WaveLAN data shown in Figure 5.13. However, the tendency to operate at a single throughput level is slightly more evident in Figure 5.14.
2000
1500
1000
500
0 0
10
20
50
60
70
Figure 5.14: All measurements averaged for all directions at each measurement location for Scenario C, WaveLAN System
65
5.4.3 WaveLAN, Scenario C, Window Averaged Data

The window averaging has somewhat distorted the overall trend in the WaveLAN data. Figure 5.15 has a higher throughput for mid-range SNR levels due to the low SIR levels measured at these points. This results in the misleading trend of the data in Figure 5.15.
2000
1500
1000
500
0 0
10
20
50
60
70
Figure 5.15: All measurements averaged over similar signal strength (i.e. RSSI) locations for Scenario C, WaveLAN System
66
5.4.4 3Com, Scenario C, All Data

Figure 5.16 presents all of the single client with interference data measured for the 3Com system. The data shown in the gure indicates that all measured locations appear to have been inuenced by the interference. All measurements have dropped in throughput while maintaining the same basic trend as was found in the single client (without interference) data shown in Figure 5.4.
2000
throughput, kbps
1500
1000
500
0 0
10
20
30 40 SNR, dB
50
60
70
Figure 5.16: All measurements for Scenario C, 3Com System
67
5.4.5 3Com, Scenario C, Spatially Averaged Data

Figure 5.17 shows the data tends to be centered about two different levels. This is probably due to the wireless LAN operating at the 5.5 and 2 Mbps transmission levels.
2000
1500
1000
500
0 0
10
20
50
60
70
Figure 5.17: All measurements averaged for all directions at each measurement location for Scenario C, 3Com System
68
5.4.6 3Com, Scenario C, Window Averaged Data

Figure 5.18, like Figure 5.17, shows the throughput dropping to two different operating levels. In Figure 5.18, the data can be seen to be trending downward in steps, probably as the wireless LAN system drops to the 5.5 and 2 Mbps transmission settings.
2000
1500
1000
500
0 0
10
20
50
60
70
Figure 5.18: All measurements averaged over similar signal strength (i.e. RSSI) locations for Scenario C, 3Com System
5.5. SCENARIO D DATA - TWO CLIENTS WITH INTERFERENCE
69
5.5
Scenario D Data - Two Clients with Interference
5.5.1 WaveLAN, Scenario D, All Data

Figure 5.19 shows all of the WaveLAN two client with interference measurements for both clients. This data shows the variation in throughput found in Figure 5.7 but also shows the tendency to operate at two different throughput levels as was seen in Figure 5.13. These two throughput levels are probably due to the wireless LAN using the 11 and 5.5 Mbps transmission settings.
2000
throughput, kbps
1500
1000
500
0 0
10
20
50
60
70
Figure 5.19: All measurements for Scenario D, WaveLAN System
70
5.5.2 WaveLAN, Scenario D, Spatially Averaged Data

Figure 5.20 indicates that the throughput has the similar trend as the single client data in which the throughput is constant above a critical SNR value and drops steadily off below that level. This is only the basic trend, as several points do not follow this pattern. Further the constant throughput reached by the WaveLAN data probably corresponds to the wireless LAN operating at the 5.5 Mbps transmission level.
2000
1500
1000
500
0 0
10
20
50
60
70
Figure 5.20: All measurements averaged for all directions at each measurement location for Scenario D, WaveLAN System
71
5.5.3 WaveLAN, Scenario D, Window Averaged Data

Figure 5.21 has tended to smear the overall features of the performance of the WaveLAN system that was presented in Figure 5.19. However, the data does have similar aspects to Figure 5.20, though the trends are not as clear.
2000
1500
1000
500
0 0
10
20
50
60
70
Figure 5.21: All measurements averaged over similar signal strength (i.e. RSSI) locations for Scenario D, WaveLAN System
72
5.5.4 3Com, Scenario D, All Data

Figure 5.22 presents the 3Com two client with interference data for both clients. The data appears to be a logical combination of the trends found in Figures 5.10 and 5.16. Figure 5.22 indicates the data has a variation in the throughput due to two users and a drop to lower throughput levels due to the presence of interference.
2000
throughput, kbps
1500
1000
500
0 0
10
20
50
60
70
Figure 5.22: All measurements for Scenario D, 3Com System
73
5.5.5 3Com, Scenario D, Spatially Averaged Data

Figure 5.23 shows that the spatial averaging has smoothed out the trend of the measurements into a single smooth drop-off in throughput below a critical point and fairly constant above the critical point. The data, thus, has a similar trend as Figure 5.4 but with a lower constant throughput then the single client data.
2000
1500
1000
500
0 0
10
20
50
60
70
Figure 5.23: All measurements averaged for all directions at each measurement location for Scenario D, 3Com System
74
5.5.6 3Com, Scenario D, Window Averaged Data

Figure 5.24 appears to have also smoothed the data as in Figure 5.23. However, the data still appears to be centered about two different throughput levels, rather than smoothed into a single trend as was the case for Figure 5.23.
2000
1500
1000
500
0 0
10
20
50
60
70
Figure 5.24: All measurements averaged over similar signal strength (i.e. RSSI) locations for Scenario D, 3Com System
5.6. ANALYSIS OF THROUGHPUT AND SNR MEASUREMENTS
75
5.6
Analysis of Throughput and SNR Measurements
As can be seen in the above gures, a distinct relationship between the signal strength and throughput of a wireless LAN is present. This relationship can be seen clearest in the window averages shown in the third graphs of each of the presented sets. An overall trend of an exponential curve is seen throughout all of the measurements. Additionally, the interference measurements show a distinct switching between discrete data bandwidths for the interference Scenarios C and D, as in the two levels of different throughput present in Figure 5.18. This is expected because the wireless LAN standard denes different data bandwidths that allow the throughput to decrease gracefully as the RF environment degrades. These observations were applied to the empirical throughput prediction models which are discussed in the next chapter. Notice that the Throughput measured at each location never exceeds a certain level. This maximum throughput is well below the 11 Mbps data bandwidth supposedly available with IEEE 802.11b products. It is normal for the throughput to be below the data bandwidth of a network connection because of the overhead in each data packet sent. Further, the manner in which the throughput is measured will affect the measured throughput. Since LANFielder measures the throughput for the non-optimistic case in which only a single packet is transmitted at any one time, the throughput measured is particularly low compared to the 11 Mbps data bandwidth available. However, the measured data can easily be compared to measurements made using different techniques by simply normalizing the throughput by the maximum measured throughput.
5.7
Interference Signal Strength Measurements and Predictions
Measurements of the signal strength from the interfering access point location and the interfering client locations to the desired client locations were measured. It would have been preferred to have also measured the interfering signal strengths at the desired access point locations, as well. This would have been extremely difcult because of the impossibility of accurately mimicking the desired access point locations with a measurement laptop. Additionally, the 3Com access point antenna was incompatible with any wireless
76
LAN card which could be used to measure signal strength. For this reason, all prior SNR measurements were used to derive a path loss model (of the form presented in [25] and given in equation 5.1) which could then be used to predict what the signal strength at the desired access point location from the interfering access point and client locations. This process is discussed further in Section 5.7.1.
5.7.1 Predicting the Interference at Signal Strength Locations

SitePlanner Model of Measurement Cases In order to predict the needed interference, the wireless prediction tool SitePlanner was used. In SitePlanner, the environments from all three cases were modeled in a three dimensions. Partitions such as walls and cubicles were divided into four different types: Primary Partitions, Secondary Partitions, Doors and Metal [26]. Each partition has a different Attenuation Loss associated with it. Primary partitions were used to model the cinder block hallways and drywall partitions. Secondary partitions were used to model the cubicle partitions. The door partitions are self-explanatory and the metal partitions were used to model the elevators. SitePlanner has facilities for using eight different partition types but these were not all used. All partitions extended from oor to ceiling except for the cubicle partitions which were modeled to extend to 1.7 meters as was true of the actual partitions. Figure 5.25 shows the environment as it was modeled as well as some of the measurements that were entered into SitePlanner. The different partition types are represented using different colors. The primary, secondary, door, and metal partitions are blue, gray, purple, and orange, respectively.
Inputting SNR Measurements into SitePlanner Using InFielder, SNR measurements made using the custom RSSI measurement software, were entered into the drawing at the precise three dimensional coordinates that they were measured at. In order to do this, the RSSI measurements were rst linearly averaged over the four cardinal directions for each measurement point and then placed into a SitePlanner logle format and imported. The measurements were imported in such a way that the measurements were associated with the correct access point location in an automatic
5.7. INTERFERENCE SIGNAL STRENGTH MEASUREMENTS AND PREDICTIONS 77
Figure 5.25: The entire measurement environment modeled in SitePlanner. manner.
Path Loss Prediction in SitePlanner SitePlanner has several different prediction models available [26] The model used for these predictions was the Wall/Floor Attenuation Factor, Single Path Loss Exponential model. Since all measurements were taken on a single oor, this model uses Equation 5.1 for all predictions. d d0
P L(d) = P L(d0 ) + 10nlog
+
i
[(Pi )(AFi )]
(5.1)
This equation predicts the path loss, P L at distance d. d0 is a reference distance, P L(d0 ) is the path loss at the reference distance, n is the path loss exponent, Pi is the number of partitions of type i and AFi is the attenuation factor for partitions of type i. SitePlanner uses the Friis Equation [24] to calculate the path loss at the reference distance d0 [26]. This formula for the value of P L(d0 ) in dB is given in equation 5.2.
78

4d0 P L(d0 ) = 10 log Optimizing Path Loss Prediction Model Parameters
(5.2)
After all of the measurements were imported into SitePlanner and the access points were correctly modeled, the Optimatic module of SitePlanner was used to calculate optimal values for n and AFi using the RSSI measurements previously entered into SitePlanner. SitePlanner uses a basic Mean Squared Error t algorithm to nd an ideal parameters based on known measured data. Using all of the 3Com and WaveLAN measured data from the desired access point to the desired client locations, the following parameters resulted in a mean error of 0.05 dB and a standard deviation of 3.27 dB. Further, 100measurements were within 10 dB of the predicted value. The calculated parameters were 3.02 for the path loss exponent, 1.70 dB for the primary partition attenuation factor, 2.41 dB for the secondary partition attenuation factor and 2.50 dB for the door partition attenuation factor. There was not enough measurements to compute an ideal value for the elevator partition attenuation factor, so the default value of 5.00 dB was used. These parameters and all of the settings used in SitePlanner are summarized in Table 5.1. It should also be noted that these results are in excellent agreement with the measurements made in [11]. Table 5.1: Summary of All Congured Parameters Used in SitePlanner
Parameter Name 3Com Access Point Transmit Power WaveLAN Access Point Transmit Power BayNetworks Access Point Transmit Power BayNetworks Client Transmit Power Access Point Heights Receiver Transmit Power Receiver Antenna Gain Receiver Height Transmit/Receive Frequency Reference Distance (d0 ) Path Loss at the Reference Distance, P L(d0 ) Path Loss Exponent, n Primary Partition Loss Factor, AF1 Secondary Partition Loss Factor, AF2 Door Partition Loss Factor, AF3 Metal Partition Loss Factor, AF4 Parameter Value 18 dBm 15 dBm 15 dBm 15 dBm 2.5 meters 15 dBm 2 dBi 0.91 meters 2.412 GHz 1 meter 40.0 dB 3.02 1.70 dB 2.41 dB 2.50 dB 5.00 dB
5.7. INTERFERENCE SIGNAL STRENGTH MEASUREMENTS AND PREDICTIONS 79 Modeling Access Points in SitePlanner The desired access points are modeled in SitePlanner as base stations and can be seen in Figure 5.25 as concentric black circles and are labeled with TX names. The access points were correctly modeled to have 15 or 18 dBm transmit powers for the WaveLAN or 3Com access points, respectively. The access points were modeled at the precise three dimensional locations that they were positioned in the actual environment. This includes the correct antenna height. Each access point had a 2 dBi omnidirectional antenna attached. This is the correct maximum gain, though not terribly accurate in terms of the actual radiation pattern of the access points. It was, however, felt that since measurements were averaged over the four cardinal directions, the omnidirectional pattern would result in an accurate representation of the measurements. In addition to the desired access point, the interfering access point and interfering client were also modeled as base stations and are shown in Figure 5.25. These base stations were also modeled in the correct three dimensional location and as having a 2 dBi omnidirectional antenna and a transmit power of 15 dBm, which matches the actual specications of the BayNetworks products.
Additional Parameters of the SitePlanner Model It was necessary to set several additional parameters in addition to the locations and transmit powers of the access points. All base stations were congured to be IEEE 802.11b access points with the DSSS channels available and channel 1 the active channel. That is, all predictions were made using a center frequency of 2412 MHz which corresponds to the center frequency of channel 1 in the DSSS portion of the IEEE 802.11 standard [30]. Several receiver parameters were also congured in SitePlanner. The mobile receiver was congured to have a height of 0.91 meters and a receiver antenna gain of 2 dBi which was included in all path loss calculations.
Predicting Interference Signal Strength Once an effective prediction model was calculated using Equation 5.25 in SitePlanner, the model was then used to predict the signal strength at the desired access point location from the interfering access point location and the interfering client locations. To do this, base stations were modeled in SitePlanner as was previously described and as is out-
80
lined in Table 5.1. These base stations were then used to predict the signal strength at the base stations which represent the desired access points. Combining these RSSI predictions with the RSSI measurements of the interfering access point and interfering client at the desired client locations resulted in four interference measurements for every desired client location. These four measurements are used to calculate a single SIR value using the technique outlined in Section 5.7.2.
5.7.2 Computing Total Interference Signal Strength

The measured interference was found through a combination of measurement and through predictions based on measurements. As a result, each throughput measurement in Cases C and D is associated with four interference signal strength measurements. These four measurement locations are illustrated in Figure 5.26. The four path loss measurements, P L, are shown using a solid arrows in the gure. For reference, the SNR for a client is based on the path loss from the desired client and the desired access point location, shown in Figure 5.26 as a dashed line.
Desired Client
PL( AP
) ,C d (C i PL
Desired Transmission
Desired AP
d)
i ,C ) d
Interfering Client
Interfering Transmission
Interfering AP
Figure 5.26: A diagram of the interference paths between the different wireless LAN users in Scenarios C and D.
PL (A P,
AP
) ,AP d Ci PL(
5.7. INTERFERENCE SIGNAL STRENGTH MEASUREMENTS AND PREDICTIONS 81 The four path loss measurements, P L(APi , APd ), P L(APi , Cd ), P L(Ci , APd ) and P L(Ci , Cd ), correspond to the four respective RSSI values used to compute a combined SIR value using Equation 5.3.
SIRnet =
RSSI(APd , Cd ) mean((RSSI(APi , APd ) + RSSI(Ci , APd )), (RSSI(APi , Cd ) + RSSI(Ci , Cd ))) (5.3)
In this equation, RSSI(Xa , Yb ) is the received signal strength at location Yb from location Xa . Xa and Yb are symbols in which X and Y are either AP or C or an access point or client location respectively. The subscript a or b is either d or i to represent the desired or interfering location respectively. Thus, RSSI(APi , Cd ) is the signal strength at the desired client from the interfering access point. Note that all RSSI values were summed and averaged by rst converting to a linear value, performing the desired operation and then converting back into dB. Equation 5.3 implements a linear average of the total interference power at the desired access point, APd , location with the total interference power at the desired client location, Cd . An average performed in this manner does not excessively decrease a strong interference signal when combined with a much weaker interference signal. An average is used because it is assumed that the access point will be receiving half the time and the client will be receiving half the time, rather than both simultaneously. This is the motivation behind Equation 5.3.
5.7.3
Advantages of SIR Calculation
This formula has the advantage that an unusually strong interfering signal is not excessively reduced by a unusually weak interfering signal. Further, Equation 5.3 is based on the average of the total interference power at the desired client and the total interference power at the desired access point. The two values are averaged because the client and access point never transmit simultaneously. It is theorized then that the throughput experienced will be related to the average interference power at the desired client and the desired server.
82
5.7.4 Resulting Plots of Measured and Predicted SIR Data

After applying Equation 5.3 to the measured and predicted interference power levels, the following plots of the data were generated. These plots use the measured throughput from Scenarios C and D and plot the throughput versus the computed net interference power and versus the SIR values. These plots are presented below.
5.7. INTERFERENCE SIGNAL STRENGTH MEASUREMENTS AND PREDICTIONS 83 Scenario C, WaveLAN Throughput versus Net Interference Power
1800 1600 1400 1200 1000 800 600 400 200 0 100
throughput, kbps
90
80 70 60 net interference power, dBm
50
40
Figure 5.27: Plots of the WaveLAN measured throughput versus the calculated net interference power averaged over all cardinal directions for each physical location
84
Scenario C, WaveLAN Throughput versus SIR
2000 1800 1600 1400 throughput, kbps 1200 1000 800 600 400 200 0 30 20 10 0 10 20 computed SIR, dB 30 40 50
Figure 5.28: Plots of the WaveLAN measured throughput versus the calculated SIR averaged over all cardinal directions for each physical location
5.7. INTERFERENCE SIGNAL STRENGTH MEASUREMENTS AND PREDICTIONS 85 Scenario C, 3Com Throughput versus Net Interference Power
1400 1300 1200 1100 throughput, kbps 1000 900 800 700 600 500 400 100 90 80 70 60 net interference power, dBm 50 40
Figure 5.29: Plots of the 3Com measured throughput versus the calculated net interference power averaged over all cardinal directions for each physical location
86
Scenario C, 3Com Throughput versus SIR
Figure 5.30: Plots of the 3Com measured throughput versus the calculated SIR averaged over all cardinal directions for each physical location
5.7. INTERFERENCE SIGNAL STRENGTH MEASUREMENTS AND PREDICTIONS 87 Scenario D, WaveLAN Throughput versus Net Interference Power
2000 1800 1600 1400 throughput, kbps 1200 1000 800 600 400 200 0 100 90 80 70 60 net interference power, dBm 50 40
Figure 5.31: Plots of the WaveLAN measured throughput versus the calculated net interference power averaged over all cardinal directions for each physical location. All data points for both clients are shown in the graph.
88
Scenario D, WaveLAN Throughput versus SIR
Figure 5.32: Plots of the WaveLAN measured throughput versus the calculated SIR averaged over all cardinal directions for each physical location. All data points for both clients are shown in the graph.
5.7. INTERFERENCE SIGNAL STRENGTH MEASUREMENTS AND PREDICTIONS 89 Scenario D, 3Com Throughput versus Net Interference Power
1400
1200
1000 throughput, kbps
800
600
400
200
0 100
90
80 70 60 net interference power, dBm
50
40
Figure 5.33: Plots of the 3Com measured throughput versus the calculated net interference power averaged over all cardinal directions for each physical location. All data points for both clients are shown in the graph.
90
Scenario D, 3Com Throughput versus SIR
Figure 5.34: Plots of the 3Com measured throughput versus the calculated SIR averaged over all cardinal directions for each physical location. All data points for both clients are shown in the graph.
5.8. MEASUREMENT RESULTS CONCLUSIONS
91
5.7.5 Analysis of Interference Measurements and Calculations

As can be seen in Figures 5.27 through 5.34, there is very little correlation between the net interference signal strength and the measured throughput measured, though there is some correlation between the SIR and the throughput. Further, by comparing scenarios C and D, it is clear that there is consistency from between the two scenarios. Thus, the throughput of a client can not be predicted directly from the net interference power, it does appear to be a useful parameter.
5.8
Measurement Results Conclusions
It has been seen from the measurement results presented in this chapter that there is denitely promise for creating empirical throughput prediction models. The presented plots of the measurements show a denite correlation between the signal to noise ratio, and to some small degree, to the net interference power. These relationships are clearest in the window averaged plots that have been presented. As a result, the measurements are used in Chapter 6 to create empirical throughput prediction models for wireless LANs.
92
Chapter 6 Empirical Throughput Prediction Models

6.1 Development of Wireless LAN Performance Models
After analyzing the data presented in the previous chapter, it appears that the throughput of a wireless LAN can be modeled and predicted. The goal of this chapter is to present models that have been developed to predict the throughput performance, T , of wireless LAN networks. The models that have been developed are presented in this chapter, along with the reasoning behind the models and the accuracy of the models, when compared to the measurements. All of the models presented were developed empirically (that is, models were tuned using the measured data), using the data that was averaged per each measurement location. That is, each data point used in the models was the average of four measurements (north, east, south, west) at a single physical location.
6.2
Optimizing Prediction Model Parameters
To t the prediction models to the data, a technique similar to the MMSE tting is used [20]. The difculty with a MMSE error t is that it requires a linear or polynomial equation to function. Since this thesis attempts to t to nonlinear models, an exact MMSE tting procedure is not valid. Therefore, the MATLAB command nlint, which is part of the Statistics Toolbox, was used. This command implements the Gauss-Newton method to nd a local minimum of a given equation using provided data points [16]. The technique 93
94
CHAPTER 6. EMPIRICAL THROUGHPUT PREDICTION MODELS
requires an initial guess to function, and does not guarantee the best t, but only a local minima. However, a good initial guess typically provides a result which is an extremely good t to the data.
6.3
Prediction Models for Scenario A One User
Two primary models were considered for this scenario. The rst of these models is a piece-wise linear t to the measured data. The second model is an exponential curve t to the measured data. Each of these models have three parameters that need to be specied: 1. Tmax , the throughput experienced in ideal circumstances (i.e. under maximum SNR and without interference) 2. A or , a slope or rate of drop-off in throughput as the SNR decreases 3. T0 or SN R0 , intercept points for the model Note that unless otherwise specied, all throughputs are in units of kbps and all SNR values are in dB. All calculations were carried out in MATLAB after having imported all of the measurement data into MATLAB from the LANFielder logle format. Details of these models are now presented in the sections below.
6.3.1
Scenario A, Linear Model
Formula for the Scenario A Linear Model A linear model was rst considered because in looking at the single user measured data, without co-channel users, it appeared that the wireless LAN tended to operate with a very steady throughput until the signal to noise ratio (SNR) dropped below a certain critical level. Below this critical SNR, represented as SN Rc , the throughput appears to drop-off linearly. This drop-off can be clearly seen in Figures 6.1, 6.2, 6.3, and 6.4. The formula for a reasonable model is given in equation 6.1.
6.3. PREDICTION MODELS FOR SCENARIO A

T max T = A SN R T
95
if if
SN R > SN Rc SN R <= SN Rc
(6.1)
In the linear prediction equation, Tmax is a critical throughput that is vendor and application specic. That is, different wireless LAN cards (e.g. WaveLAN, 3Com, etc.) running different network applications (e.g. telnet, VoIP, FTP, etc.) will have a different value for Tmax . Tmax is the throughput that the wireless LAN will provide in perfectly ideal circumstances. The parameters A and T0 are also vendor and application specic, and are used to predict the throughput of the wireless LAN when the performance is not ideal. That is, when the SNR drops below SN Rc , the throughput drops of with a slope of A and reaches a throughput of T0 when the SNR drops to 0 dB. Note that the value of SN Rc is the intersection of the two lines of the model. Thus, SN Rc can be derived from Tmax , A and T0 using Equation 6.2. Tmax + T0 A
SN Rc =
(6.2)
Calculating the Linear Model Parameters The procedure explained in Section 6.2 was used to t the parameters of Equation 6.3. The initial guesses selected for the WaveLAN data were Tmax equal to 1800, A equal to 100 and T0 equal to 0. The initial guesses for 3Com were Tmax equal to 2100, A equal to 100 and T0 equal to 0. After obtaining the optimum values for Tmax , A and T0 , the value of SN Rc was calculated using Equation 6.2. The optimized model parameters are given in Table 6.1. Further, the resulting model and the locationally averaged data are shown in Figures 6.1 and 6.2. The equations used to generate the model are shown on the graphs along with the mean and standard deviation in the error. These error statistics are also presented for this and all prediction models in this Chapter in Table 6.2. As can be seen from the statistics, both the WaveLAN and the 3Com data t to the linear model quite well. Both had mean errors of 0 and standard deviations of less than about 150 kbps. This is quite good considering LANFielder and SiteSpy have a measurement precision of about 10 kbps.
96
2000
throughput, kbps
1500
1000
500
measured points with location average linear model fit to data T=1863.6, if SNR >21.8 T=47.1*SNR+835.9, if SNR <=21.8 error: =0 =31.3
0 0
10
20
30 40 SNR, dB
50
60
70
Figure 6.1: WaveLAN Scenario A, Location Averaged Data and Linear Throughput Prediction Model
Constraining the Linear Model
As can be seen from Figures 6.1 and 6.2, the linear model of throughput does not predict zero throughput at a 0 dB SNR. For this reason, the Linear models were ret to the data, except with T constrained to be equal to zero at zero SNR which is equivalent to forcing T0 to be zero. In this manner the prediction models are simpler, more intuitive and more realistic, since throughput drops to zero when the SNR reaches zero. The model was again t using the MATLAB nlint command as described in Section 6.2 and was performed previously in Section 6.3.1 except that T0 was not considered an adjustable model parameter.
97
2000
throughput, kbps
1500
1000
500
measured points with location average linear model fit to data T=1981.3, if SNR >18.5 T=91.6*SNR+287.5, if SNR <=18.5 error: =0 =151.5
0 0
10
20
30 40 SNR, dB
50
60
70
Figure 6.2: 3Com Scenario A, Location Averaged Data and Linear Throughput Prediction Model
The resulting error statistics are given in Table 6.2 and the calculated model parameters are given in Table 6.1. resulting constrained models and the measured data are plotted in Figures 6.3 and 6.4. Table 6.2 and Figures 6.3 and 6.4 show that the mean error, , increased from 0 to 26.29 kbps for the WaveLAN data and from 0 to 1.12 kbps for the 3Com data. The error standard deviation, , increased from 31.3 kbps to 153.3 kbps for the WaveLAN data and from 151.5 kbps to 153.4 kbps for the 3Com data. As can be seen from these results, constraining the model introduces a mean error and the error standard deviation increases. Thus, the more intuitive, simpler model that results from constraining T to be zero at 0 SNR comes with a large penalty in accuracy.
98
2000
throughput, kbps
1500
1000
500
measured points with location average linear model fit to data T=1863.6, if SNR >15.4 T=121.2*SNR, if SNR <=15.4 error: =26.29 =153.3
0 0
10
20
30 40 SNR, dB
50
60
70
Figure 6.3: WaveLAN Scenario A, Location Averaged Data and Linear Constrained Throughput Prediction Model
6.3.2 Scenario A, Exponential Model

Formula for the Scenario A Exponential Model After considering a piece-wise linear model, an exponential model was also considered. The advantage of an exponential model is that it is a simple model of the behavior of the wireless LAN throughput which, above a critical SNR value is close to a maximum throughput, Tmax . Below the critical SNR value, the throughput drops off quickly to zero. However, the exponential model should provide a better match to the data because it provides a more gradual transition around the critical SNR value. The formula used for the exponential model is give in Equation 6.3
99
2000
throughput, kbps
1500
1000
500
measured points with location average linear model fit to data T=1978.1, if SNR >16.9 T=116.7*SNR, if SNR <=16.9 error: =1.12 =153.4
0 0
10
20
30 40 SNR, dB
50
60
70
Figure 6.4: 3Com Scenario A, Location Averaged Data and Linear Constrained Throughput Prediction Model
T1 = Tmax (1 exp((SN R SN R0 )))
(6.3)
As was the case for the Linear Model, the values of Tmax , and SN R0 are application and vendor specic, but can be tuned by measurement. Tmax is the throughput that the wireless LAN connection will experience in ideal circumstances. can be thought of as the slope of the exponential curve and governs the rate at which the exponential drops off. Lastly, the parameter SN R0 is the SNR value at which the wireless LAN provides zero throughput.
100
Table 6.1: The Values of the Calculated Parameters and Error Statistics for the Scenario A Throughput Prediction Models, as dened in Equations 6.1 and 6.3
Model Name WaveLAN Linear WaveLAN Linear, Constrained WaveLAN Exponential WaveLAN Exponential, Constrained 3Com Linear 3Com Linear, Constrained 3Com Exponential 3Com Exponential, Constrained Tmax 1,863.6 kbps 1836.3 kbps 1907.2 kbps 1874.7 kbps 1981.3 kbps 1978.1 2048.6 kbps 2050.9 kbps T0 or SN R0 47.1 835.9 kbps/dB kbps 121.2 0 kbps kbps/dB 0.099 -3.0 dB 0.144 91.6 kbps/dB 116.7 kbps/dB 0.097 0.095 0 dB 287.5 kbps 0 kbps 0.2 dB 0 A or SN Rc 21.8 dB 15.4 dB N/A N/A 18.5 dB 16.9 dB N/A N/A 0 kbps 26.29 kbps 0 kbps 1.71 kbps 0 kbps 1.12 kbps 0 kbps -0.2 kbps 31.3 kbps 153.3 kbps 34.6 kbps 48.7 kbps 151.5 kbps 153.4 kbps 147.3 kbps 147.3 kbps
Calculating the Exponential Model Parameters
The procedure explained in Section 6.2 was used to t the parameters of Equation 6.3. The initial guesses selected for the WaveLAN data were Tmax equal to 1800, equal to 0.1 and SN R0 equal to 0. The initial guesses for 3Com were Tmax equal to 2100, equal to 0.1 and SN R0 equal to 0. The model parameters resulting from this tting process are given in Table 6.1 and the error values are given in Table 6.2. The exponential provided a fairly good t to the data as can be seen in Figures 6.5 and 6.6. The equations used to generate the model are shown on the graphs along with the mean and standard deviation in the error. As can be seen from the error statistics and the gures, the exponential provides a good t to the data with the error statistics very similar to those for the linear model. Again, the mean error has been reduced to 0 kbps and the error standard deviation is less than about 150 kbps. The exponential, however, is closer to intersecting 0 kbps throughput when the SNR is zero than the linear models were.
101
2000
throughput, kbps
1500
1000
500
measured points with location average exponential model fit to data T=1907.2(1exp(0.099(SNR+3))) error: =0 =34.6
0 0
10
20
30 40 SNR, dB
50
60
70
Figure 6.5: WaveLAN Scenario A, Location Averaged Data and Exponential Throughput Prediction Model Constraining the Exponential Model As can be seen from Figures 6.5 and 6.6, the exponential model of throughput does not predict zero throughput at a 0 dB SNR. For this reason, the Exponential models were ret to the data, except with SN R0 forced to be equal to zero. In this manner the prediction models are simpler, more intuitive and more realistic. The model was again t using MATLAB nlint command as described in 6.2 and was performed previously in Section 6.3.2 except that SN R0 was not considered an adjustable model parameter. The resulting error statistics are given in Table 6.2 and the calculated model parameters are given in Table 6.1. Additionally, the resulting constrained models and the measured data are plotted in Figures 6.3 and 6.4.
102
2000
throughput, kbps
1500
1000
500
measured points with location average exponential model fit to data T=2048.6(1exp(0.097(SNR0.2))) error: =0 =147.3
0 0
10
20
30 40 SNR, dB
50
60
70
Figure 6.6: 3Com Scenario A, Location Averaged Data and Exponential Throughput Prediction Model As can be seen from the results of constraining the exponential model, there is a slight mean error introduced and a slight increase in the error standard deviation. Again, the use of this, simpler more intuitive constrained model has resulted in a loss in prediction accuracy.
6.3.3 Foundations for Other Scenario Models

Scenario A has provided a foundation model for use Scenarios B, C and D. Considering all of the results of the model optimizations presented, as summarized in Table 6.2, the Exponential prediction model has proven to be superior to the Linear prediction model.
103
2000
throughput, kbps
1500
1000
500
measured points model fit to data T=1874.7(1exp(0.144(SNR))) error: =1.71 =48.7
0 0
10
20
30 40 SNR, dB
50
60
70
Figure 6.7: WaveLAN Scenario A, Location Averaged Data and Exponential Constrained Throughput Prediction Model The Exponential model is conceptually simpler to implement than linear model and has fewer associated parameters. In addition, the exponential model provides slightly lower error statistics, especially when constrained and especially for the WaveLAN data. As a result the exponential model will be used as a basis for further prediction models for Scenarios B, C and D.
104
2000
throughput, kbps
1500
1000
500
measured points model fit to data T=2050.9(1exp(0.095(SNR))) error: =0.02 =147.3
0 0
10
20
30 40 SNR, dB
50
60
70
Figure 6.8: 3Com Scenario A, Location Averaged Data and Exponential Constrained Throughput Prediction Model
105
Table 6.2: A Summary of the Accuracy of the Empirical Throughput Prediction Models.
Scenario A Manufacturer WaveLAN WaveLAN WaveLAN WaveLAN 3Com 3Com 3Com 3Com WaveLAN WaveLAN 3Com 3Com WaveLAN WaveLAN 3Com 3Com D WaveLAN WaveLAN 3Com 3Com Model Linear, Equation 6.1 Linear, Constrained, Equation 6.1 Exponential, Equation 6.1 Exponential, Constrained, Equation 6.1 Linear, Equation 6.3 Linear, Constrained, Equation 6.3 Exponential, Equation 6.3 Exponential, Constrained, Equation 6.3 Two Users, Equation 6.4 Two Users, Constrained , Equation 6.4 Two Users, Equation 6.4 Two Users, Constrained, Equation 6.4 One User with Interference, Equation 6.8 One User, with Interference, Constrained, tion 6.8 One User with Interference, Equation 6.8 One User with Interference, Constrained, tion 6.8 Two User with Interference, Equation 6.10 Two User, with Interference, Constrained, tion 6.10 Two User with Interference, Equation 6.10 Two User with Interference, Constrained, tion 6.10 0 kbps 26.29 kbps 0 kbps 1.71 kbps 0 kbps 1.12 kbps 0 kbps -0.2 kbps 0 kbps 0 kbps 0 kbps 0 kbps 0 kbps 0 kbps 0 kbps 0 kbps 0 kbps 2.65 kbps 0.2 kbps -10.2 kbps 31.3 kbps 153.3 kbps 34.6 kbps 48.7 kbps 151.5 kbps 153.4 kbps 147.3 kbps 147.3 kbps 185.2 kbps 178.9 kbps 225.2 kbps 225.2 kbps 254.9 kbps 258.3 kbps 136.2 kbps 135.6 kbps 268.7 kbps 253.0 kbps 196.1 kbps 195.7 kbps
Equa-
Equa-
Equa-
Equa-
106
6.4
Prediction Models for Scenario B Two Users
Scenario B utilizes the basic one client exponential model derived for Scenario A as a base for its models. The two client model for throughput prediction introduces Tloss factors to adjust the ideal, one User throughput to match the actual throughput in the presence of another user. The exponential model given in Equation 6.3 was used as the basis for this model, rather than the piece-wise linear model of Equation 6.1, because it is simpler and more intuitive than the linear model. It also lends itself well to be modied by various situations that can cause a loss in throughput for two or more users of a shared wireless LAN.
6.4.1 Throughput Prediction Formula for Two Simultaneous Users

The calculation of the throughput of a client which is sharing the access to a wireless LAN is fairly complex. The basic concept of the throughput prediction model is to start with the normal, ideal wireless LAN throughput and subtract various losses due to known difculties with wireless LAN systems. The basic formula for two clients is shown in Equation 6.4.
T = T1 (SN R1 ) Tlosscompete (SN R2 ) Tlosshidden (P L(C1 , C2 )) Tof f set
(6.4)
As can be seen from Equation 6.4, the throughput of an individual client is based on the ideal one user throughput, T1 from Equation 6.3, which represents the single client, exponential model developed for Scenario A. This single client throughput is modied in Equation 6.4 by two loss functions and a basic throughput offset value, To f f set. SN R1 is the SNR of the considered client, SN R2 is the SNR of the other client which is sharing the access point and is simultaneously transmitting data. Tlosscompete (SN R2 ), is a loss in throughput due to the client having to compete with other users for throughput to the desired access point. The formula for this loss parameter is give in Equation 6.5. Tlosshidden is a loss in throughput due to the hidden terminal problem [12]. The formula for this loss factor is given in Equation 6.6.
Tlosscompete (SN R2 ) = Acompete [Tmax T1 (SN R2 )]
(6.5)
6.4. PREDICTION MODELS FOR SCENARIO B
107
Tlosscompete is the loss in throughput due to the congestion caused by both users attempting to use the same access point at the same time. As shown in Equation 6.5, it is a function of the ideal throughput of the other user. Thus, the more the other client is experiencing poor throughput, the more likely that client is to interfere with, and decrease the throughput of the rst client. In Equation 6.5, SN R2 is the SNR of the other client. Tmax is the maximum throughput under ideal circumstances, as dened by the results of the One Client Exponential Model from Scenario A. T1 is the ideal exponential throughput calculation function, also dened by the results from Scenario A.
Tlosshidden (N ) = Ahidden P L(C1 , C2 )
(6.6)
Tlosshidden is a loss in throughput due to the so called hidden terminal problem [12]. The hidden terminal problem is when two clients can not detect the transmissions of one another and so are unable to avoid collisions with one anthers packets. This loss factor is based on the path loss from the main, rst client, C1 , to the second client, C2 . This path loss is represented as P L(C1 , C2 ). As the path loss between the two clients increases, the clients are less likely to detect the transmissions of one another and thus the probability that packets sent from different clients will collide at the access point increases. It should be noted that the value of Ahidden will change if the RTS/CTS functionality (see Appendix A for more information) of the wireless LAN is enabled. This functionality was not enabled for the measurement made in this thesis.
6.4.2 Calculating Parameters for the Two User Model

Due to the complexity of the two user model shown in Equation 6.4, it was imperative that the nlint command in MATLAB be used to compute the best values for the parameters of the model. However, before using the same MATLAB solution technique as was applied in Scenario A using nlint, it was rst necessary to obtain the path loss between different client locations, P L(C1 , C2 ). These values are used in Equation 6.6. The path loss values were found using the SitePlanner RF Prediction model previously calculated and utilized in Section 5.7.1 and given in Equation 5.1. This prediction model allowed the accurate calculation of the path loss between different measurement locations. After the path losses between all client locations were known, it is possible to calculate the opti-
108
mal values for the two client prediction models. The results of the MATLAB calculations to calculate the optimal model parameters are given in Table 6.3. The error statistics for the model given in Equation 6.4 with the optimized model parameters are give in Table 6.2. All this information is also shown on Figures 6.9 and 6.10. These gures provide graphical insight into the accuracy of this above two client throughput prediction model. Table 6.3: The Values of the Calculated Parameters and Error Statistics for the Scenario B Throughput Prediction Models, as given in Equation 6.4
Model Name WaveLAN 2 Client Model WaveLAN 2 Client Model, Constrained 3Com 2 Client Model 3Com 2 Client Model, Constrained Acompete 0.359 0.309 Ahidden 3.48 kbps/dB 3.84 kbps/dB T0 -81 kbps -91.8 kbps 0 kbps 0 kbps 185.2 kbps 178.9 kbps
0.034 0.072
6.44 kbps/dB 6.46 kbps/dB
60.6 kbps 64.7 kbps
0 kbps 0 kbps
225.2 kbps 225.2 kbps
6.4.3 Constraining the Two User Throughput Measurement Model

In order to reduce the number of parameters needed for the two client model given in Equation 6.4, the model was constrained as was done for the Single client model. That is, the value of T1 (SN R) is forced to be zero when SNR is zero. T1 (SN R) is the single user model given in Equation 6.3 and can be constrained in the desired manner by simply setting the T0 parameter of this equation to zero. The parameters of the two user model are then re-computed using the MATLAB nlint command. The results of this constrained optimization are given in Tables 6.3 and 6.2. Additionally, the success of this technique can be viewed in Figures 6.11 and 6.12 where the theoretical and actual throughput are compared.
6.4.4 Extending the Two User Model to N Users

The two user model presented in Equation 6.4 could be generalized to multiple users by calculating the Tlosscompete and Tlosshidden values for several other operating clients rather
109
2000 1800 1600 actual throughput, kbps 1400 1200 1000 800 600 400 200 0 0 500
measured/predicted points ideal fit line T2 = T1(SNR1) 0.359*(TmaxT1(SNR2)) ... ... 3.48*PL(C1,C2) +81 error: =0 =185.2
1000 1500 predicted throughput, kbps
2000
Figure 6.9: WaveLAN Scenario B, 2 User Prediction Model. This gure shows the predicted versus actual throughput results and an ideal t line
than just one other client. A modied formula to predict the performance of a wireless LAN with N total users is presented in Equation 6.7. Further modications of this formula may be necessary, but measurements need to be performed to verify this.
N
T = T1 (SN R1 )
j=2
[Tlosscompete (SN Rj ) + Tlosshidden (P L(C1 , Cj ))] Tof f set
(6.7)
110
measured/predicted points ideal fit line T2 = T1(SNR1) 0.034*(TmaxT1(SNR2)) ... ... 6.44*PL(C1,C2) 60.6 error: =0 =225.2
2000
Figure 6.10: 3Com Scenario B, 2 User Prediction Model. This gure shows the predicted versus actual throughput results and an ideal t line
6.4.5 Analysis of the Two User Throughput Prediction Model

The results of the tting technique for this model show the results to be fairly accurate. This combined with the simple, intuitive nature of the prediction model make it a valuable contribution for wireless LAN installation design and deployment. However, there are two other very important parameters that need to be considered that were neglected in the two user model (Equation 6.4) due to a lack of available data: 1. The rst of these is the saturation of the available bandwidth. For all measurements taken and used to derive these models, there was no evidence that the clients were being limited by the users utilizing all of the available bandwidth at a single access
111
measured/predicted points constrained ideal fit line T2 = T1(SNR1) 0.309*(TmaxT1(SNR2)) ... ... 3.84*PL(C1,C2) +91.8 error: =0 =178.9
2000
Figure 6.11: WaveLAN Scenario B, 2 User Constrained Prediction Model. This gure shows the predicted versus actual throughput results and an ideal t line point. The Tlosscompete parameter could be used with some modication to model the saturation that occurs when all available bandwidth is utilized. The typical saturation point of wireless LANs has been reported to occur at approximately 2 or 3 users operating at maximum usage [1]. Therefore it was not possible to model the saturation that occurs using the measured data presented in this thesis. 2. The second unconsidered difculty is the user access patterns. All measurements taken for this thesis had users attempting to access the wireless network as often as possible. Users were always trying to transmit or receive in a continuous fashion, thus the probability of two users attempting to transmit at the same time was extremely high. However, it would extremely useful if an Erlang or similar traf-
112
measured/predicted points constrained ideal fit line T2 = T1(SNR1) 0.072*(TmaxT1(SNR2)) ... ... 6.46*PL(C1,C2) 64.7 error: =0 =225.2
2000
Figure 6.12: 3Com Scenario B, 2 User Constrained Prediction Model. This gure shows the predicted versus actual throughput results and an ideal t line c model were applied to allow wireless LANs to calculate an actual user capacity based on the shared nature of a network in which users are not attempting to transmit continuously. This sort of analysis is beyond the scope of this thesis.
6.5. PREDICTION MODELS FOR SCENARIO C
113
6.5
Prediction Models for Scenario C One User with Interference
Scenario C builds directly from the exponential Scenario A model given in Equation 6.3. The presence of an interfering client tends to lower the throughput of the desired client. As a result, the model developed for Scenario C was a modied version of the exponential model developed in Scenario A. The exponential model was chosen over the linear model because the exponential has fewer parameters and is more intuitive to use. The model developed below introduces a Tlossint factor similar to the Tloss factors introduced in the Scenario B model (Equation 6.4). This loss factor, Tlossint models the additional losses in throughput due to the presence of interference.
6.5.1 Quantifying the Interference Power

In order to predict the throughput performance of a wireless LAN in the presence of an interfering wireless LAN system, the interference must be quantied. This is more challenging than it might rst appear. The difculty is that all interference measurements were made with another wireless LAN system in use as the interferer. As a result, each measurement had two sources of interference and two recipients of interference. This is shown diagrammatically in Figure 5.26, where the intentional transmissions are represented by the dashed lines. The interfering client and interfering access point are attempting to communicate with one another while the desired client and desired access point attempt to do the same thing. Additionally, there is intentional interference between the interfering system and the desired system. In order to accurately predict the throughput experienced between the desired client and the desired access point, four signal strengths must be considered. That is, the interference power at both the desired client and the desired access point from both the interfering client and the interfering server need to be taken into account. To do this, a combination of measurements and predictions, as outlined in Section 5.7.2 were performed to nd four RSSI values for each throughput measurement. Section 5.7.2 also described how to calculate a net SIR using Equation 5.3. Thus, using the equation, each throughput measurement has an associated net SIR value associated with it. These net SIR values are used in the throughput prediction model for Scenarios C and D.
114
6.5.2 Formulation of the Scenario C Model

A Tlossint parameter is introduced in this section for the Scenario C, single client plus interference throughput prediction model. Tlossint is intended to model the additional loss in throughput caused by the interference. The main formula for the model is shown in Equation 6.8. In Equation 6.8 a users throughput is predicted from the theoretical SNR and the measured SIR values for the clients location. The SNR value is the signal to noise measurement made when the interfering client was not operating. Referring to Figure 5.26, the SIR value is actually based on the net effect of all the interference power from an interfering client and an interfering access point on the desired client and desired access points. A detailed treatment of how to compute the SIR parameter is presented in Sections 6.5.1 and 5.7.2. Also in Equation 6.8, T1 (SN R) is the exponential equation from scenario A and is dened in Equation 6.3. Tlossint is a new parameter and is the loss due to the interference from another operating wireless LAN system.
Ti (SN R, SIR) = T1 (SN R) Tlossint (SIR)
(6.8)
The basis for the Tlossint parameter of Equation 6.8 os that as the SIR at a user location decreases (i.e. as the interference power from another wireless LAN system increases) the throughput of the client will decrease. Thus, a linear dependence on the SIR is used to model Tlossint . This formulation is given in Equation 6.9.
Tlossint (SIR) = Ai SIR + Bi
(6.9)
Equation 6.9 is a linear t to the measured data. This t is accomplished using the parameters Ai and Bi . Ai is a slope in units of kbps/dB and Bi is an intercept point in units of kbps. Note that Ai is expected to be a negative value because the loss in throughput is inversely proportional to the users SIR.
6.5.3 Interference Model Parameters

For Equations 6.8 and 6.9, SIR and SN R are dB values, Tlossi nt , T1 and Ti are in kbps. The parameters for T1 (SN R) are those found to the optimum values from the data in
115
Scenario A. All parameters vary depending on the wireless LAN vendor, and may vary for different network applications. For Equations 6.8 and 6.9, it is important to note that SIR can be negative but still result in a non-negative throughput, Ti . This is reasonable because SIR is not based on the power which is passed through the wireless LAN despreader and bandwidth lter. Recall that the desired system was always used on channel 1, but the interference client was always operated on channel 3. Thus, nearly half of the power transmitted by the interfering system is not in the same frequency band as the desired system. In addition, the spreading codes used by the IEEE 802.11 wireless LANs are designed to have low cross-correlations. As a result, a wireless LAN could potentially operate despite a negative SIR value. The notion of a wireless LAN being able to operate despite the presence of a stronger interfering signal than the desired signal (i.e. a negative SIR) is somewhat counter-intuitive. However, this is because an easily predicted, shifted value for the SIR is intentionally used. By using an easily predicted model for SIR , rather than an SIR corrected (i.e. shifted) for cross-correlations, channel bandwidth overlaps, and other factors, it is easier to directly implement the resulting throughput prediction model in a software tool, such as SitePlanner, which can predict the received signal strength at a given location, but not the complexities involved in the cross-correlation of different IEEE 802.11 spreading codes.
6.5.4 Calculating the Interference Model Parameters.

The same optimization technique was used with Equations 6.8 and 6.9 of this model, as was used for the prior two models. The MATLAB nlint function was used to compute ideal values for the model parameters Ai and Bi . The values for Tmax , and SN R0 remain unchanged from the optimal values found from the data given in Scenario A. Note that in this case, a simple MMSE t [20] could have been used. However, the nlint is equally effective and in this case is guaranteed to nd a best t, rather than a local minima. The resulting values for Ai and Bi are given in Table 6.4. Additionally, the mean and standard deviations of the resulting error in the equations are given in Table 6.2. Figures 6.13 and 6.14 give an indication for how well the throughput prediction model matches the measured data. These gures show a best t line running across the diagonal and allow the comparison of the predicted and measured data points.
116
measured/predicted points ideal fit line Ti(SNR,SIR) = T1(SNR) (10.76* SIR +785.5) error: =0 =254.9
2000
Figure 6.13: WaveLAN Scenario C, One User with Interference, throughput prediction Model. This gure shows the predicted versus actual throughput results and an ideal t line
6.5.5 Constraining the One User with Interference Model

As was done for the Scenario B models, the Scenario C throughput prediction model uses Equations 6.8 and 6.9 and was constrained to force the case where zero SNR, and innite SIR results in zero throughput. This is conceptually the same as forcing SN R0 in Equation 6.3, which denes T1 (SN R), to be zero. This simplies the single user with interference, throughput prediction model to be fully specied by 4 parameters. To constrain the One User with Interference model of Equation 6.8, T1 (SN R) was congured to use the constrained exponential version of the One User, No Interference model. Then, the process of nding a best t for Ai and Bi was repeated to nd a constrained versions of the two
117
measured/predicted points ideal fit line Ti(SNR,SIR) = T1(SNR) (3.97* SIR +889.4) error: =0 =136.2
2000
Figure 6.14: 3Com Scenario C, One User with Interference, throughput prediction model. This gure shows the predicted versus actual throughput results and an ideal t line parameters. These new parameters are presented in Table 6.4 and in new plots comparing the measured and predicted versions of throughput. These plots are shown in Figures 6.15 and 6.16.
6.5.6 Analysis of the One User with Interference Model

The single user with interference model presented in Equations 6.8 and 6.9 benets from the fact that it is extremely intuitive and simple to compute. The single user interference model needs only two new parameters (i.e. Ai and Bi in Equation 6.8) and a single new variable (i.e. the SIR). However, despite only adding a single new variable compared to
118
Table 6.4: The Calculated, Ideal Values and Error Statistics for the Scenario C, Throughput Prediction Model, given in Equations 6.8 and 6.9 for One User with WLAN Interference.
Manufacture/Model WaveLAN WaveLAN, Constrained 3Com 3Com, Constrained Ai -10.76 kbps/dB -11.2 kbps/dB -3.97 kbps/dB -4.32 kbps/dB Bi 785.5 kbps 788.9 kbps 889.4 kbps 905.1 kbps 0 kbps 0 kbps 0 kbps 0 kbps 254.9 kbps 258.3 kbps 136.2 kbps 135.6 kbps
the single user model from Scenario A, it still provides a fairly accurate t to the measurement data. The accuracy could perhaps been better, but not signicantly so, even if several parameters were added to Equation 6.8 predict the throughput experienced by a user. Thus, the simple, powerful model of throughput presented in Equations 6.8 and 6.9 is extremely valuable for wireless LAN design and installations.
119
measured/predicted points constrained ideal fit line Ti(SNR,SIR) = T1(SNR) (11.2* SIR +788.9) error: =0 =258.3
2000
Figure 6.15: Constrained WaveLAN Scenario C, One User with Interference, throughput prediction Model. This gure shows the predicted versus actual throughput results and an ideal t line
120
measured/predicted points constrained ideal fit line Ti(SNR,SIR) = T1(SNR) (4.32* SIR +905.1) error: =0 =135.6
2000
Figure 6.16: Constrained 3Com Scenario C, One User with Interference, throughput prediction model. This gure shows the predicted versus actual throughput results and an ideal t line
6.6. PREDICTION MODELS FOR SCENARIO D
121
6.6
Prediction Models for Scenario D Two Users with Interference
Scenario D has been created as a synthesis of Scenarios B and C. By combining the models for throughput from Scenarios B and C, (given respectively in Equations 6.4 and 6.8) into a single model, a validation of the all models is possible. For this nal Scenario, no nonlinear curve tting was used. Instead, the models created form the work done in Scenarios B and C were combined into a single model presented in Equation 6.10 in which both the effect of interference and multiuser competition is taken into account as shown below in Equations 6.14 and 6.11. This section presents the work performed in comparing the non-optimized model with the measurements made for this Scenario D.
6.6.1 Creating the Two Client with Interference Model

In order to create the throughput prediction model for Scenario D, the Scenario B and Scenario C models were combined into a single model with a new offset value, Tof f set , which aligned the predicted and measured data. The nal equation models throughput for two clients with interference and is presented in Equation 6.10.
T = T1 (SN R1 ) Tlosscompete Tlosshidden Tlossint Tof f set
(6.10)
The Scenario D Model of Equation 6.10 is based on T1 (SN R1 ), which is the Scenario A Model given in Equation 6.3. Since there are two clients in this Scenario, SN R1 is the signal to noise ratio of the considered client, as opposed to SN R2 which is the signal to noise ratio of the client sharing the access point with the considered client. Note that both clients are desired, not inferring clients. Equation 6.10 is made up of all the Tloss parameters as the Scenario B and Scenario C equations combined. However, each of these factors have been slightly modied to keep the equations analogous to their original intent. For example, the Tlossequation has been modied, as can be seen in Equation 6.11, to use the same optimized parameter Acompete as was found in Scenario B. Acompete is still multiplied by the theoretical throughput of the other user. However, the other users theoretical throughput is now calculated using the
122
Scenario C equation which takes into account the effect of the interference on the other clients theoretical throughput.
Tlosscompete = Acompete (Tmax [T1 (SN R2 ) (Ai SIR2 + Bi )])
(6.11)
Tmax is also adjusted to be equal to the maximum theoretical throughput that will be experience in the interference environment, as is shown in Equation 6.12.
Tmax = Tmax (Ai max(SIR) + Bi );
(6.12)
No modications were made to the hidden terminal adjustment equation, Equation 6.13 nor the interference adjustment equation, Equation 6.14, which are both presented again here in Equations 6.13 and 6.14.
Tlosshidden = Ahidden P L(C1 , C2 )
(6.13)
Tlossint = (Ai SIR1 + Bi )
(6.14)
Lastly, T0 , the throughput intercept value was adjusted to align the measured and predicted results. The value of T0 as well as all other variable parameters are shown in Table 6.5. Note that the values of these parameters are the results of the optimization procedures carried out for Scenarios B and C. Thus, the values in Table 6.5 are identical to those used in Scenarios B and C as given in Tables 6.3 and 6.4 respectively.
6.6.2 Analysis of the Two User with Interference Model

The results of the modications to the Scenario B and Scenario C equations are illustrated in Figures 6.17, 6.18, 6.19, and 6.20. As can be seen from the graph, there is a quite good agreement in the predictions versus the measured throughput values. The mean error and error standard deviations of the resulting measurements are shown in Table 6.2 and conrm these results.
123
2000 1800 1600 actual throughput, kbps 1400 1200 1000 800 600 400 200 0 0 500 measured/predicted points ideal fit line error: =0.01 =268.7 1000 1500 predicted throughput, kbps 2000
Figure 6.17: WaveLAN Scenario D, 2 User with Interference, throughput prediction Model. This gure shows the predicted versus actual throughput results and an ideal t line
Table 6.5: Table of Parameter Values Used for the Scenario D Prediction Model for Equation 6.10
Parameter Acompete Ahidden Ai Bi T0 WaveLAN Value 0.359 3.48 -10.76 785.5 477.34 3Com Value 0.034 6.44 -3.97 889.4 165.37
124
2000 1800 1600 actual throughput, kbps 1400 1200 1000 800 600 400 200 0 0 500 1000 1500 predicted throughput, kbps 2000 measured/predicted points ideal fit line error: =0.23 =196.1
Figure 6.18: 3Com Scenario D, 2 User with Interference, throughput prediction model. This gure shows the predicted versus actual throughput results and an ideal t line
125
2000 1800 1600 actual throughput, kbps 1400 1200 1000 800 600 400 200 0 0 500 measured/predicted points ideal fit line error: =2.65 =253 1000 1500 predicted throughput, kbps 2000
Figure 6.19: WaveLAN Scenario D, Constrained 2 User with Interference, throughput prediction Model. This gure shows the predicted versus actual throughput results and an ideal t line
126
2000 1800 1600 actual throughput, kbps 1400 1200 1000 800 600 400 200 0 0 500 1000 1500 predicted throughput, kbps 2000 measured/predicted points ideal fit line error: =10.23 =195.7
Figure 6.20: 3Com Scenario D, Constrained 2 User with Interference, throughput prediction model. This gure shows the predicted versus actual throughput results and an ideal t line
6.7. PROCEDURE FOR OPTIMIZING MODELS IN A COMPUTER SIMULATION TOOL127
6.7
Procedure for Optimizing Models in a Computer Simulation Tool
In order to implement the above models in a computer simulation tool capable of making site-specic measurements in a full three dimensional model of an environment, the following steps would be followed: 1. Model the environment in which the wireless LAN will operate. Be sure to include the locations of access points. 2. Measure the throughput of the wireless LAN at a wide range of locations and with several different numbers of users. Measurements should be taken for several directions at each location. 3. Either measure or predict the RSSI at each measurement location. Ideally this step would be performed simultaneously with the previous step. 4. Perform spatial averaging of the all measurements 5. If RSSI measurements were made in step 3, optimize a throughput prediction model. Equation 5.1 can be used for this step. 6. Use the single client model and Equation 6.3 to calculate ideal values for Tmax , and SN R0 . 7. In step 2, if throughput measurements were made for multiple users and/or in the presence of other DSSS wireless LANs, optimize the parameters of Equations 6.8 and/or 6.4 respectively. 8. Use the models with the calculated parameters from steps 6 and 7 to predict the performance of an access point placed in a different location. 9. Repeat the previous step until the desired performance is found for a certain access point location. 10. Repeat steps 8 through 9 as often as needed for multiple access points. 11. Install the access points in the nal locations from step 9.
128
Note that some exibility in the above steps could easily be followed. However, the order given is intended for a more logical description.
6.8
Summary of Prediction Models
This chapter has presented several, easily simulated prediction models. Reviewing Table 6.2, the mean and standard deviations are all at levels that are quite reasonable for prediction purposes. As a result, it can be safely stated that the models presented here can provide a major help to non-wireless network engineers who do not comprehend wireless specications, but do understand network statistics.
Chapter 7 Predicting Throughput from a Building Model

7.1 Motivation for Predicting Throughput from a Building Model
As a nal verication of the entire concept of this thesis, throughput is predicted directly from a model of the wireless environment. This procedure reects how the models in this thesis are intended to be used. Additionally, carrying out a throughput prediction from the SitePlanner model of the environment will provide some verication of the measurement, modeling and predictions carried out throughout this thesis.
7.2
Prediction Technique
The prediction of throughput at a certain location in a modeled building environment is carried out in two steps. First, the signal strength at a given location is predicted. Next, the appropriate Scenario Throughput Prediction model is selected. The predicted signal strength is then used as an input to the throughput prediction model. Note that it should be expected that the mean and standard deviations of error should increase somewhat using this technique. This is because errors in throughput prediction are compounded 129
130
CHAPTER 7. PREDICTING THROUGHPUT FROM A BUILDING MODEL
with the errors in the RSSI prediction model. However, the error statistics should be similar for this technique as for the error statistics for the throughput prediction models based on the measured signal strength.
7.2.1 Signal Strength Prediction

Equation 5.1 is used in SitePlanner to predict the signal strength at each of the original measurement locations. These locations and the SitePlanner model of the environment are shown in Figure 5.25. The model parameters shown in Table 5.1 were used for these signal strength positions because they were found to be optimal values for predicting the path loss of wireless LAN signals in Section 5.7.1. This was done using Equation 5.1, but other prediction models were available for use in SitePlanner [26]. The signal strength has been predicted for each average location using omnidirectional antennas. Thus, the predicted signal strength values will be used with the locationally averaged throughput data.
7.2.2 Throughput Prediction from Predicted Signal Strength

The next step to predicting the throughput is to take the predicted signal strength at each location, and use it as an input to the throughput models for each of the four scenarios. Then, by comparing this predicted throughput to the measured throughput, error statistics can be calculated using the optimized models given in equations 6.3, 6.4, 6.8 and 6.10. Figures 7.1, 7.2, 7.3, 7.4, 7.5, 7.6, 7.7, 7.8 illustrate the results of this prediction process. Additionally, Table 7.1 presents a comparison of using the throughput prediction models with either a model of a building environment or using RSSI measurements.
7.3
Successful Prediction of Throughput
From Table 7.1 and the gures presented in this chapter, it is clear that the use of empirical models to predict site-specic, location based throughput is denitely feasible. As the results indicate, a non-technical networking professional could quite reasonably use the prediction models developed in this thesis to design wireless LAN networks.
7.3. SUCCESSFUL PREDICTION OF THROUGHPUT
131
2500
2000
actual throughput
1500
1000
500 measured/predicted data ideal fit line error: =75.09 =85.2 0 0 500 1000 1500 predicted throughput 2000 2500
Figure 7.1: Comparison of the t for Scenario A when throughput is predicted directly from a model of the building environment for the WaveLAN network.
132
2500
2000
actual throughput
1500
1000
500 measured/predicted data ideal fit line error: =8.79 =151.6 0 0 500 1000 1500 predicted throughput 2000 2500
Figure 7.2: Comparison of the t for Scenario A when throughput is predicted directly from a model of the building environment for the 3Com network.
133
2000 1800 1600 1400 actual throughput 1200 1000 800 600 400 200 0 0 500 measured/predicted data ideal fit line error: =95.06 =153.5 1000 1500 predicted throughput 2000 2500
Figure 7.3: Comparison of the t for Scenario B when throughput is predicted directly from a model of the building environment for the WaveLAN network.
134
2000 1800 1600 1400 actual throughput 1200 1000 800 600 400 200 0 0 500 measured/predicted data ideal fit line error: =4.11 =224.6 1000 predicted throughput 1500 2000
Figure 7.4: Comparison of the t for Scenario B when throughput is predicted directly from a model of the building environment for the 3Com network.
135
2000 1800 1600 1400 actual throughput 1200 1000 800 600 400 200 0 0 500 1000 predicted throughput 1500 2000 measured/predicted data ideal fit line error: =81.67 =256.6
Figure 7.5: Comparison of the t for Scenario C when throughput is predicted directly from a model of the building environment for the WaveLAN network.
136
Figure 7.6: Comparison of the t for Scenario C when throughput is predicted directly from a model of the building environment for the 3Com network.
137
Figure 7.7: Comparison of the t for Scenario D when throughput is predicted directly from a model of the building environment for the WaveLAN network.
138
2000 1800 1600 1400 actual throughput 1200 1000 800 600 400 200 0 0 500 1000 predicted throughput 1500 2000 measured/predicted data ideal fit line error: =4.45 =277.7
Figure 7.8: Comparison of the t for Scenario D when throughput is predicted directly from a model of the building environment for the 3Com network.
139
Table 7.1: Comparison of Throughput Model Error Statistics for Starting from a Building Model or from RSSI Measurements
Model Using Building Environment Mean Error Error Standard Deviation 75.09 kbps 85.2 kbps -8.79 kbps 151.6 kbps 81.67 kbps 256.6 kbps -15.67 kbps 146.8 kbps 95.06 kbps 153.5 kbps 4.11 kbps 224.6 kbps 108.66 kbps 377.1 kbps 4.45 kbps 277.7 kbps Using RSSI Measurements Mean Error Error Standard Deviation 0 kbps 34.6 kbps 0 kbps 147.3 kbps 0 kbps 185.2 kbps 0 kbps 225.2 kbps 0 kbps 254.9 kbps 0 kbps 136.2 kbps 0 kbps 268.7 kbps 0.2 kbps 196.1 kbps
WaveLAN Scenario A 3Com Scenario A WaveLAN Scenario B 3Com Scenario B WaveLAN Scenario C 3Com Scenario C WaveLAN Scenario D 3Com Scenario D
140
Chapter 8 Conclusions
8.1 Summary of Findings
This thesis has demonstrated that the network performance of a wireless LAN will vary with the site-specic location of a user, the number of users sharing the wireless LAN and the strength of interfering wireless LAN signals. How the performance of a wireless LAN varies is of critical importance to the professional designing a wireless LAN network. Simulation tools exist today which will allow anyone to predict the signal strength of a wireless LAN access point at any location, but this information is useless to most wireless LAN installers. This is because wireless LAN installers are typically networking professionals with little or no knowledge in regards to wireless packet networks. For this reason, this thesis has developed throughput prediction models for IEEE 802.11b wireless LANs. These models are designed to be intuitive and easy to implement in a simulation product, but still be accurate. This thesis has achieved this goal, as is outlined below.
8.1.1 Wireless LAN Measurement Software

This thesis has presented new wireless LAN measurement software products LANFielder and SiteSpy which allow consistent, non-optimistic measurements of the performance of wireless LAN networks. In addition, this patent pending software utilizes a three dimensional model of the measurement environment which allows a user to actually record the 141
142
CHAPTER 8. CONCLUSIONS
throughput in the modeled location of the precise physical location. The ability of LANFielder to store throughput measurements site-specically is key to being able to predict and adequately measure the performance of wireless LANs. The software is capable of measuring RSSI, throughput, packet error rates, and packet latencies on a location basis. As a result, the wireless LAN software developed as part of the thesis is capable of recording wireless LAN performance data in a site-specic manner. This is an important contribution to the wireless industry because it allows easy measurement of wireless LAN installations.
8.1.2 Wireless LAN Measurement Campaign

An extensive wireless LAN measurement campaign has been conducted as part of this thesis. The measurements have been extensively presented to provide installers with a concept of how a wireless LAN will perform in a variety of environments and usage scenarios. The measurements were conducted using the software presented in this thesis to allow validation of the software. The wireless data was measured in a variety of different environments and for two different IEEE 802.11b test-bed systems. Additionally, all measurements were performed for a variety of different typical usage scenarios. These scenarios were comprised of one and two user situations each with and without the presence of an interfering IEEE 802.11 DSSS system. The measurements presented show denite trends throughout the different scenarios. The thesis compared the throughput with the signal to noise ratio at each measurement location. In general this comparison revealed a constant throughput above a certain signal to noise ratio (SNR). Below this certain, critical SNR value, the measured throughput shows a steady decrease. Additionally, the measurement data has shown evidence in competition between multiple users causing a decrease in the measured throughput of wireless LAN users. The presence of interference also tended to cause the wireless LAN systems to drop to lower data transmission rates. These observations were used to to develop empirically based throughput prediction models.
8.2. FURTHER RESEARCH
143
8.1.3 Throughput Prediction Models

The throughput prediction models presented in this thesis have proven to be both effective at predicting the throughput based on just a model of a building environment or based upon RSSI measurements using wireless LAN card hardware. In the development of these models, linear, piece-wise ts to the data was rst considered. These models worked quite well, but were considered to be too complicated. Thus, exponential ts to data were instead used. These models proved to be extremely close in terms of accuracy, but were simpler to specify. To develop the throughput prediction models, the basic exponential model was rst t to the single user measurements. The resulting model was then used as the basis for further models for the two user, one user with interference and two user with interference models. To do this throughput loss factors were introduced. These loss factors were based on the expected causes for a loss in throughput: competition for shared resources, hidden terminal difculties, and the presence of strong interference resulting in transmission delays. Using these loss factors, a model for each of the measurement scenarios was created. These models were then t to the measured data. The accuracy of these models has been shown to be as good as zero mean error and an error standard deviation of 150 kbps, or as poor as a mean error of 26 kbps and a error standard deviation of 270 kbps, depending on the usage scenario. The most models, however, had zero mean error and error standard deviation of less than 200 kbps. These results imply the developed models t the measurements quite well. Since the models are also easily implemented in a computer-aided simulation environment, the models are an important contribution to wireless LAN system design.
8.2
Further Research
This work has opened up several areas of possible new research. For instance, the measurements and models presented in this thesis only cover some of the most basic usage patterns of wireless LANs. Additional measurements of wireless LANs in operation using the PCF modes, or having RTS/CTS or even WEP functionality enabled would be quite interesting to compare to the results presented in this thesis. Additionally, it would
144
CHAPTER 8. CONCLUSIONS
be useful to explore scenarios in which FHSS or high rate (i.e. IEEE 802.11b) DSSS wireless LANs or even Bluetooth or HomeRF modems were used as interference sources. Lastly, it would be interesting to measure scenarios in which a large number of users were all accessing the wireless LAN using different server computers and different usage levels. It should be noted that the data collected seems to indicate that the developed throughput prediction models seem to fair better for the WaveLAN equipment in the single client cases but better for the 3Com equipment in the interference free cases. This indicates that perhaps different models for different manufacturers would be potential area of research. This was however avoided in this thesis because a single, intuitive model for throughput prediction was desired. Regardless, the models presented in this thesis provide a useful groundwork for more advanced wireless LAN modeling.
Appendix A An Overview of the IEEE 802.11 and IEEE 802.11b Standards

A.1 Organization of the Standard
The IEEE 802.11b standard is broken into two main layers: the MAC or Media Access Control layer and the PHY or Physical Layer. These two layers allow a functional separation of the standard and, more importantly allows a single data protocol to be used with several different RF transmission techniques. Since the goal of this thesis is to predict the performance of IEEE 802.11b wireless LAN products, this chapter will present an overview of the DSSS function of the PHY layer and a basic description of the MAC layer.
A.2 Physical Layer

The PHY layer of the 802.11 standard denes the different RF transmission techniques. There are three basic transmission techniques: Frequency Hopping Spread Spectrum or FHSS, Direct Sequence Spread Spectrum or DSSS, and Diffuse Infrared. The relationship of these three standards is shown in gure A.1. Note that the diffuse infrared PHY access technique has received little attention and will be neglected in this thesis as it is not relevant to the research which has been conducted. 145
146
APPENDIX A. OVERVIEW OF IEEE 802.11 AND 802.11B
Figure A.1: A diagrammatic overview of the IEEE 802.11 standard as dened in [30] and [31].
A.2.1
DSSS Frequency Band and Channels
The two remaining PHY access techniques operate in in channels spread between 2.4 and 2.497 GHz. However, the FHSS technique has also been substantially less popular than the DSSS technique. This is mainly due to the higher bandwidth available to the DSSS implementation and the fact that the DSSS function lends itself better to interoperability between different implementations. For this reason, this thesis focuses on the more popular technique of DSSS used under the IEEE 802.11b standard. The specic channels available vary by country and the regulation agencies which controls the spectrum allocation. An description of the spectrum available in different countries is shown in Table A.1. The DSSS channels dened by the IEEE 802.11 standard for different countries are also shown in Figure A.2.
A.2.2
DSSS Advantages
The DSSS technique has two major advantages. It provides a spreading gain against narrowband interference signals and it spreads the transmitted signal across a wide range so the transmission resembles noise to a narrowband receiver. These two characteristics
A.2. PHYSICAL LAYER
147
Table A.1: World Wide Spectrum Allocation for IEEE 802.11 and 802.11b use as dened in [30] and [29].
Country United States Canada Japan France Spain Remainder of Europe Regulatory Agency FCC IC MKK Frequency Range Available 2.4 to 2.4835 GHz 2.4 to 2.4835 GHz 2.4 to 2.497 GHz 2.4465 to 2.4835 GHz 2.445 to 2.475 GHz 2.4 to 2.4835 GHz DSSS Channels Available 1 through 11 1 through 11 1 through 14 10 through 13 10 and 11 1 through 13 FHSS Channels Available 2 through 80 2 through 80 2 through 95 48 through 82 47 through 73 2 through 80
ETSI
are why DSSS was originally used by the military because it is difcult to jam and difcult to detect by narrowband radios. These two characteristics also make the DSSS technique ideal for coexisting with other narrowband users.
A.2.3
Data Bandwidths and Modulation Techniques
DSSS is currently a very popular transmission technique because it has the highest data bandwidth available. The DSSS transmission technique is dened in both the IEEE 802.11 and IEEE 802.11b standard. The 802.11b standard was introduced after the 802.11 standard to dene 2 different modulation techniques in addition to the two originally dened in the original 802.11 standard. The original 802.11 standard originally dened two data bandwidths (e.g. bit rates): 2 Mbps and 1 Mbps. These bandwidths use a Barker sequence described below. However, the data is modulated using DQPSK and DBPSK for the 2 Mbps and 1 Mbps bandwidths respectively. The IEEE 802.11b standard adds 5.5 Mbps and 11 Mbps data bandwidths to the original 2 Mbps and 1Mbps transmission techniques.
Switching Between Data Bandwidths A compliant IEEE 802.11b product can use any of the four transmission bandwidths. Ordinarily a wireless LAN card will operate at the highest possible bandwidth. However, as a user moves further away from an access point or if an interference source is present, the highest bandwidth may not provide reliable transmission of data. To combat this, the wireless LAN card will drop to a lower data transmission bandwidth. The lower rates are
148
Figure A.2: Illustration to scale of IEEE 802.11 DSSS Channels as dened in [30]. The channels are labeled by channel number and center frequency, in MHz. Note that channels 12 and 13 are not used in the USA or Canada even though they lie within the allocated frequency bands. Note also that channel 10 is dened for use in France even though it slightly exceeds the valid frequency band available in France. Lastly, also note that channel 14 is slightly offset from the other channels which all have 5 MHz spacings. All Channels have 22 MHz Bandwidths. more tolerant to noise and thus can be more reliable than the faster transmission rates. The IEEE 802.11 standard does not dene what criteria to use to decide which data transmission rate to use. The standard only requires that all compliant products support all data bandwidths for compatibility purposes. The next sections explain how the different data bandwidths are achieved.
A.2.4
1 and 2 Mbps Transmission Rates
Direct Sequence Spread Spectrum uses a PN spreading code to spread transmitted data over a wide bandwidth. This can be thought of as XORing a stream of data bits with a specic PN sequence. In the IEEE 802.11 standard, a single PN code is used by every user in the network. (See Section A.3.3 for information about multiple access techniques). This
A.2. PHYSICAL LAYER
149
PN code is the 11 bit barker sequence: +1 -1 +1 +1 -1 +1 +1 +1 -1 -1 -1. The technique of XORing data with the Barker sequence is shown in Figure A.3.
Figure A.3: Illustration of the process of spreading data using a Barker Sequence. Adapted from [36]. The gure shows how a one or a zero is transmitted as 11 bits of data represented by the original Barker sequence or the inverse of the Barker sequence. By combining data with a bandwidth at baseband of 1 MHz, spreading it and up-converting it to the desired 2.4 GHz channel results in an RF channel bandwidth of 22 MHz, as is shown in Figure A.4. The data is transmitted using either BPSK or QPSK encoding to provide the 1 Mbps and 2 Mbps transmission rates dened in the IEEE 802.11 standard.
A.2.5
5.5 Mbps and 11 Mbps Transmission Rates
The IEEE 802.11b standard denes two additional data bandwidths of 5.5 and 11 Mbps respectively. For these two transmission rates, the data is no longer spread using the Barker Sequence dened above. In order to increase the data rate to 5.5 and 11 Mbps, the IEEE 802.11b standard uses Complimentary Code Keying (CCK) or, optionally, Packet Binary Convolutional Coding (PBCC). Both techniques are discussed in the next two sections.
150
Figure A.4: Illustration of the basic IEEE 802.11 Spreading technique. Created using information from [30] and [31].
A.2.6
CCK Encoding
The IEEE 802.11b standard requires modems to support a technique known as Complementary Code Keying to simultaneous spread data across a 22 MHz channel while transmitting more data bits per 11 spread bits than the 1 or 2 bits transmitted in the plain IEEE 802.11 standard.
CCK Based 11 Mbps Transmission Technique The 11 Mbps version of CCK encoding works using an 8 chip spreading sequence. This can be done because the 8 bit sequence still runs at a rate of 11 Megachips per second, which results in a spreading factor of 11. CCK encoding, however, does not use a static PN code. It calculates a different spreading code based on the incoming data. This is done by breaking the incoming bits into symbols of 8 bits in duration. An 8 chip spreading code is found from the 8 data bits. Each chip is then encoded using the same DQPSK constellation and then transmitted. The 11 Mbps CCK technique of calculating the CCK spreading code can be broken down into two steps. First, 8 data bits are split into pairs called dibits. The four dibits are rst used to calculate four phase angles, 1 through 4 . The 2nd through 4th dibits are converted to a phase angle, using the mapping shown in Table A.2. The rst dibit is
A.2. PHYSICAL LAYER
151
converted to 1 using Equation A.1. That is, 1 (i) found as the value of 1 (i 1) for the previous symbol plus an offset angle found using Table A.2 plus 180 degrees if this is an odd symbol, 0 if it is an even symbol. In Equation A.1, i is the current symbol. At time zero, i starts at zero. This allows the mod(i, 2) portion of the formula to provide a 180 degree ( radian) shift to every odd numbered symbol. The of f set(1st dibit portion of the equation is simply a table look-up procedure based on the rst dibit, using Table A.2. Correct use of this table and Equation A.1 results in the rst of four phases, 1 . Table A.2: Mapping of dibits to Angles for CCK Modulation. Created using information from [31].
dibit 00 01 10 11 value or offset angle, radians 0 /2 3/2
1 (i) = 1 (i 1) + of f set(1st
dibit) + (mod(i, 2))
(A.1)
In the second step the four phases are used to calculate 8 complex chips using the mapping in Table A.3. The 8 complex chips are then mapped to the same QPSK constellation and transmitted. Note that the QPSK constellation is actually a DQPSK constellation. The value of 1 offsets all of the chips by the same angle, which is an offset to the QPSK constellation used for the previous set of 8 complex chips.
CCK Based 5.5 Mbps Transmission Technique The 5.5 Mbps version of CCK modulation is carried out in the same manner that the 11 Mbps version is carried out, except that only four bits are encoded per symbol instead of eight. Since, there are only four bits per symbol, the values are calculated differently. In this case the rst dibit is still used to encode 1 in exactly the same manner as before using Equation A.1. The remaining values are calculated based on the 3rd and 4th bits. The formulas for these calculations are given in Equations A.2, A.3 and A.4. Once the values have been calculated, the chips are calculated and mapped to the QPSK
152
Table A.3: Mapping of Angles into Complex Chips for CCK Modulation. Created using information from [31].
Chip Number 0 1 2 3 4 5 6 7 Formula ej(1 +2 +3 +4) ej(1 +3 +4) ej(1 +2 +4) ej(1 +4) ej(1 +2 +3 ) ej(1 +3 ) ej(1 +2 ) ej1
constellation exactly as was done for the 11 Mbps version of CCK. 2 = (3rd
bit) + /2 3 = 0
(A.2) (A.3) (A.4)
4 = (4th
bit)
The above process can be thought of in a slightly different manner. Instead of calculating the chips as above, the chips can be thought of as being calculated using the equations shown in Table A.3 modied to not include 1 . That is, each chip would be a function of just 2 , 3 , and 4 . Then, instead of encoding each chip using QPSK, the entire symbol is encoded using a DQPSK encoding based on the value of 1 as a phase difference from the previously transmitted symbol. Then the symbol is spread using the CCK chips calculated using the modied equations. This alternative way of considering the process is show in block diagram form in Figure A.5.
CCK Encoding Example To illustrate the CCK spreading process, the sequence 00011011 as an odd symbol in which the phase of the previous symbol was pi/2 will be encoded. The resulting dibits are 00, 01, 10 and 11. The phases would rst be calculated to be 1 = /2, 2 = /2, 3 = , 4 = 3/2. Next, the 8 complex chips would be calculated to be e3j/2 , ej , ej/2 , e0 = ej ,
A.2. PHYSICAL LAYER
153
Figure A.5: Conceptual block diagram of the CCK encoding process carried out by the transmitter. Created using information from [31]. e0 , e3j/2 , ej = e0 , and ej/2 . Each chip would be encoded on the QPSK constellation based on the phase of each of the chips.
A.2.7
PBCC Encoding for 11 and 5.5 Mbps Transmission Rates
The IEEE 802.11b standard also supports an optional technique known as Packet Binary Convolutional Coding (PBCC) to achieve the 5.5 and 11 Mbps data rates. The PBCC encoding technique uses a standard 1/2 rate, 64 state, rate code. The PBCC technique feeds the data bits into the 1/2 rate encoder. The 1/2 rate encoder, by denition, generates 2 output bits for each input data bit. The output of the encoder is mapped to a QPSK constellation for the 11 Mbps data rate and to a BPSK constellation for the 5.5 Mbps data rate. To provide some pseudo-randomness to this technique a pseudo-random cover code is used to vary the QPSK or BPSK constellation used. A block diagram of this technique is shown in Figure A.6 A 256 bit cover sequence is used to vary the QPSK or BPSK constellation used. The cover sequence is generated by taking the 16 bit sequence: 0011 0011 1000 1011 and rotating it 3
154
Figure A.6: A block diagram of the PBCC encoding technique. Created using information from [31]. bits to the left 15 times to generate the 256 bit sequence made up of 16 sequences of 16 bits each. Thus the 17th through 32nd bits of the full sequence are 1001 1100 0101 1001. The 33rd through 48th bits are 1110 0010 1100 1100. This 256 bit sequence is used repeatedly to vary the constellation used to transmit each chip. Specically, if the cover sequence is a 0, one constellation is used. If the cover sequence is a 1, then the constellation is rotated by +pi/2 and used. After the end of the 256 bit sequence, the sequence is repeated. Note that no spreading takes place in this technique. Rather, the data is encoded directly at the desired data rate (either 11 Mbps or 5.5 Mbps). However, the use of the cover sequence will act to randomly distribute the data transmission across the full 22 MHz channel bandwidth and the spreading can be thought of as occurring inside the QPSK or BPSK encoder.
A.2.8
PHY Layer Packet Format
The IEEE 802.11b standard denes two different packet structures that are used in the DSSS standard. There is a short and a long packet format as shown in gure A.7. The
A.3. MAC LAYER
155
Figure A.7: Basic structure of the IEEE 802.11b packet as its transmitted on the physical layer is shown here. Adapted from [31]. short packet format is intended to reduce the overhead of transmissions while the long packet format is to maintain compatibility with IEEE 802.11 networks. The PHY preamble is used to allow the receiver to get synchronized to the transmitter. The PHY header is the overhead needed by the PHY layer. The remainder of the packet contains the data passed to the PHY layer by the MAC layer as is shown in A.8. Note that different parts of the packet are transmitted at different transmission rates.
A.3 MAC Layer

A.3.1 Basic Network Layout
The IEEE 802.11 standard denes two types of networks: Adhoc and Infrastructure. Adhoc networks are self-conguring networks between mobile and portable wireless clients. Infrastructure networks use xed, interconnected access points to provide connectivity to mobile and portable wireless clients. This thesis will focus on infrastructure networks. Since these networks use xed location access points, it is import to carefully select the locations of these access points. This is the main motivation for this thesis. Infrastructure based wireless networks need to provide some level of service, either in terms of coverage area or in network performance, or in both. However, in order to carefully place
156
the access points which make up a wireless LAN infrastructure, design rules are needed. Prediction models which can be used to design wireless LAN infrastructure networks are presented later in this thesis.
A.3.2
MAC Layer Packet Structure
The basic format of packets passed to the PHY layer from the MAC layer is shown in gure A.8. Note that this is the basic format for all packets sent by the MAC layer. Some actual packets do not actually contain all of the elds. However, all elds are present in all data packets. Up to four addresses are needed because it is sometimes necessary to identify the address of the access point used by the transmitter or receiver. Thus, if two wireless LAN users are sending packets to one another but each is using a different access point, the 802.11 MAC address of both access points and both clients will be present in the four address elds.
Figure A.8: The structure of packet created at the MAC Layer. Adapted from [30].
A.3.3
Multiple Access, DCF and CSMA/CA
Regardless of the physical layer used, all IEEE 802.11 wireless LAN clients use the same channel to transmit on. This means the standard needs to dene a way in which clients know when they can transmit and when they can not. This is handled using several multiple access mechanisms. The most basic of these is the Carrier Sense Multiple Access with Carrier Avoidance (CSMA/CA) mechanism. This mechanism is dened as part of the Distributed Coordination Function (DCF) of the IEEE 802.11b standard. The DCF is the mandatory method by which clients work together and differ access to the medium so that the all users can use the same wireless channel. CSMA/CA is based on the multiple access technique used in wired Ethernet connections, Carrier Sense Multiple Access with Collision Detection, CSMA/CD. In both types
A.3. MAC LAYER Table A.4: Different Interframe Spacings (IFS) as dened in [30].
Abbreviation SIFS PIFS DIFS EIFS Meaning Short Interframe Spacing Point Coordination Function (PCF) Interframe Spacing Distributed Coordination Function (DCF) Interframe Spacing Extended Interframe Spacing
157
of CSMA users rst sense the transmission medium to see if anyone is transmitting just before transmitting a packet of data. This only partially avoids the possibility of packets being transmitted by two users at the same time. When two or more packets are transmitted simultaneously, or overlapping in time, a collision is said to have taken place. In wired Ethernet connections a user is able to detect when a collision has taken place because a network card is setup to be able to transmit and receive on different physical wires that make up the actual Ethernet cable. This is not possible in wireless Ethernet because when a wireless LAN card is transmitting it can not listen to detect if packets collide. To partially cope with the inability to detect a collision, the IEEE 802.11 standard attempts to avoid collisions using carefully designed waiting periods that allow multiple users to defer access to the shared wireless channel to one another. That is, IEEE 802.11 clients will always ensure a channel has been idle for a certain period of time before transmitting. The process of deciding how long to wait as governed by the basic DCF is illustrated via a owchart in Figure A.9. The following presents a basic overview of how the DCF progresses. 1. In Figure A.9, ow begins at the top left portion of the diagram. When a data packet is ready for transmission, a client will rst sense the medium. If it is idle and remains so for a period of time know as the Interframe Spacing (IFS) period, the packet can be immediately transmitted. A standard, unicast packet (called a directed MSDU or MAC Service Data Unit) needs to be acknowledged by the receiver by a short ACK packet. If this transmission fails by the ACK not being received, the client enters the same defer state as if the medium was initially detected to be busy during the IFS period.
158
Figure A.9: Flow chart of the process of sending data packets under the CSMA/CA based Distributed Coordination Function (DCF). Created using information in [30]. Note that their are four different types of IFS frame spacings. These are shown in Table A.4. The shorter IFS frame spacings are used for higher priority transmissions and to ensure certain events like acknowledgments (ACKs) will occur before another packet transmission. 2. If the medium was not idle or the transmission fails, the client must defer until the medium is free. This is done using a special timer known as the Network Allocation Vector (NAV). To set this timer the client reads a eld in the header of the packet currently being transmitted that tells the client how long the current user will continue to use the medium, through the current transmission, or through the current transmission and immediate transmissions after the current transmission. In this way the client does not need to continue to sense the state of the channel until the NAV timer has expired.
A.3. MAC LAYER
159
3. After the NAV timer has expired or the client has sensed that the channel is no longer busy, the client will calculate a backoff interval. This backoff interval is a uniform random number. This number is chosen from the interval between 0 and the value of the Contention Window (CW) inclusive. The contention window is initially set to be equal to CWmin which is a value dened by the PHY layer. 4. After calculating the value of the backoff interval to use, the senses the channel for an IFS period. If the channel is idle at the end of this period the client will set a backoff timer equal to the value of the backoff period calculated previously. This timer is periodically decremented while the channel continues to stay idle. If the channel becomes busy either during the IFS period or during while the backoff counter is being decremented but before it reaches zero, the client goes into a defer state without changing the value of the backoff timer. 5. If the client goes back into a defer state it goes through the same process as before in which the client waits for the medium to become idle by rst sensing the medium and then by setting the NAV timer. When the medium returns to an idle state, the client must wait for an additional IFS period before continuing to decrement the backoff counter. Note that the backoff counter does not get reset to the initially calculated backoff interval each time the client goes into the Defer state. Rather, the backoff counter is decremented whenever the medium has been idle for at least an IFS period. 6. When the medium has been idle for an IFS period and the backoff counter reaches zero, the client will transmit its data. 7. If the client discovers that the transmission has failed then the client must exponentially increase the value of CW using equation A.5. As a result, if the medium is very busy, exponential increases in the maximum backoff delay will occur and the probability of packet collisions will decrease. After increasing CW, the client generates a new value for the backoff interval and re-senses the state of the channel. 8. After either of the two transmit states have been completed successfully (by having been properly acknowledged by the receiver using an ACK packet), several things happen. First, the value of CW is reset to CWmin after successful transmission occurs. Second, the client goes through a mandatory backoff interval in which the
160
APPENDIX A. OVERVIEW OF IEEE 802.11 AND 802.11B state of the medium is ignored. The client then goes back to the initial state in which the client waits for data to be ready for transmission.
CWnew = min(2 (CWold + 1) 1, CWmax )
(A.5)
A.3.4
RTS/CTS and the Hidden Terminal Problem
The DCF implementation of IEEE 802.11 does not handle a problem referred to as the hidden terminal problem. This problem occurs when a mutual receiver is in range of two transmitters which are not in range of one another. In this case attempting to detect if the medium is free does not necessarily work because the two transmitters can not detect one anothers transmissions. Thus the packets from the two transmitters will collide at the common receiver. To combat this problem, IEEE 802.11 adds an optional RTS/CTS mechanism. In this technique instead of transmitting a data packet after waiting for a free medium, a client will transmit a short Ready To Send (RTS) packet to request the use of the medium. If this succeeds, the receiver will quickly (after a SIFS period) reply with a short Clear To Send (CTS). After the successful exchange of an RTS/CTS pair the actual transmission takes place. This method allows hidden terminals to hear either a CTS or an RTS packet and know to differ access using the NAV functionality described previously. It also means that if packets do collide only a short RTS or CTS packet is lost rather than a long data packet. It is important to note though that this functional is optional to include and is enabled in one of three modes: always on, always off or on for packet sizes above a certain threshold.
A.3.5
Additional Optional Provisions of the MAC Layer
The Point Coordination Function Another optional protocol that is part of the IEEE 802.11 standard is the Point Coordination Function (PCF). This function allows time critical or delay sensitive packets to be given priority over regular data transmissions. The PCF uses a polling procedure to setup a contention free period which takes priority over the DCF procedure. During the PCF
A.3. MAC LAYER
161
established contention free period, a single host poles clients and allows them to transmit. In this way delay sensitive packets such as voice or video can be given priority over other data.
Wired Equivalent Privacy The Wired Equivalent Protocol (WEP) is intended to provide a simple layer of protection for wireless network connections. By their very nature, wireless networks are easy to connect to and be eavesdropped on. WEP is a single shared key system in which a basic 40 bit encryption is applied to packet transmissions on the network. Without knowledge of the WEP key, packets can not easily be decoded by an unauthorized user. The use of WEP though tends to slow down transmissions and increases the overhead of packet transmissions, thereby lowering the bandwidth available. In addition, several security aws have been found in the technique [3].
162
Bibliography
[1] Bianchi, G.;Fratta, L.; Oliveri, M. Performance Evaluation and Enhancement of the CSMA/CA MAC Protocol For 802.11, Seventh IEEE International Symposium on Personal, Indoor and Mobile Radio Communications, Proceedings, (PIMRC 96), Vol. 2, p. 392-396, 1996. [2] Bing, B. Measured Performance of the IEEE 802.11 Wireless LAN, Local Computer Networks, 1999. Conference on. (LCN 99). p. 34-42. 1999. [3] Borisov, N.; Goldberg, I; Wagner, D. (In)Security of the WEP algorithm. http://www.isaac.cs.berkeley.edu/isaac/wep-faq.html Viewed: July 4, 2001 [4] Chen, S-A; Lee, Y-H; Yen, R.Y.; Zheng, Y-J; Ko, C-H; Sheu, S-T; Chen, M-H. Optimal prediction tool for wireless LAN using genetic algorithm and neural network concept. Communications, 1999. APCC/OECC 99. Fifth Asia-Pacic Conference on ... and Fourth Optoelectronics and Communications Conference, Vol. 1, p. 786 -789, 1999. [5] Demir, T.; Komar, C.; Ersoy, C. Measured Performance of an IEEE 802.11 Wireless LAN, Proceedings of the Fifteenth International Symposium on Computer and Information Sciences, Istanbul, Turkey. p. 246-254, Oct 2000. [6] Duchamp, D.,; Reynolds, N. F., Measured Performance of a Wireless LAN, Local Computer Networks, 1992. Proceedings., 17th Conference on. p. 494-499, 1992. [7] Durgin, G.; Rappaport, T. S.; Xu, H., Measurements and Models for Radio Path Loss and Penetration in and Around Homes and Trees at 5.85 GHz, IEEE Transactions on Communications, Vol. 46, No. 11, November 1998, pp. 1484-1496. [8] Feigin, J.; Pahlavan, K. Measurement Of Characteristics Of Voice Over IP In A Wireless LAN Environment, IEEE International Workshop on Mobile Multimedia Communications, 1999. (MoMuC 99), p236-240. [9] Hope, M.; Linge, N., Determining the Propagation Range of IEEE 802.11 Radio LANs for Outdoor Applications, Local Computer Networks, 1999, Conference on. (LCN 99). p. 49-50, 1999. 163
164
BIBLIOGRAPHY
[10] Kamerman, A.; Aben, G. Throughput performance of wireless LANs operating at 2.4 and 5 GHz, Personal, Indoor and Mobile Radio Communications, 2000, The 11th IEEE International Symposium on, (PIMRC 2000), Vol 1, p. 190-195, 2000. [11] Kim, S-C.; Bertoni, H. L.; Stern, M. Pulse Propagation Characteristics At 2.4 GHz Inside Buildings, IEEE Transactions on Vehicular Technology, Vol. 45, Iss. 3, p. 579-592, Aug. 1996. [12] Khurana, S.; Kahol, A.; Jayasumana, A. P. Effect of Hidden Terminals on the Performance of IEEE 802.11 MAC Protocol, IEEE Conference on Local Computer Networks Proceedings, Piscataway, NJ, p. 12-20, 1998. [13] Leskaroski, D; Mikael, W. B. Frequency Planning and Adjacent Channel Interference in a DSSS Wireless Local Area Network, Wireless Personal Communications: Bluetooth Tutorial and Other Technologies, p. 169-180, 2001. [14] Maeda, Y.; Takaya, K.; Kuwabara, N. Experimental Investigation Of Controlling Coverage Of Wireless LAN By Using Partitions With Absorbing Board, Electromagnetic Compatibility, 1999, International Symposium on, p. 674 -677, 1999. [15] Maeda, Y.; Takaya, K.; Kuwabara, N. Experimental Investigation of Propagation Characteristics of 2.4 GHz ISM-Band Wireless LAN in Various Indoor Environments, IEICE Transactions in Communications, Vol. E82-B, No. 10, Oct 1999. [16] MATLAB 6.0 Product Documentation. Supplied with the MATLAB Product from Mathworks, Inc. 2001. Available at: http://www.mathworks.com/access/helpdesk/help/helpdesk.shtml. [17] Messier, A.; Robinson; Pahlavan, K. Performance Monitoring of a Campus Area Network. 22nd Annual Conference on Local Computer Networks, Proceedings, p. 232238, 1997. [18] Motley, A. J.; Keenan, J. M. P. Personal Communication Radio Coverage in Buildings at 900 MHz and 1700 MHz, Electronics Letters. Vol 4, Issue 12, p. 763-764, Jun 1998. [19] http://www.netiq.com/Products/Network Performance/Chariot/Default.asp. NetIQ Products - Chariot Downloaded on: June 14, 2001. [20] Papoulis, A. Probability, Random Variables, and Stochastic Processes. 3rd Ed. New York: WCB/McGraw-Hill, 1991. [21] Peterson, L. L.; Davie B. S. Computer Networks, A Systems Approach. 2nd Ed. San Francisco, California: Morgan Kaufmann, 2000.
BIBLIOGRAPHY
165
[22] Prasad, A.R.; Prasad, N.R.; Kamerman, A.; Moelard, H.; Eikelenboom, A. Indoor Wireless LANs Deployment Vehicular Technology Conference Proceedings, 2000. IEEE 51st, (VTC 2000-Spring Tokyo.), Vol. 2, p. 1562-1566, 2000. [23] Prasad, N. R. IEEE 802.11 System Design Personal Wireless Communications, 2000, IEEE International conference on, p. 490-494, 2000. [24] Rappaport, T. S. Wireless Communications, Principles and Practice. 7th Ed. Upper Saddle River, NJ: Prentice-Hall, 1996. [25] Seidel, S. Y., Rappaport, T. S., 914 MHz Path Loss Prediction Models for Indoor Wireless Communications in Multioored Buildings. IEEE Transactions on Antennas and Propagation, Vol. 40, No 2, p.207-217, Feb 1992. [26] SitePlanner 2000 Product Documentation. Supplied with the SitePlanner Product from Wireless Valley Communications, Inc. 2000. [27] Takaya, K.; Maeda, Y.; Experimental And Theoretical Evaluation Of Interference Characteristics Between 2.4-GHz ISM-Band Wireless LANs, IEEE International Symposium on Electromagnetic Compatibility, 1998, Vol. 1, p. 80-85, 1998. [28] Tarng, J. H.; Liu, T. R. Effective Models in Evaluating Radio Coverage on Single Floors of Multioor Buildings, IEEE Transactions on Vehicular Technology, Vol. 48, No. 3, May 1999. [29] WaveLAN Access Point 500 and ORiNOCO Gold PCMCIA Card pDocumentation. 2001. Available at: http://www.wavelan.com/template.html?section=m116&envelope=205 [30] Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY) Specications IEEE 802.11 Standard, 1999. [31] Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY) Specications: Higher Speed Physical Layer Extension in the 2.4 GHz Band, IEEE 802.11b Standard, 1999. [32] Xylomenos, G.; Polyzos, G. C. Internet Protocol Performance Over Networks With Wireless Links, IEEE Network, Vol. 13, Iss 4, p. 55-63, July-Aug 1999. [33] Xylomenos, G.; Polyzos, G. C. TCP And UDP Performance Over A Wireless LAN, Eighteenth Annual Joint Conference of the IEEE Computer and Communications Societies, Proceedings. (INFOCOM 99), Vol. 2, p. 439-446, 1999. [34] Xylomenos, G.; Polyzos, G. C.; Mahonen, P.; Saaranen, M. TCP Performance Issues Over Wireless Links, IEEE Communications Magazine, Vol. 39, No. 4, p. 52-58, April 2000.
166
BIBLIOGRAPHY
[35] Yang, S-H.; Lee, Y-H.; Yen, R.Y.; Zheng, Y-J; Shue, S-T; Ko, C-H; Chen; M-H. A Wireless LAN Measurement Method Based on RSSI and FER, Fifth Asia-Pacic Conference on Communications and Fourth Optoelectronics and Communications Conference, Proceedings, (APCC/OECC 99), Vol. 1, p. 821-824, 1999. [36] Zyren, J; Petrick, A. Brief Tutorial on IEEE 802.11 Wireless LANs, Intersil Application Note 9829 (AN9829). http://www.intersil.com/data/an/an9/an9829/an9829.pdf Downloaded: Fall 1999, Written: Feb 1999.
Vitae
Benjamin E. Henty Benjamin Henty began his college career as an Electrical and Computer Engineering student at Carnegie Mellon University as an Andrew Carnegie Scholar. He attended Carnegie Mellon from August 1995 through May of 1999, at which time he graduated with University and College Honors. At Carnegie Mellon, Benjamin performed research as an undergraduate in the area of IEEE 802.11 wireless LAN design. The research work involved investigating the feasibility and performance of using heating and air conditioning ventilation ducts for distributing wireless LAN signals throughout a building. This research included a large number of RF and network measurements as well as antenna design work. This work was presented in a conference paper at the 8th Annual Virginia Tech Symposium on Wireless Personal Communications in 1998 and at the Carnegie Mellon Meeting of the minds in 1999. While at Carnegie Mellon, Benjamin worked as a sound engineer as part of the student activities program. From April 1998 through April 1999 he was the head supervisor for the Activities Board Technical Committee (AB Tech). The responsibilities of this position included directing 15 students in the safe use of professional sound and light systems for campus events. These events varied between national label entertainers to small student group presentations. In this position, Benjamin worked with Carnegie Mellon organizations, outside artists, and agents to organize technical details of over 150 events, or an average of one show every two days. In addition, Benjamin oversaw the repair and maintenance of all sound, light and power equipment and was scally responsible for a $35,000 budget. Benjamin received a Student Leadership Award for his work in this position. Benjamin worked as an RF Engineering intern at Daimler Chrysler Rail Systems (ADTranz) in the summer of 1999. At Daimlet Chrysler, Benjamin developed and performed several test specications on various different wireless modems. Benjamin analyzed and measured the performance of several wireless data modems, microwave and ultrasonic train detection sensors, and voice radios. These measurements were performed at the companies test track and on the Pittsburgh Airport people mover system. All results 167
168
VITAE
were written up in internal test specications and presented to other engineers at Daimler Chrysler. Benjamin began his masters work at Virginia Tech University in August of 1999 as a Bradley Fellow. Benjamin again performed research with IEEE 802.11 and IEEE 802.11b wireless LANs. This research involved using simulation tools to design indoor wireless LAN networks. The work was used as the basis for his masters thesis. The goal of his research was to develop throughput prediction models that would simplify computeraided design of wireless LAN systems. This research was performed at the Mobile and Portable Radio Research Group (MPRG) of Virginia Tech under the supervision of Dr. Theodore S. Rappaport. Benjamin became a Bradley Industrial Fellow in June of 2000 when he began to work at a Virginia Tech spinoff company, Wireless Valley Communications, Inc. The company was a start-up company consisting of approximatly 10 people. At Wireless Valley, Benjamin worked as a Wireless Network Engineer. Benjamin was the product manager of the companies wireless LAN measurement and design software tools, LANFielder and SiteSpy. These products allow location based, site-specic measurement of wireless LAN throughput, and signal strength statistics. In addition to developing, supporting and marketing these products, Benjamin supported the companies propagation channel simulation tools, SIRCIM and SMRCIM. Further, Benjamin submitted a patent, performed consulting work, made customer visits, attended conferences and gave presentations and talks on wireless LAN system design, while working at Wireless Valley. Benjamin left Wireless Valley at the end of April 2001 to complete the remainder of his Masters Degree at Virginia Tech University. This he did in August of 2001.
The Application of GTD and Ray Launching Techniques to Channel Modelling for Cordless Radio Systems
M.C. Lawton and J.P. McGeehan University of Bristol Centre for Communications Research Queens Building, University Walk Bristol BS8 lTR, United Kingdom Tel: +44 272 303727, Fax: +44 272 255265
Abstract
Propagation characteristics play a fundamental role in the design and implementation of radio systems. The application of broadband digital data services within the cordless environment requires close consideration of the dispersive nature of radio channels. A prediction algorithm is presented such that propagation characteristics can be estimated for small cell high data rate systems. Through the use of Geometric Optics and Geometric Theory of Diffraction (GTD) the algorithm performs ray launching techniques in order to evaluate reflected, transmitted, and diffracted rays from a simplified description of a given environment. Both modelled and measured results arc presented demonstrating the models ability to predict typical RMS delay spread values.
now becoming an increasingly attractive proposition for wideband modelling given the advances in computer technology and the trend towards smaller cells.
The Propagation Model
The equation used to describe the channel was first proposed by Turin [2] and has subsequently been used by several authors [3, 41. It takes the form of a bandlimited complex impulse response, h(t), given by:-
1 Introduction
The last decade has seen an unprecedented growth in the use of mobile and cordless radio systems, offering primarily voice communication. However, increasingly, work is now being directed towards digital data services. In addition, the success of mobile telephony has resulted in considerable interest in incorporating spectrally efficient small cell systems. For such systems the extent of the time dispersion and hence intersymbol interference provided by the channel becomes an important criterion limiting the data rate of the system. This paper considers the propagation characteristics of small cell high data rate systems. Rather than use a simulation approach an analytical ray launching technique has been developed. Ray launching techniques were initially used to model narrowband fading statistics by Gladstone and McGeehan [l], in both single and multi-transmitter mobile radio systems, and are
Here the transmitted pulse is mathematically described by a Dirac function and the received signal, h ( t ) ,is formed from the addition of waves from several time delayed paths each represented by an attenuated and phase shifted DiriLC waveform. The method adopted in this paper directly seeks to evaluate suitable path parameters for the amplitude & k r arrival time T k , and the arrival phase @ k , such that the propagation characteristics can be predicted. This calculation uses geometric optics and the geometric theory of diffraction.
2.1
Geometric Optics
The algorithm uses geometric optics in order to calculate analytically all reflected and transmitted paths up to and including seven reflections for an outdoor picocell application. This typically results in excess of one hundred and fifty rays. Conceptually the algorithm works by reflecting the entire room plus receiver about a chosen reflecting wall.
125
0-7803-0673-2192$3.00 1992 IEEE
A line is then drawn between the transmitter and receiver image . If this is intersected by the reflecting wall and no other walls then a purely reflected path is said to exist. The value of Tk for this path is found from the s length of the line and the speed of light. (Yk i found by using a square law attenuation of signal power with distance and incorporating a reflection loss. @k can be evaluated from the path length and knowledge of the carrier frequency with allowances being made for phase transitions resulting from reflections or diffraction.
In the event of the line between transmitter and reflected receiver being intersected by more than the intersecting plane then a path is said to exist but the wave is attenuated further by passing through a blocking wall or walls. The basic idea for the algorithm can be extended to include any number of reflections. The reflection and transmission characteristics have been evaluated as a function of arriving angle for a range of different wall materials .The algorithm has also been made more realistic by giving both transmitting and receiving antenna gain patterns. Both antennas were modelled as vertically polarised half-wave dipoles . Previous work has been reported by the authors [5) where the model was used for the prediction of radio channel characteristics for indoor cells. In such cells it was necessary to have a 3-D model such that multiple ceiling-floor reflections could be modelled. For the outdoor application described in this paper it was decided to adapt the model for a 2-D environmental description. In doing this an assumption is made that the transmitter height is below roof top height and therefore roof top dimensions need not be inputted. However whilst all the reflectors are described in terms of z and y the ground is still represented within the model. For each path found an additional path corresponding to the same path with an extra ground reflection is added. Hence receiver and transmitter heights are still needed as inputs to the model. This simplification has resulted in a reduction in the maximum possible number of purely reflected rays for a given number of reflections within a simple empty environment. However the increased simplicity of the algorithim allows the program to calculate paths up to seven reflections as opposed to maximum of five for the indoor model.
and as such has limitations. For a non-LOS (line of sight) scenario geometric optics suffers from its exclusion of the diffracted ray paths. It was felt that in the frequency band considered here (centre frequency 1.845 GHz) the effect of the diffracted rays may well be significant and thus warranted investigation. GTD owes its origins to Keller's attempts to improve on geometric optics and has been designed as an extension to a geometric optics approach. Thus the two methods can elegantly be combined to provide a single prediction tool. The formulation of diffracted paths is similar to that of reflected paths in geometric optics except that the reflection coefficent is replaced with a diffraction coefficent and alterations made to the spatial attenuation factor. Figure 1 shows the geometry for diffraction by a curved edge. The diffracted field, for either soft (perpendicular) or hard (parallel) polarization, resulting from a finitely conducting wedge can be calculated without singularity problems at the incidence and reflection boundaries
where the h s n e l transition function,
(3)
and
L=
ss'sin2& s st
(4) k = dJ*&
(5)
gf(k) = 2c0s2 2nnNf
- k),
In equation ( 5 ) , N* are the integers which most closely satisfy the equations :-
2.2
Geometric Theory of Diffraction
The application of geometric optics provides a simple and effective method for predicting the behaviour of radio channels. The method is however approximate
2nnN-
- (k) = -n
(7)
126
R, and Rn'L are the reflection coefficients for the perpendicular and parallel polarization for the 0 face, incident angle d', and for t,he n face, reflection angle nr - 4 (figure 1).
' I
231 ..
Site 1 (see figure 2)
Oblique incidence
Observation
Yt
This was the smaller site and was chosen because of its simplicity for description within the model and because it represented a confined linear cell. The transmitter was placed at the end of a roadway which provided the entrance to a large 'U' shaped building surrounding three of the entrance's four sides. The area behind the transmitter is a car park which offered many potential scatterers. Measurement of both LOS and obsructed (OBS) paths were taken. The environmental description within the model is also indicated on the map. Principal reflectors, diffraction points, receiver locations are all shown. The outer cell dimensions have been bounded by reflecting planes represented by thick solid lines. This was done because it was felt that for these antenna heights (Tx = 2.5m , Rx = 1.5m) it was more likely for dissipating rays to encounter obstacles than continue in free space.
Plan view of wedge
Figure 1: Geometry and co-ordinates for description of diffracted path at a wedge Although the mathematical expressions for the diffraction terms look somewhat cumbersome and difficult t o manipulate they can quite readily be evaluated by a computer. Calculation of the diffracted fields does however add further complexity t o the algorithm and it was felt neccessary to limit the computation performed. Consequently the algorithm described evaluates the first and second order diffraction terms along with all combinations of one reflection and one diffraction. For each diffracted path calculated the same path with an additional ground reflection is also evaluated.
% , -.Wall Receiver Locations
0 10 20 3040 50
Metres
Transmitter Location Diffraction Point
2.3
Measurement Sites
Two measurement sites are described in this paper (see figures 2 and 3). The principal reflectors for each site are shown in heavy lines with shading used to represent buildings. These sites were chosen because it was felt that they represented two quite different cordless environments.
Figure 2: Environmental description for Site 1 (Heavy lines show representation of obstacles within the model)
127
-.- Receiver Locations
Transmitter Location
I .
Metres
i
1) Diffraction Point
20
40
60
80
100
232 .. Site 2 (see figure 3)
Wall
Figure 3: Environmental description for Site 2 (Heavy lines show representation of obstacles within the model) sign criterion because for most systems the worst case Doppler shift would need to be assumed, and this is a parameter which is easily calculated.
This area was chosen in an attempt to model a more complicated scenario which might be typical of a street location. A combination of both LOS and OBS profiles was measured along the four lines shown (A-D). Various simplifications have been made for the environmental description in the model in an attempt to combine high speed with a reasonably accurate model.
3
3.1
Results
Wideband Modelling
2.4
Measurements Undertaken
Care was taken that the measurement and modelling techniques should be as similar as possible. Many impulse responses were both measured and modelled at the receiver locations indicated on the maps. These were then used to generate power delay profiles and subsequently their RMS delay spreads[7] values were calculated. The RMS delay spread is a measure of the time dispersion of a transmitted pulse. The results were then presented as a cumulative distribution and both measured and modelled results compared.
Figure 4 shows a typical Doppler scattergrams for a location at site 2. Doppler scattergrams contain a l l the information represented within a power delay profile with additional Doppler information added. This clearly shows the ability of the algorithm to evaluate individual rays. The cumulative distribution of the RMS delay spread values is shown in figures 5 and 6. The modelled and measured results agree quite closely in each case. The smaller RMS delay spread values were found in site 2 which was the smaller cell. Here there is a range between 40 nS and 130 nS with a median value of about 70 nS. Whilst there is good agreement between modelled and measured results it is apparent that in both cases the modelled results have a lower maximum RMS delay spread value. This is probably because the model
For the measured results the receiver was kept stationary. This was done so that the sounder statistics would average over a stationary channel. Also it w s felt that the limited Doppler information available a (from a moving receiver) was not an important de-
128
and close to the building at the road end.
32 .
Narrowband Modelling
In addition to wideband caracteristics of the radio channel, the narrowband characteristics have also been studied for both sites. Many signal strength values have been calculated for both LOS and OBS scenarios and their statistics have been studied. For the nonLOS case the signal strength values have been found to be well represented by the Rayleigh distribution and for the LOS case the amplitudes have been found to be well represented by the Rician distribution, with a varying K factor (the ratio of knownlscattered powers). Close to the transmitter the K factor was found to be as high as 10 dB. For more distant LOS locations the K factor was found to be around 0 dB (see figure
8).
Figure 4: Doppler Scat,tergram for Site 1 with receiver positioned 40m from transmitter
05
E 4 O 03
02
01
I , I , I I I I I
confines the cell barriers at the outer perimeter and thus may fail to find the occasional distant reflected path. In order t o consider more closely the factors affecting RMS delay spread, the simple environment of site 1 was examined in more detail. The RMS delay spread was found for many positions as the receiver was moved along the path used in the measurements. These results were then compared with measured results at identical locations (see Figure 7). The graph indicates that the model is not able to predict localised low or high values. There would appear to be two principal reasons for this. Firstly, the model does simplify the channel with insufficient information to prvdict small scale details, i.e localised scattering froiii vehicles, humans etc. Secondly, the measurement device works with a finite bandwidth so that, unlike the model, it experiences multipath effects for rays arriving with very short time delays. The effwt of this fading could have been reduced by performing the averaging with small scale movements of the transmitter. The model does however show that for this environment the RMS delay spread peaks at around 45m from the transmitter and is lower both close to the transmitter
OO
a ,
40
Bo Bo 1w RMS Delay Spread d Ins
1 P
140
Figure 5 : Modelled and measured RMS delay spread values for Site 1
09
s ::
08
05
0 0 2
01
Bo
im
im
im
i8o
180
2~
PO
240
RMS Delay Spread d InS
Figure 6: Modelled and measured RMS delay spread values for Site 2
129
Conclusions
A reflection and diffraction based algorithm has been shown to be a useful tool for the prediction of RMS delay spread values for two different outdoor environments. Previous work has shown the ability of a reflection based model to predict measured R M S delay spread values for indoor environments (51.
[4] R. Ganesh and K. Pahlavan, Statistical modelling and computer simulation of indooor radio channel, IEE Pmc. I., vol. 138, pp. 153 - 161, June 1991.
[5) M.
C. Lawton, R. L. Davies, and J. P. McGeehan, An analytical model for indoor multipath propagation in the picocellular environment, 6th Int. Conf. on Mobile Radio and Personal Comms, Warwick, UK, pp. 1- 8, December 1991.
ml However the model cannot predict s a l scale fluctuations in the RMS delay spread.
The algorithm has also been used in order to simulate the narrowband behaviour of the two sites. For the obstructed (OBS) case the signal strength values are well represented by a Rayleigh distribution. For the line of sight (LOS) the results were modelled by a Rician distribution with varying K factors depending on the transmitter-receiver separation.
(61 R. J. Leubbers, Finite Conductivity Uniform GTD Versus Knife Edge Diffraction in Prediction of Propagation Path Loss, IEEE %W. Antenn. Pmpag., vol. AP-32, pp. 70 - 76, January 1984. [7] D. C. Cox, Delay Doppler Characteristics of Multipath Propagation at 910 MHz in a Suburban Mobile Radio Environment, IEEE l h n s . Antenn. Pmpag., vol. AP-20, pp. 625 - 635, September 1972.
Acknowledgements
The authors would like to thank SERC and BT labs for the award of a SERC CASE research studentship. In addition they are grateful t o BT for the provision of their channel sounding and laboratory facilities, and in particular, they wish to acknowledge the the assistance of Paul Tattersall, Head of Mobile Propagation Group, for his contributive comments to the research programme. Finally the authors are indebted to their colleagues in the Centre for Communications Research, University of Bristol for the valuable advice and comments, and the provision of computing facilities.
-i o
S I
50
Clismmfrom W m
Figure 7: Modelled and measured variation in RMS delay spread with distance from the transmitter for Site 1
References
( 1 K. J. Gladstone and J. P. McGeehan, A Com1 puter simulation of the Effect of Fading on a QuasiSynchronous Sideband Diversity AM Mobile Radio Scheme, IEEE %ns. on Selected a m of Comms, vol. SAC-2, pp. 191-203, January 1984.
[2] G. L. Turin, F. D. Clapp, T. L. Johnston, S. B. Fine, and D. Lavry, A Statistical Model for Urban Multipath Propagation, IEEE %ns. Vehicul. Technol., vol. VT-21, pp. l - 9, February 1972.
[3] H. Suzuki, A Statistical Model for Urban Radio Propagation, IEEE %W. Vehicul. Technol., vol. VT-40, pp. 203 - 210, July 1991.
Figure 8: Rician statistics in Site 1for a LOS case with receiver-transmitter separation = 4om
130
I ;
.I
Optimization of spot pattern in indoor diffused optical wireless systems
by Madhusudhana Reddy Lebaka
DEPARTMENT OF ELECTRICAL ENGINEERING
INDIAN INSTITUTE OF TECHNOLOGY, KANPUR

June 2006
Optimization of spot pattern in indoor diffused optical wireless systems
A Thesis Submitted in Partial Fulllment of the Requirements for the Degree of Master of Technology
by Madhusudhana Reddy Lebaka
to the DEPARTMENT OF ELECTRICAL ENGINEERING
INDIAN INSTITUTE OF TECHNOLOGY, KANPUR

June 2006
ii
CERTIFICATE
It is certied that the work contained in the thesis entitled Optimization of spot pattern in indoor diused optical wireless systems by Madhusudhana Reddy Lebaka has been carried out under my supervision and that this work has not been submitted elsewhere for a degree.
(Dr. Y.N.Singh) June 2006 Associate Professor, Department of Electrical Engineering, Indian Institute of Technology, Kanpur-208016.
iii
ACKNOWLEGGEMENTS
I would like to express my gratitude to my advisor, Dr. Y. N. Singh, for his both insight and broad range of knowledge that he willingly shared with me during the research and writing of my dissertation. His advices are always very helpful, both academically and nonacademically. It has been a great pleasure working with him. I also would like to extend my sincere thanks to Mr. A. Sivabalan, for his encouragement, help and support. I express my appreciation and indebtedness to my friends G. Praveen, D.Sreenivas, P. Pathak, V. Srinivas, Anarul Islam, Rajat, and Rajeev who helped me in many ways during my thesis work. Madhusudhana Reddy Lebaka
iv
DEDICATED
to
MY PARENTS and SISTER
ABSTRACT
Simulated Annealing(SA) algorithm is proposed to be used in the optimization of the spot pattern for the indoor diuse optical wireless systems.The channel Response is analyzed using conventional grid-based patterns and a eld of view (FOV) of 30 is found to give a good performance balance in the uniformity of the received power distribution and multipath dispersion. Using the algorithm, the optimized spot pattern is found for minimizing the ratio of standard deviation of the received power to the average received power level. It resulted in more than 50% improvement.
Contents
1 INTRODUCTION 1.1 1.2 Thesis Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Thesis Organization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 2 3 4 4 5 6 6 7 7 10 11 11 12 13 13 14 14 16
2 REVIEW OF INDOOR INFRARED WIRELESS SYSTEMS 2.1 2.2 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Basic Optical Wireless System . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2.1 2.2.2 2.2.3 2.3 Transmitter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Free Space Medium . . . . . . . . . . . . . . . . . . . . . . . . . . . . Receiver . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Comparison between Radio and Infrared . . . . . . . . . . . . . . . . . . . . 2.3.1 2.3.2 2.3.3 Link Conguration . . . . . . . . . . . . . . . . . . . . . . . . . . . . Line of Sight Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . Diuse Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.4 2.5 2.6
Current Infrared Communication System
The IrDA Standard for Very Short Distance Point-to-Point Systems . . . . . Other Standards used in Diused Infrared Systems . . . . . . . . . . . . . . 2.6.1 2.6.2 Dierent IEEE 802 Standards . . . . . . . . . . . . . . . . . . . . . . Scope of Eort of IEEE 802.11 Standard . . . . . . . . . . . . . . . .
3 INFRARED CHANNEL MODELING
vi
vii 3.1 Multi-Receiver Channel Estimation . . . . . . . . . . . . . . . . . . . . . . . 3.1.1 3.1.2 3.1.3 3.1.4 3.2 Site and Link Model . . . . . . . . . . . . . . . . . . . . . . . . . . . Channel Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Source and Receiver . . . . . . . . . . . . . . . . . . . . . . . . . . . Environment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16 16 17 18 18 19 20 21 21 22 22 23 23 24 24 29 29 37 37
Impulse Response Calculation . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2.1 3.2.2 3.2.3 3.2.4 Decomposition into Bounces . . . . . . . . . . . . . . . . . . . . . . . Discretization into Facets . . . . . . . . . . . . . . . . . . . . . . . . Multi-Receiver Iteration . . . . . . . . . . . . . . . . . . . . . . . . . Multi-Transmitter Iteration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.3 3.4
Implementation And Computational Eciency
Simulations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.4.1 3.4.2 Simulation Setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Design Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.5
Impulse Response Calculations . . . . . . . . . . . . . . . . . . . . . . . . . .
4 OPTIMIZED SIGNAL POWER DISTRIBUTION 4.1 4.2 Signal Power Distribution . . . . . . . . . . . . . . . . . . . . . . . . . . . . Optimization of Spot Pattern . . . . . . . . . . . . . . . . . . . . . . . . . . 4.2.1 4.3 Simulated Annealing Technique . . . . . . . . . . . . . . . . . . . . .
Comparison of Results of Conventional Grid Patterns And Optimized Patterns 41 43 43 44
5 CONCLUSIONS AND FUTURE SCOPE 5.1 5.2 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Future Scope . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
List of Figures
2.1 2.2 3.1 3.2 3.3 4.1 4.2 4.3 Block Diagram of an Optical wireless Communication link . . . . . . . . . . Classication of Infrared Links . . . . . . . . . . . . . . . . . . . . . . . . . . Site and Link Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Impulse Response Calculation . . . . . . . . . . . . . . . . . . . . . . . . . . Spot patterns . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Signal Power Distribution for 2 2 Spot Pattern With Various FOVs . . . . Average delay spread as FOV varied with dierent patterns . . . . . . . . . 5 10 17 19 25 32 34 35
Signal Power Distribution for Various Spot Patterns With FOVs 30 . . . . .
4.4 Signal Power Distribution for Various Spot Patterns With FOVs at 30 , N=60,000 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.5 Optimized Patterns for 20 Fov . . . . . . . . . . . . . . . . . . . . . . . . . 36 38 39 39 40 40
4.6 Optimized Patterns for 30 Fov . . . . . . . . . . . . . . . . . . . . . . . . . 4.7 Optimized Patterns for 50 Fov . . . . . . . . . . . . . . . . . . . . . . . . .
4.8 Optimized Patterns for 75 Fov . . . . . . . . . . . . . . . . . . . . . . . . . 4.9 Optimized Patterns for 90 Fov . . . . . . . . . . . . . . . . . . . . . . . . .
4.10 Comparison of Results for Optimized Patterns to that of Conventional Grid Patterns . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
viii
List of Tables
2.1 2.2 3.1 Comparison between LEDs and LDs . . . . . . . . . . . . . . . . . . . . . . Comparison between Infrared and Radio Channels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6 8 26 26 27 27 27 28 30 31 31 33 35 41
Metrics from Impulse Response Calculation at 10 FOV
3.2 Metrics from Impulse Response Calculation at 20 FOV 3.3 Metrics from Impulse Response Calculation at 30 FOV 3.4 Metrics from Impulse Response Calculation at 50 FOV 3.5 Metrics from Impulse Response Calculation at 75 FOV 3.6 Metrics from Impulse Response Calculation at 90 FOV
4.1 Signal Power Distribution Parameters for Single Spot . . . . . . . . . . . . . 4.2 Signal Power Distribution Parameters for Two Spots . . . . . . . . . . . . . 4.3 Signal Power Distribution Parameters for 2 2 Spot Pattern . . . . . . . . . 4.4 Signal Power Distribution Parameters for 2 4 Spot Pattern . . . . . . . . . 4.5 Signal Power Distribution Parameters for various Spot Pattern at 30 FOV . 4.6 Signal Power Distribution Parameters for Optimized Spot Patterns . . . . .
ix
GLOSSARY
APD BER CGH DD FOV IM IrDA LAN LD LED LOS MAC OOK OSI PIN PPM RZ SA Avalanche Photo Diode Bit Error Rate Computer Generated Hologram Direct Detection Field of View Intensity Modulation Infrared Data Association Local Area Network Laser Diode Light Emitting Diode Line Of Sight Medium Access Control sub-Layer On O Keying Open Systems Interconnection P Intrinsic N Pulse Position Modulation Return to Zero Simulated Annealing
Chapter 1 INTRODUCTION
The emergence of portable computing devices such as laptops, palmtops and personal digital assistants has fulled the demand for mobile connectivity and hence, led the development of wireless local area networks(LANs). Wireless communications have always oered an alternate solution to the xed network, which introduce diculties in construction and rewiring during the initial system setup and later during the expansion phases. The benets of an alternative exible network are many in terms mobility and portability. Trends in communication suggest that the future network consist of ber-optic network and wide band wireless short-range access over a wireless channel. Portable terminals should have access to all of the services that are available on high-speed networks. Unlike their wide counterparts, portable devices are subject to severe limitations on power consumption,size and weight. The desire for inexpensive, high-speed links satisfying these requirements has motivated the recent interest in infrared wireless communication. Traditionally, radio frequency transmission was used in wireless applications. However, the RF spectrum is so congested that it is very dicult to accommodate high bit rate applications. The optical systems with low implementation complexity and no spectrum license requirements provide a possible solution. The optical infrared energy can typically conne within the communication environment. This eliminates the problems of interference generated by neighboring users and oers a degree of security at the physical layer level. The 1
2 transmission equipment and optical wavelength can be reused at other parts of the building. Optical wireless systems also oer immunity from signal fading, which is a major problem in RF communication systems. As such, indoor infrared communications has recently gained importance, especially in view of the increased data mobility requirements of users for both computing and communications. Infrared communications were mainly studied during 1960s as an alternative for mobile radio communications. But, no real system for indoor coverage using infrared links was seriously considered. Recently, infrared has gained ground with operators seeking to cover area that require high bit rate services such as the oce areas. Serious work on indoor optical wireless systems started during early 90s. At present it is a very rapidly developing research are and there is huge interest and commitment in this area by major communication industries due to its enormous commercial applications. In order to correctly and reticently deploy a communication system, the system designer must have a sound knowledge of the channel. This can be done through experimentation and modeling with methods that are accurate enough for all signicant channel characteristics. This thesis deals mainly with the study of dierent spot patterns in indoor diused optical wireless systems. Simulation studies are done on the various properties of the optical signal, and receiver. The main objectives of the work are discussed in the following section.
1.1
Thesis Objectives
The Objective of the thesis is to optimize the spot patterns so as to receive uniform power all through the receiving plane. Multiple Bounce Impulse Response Model and Simulate annealing algorithm are used for simulation. The objectives are summarized as follows. To nd RMS delay spread of the transmitted optical signal due to multipath propagation. To study the eect of eld of view (FOV) on the received signal.
3 To study the eect of spot pattern on the received signal. To nd the optimized spot pattern at various FOVs, for which, the received power is uniform through out the receiving plane.
1.2
Thesis Organization
Chapter 2 gives a review of the indoor optical wireless systems. The basic components of optical wireless systems and link congurations are discussed. A brief literature survey of infrared channel modeling and characterization is also presented. Chapter 3 summarizes theoretical background of the indoor channel characterization and simulation studies of channel. Chapter 4 deals with simulation studies of power distribution in the receiving plane for various spot patterns and optimization patterns. The results are also discussed. Chapter 5 presents the conclusions of the work and future improvements that may be possible.
Chapter 2 REVIEW OF INDOOR INFRARED WIRELESS SYSTEMS

2.1 Introduction
The wireless systems widely used today utilizes radio waves, but this alone may not be sucient for the future. Modern digital communication systems require high data rate capabilities. However, the radio spectrum is regulated and saturated. An alternative medium is infrared communications, which appear to be very attractive way to complement radio. Infrared communications have found many areas of use. Almost every home includes a television set which is controlled by an infrared remote control. Also, there is a standard for communication between two computers using infrared. This standard is called IrDA (Infrared Data Association), and has been included in almost every portable computer and mobile phone. This supports speed up to 115kb/s but the limiting factor is distance between the terminals should be low, and a line of sight (LOS) is always needed. Infrared communications are very successful in the eld of directed communication between two buildings. although very high data rates up to hundreds of Mb/s can be achieved, it has disadvantage of being unsustainable in heavy weather conditions such as snowfall or fog. The link cannot be sustained then, due to the increased path loss between terminals. 4
Figure 2.1: Block Diagram of an Optical wireless Communication link This chapter presents a small review of the indoor infrared wireless communication system. It explains the basic components, link congurations, a comparison with radio, and literature survey of the important works in this eld. nally, Some of the practical systems reported are discussed.
2.2
Basic Optical Wireless System
The basic sub system of an optical wireless system are transmitter based on either (LEDs or LDs), the channel (the medium between the transmitter and the receiver)and the receiver based on (PIN or APD based). The link length can vary from a few meters to a few km. The block diagram of a typical optical wireless communication link is shown in Fig.2.1. The electrical information signal produced by the source modulates an optical carrier. The one commonly used is Intensity Modulation (IM). The modulated optical carrier is propagated through the channel. At the receiver, the received optical eld is optically collected and converted back to an electrical signal by a detector, which is further processed by electronic stages to recover the original information with an acceptable level of error.
2.2.1
Transmitter
The transmitter side consists of the data source, driver circuits and the light source. There are two basic light sources, the LD and the LED. Since eye safety is a major consideration for indoor applications, the optical source need to be chosen carefully. LEDs are the preferred choice for indoor use. Since the light intensity of an LED may not be sucient for most applications, LED arrays may have to be used. Another popular solution is to use a laser diode with a diuser. A comparison between the characteristics of LEDs and the LDs [1] are given in table 2.1 Characteristics Spectral Width Modulation Bandwidth E/O Conversion Eciency Eye Safety Light Emitting Diodes 25-100 nm Tens of KHz to tens of MHz 10 to 20% Generally considered eye safe Laser Diodes < 105 to 5 nm Tens of MHz to tens of GHz 30 to 70% Must be rendered eye-safe. Especially for < 1400nm Cost Low Moderate to high
Table 2.1: Comparison between LEDs and LDs
2.2.2
Free Space Medium
The link budget for a free space link are strongly determined by the atmospheric loss along the propagation path, which comprises of free space loss, clear air absorption, scattering, refraction, and scintillation. All forms of optical wireless systems experience free space loss. Other sources of atmospheric losses are applicable to long distance systems only. Free space loss denes the portion of optical power arriving at the receiver that is usefully captured within the receivers aperture.
2.2.3
Receiver
The function of the receiver is to convert optical signal back to the original electrical signal. Receiver consists of a PIN or APD photodetector, preamplier receiver circuit. PIN or APD operate in reverse bias, also known as photoconductive mode of operation. The advantages of photoconductive operation are high speed, lower capacitance, and better linearity. The photodetector and amplier are sources of noise at the receiver side. The noise in a photodiode can be of two types; the rst one is called the shot noise due to dark current and received optical signal. and the second one is called thermal noise due to shunt resistance, also known as Johnson noise. The important requirements on the photodiodes for optical wireless communication are high quantum eciency, fast response time, low capacitance, low dark current, and low avalanche excess noise. It is desirable to use a large area photodetector since shot noise can be reduced, SNR is proportional to the detector area. However large area detectors have high capacitance, which can limit receiver bandwidth and greatly increase thermal noise. It is desirable to reduce the required detector physical area by the use of an optical concentrator, which accepts light from a large collection area and concentrates it to the smaller detector area.
2.3
Comparison between Radio and Infrared
Infrared communication oer advantages and some disadvantages when compared with the analogous radio links. Table 2.2 gives a comparison between the two systems [2]. As seen from the comparison, infrared shares some properties with radio, although there are some signicant dierences. These are discussed in detail below. Multipath fading is not present in infrared channels, since the area of the receiving photodetector spans several thousands of wavelengths. the output current of the photodiode will be proportional to the integral of the squared electric eld over the entire area of photodiode surface. thus a single photodetector can provide a signicant spatial diversity, which will prevent multipath fading. Multipath dispersion is present in
8 Property Path Loss Multipath Fading Multipath Dispersion Bandwidth Limitation Dominant Noise Signal Transmitted Wall penetration Radio High Yes Yes Regulatory Interference Amplitude Yes Infrared High No Yes Photodiode Background Noise Power No
Table 2.2: Comparison between Infrared and Radio Channels infrared channels also. If a short pulse is sent through the channel, the pulse broadens due to dispersion and inter symbol interference occurs. The dispersion is the result of the reective properties of the channel, which transforms a narrow pulse into wider version. This limits the data rate and increases the bit error rate. In radio, the main limitation of bandwidth is due to regulation, since the bandwidth has to be licensed before being used. In infrared limitation is posed by the channel and various devices. The dominant noise in cellular communications is interference from adjacent cells. In infrared systems, noise comes from the background. The three main sources of ambient light are sunlight, incandescent lamps and uorescent lamps [3],[4],[5]. Infrared links cannot sustain operations in direct sunlight, unless spatial lters or very directional sources and receivers are used. Articial sources of ambient light produce a periodic interference signal in infrared wireless receivers which, if ignored, has the potential to degrade link performance. These lights have a wide spectrum and they also emit higher power infrared radiation during warm up thus degrading infrared communication. Although the infrared channel is related to the linear gaussian channel, there is a dierence between the two. The input of infrared channel represents power and not
9 amplitude unlike radio. This gives raise to two constraints; rst, the signal must be positive and its average value must not exceed a prescribed value[6] for safety reasons. The transmitted power in the infrared domain is described by 1 T 2T lim where x(t) is the instantaneous value of the transmitter signal(Watts). P is the maximum value of the power allotted to be transmitted due to safety reasons. T is the duration of communication(sec). Infrared radiation does not penetrate opaque objects, and walls eciently separate two cells. Security is high since the radiation is conned within the cell. On the other hand, less coverage can be achieved with an infrared link, and base stations have to be placed within every room of operation. The following points summarize the advantages and drawbacks of infrared communications over radio. Infrared oers unregulated bandwidth: 200Thz in the 700-1500nm range. There is no interference from adjacent cells and security is very high, No multipath fading. Higher capacity per unit volume. Cost eective at rates above 100 Mb/s. The draw backs are Multipath dispersion is present.
T
x(t)dt P
T
(2.1)
10
Figure 2.2: Classication of Infrared Links Large amount of radiations can be harmful to humans. Strong interference from daylight, light sources, remote controls and other devices.
2.3.1
Link Conguration
Indoor optical wireless can employ various designs, and it is convenient to classify them according to two criteria.This classication scheme is shown in Fig.2.2[1]. The rst section is directionality of the transmitter and receiver. Directed links employ directional transmitters and receivers, which must be aimed in order to establish a link. Non directed links employ wide-angle transmitters and receivers, alleviating the need for pointing. Directed link design maximizes power eciency, since it minimizes path loss and reception of ambient noise. On the other hand, non directed links may be more convenient to use, particularly for mobile terminals, since they do not require aiming to the transmitter or receiver. It is also possible to establish hybrid links, which combine transmitters and receivers having dierent degrees of directionality. The second classication criterion to whether the link relies upon the existence of an uninterrupted LOS path between the transmitter and receiver. LOS links rely upon such a path, while non-LOS relies upon reection of the light from the ceiling or some other diusely
11 reecting surface. LOS link design maximizes power eciently and minimizes multipath distortion. Non-LOS link design increases link robustness and ease of use, allowing the link to operate even when barriers are there between the transmitter and receiver. The greatest robustness and ease of use is achieved by the non directed-non-LOS link design which is often referred to as diuse link.
2.3.2
Line of Sight Systems
Line of Sight systems employ high degree of directionality of the transmitter and receiver and uninterrupted LOS. LOS link design maximizes power eciency and minimizes multipath distortion. The performance of the link relies up on the existence of an uninterrupted LOS path between the transmitter and receiver. The main drawback is the lack of mobility and susceptibility to blocking. The optical sources must be Class 1 eye safe; hence LEDs have to be used in place of LDs which limits the capacity to few megabits per second. These types of links are used for applications in which the terminals are relatively xed, such as desktop computers in oce. One of the most widely adopted standards is the IrDA.
2.3.3
Diuse Systems
The greatest robustness and ease of use is achieved by the non-directed, non-LOS design, which is often referred to as diuse link. However, diuse systems have a higher path loss than their LOS counter parts, requiring higher transmitter power and a receiver having large light collection area. This type of topology overcomes the blocking problem by relying on the high reectivity of common building materials, so that a signicant fraction of the received signal arrives at the receiver from a number of angles. Such as topology is extremely exible and can be used in either organized or ad-hoc networks. However the bit rate that can be achieved is very less compared to the LOS systems. In addition, optical losses associated with the link are much greater than LOS systems.
12 In case of an entirely diuse link, the optical power launched into a closed room is scattered by the walls, ceiling, oor and furniture. After some reections the irradiance is almost uniform, so that detector need not be oriented towards the transmitter. Full mobility within the room is allowed and there is no shadowing eects caused by moving persons or machines. There are two limitations for diuse links: The optical power has to be large enough to cover the whole volume, and multipath dispersion limits the data rate. Multipath propagation causes the a spread of transmitted pulse, which may result in loss of pulse amplitude and inter symbol interference. Hence, there is a maximum transmission speed which depends on the room size and reection coecients inside the room.
2.4
Current Infrared Communication System
At present, most infrared links are of the directed-LOS or hybrid-LOS designs. The low path loss of these designs minimizes the transmitter power requirement and permits the use of a simple, low-cost receiver. Typically, these links transmit using a single LED, which emits an average power of several tens of mW that is concentrated within a semiangle of 15 to 30 . The LED emission wavelength typically lies between 850 and 950 nm. This wavelength matches the responsivity peak of the silicon p-i-n photodiode. In hybrid-LOS link designs, the photodiode is most often encapsulated in a planocylindrical or hemispherical plastic lens that serves to concentrate the received light, while maintaining a relatively wide FOV, e.g., a semiangle of the order of 60 . Directed-LOS links employ an optical concentrator that restricts the FOV, usually with the goal of providing a higher degree of optical concentration. Directed-LOS and hybrid-LOS links are relatively free from multipath distortion, sometimes permitting them to achieve bit rates above 100 Mb/s while maintaining a very simple design. These link designs are well-suited for point-to-point and some point-tomultipoint applications, but are not suited for multiple-access networks, since it is dicult to establish full bidirectional connectivity between more than two transceivers. DirectedLOS and hybrid-LOS links have been used for many years in remote-control units and other
13 unidirectional, low-bit-rate applications.
2.5
The IrDA Standard for Very Short Distance Pointto-Point Systems
Over the past seven years, the IrDA has established standards for short-range, half-duplex LOS links operating at bit rates up to 4 Mb/s [12]. Two of the key features of IrDA transceivers are low cost and low power consumption (under 1 W while transmitting, and under 100 mW when idle or receiving). At present, more than 160 companies worldwide are members of IrDA. There are several key components of the IrDA standards [12]. The IrDA Serial Infrared Physical Layer denes standards for half-duplex links at several bit rates up to 4 Mb/s. 4 Mb/s links employ 4-pulse-position modulation (4-PPM), while 1.152 Mb/s links utilize on-o keying (OOK) with return-to-zero (RZ) pulses having a duty cycle of 0.25. Links operating at bit rates of 115.2 kb/s and below employ OOK with RZ pulses having a duty cycle of 0.1875 (shorter pulses are permitted in some cases). IrDA-compliant transmitters must emit at a wavelength between 850 and 900 nm into a semiangle (at half-power) of 15 to 30 . Compliant receivers must have a FOV (semiangle at half-eective light-collection area) of at least 15 . Most IrDA receivers have a much larger FOV, so that most IrDA links are of the hybrid-LOS type. IrDA links are required to achieve a bit-error rate (BER) not exceeding 109 (108 for 4 Mb/s links) over a range of at least 1 m, but many links achieve a range as long as 3 m.
2.6
Other Standards used in Diused Infrared Systems
The main importance of standards is to provide wireless connectivity to automate machinery, equipment or stations that require rapid deployment, which may be portable or hand held or which may be mounted on moving vehicles with local area. The other importance is to
14 oer a standard for use by regulating bodies to standardize access to one or more frequency bands for the purpose of local communication.
2.6.1
Dierent IEEE 802 Standards
The IEEE has produced several standards for Local Area Networks. These standards, collectively known as IEEE 802, include CSMA/CD, CSMA/CA, Token bus, Token ring, etc. The various standards dier at the physical layer and Medium Access Control (MAC) sublayer, but are compatible at the data link layer. The IEEE 802 standards have been adopted by American National Standard (ANSI), by NIST as government standards, and by ISO as international standards (known as ISO-8802). IEEE 802.11 standard describes Diused IR & RF Wireless Local Area Network.
2.6.2
Scope of Eort of IEEE 802.11 Standard
The scope of eort of IEEE 802.11 standard is to dene physical layer wireless data networking and physical layer transmissions (PHY) by radio or infrared and Medium Access Control (MAC) Protocol compatible with the existing standards for higher layers. The IEEE standard for wireless LAN started in 1988 as IEEE 802.4L as a part of the IEEE 802.4 Token bus wired LAN standard. In 1990, the IEEE 802.4L changed its name to IEEE 802.11 to form a stand alone Wireless LAN standard in the IEEE 802 LAN standard organization and several draft standards have been published for review. With long eort, the IEEE 802.11 group developed framework to incorporate wireless specic issues such as power control, frequency management, roaming and PHY and MAC Protocol sub-layer in LAN standards. The IEEE 802.11 draft standard describes mandatory support for 1 Mb/s Wireless LAN with optional support for a 2 Mb/s data transmission rate. at present there are two emerging LAN standards in use, as mentioned below. 1. European Telecommunication standards institutes High Performance European Radio
15 LAN (HIPERLAN), for data rates upto 23.529 Mb/s. 2. The IEEE 802.11 standard for Wireless LAN. Both draft standards cover the PHY and MAC sub-layer of the open systems interconnection (OSI) seven layer reference model. Through judicious use of the technologies employed in currently available systems, it is possible to enhance the performance of wireless infrared systems signicantly. It appears likely that 10-Mb/s diuse links and low-cost LOS links operating at tens of Mb/s can be achieved. Even higher bit rates will be desirable in future applications. Recent research work suggests that using new techniques, low-cost infrared links operating in the 100 Mb/s range may be achieved.
Chapter 3 INFRARED CHANNEL MODELING

This chapter describes the tools used to model the infrared channel. The channel under study is the infrared diuse channel, in which there is no need for LOS between the transmitter and receiver. Communication may rely on scattered radiation from the walls of the room. In this way, mobility is increased, and shadowing does not have a large impact on the link.
3.1
Multi-Receiver Channel Estimation
In this discussion we limit considerations to rectangular rooms, although techniques can be easily extended to other geometries. Next we dene the models for the source, reectors, and receiver.
3.1.1
Site and Link Model
We model optical wireless channels formed by a transmitter and receiver placed inside a reective environment, as depicted in Fig. 3.1(b). The transmitter or source Sj is a LD or a LED transmitting a signal Xj (t) using IM. We rst consider a collection of receivers, each with a photodiode with responsivity r and using DD. These receivers may be either a group 16
17 of receivers being used as an angle-diversity receiver, as in [13], or they might represent a collection of alternative single receiver locations that are being considered together. We will show that considering all receiver locations and orientations concurrently will bring substantial savings in channel estimation computation time.
(a) Environment
(b) Source and Receiver
Figure 3.1: Site and Link Model
3.1.2
Channel Model
When source Sj is transmitting receiver Ri is receiving, the current from the photodiode is Yij (t) = rXj (t) hij (t) + Nj (t), (3.1)
where denotes convolution, hij (t) the impulse response of the channel between source Sj and receiver Ri , and Ni (t) the noise at the receiver. The baseband impulse response hij (t) for IM/DD communication [14] is completely determined properties of source Sj , receivers Ri , and environment E. Hence, we will write hij (t) more specically as hE (t; Sj , Ri ). Although we consider multiple transmitters and receivers, here its been restricted to [1] a single transmitter and a single receiver at a time. If multielement transmitters are employed, then the signal
18 received by receiver Ri would be

J
Yij (t) =
j=1
(rXj (t) hij (t)) + Nj (t),
(3.2)
where the transmitted signals Xj (t) might be carrying the same or dierent information sequences. A multielement receiver employing combining would receive the signal
I
Yi (t) =
i=1
(i Yij (t i )).
(3.3)
3.1.3
Source and Receiver
The source Sj is described by a position vector sj , an orientation vector nsj and a radiant r intensity pattern T (), where we assume for simplicity that the radiant intensity pattern has axial symmetry about the normal. A typical model for radiant intensity pattern is the Lambertian of order n given by T () = n+1 cosn (). 2 (3.4)
The receiver Ri is described by a position vector ri , an orientation vector nri , an optical r collection area Ari , and an eective area at incident angles of Ai () = Ari gi () is again modeled as axial symmetric. This allows for a very general description of the receiver optical system. A typical model for a bare photodiode is that g() = cos(); the cosine dependence models the decline in eective area for light incident on planar detectors at non-normal incidence.
3.1.4
Environment
The environment E is modeled as a set of Nb rectangular boxes {B1 , ..., BN b }, as depicted in Fig.3.1(a). The rst box B1 represents the universe in which all other boxes and all sources and receivers are contained. This can represent a single room, a oor, or even an entire building. Interior objects are described by single boxes or combinations of boxes. This method allows for inclusion of such objects as wall partitions, doorways, desks, chairs, and
19
Figure 3.2: Impulse Response Calculation people. The boxes are further modeled as having six opaque internal faces and six opaque external faces. Only the exterior faces of the internal boxes {B2 , ..., BN b } are relevant, and only the internal faces of the universe box B1 are relevant, for a total of 6Nb reecting faces. Each face Fi is modeled as a diuse reective surface (Lambertian) of reectivity F i The receivers and transmitters are not included as boxes, so their packaging must be explicitly included if it is signicant to the problem at hand.
3.2
Impulse Response Calculation
Impulse response calculation follows the basic methodology outlined in [15] with extensions for arbitrary transmitter and receiver gains and multiple transmitters and receivers. The calculation involves decomposition into bounces, discretization into facets, and nally multi-receiver iteration. We then present an equivalent formulation for a multi-transmitter calculation.
20
3.2.1
Decomposition into Bounces
All transmitted light arriving at the receiver has undergone a denite number of reections or bounces. Hence, we can decompose the impulse response hE (t; Sj , Ri ) as
hE (t; Sj , Ri ) =
k=0
hk (t; Sj , Ri ), E
(3.5)
where hk (t; Sj , Ri ) is the impulse response due to signal light undergoing exactly k E bounces during its path from the source Sj to the receiver Ri . The line of sight impulse response h0 (t; Sj , Ri ) is given by E
2 h0 (t; Sj , Ri ) = V ( sj , ri , E)T (ij )(Ari g(ij )/Dij )(t Dij /c), r r E
(3.6)
where Dij = |( sj ri )| is the distance between the source and receiver. The visibility r r function V ( sj , ri , E) is 1 when the LOS path between Sj and Ri is unobstructed, and is r r zero otherwise. Now, the K-bounce response can be calculated using the (K-1)- bounce response using hk (t; Sj , Ri ) = E d r hk1 (t; Sj , d r ) h0 (t; d s , Ri ), E E (3.7)
where the integral is over all surfaces in E and is the surface reectivity function. (see Fig. 3.2). The quantities d
r
and d
represent a dierential surface of area dr2 that is rst
acting as a receiver from the source Sj and then as a source to the receiver Ri . The surfaces act as receivers with g() = cos() 1{ < /2} and as rst-order Lambertian transmitters. To estimate hE (t; Sj , Ri ) using (3.7), we consider only M bounces so that
M
hE (t; Sj , Ri )
k=0
hk (t; Sj , Ri ). E
(3.8)
The contributions to the overall impulse response from the k-bounce impulse response will decline for increasing k so that excellent approximations can be obtained for M ranging from 3 to 10, as discussed in [15].
21
3.2.2
Discretization into Facets
The integration in (3.7) is approximated by representing each face Fi at a spatial partitioning factor P , i.e. each face is divided into small elements of size 1/P 1/P m2 . Hence we estimate hk (t; Sj , Ri ) using E
N
hk (t; Sj , Ri ) E
n=1
r hk1 (t; Sj , E n
r n)
h0 (t; E
s n , Ri ),
(3.9)
where
r n
and
s n
represent element n acting as a receiver and source, respectively. The number
of elements N is given by
Nb
N = 2P where Bi is the ith box.
2 i=1
(Lx,Bi Ly,Bi + Ly,Bi Lz,Bi + Lz,Bi Lx,Bi ),
(3.10)
3.2.3
Multi-Receiver Iteration
r m
In (3.9) with ri =
N
to obtain
N r n)
hk (t; Sj , E where
r m)
n=1
r hk1 (t; Sj , E n
h0 (t; E
s r n, m)
=
n=1
mn hk1 (t mn ; Sj , E
r n ),
(3.11)
r T (mn )g(mn ) mn = V ( s , r , E) n r n r n 2 P 2 Dmn
(3.12)
and mn = Dmn /c. The angles mn , mn and Dmn are the receivers angle to the source, the sources angle to the receiver and the source-to-receiver distance, respectively, for the source
s n
, for the receiver
r n. r n ).
Hence, to calculate hk (t; Sj , Ri ), rst calculate N impulse responses of h0 (t; Sj , E E these, compute h1 (t; Sj , E possible to calculate
N r n)
using
and continued until hk1 (t; Sj , E
r n)
is obtained, at this point it is
hk (t; Sj , Ri ) E
n=1
r hk1 (t; Sj , E n
r n)
h0 (t; E
s n , Ri )
(3.13)
for each receiver Ri . A key observation is that previously calculated impulse response do not depend on impulse response calculated later, and thus can be computed only once,
22 independently of the number of receivers involved. Hence, the only calculation required for each receiver is the collection stage given by (3.13). The nal calculation stage is to combine the k-bounce impulse responses into an estimate of the overall impulse response, as shown in (3.8).
3.2.4
Multi-Transmitter Iteration
The approach of above section is the most natural and ecient when there are more receiver locations than transmitter locations. The following is an equivalent impulse response calculation when the situation is reversed. i.e. more transmitter locations than receiver locations. Rewriting (3.7)as hk (t; Sj , Ri ) = E and thus (3.9) as
N E
d r h0 (t; Sj , d r ) hk1 (t; d s , Ri ) E E
(3.14)
hk (t; Sj , Ri ) E in (3.15) substituting Sj =

N s m
n=1
r h0 (t; Sj , E n
r n)
hk1 (t; E
s n , Ri )
(3.15)
,
N
hk (t; s , Ri ) E m
n=1

r n
h0 (t; s , r ) E m n
hk1 (t; s , Ri ) E n
=
n=1
mn hk1 (t mn ; E
s n , Ri )
(3.16)
where mn and mn are dened as before. Hence, to calculate hk (t; Sj , Ri ), rst calculate N impulse responses of h0 (t; E E these, compute h1 (t; E
r n , Ri ) r n , Ri ).
using
and continued until hk1 (t; , E

N
r n , Ri )
is obtained, at this point it
is possible to calculate hk (t; Sj , Ri ) E

n=1
r hk1 (t; Sj , E n
r n)
h0 (t; E
s n , Ri )
(3.17)
3.3
Implementation And Computational Eciency
The computation time primarily depends on (a) the maximum number of bounces M, (b) the number of partitions N in the room, and (c) the number of receivers I. From a derivation
23 given in [22], for a single receiver, the computation time is O(N 2 .M 2 ), while the computation time per receiver T(M,N,I)/I for large I is O(N.M). For a 5 5 m2 empty room , and divided into 2400 facets(N). The time Computational time required to calculate a two-bounce impulse response in this scenario for a single receiver is 25 seconds, whereas we can calculate impulse responses for 2500 dierent receivers (for the same transmitter) in 38 seconds, resulting in a speedup, when compared with the impulse response calculated individually for 2500 receivers, which results in (2500 25) seconds. The calculations were performed on a 1.6 GHz Pentium 4 processor. For very large collections of receivers, the computation time will be dominated by the time to calculate the receiver impulse responses hk (t; Sj , Ri ) for each receiver Ri , and not by the time to compute the E surface responses hk (t; Sj , E
r m ).
For one-bounce responses, the speedup is much more modest
because the most time-consuming operation (calculation of N surface responses from N previous surface responses) is not needed. We are only saved from having to recompute a collection of zero-bounce responses.
3.4
Simulations
Simulation results of infrared channel impulse response are described here.
3.4.1
Simulation Setup
The simulation environment conditions are described here. 1. Room dimensions of 5m 5m 5m are used in the simulation. 2. Reection coecients [9] of ceiling and walls are considered as 0.8, where as bottom is considered as 0.3. 3. A Lambertian mode number of 1 is used for the diusing surfaces. 4. To collect the transmitted impulse response, a 1cm2 active area photodetector directed vertically at the ceiling is used.
24 5. The transmitter is at the center of the room and directed vertically at the ceiling. Total power of transmitted pulse is 1W. All the calculations in this thesis are two-bounce impulse responses (k=2). Many assumptions are made for simplicity of calculations. 1. The room is considered as dark room, without any interference, also there are no windows in room to prevent sunlight. 2. There is no loss of power from transmitter to the top(ceiling) of the room. i.e. there is no path loss in this part of transmission.
3.4.2
Design Parameters
A metric to determine the performance of a wireless network is the RMS delay spread, = 2 ( )2 [17], where = P ( ). / P ( ). is the delay associated with the signal,
relative to the rst arriving signal, P ( ) is the associated power of the signal, is the mean delay spread, and 2 is the second moment of the delay spread. A higher delay spread limits the data rate that can be transmitted, and a conservative estimate of the data rate can be found by calculating the inverse of 10. This data rate does not take into account equalization and other forms of signal conditioning which could be used to increase the data rate.
3.5
Impulse Response Calculations
To analyze the eect of spot patterns and receiver FOV on the system response, the model was run with 4 dierent spot patterns and 6 dierent receiver FOVs. Four grid-based designs as shown in Fig.3.3: 2x2, with a 2m separation between the spots; 2x4, with a 1.65m separation; Patterns consisting of a single spot (PT) as well as the case of two spots is considered for simulation. Each pattern had a total of 1W which was equally distributed into all the spots in the design. FOV values of 10o , 20o , 30o , 50o , 75o and 90o were used.
25 With the detector placed at 2 positions within the room, the received power and delay spread results are tabulated in table 3.1 to table 3.6. At position A, the detector is placed at the exact room center with coordinates (2.5, 2.5, 0.0) and at position B, the detector is placed at the corner of the room with coordinates (0.5, 0.5, 0.0).
(a) single spot
(b) Two spots
(c) 2 2 Spot Pattern
(d) 2 4 Spot Pattern
Figure 3.3: Spot patterns The received power is very low for most of the transmitting spot patterns at 10 FOV of the receiver and the placement at the corner, resulting in only 2nd order bounces from the walls to be received by the detector and thus the generally lower receiver power. The
26 2 2 pattern results in more signal power being received as the wider coverage of the pattern allows some of the 1st order signals from the ceiling to be reected onto the detector. However, the tradeo is that the impulse response is longer compared as to when other patterns are used. When the 10 FOV detector is placed at the room center, the limited FOV only allows 1st order signals from the ceiling to be received. As the 2 2 and 2 4 patterns do not have spots that are within the receiver FOV, no signal is received resulting in a lost connection at the center when these 2 patterns are used. When a single spot lies within the FOV, a negligible delay spread results due to the restriction of multi-paths entering the detector. The greatest amount of power is received when a single spot is used. This is since all the power is concentrated within that single spot and when the single spot falls within the receiver FOV. Spot Pattern FOV (deg) Single Two 22 24 10 10 10 10 Power Received(W ) Pos A 0.0866 0.0086 0.0 0.0 Pos B 0.0 1.9986e-8 7.267e-8 2.399e-7 Delay Spread (ns) Pos A 0.0 0.1674 NA NA Pos B NA 0.2498 0.2781 0.2546
Table 3.1: Metrics from Impulse Response Calculation at 10 FOV
Spot Pattern FOV (deg) Single Two 22 24 20 20 20 20
Power Received (W ) Pos A 0.1765 0.0164 0.0140 0.0105 Pos B 2.7450e-6 6.6078e-8 0.0070 0.0052
Delay Spread (ns) Pos A 0.1601 0.1696 0.17242 0.1723 Pos B 0.2487 .2440 0.1724 0.01724
27
Power Received (W ) Pos A 0.227 0.01989 0.0197 0.0151 Pos B 0.01447 0.01291 0.0143 0.0075
Delay Spread (ns) Pos A 0.1628 0.1722 0.1769 0.1755 Pos B 0.1834 0.1827 0.1828 0.1759
28 Power Received (W ) Pos A 0.227 0.02172 0.0202 0.0182 Pos B 0.0172 0.01651 0.0162 0.0147 Delay Spread (ns) Pos A 0.17280 0.1743 0.1775 0.1811 Pos B 0.1853 0.1869 0.1866 0.1901
Table 3.6: Metrics from Impulse Response Calculation at 90 FOV The power received is more as FOV increases, this is due to receiver will detect more multipaths. At 90 FOV it will detect all the signals which fall on the receiver area. However delay spread also increases with increase in FOV.
Chapter 4 OPTIMIZED SIGNAL POWER DISTRIBUTION

4.1 Signal Power Distribution
2500 receivers are placed in the room to calculate signal power distribution and to visualize how the signal power is distributed in the room. To analyze the results, another metric, the standard deviation of the received power and average delay spread are taken into consideration. From the results, particularly at the narrower FOVs, certain positions may be unable to receive any signals if no sources, either 1st order from the ceiling or 2nd order from the walls, within the receiver FOV. This is particularly prevalent at the corners, which tend to suer a sharp drop-o in power as compared to positions near the room center. A low standard deviation relative to the average received power would imply that the signal power distribution throughout the receiving plane, in this case the oor, would be more uniform than a high relative standard deviation, and would reduce the drop-o at the corners. Furthermore, a more uniform signal distribution would simplify the design of the receiver electronics by reducing the required dynamic power range of the electronics. The receiver FOV has a large impact on the system performance. As the FOV is reduced, 29
30 the number of multipaths reaching the detector is reduced, which reduces the delay spread, limiting multipath dispersion eects and increasing the possible data rates. However, the tradeo is that with a reduction in FOV, few paths reaching the detector means that the average received signal power is also reduced accordingly. At 90 , as earlier mentioned, all signals reaching the detector are accepted. When the FOV is reduced to 45 however, the weaker 2nd order signals from the walls are rejected more than the stronger 1st order signals from the ceiling. For example, when the receiver is placed at the room center, a 45 FOV describes a circle of radius 2.5m at the ceiling. Eectively this means only 1st order signals from the ceiling is accepted. When the FOV is increased to 90 at the same position, the additional power at the receiver is due to the acceptance of 2nd order signals from reections o the room walls. FOV(deg) Average Power(P ) Std.Dev of Power Ratio W 10 20 30 50 75 90 0.09157 0.3599 0.7125 0.1601 0.7628 0.763094 (P ) W 0.2866 0.4426 0.2459 0.1373 0.13729 0.137322 3.130 1.22968 0.35027 0.18064 0.1799 0.17995
Table 4.1: Signal Power Distribution Parameters for Single Spot The variation of spot pattern do not have a eect on the performance of the system with a variation of within 10% when dierent patterns are used. However, the choice of patterns aects the p standard deviation values, with a variation of more than 50% with the standard deviation values, particularly at narrower FOVs. At an FOV value of 10 , p is more than p . A large variation in received power implies that some locations are not receiving any signals. Although an extremely low average delay spread mut is possible at this FOV, alignment is needed for the spot distribution with the room
31
FOV(deg)
Average Power(P ) Std.Dev of Power Ratio W (P ) W 0.19207 0.3015 0.21195 0.1183 0.11845 0.11848 2.0975 0.86414 0.35405 0.1628 0.1623 0.1622
10 20 30 50 75 90
0.09157 0.3490 0.6212 0.7268 0.7295 0.7297
Table 4.2: Signal Power Distribution Parameters for Two Spots
FOV(deg)
Average Power(P ) Std.Dev of Power Ratio W (P ) W 0.1193 0.1608 0.20918 0.1004 0.10062 0.10063 1.303 0.4759 0.33862 0.1444 0.1441 0.1440
10 20 30 50 75 90
0.09157 0.3380 0.5406 0.6592 0.6978 0.6980
Table 4.3: Signal Power Distribution Parameters for 2 2 Spot Pattern
32
(a) FOV 10
(b) FOV 20
(c) FOV 30
(d) FOV 50
(e) FOV 75
(f) FOV 90
Figure 4.1: Signal Power Distribution for 2 2 Spot Pattern With Various FOVs
33 FOV(deg) Average Power(P ) Std.Dev of Power Ratio W 10 20 30 50 75 90 0.07040 0.26361 0.4640 0.6279 0.63003 0.6302 (P ) W ) 0.6300 0.9220 0.1362 0.75064 0.7479 0.74806 0.8949 0.34977 0.3052 0.11953 0.11871 0.1187
Table 4.4: Signal Power Distribution Parameters for 2 4 Spot Pattern size to ensure that the receiver is able to receive signals, which may be dicult to ensure practically. The single point and 2 2 spot patterns are clearly unsuitable for use at narrow FOVs as p is larger than p . From Table 4.1, it can be observed that use of a wide FOV of 50 and 90 , allows a low p to be achieved. However, the tradeo is a higher delay spread due to a larger number of multipaths being allowed to reach the detector. A maximum delay spread of 1.85ns is used as a cuto value. Using a conservative estimate of 1/(10 x ), this average delay spread value gives a maximum data rate of approximately 54 Mbit/s. Use of equalizers and other communication techniques such as diversity can be used to increase the data rate further but this value is used as a performance benchmark . From the Fig.4.2, the maximum FOV within this benchmark is seen to be 30 . A larger FOV would also simplify the design of the spot pattern. Using a 20 FOV, for a detector placed at the corner to be able to see a spot, the spot must be less than 2.5m away from the corner. If a 30 FOV detector is used, the spot only need to be less than 6.5m away from the corner, reducing the likelihood of a dropped link if the transmitter is oset from the center. Based on the arguments given above it can be concluded that a receiver FOV of 30 would give a good balance of delay spread and signal power distribution performance. The performance of the various spot patterns at a 30 FOV is summarized in Fig.4.3. The high standard deviation of the single point and 2 2 spot patterns for received power
34
Figure 4.2: Average delay spread as FOV varied with dierent patterns indicate a large variation in the system performance, making the pattern clearly unsuitable for use at this FOV value. The Signal power distribution of various Spot patterns with N = 60,000 at 30 FOV are shown in Fig.4.4. The N value of earlier simulations is 15,000. As the value of N increased , The approximations are more ideal. The values of average received power increases. However, the computational cost also increases. Time taken for calculating signal power distribution with N value 15,000 is 383seconds. With N value 60,000 it is 1593seconds. On the whole, Ratio doesnt change much. So we used N=15,000 to nd optimized spot pattern.
35
(a) Single Spot
(b) Two Spots
Figure 4.3: Signal Power Distribution for Various Spot Patterns With FOVs 30 Spot Pattern Average Power(P ) Std.Dev of Power Ratio W Single Spot Two Spots 22 24 0.72005 0.62693 0.5459 0.4498 (P ) W 0.2429 0.21843 0.20814 0.13521 0.3374 0.3484 0.3812 0.3005
Table 4.5: Signal Power Distribution Parameters for various Spot Pattern at 30 FOV
36
(a) Single Spot
(b) Two Spots
Figure 4.4: Signal Power Distribution for Various Spot Patterns With FOVs at 30 , N=60,000
37
4.2
4.2.1
Optimization of Spot Pattern

Simulated Annealing Technique
Using the optimum FOV as determined in above Section , the spot pattern is next optimized using a simulated annealing (SA) technique for this optimum FOV of 30 . The simulated annealing technique was proposed by Kirkpatrick [18] as a means of nding a global minimum in an optimization process. It is analogous to the annealing process of a molten metal, in which a material nds the lowest possible energy state during the cooling process. The chief advantage of using the SA technique is that it avoids being trapped in a local minimum by introducing a temperature coecient that is slowly reduced. The governing equation is P = exp(df /T ), where P is the probability of accepting an undesired result df, and T is the current temperature. The technique uses a random search process. If a desired result is obtained, the technique accepts the change as the best obtained result. If an undesired result is obtained, that undesired result is accepted with a probability of P. This avoids the algorithm from working on the assumption that the rst minimum is the global minimum, as can be the case when a simple iterative minimization (IM) process is utilized, which simply only accepts desirable results with no probability of accepting undesired results. After a number of iterations, T is reduced by a factor of 0.8 [19]. The initial temperature T0 should result in an average acceptance probability 0 of about 0.8 [20], and this can be determined from df /ln(0 ). In previous work by Yao, Chen and Lim [21], the SA approach was found to be able to optimize the hologram mask to convert a point source into an extended source, with the lowest cost function as compared to the error reduction and input output methods. In this application, the SA algorithm is used to determine the optimum distribution of spots on the room ceiling to achieve a desired performance metric. The algorithm begins with a randomly generated array of spots. A dark pixel represents no illumination by the transmitter, while a light pixel represents a diusing spot. At each iteration, four pixels which are symmetric about the origin are inverted. If the pixels were previously dark, they are made light, and
38 if light, are made dark. The relevant metric such as the ratio of standard deviation of the received power to the average received power is calculated, and compared with the previous result to give df, where an undesired result may or may not be accepted depending on P. The iterations are repeated till the end of one cycle, whereby the annealing temperature is reduced, and the cycle of iterations begins for the new temperature. The process can run till either no improvement can be observed or until a set number of cycles have been reached. In this simulation, initial temperature T0 is considered as 1, minimum temperature as 1e-8. reduction factor is 0.8. The maximum number of iterations limited to 1600, considering time complexity. Average acceptance probability as exp((df )/(k T ). The simulation started with randomly generated spot pattern. At rst, the spots at four corners were inverted and the ratio is been compared with the parent, if it is desired result it will be accepted otherwise it will accepted with probability exp((df )/(k T ). This process continued until 1600 iterations, which results in inversion of all spots at least twice. Since top surface (ceiling) is divided into 2500 spots, it will complexly invert all spots nearly 3 times. 1600 iterations are chosen considering time taken for simulation, which is 6 days.
(a) Optimized Spot Pattern
(b) Optimized Field Pattern
Figure 4.5: Optimized Patterns for 20 Fov Figures 4.5 to 4.9 describes optimized spot patterns at various FOVs and their corre-
39
Figure 4.6: Optimized Patterns for 30 Fov
40
41 sponding power distribution in the room. All the optimized patterns are having more white spots at corner rather than at center. The signal power distribution is approximately uniform over receiving surface. Average received power P and Standard deviation of the average received power P for the optimizing patterns with various FOVs are tabulated. This again shows as FOV increases P increases and P decreases. Hence, the ratio decreases. FOV(deg) Average Power(P ) Std.Dev of Power Ratio W 20 30 50 75 90 0.59210 0.30207 0.46944 0.50958 0.49026 (P ) W 0.057006 0.029246 0.027344 0.027253 .022403 0.0963 0.0938 0.0582 0.0535 0.0457
Table 4.6: Signal Power Distribution Parameters for Optimized Spot Patterns
4.3
Comparison of Results of Conventional Grid Patterns And Optimized Patterns
A comparison is made between the graphs of the conventional grid designs and the optimized results. Using the SA, a reduction of p by a factor of 10 can be achieved. It can be generally observed that this is achieved by having a higher spot density at the edges and corners rather than at the center. Figure 4.9 shows the comparison of results of optimized spot patterns and conventional grid patterns. Ratio of Standard deviated power to the average received power in the optimized patterns is very less. Though, the ratio graph comes down for some of the higher conventional grid patterns like 4 4, 10 10, delay spread is more. So the technique of using Optimization of spot pattern for a given FOV and specied delay spread(in order to
42
Figure 4.10: Comparison of Results for Optimized Patterns to that of Conventional Grid Patterns achieve minimum data rate) will give much better performance than any other conventional grid patterns.
Chapter 5 CONCLUSIONS AND FUTURE SCOPE

5.1 Conclusions
The aim of the thesis is to optimize the spot patterns so as to receive uniform power all through the receiving plane. Multiple Bounce Impulse Response Model and Simulated Annealing technique are used for simulation. conclusions are summarized as follows. Impulse response of the indoor optical wireless channel for multiple receivers and multiple transmitters was derived. Delay spread of the transmitted optical signal due to multipath propagation is found. The eect of FOV on the received signal is observed and compared, FOV value of 30 is found to be the optimum value. The eect of spot pattern on the received signal is observed and discussed. If spot pattern is changed all the parameters will change. Positions of spots have more impact than the number of spots.
43
44 The optimized spot pattern at various FOVs, for which, the received power is approximately uniform through out the receiving plane are obtained. SA technique is been is used for this purpose.
5.2
Future Scope
In this thesis, during optimization process delay spread is been used as a constraint. To nd the eect of delay spread of the signal we can give weighting factors to the delay spread, and to the ratio of standard deviation of average received power to the average received power. By this, we can get optimized pattern in a much better way, which reduces the ratio much less with specied delay spread. Computer Generated Hologram (CGH) can be designed [21] to generate particular optimized spot pattern. Once we design CGH, we can simulates whole system and performance can be compared with the existing models of diversity reception.
References
[1] J.M. Kahn and J.R. Barry, Wireless ifrared communications, Proc. IEEE, Feb. 1997, vol. 85, pp. 265-298. [2] D. Mavrakis: Measurement and prediction of the wideband indoor radio and infrared channels, Ph.D. dissertation, University of Surrey, Oct. 2002. [3] A.C. Boucouvalas, Indoor ambeint light noise and its eects on wireless optical links, IEE Proc.: Optoelectronics, Dec. 1996, vol. 143, pp.334-338. [4] A.J.C. Moreira, R.T. Valadas, and A.M. De Oliveira Durate, Performance of infrared trnasmission systems under ambient light interference, IEE Proc.: Optoelectronics, vol. 143, Dec. 1996, pp 339-346. [5] Z. Ghassemloy, Indoor optical wireless communication systems- Part 1: Review, http://soe.unn.ac.uk/ocr/downloads/part1-rev.pdf,2003. [6] D.J.T. Heatley, D.R. Wisely, I. Neild, and P. Cochrane, Optical wireless: The story so far, IEEE Commun. Mag., Dec. 1998, vol. 36, pp. 72-82. [7] F.R. Gfeller and U.H. Bapst, Wireless in-house data communication via diuse infrared radiation, Proc. IEEE, Apr. 1993, vol. 67, pp. 1474-1486. [8] J.R. Barry, J.M. Khan, W.J. Krause, E.A. Lee, and D.G. Messerschmitt, Simulation of multipath impulse response for wireless optical channels, IEEE J. Select. Areas Commun., Apr. 1993, vol. 11, no.3, pp. 367-379. 45
46 [9] J.R. Barry, Wireless infrared communication, Boston: Kluwer Academic, 1994. [10] J.B. Carruthers and J.M. Kahn, Modelling of non-directed wireless infrared channels, IEEE Trans. Commun., Oct 1997, vol. 45, pp.1260-1268. [11] V. Jungnickel, T. Haustein, A. Forck and C. von Helmolt, 155Mbit/s wireless transmission with imaging infrared receiver, Electron. Lett., 2001, vol. 37, pp. 314-315. [12] Infrared Data Association standards can be obtained at the organizations home page on the Web http://irda.org [13] F. Lopez-Hermandez and M. Betancor, DUSTIN: algorithm for calculation of impulse response on IR wireless indoor channels, Electron. Lett., Oct. 1997, vol. 33, pp. 18041806. [14] J. M. Kahn, W. J. Krause, and J. B. Carruthers, Experimental characterization of non-directed indoor infrared channels, IEEE Trans. Commun., February-March-April 1995, vol. 43, no. 2-3-4, pp. 1613-1623. [15] J. B. Carruthers and P. Kannan, Iterative site-based modeling for wireless infrared channels, IEEE Transactions on Antennas and Propagation, May 2002, vol. 50, pp. 759 -765. [16] H. Hashemi, G. Yun, M. Kavehrad, F. Behbahani, and P. Galko, Frequency response measurements of the wireless indoor channel at infrared optics, in International Zurich Seminar on Digital Communications, Mar. 1994. [17] T.S. Rappaport, Wireless Communications, Principles and Practices, Prentice-Hall, Upper Saddle River, 2002. [18] S. Kirkpatrick, C. D. Jr. Gerlatt, and M. P. Vecchi, Optimization by simulated annealing, Science, 1983, vol. 220, pp. 671-680.
47 [19] S. Kirkpatrick, C. D. Jr. Gerlatt, and M. P. Vecchi, Optimization by simulated annealing, IBM Research Report RC 9355, 1982. [20] S. Kirkpatrick, Optimization by simulated asnnealing - Quantitative Studies, J. Stat. Phys, 1984, vol. 34, pp. 975-986. [21] J. P. Yao, G. Chen and T. K. Lim, Holographic diuser for diuse infrared wireless home networking, Opt. Eng, 2003, vol. 42, pp.317-324. [22] P. Kannan, Iterative site-based modeling for wireless infrared channels: an analysis and implementation, Masters thesis, Boston University, Dept. of Electrical and Computer Engineering, 2001. [23] J.B. Carruthers and J. M. Kahn, Angle diversity for nondirected wireless infrared communication, IEEE Trans. Commun., June 2000. [24] F. J. Lopez-Hernandez, R. Perez-Jimenez, and A. Santamaria, Ray-tracing algorithms for fast calculation of the channel impulse response on diuse IR wireless indoor channels, Opt. Eng., 2000, vol. 39, pp. 27752780. [25] W. A. Arbaugh, Wireless security is dierent, Computer, 2000, vol. 36, pp. 99-101. [26] Jerey B. Carruthers, Sarah M. Carroll, Prasanna Kannan, Propagation modeling for indoor optical wireless communications using fast multireceiver channel estimation. DRAFT, Oct. 2002. [27] P. L. Eardley, D. R. Wisely, D. Wood and P. McLee, Holograms for optical wireless LANs, Opt. Eng., 2002, vol. 41, pp. 899-910. [28] J. P. Yao, G. Chen and T. K. Lim, Holographic diuser for diuse infrared wireless home networking, Opt. Eng., Feb. 2003, vol. 42, pp. 317-324. [29] K. Smitha and J.John, Propagation measurements of indoor infrared channels, Proc. Photonics-2004, Dec. 2004.
48 [30] A.Sivabalan, and J. John, Modeling and simulation of indoor optical wireless channels, Proc. IEEE, TENCON 2003. [31] Smitha K, Studies on propagation properties of indoor optical wireless diuse channels, Masters thesis, May 2005. [32] N. Janardhan, Studies on diused indoor optical wireless systems, Masters thesis, Feb. 2001. [33] Chaturi Singh, J. John, and Y. N. Singh, Simulation and optimization of the intensity prole of an optical transmitter for high-speed wireless local area networks., Proc. Phtonics, Cochin, Dec. 2004. [34] Chaturi Singh, J. John, Y. N. Singh, K. K. Tripathi, Design aspects of highperformance indoor optical wireless transreceivers., ICPWC 2005, New Delhi, Jan. 2005.
Propagation modelling for indoor wireless cornrnunication

by W.K. Tam and V.N. Tran
It is important to characterise the indoor radio propagation channel to ensure satisfactory performance of a wireless communication system. Site measurements can be costly; propagation models have been developed as a suitable low-cost alternative. The existing models can be classified into two major classes: statistical models and site-specific propagation models. Statistical models rely on measurement data; site-specific propagation models are based on electromagnetic-wave propagation theory. The ray-tracing technique is very useful in site-specific propagation modelling. This paper gives an overview of indoor propagation modelling and concentrates on a discussion of the raytracing modelling technique because of its practical appeal and its applicability t o any environment.
1
Introduction
The explosion in wireless communications has resulted in new technologies and new applications for the personal use of radio frequencies;personal communication systems (PCS) are now being developed worldwide. An important consideration for the successful implementation of a PCS is indoor wireless communication. This covers a wide variety of situations ranging from communication with individuals walking in residential or office buildings, hospitals, factories, etc. to fixed stations sending control messages to robots in motion in assembly lines in a factory environment. The indoor environment is prone to interference. Owing to reflection, refraction and scattering of radio waves by structures inside a building, the transmitted signal most often reaches the receiver by more than one path, resulting in a phenomenon known as multipath fading. Multipath causes deep fading and pulse spreading of the signal and hence intersymbol interference can be caused in a digital radio system. Although multipath interference seriously degrades the performance of communication svstems. little can be done to eliminate it. However, if we characterise the multipath medium well, the siting of transmitters can be selected to achieve good propagation performance and hence to achieve better connectivity. In the extreme, if service providers had to characterise the propagation medium by taking radio propagation measurements of every building in which they deployed their system, the installation cost would be very high due to the labour costs involved.
Therefore, there has been a special interest in developing a propagation model to predict the propagation characteristics of the indoor environment. Once a propagation model has been verified, an environment can be quickly entered into such a model to provide propagation characteristics for initial evaluation. It remains to carry out quick measurements at positions where the signal is poor. This approach is much cheaper than an exhaustive measurement programme and thus installation costs are dramatically reduced. Many propagation models have been developed to characterise indoor radio propagation. They can be classified into two major classes: statistical models and sitespecific propagation models. The general statistical impulse-response modelling of the multipath fading channel was first suggested by Turin.' It has been subsequently used in the measurement, modelling, and simulation of the mobile radio channel by investigators following Turin's line of workJ4 and by other researchers.'" More recently, the statistical impulseresponse approach has been used directly or indirectly in .. indoor radio propagation channel modelling.'-" Keenan and Motley" attempted to formulate a statistical lossdistance model to predict radio coverage based on a few building parameters. Some other statistical loss-distance models'' have been developed recently with different formulas and statistical data, In contrast, site-specific propagation models are based on the use of electromagnetic wave propagation theory to characterise indoor radio Propagation. Recent reports of the application of the analytical ray-tracing technique to site-specific
''
ELECTRONICS Rr COMMUNICATION ENGINEERING JOURNAL OCTOBER 1995
22 1
Authorized licensed use limited to: Aalborg Universitetsbibliotek. Downloaded on September 21, 2009 at 08:36 from IEEE Xplore. Restrictions apply.
indoor radio propagation modelling have appeared in a number of papers.'62' This technique has been proposed to predict path loss, the timeinvariant impulse response, and the RMS delay spread. It promises to provide fast and accurate prediction of indoor radio coverage and channel impulse response. Generally, sitespecific propagation modelling is preferred in many practical situations.
wall
signal encounters. Consequently, the signal decay in an indoor environment can be characterised by the construction materials.
Fading Properties
In a typical indoor radio system, a fixed base station antenna installed in an elevated position communicates with a number of portable radios or fixed wall receiving terminals inside a building. Owing to the Fig. 1 Multipath propagation reflection, refraction and 2 Indoor electromagnetic-wave propagation scattering of radio waves by structures inside a building, the transmitted signal most often reaches a Indoor radio channels do not suffer from the receiver by more than one path, resulting in a environmental effects of snow, rain, hail, clouds or phenomenon known as multipath fading, as shown in temperature inversion as do outdoor radio channels but Fig. 1.The signal components arriving from the indirect because of the variation of building size, shape, structure, paths and the direct path (if it exists) combine and layout of rooms and, most importantly, the type of produce a distorted version of the transmitted signal. In construction materials, electromagnetic-wave propagation narrow-band transmission, the multipath medium inside a building is a more complex multipath structure causes fluctuation in the envelope and phase of the than that of terrestrial mobile radio channels. For example, received signal. In wide-band pulse transmission, the a factory building is quite different to an office building effect is to produce a series of delayed and attenuated both in its structure and in the materials used. The pulses (echoes) for each transmitted pulse. Fig. 2 is a variation of type of materials used in internal partitions, simulated impulse response for a laboratory to outside walls, ceilings and floors, as we!l as the size and illustrate the multipath fading effect on wide-band pulse percentage of windows, age of building, people density and transmission. activity are also factors which complicate indoor In the case of continuous-wave signal transmission, it electromagnetic-wave propagation. has been shown that the statistics of the received signal envelope are described by a Rayleigh distribution if no Path loss dominant signal path (i.e. a line-of-sight path or strong The sqatial distribution of power at a distance d from a reflected path) exists between the receiver and transmitter is, in general, a decreasing function of d. This transmitter. If a dominant path does exist the statistics of function is represented by a distance power law of the form the signal envelope are Rician. P= l l d " . For free space, m is equal to 2 and it is said that the In digital pulse transmission, the delay spread of a power gain follows an inverse square law. In an enclosed multipath signal is important since it affects the data transmission rate. Three parameters are often used to environment, however, this is not true anymore. Saleh and Valenzuela' and Bultitude" showed that when the describe the temporal spread of the channel: the mean transmitter and receiver were placed in the same hallway, excess delay, the RMS delay spread and the excess-delay in sight of each other, the power decayed with a value of m spread. Mean excess delay describes the average ranging of 1.5 to 1.8; when the receiver was located within propagation delay relative to the first-amving signal a room off the hallway, m ranged from 3 to 4. component. The RMS delay spread measures the temporal spread of the power delay profile about the mean excess The path loss also varies with frequency. Owen and delay. The excess-delay spread (X dB) indicates the Pundey' made measurements inside office building at maximum delay, relative to the first-arriving signal 900 MHz and 1650 MHz. The measurement results component, at which the multipath energy falls to X dB indicate that loss through floors is greater at the higher frequency. It is found that at wavelengths in the millimetre below the peak received level. These parameters are range the radio wave cannot penetrate most common loosely related to outage and bit error rates for different building materials such as brick and concrete block and digital modulation schemes that do not use equalisation. A rule of thumb is that a bit-error rate (BER) of less than that signal attenuation occurs more rapidly with distance. will occur if the channel RMS delay spread is less than 0.2 Therefore the millimetre waveband seems to be a good of the symbol duration. choice for providing broadband services in a high-capacity Saleh and Valenzuela* observed from their frequency-reuse environment. measurement at 1.5 GHz in a two-storey narrow building AIe~ande?','~has given the values of m according to the that the RMS time delay spread values extended up to building materials used in the environment. The degree 01 50 ns. Devasirvatham,L" making measurements in a large signal attenuation depends on the type of materials the 222 ELECTRONICS & COMMUNICATION ENGINEERING JOURNAL OCTOBER 1995
building at 850 MHz, observed median RMS time delay spread values of 125 ns. Rappapoff6 reported results of measurement at 1300 MHz in five factory buildings. Multipath spreads ranged from 40 to 800 ns. Mean excessdelay and RMS delay spread values ranged from 30 to 300 ns. The different results obtained were due to the differencein the type of building measured. Delay spreads were found to be affected by factory inventory, buildingconstruction materials, building age, wall locations, and ceiling heights. Buildings with more metal material have larger delay spreads.
3
Overview of indoor propagation models
Both of the two main classes of indoor radio propagation model - statistical and site-specific -have strengths and limitations when applied to the design and installation of indoor wireless systems.
Statistical models A general statistical impulse-response model for the multipath fading channel was first suggested byTurin'.'for outdoor radio propagation. Recently, this statistical impulse-response approach has been used for indoor radio channel propagation modelling. Saleh and Valenzuela' used their measurement results in a medium-size two-storey office building, together with measurement results from other researchers, to develop a statistical model of the indoor radio channel for the simulation and analysis of various indoor communication schemes. The model was shown to fit the measurements and may be extended to other buildings by adjusting its parameters. The model assumes that the multipath components arrive in clusters. The received amplitude of each component is an independent Rayleigh random variable, with a variance that decays exponentially with propagation delay, as well as with time delays, within a cluster. The corresponding phase angles for each component are independent uniform random variables The over [0,2n]. clusters and multipath components within a cluster form a Poisson-arrival process with different
the clusters have exponentially distributed inter-arrival times. The formation of the clusters is related to the building structure, while the multipath components within each cluster are formed by multiple reflection from objects in the vicinity of the transmitter and the receiver. The model has enough flexibility to permit reasonably accurate fitting of the measured channel responses and is simple enough for simulation. The model is successful for application in office environments but its application to multipath data collected in several factory environments has been
0
statistical model for indoor radio propagation based on the results of extensive multipath propagation measurements in two office buildings. The data base for this model development consists of 12000 impulse response profiles of the channel collected in these office buildings. In this model, the data amval time is modelled as a modified Poisson distribution and the amplitudes were found to be lognormally-distributed over both local and global areas, with a log-mean value that decreases almost linearly with increasing excess-delay. The simulation results of the model agreed with measurement results in office environments. Unfortunately, application of this model to impulse- response data collection in factory environments was not successful. Keenan and Motley" formulated a radio coverage prediction model based on a few building parameters. The parameters in the formulae of this model were derived from measurement data. The model provides a quick and simple way to predict the path loss, in decibels, in an indoor environment. It is useful for an initial coverage prediction. However, the model only provides path loss information and may not work well in some complex indoor environments. Seidel and Rappaport" have proposed path loss models based on data measured at 914 MHz.The models are based on a simple d" exponential path loss against distance relationship. In open-plan buildings, the path loss exponent n is close to 2. For environments with many more obstructions between the transmitter and the receiver, the path loss exponent can be much higher. Statistical distance-dependent path loss models are useful for understanding the propagation of radio waves in buildings. However, exhaustive measurements were required to obtain the data to determine the appropriate parameters for the models for these particular buildings. Rappaport et al.' have developed a statistical channel impulse response model for the design of radio communication systems for use in factories and open-plan buildings. The model, called SIRCIM (simulation of indoor radio-channel impulse-response models), is being used by many researchers to generate, on a computer, impulse-
-50
-100
-150 -200 -250
H
a ,
-300
-350
100
200
300
400
500
600
700
800
time ns
ELECTRONICS & COMMUNICATION ENGINEERING JOURNAL OCTOBER 1995
223
signal ray
reflectedray
parallel plate (partition)
transmitted ray
signal ray
reflected ray
parallel plate,, (partition)
E,
transmitted ray
Fig 3 A signal wave incident obliquely on a parallel plate: (a) electric vector perpendicular to plane of incidence; (b) electric vector parallel to plane of incidence
response and path-loss measurements in office and factory buildings. The model incorporates first- and second-order statistics to characterise the discrete impulse responses of indoor radio channels for both line-of-sight (LOS) and obstructed (OBS) topographies. SIRCIM can be applied throughout the low microwave band, and the code is written to work up to 60 GHz. Data files produced by SIRCIM contain amplitudes, phases, time delays, and path loss for individual multipath components, as well as large scale path loss, so that a complete propagation model is created. The most salient feature of the model is that it reproduces multipath channel conditions that are very realistic since they are based on real-world measurements, and may thus be used for meaningful system analysis in factories and open-planbuildings. It is possible to simulate other multipath channels, such as office buildings, based on the framework of the model by changing the values of the model. This model is attractive for the characterisation of indoor propagation channels. However, it relies on extensive measurement data to determine the appropriate parameters to model a particular type of indoor environment.
Site-specific propagation models Sitespecific propagation models are based on electromagnetic-wavepropagation theory to characterise indoor radio propagation. Unlike statistical models, site specific propagation models do not rely on extensive measurement, but a greater detail of the indoor environment is required to obtain an accurate prediction of signal propagation inside a building. In theory, electromagnetic-wave propagation characteristics could be exactly computed by solving Maxwell's equations with the building geometry as boundary conditions. Unfortunately, this approach requires very complex mathematical operations and requires considerable computing power, beyond that of current microcomputers. Hence it is not economical for the characterisation of indoor radio wave propagation. Therefore, approximate numerical methods are of interest. Ray tracing is an intuitively appealing method for calculating radio signal strength, timeinvariant impulse response, RMS delay spread and related parameters in an indoor environment.'621 The concept of ray-tracingmodelling is based on the fact that high-frequency radio waves behave in a ray-like fashion. Therefore signal propagation can be modelled as ray propagation. By using the concept of ray-tracing,rays may be launched from a transmitter location and the interaction of the rays with partitions within a building modelled using well-known reflection and transmission theory. Two types of ray-tracing methods - the image and the bruteforce ray-tracingmethod"- are being used in the characterisation of indoor electromagnetic-wavepropagation. For scatters bounded by plane faces it is convenient to employ the image method to mirror the radio wave source at a particular face. The point where the mirror face intersects the line connecting the transmitter image and the receiver is the point at which specular reflection occurs. This method is well suited to radio propagation analysis in the case of geometries of low complexity and where a low number of reflections are considered. The bruteforce ray-tracingmethod considers a bundle of transmitted rays that may or may not reach the receiver. The number of rays employed and the distance from the transmitter to the receiver location determine the available spatial resolution and hence the accuracy of the model. This method requires more computing power than the image method. Ray tracing can be much less demanding of computation than methods based on Maxwell's equations. With the computing powers currently available on personal computers and workstations, the ray-tracing approach provides a challenging but feasible method of propagation modelling. Reliable sitespecific ray-tracing propagation prediction models for each building based on its detailed geometry and construction can be very effective tools in designing indoor communication systems.
4
Modelling by ray tracing
The ray-tracing approach approximates the scattering of
224
electromagnetic waves by simple reflection and refraction. The degree of transmission and reflection of a signal through and off an obstacle is related to the complex permittivities of the obstacle.
where
Transmission and reflection ofa radio signal When a signal is transmitted through or reflected off a wall or a partition, the degree of signal attenuation and the amount of phase change depend on the complex transmission and reflection coefficients, respectively. These are computed from the complex permittivities of the materials the signal rays encounter. Other factors affecting the transmission and reflection of the signal are the angle of incidence and the relative polarisation. The complex transmission coefficient is defined as the ratio of the transmitted to the incident electric-field strengths and the complex reflection coefficient is defined as the ratio of the reflected to the incident electric-field strengths. Referring to Fig. 3, a signal ray is incident on a parallel plate, which may be a partition or a wall. In Fig. 3a, the signal has a horizontal polarisation with reference to the parallel plate. The complex transmission and reflection coefficients for horizontal polarisation are:
t, = ta x tBis a complex transmission coefficient ta = the complex transmission coefficient at A tB= the complex transmission coefficient at B a = attentuation factor and d = the distance travelled by a signal ray in the material.
The attenuation factor, a, is given as:
where
E', =
E','
real part of the relative complex permittivity of the material imaginary part of the relative complex permittivity of the material and wavelength.
Line-ofsight signal strength calculation Fig. 4 shows a signal transmitted through a partition. The received signal strength at the receiver from a line-ofsight path is given by:
reflection coef. r = E, =
~
CO%~
(&- r n L @ l ) '
~
E,
cos $1 + (E" - Sk5#JI) '
(2) where
In Fig. 36 the signal is polarised vertically with respect to the parallel plate. The complex transmission and reflection coefficients for vertical polarisation are:
E, = the source signal strength from the transmitter E,~,, = received lineof-sight signal strength at the
receiver do = lineof-sight distance from transmitter to receiver t, = complex transmission coefficient h = wavelength of the signal.
transmission coef. t = Et
-
E,
2 COS@, cos + (1- %sin?@l) '
(3)
(3"
reflection coef. r =
Reflected signal strength calculation Fig. 5 shows a first order reflection path of the signal. The signal strength at the receiver due to this path is given by:
E,
~
@s@, - ($ =
(4)
E,
&COS&+
(&-sin2@1)"
partition transmitter
In eqns. 1-4, E is the complex permittivity of the parallel plate, E~ is the permittivity of free space, is the angle of incidence, and E,, E, and Et are the incident, reflected and transmitted electric field strengths, respectively. When a signal ray, as shown in Fig. 3, encounters a parallel plate (a wall or partition) with thickness D the signal is attenuated while going through the material. The complex transmission coefficient is then given as
-_
I -
d, - _ .
receiver
() 5
Fig. 4
Line-of-sight signal
ELECTRONICS 81COMMUNICATION ENGINEERING JOURNAL OCTOBER 1995
225
Fig. 6 shows a second order reflection path of the signal. The signal strength at the receiver due to this path is given by:
and the multiple reflection signal strength at the receiver is given by:
where
E,
signal strength from the transmitter first order reflected signal strength at receiver E2R = second order reflected signal strength at receiver E m R = multiple reflection signal strength at receiver 11 + l2 +...+ Lk, k any positive integer = total reflection distance r,,, r,, ..., r,, = complex reflection coefficientsat each reflection point 1.2, ...,m, respectively til, tc2,..., t,, = complex transmission coefficients at each wall or partition 1.2, ....,n,respectively d = wavelength of the signal.
=
=
EL^
Raytracing method In ray-tracingmethods, the locations of transmitters and receivers are assigned to points referenced by three
Fig. 5
First order reflected signal
'
Fig. 6 Second order reflected signal
dimensional co-ordinates. The walls, partitions, ceilings and floors in an indoor environment are usually modelled as plane surfaces of given thickness and complex permittivity. For simplication, curved surfaces can be modelled as piecewise planar surfaces. The rays from the transmitter antenna are reflected off walls, partitions, ceilings, floors and tables etc. or transmitted through walls and partitions etc. to arrive at the receiver. As already mentioned, two common methods - the image method and the brute-force method - have been developed to trace the rays from the transmitter to the receiver. (a) Image rneth~d:'~.'~ method assumes every This plane face in an indoor environment to be a mirror. For line-of-sight propagation, it is easy to trace the ray by connecting the transmitter and receiver. For singlereflection propagation, the radio source is mirrored at a particular face.The point of intersection of the mirror face and the line connecting the transmitter image to the receiver is the point where specular reflection occurs. The single-reflection propagation path can then be obtained by connecting the source point, reflection point and receiver point. For repeated reflection, the image of the radio source with reference to a particular plane face is found first. The next step is to find an image of the source image with reference to another plane face where the second point of reflection will be located. Following the same rule, all the points of reflection at the relevant plane faces can be obtained. The multiple-reflection propagation paths can then be obtained. The method described above, starting at the source image, is referred to as the forward ray-tracing method. It is also possible to be start at the receiver image and trace back to the transmitter. This is called the backward ray-tracing method. Consider a rectangular room as shown in Fig. 7. The transmitting antenna is located at point T, while the receiving antenna is located at point R To trace the path from T to R, reflecting off walls 2, 4 and 3, three images have to be found. First the first order image, 13. of the receiver antenna in wall 3 is found. Then, the second order image, at point is found by reflecting the first order image in the semi-intiniteplane containing wall 4. Finally, the co-ordinates of the highest order image, at point 13,4,2. are found by reflecting in wall 2. Once all the images have been found, the complete path and all the reflection points can be found as shown in Fig. 7. The signal strength of this propagation path can then be calculated using eqn. 10. (b) Brute-force rayfracing method:'* This method accounts for all possible propagation paths. The transmitters and receivers are modelled as points at discrete locations in three-dimensional space. All the possible angles of departure and arrival at the transmitters and receivers are considered to determine all possible rays that may leave the transmitter and arrive at the receiver. Ray tracing is accomplished by an exhaustive search of a ray tree taking into acco.int decomposition of the ray at each planar intersection. First the model determines whether a line-of-sight path exists and if so computes the
226
received signal. Next, the model traces a source ray in a specified ,,A direction and detects whether an object intersection occurs. If no intersection is found, the process stops and a new source ray in a direction making an angle with the 4 original ray is initiated. Once an intersection has occurred, 7 a check is made to see whether the ray can be considered to have reached any of the specified receiver locations. If the ray is found to reach a receiver location, the received signal is computed. After checking the reception, the incident ray is divided into a transmitted and a reflected ray, each of which is traced to the next intersection in the same way. This recursion continues until the ray intensity falls below a specified threshold or no further intersections occur. As multiple scattering of a ray will not contribute significantly to the received power since the amplitude of these scattered rays decreases rapidly with distance, the existing ray-tracing models using this approach do not include multiple scattering and also do not trace scattered 5 Conclusion rays recursively. At each step in the creation of a ray tree, the In this paper, we have pointed out the importance of corresponding ray segment is tested to see whether it can propagation models in the development of indoor wireless be considered to have reached specified receiving communications. Propagation models provide estimates of locations. To do this, a reception sphere is constructed signal strength and time dispersion in many indoor about the receiving location with a radius proportional to environments. These data are valuable in the design and the unfolded path length from transmitter to receiver and installation of indoor radio systems. the angular spacing between neighbouring rays at the Propagation models can be classified into statistical source. If the ray intersects the sphere, the ray is taken as models and site-specific propagation models. Statistical contributing to the received signal. Otherwise, the ray is models need extensive measurement data and do not treated as not having reached the receiver location. provide site-specific information. Site-specific propagation models provide site-specific information hut require The reception sphere effectively accounts for the considerable detail of the layout of the indoor divergence of the rays from the source. For sufficiently small ray separation 0, the ray intercepting the sphere will environment. be an accurate measure of the ray that would pass directly Statistical models can he used for preliminary design through the receiving point, The physical interpretation of and analysis whereas site-specific propagation models the reception sphere can be justified with the aid of Fig. 8. can be used for fast and accurate prediction of indoor This Figure is a two-dimensional representation of a ray being traced. adjacent ray Two adjacent rays launched at t 0 relative to the test ray are also shown. correct Note that in three dimensions any reception sphere ray will have more than two adjacent rays and angular separation of the d adjacent rays will not necessarily coincide with the co-ordinate axes. As shown in Fig. 8, a reception sphere with the correct radius (= / , W 2 ) can receive exactly one of the rays. If the radius is too large. two of the rays could be received and would, in effect, count the same ray path twice. Likewise, if the radius is transmiller I too small, it is possible that none of the rays will intercept the sphere and undersized reception sphere I the ray path energy will he excluded. oversized reception sphere The path loss error due to perceiving two rays would be a few decibels. A missed specular ray could lead to a Fig. 8 Two-dimensional view of the reception sphere. The unfolded ray path length is d. The reception sphere radius varies with Hand d much larger error if a significant
I\
ELECTRONICS Kr COMMUNICATION ENGINEERING JOURNAL OCTOBER 1995
227
r a d i o coverage. Site-specific propagation modelling b y solving Maxwells e q u a t i o n s i s costly a n d impractical. Ray t r a c i n g i s p r o m i s i n g as an a p p r o x i m a t e m e t h o d for site-specific propagation modelling. Ray t r a c i n g m a k e s u s e of the fact that all objects of i n t e r e s t within the propagation e n v i r o n m e n t a r e l a r g e r t h a n a wavelength, so it is suitable f o r application in the h i g h e r radio f r e q u e n c y r a n g e s . The inclusion of diffraction t h e o r y c a n b r o a d e n i t s application to l o w e r radio frequencies. The a c c u r a c y of ray-tracing t e c h n i q u e s d e p e n d s heavily on t h e a c c u r a c y a n d detail of the site-specific representation of the propagation m e d i u m . The availability of fast interactive-computing e n v i r o n m e n t s a n d high-accuracy g r a p h i c s d a t a b a s e s greatly i m p r o v e s the efficiency a n d a c c u r a c y of ray-tracing modelling. It i s r e c o m m e n d e d t h a t high-accuracy ray-tracing m o d e l s he developed. Nevertheless, a simple, fast a n d low-cost PC-based rayt r a c i n g propagation m o d e l i s helpful to e n g i n e e r s in the preliminary d e s i g n a n d analysis of i n d o o r wireless s y s t e m s . The d e v e l o p m e n t of s u c h a m o d e l i s highly desirable.
communications, IEEE Commun. Mag., June 1987, 25, (6). pp.5-12 13 KEENAN, J.M.. and MOTLEY, AJ.: Radio coverage in buildings, Er. Telerom. Technol. J.. January 1990, 8, ( l ) , pp.19-24 14 TANG, Y., and SOBOL, H.: Modeling of indoor microwave propagation for PCS system. Proc. IEEE Int. Conf. on Communication 93. vol. 3, Geneva, Switzerland, 22nd-26th May 1993, pp.161CL1614 15 RAPPAPORT, T.S.. and SANDHU, S.: Radio-wave propagation for emerging wireless personal-communication systems, IEEE Antennas & Propagation Magazine, October 1994,36, (5), pp.14-24 16 VALENZUELA, R A : A ray tracing approach to predicting indoor wireless transmission. Proc. 43rd IEEE Vehicular Technology Conference,NJ. USA, 1993, pp.214-218 17 MCKOWN, J.W., and HAMILTON, R.L.: Ray tracing as a design tool for radio networks. IEEE Network, November 1991,5, (6),pp.27-30 18 SEIDEL, S.Y., and RAPPAP0RT.T.S.: Araytracingtechnique to predict path loss and delay spread inside buildings. Proc IEEE GLOBECOM 92 Conference, Orlando, USA, 6th-9th December, 1992,pp.649-653 19 HOLT, T., PAHLAVAN, IC, and LEE, J.F.: Ray tracinR algorithm for indoor radio propagation modeling. Proc 3rd References IEEE Int. Symp. on Personal, Indoor and Mobile Radio Communications, Boston, MG 19th-21st October 1992 1 TURIN. G.L.: Communication through noisy, random20 DRIESSEN, P.F., GIMERSKY, M., and RHODES, T.: Ray multipath channels. 1956 IRE Convention Record, part 4, model of indoor propagation. Proc. 2nd Annual Virginia pp.154-166 Technology Symp. on Wireless Personal Communication, 2 TURIN.G.L.,CLAPP,F.D.,JOHNSTON,T.L.,FINE,S.B.,and Blacksburg, VA, 17th-19th June 1992 LAVRY. D.: A statistical model of urban multipath 21 LAURENSON, D.I., SHE1KH.A.U.H.. and McLAUGHLIN. S.: propagation, IEEE Trans. Veh. Technol., February 1972, Characterization of the indoor mobile channel using a ray VT-21, pp.1-9 tracing technique. Proc. 1992 IEEE Int. Conf. on Selected 3 SUZUKI, H.: A statistical model for urban radio propagation, Topics in Wireless Communications, Vancouver, BC, IEEE Trans. Commun.,July 1977,COM-25, pp.673-680 25th-26th June 1992,pp.6568 4 HASHEMI, H.: Simulation of the urban radio propagation 22 OWEN, F.C., and PUNDEY, C.D.: In-building propagation at channel, IEEE Trans. Veh. Technol., August 1979, VT-28, 900 MHz and 1650 MHz lor digital cordless telephone. 6th pp.213-224 Int. Conf. on Antennas and Propagation, ICAP 89, Part 2: 5 BAJWA, AS.: UHFwideband statistical model and simulation Propagation, 1989, pp.276-281 of mobile radio multipath propagation effects, IEE Proc. F., 23 ALEXANDER, SE.: Characterising buildings for propagation August 1985,132, (51, pp.327-333 September 1983. 19, (20). p.860 at 900 MHz, Electron. Lett., 6 RAPPAPORT, T.S., SEIDEL. S.Y., and SINGH, R : 900 MHz 24 ALEXANDER, S E : The propagation of radio signals at 900 multipath propagation measurements for US digital cellular MHz within buildings. IEE Colloquium Digest 1986/030, radio telephone, IEEE Trans. Veh. Technol., May 1990, Propagation in confined spaces and tunnels, IEE, London, VT-39, (2). pp.132-139 UK 1986,pp.7/1-7/4 7 RAPPAPORT, T.S., SEIDEL, S.Y., and TAKAMIZAWA, IC: 25 DEVASIRVATHAM, D.M.J.: Time delay spread Statistical channel impulse response models for factory and measurements of wideband radio signals within a building. open plan building radio communication system design, Electron. Lett.,8th November 19&?4,20, (23). pp.951-952 IEEE Trans. Commun.. May 1991.39, (5). pp.794-807 26 RAPPAPORT, T.S.: Characterization of UHF multipath radio 8 SALEH. A A M . , and VALENZUELA, RA.: A statistical model channels in factory buildings, IEEE Trans. Antennas & lor indoor multipath propagation, IEEE J. Sel. Areas Propagation,August 1989,AF-37, pp.1058-1069 Commun., February 1987, SAC-5, (2), pp.12&137 27 SEIDEL, S.Y., and RAPPAPORT, T.S.: 914 MHz path loss 9 HASHEMI, H., THOLL. D., and MORRISON, G.: Statistical prediction models for indoor wireless communications in modelling of the indoor radio propagation channel part I. multifloored buidlings,IEEE Trans. Antennas & Propagation. Proc. IEEE Vehicular Technology Conference, WC92, February 1992,AP-40, (2). pp.207-217 Denver, CO, May 1992, pp.33&342 10 HASHEMI, H., LEE, D., and EHMAN. D.: Statistical modeling of the indoor radio propagation channel: part 11. Proc. IEEE Vehicular Technology Conference, WC92, Denver, CO., May 1992, pp.839843 11 HASHEMI, H.: Impulse response modelling of indoor radio propagation channels, IEEE J. Sel. Areas Commun.. September 1993, SAC-1 1, pp.1788-1796 12 BULTITUDE, R.J.C.: Measurement, characterization and modeling of indoor 800/900 MHz radio channels for digital
0IEE:1995
First received 1st March and in revised form 14th July 1995
The authors are with the Department of Communication and Electronic Engineering, Royal Melbourne Institute of Technology, GPO Box 247fiV, Melbourne, Victoria 3001, Australia.
228
662
IEEE TRANSACTIONS ON MICROWAVE THEORY AND TECHNIQUES, VOL. 50, NO. 3, MARCH 2002
Propagation Prediction Models for Wireless Communication Systems

Magdy F. Iskander, Fellow, IEEE, and Zhengqing Yun, Member, IEEE
Invited Paper
AbstractA comprehensive review of the propagation prediction models for terrestrial wireless communication systems is presented in this paper. The classic empirical models are briefly described and the focus is placed on the application of ray-tracing techniques to the development of deterministic propagation models. Schemes to increase the computational efficiency and accuracy are discussed. Traditional statistical models are also briefly reviewed for completeness. New challenges to the propagation prediction are described and some new approaches for meeting these challenges are presented. Index TermsChannel characterization, delay spread, path loss, propagation prediction model, ray tracing, wireless communications.
I. INTRODUCTION EINRICH Rudolf Hertz observed in 1886 [1] the transmission of electromagnetic waves and, hence, realized the long-debated Maxwells predictions of wave propagation. The first milestone on the road to wireless communications, however, was realized by Guglielmo Marconi who conducted his famous experiments from 1894 to 1901. Marconi demonstrated in 1901 that the radio wave could provide continuous contact with ships sailing the English Channel [2]. After that, two-way radio communications and broadcasting systems were developed in the 1930s and 1940s. In the 1960s and 1970s, the cellular concept was developed in Bell Laboratories, Holmdel, NJ [3]. The first generation of wireless mobile communication systems appeared in the 1980s and was based on analog technology with FM modulation. Examples of first-generation cellular systems are the Nordic Mobile Telephone (NMT) and Advanced Mobile Phone System (AMPS). In the early 1990s, the second-generation (2G) digital cellular systems were developed with varying standards. Examples include the Goupe Special Mobile [(GSM), now Global System for Mobile Communications)] in the U.K., IS-54/136 and IS-95 in the U.S., and the Personal Digital Cellular (PDC) in Japan. In general, the 2G systems have improved spectral efficiency and voice quality.
Manuscript received October 26, 2001. M. F. Iskander was with the Electrical and Computer Engineering Department, University of Utah, Salt Lake City, UT 84112-9206 USA. He is now with the Hawaii Center for Advanced Communications, College of Engineering, University of Hawaii, Honolulu, HI 96822 USA. Z. Yun is with the Electrical and Computer Engineering Department, University of Utah, Salt Lake City, UT 84112-9206 USA. Publisher Item Identifier S 0018-9480(02)01987-7.
The third generation (3G) of wireless communications are currently being developed in different regions of the world. The 3G systems will provide multimedia services and satisfy more requirements such as applications and communications anytime and anywhere [4]. To this end, wide-band and broad-band radio technologies will be necessary. The examples of 3G standards are International Mobile Telecommunications 2000 (IMT-2000), CDMA-2000, and NTT DoCoMo W-CDMA systems. Although the 3G will begin service in 2001/2002 and reach full deployment by 2005, the fourth-generation (4G) systems are currently being discussed [5]. The 4G system will provide an all-IP network that integrates several services available at present and provides new ones, including broadcast, cellular, cordless, WLAN, and short-range communication systems. The general trend in the development of wireless communication is the use of higher data rates (broader frequency band), propagation in more complex environments, employment of smart antennas, and use of multiple-input multiple-output (MIMO) systems. A. Importance of Propagation Prediction Before implementing designs and confirming planning of wireless communication systems, accurate propagation characteristics of the environment should be known. Propagation prediction usually provides two types of parameters corresponding to the large-scale path loss and small-scale fading statistics. The path-loss information is vital for the determination of coverage of a base-station (BS) placement and in optimizing it. The small-scale parameters usually provide statistical information on local field variations and this, in turn, leads to the calculation of important parameters that help improve receiver (Rx) designs and combat the multipath fading. Without propagation predictions, these parameter estimations can only be obtained by field measurements which are time consuming and expensive. The following subsections provide a brief description of deterministic models, statistical models, and challenges facing the development of accurate and sufficiently general propagation prediction models. B. Empirical, Theoretical, and Site-Specific Models The path-loss prediction models can be roughly divided into three types, i.e., the empirical, theoretical, and site-specific
00189480/02$17.00 2002 IEEE
ISKANDER AND YUN: PROPAGATION PREDICTION MODELS FOR WIRELESS COMMUNICATION SYSTEMS
663
models. Empirical models are usually a set of equations derived from extensive field measurements [6], [7]. Empirical models are simple and efficient to use. They are accurate for environments with the same characteristics as those where the measurements were made. The input parameters for the empirical models are usually qualitative and not very specific, e.g., a dense urban area, a rural area, and so on. One of the main drawbacks of empirical models is that they cannot be used for different environments without modification, and sometimes they are simply useless. For example, the empirical model for macrocells cannot be used for indoor picocells. The output parameters are basically range specific, not site specific. Site-specific models are based on numerical methods such as the ray-tracing method [8], [9] and the finite-difference time-domain (FDTD) method. The input parameters can be very detailed and accurate. The disadvantages of the site-specific methods are the large computational overhead that may be prohibitive for some complex environments. Theoretical models are derived physically assuming some ideal conditions. For example, the over-rooftop diffraction model is derived using physical optics assuming uniform heights and spacing of buildings. Theoretical models are more efficient than the site-specific models and more site-specific than the empirical models [10]. C. Statistical Models Small-scale propagation parameters are usually characterized by some statistics, such as rms delay spread, coherence bandwidth, Doppler spread, and coherence time. These parameters directly affect the design of Rxs and affect estimated values of the bit error rate. They also facilitate simulations of communication systems and provide performance measure of quality of service (QoS). Due to the length limitation of this paper, the review of this rich research area will be very brief. Readers are referred to the references for detailed information on specific areas of interest. D. Challenges to the Propagation Modeling Wireless communication channels are inherently frequency dispersive, time varying, and space selective, although only one or two of these dependencies will appear in some cases. The fast evolution of wireless communications has lead to the use of higher frequency bands, smaller cell sizes, and smart antenna systems, making the propagation prediction issues more challenging. In macrocells, since the transmitting antenna is usually located on a high tower, simple empirical and statistical models are widely used with satisfactory accuracy. As for the microcells and especially for picocells, the height of the transmitting antenna may be lower than the average height of the buildings in the regions involved. In this case, the geometry of the buildings and terrains will greatly affect the propagation of the radio waves, causing wide shadow regions. The outdoor radio wave propagates through reflections from vertical walls and ground, diffractions from vertical and horizontal edges of buildings, and scattering from nonsmooth surfaces, and all possible combinations. There is no general empirical and statistical model that
can be used for prediction of these complicated propagation environments. Smart antenna systems exploiting space diversity require information on the angle of arrival of the multipath in addition to the usual parameters such as path loss and delay spread. A MIMO system uses the multipath to provide higher capacity [11], [12], completely different from the classical systems where multipath is considered harmful. All these new systems involve space-time and space-frequency channel models. To deal with the new complex propagation environments, site-specific models have been developed based on ray-tracing techniques. In a basic ray-tracing algorithm, the main task is to determine the trajectory of a ray launched from a transmitting antenna. This procedure involves the calculation of the intersection of a ray with a surface (in three-dimensional (3-D) cases) or a ray with an edge segment (in two-dimensional (2-D) cases). The computation time might be huge or even beyond the capability of present computers if the propagation environment is large and/or complex. The computation efficiency is then the biggest obstacle against the application of ray-tracing methods. An efficient ray-tracing procedure is also important for improving the prediction accuracy since more types of rayssuch as reflected, transmitted, diffracted and scattered rays and their combinationscan be taken into account. The accuracy of propagation prediction involves many aspects. These include the accuracy of locations and sizes of buildings and accurate knowledge of the electric parameters of walls and other objects involved. Trees, large posts, traffic, and pedestrians in outdoor cases and furniture in indoor cases can also influence the results and make a difference. Recently, accurate characterization of complex wall structures including metal-framed windows is receiving attention due to the requirement of a more accurate prediction of the indoor/outdoor propagation mechanism. To meet these challenges, existing prediction methods should be modified and improved, and new procedures and techniques have to be developed. E. Objective This paper will first give a brief review of widely used empirical and simplified theoretical propagation models. These models are mainly used for macrocells and microcells. Then a detailed review will be given on the research and application of that ray-tracing method that is gaining importance for propagation simulation of microcells and picocells. Some full-wave prediction methods will also be briefly described. It should be pointed out that we will focus on the deterministic prediction models for path loss. Due to the length limitation of this paper, statistical models for multipath fading will only be very briefly included. The readers are encouraged to read some review papers [13][19] and books [20][27] to complement the brief review included in this paper. II. EMPIRICAL AND THEORETICAL MODELS A. Definition of Path Loss and Free-Space Propagation The path loss at a point is defined as the ratio of transmitted , over the received power at . For power at
664
free-space propagation, the path loss can be simply expressed as [26] dB (1)
and are the gains of the transmitting antenna (Tx) where and receiving antenna ( ), respectively, is the distance beand , and is the wavelength in free space. tween B. Okumura Model and Hata Model The Okumura model [6] is an empirical model based on extensive measurements made in Japan at several frequencies in the range from 1501920 MHz (it is also extrapolated up to 3000 MHz). Okumuras model is basically developed for macrocells with cell diameters from 1 to 100 km. The heights of the BS antenna are between 301000 m. The Okumura model takes into account some of the propagation parameters such as the type of environment and the terrain irregularity. The basic prediction formula is as follows: dB
Fig. 1. Multiple slope regression fit to the two-ray model [37].
where (dB) is the median value of the propagation path is the free-space path loss, and can be calculated using loss, is the median attenuation value relative to free space (1), and are the height gain facin an urban area, is the correction tors of BS and mobile antennas, and and are determined factor due the environment. and by looking up curves derived from measurements. are calculated using simple formulas. Terrain information can be qualitatively included in the Okumura model. For example, the propagation environments are categorized as open area, quasi-open area, and suburban area. Other information such as terrain modulation height and average slope of terrain can also be included. Illustrative examples using the Okumura model can be found in, e.g., [26], [27]. The Hata model [7] is a formula-based Okumura model (graphics-based) and can be used more effectively. The frequencies range from 150 to 1500 MHz. It has been extended to cover the frequency band from 1500 to 2000 MHz in the COST 231 project [19]. C. Over-Rooftop Models Over-rooftop models are typical theoretical models [10], [28] that are more precise than the Okumura model for the description of the urban environments. Based on the physical optics and some assumptions made for the geometry of the buildings and heights of BS antennas, formulas are derived that give the average received signal for mobiles at street level. Typical assumptions are that the heights of the buildings are equal and the spacing between the buildings is identical. In over-rooftop models, the path loss in decibels is the sum of ). The excess free-space loss and the so-called excess loss ( , i.e., the loss is further divided into two parts diffraction of the fields at the rooftop before the mobile down to the street level, and the reduction of the field at this rooftop as a result of propagation over the previous rows of buildings.
Saunder and Bonar [29], [30] also investigated the over-rooftop propagation for the case of arrays of buildings. In [29], more general situations are taken into account, e.g., the short-range case where the method in [10] gives incorrect results. In [30], efforts were made to extend the existing model to deal with more realistic situations, i.e., the building heights and spacing are irregular. Ikegami et al., using a single diffraction mechanism, also studied the over-rooftop propagation [31]. Vogler [32] proposed another over-rooftop method for buildings with variant heights and spacing. Bertonis method [10], [28] and Voglers formulation were combined in [33] so that an efficient and accurate prediction model for rooftop propagation is obtained. Other improvement of over-rooftop propagation models can be found in [34]. A comparison between several over-rooftop propagation models for two types of building profiles is presented in [35], i.e., buildings with equal heights and uniform spacing, and buildings with irregular heights and spacing. It is found that the two building profiles have their own merits and will give more accurate results if properly used for different environments (including building geometry, antenna heights relative to average building heights, etc.). D. Two-Slope Model for Microcellular Environments This model is measurement based and is used for line-of-sight (LOS) propagation in an urban area. The model is based on a two-ray propagation mechanism, i.e., the LOS ray and the reflection ray from the ground [36], [37]. This model is characterized by the fact that a break point exists that clearly separates the different properties of propagation in near and far regions relative to the BS, as shown in Fig. 1. Using regression analysis of the measured data in the San Francisco Bay area, it is shown that the slope before the break point is less than two, while the slope after the break point is greater than two [37]. The two-ray model for LOS propagation was extended in [38] to take into account the effects of traffic and high obstacles such as posts. It is shown that when the heights of traffic and some obstacles are included in the model, better accuracy can be obtained compared with the experimental results.
665
(a) Fig. 3. Fig. 2. Ray launching procedure. It is shown that some rays are launched from the Tx and reflected by the walls. It is also shown that at the end of the process, a fewer number of rays may be received by the Rx.
(b)
Ray cone and ray tube. (a) Ray cone. (b) Ray tube.
E. Other Models Rustako et al. [36] proposed a six-ray model for an urban LOS area that was shown to be accurate compared with measured results. These rays are the direct (LOS) ray, the ground-reflected rays, two rays with one-wall reflection, and two rays with double-wall reflections. Lee proposed an area-to-area model [25] for flat-terrain regions. A set of nominal conditions is assumed and, when the realistic situation is different from the assumptions, correction factors are calculated and included in the prediction formula. Other models can be found in [39][42] and, specifically, for indoor prediction models, the reader is referred to [17], [18], [43][56]. III. RAY-TRACING MODELS Ray theory emerged as a highly promising procedure for providing an accurate site-specific means to obtain useful simulation results [8], [9], [57], [58]. It should be noted that the ray-tracing method also serves as a starting point for statistical modeling [59][62]. According to the ray optics and the uniform theory of diffraction (UTD), propagation mechanisms may include direct (LOS), reflected, transmitted, diffracted, scattered, and some combined rays, which, in fact, complicates and, in many realistic propagation environments, slows down the calculation procedure. In this section, some of the more commonly used ray-tracing methods will be briefly described. A. Shooting-and-Bouncing Ray (SBR) Launching Algorithm The basic procedure of a ray-tracing method is the SBR algorithm [63]. First, a ray is launched from the transmitting antenna (Tx), then the ray is traced to see if it hits any object or is received by the receiving antenna. When an object is hit, reflection, transmission, diffraction, or scattering will occur, depending on the geometry and the electric properties of the object. When a ray is received by a receiving antenna, the electric field (power) associated with the ray is calculated. A schematic of the SBR method is shown in Fig. 2. This algorithm has some fundamental issues that need to be considered. The first is how to launch a ray. The second is how to determine if a ray hits an object. Third, if there are several possible objects that can be hit by the ray, how is it determined which one is really hit? The fourth is how to determine whether a ray is received. In the following, ray launching and reception criteria as well as ray intersection with an object will be reviewed. 1) Ray Launching Model and Reception Criteria: A ray is actually a ray tube and is usually a cone, as shown in Fig. 3 [8],
[64][66]. When ray cones are used to cover the spherical wavefront at the receiving location, these cones have to overlap [8]. When ray tubes [see Fig. 3(b)] are used, the spherical wavefront can be covered without the overlapping of ray tubes. To determine whether a ray is received or not by a receiving antenna, one has to check if the receiving point is inside the ray cone or tube. If yes, the ray will be received; otherwise, it will not. For the ray-cone scheme, the reception test can be easily carried out by using a reception sphere centered at the [8], where is the receiving point with radius equal to angle between two adjacent rays and is the total length of the ray. Since ray cones are overlapped, when a receiving point is located in the overlapping area between the ray cones, the Rx will then receive two rays and ray double counting occurs [67]. This gives errors, and some procedures are proposed to deal with this issue [67], [68]. 2) Intersection Test of a Ray With an Object: To determine if a ray hits an object, one has to test the intersection of a ray with the object. This is a classic problem in computational geometry and graphics [69]. A naive SBR method tests all the objects to determine whether a ray hits an object. When the number of objects is large, the testing can be very time consuming and inefficient. It is pointed out in [70] that intersection testing can consume more than 90% of CPU time for a naive SBR algorithm. B. Image Method The image method is a simple and accurate method for determining the ray trajectory between the transmitter (Tx) and Rx. Fig. 4 shows the basic idea of the image method. For this simple is first determined ( in case, the image of Tx due to due to is calculated ( ). Fig. 4). Then the image of , one can find a reflection point ( ) on Connecting Rx and . Another reflection point ( ) is the intersection point of with the line connecting and . The image method is accurate, but suffers from inefficiency when the number of walls involved is large and reflection times are high. For realistic applications, special techniques such as the hybrid and acceleration methods have to be used to reduce the computation time. C. Hybrid Method Tan et al. [71] proposed a hybrid method combining the image and SBR methods. The SBR method is used to quickly identify a possible ray trajectory from Tx to Rx. When the trajectory is found, a series of walls involved can be determined. The exact reflection positions can then be accurately found by the image method. This method has the advantages of the SBR (efficient) and image (accurate) methods.
666
Fig. 4. Illustration of the image method.
D. Acceleration of Ray-Tracing Algorithms The ray-tracing method is simple and is most widely used in the area of site-specific propagation prediction. However, the ray-tracing method can be very computationally inefficient. This is why there are many publications focusing on the acceleration of the ray-tracing algorithms. There are several ways to achieve the acceleration. The first is to reduce the number of objects on which actual ray-object intersection will be performed. The second is to accelerate the calculation of the intersection test. All acceleration methods concern the preprocessing of the propagation environments and/or the positions of Tx and/or Rx. In this section, we will provide a brief summary of these efforts. 1) Angular Z-Buffer (AZB) [70], [72]: This method is based on the light buffer technique used in computer graphics. The basic idea is to divide the space into angular regions according to a source point. The source point can be a Tx or an image of it related to a reflection plane. When a ray is launched from the source point, only those objects located in the angular region containing the ray need to be tested for ray intersection. This method can accelerate the ray-tracing algorithm, but, when multiple reflections are needed, the preprocessing is not easy. This is because there are many source points (including the Tx and a large number of its images) and an AZB should be established for each of them. 2) Ray-Path Search Algorithm: Based on the idea that raytracing routines should be applied only to those areas where rays are likely to exist, the ray-path search algorithm in [73] and [74] employs the visibility graph to limit the intersection test. The visibility graph contains several layers. The first layer includes all objects visible to the Tx (for LOS rays). The second layer contains objects visible to the first layer (for transmitted, reflected, and diffracted rays). Further layers are of similar recursive relationship. Since the determination of visibility between two objects is not easy, acceleration methods such as bounding boxes are employed for establishing the visibility graph. When a ray is launched from the Tx, only those objects in the first layer of the visibility graph need to be tested for the first intersection. To determine the th intersection of the ray, only objects in the th layer need to be tested, thus leading to saving of computation time.
Fig. 5. Image generation and the illumination zones of the images: the basic concept and path tracing using illumination zones [76].
This method has similar drawbacks to the AZB method, i.e., when interaction levels are high, the establishment of the visibility graph will be much more time consuming and complicated. A similar procedure for the image method is presented in [75][77]. This technique generates an image table for each BS location. These images take into consideration the various wall reflections, transmissions, and edge diffractions. To further accelerate the ray-tracing procedure, accurate illumination zone concepts are used to restrict each image to an illuminated area. Thus, only objects inside the illumination zone need intersection testing. Fig. 5 shows the illumination zone concept, where is the image of the BS with respect to Wall 1, and (second-order image) is the image of with respect to W2. . The mobile station (MS) is in the illumination zone of In [78], an efficient method for visibility list construction is developed. This method is especially designed for a large number of receiving points. Some unnecessary repetition calculations are avoided by using a so-called point-to-area algorithm. A dynamic calculation of the visibility list is used to accelerate the 3-D ray-tracing procedure. 3) Dimension Reduction Method: To achieve efficient raytracing procedures and retain acceptable accuracy, ray-tracing algorithms may be carried out in nonfull 3-D geometries. Examples of this approach may include the 2-D/two-and-one-half dimensional (2.5-D) method, the vertical plane launch (VPL) method, and so on. The following is a brief summary of some of these methods. a) 2-D/2.5-D Method: When the heights of buildings in a region are much larger than the height of the Tx, the main propagation is a lateral one. In this case, the complex 3-D environment can be approximated by much simpler 2-D structures and a significant saving in computation time can be achieved. Rizk et al. [79] presented a 2-D ray-tracing modeling method for microcellular environments. Based on the image method, the obtained prediction results compared well with measurement data. b) VPL Method: The VPL technique is proposed in [9]. The usual 2-D ray tracing is used in the horizontal plane.
667
Fig. 6. Schematic illustration of the ray-launching procedure in the VPL method [9].
Each ray in the 2-D case represents a vertical propagation plane. When a ray hits a vertical wall, specular reflection from the vertical wall and diffraction from the rooftop horizontal edge can occur. When the ray hits a vertical edge, diffraction also occurs. The over-rooftop diffraction creates two vertical propagation planes, one in the same direction as the incident ray and the other in the direction of reflection. Diffraction from the vertical edge creates a new source and many new rays in 2-D planes should be launched. These rays are further traced in a similar manner until some criteria are reached. Fig. 6 is a schematic illustration of the VPL method [9]. Rizk et al. [80] compared the results using lateral, full-3-D, and VPL methods. It is found that when the average building heights are around the Tx height, VPL can give very good predictions. 4) Space-Division Method: The space-division method is widely used in computer graphics. The basic idea is to first create a grid (usually rectangular) in the propagation environment, and then establish a lookup table registering objects residing in each grid cell. When a ray is launched, it is traced in the grid. For each grid the ray is traversing, the lookup table is checked to see if any objects reside in the grid. If yes, the ray is tested for intersection with these objects. If any object is hit, a reflected (or diffracted) and/or a transmitted ray will be created and the new rays will be further traced. The space-division method can give fast ray traversing and efficient ray tracing. This is due to the fact that the algorithm for traversingthegridcanbefastandtheintersectiontestisperformed only on a small number of objects. Two types of space-division methods that have been applied to propagation in urban environments will be summarized in the following subsections. a) Rectangular Division: The rectangular division method has the advantage that the ray traversing in the grid is
very fast [81]. Yun et al. [82] developed an efficient ray-tracing method employing the fast ray-traversing algorithm for rectangular grid [81]. Different from the usual space-division method, the new method requires that the wall should be exactly located on the grid lines to acquire best efficiency. The new method labels each grid cell according to the room (or building) identification number, i.e., cells in the same room have the same identifications. When a ray is traversing from one grid cell to another, and the label of the new cell is different from the previous cell, the wall between the two cells will be hit. In this case, the intersection test that is used in usual space-division methods is completely avoided. As a result, a significant reduction in CPU time is expected. Specifically, the CPU time for the new method was shown to be around 15% of that of the visibility method for a realistic indoor environment [82]. b) Triangular Division: In [83] and [84], triangular division methods were proposed. It is shown that a great improvement in CPU time was also achieved when using this method. In [84], the triangulation strategies are described in detail, and a ray-traversing algorithm is developed. It should be pointed out that the most important feature of the triangular grid method developed in [84] is that, when finding the segment hit by the ray, only two cross products of two vectors need to be calculated. Fig. 7 shows the basic ray-traversal procedure. When a ray, is leaving the first intersection edge i.e., , from the source in the triangle , we need to determine which edge, i.e., or , will be hit next. We can build a unit vector from and pointing to , the opposite the intersection point on . The sign of the cross-product between and vertex to edge then determines the edge to be hit. If the sign is positive, the ray hits ; otherwise, the edge will be hit.
668
Fig. 7.
Determination of the edge to be hit by a ray.
It is shown in [85] that the CPU time for the triangular grid method is about 30% of that of the visibility method for an indoor environment. The triangular grid method was also used to calculate the path loss of Munich City, and the results were compared with the measured results provided by the COST 231 project. It is shown that the average error standard deviation is 7.2 and is very good compared with other results in the COST 231 project [86]. Estimates of the saving in CPU time were very significant and more than an order of magnitude [87]. E. Improvement of Accuracy of Ray-Tracing Algorithms The ray-tracing method can provide site-specific predictions. Due to the fact that the environmental database may not be accurate and the materials of the objects in the region of interest may not be known, the ray-tracing method can only provide approximate results for realistic propagation environments. Another factor affecting the accuracy of the ray-tracing procedure is the incomplete account for all kinds of rays. This is because the more rays taken into account, the more computation time will be needed, leading to unacceptable efficiency. Examples of techniques used to improve the accuracy of ray-tracing algorithms are described in the following sections. 1) Additional Ray MechanismsEffect of Diffractions: Diffractions from vertical and horizontal edges of buildings are important contributions to the received power. The over-rooftop propagation is mainly due to diffractions from the horizontal edges. Methods for calculation of diffraction coefficients for metal or materials with finite conductivity were developed [88][91]. A comparison among the perfectly absorbing wedge (PAW) method, UTD, and UTD heuristic methods can be found in [79]. It is found that errors given by these three methods are comparable. Rizk et al. [92] proposed a method to include the slope diffraction from wedges to improve the accuracy of calculation of the diffracted field in transition regions using classical UTD. Several decibels (approximately 6 dB) of improvement can be achieved. The diffraction from building corners (wedges) is taken into account in [93]. New diffraction coefficients for objects with finite conductivity are developed [94]. The artificial dip in the usual diffraction calculation is removed. Comparison with FDTD shows that the results of the new method are of good accuracy. 2) Ray-Launching Models: In the SBR method, the detection of reception of rays depends on how these rays are
launched. When ray cones are used, it is usually assumed that the angles between adjacent rays are equal. This is true if the spherical wavefront is approximated by a regular icosahedron and rays are launched from each of the vertices of the icosahedron. Since the number of the vertices of the icosahedron is 12, the number of rays launched is usually too few to get satisfactory accuracy. To launch more rays, the icosahedron is tessellated, but the angles between adjacent rays will no longer be equal. Fortunately, the difference between these angles is not significant [8] and, in practice, they are assumed to be equal. It is pointed out earlier in this paper that the ray-cone model has the double-counting problem [67]. Several methods have been proposed to reduce or avoid its effect. In [67], a method of distributed wavefronts was developed to remedy this problem. Instead of counting hit-or-miss rays, the method in [67] takes the contribution of all nearby rays into account. The total field received by an antenna is the weighted sum of several wavefronts. Rays near the Rx contribute more power than those far away from the Rx. The distributed wavefronts method improves the accuracy of the calculated fields, but is relatively complex to realize (counts and keeps record of a large number of rays) and is also inherently inaccurate. Yun et al. [68] proposed a simple method to avoid the ray double-counting problem. It is based on the fact that a ray from Tx to Rx is uniquely defined by a sequence of objects involved. Thus, when a ray is received, this sequence is stored. When another ray is received, the corresponding sequence will be checked with the existing sequences. If there is an identical sequence, the new received ray is a double-counted ray and should not be taken into account. This method adds little extra effort for the existing SBR code, but avoids ray double counting and, hence, improves the accuracy. For the ray-tube launching model, the power is assigned to each tube according to its cross-sectional area. There is no need to make uniform ray distribution in theory in this case. It does not have the problem of double counting. Usually, the information of the cross section needs to be kept on track, and the reception scheme is different from the reception sphere model. 3) Other Issues: Rizk et al. [95] investigated the influence of database accuracy for ray-tracing techniques based on 2-D environments. The building layouts built with different maps, the materials assumed for the walls, the dimensions and locations of buildings, and the positions of Txs are examined. In [96], the relationship between the received power and different ray combinations and interactions, wall material characteristics, antenna position offsets, and database inaccuracies were investigated. In [97] and [98], comparisons between experiment and raytracing results were performed and it was found that the raytracing method was suitable for predicting signal and interference levels. Rizk et al. [79] studied the repeatability of propagation measurements. It is pointed out that the divergence between the repeated measurements will be as large as 5 dB over a distance less than 50 m. The standard deviation between repeated measurements is about 3 dB. It is thus concluded that the comparison between measurements and predictions should emphasize the global tendency, instead of details.
669
Rizk et al. [99] examined the effects of lamppost and panel scattering in urban propagation simulation. The metal lamppost is approximated by a cylinder and the panel is represented by a finite plate. It is found that the scattering by metallic cylinders is as important as the reflections and diffractions. The panel can have a nonnegligible effect on the simulation results. Tree effects of tree and vegetation on the propagation characteristics are investigated in [100][102]. The tree canopy is simulated as an elliptical cylinder horizontally placed to nearby buildings [101]. It is found that ten rows of wide trees may cause 45-dB extra path loss. IV. STATISTICAL MODELS Small-scale channel modeling is concerned with multipath fading and is usually investigated using statistical methods. An impulse response model is often employed and, for time-invariant cases, its transfer function has the form
where and are the amplitude, angle of arrival, and time delay of the th multipath, respectively [26]. Usually, parameters such as the time delay spread, the coherence bandwidth, Doppler spread, and coherence time are of interest. Turin et al. [103] studied the statistical model for urban multipath propagation based on experiments. It was found that the excess delay forms a Poisson sequence. The multipath spread was found to be almost totally dependent on the local environment of the Rx, independent of the distance between the Tx and Rx. Suzuki [104] further investigated Turins experimental data and proposed a modified Poisson process to model the path arrival time. Turins mathematical model was employed by Hashemi for the development of a simulation program for urban multipath propagation [105]. Saleh et al. proposed a statistical model for indoor multipath propagation [106]. The model is based on experimental results taken for a medium-sized two-story office building. The new model introduced the cluster-ray concept, i.e., the rays arrive in clusters. The cluster arrival times are modeled by a Poisson process. Within each cluster, rays are also arriving according to a Poisson process. Rappaport et al. [44], [47] developed a measurement-based statistical model for factory buildings. The effect of TxRx separation distance, small-scale Rx movement, and correlation of multipath amplitudes on local areas were examined. V. NEW TRENDS IN PROPAGATION PREDICTIONS As the fast development of wireless communications continues to progress, new techniques are employed to increase the capacity and the QoS of the deployed systems. The application of smart antennas and MIMO systems requires a spatiotemporal characterization of wireless channel. In addition to the path loss and time delay spread, angle-of-arrival and joint spatio-temporal models become necessary for the development of modern wireless communication systems [107], [108].
The currently developed spatio-temporal models usually assume certain distribution of scatterers around mobile stations and/or BSs and then derive the joint time-of-arrival and angle-of-arrival density functions. The scatterer distribution can lie in a circular area [22], an elliptical area [108], [109], or in clusters [110]. It is also found that, based on indoor measurements, rays can also arrive in clusters and a joint spatio-temporal model is developed for indoor environments [111]. As the more accurate modeling of outdoor-to-indoor propagation is needed, characterization of wall structures is receiving more attention. Walls are usually represented by half-space materials [8] or slab and layered models [76]. Measurement results are widely used for calculation of reflection and transmission through these walls [112][116]. Analytical and numerical characterization of walls is also available. Honcharenko and Bertoni [117] investigated the reflection and transmission properties of concrete walls with periodic structures. Chu et al. [118], [119] studied some composite wall structures using periodic surface-integral formulation. Both TE and TM incidences were treated. Chiu et al. [120], using the filament-current and thin-current assumptions, analyzed structures with laminated ( -ply) structures. The finiteelement method is used in [121] for the analysis of propagation into reinforced concrete walls. Dalke et al. [122], using the FDTD method, analyzed the propagation properties of reinforced concrete structures. It is found that the transmitted field has resonance and nulls. These resonance and nulls are dependent on the structure and the thickness of the wall. These resonance-type effects are very important and may have critical impact on wide-band and ultra-wide-band communication systems. Holloway et al. [123] proposed a different method to characterize complex walls with periodic structures. Holloway used the homogenized method to replace the complex wall with three layers of materials. The electrical parameters of the center layer are dependent on the angle of incidence. The resonant property is predicted and accurately characterized using this method. Fig. 8 shows a comparison between reflection/transmission coefficients calculated using Holloways method and the FDTD method [124]. The resonant effects can be clearly seen in this figure. The possible use of FDTD in these calculations now opens the door for calculating a wide variety of these walls and the implementation of the obtained results in urban propagation modeling codes. Another structure of interest is the metal-framed glass (windows) with periodic geometry. Since the size of the periodic cell is usually larger than the wavelength, caution has to be taken for its characterization. An equivalent-ray method was proposed by the authors to accurately characterize transmission through windows [125]. The proposed method represents each periodic cell (window) with an equivalent ray representation with amplitudes precalculated using UTD or a numerical method such as FDTD. When a ray hits the structure, it is replaced by precalculated rays entering the building (90 rays for 180 diffraction pattern). This is certainly different from accounting for transmission through windows by using a single complex transmission coefficient parameter. Preliminary results show that the method gives very
670
Fig. 8. Reflection and transmission coefficients for a composite wall: FDTD versus Holloways method. D is the period and is the wavelength.
good accuracy compared with a full FDTD simulation of the entire window system [125]. This new equivalent ray-tracing representation of windows and metal-framed structures, therefore, provides a significant step toward integration of indoor and outdoor propagation prediction models. VI. CONCLUSIONS The tremendous development in wireless communications leads to the emergence of new ideas and techniques to increase capacity and improve the QoS. Smaller cell sizes, higher frequencies, and more complex environments need to be more accurately modeled and site-specific propagation prediction models need to be developed to achieve optimum design of next-generation communication systems. New techniques such as smart antennas and multiinput and multioutput systems need new propagation prediction models to characterize the joint spatio-temporal channel. This paper presented a review of the state-of-the-art propagation prediction models that range from early simple empirical formulas to modern site-specific ray-tracing-based models. It is shown that the ray-tracing method can provide path loss, time of arrival, angle of arrival, and even some statistic parameters for propagations in complex environments. New challenges were briefly discussed and new methods to meet these challenges were described. Specifically, new efforts to characterize walls of complex structures and develop equivalent ray-tracing models for windows and metal-framed structures were highlighted. These new developments, together with computationally efficient ray-tracing methods, are expected to lead the way toward the development of an integrated indoor/outdoor urban propagation model that takes into account the complex indoor/outdoor interface issues. REFERENCES
[1] A. J. Schwab and P. Fischer, Maxwell, Hertz, and German radio-wave history, Proc. IEEE, vol. 86, pp. 13121318, July 1998. [2] G. C. Corazza, Marconis history, Proc. IEEE, vol. 86, pp. 13071311, July 1998.
[3] Q. Bi, G. I. Zysman, and H. Menkes, Wireless mobile communications at the start of the 21st century, IEEE Commun. Mag., vol. 39, pp. 110116, Jan. 2001. [4] M. Dinis and J. Fernandes, Provision of sufficient transmission capacity for broadband mobile multimedia: A step toward 4G, IEEE Commun. Mag., vol. 39, pp. 4654, Aug. 2001. [5] B. G. Evans and K. Baughan, Visions of 4G, Electron. Commun. Eng. J., vol. 12, no. 6, pp. 293330, Dec. 2000. [6] Y. Okumura, E. Ohmori, T. Kawano, and K. Fukuda, Field strength variability in VHF and UHF land mobile service, Rev. Elect. Comm. Lab., vol. 16, no. 910, pp. 825873, Sept.Oct. 1968. [7] M. Hata, Empirical formula for propagation loss in land mobile radio services, IEEE Trans. Veh. Technol., vol. VT-29, pp. 317325, Aug. 1980. [8] S. Y. Seidel and T. S. Rappaport, Site-specific propagation prediction for wireless in-building personal communication system design, IEEE Trans. Veh. Technol., vol. 43, pp. 879891, Nov. 1994. [9] G. Liang and H. L. Bertoni, A new approach to 3-D ray tracing for propagation prediction in cities, IEEE Trans. Antennas Propagat., vol. 46, pp. 853863, June 1998. [10] J. Walfisch and H. L. Bertoni, A theoretical model of UHF propagation in urban environments, IEEE Trans. Antennas Propagat., vol. 36, pp. 17881796, Dec. 1988. [11] G. J. Foschini, Layered spacetime architecture for wireless communication in a fading environment when using multi-element antennas, Bell Syst. Tech. J., vol. 1, no. 2, pp. 4159, Autumn 1996. [12] V. Tarokh, N. Seshadri, and A. R. Calderbank, Spacetime codes for high data rate wireless communication: Performance criterion and code construction, IEEE Trans. Inform. Theory, vol. 44, pp. 744765, Mar. 1998. [13] T. Kurner, D. J. Cichon, and W. Wiesbeck, Concepts and results for 3D digital terrain-based wave propagation models: An overview, IEEE J. Select. Areas Commun., vol. 11, pp. 10021012, Sept. 1993. [14] H. L. Bertoni, W. Honcharenko, L. R. Maciel, and H. H. Xia, UFF propagation prediction for wireless personal communications, Proc. IEEE, vol. 82, pp. 13331359, Sept. 1994. [15] G. Liang and H. L. Bertoni, Review of ray modeling techniques for site specific propagation prediction, in Wireless Communications: TDMA Versus CDMA, S. G. Glisic and P. A. Leppanen, Eds. Norwell, MA: Kluwer, 1997. [16] S. Stein, Fading channel issues in system engineering, IEEE J. Select. Areas Commun., vol. SAC-5, pp. 6889, Feb. 1987. [17] D. Molkar, Review on radio propagation into and within buildings, Proc. Inst. Elect. Eng., pt. H, vol. 138, pp. 6173, Feb. 1991. [18] H. Hashemi, The indoor radio propagation channel, Proc. IEEE, vol. 81, pp. 943968, July 1993. [19] D. J. Cichon and T. Kurner. Propagation prediction models. COST 231 Final Rep. [Online]. Available: http://www.lx.it.pt/cost231/ [20] W. C. Jakes, Ed., Microwave Mobile Communications. New York: Wiley, 1974. [21] G. L. Stuber, Principles of Mobile Communications, 2nd ed. Norwell, MA: Kluwer, 2001. [22] J. C. Liberti and T. S. Rappaport, Smart Antennas for Wireless Communications: IS-95 and Third Generation CDMA Applications. Englewood Cliffs, NJ: Prentice-Hall, 1999. [23] H. L. Bertoni, Radio Propagation for Modern Wireless Systems. Englewood Cliffs, NJ: Prentice-Hall, 2000. [24] N. Blaunstein, Radio Propagation in Cellular Networks. Norwood, MA: Artech House, 2000. [25] W. Y. C. Lee, Mobile Cellular Telecommunications Systems. New York: McGraw-Hill, 1989. [26] T. S. Rappaport, Wireless Communications: Principle and Practice. Englewood Cliffs, NJ: Prentice-Hall, 1996. [27] J. M. Hernando and F. Perez-Fontan, Introduction to Mobile Communications Engineering. Norwood, MA: Artech House, 1999. [28] L. R. Maciel, H. L. Bertoni, and H. H. Xia, Unified approach to prediction of propagation over buildings for all ranges of base station antenna height, IEEE Trans. Veh. Technol., vol. 42, pp. 4145, Feb. 1993. [29] S. R. Saunder and F. R. Bonar, Explicit multiple building diffraction attenuation function for mobile radio wave propagation, Electron. Letts., vol. 27, no. 14, pp. 12761277, July 1994. , Prediction of mobile radio wave propagation over buildings of [30] irregular heights and spacings, IEEE Trans. Antennas Propagat., vol. 42, pp. 137144, Feb. 1994. [31] F. Ikegami, T. Takeuchi, and S. Yoshida, Theoretical prediction of mean field strength of urban mobile radio, IEEE Trans. Antennas Propagat., vol. 39, pp. 299302, Mar. 1991.
671
[32] L. E. Vogler, An attenuation function for multiple knife-edge diffraction, Radio Sci., vol. 19, no. 6, pp. 15411546, Nov./Dec. 1982. [33] N. C. Goncalves and L. M. Correia, A propagation model for urban microcellular systems at the UHF band, IEEE Trans. Veh. Technol., vol. 49, pp. 12941302, July 2000. [34] L. Juan-Llacer and N. Cardona, UTD solution for the multiple building diffraction attenuation function for mobile radiowave propagation, Electron. Lett., vol. 33, no. 1, pp. 9293, Jan. 1997. [35] L. Juan-Llacer, L. Ramos, and N. Cardona, Application of some theoretical models for coverage prediction in macrocell urban environments, IEEE rans. Veh. Technol., vol. 48, pp. 14631468, Sept. 1999. [36] A. J. Rustako, N. Amitay, G. J. Owens, and R. R. Roman, Radio propagation at microwave frequencies for line-of-sight microcellular mobile and personal communications, IEEE Trans. Veh. Technol., vol. 40, pp. 203210, Feb. 1991. [37] H. H. Xia, H. L. Bertoni, L. R. Maciel, A. Lindsay-Stewart, and R. Rowe, Radio propagation characteristics for line-of-sight microcellular and personal communications, IEEE Trans. Antennas Propagat., vol. 41, pp. 14391447, Oct. 1993. [38] Y. Oda, K. Tsunekawa, and M. Hata, Advanced LOS path-loss model in microcellular mobile communications, IEEE Trans. Veh. Technol., vol. 49, pp. 21212125, Nov. 2000. [39] R. Edwards and J. Durkin, Computer prediction of service areas for VHF mobile radio network, Proc. Inst. Elect. Eng., vol. 116, pp. 14831500, Sept. 1969. [40] A. J. Goldsmith and L. J. Greenstein, A measurement-based model for predicting coverage areas of urban microcells, IEEE J. Select. Areas Commun., vol. 11, pp. 10131023, Sept. 1993. [41] J. Berg, A recursive method for street microcell path loss calculations, in Proc. PIMRC95, pp. 140143. [42] W. C. Y. Lee and D. J. Y. Lee, Microcell prediction in dense urban area, IEEE Trans. Veh. Technol., vol. 47, pp. 246253, Feb. 1998. [43] S. E. Alexander, Radio propagation within buildings, Electron. Lett., vol. 18, no. 21, pp. 913914, 1982. [44] T. S. Rappaport, Characterization of UHF multipath radio channels in factory buildings, IEEE Trans. Antennas Propagat., vol. 37, pp. 10581069, Aug. 1989. [45] D. C. Cox, R. R. Murray, and A. W. Norris, Measurements of 800 MHz radio transmission into buildings with metallic walls, Bell Syst. Tech.l J., vol. 62, no. 9, pp. 26952717, Nov. 1983. [46] D. J. Devasirvatham, M. J. Krain, and D. A. Rappaport, Radio propagation measurements at 850 MHz, 1.7 GHz, and 4.0 GHz inside two dissimilar office buildings, Electron. Lett., vol. 26, no. 7, pp. 445447, 1990. [47] T. S. Rappaport, S. Y. Seidel, and K. Takamizawa, Statistical channel impulse response models for factory and open plan building radio communicate system design, IEEE Trans. Commun., vol. 39, pp. 794807, May 1991. [48] S. Y. Seidel, T. S. Rappaport, M. J. Feuerstein, K. L. Blackard, and L. Grindstaff, The impact of surrounding buildings on propagation for wireless in-building personal communications system design, in IEEE 42nd Veh. Technol. Conf., vol. 2, 1992, pp. 814818. [49] S. Y. Seidel and T. S. Rappaport, 914 MHz path loss prediction models for indoor wireless communications in multifloored buildings, IEEE Trans. Antennas Propagat., vol. 40, pp. 207217, Feb. 1992. [50] R. Hoppe, G. Wolffe, and F. M. Landstorfer, Measurement of building penetration loss and propagation models for radio transmission into buildings, in IEEE Veh. Technol. Conf., Sept. 1999, pp. 22982302. [51] A. M. D. Turkmani and A. F. Toledo, Propagation into and within buildings at 900, 1800, and 2300 MHz, in IEEE Veh. Technol. Conf., Dec. 1992, pp. 633636. [52] S. Kim, H. L. Bertoni, and M. Stern, Pulse propagation characteristics at 2.4 GHz inside buildings, IEEE Trans. Veh. Technol., vol. 45, pp. 579592, Aug. 1996. [53] D. Akerbert, Properties of a TDMA picocellular office communications systems, in IEEE GLOBECOM, Dec. 1988, pp. 13431349. [54] G. D. Durgin, T. S. Rappaport, and H. Xu, 5.85-GHz radio path loss and penetration loss measurements in and around homes and trees, IEEE Commun. Lett., vol. 2, pp. 7072, Mar. 1998. , Measurements and models for radio path loss and penetration [55] loss in and around homes and trees at 5.85 GHz, IEEE Trans. Commun., vol. 46, pp. 14841496, Nov. 1998. [56] D. J. Y. Lee and W. C. Y. Lee, Propagation prediction in and through buildings, IEEE Trans. Veh. Technol., vol. 49, pp. 15291533, Sept. 2000.
[57] L. Piazzi and H. L. Bertoni, Achievable accuracy of site-specific path-loss predictions in residential environments, IEEE Trans. Antennas Propagat., vol. 48, pp. 922930, May 1999. [58] S. Kim, B. J. Guarino, T. M. Willis, III, V. Erceg, S. J. Fortune, R. A. Valenzuela, L. W. Thomas, J. Ling, and J. D. Moore, Radio propagation measurements and prediction using three-dimensional ray tracing in urban environments at 908 MHz and 1.9 GHz, IEEE Trans. Veh. Technol., vol. 48, pp. 931946, May 1999. [59] J. E. Dietert, S. Karger, and B. Rembold. (2000) Statistical channel modeling based on ray-tracing simulations. COST 259, TD(00)004. [Online]. Available: http://www.lx.it.pt/cost259/. [60] J. E. Dietert and B. Rembold. (2000) Stochastic channel model for outdoor applications based on ray-tracing simulations. COST 259, TD(00)005. [Online]. Available: http://www.lx.it.pt/cost259/. [61] Y. L. C. de Jong and M. H. A. J. Herben, Prediction of local mean power using 2-D ray-tracing-based propagation models, IEEE Trans. Veh. Technol., vol. 50, pp. 325331, Jan. 2001. [62] R. P. Torres, S. Loredo, L. Valle, and M. Domingo, An accurate and efficient method based on ray-tracing for the prediction of local flat-fading statistics in picocell radio channels, IEEE J. Select. Areas Commun., vol. 19, pp. 170178, Feb. 2001. [63] H. Ling, R. Chou, and S. Lee, Shooting and bouncing rays: Calculating the RCS of an arbitrarily shaped cavity, IEEE Trans. Antennas Propagat., vol. 37, pp. 194205, Feb. 1989. [64] S. Chen and S. Jeng, An SBR/image approach for radio wave propagation in indoor environments with metallic furniture, IEEE Trans. Antennas Propagat., vol. 45, pp. 98106, Jan. 1997. [65] C. Yang, B. Wu, and C. Ko, A ray-tracing method for modeling indoor wave propagation and penetration, IEEE Trans. Antennas Propagat., vol. 46, pp. 907919, June 1998. [66] H. Suzuki and A. S. Mohan, Ray tube tracing method for predicting indoor channel characteristic map, Electron. Lett., vol. 33, no. 17, pp. 14951496, 1997. [67] G. Durgin, N. Patwari, and T. S. Rappaport, Improved 3D ray launching method for wireless propagation prediction, Electron. Lett., vol. 33, no. 16, pp. 14121413, 1997. [68] Z. Yun, M. F. Iskander, and Z. Zhang, Development of a new shooting-and-bouncing ray (SBR) tracing method that avoids ray double counting, in IEEE AP-S Int. Symp. Dig., vol. 1, July 2001, pp. 464467. [69] J. ORourke, Computational Geometry in C. Cambridge, U.K.: Cambridge Univ. Press, 1993. [70] M. F. Catedra, J. Perez, F. S. de Anana, and O. Gutierrez, Efficient ray-tracing techniques for three-dimensional analyses of propagation in mobile communications: Application to picocell and microcell scenarios, IEEE Anntenas Propagat. Mag., vol. 40, pp. 1528, Apr. 1998. [71] S. Y. Tan and H. S. Tan, A microcellular communications propagation model based on the uniform theory of diffraction and multiple image theory, IEEE Trans. Antennas Propagat., vol. 44, pp. 13171326, Oct. 1996. [72] F. S. de Adana, O. G. Blonco, I. G. Diego, J. P. Arriaga, and M. F. Catedra, Propagation model based on ray tracing for the design of personal communication systems in indoor environments, IEEE Trans. Veh. Technol., vol. 49, pp. 21052112, Nov. 2000. [73] F. A. Agelet, F. P. Fontan, and A. Formella, Fast ray-tracing for microcellular and indoor environments, IEEE Trans. Magn., vol. 33, pp. 14841487, Mar. 1997. [74] F. A. Agelet, A. Formella, J. M. H. Rabanos, F. I. de Vicente, and F. P. Fontan, Efficient ray-tracing acceleration techniques for radio propagation modeling, IEEE Trans. Veh. Technol., vol. 49, pp. 20892104, Nov. 2000. [75] M. C. Lawton and J. P. McGeehan, The application of a deterministic ray launching algorithm for the prediction of radio channel characteristics in small-cell environments, IEEE Trans. Veh. Technol., vol. 43, pp. 955969, Nov. 1994. [76] G. E. Athanasiadou, A. R. Nix, and J. P. McGeehan, A microcellular ray-tracing propagation model and evaluation of its narrow-band and wide-band predictions, IEEE J. Select. Areas Commun., vol. 18, pp. 322335, Mar. 2000. [77] G. E. Athanasiadou and A. R. Nix, A novel 3-D indoor ray-tracing propagation model: The path generator and evaluation of narrow-band and wide-band predictions, IEEE Trans. Veh. Technol., vol. 49, pp. 11521168, July 2000. [78] W. M. OBrien, E. M. Kenny, and P. J. Cullen, An efficient implementation of a three-dimensional microcell propagation tool for indoor and outdoor urban environments, IEEE Trans. Veh. Technol., vol. 49, pp. 622630, Mar. 2000.
672
[79] K. Rizk, J. Wagen, and F. Gardiol, Two-dimensional ray-tracing modeling for propagation in microcellular environments, IEEE Trans. Veh. Technol., vol. 46, pp. 508517, May 1997. [80] K. Rizk, R. Valenzuela, S. Fortune, D. Chizhik, and F. Gardiol, Lateral, full-3D and vertical plane propagation in microcells and small cells, in 48th IEEE Veh. Technol. Conf., vol. 2, 1998, pp. 9981002. [81] J. G. Cleary and G. Wyvill, Analysis of an algorithm for fast ray tracing using uniform space subdivision, Vis. Comput., vol. 4, pp. 6583, 1998. [82] Z. Yun, M. F. Iskander, and Z. Zhang, Fast ray tracing procedure using space division with uniform rectangular grid, Electron. Lett., vol. 36, no. 10, pp. 895897, May 2000. [83] S. F. Fortune, D. M. Gay, B. W. Kernighan, O. Landron, R. A. Valenzuela, and M. H. Wright, WISE design of indoor wireless systems: Practical computation and optimization, IEEE Comput. Sci. Eng. Mag., pp. 5868, Spring 1995. [84] Z. Zhang, Z. Yun, and M. F. Iskander, Ray tracing method for propagation models in wireless communication systems, Electron. Lett., vol. 36, no. 5, pp. 464465, Mar. 2000. [85] Z. Yun, M. F. Iskander, and Z. Zhang, A fast indoor/outdoor ray tracing procedure using combined uniform rectangular and unstructured triangular grids, in Proc. IEEE AP-S Inte. Symp. Dig., July 2000, pp. 11341137. [86] Z. Yun, Z. Zhang, and M. F. Iskander, A ray-tracing method based on the triangular grid approach and application to propagation prediction in urban environments, IEEE Trans. Antennas Propagat., to be published. , Advanced and computationally efficient ray-tracing method for [87] propagation prediction in urban environment, in Int. Electromag. Advanced Applicat. Conf., Turin, Italy, Sept. 1014, 2001. [88] R. G. Kouyoumjian and P. H. Pathak, A uniform geometrical theory of diffraction for an edge in a perfectly conducting surface, Proc. IEEE, vol. 62, pp. 14481461, Nov. 1974. [89] R. J. Luebbers, Finite conductivity uniform UTD versus knife diffraction prediction of propagation path loss, IEEE Trans. Antennas Propagat., vol. AP-32, pp. 7076, Jan. 1984. [90] , Comparison of lossy wedge diffraction coefficients with application to mixed path propagation loss prediction, IEEE Trans. Antennas Propagat., vol. 36, pp. 10311034, July 1988. [91] , A heuristic UTD slope diffraction coefficient for rough lossy edges, IEEE Trans. Antennas Propagat., vol. 37, pp. 206211, Feb. 1989. [92] K. Rizk, R. Valenzuela, D. Chizhik, and F. Gardiol, Application of the slope diffraction method for urban microwave propagation prediction, in 48th IEEE Veh. Technol. Conf., vol. 2, 1998, pp. 11501154. [93] V. Erceg, A. J. Rustako, and R. S. Roman, Diffraction around corners and its effects on the microcell coverage area in urban and suburban environments at 900 MHz, 2 GHz, and 6 GHz, IEEE Trans. Veh. Technol., vol. 43, pp. 762766, Aug. 1994. [94] K. A. Remley, H. R. Anderson, and A. Weisshar, Improving the accuracy of ray-tracing techniques for indoor propagation modeling, IEEE Trans. Veh. Technol., vol. 49, pp. 23502357, Nov. 2000. [95] K. Rizk, J. F. Wagen, and F. Gardiol, Influence of database accuracy on two-dimensional ray-tracing-based predictions in urban microcells, IEEE Trans. Veh. Technol., vol. 49, pp. 631642, Mar. 2000. [96] G. E. Athanasiadou and A. R. Nix, Investigation into the sensitivity of the power predictions of a microcellular ray tracing propagation model, IEEE Trans. Veh. Technol., vol. 49, pp. 11401151, July 2000. [97] V. Degli-Esposti, G. Lombardi, C. Passerini, and G. Riva, Wide-band measurement and ray-tracing simulation of the 1900-MHz indoor propagation channel: Comparison criteria and results, IEEE Trans. Antennas Propagat., vol. 49, pp. 11011109, July 2001. [98] V. Erceg, S. J. Fortune, J. Ling, A. J. Rustako, Jr., and R. A. Valenzuela, Comparisons of a computer-based propagation prediction tool with experimental data collected in urban microcellular environments, IEEE J. Select. Areas Commun., vol. 15, pp. 677684, May 1997. [99] K. Rizk, J. Wagen, J. Li, and F. Gardiol, Lamppost and panel scattering compared to building reflection and diffraction, in COST 259 TD, Turin, Italy, May 1996, pp. 158167. [100] M. O. Al-Nuaimi and R. B. L. Stephens, Estimation of the effects of hilltop, singly distributed, trees on the path loss of microwave signals, Electron. Lett., vol. 33, no. 10, pp. 873874, May 1997. [101] S. A. Torrico, H. L. Bertoni, and R. H. Lang, Modeling tree effects on path loss in a residential environment, IEEE Trans. Antennas Propagat., vol. 46, pp. 872880, June 1998.
[102] J. C. R. Dal Bello, G. L. Siqueira, and H. L. Bertoni, Theoretical analysis and measurement results of vegetation effects on path loss for mobile cellular communication systems, IEEE Trans. Veh. Technol., vol. 49, pp. 12851293, July 2000. [103] G. L. Turin, F. F. Clapp, T. L. Johnston, S. B. Fine, and D. Lavry, A statistical model of urban multipath propagation, IEEE Trans. Veh. Technol., vol. VT-21, pp. 19, Feb. 1972. [104] H. Suzuki, A statistical model for urban radio propagation, IEEE Trans. Commun., vol. COM-25, pp. 673680, July 1977. [105] H. Hashemi, Simulation of the urban radio propagation channel, IEEE Trans. Veh. Technol., vol. VT-28, pp. 213225, Aug. 1979. [106] A. A. M. Saleh and R. A. Valenzuela, A statistical model for indoor multipath propagation, IEEE J. Select. Areas Commun., vol. SAC-5, pp. 128137, Feb. 1987. [107] G. G. Raleigh and J. M. Cioffi, Spatiotemporal code for wireless communication, IEEE Trans. Commun., vol. 46, pp. 357366, Mar. 1998. [108] J. C. Liberti and T. S. Rappaport, A geometrically based model for line-of-sight multipath radio channels, in IEEE Veh. Technol. Conf., Apr. 1996, pp. 844848. [109] M. Lu, T. Lo, and J. Litva, A physical spatiotemporal model of multipath propagation channels, in IEEE Veh. Technol. Conf., 1997, pp. 810814. [110] P. Zetterberg and B. Ottersten, The spectrum efficiency of a base station antenna array system for spatially selective transmission, in IEEE Veh. Technol. Conf., 1994, pp. 15171521. [111] W. H. Spencer, B. D. Jeffs, M. A. Jensen, and A. L. Swindlehurst, Modeling the statistical time and angle of arrival characteristics of an indoor multipath channel, IEEE J. Select. Areas Commun., vol. 18, pp. 347359, Mar. 2000. [112] A. Davidson and C. Hill, Measurement of building penetration into medium buildings at 900 and 1500 MHz, IEEE Trans. Veh. Technol., vol. 46, pp. 161168, Feb. 1997. [113] J. Berg, Building penetration loss along urban street microcells, in Proc. PIMRC96, pp. 795797. [114] O. Landron, M. J. Feuerstein, and T. S. Rappaport, A comparison of theoretical and empirical reflection coefficients for typical exterior wall surfaces in a mobile radio environment, IEEE Trans. Antennas Propagat., vol. 44, pp. 341351, Mar. 1996. [115] J. Lahteenmaki and T. Karttaavi, Measurement of dielectric parameters of wall materials at 60 GHz band, Electron. Lett., vol. 32, no. 16, pp. 14421444, Aug. 1996. [116] K. Sato, T. Manabe, T. Ihara, H. Saito, S. Ito, T. Tanka, K. Sugai, N. Homi, Y. Murakami, M. Shibayama, Y. Konishi, and T. Kimura, Measurements of reflection and transmission characteristics of interior structures of office buildings in the 60-GHz band, IEEE Trans. Antennas Propagat., vol. 45, pp. 17831792, Dec. 1997. [117] W. Honcharenko and H. L. Bertoni, Transmission and reflection characteristics at concrete block walls in the UHF band proposed for future PCS, IEEE Trans. Antennas Propagat., vol. 42, pp. 232239, Feb. 1994. [118] H. Chu, S. Jeng, and C. H. Chen, Reflection and transmission characteristics of lossy periodic composite structures, IEEE Trans. Antennas Propagat., vol. 44, pp. 580587, Mar. 1996. , Reflection and transmission characteristics of single-layer peri[119] odic composite structures for TE case, IEEE Trans. Antennas Propagat., vol. 45, pp. 10651070, July 1997. [120] H. Chiu, H. Chu, and C. H. Chen, Propagation modeling of periodic laminated composite structures, IEEE Trans. Electromag. Compat., vol. 40, pp. 218224, Aug. 1998. [121] E. Richalot, M. Bonilla, M. Won, V. Fouad-Hanna, H. Baudrand, and J. Wiart, Electromagnetic propagation into reinforced-concrete walls, IEEE Trans. Microwave Theory Tech., vol. 48, pp. 357366, Mar. 2000. [122] R. A. Dalke, C. L. Holloway, P. McKenna, M. Johansson, and A. S. Ali, Effects of reinforced concrete structures on RF communications, IEEE Trans. Electromag. Compat., vol. 42, pp. 486496, Nov. 2000. [123] C. L. Holloway, P. L. Perini, R. R. DeLyser, and K. C. Allen, Analysis of composite walls and their effects on short-path propagation modeling, IEEE Trans. Veh. Technol., vol. 46, pp. 730738, Aug. 1997. [124] M. F. Iskander, Z. Yun, and Z. Zhang, Outdoor/indoor propagation modeling for wireless communications systems, in IEEE AP-S Int. Symp. Dig., USNC/URSI Nat. Radio Sci. Meeting, vol. 2, July 813, 2001, pp. 150153. [125] Z. Zhang, R. K. Sorensen, Z. Yun, M. F. Iskander, and J. F. Harvey, A ray-tracing approach for indoor/outdoor propagation through window structures, IEEE Trans. Antennas Propagat., to be published.
673
Magdy F. Iskander (S72M76SM84F93) is currently a Professor of electrical engineering at the University of Utah, Salt Lake City. From 1997 to 1999, he was the Program Director in the Electrical and Communication Systems Division, National Science Foundation (NSF). While with the NSF, he formulated and directed a Wireless Information Technology initiative in the Engineering Directorate and funded over 29 projects in the microwave/millimeter-wave devices, RF microelectromechanical systems (MEMS) technology, propagation modeling, and antennas areas. He was the University of Utah, where he was the Director of the Center of Excellence for Multimedia Education and Technology (CAEME). The NSF is currently funding the CAEME to formulate the Conceptual Learning of Engineering (CoLoE) Consortium and develop educational software that implements this concept. In 1986, he established the Clinic Program to attract industrial support for projects for engineering students and has been the Director since its inception. To date, the program has attracted over 115 projects sponsored by 37 corporations from across the U.S. The Clinic Program now has an endowment for scholarships and a professorial chair at the University of Utah. He is currently with the Hawaii Center for Advanced Communications, College of Engineering, University of Hawaii, Honolulu, HI. He has authored or co-authored over 170 papers in technical journals, holds nine patents, and has made numerous presentations in technical conferences. He authored Electromagnetic Fields and Waves (Englewood Cliffs, NJ: Prentice-Hall, 1992), edited the CAEME Software Books, Vol. I (1991) and CAEME Software Books, Vol. II (1994), and edited four other books on microwave processing of materials. He has also edited two special issues of the Journal of Microwave Power, a special issue of the ACES Journal, and the 1995 and 1996 Proceedings of the International Conference on Simulation and Multimedia in Engineering Education. He is the founding editor of Computer Applications in Engineering Education (CAE). This journal received the 1993 Excellence in Publishing Award presented by the American Association of Publishers. His ongoing research contracts include propagation models for wireless communication, funded by the Army Research Office and the NSF, low-cost phased array antennas, funded by both the Army Research Laboratory and the NSF, microwave processing of materials, funded by Corning Inc., and the conceptual learning of engineering project funded by the NSF. He has spent sabbatical and other short leaves at the Polytechnic University of New York, Ecole Superieure DElectricite, University of California at Los Angeles, Harvey Mudd College, Tokyo Institute of Technology, Polytechnic University of Catalunya, and at several universities in China. Dr. Iskander was a member of the National Research Council Committee on Microwave Processing of Materials. He was a member of the WTEC panel on Wireless Information Technology, and the chair of the Panel on Asia Telecommunications sponsored by the Department of Defense (DoD) and organized by the International Technology Research Institute (ITRI) in 20002001. As part of these studies, he visited many wireless companies in Europe, Japan, and several telecommunications institutions and companies in Taiwan, Hong Kong, and China. He was the general chair of the 1996 Frontiers in Education Conference, Salt Lake City, UT. He was a member of the IEEE Antennas and Propagation Society (IEEE AP-S) AdCom (19971999) and the general chair of the 2000 IEEE AP-S Symposium and URSI Meeting, Salt Lake City, UT. He was a distinguished lecturer for the IEEE AP-S (19941997) and was the 2001 vice president of the IEEE AP-S. While serving as a distinguished lecturer, he gave lectures in Brazil, France, Spain, China, Japan, and at a large number of U.S. universities and IEEE chapters. He organized the first Wireless Grantees Workshop sponsored by the NSF in 2001. He was the recipient of the 1985 Curtis W. McGraw American Society for Engineering Education (ASEE) National Research Award, the 1991 ASEE George Westinghouse National Education Award, the 1992 Richard R. Stoddard Award presented by the IEEE Electromagnetic Compatibility Society, and the 2000 University Distinguished Teaching Award.
Zhengqing Yun (M98) received the Ph.D. degree in electrical engineering from Chongqing University, Chongqing, China, in 1994. From 1995 to 1997, he was a Post-Doctoral Fellow with the State Key Laboratory of Millimeter Waves, Southeast University, Nanjing, China. Since 1997, he has been with the Electrical Engineering Department, University of Utah, Salt Lake City, where he is currently a Research Assistant Professor. His recent research interests include development of numerical methods and modeling of radio propagation for wireless communications systems, design, and simulation of antennas. Dr. Yun was the recipient of the 1997 Science and Technology Progress Award (First Class) presented by The State Education Commission of China.
INDOOR PROPAGATION MODELING AT 2.4 GHZ FOR IEEE 802.11 NETWORKS Dinesh Tummala, B.S.
Thesis Prepared for the Degree of MASTER OF SCIENCE
UNIVERSITY OF NORTH TEXAS December 2005
APPROVED: Robert Akl, Major Professor Albert B. Grubbs, Committee Member and Chair of the Department of Engineering Technology Robert G. Hayes, Committee Member Vijay Vaidyanathan, Committee Member Oscar N. Garcia, Dean of the College of Engineering Sandra L. Terrell, Dean of the Robert B. Toulouse School of Graduate Studies
Tummala, Dinesh. Indoor Propagation Modeling at 2.4 GHz for IEEE 802.11 Networks. Master of Science (Engineering Technology), December 2005, 109 pp., 18 tables, 80 figures, 47 titles. Indoor use of wireless systems poses one of the biggest design challenges. It is difficult to predict the propagation of a radio frequency wave in an indoor environment. To assist in deploying the above systems, characterization of the indoor radio propagation channel is essential. The contributions of this work are two-folds. First, in order to build a model, extensive field strength measurements are carried out inside two different buildings. Then, path loss exponents from log-distance path loss model and standard deviations from log-normal shadowing, which statistically describe the path loss models for a different transmitter receiver separations and scenarios, are determined. The purpose of this study is to characterize the indoor channel for 802.11 wireless local area networks at 2.4 GHz frequency. This thesis presents a channel model based on measurements conducted in commonly found scenarios in buildings. These scenarios include closed corridor, open corridor, classroom, and computer lab. Path loss equations are determined using log-distance path loss model and log-normal shadowing. The chi-square test statistic values for each access point are calculated to prove that the observed fading is a normal distribution at 5% significance level. Finally, the propagation models from the two buildings are compared to validate the generated equations.
ACKNOWLEDGEMENTS I express my gratitude of thanks to my major advisor and Chair of Department of Engineering Technology Dr. Albert B.Grubbs for his support, confidence, encouragement, and guidance throughout my graduate program and in completing this thesis. From the formative stages of this thesis, to the final draft, I am indebted to my supervisor, Dr. Robert Akl, whose expertise, understanding, and patience, added considerably to my graduate experience. I thank the members of my committee, Dr. Robert G. Hayes, and Dr. Vijay Vaidyanathan for the assistance, careful reading and suggestions they provided at all levels of the research project. I would also like to thank my supervisors at work Mr. Ken Brinkley and Mr. Larry Guay for being supportive. I really appreciate the flexible work schedule you provided. I would also like to thank my family for the support they provided me through my entire life and in particular, I must acknowledge my wife and best friend, Swathi, without whose love, encouragement and editing assistance, I would not have finished this thesis. Finally, but not least, I want to thank my parents Mr. Gopala Krishna and Mrs. Vijaya. Thanks for encouraging me to be an independent thinker, and having confidence in my abilities to go after new things that inspired me. Thanks Dad for taking time to work with me on my mathematics (You were the best mathematics teacher I ever had). Thanks for teaching me that it is important to try to leave the world just a little better than when you came into it. And, of course, thank you both for your constant support through the ups and downs of my academic career. It has been bumpy at times, but your confidence in me has enhanced my ability to get through it all and succeed in the end.
ii
TABLE OF CONTENTS Page ACKNOWLEDGEMENTS..........................................................................................................ii LIST OF TABLES........................................................................................................................ v LIST OF FIGURES ..................................................................................................................... vi Chapters 1. INTRODUCTION ................................................................................................ 1 1.1 1.2 1.3 1.4 2. Communication Systems .......................................................................... 2 Problem Statement and Objectives ........................................................... 3 Research Questions................................................................................... 3 Organization.............................................................................................. 4
INDOOR PROPAGATION MODELING ........................................................... 5 2.1 2.2 2.3 2.4 2.5 Introduction............................................................................................... 5 Line of Sight Propagation ......................................................................... 5 Channel Modeling..................................................................................... 8 Propagation Models .................................................................................. 8 Empirical Models...................................................................................... 9 2.5.1 Log-Distance Path Loss Model................................................... 10 2.5.2 2.6 2.7 2.8 2.9 Log-Normal Shadowing.............................................................. 10 Two-Ray Model ...................................................................................... 11 Indoor RF Propagation and Wireless Local Area Network Technology ................................................................................................................. 13 IEEE 802.11 Standard............................................................................. 15 Frequency Range and Channel Allocation for 802.11b.......................... 16
3.
STATISTICAL ANALYSIS .............................................................................. 19 3.1 3.2 3.3 3.3.1 Normal Plot............................................................................................. 19 Standard Normal Distribution................................................................. 20 Chi-square Goodness-of-Fit Test............................................................ 20 Deciding What Value of X2 is Critical for Accepting or Rejecting a Hypothesis............................................................................................... 22
iii
3.3.2 3.4 4.
Finding p-Value for Chi-square Test ...................................................... 23
Curve Fitting ....................................................................................................... 24 EXPERIMENTAL SETUP................................................................................. 27 4.1 4.2 Introduction............................................................................................. 27 Propagation Environment ....................................................................... 27 4.2.1 Description of Measurement Scenarios for Building One.......... 31 4.2.2 Description of Measurement Scenarios for Building Two ......... 34 4.3 4.4 4.5 Hardware and Software Description....................................................... 34 Data Acquisition ..................................................................................... 36 Numerical Analysis................................................................................. 36
5.
NUMERICAL ANALYSIS ................................................................................ 38 5.1 5.2 Introduction............................................................................................. 38 Analysis of Results ................................................................................. 40 5.2.1 Scenario 1: Building One Closed Corridor................................. 40 5.2.2 Scenario 2: Building One Open Corridor ................................... 50 5.2.3 Scenario 3: Building One Classroom.......................................... 59 5.2.4 Scenario 4: Building One Computer Lab.................................... 71 5.3 Summary of Results................................................................................ 80
6.
COMPARISON WITH SIMILAR SCENARIOS .............................................. 83 6.1 6.2 6.3 6.4 6.5 6.6 Introduction............................................................................................. 83 Scenario 1: Building Two Closed Corridor ............................................ 83 Scenario 2: Building Two Classroom ..................................................... 93 Summary ............................................................................................... 103 Conclusion ............................................................................................ 105 Future Work .......................................................................................... 106
BIBLIOGRAPHY......................................................................................................... 107
iv
LIST OF TABLES Page 2.1 2.2 3.1 3.2 5.1 5.2 5.3 5.4 5.5 5.6 5.7 5.8 6.1 6.2 6.3 6.4 6.5 6.6 Path loss exponents for different environments [1] ........................................................ 11 Frequency and channel assignments............................................................................... 17 Area under the normal curve from 0 to X....................................................................... 21 Critical values of chi-square distribution with degrees of freedom................................ 25 Closed corridor X2 table for AP1 (D-Link).................................................................... 46 Closed corridor X2 table for AP2 (LinkSys) ................................................................. 47 Open corridor X2 table for AP1 (D-Link) ...................................................................... 57 Open corridor X2 table for AP2 (LinkSys)..................................................................... 58 Classroom X2 table for AP1 (D-Link)............................................................................ 67 Classroom X2 table for AP2 (LinkSys) .......................................................................... 68 Computer lab X2 table for AP1 (D-Link)....................................................................... 77 Computer lab X2 table for AP2 (LinkSys) ..................................................................... 78 Closed corridor X2 table for AP1 (D-Link).................................................................... 90 Closed corridor X2 table for AP2 (LinkSys) .................................................................. 91 Closed corridor results from Building One and Building Two....................................... 92 Classroom X2 table for AP1 (D-Link).......................................................................... 100 Classroom X2 table for AP2 (LinkSys) ........................................................................ 101 Classroom results from Building One and Building Two ............................................ 103
LIST OF FIGURES Page 1.1 2.1 2.2 4.1 4.2 4.3 4.4 4.5 4.6 4.7 4.8 4.9 5.1 5.2 5.3 5.4 5.5 5.6 5.7 5.8 5.9 5.10 5.11 Basic communication system............................................................................................ 2 Two-ray model................................................................................................................ 13 802.11b channel overlap ................................................................................................. 18 College of Engineering, first floor overview .................................................................. 28 College of Engineering, second floor overview ............................................................. 29 College of Engineering, second floor detail ................................................................... 30 Building One closed corridor.......................................................................................... 31 Building One open corridor ............................................................................................ 32 Building One classroom.................................................................................................. 33 Building One computer lab............................................................................................. 33 Building Two closed corridor ......................................................................................... 35 Building Two classroom ................................................................................................. 35 Numerical analysis.......................................................................................................... 39 Building One, closed corridor loss for AP1 (D-Link) .................................................... 40 Building One, closed corridor loss for AP2 (LinkSys)................................................... 41 Building One, closed corridor mean loss for AP1 and AP2 ........................................... 42 Building One, closed corridor curve fitting for AP1 (D-Link)....................................... 43 Building One, closed corridor curve fitting for AP2 (LinkSys) ..................................... 43 Building One, closed corridor normal probability plot for AP1 (D-Link) ..................... 44 Building One, closed corridor normal probability plot for AP2 (LinkSys).................... 45 Building One, closed corridor standard normal distribution for AP1 (D-Link) ............. 45 Building One, closed corridor standard normal distribution for AP2 (LinkSys)............ 48 Comparison of closed corridor mean loss with two-ray model ...................................... 49
vi
5.12 5.13 5.14 5.15 5.16 5.17 5.18 5.19 5.20 5.21 5.22 5.23 5.24 5.25 5.26 5.27 5.28 5.29 5.30 5.31 5.32 5.33 5.34 5.35
Comparison of closed corridor AP1 and AP2 loss with mean signal and two-ray model ......................................................................................................................................... 50 Building One, open corridor loss for AP1 (D-Link)....................................................... 51 Building One, open corridor loss for AP2 (LinkSys) ..................................................... 51 Building One, open corridor mean loss for AP1 and AP2.............................................. 52 Building One, open corridor curve fitting for AP1 (D-Link) ......................................... 53 Building One, open corridor curve fitting for AP2 (LinkSys)........................................ 54 Building One, open corridor normal probability plot for AP1 (D-Link)........................ 54 Building One, open corridor normal probability plot for AP2 (LinkSys) ...................... 55 Building One, open corridor standard normal distribution for AP1 (D-Link)................ 56 Building One, open corridor standard normal distribution for AP2 (LinkSys) .............. 56 Comparison of open corridor mean loss with two-ray model......................................... 60 Comparison of open corridor AP1 and AP2 loss with mean signal and two-ray model ......................................................................................................................................... 60 Building One, classroom loss for AP1 (D-Link) ............................................................ 61 Building One, classroom loss for AP2 (LinkSys)........................................................... 61 Building One, classroom mean loss for AP1 and AP2 ................................................... 62 Building One, classroom curve fitting for AP1 (D-Link)............................................... 63 Building One, classroom curve fitting for AP2 (LinkSys) ............................................. 64 Building One, classroom normal probability plot for AP1 (D-Link) ............................. 65 Building One, classroom normal probability plot for AP2 (LinkSys)............................ 65 Building One, classroom standard normal distribution for AP1 (D-Link) ..................... 66 Building One, classroom standard normal distribution for AP2 (LinkSys).................... 66 Comparison of classroom mean loss with two ray-model .............................................. 70 Comparison of classroom AP1 and AP2 loss with mean signal and two-ray model...... 70 Building One, computer lab loss for AP1 (D-Link) ....................................................... 71
vii
5.36 5.37 5.38 5.39 5.40 5.41 5.42 5.43 5.44 5.45 5.46 6.1 6.2 6.3 6.4 6.5 6.6 6.7 6.8 6.9 6.10 6.11 6.12 6.13
Building One, computer lab loss for AP2 (LinkSys)...................................................... 72 Building One, computer lab mean loss for AP1 and AP2 .............................................. 72 Building One, computer lab curve fitting for AP1 (D-Link) .......................................... 74 Building One, computer lab curve fitting for AP2 (LinkSys) ........................................ 74 Building One, computer lab normal probability plot for AP1 (D-Link)......................... 75 Building One, computer lab normal probability plot for AP2 (LinkSys) ....................... 76 Building One, computer lab standard normal distribution for AP1 (D-Link) ................ 76 Building One, computer lab standard normal distribution for AP2 (LinkSys)............... 79 Comparison of computer lab mean loss with two ray model.......................................... 80 Comparison of computer lab AP1 and AP2 loss with mean signal and two ray model ......................................................................................................................................... 81 Results at a glance........................................................................................................... 82 Building Two, closed corridor loss for AP1 (D-Link).................................................... 84 Building Two, closed corridor loss for AP2 (LinkSys) .................................................. 84 Building Two, closed corridor mean loss for AP1 and AP2........................................... 85 Building Two, closed corridor curve fitting for AP1 (D-Link) ...................................... 86 Building Two, closed corridor curve fitting for AP2 (LinkSys)..................................... 87 Building Two, closed corridor normal probability plot for AP1 (D-Link)..................... 88 Building Two, closed corridor normal probability plot for AP2 (LinkSys) ................... 88 Building Two, closed corridor standard normal distribution for AP1 (D-Link)............. 89 Building Two, closed corridor standard normal distribution for AP2 (LinkSys) ........... 89 Comparison of closed corridor mean loss with two-ray model ...................................... 93 Comparison of closed corridor AP1 and AP2 loss with mean signal and two-ray model ......................................................................................................................................... 94 Building Two, classroom loss for AP1 (D-Link)............................................................ 94 Building Two, classroom loss for AP2 (LinkSys) .......................................................... 95
viii
6.14 6.15 6.16 6.17 6.18 6.19 6.20 6.21 6.22
Building Two, classroom mean loss for AP1 and AP2 .................................................. 96 Building Two, classroom curve fitting for AP1 (D-Link) .............................................. 97 Building Two, classroom curve fitting for AP2 (LinkSys)............................................. 97 Building Two, classroom normal probability plot for AP1 (D-Link)............................. 98 Building Two, classroom normal probability plot for AP2 (LinkSys) ........................... 99 Building Two, classroom standard normal distribution for AP1 (D-Link) .................... 99 Building Two, classroom standard normal distribution for AP2 (LinkSys)................. 102 Comparison of classroom mean loss with two-ray model ............................................ 104 Comparison of classroom AP1 and AP2 loss with mean signal and two-ray model.... 104
ix
CHAPTER 1
INTRODUCTION Wireless communication is one of the most active areas of technology development of our time. This development is being driven primarily by the transformation of what has been largely a medium for supporting voice telephony into a medium for supporting other services, such as the transmission of video, images, text, and data. Thus, similar to the developments in wireline capacity in the 1990s, the demand for new wireless capacity is growing at a very rapid pace. The impact of wireless technology has been and will continue to be profound. The convergence of dierent standards that dene how wireless devices interact will allow the creation of a global wireless network that will deliver a wide variety of services. Cellular phones are currently the most obvious sign of the advent of wireless technology, but mobile telephones are only the tip of the cellular revolution. The rst rush to wireless was for voice. Now, the attention is on data. Presently, there are many types of wireless networks in use around the world. Most new devices have access to the internet. A big part of this market is the wireless internet. The Internet is increasingly becoming a multimedia experience. For wireless networks to compete with their xed counterparts, wireless networks need to obtain higher data rates [2]. Wireless Local Area Networks (WLANs) provide network services where it is dicult or too expensive to deploy a xed infrastructure. WLANs can coexist with xed infrastructure to provide mobility and exibility to users. The primary WLAN standards are IEEE 802.11 [3] and Europes HyperLAN [4]. The 802.11 protocol set, popularly known as Wi-Fi, includes wireless network standards that allow data transmission up to a theoretical 54 Mbps [5]. WLANs operate mainly in a indoor environment. It is very dicult to predict how a 1
Figure 1.1: Basic communication system. RF wave travels in an indoor environment. So there is a need for developing an indoor propagation model to predict RF wave behavior more accurately.
1.1 Communication Systems Any communication system can be viewed as a link between a source and a destination where information is sent from the source and received at the destination. The intervening stages are shown in Figure 1.1. The transmitter takes the information from the source and codes it in a form suitable for transfer over the channel such that the cost of transmission is minimal. In this context, cost is a function of the bandwidth used, the time taken to perform the communication, the degree to which the transmission interferes with other transmissions occurring simultaneously and the amount of information that is lost in the communication process. The channel is a description of how the communications medium alters the signal that is being transmitted. Finally the receiver takes the signals that have been altered by the channel, and attempts to recover the information that was sent by the source. This recovered signal is passed to the destination as the received information. For a radio communication system, the channel describes how the electromagnetic propagation of a transmitted signal provides that signal at the receiver. In a mobile communication system, the channel changes according to the movement of the communicating entities and other objects that have an eect on the electromagnetic elds at the receiver. The purpose of this study is to characterize the indoor channel for 802.11b wireless local area networks at 2.4 GHz frequency [6].
1.2 Problem Statement and Objectives WLANs are rapidly gaining popularity. These networks are primarily targeted for indoor use, and are most often based on either the IEEE 802.11 Ethernet-type protocols or the Bluetooth Special Interest Group (SIG), both using the unlicensed bands at 2.4 to 2.5 GHz, IEEE 802.11b [6] and Bluetooth [7]), or at 5.15 to 5.85 GHz, IEEE 802.11a [8]. The European HiperLAN standard is also designed for operation around 5.2 to 5.8 GHz [4]. Indoor use of wireless systems poses one of the biggest design challenges. It is dicult to predict the propagation of a RF wave in an indoor environment [9][10]. To assist in deploying the above systems, characterization of the indoor radio propagation channel is essential. The proposed innovation characterizes the indoor channel by developing a propagation model at 2.4 GHz for dierent scenarios commonly found in buildings. By using this propagation model, network analysis and simulation can be developed. This will facilitate faster and more ecient deployment of wireless networks.
1.3 Research Questions The research questions addressed in this study are stated below for hypotheses testing. 1. Is multipath fading observed in indoor radio propagation at 2.4 GHz distributed normally? Null Hypotheses: Multipath fading observed in indoor radio propagation at 2.4 GHz is not a normal distribution. Alternative Hypotheses: Multipath fading observed in indoor radio propagation at 2.4 GHz is a normal distribution.
Normal probability plot and Chi-square Goodness of t tests are used to verify the above hypotheses. 2. Does indoor radio propagation at 2.4 GHz depend on the indoor environment? Indoor environment or indoor channel consists of hard and soft partitions (Section 2.7). Null Hypotheses: Indoor radio propagation at 2.4 GHz is not dependent on the indoor environment. Alternative Hypotheses: Indoor radio propagation at 2.4 GHz is dependent on the indoor environment.
1.4 Organization This thesis creates a model of an indoor propagation channel at 2.4 GHz for dierent scenarios commonly found in buildings. The remainder of this work consists of the following chapters. Chapter 2 reviews the published work that has been conducted in the eld of channel measurement and modeling. Concepts in channel modeling and measurements are examined. Chapter 3 describes the statistical analysis required to measure the quality of the models developed in this study. Chapter 4 describes the experimental setup, the software and hardware used for measurements, and the propagation environment in which measurements are done. A detailed description of each measurement scenario is discussed. Chapter 5 does a numerical analysis of measurements conducted in each scenario. The study determines equations that describe path loss for each scenario. Chapter 6 compares the measurements from scenarios in dierent buildings and compares them to measurements from similar scenarios in Chapter 5. Conclusions and possible direction of future work in this eld are also presented.
CHAPTER 2
INDOOR PROPAGATION MODELING 2.1 Introduction A signal radiated from an antenna travels along one of the three routes: ground wave, sky wave, or line of sight (LOS). Based on the operating frequency range, one of the three predominates. LOS propagation is the mode of propagation which is of interest in this paper.
2.2 Line of Sight Propagation At frequencies higher than 30 MHz, LOS is the dominant propagation mode. The ionosphere reects less of the signal as the frequency is increased beyond 30 MHz. A signal can thus be transmitted either to a satellite or to a receiving antenna which is in the line of sight of the transmitting antenna. In a communication system, a received signal will dier from the transmitted signal due to various transmission impairments. The most signicant transmission impairments for LOS transmission are [2]: Attenuation: The strength of a signal falls o with distance over any transmission medium. This reduction in strength or attenuation is logarithmic for guided media. Whereas attenuation is a more complex function of distance and the makeup of the atmosphere for an unguided media. Free space loss: In any wireless communication, the signal disperses with distance. A receiving antenna will receive less signal power the farther it is from the transmitting antenna. Assuming all the sources of impairments are nullied the transmitted signal
attenuates over distance because the signal is being spread over a larger and larger area. This form of attenuation is known as free space loss. Fading: Fading refers to the time variation of received signal power caused by changes in the transmission medium or path. Fading is the most challenging technical problem in designing a communication system. In a xed environment, fading is aected by changes in atmospheric conditions. Whereas in a mobile environment where either the receiving or transmitting antenna is in motion relative to the other, the relative location of various obstacles changes with time, causing complex transmission eects. Multipath: Multipath is dened as a propagation phenomenon that results in radio signals reaching the receiving antenna by two or more paths. The direct and reected signals are often opposite in phase, which can result in a signicant signal loss due to mutual cancelation in some circumstances. Depending on the dierences in the path lengths of direct and reected waves, the composite signal can be either larger or smaller than the direct signal. Multipath is most troublesome indoors and in areas where many metallic surfaces are present. Multipath is caused by the following propagation mechanisms: Reection: Reection occurs when a propagating electromagnetic wave impinges upon an object which has very large dimensions when compared to the wavelength of the propagating wave. Reections occur from the surface of the earth and from buildings and walls. The reected waves may interfere constructively or destructively at the receiver. Diraction: Diraction occurs when the radio path between the transmitter and receiver is obstructed by a surface that is large compared to the wavelength of the radio wave. The secondary waves resulting from the obstructing surface are present throughout the space and even behind the obstacle, giving rise to a bend6
ing of waves around the obstacle, even when a line-of-sight path does not exist between transmitter and receiver. Scattering: Scattering occurs when the medium through which the wave travels consists of objects with dimensions that are small compared to the wavelength, and where the number of obstacles per unit volume is large. Scattered waves are produced by rough surfaces, small objects, or by other irregularities in the channel. These three propagation eects inuence system performance in various ways depending on local conditions and as a mobile unit moves through the medium. Diraction and scattering are generally minor eects if there is a clear LOS between transmitter and receiver although reection may have a signicant impact. In cases where there is no LOS, diraction and scattering are the primary means of signal reception. Refraction: Refraction is dened as a change in direction of an electromagnetic wave resulting from changes in the velocity of propagation of the medium through which it passes. This may result in a situation in which only a fraction or no part of the line of sight wave reaches the receiving antenna. Noise: In any transmission event, a received signal will consist of the transmitted signal, modied by various distortions imposed by the transmission medium, plus additional unwanted signals that are inserted by the medium. These unwanted signals are referred to as noise or interference. Noise is the major limiting factor in any communications system performance. Atmospheric absorption: Atmospheric absorption is an additional loss due to the presence of dierent atmospheric elements such as water vapor and oxygen etc. A peak attenuation occurs in the vicinity of 22 GHz due to water vapor. At frequencies below 15 GHz, the attenuation is less. 7
2.3 Channel Modeling In order to evaluate the eectiveness of coding and processing techniques for a given channel prior to construction, a model of the channel must be developed that adequately describes the environment. Such analysis reduces the cost of developing a complex system by reducing the amount of hardware required for evaluation of performance. Indoor channels are highly dependent upon the placement of walls and partitions within the building. As placement of these walls and partitions dictates the signal path inside a building. In such cases, a model of the environment is a useful design tool in constructing a layout that leads to ecient communication strategies. To achieve this aim, a channel model of an indoor environment must be applied to various layout plans of oces which will lead to the characterization of design methodologies. A channel model is useful in determining the mechanisms by which propagation in the indoor environment occurs, which in turn is useful in the development of a communication system. By examining the details of how a signal is propagated from the transmitter to the receiver for a number of experimental locations, a generic model may be developed that highlights the important characteristics of a given indoor environment. Generic models of indoor communications can then be applied to specic situations to describe the operation of a radio system, and may also be used to generate building designs that are particularly suited to supporting radio communication systems.
2.4 Propagation Models In the literature, there are numerous experimental and theoretical studies of indoor propagation [11][12][13][14][15][16][10]. These models tend to focus on a particular characteristic like temporal fading or inter-oor losses. This study aims at developing an indoor propagation model from measurements taken using 802.11b compliant access point and client adapters.
The study focuses on generating dierent loss equations for dierent scenarios in indoor environments. Using these equations, an accurate model can be developed to visualize the propagation phenomenon for dierent buildings with changing indoor environments. A propagation model is a set of mathematical expressions,diagrams, and algorithms used to represent the radio characteristics of a given environment [17]. The prediction models can be either empirical (also called statistical) or theoretical (also called deterministic), or a combination of these two. While the empirical models are based on measurements, the theoretical models deal with the fundamental principles of radio wave propagation phenomena. In the empirical models, all environmental inuences are implicitly taken into account regardless of whether they can be separately recognized. This is the main advantage of these models. Because deterministic models are based on the principles of physics they may be applied to dierent environments without aecting the accuracy. In practice, their implementation usually requires a huge database of environmental characteristics, which is sometimes either impractical or impossible to obtain. The algorithms used by deterministic models are usually very complex and lack computational eciency. For that reason, the implementation of the deterministic models is commonly restricted to smaller areas of microcell or indoor environments. On the basis of the radio environment, the prediction models can be classied into two main categories, outdoor and indoor propagation models. Further, in respect to the size of the coverage area, the outdoor propagation models can be subdivided into two additional classes, macrocell and microcell prediction models. A discussion of one empirical and one theoretical model used in this study is presented in the following sections.
2.5 Empirical Models Both theoretical and measurement based propagation models indicate that average received signal power decreases logarithmically with distance. Empirical models help in reducing
computational complexity as well as increasing the accuracy of the predictions [17]. The empirical model used in this study is Log-distance Path Loss Model.
2.5.1 Log-distance Path Loss Model In both indoor and outdoor environments the average large-scale path loss for an arbitrary Transmitter-Receiver (T-R) separation is expressed as a function of distance by using a path loss exponent, n [1]. The average path loss P L(d) for a transmitter and receiver with separation d is:
P L(d) or
d d0
(2.1)
P L(dB) = P L(d0 ) + 10 n log
d , d0
(2.2)
where n is the path loss exponent which indicates the rate at which path loss increases with distance d. Close in reference distance (d0 ) is determined from measurements close to the transmitter. The plot for distance d versus path loss P L on a log-log scale is a straight line with a slope equal to 10n. This value of n depends on the specic propagation environment, i.e., type of construction material, architecture, location within building. Lower the value of n lower the signal loss. The values of n range from 1.2 (Waveguide eect) to 6 [17]. For example, in free space, n is equal to 2, and when obstructions are present, n will have a larger value. Table 2.1 lists typical path loss exponents obtained in various radio environments.
2.5.2 Log-Normal Shadowing Random shadowing eects occuring over a large number of measurement locations which have the same T-R separation, but dierent levels of clutter on the propagation path is referred to as Log-Normal Distribution [1]. This phenomenon is referred to as log-normal
10
Table 2.1: Path loss exponents for dierent environments [1].

Environment Free Space Urban area cellular radio Shadowed urban cellular radio In building line-of sight Obstructed in building Obstructed in factories Path Loss Exponent, n 2 2.7 to 3.5 3 to 5 1.6 to 1.8 4 to 6 2 to 3
shadowing. Variations in environmental clutter at dierent locations having the same T-R separation is not accounted for in equation (2.2). This leads to measured signals which are vastly dierent than the average value predicted by (2.2). To account for the variations described above equation (2.2) is modied as:
P L(dB) = P L(d0 ) + 10 n log
d + X , d0
(2.3)
where X is a zero-mean Gaussian distributed random variable with standard deviation . The close-in reference distance d0 , the path loss exponent n, and the standard deviation , statistically describe the path loss model for an arbitrary location having a specic T-R separation. This model can be used in computer simulation to provide received power levels for random locations in communication system design and analysis.
2.6 Two-Ray Model Site specic propagation models are based on electromagnetic-wave propagation theory to characterize indoor radio propagation. Unlike statistical models, site specic propagation models do not rely on extensive measurement, but a greater detail of the indoor environment is required to obtain an accurate prediction of signal propagation inside a building.
11
In theory, electromagnetic-wave propagation characteristics could be exactly computed by solving Maxwells equations with the building geometry as boundary conditions. Unfortunately, this approach requires very complex mathematical operations and requires considerable computing power, beyond that of current microcomputers. Hence it is not economical for the characterization of indoor radio wave propagation. Therefore, approximate numerical methods are of interest. Ray tracing is an intuitively appealing method for calculating radio signal strength, time-invariant impulse response, root mean square (RMS) delay spread and related parameters in an indoor environment [18][19][20][21]. The concept of ray-tracing modeling is based on the fact that high-frequency radio waves behave in a ray-like fashion. Therefore, signal propagation can be modeled as ray propagation. By using the concept of ray-tracing, rays may be launched from a transmitter location and the interaction of the rays with partitions within a building modeled using well-known reection and transmission theory. Ray tracing can be much less demanding of computation than methods based on Maxwells equations. With the computing powers currently available on personal computers and workstations, the ray-tracing approach provides a challenging but feasible method of propagation modeling. Reliable site specic ray-tracing propagation prediction models, for each building that is based on its detailed geometry and construction, can be very eective tools in designing indoor communication systems. The ray-tracing approach approximates the scattering of electromagnetic waves by simple reection and refraction. The degree of transmission and reection of a signal through and o an obstacle is related to the complex permittivities of the obstacle. One of the propagation models based on ray-optic theory is a Two-Ray model. Two-Ray model is used in this study because all the scenarios considered in this study have one reecting surface, i.e. we have a direct path and reected path. It is used for modeling of Line of Sight radio channel as shown in Figure 2.1. The transmitting antenna of height h1 and the receiving antenna of
12
Figure 2.1: Two-Ray Model. height h2 are placed at distance d from each other. The received signal Pr for isotropic antennas, obtained by summing the contribution from each ray, can be expressed as Pr = Pt 4 1 1 (jkr1 ) e + () e(jkr2 ) r1 r2
2
(2.4)
where Pt is the transmitted power, r1 is the direct distance from the transmitter to the receiver, r2 is the distance through reection on the ground, and () is the reection coecient depending on the angle of incidence and the polarization. The reection coecient is given by cos a () = cos + a
r r
sin2 , sin2
(2.5) is
where = 90- and a = 1/ or 1 for vertical or horizontal polarization, respectively.
a relative dielectric constant of the reected surface. The signal strengths from theoretical and empirical models are compared in this study.
2.7 Indoor RF Propagation and Wireless Local Area Network Technology Indoor use of wireless systems poses one of the biggest design challenges. RF propagation obstacles can be termed hard partitions if they are part of the physical or structural components of a building. On the other hand, obstacles formed by oce furniture and xed or movable structures that do not extend to a buildings ceiling are considered soft partitions. Radio signals eectively penetrate both kinds of obstacles or partitions in ways that are very 13
hard to predict. An indoor propagation model is needed to predict the propagation of a transmitted signal in this environment. WLANs are implemented as an extension to wired LANs within a building and can provide the nal few meters of connectivity between a wired network and the mobile user. WLAN congurations vary from simple, independent, peer-to-peer connections between a set of PCs, to more complex, intra-building infrastructure networks. There are also point-topoint and point-to-multipoint wireless solutions. A point-to-point solution is used to bridge two local area networks, and to provide an alternative to cable between two geographically distant locations (up to 30 miles). Point-to-multi-point solutions connect several, separate locations to one single location or building. Both point-to-point and point-to-multipoint can be based on the 802.11b standard or on more costly infrared-based solutions that can provide throughput rates up to 622 Mbps. In a typical WLAN infrastructure conguration, there are two basic components: Access Points: An access point or a base station connects to a LAN by means of Ethernet cable. Usually installed in the ceiling, access points receive, buer, and transmit data between the WLAN and the wired network infrastructure. A single access point supports on average twenty users and has a coverage varying from 20 meters in areas with obstacles (walls, stairways, elevators) up to 100 meters in areas with clear line of sight. A building may require several access points to provide complete coverage and allow users to roam seamlessly between access points. Wireless Client Adapter: A wireless adapter connects users via an access point to the rest of the LAN. A wireless adapter can be a PC card in a laptop, an ISA or PCI adapter in a desktop computer, or fully integrated within a handheld device.
14
2.8
IEEE 802.11 Standard
IEEE 802.11 is a family of specications for WLANs developed by the Institute of Electrical and Electronics Engineers. The 802.11 standard species parameters for both the physical and medium access control (MAC) layers of a WLAN [3]. The physical layer handles the transmission of data between nodes. The MAC layer consists of protocols responsible for maintaining the use of the shared medium. Work on 802.11 began in 1987 within the IEEE 802.4 group. There are three physical layers for WLANs: two radio frequency specications (RF direct sequence and frequency hopping spread spectrum) and one infrared. Most WLANs operate in the 2.4 GHz license-free frequency band and have throughput rates up to 2 Mbps. There are various versions of the 802.11 standard. A brief description of the more popular revisions is given below. 802.11a: 802.11a operates at radio frequencies between 5 GHz and 6 GHz [8]. The modulation scheme used is orthogonal frequency-division multiplexing (OFDM). OFDM, also called multicarrier modulation, uses multiple carrier signals at dierent frequencies, sending some of the bits on each channel. This is similar to Frequency Division Multiplexing (FDM). The only dierence between FDM and OFDM is that in OFDM all the sub-channels are dedicated to a single data source. The data rates vary based on the noise level, distance from the transmitting antenna, and the propagation environment. Possible data rates for 802.11a are 6, 9, 12, 18, 24, 36, 48, and 54 Mbps. Maximum range for this standard is 200 feet. 802.11b: 802.11b often called Wi-Fi, being the most popular of all the standards, operates in the 2.4 GHz frequency [6]. It is an extension of the 802.11 standard. Typical data rates for 802.11b are 5.5 and 11 Mbps. The modulation scheme used is Direct Sequence Spread Spectrum. The chipping rate is 11 MHz, the same as in 15
802.11, providing the same occupied bandwidth. Although the data rates are slower than 802.11a, the range is higher, up to 300 feet. The frequency band used (2.4 GHz) can have signicant interference problems from such devices as microwave, cordless phones, and Bluetooth devices. 802.11g: 802.11g is the newest member of the 802.11 family. This standard combines the best of 802.11a and 802.11b. Like 802.11b, 802.11g operates in the 2.4 GHz frequency and can achieve ranges up to 300 feet, but like 802.11a, it reaches speeds up to 54 Mbps. 802.11g uses a hybrid complementary code keying OFDM modulation [5]. 802.11i: 802.11b uses Wired Equivalent Privacy (WEP) protocol to address security concerns. WEP itself is more or less an implementation of encryption with built-in message authentication and data integrity systems. The sheer number and variety of vulnerabilities discovered within WEP shows what could arise when security is not designed from the ground up. The future of wireless LAN security is currently being entrusted to 802.11i [22] [23]. IEEE is developing this wireless LAN standard, which focuses strictly on security and improving upon the protocols oered by the previous 802.11 standards. There are three main areas that the IEEE 802.11i wants to improve on over 802.11b: 1) authentication, 2) key management and 3) data transfer. All of these areas were severely lacking in WEP.
2.9 Frequency Range and Channel Allocation for 802.11b As mentioned above, 802.11b networks operate in the ultra high frequency band, specically between 2.4 and 2.5 GHz. Transmission does not take place at a single frequency. There are a total of fourteen channels for use, and the modulation technique spreads the transmission over multiple channels for eective use of the frequency spectrum. Table 2.2 gives the frequency of each channel [24]. The United States uses channels 1-11, Europe uses channels
16
Table 2.2: Frequency and channel assignments.

CHANNEL 1 2 3 4 5 6 7 FREQUENCY 2.412 GHz 2.417 GHz 2.422 GHz 2.427 GHz 2.432 GHz 2.437 GHz 2.442 GHz CHANNEL 8 9 10 11 12 13 14 FREQUENCY 2.447 GHz 2.452 GHz 2.457 GHz 2.462 GHz 2.467 GHz 2.472 GHz 2.484 GHz
1-13, and Japan uses channels 1-14. Interference is maximum between any two adjacent channels. For instance, data transmission and the eect of interference due to adjacent channels can be compared to conversations happening in 11 rooms. Each room has people having dierent conversations. People in room one can hear the conversations in rooms one, two, three, four, and ve. People in room six can hear the conversations of rooms two through ten, but cannot hear anything from rooms one and eleven. People in room eleven can hear the conversations of rooms seven, eight, nine, and ten. For people to have noise free conversations they should be in rooms one, six, and eleven. In 802.11b, there are only three non-overlapping channels, channels one, six, and eleven as shown in Figure 2.2.
17
Figure 2.2: 802.11b channel overlap.
18
CHAPTER 3
STATISTICAL ANALYSIS 3.1 Normal Plot A normal probability plot is a useful graph for assessing whether data comes from a normal distribution. Many statistical procedures make the assumption that the underlying distribution of the data is normal, so this plot can provide some assurance that the assumption of normality is not being violated. Normal plot is a scatterplot of the percentiles of the data versus the percentiles of a population having the normal distribution. If the data comes from a normal population, the resulting points should fall closely along a straight line. If the observations fall along a relatively straight line (using the so-called fat pencil test), then the data is probably normal [25][26]. To further conrm if the distribution is normal a goodness of t test can be used. These plots are produced by using following steps[26]: 1. The observations are ranked from smallest to largest, x1, x2 , ...xn. 2. The ordered observations xj are plotted against their observed cumulative frequency on a graph with the y-axis appropriately scaled for the hypothesized (Normal) distribution. 3. If the hypothesized distribution adequately describes the data, the plotted points fall approximately along a straight line. If the plotted points deviate signicantly from the straight line, then the hypothesized distribution is not appropriate. 4. In assessing the closeness of the points to a straight line, the fat pencil test is often used. If the points are all covered by the imaginary pencil, then the hypothesized (Normal) distribution is likely to be appropriate. 19
3.2 Standard Normal Distribution A useful transformation in statistics is standardization. Sometimes called converting to Z-scores it has the eect of transforming the original distribution to one in which the mean becomes zero and the standard deviation becomes 1. A Z-score quanties the original score in terms of the number of standard deviations that the score is from the mean of the distribution. (3.1) is used to convert from an original or raw score to a Z-score.
Z =
(Raw Score) (Mean) . (Standard Deviation)
(3.1)
The process of converting or transforming scores on a variable to Z-scores is called standardization. Any distribution can be converted to a standardized distribution. However the symmetry of the original distribution remains unchanged. If the original distribution was skewed to start with, it will still be skewed after the Z-score transformation. In the special case where the original distribution can be considered normal, standardizing will result in what is known as the standard normal distribution. The advantage of this is that tables exist in any statistics textbook for the area under the curve for the standard normal distribution. Table 3.1 contains the area under the standard normal curve from 0 to Z. This can be used to compute the cumulative distribution function values for the standard normal distribution.
3.3 Chi-square Goodness-of-Fit Test When an analyst attempts to t a statistical model to observed data, he or she may wonder how well the model actually reects the data. How close are the observed values to those which would be expected under the tted model? One statistical test that addresses this issue is the chi-square goodness-of-t test. The chi-square test is used to test if a sample of data came from a population with a specic distribution. The chi-square goodness-of-t test is applied to binned data. This is actually not a restriction since for non-binned data you
20
Table 3.1: Area under the normal curve from 0 to X.

X 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.8 1.9 2.0 2.1 2.2 2.3 2.4 2.5 2.6 2.7 2.8 2.9 3.0 0.00 0.00000 0.03983 0.07926 0.11791 0.15542 0.19146 0.22575 0.25804 0.28814 0.31594 0.34134 0.36433 0.38493 0.40320 0.41924 0.43319 0.44520 0.45543 0.46407 0.47128 0.47725 0.48214 0.48610 0.48928 0.49180 0.49379 0.49534 0.49653 0.49744 0.49813 0.49865 0.01 0.00399 0.04380 0.08317 0.12172 0.15910 0.19497 0.22907 0.26115 0.29103 0.31859 0.34375 0.36650 0.38686 0.40490 0.42073 0.43448 0.44630 0.45637 0.46485 0.47193 0.47778 0.48257 0.48645 0.48956 0.49202 0.49396 0.49547 0.49664 0.49752 0.49819 0.49869 0.02 0.00798 0.04776 0.08706 0.12552 0.16276 0.19847 0.23237 0.26424 0.29389 0.32121 0.34614 0.36864 0.38877 0.40658 0.42220 0.43574 0.44738 0.45728 0.46562 0.47257 0.47831 0.48300 0.48679 0.48983 0.49224 0.49413 0.49560 0.49674 0.49760 0.49825 0.49874 0.03 0.01197 0.05172 0.09095 0.12930 0.16640 0.20194 0.23565 0.26730 0.29673 0.32381 0.34849 0.37076 0.39065 0.40824 0.42364 0.43699 0.44845 0.45818 0.46638 0.47320 0.47882 0.48341 0.48713 0.49010 0.49245 0.49430 0.49573 0.49683 0.49767 0.49831 0.49878 0.04 0.01595 0.05567 0.09483 0.13307 0.17003 0.20540 0.23891 0.27035 0.29955 0.32639 0.35083 0.37286 0.39251 0.40988 0.42507 0.43822 0.44950 0.45907 0.46712 0.47381 0.47932 0.48382 0.48745 0.49036 0.49266 0.49446 0.49585 0.49693 0.49774 0.49836 0.49882 0.05 0.01994 0.05962 0.09871 0.13683 0.17364 0.20884 0.24215 0.27337 0.30234 0.32894 0.35314 0.37493 0.39435 0.41149 0.42647 0.43943 0.45053 0.45994 0.46784 0.47441 0.47982 0.48422 0.48778 0.49061 0.49286 0.49461 0.49598 0.49702 0.49781 0.49841 0.49886 0.06 0.02392 0.06356 0.10257 0.14058 0.17724 0.21226 0.24537 0.27637 0.30511 0.33147 0.35543 0.37698 0.39617 0.41308 0.42785 0.44062 0.45154 0.46080 0.46856 0.47500 0.48030 0.48461 0.48809 0.49086 0.49305 0.49477 0.49609 0.49711 0.49788 0.49846 0.49889 0.07 0.02790 0.06749 0.10642 0.14431 0.18082 0.21566 0.24857 0.27935 0.30785 0.33398 0.35769 0.37900 0.39796 0.41466 0.42922 0.44179 0.45254 0.46164 0.46926 0.47558 0.48077 0.48500 0.48840 0.49111 0.49324 0.49492 0.49621 0.49720 0.49795 0.49851 0.49893 0.08 0.03188 0.07142 0.11026 0.14803 0.18439 0.21904 0.25175 0.28230 0.31057 0.33646 0.35993 0.38100 0.39973 0.41621 0.43056 0.44295 0.45352 0.46246 0.46995 0.47615 0.48124 0.48537 0.48870 0.49134 0.49343 0.49506 0.49632 0.49728 0.49801 0.49856 0.49896 0.09 0.03586 0.07535 0.11409 0.15173 0.18793 0.22240 0.25490 0.28524 0.31327 0.33891 0.36214 0.38298 0.40147 0.41774 0.43189 0.44408 0.45449 0.46327 0.47062 0.47670 0.48169 0.48574 0.48899 0.49158 0.49361 0.49520 0.49643 0.49736 0.49807 0.49861 0.49900
21
can simply calculate a histogram or frequency table before generating the chi-square test. However, the value of the chi-square test statistic are dependent on how the data is binned. The chi-square test is dened for the hypothesis: H0 : The data follow a specied distribution. Ha : The data do not follow the specied distribution. Test Statistic: For the chi-square goodness-of-t computation, the data are divided into k bins and the test statistic is dened as in
k
X2 =
i=1
(Oi Ei )2 , Ei
(3.2)
where Oi is the observed frequency for bin i and Ei is the expected frequency for bin i. Put simply if the computed test statistic X 2 is large, then the observed and expected values are not close and the model is a poor t to the data.
3.3.1 Deciding what value of X 2 is critical for accepting or rejecting a hypothesis Before understanding how to reject or accept a hypothesis we need to understand two terms which help in this process. Degrees of Freedom: The number of independent pieces of information that go into the estimate of a parameter is called the degrees of freedom (df ) [27]. In general, the degrees of freedom of an estimate is equal to the number of independent scores that go into the estimate minus the number of parameters estimated as intermediate steps or constraints in the estimation of the parameter itself. For example, if the variance, 2, is to be estimated from a random sample of N independent scores, then the degrees of freedom is equal to the number of independent scores (N ) minus the number of parameters estimated as intermediate steps (one, mean) and is therefore equal to N 1. In chi-square test statistic with k data bins, degrees of freedom can be calculated as df = No. of data bins (k) No. of constraints 22 (3.3)
For chi-square test statistic with k data bins the only constraint is No. of observed data points = No. of expected data points Therefore, df for a chi-square test statistic with k data bins is equal to k 1. Level of Signicance or p-value: The p-value or calculated probability is the estimated probability of rejecting the null hypothesis (H0 ) of a study question when that hypothesis is true [27]. The term signicance level is used to refer to a pre-chosen probability and the term p-value is used to indicate a probability that one calculates after a given study. The smaller the p-value, the more strongly the test conrms the null hypothesis. A pvalue of .05 or less conrms the null hypothesis at the 5% level that is, the statistical assumptions used imply that only 5% of the time would the supposed statistical process produce a nding this extreme if the null hypothesis were false. A p < 0.05 is considered statistically signicant and p < 0.001 is considered statistically highly signicant. (3.4)
3.3.2 Finding p-value for Chi-square Test By using (3.2) and (3.3) we can get values for chi-square test statistic X 2 and degrees of freedom df . The nal step is to refer to a professionally prepared table (Table 3.2) of the probabilities of X 2 values. These tables come in a variety of sizes, depending upon how many subdivisions (columns) are present, and how high the degrees of freedom go. The table lists the degrees of freedom as the headings to the rows. Across the top are probability gures - the probability of the Chi-Square. The interior of the table consists of the sum of the X 2 values themselves. How to use Table 3.2: 1. Find the degrees of freedom from the data and look in the left-hand column of the table.
23
2. Scan across the row of X 2 values beside the df number for two values which bracket the calculated X 2 number. This means that one of the gures will be larger, and the other will be smaller. If the table were subdivided into enough columns, one might have found the exact calculated value on the table. Generally, one has to be satised with nding the bracketing numbers. 3. Look up at the top of the table to see which probabilities correspond to the bracketing X 2 values. If the exact X 2 value is not found on this table, its probability would have fallen somewhere between these two.
3.4 Curve Fitting Field data is often accompanied by noise. Even though all control parameters (independent variables) remain constant, the resultant outcomes (dependent variables) vary. A process of quantitatively estimating the trend of the outcomes, also known as regression or curve tting, therefore becomes necessary. The curve tting process ts equations of approximating curves to the raw eld data. Nevertheless, for a given set of data, the tting curves of a given type are generally not unique. Thus, a curve with a minimal deviation from all data points is desired. This besttting curve can be obtained by the method of least squares [26]. For Log-distance Path Loss Model discussed in section 2.5.1 a linear least squares curve tting is used. The least-squares line uses a straight line equation as shown below
Y = P1 X + P2 ,
(3.5)
to approximate the given set of data, (x1 ,y1), (x2,y2), ..., (xn ,yn ), where n >=2. The best tting curve f (x) has the least square error, i.e.,
n n
=
i=1
[yi f (xi )]2 =

i=1
[yi (P1 xi + P2 )]2 .
(3.6)
24
Table 3.2: Critical values of chi-square distribution with degrees of freedom.

Probability of exceeding the critical value df 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 0.995 0.01 0.072 0.207 0.412 0.676 0.989 1.344 1.735 2.156 2.603 3.074 3.565 4.075 4.601 5.142 5.697 6.265 6.844 7.434 8.034 8.643 9.26 9.886 10.52 11.16 11.808 12.461 13.121 13.787 0.99 0.02 0.115 0.297 0.554 0.872 1.239 1.646 2.088 2.558 3.053 3.571 4.107 4.66 5.229 5.812 6.408 7.015 7.633 8.26 8.897 9.542 10.196 10.856 11.524 12.198 12.879 13.565 14.256 14.953 0.975 0.001 0.051 0.216 0.484 0.831 1.237 1.69 2.18 2.7 3.247 3.816 4.404 5.009 5.629 6.262 6.908 7.564 8.231 8.907 9.591 10.283 10.982 11.689 12.401 13.12 13.844 14.573 15.308 16.047 16.791 0.95 0.004 0.103 0.352 0.711 1.145 1.635 2.167 2.733 3.325 3.94 4.575 5.226 5.892 6.571 7.261 7.962 8.672 9.39 10.117 10.851 11.591 12.338 13.091 13.848 14.611 15.379 16.151 16.928 17.708 18.493 0.9 0.016 0.211 0.584 1.064 1.61 2.204 2.833 3.49 4.168 4.865 5.578 6.304 7.042 7.79 8.547 9.312 10.085 10.865 11.651 12.443 13.24 14.041 14.848 15.659 16.473 17.292 18.114 18.939 19.768 20.599 0.1 2.706 4.605 6.251 7.779 9.236 10.645 12.017 13.362 14.684 15.987 17.275 18.549 19.812 21.064 22.307 23.542 24.769 25.989 27.204 28.412 29.615 30.813 32.007 33.196 34.382 35.563 36.741 37.916 39.087 40.256 0.05 3.841 5.991 7.815 9.488 11.07 12.592 14.067 15.507 16.919 18.307 19.675 21.026 22.362 23.685 24.996 26.296 27.587 28.869 30.144 31.41 32.671 33.924 35.172 36.415 37.652 38.885 40.113 41.337 42.557 43.773 0.025 5.024 7.378 9.348 11.143 12.833 14.449 16.013 17.535 19.023 20.483 21.92 23.337 24.736 26.119 27.488 28.845 30.191 31.526 32.852 34.17 35.479 36.781 38.076 39.364 40.646 41.923 43.195 44.461 45.722 46.979 0.01 6.635 9.21 11.345 13.277 15.086 16.812 18.475 20.09 21.666 23.209 24.725 26.217 27.688 29.141 30.578 32 33.409 34.805 36.191 37.566 38.932 40.289 41.638 42.98 44.314 45.642 46.963 48.278 49.588 50.892 0.005 7.879 10.597 12.838 14.86 16.75 18.548 20.278 21.955 23.589 25.188 26.757 28.3 29.819 31.319 32.801 34.267 35.718 37.156 38.582 39.997 41.401 42.796 44.181 45.559 46.928 48.29 49.645 50.993 52.336 53.672
25
The MATLAB curve tting tool box is used to nd the linear t in the discussions to follow in Chapter 5.
26
CHAPTER 4
EXPERIMENTAL SETUP 4.1 Introduction This chapter outlines the experimental procedure involved, the hardware and software used, the propagation environment, dierent scenarios considered, and the research design used for generating an indoor propagation model. The experimental procedure involves identifying dierent measurement scenarios, measurement of signal levels in each scenario, followed by numerical analysis of the data collected.
4.2 Propagation Environment The power received by a mobile receiver is inuenced by the characteristics of the propagation environment. If a generalization study for indoor propagation is to be made, it is essential that we identify the features that inuence the propagation characteristics in dierent buildings. In this section, a brief description is given of the buildings where the measurements were made. Experiments are conducted in two dierent buildings. In the rst building experiments are conducted for four dierent scenarios. The College of Engineering, University of North Texas, Research Park was chosen for conducting the rst set of experiments. The facility is a two oor building. Figure 4.1 gives an overview of the rst oor, Figure 4.2 gives an overview of the second oor, and Figure 4.3 gives a detailed view of the part of the second oor where the experiments will be conducted. The second set of experiments are conducted in General Academic Building (GAB) in the University of North Texas main campus. Signal measurements are conducted for only two scenarios in this building.
27
Figure 4.1: College of Engineering, rst oor overview. 28
Figure 4.2: College of Engineering, second oor overview. 29
Figure 4.3: College of Engineering, second oor detail.
30
Figure 4.4: Building One Closed Corridor. 4.2.1 Description of Measurement Scenarios for Building One Four dierent scenarios are considered for measurements. The scenarios used will help in developing signal loss equations, by which a generalization for propagation in an indoor environment at 2.4 GHz can be obtained. The scenarios are described as follows: Closed Corridor: A closed corridor on the second oor of building one is used for signal measurements. This corridor is closed on both sides with walls. This corridor is 9 high and 87 wide. Signal measurements are taken at every ve feet interval in the middle of the corridor. Figure 4.4 gives a blue print of the corridor and also shows the dierent locations at which measurements are taken. Open corridor: An open corridor on the second oor of building one is used for signal 31
Figure 4.5: Building One Open Corridor. measurements. The corridor is open on one side and closed with a wall on the other side. This corridor is 125 high and 147 wide. Signal measurements are taken at every ve feet interval in the middle of the corridor. Figure 4.5 gives a blue print of the corridor and also shows the dierent locations at which measurements are taken. Class room: A lecture room with furniture and computers is considered for signal measurements. This room is 40 X 30X 9. Signal measurements are taken at every ve feet interval diagonally across the room. Figure 4.6 gives a blue print of the room and also shows the dierent locations at which measurements are taken. Computer lab: A computer lab with furniture and computers is considered for signal measurements. This room is 45 X 30X 9. Signal measurements are taken at every ve feet interval diagonally across the room. Figure 4.7 gives a blue print of the room 32
Figure 4.6: Building One Classroom.
Figure 4.7: Building One Computer Lab.
33
and also shows the dierent locations at which measurements are taken.
4.2.2 Description of Measurement Scenarios for Building Two Two scenarios in building two that are similar to building one scenarios closed corridor and classroom are considered for measurements. The scenarios used will help in comparing the results from building one. The scenarios are described as follows: Closed Corridor: A closed corridor on the third oor of building two is used for signal measurements. This corridor is closed on both sides with walls. This corridor is 89.25 high and 86 wide. Signal measurements are taken at every ve feet interval in the middle of the corridor. Figure 4.8 gives a blue print of the corridor and also shows the dierent locations at which measurements are taken. Class room: A lecture room with furniture and computers is considered for signal measurements. This room is 30 X 24X 89.25. Signal measurements are taken at every ve feet interval diagonally across the room. Figure 4.9 gives a blue print of the room and also shows the dierent locations at which measurements are taken. In all of the above described scenarios the measurements of the two access points are done alternately. A laptop with the wireless client adapter is moved away from the access point at xed distance intervals and the signal strength at each interval is measured by rotating the laptop along its axis.
4.3 Hardware and Software Description The measurements are done using two separate access points. The access points are from dierent manufacturers. One is LinkSys [28], the other is D-Link [29]. Both access points are IEEE 802.11b compliant. The access points operate at 2.4 GHz and provide a bandwidth
34
Figure 4.8: Building Two, Closed Corridor.
Figure 4.9: Building Two, Classroom.
35
of 11 Mbps. A laptop attached with a wireless client adapter is used to measure the signal strength. The client adapter used is manufactured by LinkSys. The signal measurements were done using the software NetStumbler [30] which is a tool for Windows that allows one to measure the signal level of WLANs using 802.11a, 802.11b or 802.11g.
4.4 Data Acquisition Using NetStumbler measurements were taken for the above described scenarios. In each scenario the signal strength was measured for both (LinkSys and D-Link) access points at regular increments of distance. At each interval signal measurements were taken by rotating the laptop twice along its axis.
4.5 Numerical Analysis Numerical analysis is performed to generate a path loss model on the collected data in each scenario. The data is subjected to following steps of numerical analysis. 1. Use Curve tting to evaluate path loss exponent n, in Log-distance Path Loss Model: A scatter plot for signal levels at dierent distance intervals is drawn with distance ratio d/d0 as X-axis and path loss P L as Y -axis on a loglog scale. As discussed in Section 2.5.3, a linear equation of the form (3.5) is generated using MATLAB curve tting tool. The slope of the linear equation generated is equal to 10n where n is the path loss exponent, which indicates the rate at which path loss increases with distance d. Now that we know n (path loss exponent) for a scenario, we will be able to relate distance d with path loss P L as shown in (2.1). 2. Evaluate Standard deviation , for Log-Normal Distribution:
36
Before we evaluate Standard deviation for Log-Normal Distribution we need to prove that the shadowing eect over the range of distance intervals is a normal distribution. This is achieved using Normal Plot, Standard Normal Distribution, and Chi-square goodness-of-t test. Loss dierential can be dened as dierence between mean path loss at each distance interval compared to the measured data at each interval. As discussed in Section 2.10.1 if the scatter plot of Loss dierential is a straight line we can say the distribution is normal. Once the normality of the distribution is established, mean and standard deviation are calculated. 3. Standard Normal Distribution & Chi-Square Goodness-of-Fit Test: Section 3.2 describes standardization of a distribution by converting raw scores into Z-scores using (3.1). Standardizing Loss dierential will result in a standard normal distribution with zero mean and a standard deviation of one. Since the shape of the distribution is not eected by standardization, Chi-square goodness-of-t test is calculated to further verify numerically, if the distribution is normal.
37
CHAPTER 5
NUMERICAL ANALYSIS 5.1 Introduction This chapter describes the numerical analysis for scenarios in Building One as described in Chapter 4. In order to extract useful information from the raw measurement data, data processing is necessary and includes the following steps (Figure 5.1): 1. Calculate the mean signal levels. 2. Using least squares method to calculate curve tting. 3. Use curve tting to evaluate path loss exponent n, in Log-distance Path Loss Model. 4. Using Normal Plot, verify normal distribution. 5. Evaluate standard deviation , for log-normal distribution, and determine standard normal distribution. 6. Calculate Chi-square Goodness-of-Fit Test. 7. Compare with two-ray model. The close-in reference distance d0 , the path loss exponent n, and the standard deviation , statistically describe the path loss model for a specic T-R separation. Chi-square Goodnessof-Fit Test and standard normal distribution are used to quantitatively evaluate the quality of Normal Distribution.
38
Figure 5.1: Numerical Analysis. 39
Figure 5.2: Building One, closed corridor loss for AP1 (D-Link). 5.2 Analysis of Results
5.2.1 Scenario 1: Building One Closed Corridor Figure 5.2 and Figure 5.3 gives the loss (in dBm) versus T-R (Transmitter-Receiver) separation (in meters) for AP1 (D-Link) and AP2 (LinkSys) respectively. The experiment was conducted in a closed corridor at The College of Engineering, University of North Texas, Research Park (Figure 4.4). As seen from Figure 5.2 and Figure 5.3 loss increases as one goes further away from the access point. The laptop used to measure the signal strength is moved at dierent distance intervals and at each distance interval the laptop is rotated twice along its axis. Due to this rotation during measurements, the observed signal strength and loss variation is more than when the laptop is not rotated. 1. Calculate mean signal levels. Mean values of loss at each distance interval is calculated. Figure 5.4 shows a plot between mean loss for both the access points (D-Link and LinkSys) versus T-R separation. As seen from the gure, signal strength deteriorates with increasing T-R
40
Figure 5.3: Building One, closed corridor loss for AP2 (LinkSys). separation. In the remaining part of the numerical analysis an equation is determined which describes the relationship between loss and T-R separation in a closed corridor. 2. Using least squares method to calculate curve tting. MATLAB curve tting tool which is a graphical user interface tool is used to visually explore data and ts as scatter plots. A linear t is calculated using least squares method. Figures 5.5 and 5.6 show the linear curve tting for the data collected using D-Link (AP1) and LinkSys (AP2) access points. Listed below are the output results from curve tting. Closed corridor curve tting for AP1 (D-Link) Linear model Poly1: f(x) = p1 * x + p2 Coecients (with 95 percent condence bounds): p1 = 15.72 (15.68, 15.77) p2 = 0 (xed at bound) 41
Figure 5.4: Building One, closed corridor mean loss for AP1 and AP2. Closed corridor curve tting for AP2 (LinkSys) Linear model Poly1: f(x) = p1 * x + p2 Coecients (with 95 percent condence bounds): p1 = 15.8 (15.76, 15.85) p2 = 0 (xed at bound) 3. Use curve tting to evaluate path loss exponent n, in Log-distance Path Loss Model. Curve tting gives a linear equation of the form (3.5) which when compared to (2.2) helps in determining the value for path loss exponent n. From Figures 5.5 and 5.6 and output results listed above we get the slope of the curve to be 15.72 for D-Link and 15.8 for LinkSys. From (2.2), the slope of the tted curve is 10n, i.e., calculated path loss exponents (n) for D-link and LinkSys access points in closed corridor is 1.572 and 1.58. The next step is to evaluate Log normal shadowing. Before evaluating Log normal shadowing, normal probability plot is used to verify if the distribution is normal. 42
Figure 5.5: Building One, closed corridor curve tting for AP1 (D-Link).
Figure 5.6: Building One, closed corridor curve tting for AP2 (LinkSys).
43
Figure 5.7: Building One, closed corridor normal probability plot for AP1 (D-Link). 4. Using Normal Plot, verify normal distribution. In this part of the analysis we calculate the loss dierential, which is a measure of variation or scatter of loss compared to the mean value of loss calculated above. Normal probability plot is used to verify if the variation of loss (loss dierential) is distributed normally. The normal plots from Figure 5.7 and Figure 5.8 indicate that the assumption of normality is not unreasonable. Both plots pass the fat pencil test. 5. Evaluate standard deviation , for log-normal distribution, and determine standard normal distribution. Mean , and standard deviation of loss dierential is calculated for both the access points. Mean is calculated to be 0 and standard deviation for D-Link access point is 3.9849 and for LinkSys standard deviation is 4.022. Using (3.1) loss dierential is standardized. Figure 5.9 and Figure 5.10 show the plots for standard normal distributions for AP1 and AP2. 6. Calculate Chi-square Goodness-of-Fit Test. As discussed in section 3.3 to evaluate goodness of t we need to calculate test statistic 44
Figure 5.8: Building One, closed corridor normal probability plot for AP2 (LinkSys).
Figure 5.9: Building One, closed corridor standard normal distribution for AP1 (D-Link).
45
Table 5.1: Closed corridor X 2 table for AP1 (D-Link).
Bins -5.25 to -4.75 -4.75 to -4.25 -4.25 to -3.75 -3.75 to -3.25 -3.25 to -2.75 -2.75 to -2.25 -2.25 to -1.75 -1.75 to -1.25 -1.25 to -0.75 -0.75 to -0.25 -0.25 to 0.25 0.25 to 0.75 0.75 to 1.25 1.25 to 1.75 1.75 to 2.25 2.25 to 2.75 2.75 to 3.25 3.25 to 3.75 3.75 to 4.25 4.25 to 4.75 4.75 to 5.25 Total Observations
Area under standard normal curve 1.00E-06 1.00E-05 7.80E-05 0.000489 0.002403 0.009245 0.027835 0.065591 0.12098 0.17467 0.19741 0.17467 0.12098 0.065591 0.027835 0.009245 0.002403 0.000489 7.80E-05 1.00E-05 1.00E-06 -
Expected = Total Obs * Area under STD curve 0.010009 0.10009 0.7807 4.8944 24.052 92.533 278.6 656.5 1210.9 1748.2 1975.9 1748.2 1210.9 656.5 278.6 92.533 24.052 4.8944 0.7807 0.10009 0.010009 10009.0404
Observed 0 0 1 6 23 91 277 613 1147 1692 2019 1854 1215 654 277 100 25 13 2 0 0 10009
Chi Statistic 0.010009 0.10009 0.0616 0.24974 0.045981 0.025404 0.0091947 2.8824 3.3688 1.8087 0.93984 6.399 0.014095 9.52E-03 0.0091947 0.60252 0.037395 13.424 1.9043 0.10009 0.010009 -
46
Table 5.2: Closed corridor X 2 table for AP2 (LinkSys).
Observed 1 0 1 8 22 92 260 611 1139 1710 2043 1859 1217 637 268 94 27 13 5 1 1 10009
Chi Statistic 97.92 0.10009 0.0616 1.9706 0.17501 0.0030725 1.2418 3.1535 4.2656 0.83609 2.2782 7.0183 0.031045 0.57923 0.40334 0.023251 0.36143 13.424 22.803 8.0911 97.92 -
47
Figure 5.10: Building One, closed corridor standard normal distribution for AP2 (LinkSys). X 2 and degrees of freedom df . Table 5.1 and 5.2 are used for calculating Chi-square test statistic X 2 for AP1 and AP2. To nd the expected value for a particular bin, the area under standard normal curve for the bin is multiplied by total observations. Chi-square statistic is calculated using (3.2). To evaluate if the distribution is normal the bins that are used to calculate X 2 are from -3.25 to 3.25, a total of 13 bins. As discussed in section 3.3, the degrees of freedom df = 13 - 1 = 12. From Table 5.1 X 2 for AP1 (D-Link)= 16.152, and from Table 5.2 X 2 for AP2 (LinkSys)= 20.3699 By using Table 3.2 we can identify to p-value associated with X 2 and df values. We now have our Chi-square statistic X 2 for AP1 and AP2 (16.152 and 20.3699), our predetermined level of signicance (0.05), and our degrees of freedom (df =12). Finding the Chi-square distribution table with 12 degree of freedom and reading along the row we nd our value of X 2 for AP1 = 16.152, lies between 6.304 and 18.549. The corresponding probability is 0.9 < p < 0.1. This is below the conventionally accepted signicance level of 0.05 or 5%, so the null hypothesis that loss dierential is normal is veried for AP1 (D-Link). 48
Figure 5.11: Comparison of closed corridor mean loss with two-ray model. Finding the Chi-square distribution with 12 degree of freedom and reading along the row we nd our value of X 2 for AP2 = 20.3699, lies between 18.549 and 21.026. The corresponding probability is 0.10 < p < 0.05. This is below the conventionally accepted signicance level of 0.05 or 5%, so the null hypothesis that loss dierential is normal is veried for AP2 (LinkSys). 7. Compare with two-ray model. A graphical comparison is made between measured data and the two-ray model. Figure 5.11 and Figure 5.12 show the comparison between the theoretical two-ray model and measured data. It can be seen from these plots that signal deteriorates at a faster rate in two-ray model than in actual measurements.
49
Figure 5.12: Comparison of closed corridor AP1 and AP2 loss with mean signal and two-ray model. 5.2.2 Scenario 2: Building One Open Corridor Figure 5.13 and Figure 5.14 gives the loss (in dBm) versus T-R (Transmitter-Receiver) separation (in meters) for AP1 (D-Link) and AP2 (LinkSys) respectively. The experiment was conducted in an open corridor at The College of Engineering, University of North Texas, Research Park (Figure 4.5). As seen from Figure 5.13 and Figure 5.14 loss increases with distance relative to the access point. 1. Calculate mean signal levels. Mean values of loss at each distance interval is calculated. Figure 5.15 shows a plot between mean loss for both the access points (D-Link and LinkSys) versus T-R separation. As seen from the gure, signal strength deteriorates with increasing T-R separation. In the remaining part of the numerical analysis an equation is determined which describes the relationship between loss and T-R separation in an open corridor. 2. Using least squares method to calculate curve tting. Using MATLAB curve tting tool, a linear t is calculated using least squares method. 50
Figure 5.13: Building One, open corridor loss for AP1 (D-Link).
Figure 5.14: Building One, open corridor loss for AP2 (LinkSys).
51
Figure 5.15: Building One, open corridor mean loss for AP1 and AP2. Figures 5.16 and 5.17 show the linear curve tting for the data collected using D-Link (AP1) and LinkSys (AP2) access points. Listed below are the output results from curve tting. Open corridor curve tting for AP1 (D-Link). Linear model Poly1: f(x) = p1 * x + p2 Coecients (with 95 percent condence bounds): p1 = 16.88 (16.84, 16.92) p2 = 0 (xed at bound) Open corridor curve tting for AP2 (LinkSys). Linear model Poly1: f(x) = p1 * x + p2 Coecients (with 95 percent condence bounds):
52
Figure 5.16: Building One, open corridor curve tting for AP1 (D-Link). p1 = 16.3 (16.26, 16.34) p2 = 0 (xed at bound) 3. Use curve tting to evaluate path loss exponent n, in Log-distance Path Loss Model. From Figures 5.16 and 5.17 and output results listed above we get the slope of the tted curve to be 16.88 for D-Link and 16.3 for LinkSys. From (2.2), the slope of the tted curve is 10n, i.e., calculated path loss exponents (n) for D-link and LinkSys access points in open corridor is 1.688 and 1.63. 4. Using Normal Plot, verify normal distribution. As discussed in Scenario 1 (Closed corridor), normal probability plot is used to verify if the variation of loss (loss dierential) is distributed normally. The normal plots from Figure 5.18 and Figure 5.19 indicate that the assumption of normality is not unreasonable. Both plots pass the fat pencil test. 5. Evaluate standard deviation , for log-normal distribution, and determine standard normal distribution.
53
Figure 5.17: Building One, open corridor curve tting for AP2 (LinkSys).
Figure 5.18: Building One, open corridor normal probability plot for AP1 (D-Link).
54
Figure 5.19: Building One, open corridor normal probability plot for AP2 (LinkSys). Mean , and standard deviation of loss dierential is calculated for both the access points. Mean is calculated to be 0 and standard deviation for D-Link access point is 3.5773 and for LinkSys standard deviation is 3.2642. Using (3.1) loss dierential is standardized. Figure 5.20 and Figure 5.21 show the plots for standard normal distributions for AP1 and AP2. 6. Calculate Chi-square Goodness-of-Fit Test As discussed in section 3.3 to evaluate goodness of t we need to calculate test statistic X 2 and degrees of freedom df . Table 5.3 and 5.4 are used for calculating Chi-square test statistic X 2 for AP1 and AP2. To nd the expected value for a particular bin, the area under standard normal curve for the bin is multiplied by total observations. Chi-square test statistic is calculated using (3.2). To evaluate if the distribution is normal, the bins that are used to calculate X 2 are from -3.25 to 3.25, a total of 13 bins. As discussed in section 3.3, degrees of freedom df = 13 - 1 = 12. From Table 5.3 X 2 for AP1 (D-Link)= 20.0012, and from Table 5.4 X 2 for AP2 (LinkSys)= 37.6687 By using Table 3.2 we can identify to p-value associated with
55
Figure 5.20: Building One, open corridor standard normal distribution for AP1 (D-Link).
Figure 5.21: Building One, open corridor standard normal distribution for AP2 (LinkSys).
56
Table 5.3: Open corridor X 2 table for AP1 (D-Link).
Expected = Total Obs * Area under STD curve 0.009133 0.09133 0.71237 4.466 21.947 84.435 254.22 599.04 1104.9 1595.2 1803 1595.2 1104.9 599.04 254.22 84.435 21.947 4.466 0.71237 0.09133 0.009133 9133.041666
Observed 0 0 1 12 26 81 233 582 1023 1538 1884 1683 1098 599 250 88 22 12 1 0 0 9133
Chi Statistic 0.009133 0.09133 0.11613 12.709 0.74864 0.13971 1.7708 0.48486 6.0697 2.0528 3.6414 4.8297 0.042991 3.03E-06 0.069954 0.15056 0.00012994 12.709 0.11613 0.09133 0.009133 -
57
Table 5.4: Open corridor X 2 table for AP2 (LinkSys).
Observed 0 1 1 6 21 76 260 556 1061 1544 1821 1759 1020 617 263 89 33 5 0 0 0 9133
Chi Statistic 0.009133 9.0406 0.11613 0.52687 0.040829 0.84257 0.13155 3.0927 1.7436 1.6449 0.18024 16.814 6.5225 0.53831 0.30344 0.24685 5.567 0.063841 0.71237 0.09133 0.009133 -
58
X 2 and df values. We now have our chi square statistic X 2 for AP1 and AP2 (20.0012 and 37.6687), our predetermined level of signicance (0.05), and our degrees of freedom (df = 12). Finding the Chi-square distribution with 12 degree of freedom and reading along the row we nd our value of X 2 for AP1 = 20.0012, lies between 18.549 and 21.026. The corresponding probability is 0.10 < p < 0.05. This is below the conventionally accepted signicance level of 0.05 or 5%, so the null hypothesis that loss dierential is normal is veried for AP1 (D-Link). Finding the Chi-square distribution with 12 degree of freedom and reading along the row we nd our value of X 2 for AP2 = 37.6687, lies beyond 28.3. So the corresponding probability p-value is less that 0.005. This is not an acceptable signicance level. 7. Compare with two-ray model. A graphical comparison is made between measured data and two-ray model. Figure 5.22 and Figure 5.23 show the comparison between the theoretical two-ray model and measured data. It can be seen from these plots that signal strength variation is similar in two-ray model and open corridor measurements.
5.2.3 Scenario 3: Building One Classroom Figure 5.24 and Figure 5.25 gives the loss (in dBm) versus T-R (Transmitter-Receiver) separation (in meters) for AP1 (D-Link) and AP2 (LinkSys) respectively. The experiment was conducted in an classroom at The College of Engineering, University of North Texas, Research Park (Figure 4.6). As seen from Figure 5.24 and Figure 5.25 loss increases as one goes further away from the access point. 1. Calculate mean signal levels.
59
Figure 5.22: Comparison of open corridor mean loss with two-ray model.
Figure 5.23: Comparison of open corridor AP1 and AP2 loss with mean signal and two-ray model.
60
Figure 5.24: Building One, classroom loss for AP1 (D-Link).
Figure 5.25: Building One, classroom loss for AP2 (LinkSys).
61
Figure 5.26: Building One, classroom mean loss for AP1 and AP2. Mean values of loss at each distance interval is calculated. Figure 5.26 shows a plot between mean loss for both the access points (D-Link and LinkSys) versus T-R separation. As seen from the gure, signal strength deteriorates with increasing T-R separation. In the remaining part of the numerical analysis an equation is determined which describes the relationship between loss and T-R separation in a classroom. 2. Using least squares method to calculate curve tting. Figures 5.27 and 5.28 show the linear curve tting for the data collected using D-Link (AP1) and LinkSys (AP2) access points. Listed below are the output results from curve tting. Classroom curve tting for AP1 (D-Link). Linear model Poly1: f(x) = p1 * x + p2 Coecients (with 95 percent condence bounds): p1 = 12.58 (12.44, 12.72) 62
Figure 5.27: Building One, classroom curve tting for AP1 (D-Link). p2 = 0 (xed at bound) Classroom curve tting for AP2 (LinkSys). Linear model Poly1: f(x) = p1 * x + p2 Coecients (with 95 percent condence bounds): p1 = 12.63 (12.49, 12.78) p2 = 0 (xed at bound) 3. Use curve tting to evaluate path loss exponent n, in Log-distance Path Loss Model. From Figures 5.27 and 5.28 and output results listed above we get the slope of the tted curve to be 12.58 for D-Link and 12.63 for LinkSys. From (2.2), the slope of the tted curve is 10n, i.e., calculated path loss exponents (n) for D-link and LinkSys access points in a classroom is 1.258 and 1.263. 4. Using Normal Plot, verify normal distribution.
63
Figure 5.28: Building One, classroom curve tting for AP2 (LinkSys). As discussed in Scenario 1 (Closed corridor), normal probability plot is used to verify if the variation of loss (loss dierential) is distributed normally. The normal plots from Figure 5.29 and Figure 5.30 indicate that the assumption of normality is not unreasonable. Both plots pass the fat pencil test. 5. Evaluate standard deviation , for log-normal distribution, and determine standard normal distribution. Mean , and standard deviation of loss dierential is calculated for both the access points. Mean is calculated to be 0 and standard deviation for D-Link access point is 3.7607 and for LinkSys standard deviation is 4.053. Using (3.1) loss dierential is standardized. Figure 5.31 and Figure 5.32 show the plots for standard normal distributions for AP1 and AP2. 6. Calculate Chi-square Goodness-of-Fit Test. Table 5.5 and 5.6 are used for calculating Chi-square test statistic X 2 for AP1 and AP2. To nd the expected value for a particular bin, the area under standard normal curve for the bin is multiplied by total observations. Chi-square test statistic is calculated 64
Figure 5.29: Building One, classroom normal probability plot for AP1 (D-Link).
Figure 5.30: Building One, classroom normal probability plot for AP2 (LinkSys).
65
Figure 5.31: Building One, classroom standard normal distribution for AP1 (D-Link).
Figure 5.32: Building One, classroom standard normal distribution for AP2 (LinkSys).
66
Table 5.5: Classroom X 2 table for AP1 (D-Link).
Observed 0 2 2 6 15 50 142 320 590 926 1103 987 605 324 137 44 16 2 3 2 0 5276
Chi Statistic 0.005276 71.868 6.1314 4.5336 0.42519 0.030684 0.16067 1.9622 3.6519 0.021606 3.6253 4.6502 1.7352 1.41E+00 0.66166 0.46777 0.87032 0.13037 16.281 71.868 0.005276 -
67
Table 5.6: Classroom X 2 table for AP2 (LinkSys).
Observed 0 0 0 1 15 46 149 333 631 916 1064 981 580 351 125 58 20 6 0 0 0 5276
Chi Statistic 0.005276 0.05276 0.41153 0.96757 0.42519 0.15806 0.031258 0.49273 0.083032 0.033279 0.48385 3.8368 5.3214 0.070573 3.2531 1.7441 4.2284 4.5336 0.41153 0.05276 0.005276 -
68
using (3.2). To evaluate if the distribution is normal, the bins that are used to calculate X 2 are from -3.25 to 3.25, a total of 13 bins. As discussed in section 3.3, degrees of freedom df = 13 - 1 = 12. From Table 5.5 X 2 for AP1 (D-Link)= 19.6687, and from Table 5.6 X 2 for AP2 (LinkSys)= 20.1618 By using Table 3.2 we can identify to p-value associated with X 2 and df values. We now have our chi-square test statistic X 2 for AP1 and AP2 (19.6687 and 20.1618), our predetermined level of signicance (0.05), and our degrees of freedom (df =12). Finding the Chi-square distribution with 12 degree of freedom and reading along the row we nd our value of X 2 for AP1 = 19.6687, lies between 18.549 and 21.026. The corresponding probability is 0.10 < p < 0.05. This is below the conventionally accepted signicance level of 0.05 or 5%, so the null hypothesis that loss dierential is normal is veried for AP1 (D-Link). Finding the Chi square distribution table with 12 degree of freedom and reading along the row we nd our value of X 2 for AP2 = 20.1618, lies between 18.549 and 21.026. The corresponding probability is 0.10 < p < 0.05. This is below the conventionally accepted signicance level of 0.05 or 5%, so the null hypothesis that loss dierential is normal is veried for AP2 (LinkSys). 7. Compare with two-ray model. A graphical comparison is made between measured data and two-ray model. Figure 5.33 and Figure 5.34 show the comparison between the theoretical two-ray model and measured data. It can be seen from these plots that signal strength deteriorates much higher in two-ray model when compared to classroom measurements.
69
Figure 5.33: Comparison of classroom mean loss with two ray-model.
Figure 5.34: Comparison of classroom AP1 and AP2 loss with mean signal and two-ray model.
70
Figure 5.35: Building One, computer lab loss for AP1 (D-Link). 5.2.4 Scenario 4: Building One Computer Lab Figure 5.35 and Figure 5.36 gives the loss (in dBm) versus T-R (Transmitter-Receiver) separation (in meters) for AP1 (D-Link) and AP2 (LinkSys) respectively. The experiment was conducted in an computer lab at The College of Engineering, University of North Texas, Research Park (Figure 4.7). As seen from Figure 5.35 and Figure 5.36 loss increases as one goes further away from the access point. 1. Calculate mean signal levels. Mean values of loss at each distance interval is calculated. Figure 5.37 shows a plot between mean loss for both the access points (D-Link and LinkSys) versus T-R separation. As seen from the gure, signal strength deteriorates with increasing T-R separation. In the remaining part of the numerical analysis an equation is determined which describes the relationship between loss and T-R separation in a computer lab. 2. Using least squares method to calculate curve tting. Using MATLAB curve tting tool, a linear t is calculated using least squares method. Figures 5.38 and 5.39 show the linear curve tting for the data collected using D-Link 71
Figure 5.36: Building One, computer lab loss for AP2 (LinkSys).
Figure 5.37: Building One, computer lab mean loss for AP1 and AP2.
72
(AP1) and LinkSys (AP2) access points. Listed below are the output results from curve tting. Computer lab curve tting for AP1(D-Link) Linear model Poly1: f(x) = p1 * x + p2 Coecients (with 95 percent condence bounds): p1 = 14.47 (14.29, 14.65) p2 = 0 (xed at bound) Computer lab curve tting for AP2(LinkSys) Linear model Poly1: f(x) = p1 * x + p2 Coecients (with 95 percent condence bounds): p1 = 14.28 (14.09, 14.47) p2 = 0 (xed at bound) 3. Use curve tting to evaluate path loss exponent n, in Log-distance Path Loss Model. From Figures 5.38 and 5.39 and output results listed above we get the slope of the tted curve to be 14.47 for D-Link and 14.28 for LinkSys. From (2.2), the slope of the tted curve is 10n, i.e. calculated path loss exponents (n) for D-link and LinkSys access points in a computer lab is 1.447 and 1.428. 4. Using Normal Plot, verify normal distribution. As discussed in Scenario 1 (Closed corridor), normal probability plot is used to verify if the variation of loss (loss dierential) is distributed normally. The normal plots 73
Figure 5.38: Building One, computer lab curve tting for AP1 (D-Link).
Figure 5.39: Building One, computer lab curve tting for AP2 (LinkSys).
74
Figure 5.40: Building One, computer lab normal probability plot for AP1 (D-Link). from Figure 5.40 and Figure 5.41 indicate that the assumption of normality is not unreasonable. Both plots pass the fat pencil test. 5. Evaluate standard deviation , for log-normal distribution, and determine standard normal distribution. Mean , and standard deviation of loss dierential is calculated for both the access points. Mean is calculated to be 0 and standard deviation for D-Link access point is 3.7049 and for LinkSys standard deviation is 3.8460. Using (3.1) loss dierential is standardized. Figure 5.42 and Figure 5.43 show the plots for standard normal distributions for AP1 and AP2. 6. Calculate Chi-square Goodness-of-Fit Test. As discussed in section 3.3 to evaluate goodness of t we need to calculate test statistic X 2 and degrees of freedom df . Table 5.7 and 5.8 are used for calculating Chi-square test statistic X 2 for AP1 and AP2. To nd the expected value for a particular bin, the area under standard normal curve for the bin is multiplied by total observations. Chi-square test statistic is calculated using (3.2). To evaluate if the distribution is 75
Figure 5.41: Building One, computer lab normal probability plot for AP2 (LinkSys).
Figure 5.42: Building One, computer lab standard normal distribution for AP1 (D-Link).
76
Table 5.7: Computer lab X 2 table for AP1 (D-Link).
Observed 0 0 0 1 10 19 74 165 308 452 558 496 293 175 63 32 5 4 2 0 0 2657
Chi Statistic 0.002657 0.02657 0.20725 0.068934 2.047 1.2603 2.43E-05 0.49365 0.56183 0.31483 2.1362 2.1944 2.516 3.01E-03 1.6235 2.251 0.30034 5.6139 15.508 0.02657 0.002657 -
77
Table 5.8: Computer lab X 2 table for AP2 (LinkSys).
Observed 0 0 0 0 6 18 83 167 309 468 537 496 287 170 69 33 12 1 0 0 1 2657
Chi Statistic 0.002657 0.02657 0.20725 1.2993 0.023188 1.754 1.1056 0.30371 0.48133 0.032983 0.29663 2.1944 3.6897 0.10488 0.33232 2.8972 4.9384 0.068934 0.20725 0.02657 374.37 -
78
Figure 5.43: Building One, computer lab standard normal distribution for AP2 (LinkSys). normal, the bins that are used to calculate X 2 are from -3.25 to 3.25, a total of 13 bins. As discussed in section 3.3, degrees of freedom df = 13 - 1 = 12. From Table 5.7 X 2 for AP1 (D-Link)= 15.7022, and from Table 5.8 X 2 for AP2 (LinkSys)= 18.1544 By using Table 3.2 we can identify to p-value associated with X 2 and df values. We now have our chi-square test statistic X 2 for AP1 and AP2 (15.7022 and 18.1544), our predetermined level of signicance (0.05), and our degrees of freedom (df =12). Finding the Chi-square distribution with 12 degree of freedom and reading along the row we nd our value of X 2 for AP1 = 15.7022, lies between 6.304 and 18.549. The corresponding probability is 0.9 < p < 0.1. This is below the conventionally accepted signicance level of 0.05 or 5%, so the null hypothesis that loss dierential is normal is veried for AP1 (D-Link). Finding the Chi-square distribution with 12 degree of freedom and reading along the row we nd our value of X 2 for AP2 = 18.1544, lies between 6.304 and 18.549. The corresponding probability is 0.9 < p < 0.1. This is below the conventionally accepted
79
Figure 5.44: Comparison of computer lab mean loss with two ray model. signicance level of 0.05 or 5%, so the null hypothesis that loss dierential is normal is veried for AP2 (LinkSys). 7. Compare with two-ray model. A graphical comparison is made between measured data and two-ray model. Figure 5.44 and Figure 5.45 show the comparison between the theoretical two-ray model and measured data. It can be seen from these plots that signal strength deteriorates much higher in two-ray model when compared to computer lab measurements.
5.3 Summary of Results In this chapter a numerical analysis is performed on measurements conducted in Building One. Figure 5.46 lists the experimental results for all the scenarios from Building One. For Building One Closed Corridor, Path loss exponent (n) is 1.572 for AP1 and 1.58 for AP2, Standard deviation () is 3.9849 for AP1 and 4.022 for AP2. Now the Chi-square test statistic (X 2 ) values for both AP1 and AP2 (16.152 for AP1 and 20.3699 for AP2) are acceptable values to prove that the observed fading is a normal distribution at 5% signicance
80
Figure 5.45: Comparison of computer lab AP1 and AP2 loss with mean signal and two ray model. level. So, for Building One Closed Corridor Alternative Hypotheses is accepted for the rst research question. For Building One Open Corridor, Path loss exponent (n) is 1.688 for AP1 and 1.63 for AP2, Standard deviation () is 3.5773 for AP1 and 3.2642 for AP2. Now the Chi-square test statistic (X 2 ) value for AP1 (20.0012) is acceptable value to prove that the observed fading is a normal distribution at 5% signicance level but for AP2 Chi-square test statistic (X 2 ) value of 37.6687 is not an acceptable value to prove that the observed fading is a normal distribution at 5% signicance level. This is an exception in this study. So, for Building One Open Corridor Alternative Hypotheses is accepted for the rst research question. For Building One Classroom, Path loss exponent (n) is 1.258 for AP1 and 1.263 for AP2, Standard deviation () is 3.7607 for AP1 and 4.053 for AP2. Now the Chi-square test statistic (X 2 ) values for both AP1 and AP2 (19.6687 for AP1 and 20.1618 for AP2) are acceptable values to prove that the observed fading is a normal distribution at 5% signicance level. So, for Building One Classroom Alternative Hypotheses is accepted for the rst research question.
81
Figure 5.46: Results at a glance. For Building One Computer lab, Path loss exponent (n) is 1.447 for AP1 and 1.428 for AP2, Standard deviation () is 3.7049 for AP1 and 3.846 for AP2. Now the Chi-square test statistic (X 2 ) values for both AP1 and AP2 (15.7022 for AP1 and 18.1544 for AP2) are acceptable values to prove that the observed fading is a normal distribution at 5% signicance level. So, for Building One Computer lab Alternative Hypotheses is accepted for the rst research question. Varying values for Path loss exponent (n) and Standard deviation () for dierent scenarios signies that the path loss and multi path fading vary from scenario to scenario. So, for Building One Alternative Hypotheses is accepted for the second research question.
82
CHAPTER 6
COMPARISON WITH SIMILAR SCENARIOS 6.1 Introduction This chapter describes the numerical analysis for scenarios in Building Two as described in Chapter 4 and compares the results with similar scenarios in Building One.
6.2 Scenario 1: Building Two Closed Corridor Figure 6.1 and Figure 6.2 give the loss (in dBm) versus T-R (Transmitter-Receiver) separation (in meters) for AP1 (D-Link) and AP2 (LinkSys) respectively. The experiment was conducted in a closed corridor at General Academic Building (GAB) in the University of North Texas main campus. As seen from Figure 6.1 and Figure 6.2 loss increases as one goes further away from the access point. 1. Calculate mean signal levels. Mean values of loss at each distance interval is calculated. Figure 6.3 shows a plot between mean loss for both the access points (D-Link and LinkSys) versus T-R separation. As seen from the gure, signal strength deteriorates with increasing T-R separation. In the remaining part of the numerical analysis an equation is determined which describes the relationship between loss and T-R separation in a closed corridor for Building Two. 2. Using least squares method to calculate curve tting. Using MATLAB curve tting tool a linear t is calculated using least squares method. Figures 6.4 and 6.5 show the linear curve tting for the data collected using D-Link
83
Figure 6.1: Building Two, closed corridor loss for AP1 (D-Link).
Figure 6.2: Building Two, closed corridor loss for AP2 (LinkSys).
84
Figure 6.3: Building Two, closed corridor mean loss for AP1 and AP2. (AP1) and LinkSys (AP2) access points. Listed below are the output results from curve tting. Closed corridor curve tting for AP1 (D-Link). Linear model Poly1: f(x) = p1 * x + p2 Coecients (with 95 percent condence bounds): p1 = 16.66 (16.56, 16.76) p2 = 0 (xed at bound) Closed corridor curve tting for AP2 (LinkSys). Linear model Poly1: f(x) = p1 * x + p2 Coecients (with 95 percent condence bounds): p1 = 16.48 (16.38, 16.58) 85
Figure 6.4: Building Two, closed corridor curve tting for AP1 (D-Link). p2 = 0 (xed at bound) 3. Use curve tting to evaluate path loss exponent n, in Log-distance Path Loss Model. From Figures 6.4 and 6.5 and output results listed above we get the slope of the tted curve to be 16.66 for D-Link and 16.48 for LinkSys. From (2.2), the slope of the tted curve is 10n, i.e. calculated path loss exponents (n) for D-link and LinkSys access points in a closed corridor for Building Two is 1.666 and 1.648. From Section 5.2.1 the calculated path loss exponents for D-Link and LinkSys access points in a closed corridor for Building One are 1.572 and 1.58. It can be said that signal strength deteriorates at a faster rate in Building Two. Even though the width of closed corridors in both Building One and Building Two are almost the same, the dierence in path loss exponents can be attributed to dierences in material used for construction and variation of multi-path caused by height of corridors. 4. Using Normal Plot, verify normal distribution. Normal probability plot is used to verify if the variation of loss (loss dierential) is distributed normally. The normal plots from Figure 6.6 and Figure 6.7 indicate that 86
Figure 6.5: Building Two, closed corridor curve tting for AP2 (LinkSys). the assumption of normality is not unreasonable. Both plots pass the fat pencil test. 5. Evaluate standard deviation , for log-normal distribution, and determine standard normal distribution. Mean , and standard deviation of loss dierential is calculated for both the access points. Mean is calculated to be 0 and standard deviation for D-Link access point is 3.9275 and for LinkSys standard deviation is 3.7699. Using (3.1) loss dierential is standardized. Figure 6.8 and Figure 6.9 show the plots for standard normal distributions for AP1 and AP2. 6. Calculate Chi-square Goodness-of-Fit Test. As discussed in section 3.3 to evaluate goodness of t we need to calculate test statistic X 2 and degrees of freedom df . Table 6.1 and 6.2 are used for calculating Chi-square test statistic X 2 for AP1 and AP2. To evaluate if the distribution is normal, the bins that are used to calculate X 2 are from -3.25 to 3.25, a total of 13 bins. As discussed in section 3.3, degrees of freedom df = 13 - 1 = 12. From Table 6.1 X 2 for AP1 (D-Link)= 11.7578, and from Table 6.2 X 2 for AP2 87
Figure 6.6: Building Two, closed corridor normal probability plot for AP1 (D-Link).
Figure 6.7: Building Two, closed corridor normal probability plot for AP2 (LinkSys).
88
Figure 6.8: Building Two, closed corridor standard normal distribution for AP1 (D-Link).
Figure 6.9: Building Two, closed corridor standard normal distribution for AP2 (LinkSys).
89
Observed 0 0 0 3 14 49 140 309 582 908 1069 918 618 331 138 59 13 4 1 0 0 5156
Chi Statistic 0.005156 0.05156 0.40217 0.090894 0.20925 0.037265 0.0862 2.519 2.7961 0.061169 2.5693 0.33704 0.053237 1.53E-01 0.2121 2.6943 0.030046 0.86726 0.88869 0.05156 0.005156 -
90
Observed 0 0 0 4 16 48 148 302 577 899 1069 927 609 355 136 50 12 4 0 0 0 5156
Chi Statistic 0.005156 0.05156 0.40217 0.86726 1.0519 0.0023232 0.14002 3.8722 3.5057 0.0027646 2.5693 0.7752 0.34939 0.83584 0.39374 0.11416 0.012268 0.86726 0.40217 0.05156 0.005156 -
91
Table 6.3: Closed corridor results from Building One and Building Two.
Building One AP1 (D-Link). Path loss exponent (n) AP2 (LinkSys). Path loss exponent (n) AP1 (D-Link). Mean () AP2 (LinkSys). Mean () AP1 (D-Link). SD () AP2 (LinkSys). SD () 1.572 1.58 0 0 3.9849 4.022 Building Two 1.666 1.648 0 0 3.9275 3.7699
(LinkSys)= 13.6248. By using Table 3.2 we can identify to p-value associated with X 2 and df values. We now have our chi-square test statistic X 2 for AP1 and AP2 (11.7578 and 13.6248), our predetermined level of signicance (0.05), and our degrees of freedom (df =12). From Table 3.2 we can say X 2 for AP1 = 11.7578, lies between 6.304 and 18.549. The corresponding probability is 0.9 < p < 0.1. This is below the conventionally accepted signicance level of 0.05 or 5%, so the null hypothesis that loss dierential is normal is veried for AP1 (D-Link). Now for AP2 our value of X 2 = 13.6248, lies between 6.304 and 18.549. The corresponding probability is 0.9 < p < 0.1. This is below the conventionally accepted signicance level of 0.05 or 5%, so the null hypothesis that loss dierential is normal is veried for AP2 (LinkSys). Table 6.3 lists the results for AP1 and AP2 from Building One and Building Two. The dierence in path loss exponents for a closed corridor in Building One and Building Two is about 5%. So, it can be said that path loss models from Building Two validate the path loss models developed in Building One. 7. Compare with two-ray model.
92
Figure 6.10: Comparison of closed corridor mean loss with two-ray model. A graphical comparison is made between measured data and two-ray model. Figure 6.10 and Figure 6.11 show the comparison between the theoretical two-ray model and measured data. It can be seen from these plots that signal strength deteriorates at a faster rate in two-ray model than in actual measurements.
6.3 Scenario 2: Building Two Classroom Figure 6.12 and Figure 6.13 gives the loss (in dBm) versus T-R (Transmitter-Receiver) separation (in meters) for AP1 (D-Link) and AP2 (LinkSys) respectively. The experiment was conducted in an classroom at General Academic Building (GAB) in the University of North Texas main campus. As seen from Figure 6.12 and Figure 6.13 loss increases with the distance from the access point. 1. Calculate mean signal levels. Mean values of loss at each distance interval is calculated. Figure 6.14 shows a plot between mean loss for both the access points (D-Link and LinkSys) versus T-R separation. As seen from the gure, signal strength deteriorates with increasing T-R
93
Figure 6.11: Comparison of closed corridor AP1 and AP2 loss with mean signal and two-ray model.
Figure 6.12: Building Two, classroom loss for AP1 (D-Link).
94
Figure 6.13: Building Two, classroom loss for AP2 (LinkSys). separation. In the remaining part of the numerical analysis an equation is determined which describes the relationship between loss and T-R separation in a classroom for Building Two. 2. Using least squares method to calculate curve tting. Using MATLAB curve tting tool a linear t is calculated using least squares method. Figures 6.15 and 6.16 show the linear curve tting for the data collected using D-Link (AP1) and LinkSys (AP2) access points. Listed below are the output results from curve tting. Classroom curve tting for AP1 (D-Link). Linear model Poly1: f(x) = p1 * x + p2 Coecients (with 95 percent condence bounds): p1 = 13 (12.74, 13.26) p2 = 0 (xed at bound) 95
Figure 6.14: Building Two, classroom mean loss for AP1 and AP2. Classroom curve tting for AP2 (LinkSys). Linear model Poly1: f(x) = p1 * x + p2 Coecients (with 95 percent condence bounds): p1 = 12.98 (12.72, 13.25) p2 = 0 (xed at bound) 3. Use curve tting to evaluate path loss exponent n, in Log-distance Path Loss Model. From Figures 6.15 and 6.16 and output results listed above we get the slope of the tted curve to be 13.0 for D-Link and 12.98 for LinkSys. From (2.2), the slope of the tted curve is 10n, i.e. calculated path loss exponents (n) for D-link and LinkSys access points in a classroom is 1.3 and 1.298. From Section 5.2.1 the calculated path loss exponents for D-Link and LinkSys access points in a classroom for Building One are 1.258 and 1.263. It can be said that signal strength deteriorates at a faster rate in Building Two. Even though the width of dimensions of classrooms in both Building 96
Figure 6.15: Building Two, classroom curve tting for AP1 (D-Link).
Figure 6.16: Building Two, classroom curve tting for AP2 (LinkSys).
97
Figure 6.17: Building Two, classroom normal probability plot for AP1 (D-Link). One and Building Two are almost the same, the dierence in path loss exponents can be attributed to dierences in material used for construction and types of furniture in the two rooms. 4. Using Normal Plot, verify normal distribution. Normal probability plot is used to verify if the variation of loss (loss dierential) is distributed normally. The normal plots from Figure 6.17 and Figure 6.18 indicate that the assumption of normality is not unreasonable. Both plots pass the fat pencil test. 5. Evaluate standard deviation , for log-normal distribution, and determine standard normal distribution. Mean , and standard deviation of loss dierential is calculated for both the access points. Mean is calculated to be 0 and standard deviation for D-Link access point is 4.127 and for LinkSys standard deviation is 4.2860. Using (3.1) loss dierential is standardized. Figure 6.19 and Figure 6.20 show the plots for standard normal distributions for AP1 and AP2. 6. Calculate Chi-square Goodness-of-Fit Test. 98
Figure 6.18: Building Two, classroom normal probability plot for AP2 (LinkSys).
Figure 6.19: Building Two, classroom standard normal distribution for AP1 (D-Link).
99
Observed 0 0 0 0 6 21 59 134 256 339 446 408 225 132 56 21 5 2 1 1 0 2112
Chi Statistic 0.002112 0.02112 0.16474 1.0328 0.16854 0.11136 0.00076798 0.14802 0.00095691 2.4226 2.026 4.1454 3.6421 3.08E-01 0.13218 0.11136 0.0011124 0.90585 4.2351 45.37 0.002112 -
100
Observed 0 0 1 2 7 17 62 123 238 374 419 413 233 133 62 18 9 1 0 0 0 2112
Chi Statistic 0.002112 0.02112 4.2351 0.90585 0.73005 0.32664 0.17555 1.7406 1.1994 0.070658 0.010215 5.2733 1.9823 0.22061 0.17555 0.11918 3.0353 0.0010397 0.16474 0.02112 0.002112 -
101
Figure 6.20: Building Two, classroom standard normal distribution for AP2 (LinkSys). As discussed in section 3.3 to evaluate goodness of t we need to calculate test statistic X 2 and degrees of freedom df . Table 6.4 and 6.5 are used for calculating Chi-square test statistic X 2 for AP1 and AP2. To nd the expected value for a particular bin, the area under standard normal curve for the bin is multiplied by total observations. Chi-square test statistic is calculated using (3.2). To evaluate if the distribution is normal, the bins that are used to calculate X 2 are from -3.25 to 3.25, a total of 13 bins. As discussed in section 3.3, degrees of freedom df = 13 - 1 = 12. From Table 6.4 X 2 for AP1 (D-Link)= 13.2181, and from Table 5.6 X 2 for AP2 (LinkSys)= 15.0594. By using Table 3.2 we can identify to p-value associated with X 2 and df values. We now have our chi square statistic X 2 for AP1 and AP2 (13.2181 and 15.0594), our predetermined level of signicance (0.05), and our degrees of freedom (df =12). From Table 3.2 we can say X 2 for AP1 = 13.2181, lies between 6.304 and 18.549. The corresponding probability is 0.9 < p < 0.1. This is below the conventionally accepted signicance level of 0.05 or 5%, so the null hypothesis that loss dierential is normal is veried for AP1 (D-Link). 102
Table 6.6: Classroom results from Building One and Building Two.
Now for AP2 our value of X 2 = 15.0594, lies between 6.304 and 18.549. The corresponding probability is 0.9 < p < 0.1. This is below the conventionally accepted signicance level of 0.05 or 5%, so the null hypothesis that loss dierential is normal is veried for AP2 (LinkSys). Table 6.6 lists the results for AP1 and AP2 from Building One and Building Two. The dierence in path loss exponents for a classroom in Building One and Building Two is less than 5%. So, it can be said that path loss models from Building Two validate the path loss models developed in Building One. 7. Compare with two-ray model. A graphical comparison is made between measured data and two-ray model. Figure 6.21 and Figure 6.22 show the comparison between the theoretical two-ray model and measured data. It can be seen from these plots that signal strength deteriorates much higher in two-ray model when compared to classroom measurements.
6.4 Summary Tables 6.3 and 6.6 compare Path loss exponent (n), mean () and Standard deviation () for Closed Corridor and Classroom scenarios in Building One and Building Two. It can be 103
Figure 6.21: Comparison of classroom mean loss with two-ray model.
104
seen that the Path loss exponent (n), mean () and Standard deviation () of the respective scenarios are approximately equal. So the path loss models developed in chapter 5 can be used in generation of Propagation models for indoor environment.
6.5 Conclusion In this study indoor propagation models were developed for an Open Corridor, a Closed Corridor, a Classroom and a Computer lab. In Chapter 2 a brief description of dierent types of propagation models (Empirical & Deterministic models) was given. Chapter 2 also gives a description of Log-distance Path Loss Model and Log-Normal Shadowing (Empirical model), and Two-Ray Model (Deterministic model). These models are later used in determining path loss equations for dierent scenarios. In Chapter 3 statistical tools used to support the ndings of the study are described. A detailed description of Normal Plot, Standard Normal Distribution, and Chi-square Goodnessof-Fit Test is given. The chapter also discusses Curve Fitting which was used to calculate the Path Loss exponent (n). In Chapter 4 a detailed description of the measurement scenarios, the experimental setup, the software, and hardware used for measurements was given. In Chapter 5 a numerical analysis of measurements in each scenario was conducted and the study determined equations that describe path loss for each scenario. This chapter concludes by accepting Alternative Hypotheses for both the Research Questions. Chapter 6 does a comparative study of the measurement scenarios in Building One and Building Two. The chapter concludes that there is no dierence in path loss models calculated in Building One and Building Two.
105
6.6 Future Work We conclude by outlining possible directions for future research: 1. A further extension to this study could be developing an algorithm for Optimal Access Point Selection and Channel Assignment. The path loss equations, and multipath fading calculated in this study can be used to simulate propagation in indoor environments. By combining these two studies a GUI based model can be developed for ecient positioning of Access Points in an indoor environment. A user can specify the maximum number of access points, and the algorithm would nd the optimal placement of access points in a given service area. 2. IEEE 802.11b and 802.11g use the unlicensed ISM frequency. Many electric appliances such as microwave ovens, cordless phones, and Bluetooth devices use the same frequency band. The Propagation models can further be extended to consider interferences between these electric devices and wireless LANs.
106
BIBLIOGRAPHY [1] T.S. Rappaport. Wireless Communications Principles and Practice. Prentice Hall, 1996. [2] William Stallings. Wireless Communications and Networks. Prentice Hall, 2002. [3] IEEE. Wireless LAN medium access control (MAC) and physical layer (PHY) specications. IEEE Standard 802.11, 1997. [4] G.A. Halls. HIPERLAN: the high performance radio local area network standard. Electronics and Communication Engineering Journal, 6:289296, December 1994. [5] IEEE. Wireless LAN medium access control (MAC) and physical layer (PHY) specications - Amendment 4: Further higher-speed physical layer extension in the 2.4 GHz band. IEEE Standard 802.11g, 2003. [6] IEEE. Wireless LAN medium access control (MAC) and physical layer (PHY) specications - Amendment 2: Higher speed physical layer (PHY) extension in the 2.4 GHz band. IEEE Standard 802.11b, 1999. [7] SIG. Bluetooth specication version 1.1. http://www.bluetooth.com/dev/specications.asp, February 2001. [8] IEEE. Wireless LAN medium access control (MAC) and physical layer (PHY) specications - Amendment 1: High-speed physical layer in the 5 GHz band. IEEE Standard 802.11a, 1999. [9] D. Dobkin. Indoor propagation issues for wireless LANs. RF Design; www.rfdesign.com, September 2002. [10] H.J. Zepernick and T.A. Wysocki. Multipath channel parameters for the indoor radio at 2.4 GHz ISM band. IEEE Veh. Technol. Conf., 1:190193, May 1999.
107
[11] A.F. Agelet, A. Formella, J.M.H. Rabanos, I.F. de Vicente, and P.F. Fontan. Ecient ray-tracing acceleration techniques for radio propagation modeling. IEEE Veh. Technol. Conf., 49:20892104, November 2000. [12] M. Iskander, Z. Yun, and Z. Zhang. Outdoor/indoor propagation modeling for wireless communications systems. IEEE Antennas and Propagation Society International Symposium, 2:150153, July 2001. [13] S.P.T. Kumar, B. Farhang-Boroujeny, S. Uysal, and C.S. Ng. Microwave indoor radio propagation measurements and modeling at 5 GHz for future wireless LAN systems. Microwave Conference, 1999 Asia Pacic, 3:606609, 30 November-3 December. 1999. [14] K.A. Remley, H.R. Anderson, and A. Weisshar. Improving the accuracy of ray-tracing techniques for indoor propagation modeling. IEEE Veh. Technol. Conf., 49:2350 2358, November 2000. [15] W.K. Tam and V.N. Tran. Multi-ray propagation model for indoor wireless communications. Electronics and Communication Engineering Journal, 32:135137, January 1996. [16] J. Tarng and T. Liu. Eective models in evaluating radio coverage on single oors of multioor building. IEEE Veh. Technol. Conf., 48:782789, 1999. [17] A. Neskovic, N. Neskovic, and G. Paunovic. Modern approaches in modeling of IEEE Communications Surveys,
mobile radio systems propagation environment. http://www.comsoc.org/pubs/surveys, 2000.
[18] T. Holt, I. C. Pahlavan, and J. F. Lee. Ray tracing algorithm for indoor radio propagation modeling. 3rd IEEE Int. Symp. on Personal, Indoor and Mobile Radio Communications, October 1992.
108
[19] J. W. McKown and R. L. Hamilton. Ray tracing as a design tool for radio networks. IEEE Network, 6:2730, November 1991. [20] S. Y. Seidel and T. S. Rappaport. A ray tracing technique to predict path loss and delay spread inside buildings. IEEE Global Telecommunications Conf., 2:649653, December 1992. [21] R. Valenzuela. A ray tracing approach to predicting indoor wireless transmission. IEEE Veh. Technol. Conf., 43rd:214218, May 1993. [22] IEEE. Wireless LAN medium access control (MAC) and physical layer (PHY) specications Amendment 6: Medium Access Control (MAC) security enhancements. IEEE Standard 802.11i, 2004. [23] J. C. Chen, M. C. Jiang, and Y. W. Liu;. Wireless LAN security and IEEE 802.11i. Wireless Communications, IEEE, 12:2736, February 2005. [24] Cyrus Peikari and Seth Fogie. Maximum Wireless Security. Sams, 2002. [25] Statit Software:Process Management Tools and Applications. http://www.statit.com, November 1999. [26] The MathWorks Inc. MATLAB HELP. http://www.mathworks.com. [27] StatsDirect Statistical Software. Statistical help. http://www.statsdirect.com. [28] LinkSys A Division of Cisco Systems, Inc. http://www.LinkSys.com/. [29] D-Link Corporation/D-Link Systems, Inc. http://www.dlink.com/. [30] M. Milner. Network stumbler version 0.4.0. NetStumbler; www.netstumbler.com, 20012004. Statit Bulletin.
109
INDOOR PROPAGATION MODELING AT 2.4 GHZ FOR IEEE 802.11 NETWORKS Dinesh Tummala, B.S.
Thesis Prepared for the Degree of MASTER OF SCIENCE
UNIVERSITY OF NORTH TEXAS December 2005
APPROVED: Robert Akl, Major Professor Albert B. Grubbs, Committee Member and Chair of the Department of Engineering Technology Robert G. Hayes, Committee Member Vijay Vaidyanathan, Committee Member Oscar N. Garcia, Dean of the College of Engineering Sandra L. Terrell, Dean of the Robert B. Toulouse School of Graduate Studies
Tummala, Dinesh. Indoor Propagation Modeling at 2.4 GHz for IEEE 802.11 Networks. Master of Science (Engineering Technology), December 2005, 109 pp., 18 tables, 80 figures, 47 titles. Indoor use of wireless systems poses one of the biggest design challenges. It is difficult to predict the propagation of a radio frequency wave in an indoor environment. To assist in deploying the above systems, characterization of the indoor radio propagation channel is essential. The contributions of this work are two-folds. First, in order to build a model, extensive field strength measurements are carried out inside two different buildings. Then, path loss exponents from log-distance path loss model and standard deviations from log-normal shadowing, which statistically describe the path loss models for a different transmitter receiver separations and scenarios, are determined. The purpose of this study is to characterize the indoor channel for 802.11 wireless local area networks at 2.4 GHz frequency. This thesis presents a channel model based on measurements conducted in commonly found scenarios in buildings. These scenarios include closed corridor, open corridor, classroom, and computer lab. Path loss equations are determined using log-distance path loss model and log-normal shadowing. The chi-square test statistic values for each access point are calculated to prove that the observed fading is a normal distribution at 5% significance level. Finally, the propagation models from the two buildings are compared to validate the generated equations.
ACKNOWLEDGEMENTS I express my gratitude of thanks to my major advisor and Chair of Department of Engineering Technology Dr. Albert B.Grubbs for his support, confidence, encouragement, and guidance throughout my graduate program and in completing this thesis. From the formative stages of this thesis, to the final draft, I am indebted to my supervisor, Dr. Robert Akl, whose expertise, understanding, and patience, added considerably to my graduate experience. I thank the members of my committee, Dr. Robert G. Hayes, and Dr. Vijay Vaidyanathan for the assistance, careful reading and suggestions they provided at all levels of the research project. I would also like to thank my supervisors at work Mr. Ken Brinkley and Mr. Larry Guay for being supportive. I really appreciate the flexible work schedule you provided. I would also like to thank my family for the support they provided me through my entire life and in particular, I must acknowledge my wife and best friend, Swathi, without whose love, encouragement and editing assistance, I would not have finished this thesis. Finally, but not least, I want to thank my parents Mr. Gopala Krishna and Mrs. Vijaya. Thanks for encouraging me to be an independent thinker, and having confidence in my abilities to go after new things that inspired me. Thanks Dad for taking time to work with me on my mathematics (You were the best mathematics teacher I ever had). Thanks for teaching me that it is important to try to leave the world just a little better than when you came into it. And, of course, thank you both for your constant support through the ups and downs of my academic career. It has been bumpy at times, but your confidence in me has enhanced my ability to get through it all and succeed in the end.
ii
TABLE OF CONTENTS Page ACKNOWLEDGEMENTS..........................................................................................................ii LIST OF TABLES........................................................................................................................ v LIST OF FIGURES ..................................................................................................................... vi Chapters 1. INTRODUCTION ................................................................................................ 1 1.1 1.2 1.3 1.4 2. Communication Systems .......................................................................... 2 Problem Statement and Objectives ........................................................... 3 Research Questions................................................................................... 3 Organization.............................................................................................. 4
INDOOR PROPAGATION MODELING ........................................................... 5 2.1 2.2 2.3 2.4 2.5 Introduction............................................................................................... 5 Line of Sight Propagation ......................................................................... 5 Channel Modeling..................................................................................... 8 Propagation Models .................................................................................. 8 Empirical Models...................................................................................... 9 2.5.1 Log-Distance Path Loss Model................................................... 10 2.5.2 2.6 2.7 2.8 2.9 Log-Normal Shadowing.............................................................. 10 Two-Ray Model ...................................................................................... 11 Indoor RF Propagation and Wireless Local Area Network Technology ................................................................................................................. 13 IEEE 802.11 Standard............................................................................. 15 Frequency Range and Channel Allocation for 802.11b.......................... 16
3.
STATISTICAL ANALYSIS .............................................................................. 19 3.1 3.2 3.3 3.3.1 Normal Plot............................................................................................. 19 Standard Normal Distribution................................................................. 20 Chi-square Goodness-of-Fit Test............................................................ 20 Deciding What Value of X2 is Critical for Accepting or Rejecting a Hypothesis............................................................................................... 22
iii
3.3.2 3.4 4.
Finding p-Value for Chi-square Test ...................................................... 23
Curve Fitting ....................................................................................................... 24 EXPERIMENTAL SETUP................................................................................. 27 4.1 4.2 Introduction............................................................................................. 27 Propagation Environment ....................................................................... 27 4.2.1 Description of Measurement Scenarios for Building One.......... 31 4.2.2 Description of Measurement Scenarios for Building Two ......... 34 4.3 4.4 4.5 Hardware and Software Description....................................................... 34 Data Acquisition ..................................................................................... 36 Numerical Analysis................................................................................. 36
5.
NUMERICAL ANALYSIS ................................................................................ 38 5.1 5.2 Introduction............................................................................................. 38 Analysis of Results ................................................................................. 40 5.2.1 Scenario 1: Building One Closed Corridor................................. 40 5.2.2 Scenario 2: Building One Open Corridor ................................... 50 5.2.3 Scenario 3: Building One Classroom.......................................... 59 5.2.4 Scenario 4: Building One Computer Lab.................................... 71 5.3 Summary of Results................................................................................ 80
6.
COMPARISON WITH SIMILAR SCENARIOS .............................................. 83 6.1 6.2 6.3 6.4 6.5 6.6 Introduction............................................................................................. 83 Scenario 1: Building Two Closed Corridor ............................................ 83 Scenario 2: Building Two Classroom ..................................................... 93 Summary ............................................................................................... 103 Conclusion ............................................................................................ 105 Future Work .......................................................................................... 106
BIBLIOGRAPHY......................................................................................................... 107
iv
LIST OF TABLES Page 2.1 2.2 3.1 3.2 5.1 5.2 5.3 5.4 5.5 5.6 5.7 5.8 6.1 6.2 6.3 6.4 6.5 6.6 Path loss exponents for different environments [1] ........................................................ 11 Frequency and channel assignments............................................................................... 17 Area under the normal curve from 0 to X....................................................................... 21 Critical values of chi-square distribution with degrees of freedom................................ 25 Closed corridor X2 table for AP1 (D-Link).................................................................... 46 Closed corridor X2 table for AP2 (LinkSys) ................................................................. 47 Open corridor X2 table for AP1 (D-Link) ...................................................................... 57 Open corridor X2 table for AP2 (LinkSys)..................................................................... 58 Classroom X2 table for AP1 (D-Link)............................................................................ 67 Classroom X2 table for AP2 (LinkSys) .......................................................................... 68 Computer lab X2 table for AP1 (D-Link)....................................................................... 77 Computer lab X2 table for AP2 (LinkSys) ..................................................................... 78 Closed corridor X2 table for AP1 (D-Link).................................................................... 90 Closed corridor X2 table for AP2 (LinkSys) .................................................................. 91 Closed corridor results from Building One and Building Two....................................... 92 Classroom X2 table for AP1 (D-Link).......................................................................... 100 Classroom X2 table for AP2 (LinkSys) ........................................................................ 101 Classroom results from Building One and Building Two ............................................ 103
LIST OF FIGURES Page 1.1 2.1 2.2 4.1 4.2 4.3 4.4 4.5 4.6 4.7 4.8 4.9 5.1 5.2 5.3 5.4 5.5 5.6 5.7 5.8 5.9 5.10 5.11 Basic communication system............................................................................................ 2 Two-ray model................................................................................................................ 13 802.11b channel overlap ................................................................................................. 18 College of Engineering, first floor overview .................................................................. 28 College of Engineering, second floor overview ............................................................. 29 College of Engineering, second floor detail ................................................................... 30 Building One closed corridor.......................................................................................... 31 Building One open corridor ............................................................................................ 32 Building One classroom.................................................................................................. 33 Building One computer lab............................................................................................. 33 Building Two closed corridor ......................................................................................... 35 Building Two classroom ................................................................................................. 35 Numerical analysis.......................................................................................................... 39 Building One, closed corridor loss for AP1 (D-Link) .................................................... 40 Building One, closed corridor loss for AP2 (LinkSys)................................................... 41 Building One, closed corridor mean loss for AP1 and AP2 ........................................... 42 Building One, closed corridor curve fitting for AP1 (D-Link)....................................... 43 Building One, closed corridor curve fitting for AP2 (LinkSys) ..................................... 43 Building One, closed corridor normal probability plot for AP1 (D-Link) ..................... 44 Building One, closed corridor normal probability plot for AP2 (LinkSys).................... 45 Building One, closed corridor standard normal distribution for AP1 (D-Link) ............. 45 Building One, closed corridor standard normal distribution for AP2 (LinkSys)............ 48 Comparison of closed corridor mean loss with two-ray model ...................................... 49
vi
5.12 5.13 5.14 5.15 5.16 5.17 5.18 5.19 5.20 5.21 5.22 5.23 5.24 5.25 5.26 5.27 5.28 5.29 5.30 5.31 5.32 5.33 5.34 5.35
Comparison of closed corridor AP1 and AP2 loss with mean signal and two-ray model ......................................................................................................................................... 50 Building One, open corridor loss for AP1 (D-Link)....................................................... 51 Building One, open corridor loss for AP2 (LinkSys) ..................................................... 51 Building One, open corridor mean loss for AP1 and AP2.............................................. 52 Building One, open corridor curve fitting for AP1 (D-Link) ......................................... 53 Building One, open corridor curve fitting for AP2 (LinkSys)........................................ 54 Building One, open corridor normal probability plot for AP1 (D-Link)........................ 54 Building One, open corridor normal probability plot for AP2 (LinkSys) ...................... 55 Building One, open corridor standard normal distribution for AP1 (D-Link)................ 56 Building One, open corridor standard normal distribution for AP2 (LinkSys) .............. 56 Comparison of open corridor mean loss with two-ray model......................................... 60 Comparison of open corridor AP1 and AP2 loss with mean signal and two-ray model ......................................................................................................................................... 60 Building One, classroom loss for AP1 (D-Link) ............................................................ 61 Building One, classroom loss for AP2 (LinkSys)........................................................... 61 Building One, classroom mean loss for AP1 and AP2 ................................................... 62 Building One, classroom curve fitting for AP1 (D-Link)............................................... 63 Building One, classroom curve fitting for AP2 (LinkSys) ............................................. 64 Building One, classroom normal probability plot for AP1 (D-Link) ............................. 65 Building One, classroom normal probability plot for AP2 (LinkSys)............................ 65 Building One, classroom standard normal distribution for AP1 (D-Link) ..................... 66 Building One, classroom standard normal distribution for AP2 (LinkSys).................... 66 Comparison of classroom mean loss with two ray-model .............................................. 70 Comparison of classroom AP1 and AP2 loss with mean signal and two-ray model...... 70 Building One, computer lab loss for AP1 (D-Link) ....................................................... 71
vii
5.36 5.37 5.38 5.39 5.40 5.41 5.42 5.43 5.44 5.45 5.46 6.1 6.2 6.3 6.4 6.5 6.6 6.7 6.8 6.9 6.10 6.11 6.12 6.13
Building One, computer lab loss for AP2 (LinkSys)...................................................... 72 Building One, computer lab mean loss for AP1 and AP2 .............................................. 72 Building One, computer lab curve fitting for AP1 (D-Link) .......................................... 74 Building One, computer lab curve fitting for AP2 (LinkSys) ........................................ 74 Building One, computer lab normal probability plot for AP1 (D-Link)......................... 75 Building One, computer lab normal probability plot for AP2 (LinkSys) ....................... 76 Building One, computer lab standard normal distribution for AP1 (D-Link) ................ 76 Building One, computer lab standard normal distribution for AP2 (LinkSys)............... 79 Comparison of computer lab mean loss with two ray model.......................................... 80 Comparison of computer lab AP1 and AP2 loss with mean signal and two ray model ......................................................................................................................................... 81 Results at a glance........................................................................................................... 82 Building Two, closed corridor loss for AP1 (D-Link).................................................... 84 Building Two, closed corridor loss for AP2 (LinkSys) .................................................. 84 Building Two, closed corridor mean loss for AP1 and AP2........................................... 85 Building Two, closed corridor curve fitting for AP1 (D-Link) ...................................... 86 Building Two, closed corridor curve fitting for AP2 (LinkSys)..................................... 87 Building Two, closed corridor normal probability plot for AP1 (D-Link)..................... 88 Building Two, closed corridor normal probability plot for AP2 (LinkSys) ................... 88 Building Two, closed corridor standard normal distribution for AP1 (D-Link)............. 89 Building Two, closed corridor standard normal distribution for AP2 (LinkSys) ........... 89 Comparison of closed corridor mean loss with two-ray model ...................................... 93 Comparison of closed corridor AP1 and AP2 loss with mean signal and two-ray model ......................................................................................................................................... 94 Building Two, classroom loss for AP1 (D-Link)............................................................ 94 Building Two, classroom loss for AP2 (LinkSys) .......................................................... 95
viii
6.14 6.15 6.16 6.17 6.18 6.19 6.20 6.21 6.22
Building Two, classroom mean loss for AP1 and AP2 .................................................. 96 Building Two, classroom curve fitting for AP1 (D-Link) .............................................. 97 Building Two, classroom curve fitting for AP2 (LinkSys)............................................. 97 Building Two, classroom normal probability plot for AP1 (D-Link)............................. 98 Building Two, classroom normal probability plot for AP2 (LinkSys) ........................... 99 Building Two, classroom standard normal distribution for AP1 (D-Link) .................... 99 Building Two, classroom standard normal distribution for AP2 (LinkSys)................. 102 Comparison of classroom mean loss with two-ray model ............................................ 104 Comparison of classroom AP1 and AP2 loss with mean signal and two-ray model.... 104
ix
CHAPTER 1
INTRODUCTION Wireless communication is one of the most active areas of technology development of our time. This development is being driven primarily by the transformation of what has been largely a medium for supporting voice telephony into a medium for supporting other services, such as the transmission of video, images, text, and data. Thus, similar to the developments in wireline capacity in the 1990s, the demand for new wireless capacity is growing at a very rapid pace. The impact of wireless technology has been and will continue to be profound. The convergence of dierent standards that dene how wireless devices interact will allow the creation of a global wireless network that will deliver a wide variety of services. Cellular phones are currently the most obvious sign of the advent of wireless technology, but mobile telephones are only the tip of the cellular revolution. The rst rush to wireless was for voice. Now, the attention is on data. Presently, there are many types of wireless networks in use around the world. Most new devices have access to the internet. A big part of this market is the wireless internet. The Internet is increasingly becoming a multimedia experience. For wireless networks to compete with their xed counterparts, wireless networks need to obtain higher data rates [2]. Wireless Local Area Networks (WLANs) provide network services where it is dicult or too expensive to deploy a xed infrastructure. WLANs can coexist with xed infrastructure to provide mobility and exibility to users. The primary WLAN standards are IEEE 802.11 [3] and Europes HyperLAN [4]. The 802.11 protocol set, popularly known as Wi-Fi, includes wireless network standards that allow data transmission up to a theoretical 54 Mbps [5]. WLANs operate mainly in a indoor environment. It is very dicult to predict how a 1
Figure 1.1: Basic communication system. RF wave travels in an indoor environment. So there is a need for developing an indoor propagation model to predict RF wave behavior more accurately.
1.1 Communication Systems Any communication system can be viewed as a link between a source and a destination where information is sent from the source and received at the destination. The intervening stages are shown in Figure 1.1. The transmitter takes the information from the source and codes it in a form suitable for transfer over the channel such that the cost of transmission is minimal. In this context, cost is a function of the bandwidth used, the time taken to perform the communication, the degree to which the transmission interferes with other transmissions occurring simultaneously and the amount of information that is lost in the communication process. The channel is a description of how the communications medium alters the signal that is being transmitted. Finally the receiver takes the signals that have been altered by the channel, and attempts to recover the information that was sent by the source. This recovered signal is passed to the destination as the received information. For a radio communication system, the channel describes how the electromagnetic propagation of a transmitted signal provides that signal at the receiver. In a mobile communication system, the channel changes according to the movement of the communicating entities and other objects that have an eect on the electromagnetic elds at the receiver. The purpose of this study is to characterize the indoor channel for 802.11b wireless local area networks at 2.4 GHz frequency [6].
1.2 Problem Statement and Objectives WLANs are rapidly gaining popularity. These networks are primarily targeted for indoor use, and are most often based on either the IEEE 802.11 Ethernet-type protocols or the Bluetooth Special Interest Group (SIG), both using the unlicensed bands at 2.4 to 2.5 GHz, IEEE 802.11b [6] and Bluetooth [7]), or at 5.15 to 5.85 GHz, IEEE 802.11a [8]. The European HiperLAN standard is also designed for operation around 5.2 to 5.8 GHz [4]. Indoor use of wireless systems poses one of the biggest design challenges. It is dicult to predict the propagation of a RF wave in an indoor environment [9][10]. To assist in deploying the above systems, characterization of the indoor radio propagation channel is essential. The proposed innovation characterizes the indoor channel by developing a propagation model at 2.4 GHz for dierent scenarios commonly found in buildings. By using this propagation model, network analysis and simulation can be developed. This will facilitate faster and more ecient deployment of wireless networks.
1.3 Research Questions The research questions addressed in this study are stated below for hypotheses testing. 1. Is multipath fading observed in indoor radio propagation at 2.4 GHz distributed normally? Null Hypotheses: Multipath fading observed in indoor radio propagation at 2.4 GHz is not a normal distribution. Alternative Hypotheses: Multipath fading observed in indoor radio propagation at 2.4 GHz is a normal distribution.
Normal probability plot and Chi-square Goodness of t tests are used to verify the above hypotheses. 2. Does indoor radio propagation at 2.4 GHz depend on the indoor environment? Indoor environment or indoor channel consists of hard and soft partitions (Section 2.7). Null Hypotheses: Indoor radio propagation at 2.4 GHz is not dependent on the indoor environment. Alternative Hypotheses: Indoor radio propagation at 2.4 GHz is dependent on the indoor environment.
1.4 Organization This thesis creates a model of an indoor propagation channel at 2.4 GHz for dierent scenarios commonly found in buildings. The remainder of this work consists of the following chapters. Chapter 2 reviews the published work that has been conducted in the eld of channel measurement and modeling. Concepts in channel modeling and measurements are examined. Chapter 3 describes the statistical analysis required to measure the quality of the models developed in this study. Chapter 4 describes the experimental setup, the software and hardware used for measurements, and the propagation environment in which measurements are done. A detailed description of each measurement scenario is discussed. Chapter 5 does a numerical analysis of measurements conducted in each scenario. The study determines equations that describe path loss for each scenario. Chapter 6 compares the measurements from scenarios in dierent buildings and compares them to measurements from similar scenarios in Chapter 5. Conclusions and possible direction of future work in this eld are also presented.
CHAPTER 2
INDOOR PROPAGATION MODELING 2.1 Introduction A signal radiated from an antenna travels along one of the three routes: ground wave, sky wave, or line of sight (LOS). Based on the operating frequency range, one of the three predominates. LOS propagation is the mode of propagation which is of interest in this paper.
2.2 Line of Sight Propagation At frequencies higher than 30 MHz, LOS is the dominant propagation mode. The ionosphere reects less of the signal as the frequency is increased beyond 30 MHz. A signal can thus be transmitted either to a satellite or to a receiving antenna which is in the line of sight of the transmitting antenna. In a communication system, a received signal will dier from the transmitted signal due to various transmission impairments. The most signicant transmission impairments for LOS transmission are [2]: Attenuation: The strength of a signal falls o with distance over any transmission medium. This reduction in strength or attenuation is logarithmic for guided media. Whereas attenuation is a more complex function of distance and the makeup of the atmosphere for an unguided media. Free space loss: In any wireless communication, the signal disperses with distance. A receiving antenna will receive less signal power the farther it is from the transmitting antenna. Assuming all the sources of impairments are nullied the transmitted signal
attenuates over distance because the signal is being spread over a larger and larger area. This form of attenuation is known as free space loss. Fading: Fading refers to the time variation of received signal power caused by changes in the transmission medium or path. Fading is the most challenging technical problem in designing a communication system. In a xed environment, fading is aected by changes in atmospheric conditions. Whereas in a mobile environment where either the receiving or transmitting antenna is in motion relative to the other, the relative location of various obstacles changes with time, causing complex transmission eects. Multipath: Multipath is dened as a propagation phenomenon that results in radio signals reaching the receiving antenna by two or more paths. The direct and reected signals are often opposite in phase, which can result in a signicant signal loss due to mutual cancelation in some circumstances. Depending on the dierences in the path lengths of direct and reected waves, the composite signal can be either larger or smaller than the direct signal. Multipath is most troublesome indoors and in areas where many metallic surfaces are present. Multipath is caused by the following propagation mechanisms: Reection: Reection occurs when a propagating electromagnetic wave impinges upon an object which has very large dimensions when compared to the wavelength of the propagating wave. Reections occur from the surface of the earth and from buildings and walls. The reected waves may interfere constructively or destructively at the receiver. Diraction: Diraction occurs when the radio path between the transmitter and receiver is obstructed by a surface that is large compared to the wavelength of the radio wave. The secondary waves resulting from the obstructing surface are present throughout the space and even behind the obstacle, giving rise to a bend6
ing of waves around the obstacle, even when a line-of-sight path does not exist between transmitter and receiver. Scattering: Scattering occurs when the medium through which the wave travels consists of objects with dimensions that are small compared to the wavelength, and where the number of obstacles per unit volume is large. Scattered waves are produced by rough surfaces, small objects, or by other irregularities in the channel. These three propagation eects inuence system performance in various ways depending on local conditions and as a mobile unit moves through the medium. Diraction and scattering are generally minor eects if there is a clear LOS between transmitter and receiver although reection may have a signicant impact. In cases where there is no LOS, diraction and scattering are the primary means of signal reception. Refraction: Refraction is dened as a change in direction of an electromagnetic wave resulting from changes in the velocity of propagation of the medium through which it passes. This may result in a situation in which only a fraction or no part of the line of sight wave reaches the receiving antenna. Noise: In any transmission event, a received signal will consist of the transmitted signal, modied by various distortions imposed by the transmission medium, plus additional unwanted signals that are inserted by the medium. These unwanted signals are referred to as noise or interference. Noise is the major limiting factor in any communications system performance. Atmospheric absorption: Atmospheric absorption is an additional loss due to the presence of dierent atmospheric elements such as water vapor and oxygen etc. A peak attenuation occurs in the vicinity of 22 GHz due to water vapor. At frequencies below 15 GHz, the attenuation is less. 7
2.3 Channel Modeling In order to evaluate the eectiveness of coding and processing techniques for a given channel prior to construction, a model of the channel must be developed that adequately describes the environment. Such analysis reduces the cost of developing a complex system by reducing the amount of hardware required for evaluation of performance. Indoor channels are highly dependent upon the placement of walls and partitions within the building. As placement of these walls and partitions dictates the signal path inside a building. In such cases, a model of the environment is a useful design tool in constructing a layout that leads to ecient communication strategies. To achieve this aim, a channel model of an indoor environment must be applied to various layout plans of oces which will lead to the characterization of design methodologies. A channel model is useful in determining the mechanisms by which propagation in the indoor environment occurs, which in turn is useful in the development of a communication system. By examining the details of how a signal is propagated from the transmitter to the receiver for a number of experimental locations, a generic model may be developed that highlights the important characteristics of a given indoor environment. Generic models of indoor communications can then be applied to specic situations to describe the operation of a radio system, and may also be used to generate building designs that are particularly suited to supporting radio communication systems.
2.4 Propagation Models In the literature, there are numerous experimental and theoretical studies of indoor propagation [11][12][13][14][15][16][10]. These models tend to focus on a particular characteristic like temporal fading or inter-oor losses. This study aims at developing an indoor propagation model from measurements taken using 802.11b compliant access point and client adapters.
The study focuses on generating dierent loss equations for dierent scenarios in indoor environments. Using these equations, an accurate model can be developed to visualize the propagation phenomenon for dierent buildings with changing indoor environments. A propagation model is a set of mathematical expressions,diagrams, and algorithms used to represent the radio characteristics of a given environment [17]. The prediction models can be either empirical (also called statistical) or theoretical (also called deterministic), or a combination of these two. While the empirical models are based on measurements, the theoretical models deal with the fundamental principles of radio wave propagation phenomena. In the empirical models, all environmental inuences are implicitly taken into account regardless of whether they can be separately recognized. This is the main advantage of these models. Because deterministic models are based on the principles of physics they may be applied to dierent environments without aecting the accuracy. In practice, their implementation usually requires a huge database of environmental characteristics, which is sometimes either impractical or impossible to obtain. The algorithms used by deterministic models are usually very complex and lack computational eciency. For that reason, the implementation of the deterministic models is commonly restricted to smaller areas of microcell or indoor environments. On the basis of the radio environment, the prediction models can be classied into two main categories, outdoor and indoor propagation models. Further, in respect to the size of the coverage area, the outdoor propagation models can be subdivided into two additional classes, macrocell and microcell prediction models. A discussion of one empirical and one theoretical model used in this study is presented in the following sections.
2.5 Empirical Models Both theoretical and measurement based propagation models indicate that average received signal power decreases logarithmically with distance. Empirical models help in reducing
computational complexity as well as increasing the accuracy of the predictions [17]. The empirical model used in this study is Log-distance Path Loss Model.
2.5.1 Log-distance Path Loss Model In both indoor and outdoor environments the average large-scale path loss for an arbitrary Transmitter-Receiver (T-R) separation is expressed as a function of distance by using a path loss exponent, n [1]. The average path loss P L(d) for a transmitter and receiver with separation d is:
P L(d) or
d d0
(2.1)
P L(dB) = P L(d0 ) + 10 n log
d , d0
(2.2)
where n is the path loss exponent which indicates the rate at which path loss increases with distance d. Close in reference distance (d0 ) is determined from measurements close to the transmitter. The plot for distance d versus path loss P L on a log-log scale is a straight line with a slope equal to 10n. This value of n depends on the specic propagation environment, i.e., type of construction material, architecture, location within building. Lower the value of n lower the signal loss. The values of n range from 1.2 (Waveguide eect) to 6 [17]. For example, in free space, n is equal to 2, and when obstructions are present, n will have a larger value. Table 2.1 lists typical path loss exponents obtained in various radio environments.
2.5.2 Log-Normal Shadowing Random shadowing eects occuring over a large number of measurement locations which have the same T-R separation, but dierent levels of clutter on the propagation path is referred to as Log-Normal Distribution [1]. This phenomenon is referred to as log-normal
10
Table 2.1: Path loss exponents for dierent environments [1].

Environment Free Space Urban area cellular radio Shadowed urban cellular radio In building line-of sight Obstructed in building Obstructed in factories Path Loss Exponent, n 2 2.7 to 3.5 3 to 5 1.6 to 1.8 4 to 6 2 to 3
shadowing. Variations in environmental clutter at dierent locations having the same T-R separation is not accounted for in equation (2.2). This leads to measured signals which are vastly dierent than the average value predicted by (2.2). To account for the variations described above equation (2.2) is modied as:
P L(dB) = P L(d0 ) + 10 n log
d + X , d0
(2.3)
where X is a zero-mean Gaussian distributed random variable with standard deviation . The close-in reference distance d0 , the path loss exponent n, and the standard deviation , statistically describe the path loss model for an arbitrary location having a specic T-R separation. This model can be used in computer simulation to provide received power levels for random locations in communication system design and analysis.
2.6 Two-Ray Model Site specic propagation models are based on electromagnetic-wave propagation theory to characterize indoor radio propagation. Unlike statistical models, site specic propagation models do not rely on extensive measurement, but a greater detail of the indoor environment is required to obtain an accurate prediction of signal propagation inside a building.
11
In theory, electromagnetic-wave propagation characteristics could be exactly computed by solving Maxwells equations with the building geometry as boundary conditions. Unfortunately, this approach requires very complex mathematical operations and requires considerable computing power, beyond that of current microcomputers. Hence it is not economical for the characterization of indoor radio wave propagation. Therefore, approximate numerical methods are of interest. Ray tracing is an intuitively appealing method for calculating radio signal strength, time-invariant impulse response, root mean square (RMS) delay spread and related parameters in an indoor environment [18][19][20][21]. The concept of ray-tracing modeling is based on the fact that high-frequency radio waves behave in a ray-like fashion. Therefore, signal propagation can be modeled as ray propagation. By using the concept of ray-tracing, rays may be launched from a transmitter location and the interaction of the rays with partitions within a building modeled using well-known reection and transmission theory. Ray tracing can be much less demanding of computation than methods based on Maxwells equations. With the computing powers currently available on personal computers and workstations, the ray-tracing approach provides a challenging but feasible method of propagation modeling. Reliable site specic ray-tracing propagation prediction models, for each building that is based on its detailed geometry and construction, can be very eective tools in designing indoor communication systems. The ray-tracing approach approximates the scattering of electromagnetic waves by simple reection and refraction. The degree of transmission and reection of a signal through and o an obstacle is related to the complex permittivities of the obstacle. One of the propagation models based on ray-optic theory is a Two-Ray model. Two-Ray model is used in this study because all the scenarios considered in this study have one reecting surface, i.e. we have a direct path and reected path. It is used for modeling of Line of Sight radio channel as shown in Figure 2.1. The transmitting antenna of height h1 and the receiving antenna of
12
Figure 2.1: Two-Ray Model. height h2 are placed at distance d from each other. The received signal Pr for isotropic antennas, obtained by summing the contribution from each ray, can be expressed as Pr = Pt 4 1 1 (jkr1 ) e + () e(jkr2 ) r1 r2
2
(2.4)
where Pt is the transmitted power, r1 is the direct distance from the transmitter to the receiver, r2 is the distance through reection on the ground, and () is the reection coecient depending on the angle of incidence and the polarization. The reection coecient is given by cos a () = cos + a
r r
sin2 , sin2
(2.5) is
where = 90- and a = 1/ or 1 for vertical or horizontal polarization, respectively.
a relative dielectric constant of the reected surface. The signal strengths from theoretical and empirical models are compared in this study.
2.7 Indoor RF Propagation and Wireless Local Area Network Technology Indoor use of wireless systems poses one of the biggest design challenges. RF propagation obstacles can be termed hard partitions if they are part of the physical or structural components of a building. On the other hand, obstacles formed by oce furniture and xed or movable structures that do not extend to a buildings ceiling are considered soft partitions. Radio signals eectively penetrate both kinds of obstacles or partitions in ways that are very 13
hard to predict. An indoor propagation model is needed to predict the propagation of a transmitted signal in this environment. WLANs are implemented as an extension to wired LANs within a building and can provide the nal few meters of connectivity between a wired network and the mobile user. WLAN congurations vary from simple, independent, peer-to-peer connections between a set of PCs, to more complex, intra-building infrastructure networks. There are also point-topoint and point-to-multipoint wireless solutions. A point-to-point solution is used to bridge two local area networks, and to provide an alternative to cable between two geographically distant locations (up to 30 miles). Point-to-multi-point solutions connect several, separate locations to one single location or building. Both point-to-point and point-to-multipoint can be based on the 802.11b standard or on more costly infrared-based solutions that can provide throughput rates up to 622 Mbps. In a typical WLAN infrastructure conguration, there are two basic components: Access Points: An access point or a base station connects to a LAN by means of Ethernet cable. Usually installed in the ceiling, access points receive, buer, and transmit data between the WLAN and the wired network infrastructure. A single access point supports on average twenty users and has a coverage varying from 20 meters in areas with obstacles (walls, stairways, elevators) up to 100 meters in areas with clear line of sight. A building may require several access points to provide complete coverage and allow users to roam seamlessly between access points. Wireless Client Adapter: A wireless adapter connects users via an access point to the rest of the LAN. A wireless adapter can be a PC card in a laptop, an ISA or PCI adapter in a desktop computer, or fully integrated within a handheld device.
14
2.8
IEEE 802.11 Standard
IEEE 802.11 is a family of specications for WLANs developed by the Institute of Electrical and Electronics Engineers. The 802.11 standard species parameters for both the physical and medium access control (MAC) layers of a WLAN [3]. The physical layer handles the transmission of data between nodes. The MAC layer consists of protocols responsible for maintaining the use of the shared medium. Work on 802.11 began in 1987 within the IEEE 802.4 group. There are three physical layers for WLANs: two radio frequency specications (RF direct sequence and frequency hopping spread spectrum) and one infrared. Most WLANs operate in the 2.4 GHz license-free frequency band and have throughput rates up to 2 Mbps. There are various versions of the 802.11 standard. A brief description of the more popular revisions is given below. 802.11a: 802.11a operates at radio frequencies between 5 GHz and 6 GHz [8]. The modulation scheme used is orthogonal frequency-division multiplexing (OFDM). OFDM, also called multicarrier modulation, uses multiple carrier signals at dierent frequencies, sending some of the bits on each channel. This is similar to Frequency Division Multiplexing (FDM). The only dierence between FDM and OFDM is that in OFDM all the sub-channels are dedicated to a single data source. The data rates vary based on the noise level, distance from the transmitting antenna, and the propagation environment. Possible data rates for 802.11a are 6, 9, 12, 18, 24, 36, 48, and 54 Mbps. Maximum range for this standard is 200 feet. 802.11b: 802.11b often called Wi-Fi, being the most popular of all the standards, operates in the 2.4 GHz frequency [6]. It is an extension of the 802.11 standard. Typical data rates for 802.11b are 5.5 and 11 Mbps. The modulation scheme used is Direct Sequence Spread Spectrum. The chipping rate is 11 MHz, the same as in 15
802.11, providing the same occupied bandwidth. Although the data rates are slower than 802.11a, the range is higher, up to 300 feet. The frequency band used (2.4 GHz) can have signicant interference problems from such devices as microwave, cordless phones, and Bluetooth devices. 802.11g: 802.11g is the newest member of the 802.11 family. This standard combines the best of 802.11a and 802.11b. Like 802.11b, 802.11g operates in the 2.4 GHz frequency and can achieve ranges up to 300 feet, but like 802.11a, it reaches speeds up to 54 Mbps. 802.11g uses a hybrid complementary code keying OFDM modulation [5]. 802.11i: 802.11b uses Wired Equivalent Privacy (WEP) protocol to address security concerns. WEP itself is more or less an implementation of encryption with built-in message authentication and data integrity systems. The sheer number and variety of vulnerabilities discovered within WEP shows what could arise when security is not designed from the ground up. The future of wireless LAN security is currently being entrusted to 802.11i [22] [23]. IEEE is developing this wireless LAN standard, which focuses strictly on security and improving upon the protocols oered by the previous 802.11 standards. There are three main areas that the IEEE 802.11i wants to improve on over 802.11b: 1) authentication, 2) key management and 3) data transfer. All of these areas were severely lacking in WEP.
2.9 Frequency Range and Channel Allocation for 802.11b As mentioned above, 802.11b networks operate in the ultra high frequency band, specically between 2.4 and 2.5 GHz. Transmission does not take place at a single frequency. There are a total of fourteen channels for use, and the modulation technique spreads the transmission over multiple channels for eective use of the frequency spectrum. Table 2.2 gives the frequency of each channel [24]. The United States uses channels 1-11, Europe uses channels
16
Table 2.2: Frequency and channel assignments.

CHANNEL 1 2 3 4 5 6 7 FREQUENCY 2.412 GHz 2.417 GHz 2.422 GHz 2.427 GHz 2.432 GHz 2.437 GHz 2.442 GHz CHANNEL 8 9 10 11 12 13 14 FREQUENCY 2.447 GHz 2.452 GHz 2.457 GHz 2.462 GHz 2.467 GHz 2.472 GHz 2.484 GHz
1-13, and Japan uses channels 1-14. Interference is maximum between any two adjacent channels. For instance, data transmission and the eect of interference due to adjacent channels can be compared to conversations happening in 11 rooms. Each room has people having dierent conversations. People in room one can hear the conversations in rooms one, two, three, four, and ve. People in room six can hear the conversations of rooms two through ten, but cannot hear anything from rooms one and eleven. People in room eleven can hear the conversations of rooms seven, eight, nine, and ten. For people to have noise free conversations they should be in rooms one, six, and eleven. In 802.11b, there are only three non-overlapping channels, channels one, six, and eleven as shown in Figure 2.2.
17
Figure 2.2: 802.11b channel overlap.
18
CHAPTER 3
STATISTICAL ANALYSIS 3.1 Normal Plot A normal probability plot is a useful graph for assessing whether data comes from a normal distribution. Many statistical procedures make the assumption that the underlying distribution of the data is normal, so this plot can provide some assurance that the assumption of normality is not being violated. Normal plot is a scatterplot of the percentiles of the data versus the percentiles of a population having the normal distribution. If the data comes from a normal population, the resulting points should fall closely along a straight line. If the observations fall along a relatively straight line (using the so-called fat pencil test), then the data is probably normal [25][26]. To further conrm if the distribution is normal a goodness of t test can be used. These plots are produced by using following steps[26]: 1. The observations are ranked from smallest to largest, x1, x2 , ...xn. 2. The ordered observations xj are plotted against their observed cumulative frequency on a graph with the y-axis appropriately scaled for the hypothesized (Normal) distribution. 3. If the hypothesized distribution adequately describes the data, the plotted points fall approximately along a straight line. If the plotted points deviate signicantly from the straight line, then the hypothesized distribution is not appropriate. 4. In assessing the closeness of the points to a straight line, the fat pencil test is often used. If the points are all covered by the imaginary pencil, then the hypothesized (Normal) distribution is likely to be appropriate. 19
3.2 Standard Normal Distribution A useful transformation in statistics is standardization. Sometimes called converting to Z-scores it has the eect of transforming the original distribution to one in which the mean becomes zero and the standard deviation becomes 1. A Z-score quanties the original score in terms of the number of standard deviations that the score is from the mean of the distribution. (3.1) is used to convert from an original or raw score to a Z-score.
Z =
(Raw Score) (Mean) . (Standard Deviation)
(3.1)
The process of converting or transforming scores on a variable to Z-scores is called standardization. Any distribution can be converted to a standardized distribution. However the symmetry of the original distribution remains unchanged. If the original distribution was skewed to start with, it will still be skewed after the Z-score transformation. In the special case where the original distribution can be considered normal, standardizing will result in what is known as the standard normal distribution. The advantage of this is that tables exist in any statistics textbook for the area under the curve for the standard normal distribution. Table 3.1 contains the area under the standard normal curve from 0 to Z. This can be used to compute the cumulative distribution function values for the standard normal distribution.
3.3 Chi-square Goodness-of-Fit Test When an analyst attempts to t a statistical model to observed data, he or she may wonder how well the model actually reects the data. How close are the observed values to those which would be expected under the tted model? One statistical test that addresses this issue is the chi-square goodness-of-t test. The chi-square test is used to test if a sample of data came from a population with a specic distribution. The chi-square goodness-of-t test is applied to binned data. This is actually not a restriction since for non-binned data you
20
Table 3.1: Area under the normal curve from 0 to X.

X 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.8 1.9 2.0 2.1 2.2 2.3 2.4 2.5 2.6 2.7 2.8 2.9 3.0 0.00 0.00000 0.03983 0.07926 0.11791 0.15542 0.19146 0.22575 0.25804 0.28814 0.31594 0.34134 0.36433 0.38493 0.40320 0.41924 0.43319 0.44520 0.45543 0.46407 0.47128 0.47725 0.48214 0.48610 0.48928 0.49180 0.49379 0.49534 0.49653 0.49744 0.49813 0.49865 0.01 0.00399 0.04380 0.08317 0.12172 0.15910 0.19497 0.22907 0.26115 0.29103 0.31859 0.34375 0.36650 0.38686 0.40490 0.42073 0.43448 0.44630 0.45637 0.46485 0.47193 0.47778 0.48257 0.48645 0.48956 0.49202 0.49396 0.49547 0.49664 0.49752 0.49819 0.49869 0.02 0.00798 0.04776 0.08706 0.12552 0.16276 0.19847 0.23237 0.26424 0.29389 0.32121 0.34614 0.36864 0.38877 0.40658 0.42220 0.43574 0.44738 0.45728 0.46562 0.47257 0.47831 0.48300 0.48679 0.48983 0.49224 0.49413 0.49560 0.49674 0.49760 0.49825 0.49874 0.03 0.01197 0.05172 0.09095 0.12930 0.16640 0.20194 0.23565 0.26730 0.29673 0.32381 0.34849 0.37076 0.39065 0.40824 0.42364 0.43699 0.44845 0.45818 0.46638 0.47320 0.47882 0.48341 0.48713 0.49010 0.49245 0.49430 0.49573 0.49683 0.49767 0.49831 0.49878 0.04 0.01595 0.05567 0.09483 0.13307 0.17003 0.20540 0.23891 0.27035 0.29955 0.32639 0.35083 0.37286 0.39251 0.40988 0.42507 0.43822 0.44950 0.45907 0.46712 0.47381 0.47932 0.48382 0.48745 0.49036 0.49266 0.49446 0.49585 0.49693 0.49774 0.49836 0.49882 0.05 0.01994 0.05962 0.09871 0.13683 0.17364 0.20884 0.24215 0.27337 0.30234 0.32894 0.35314 0.37493 0.39435 0.41149 0.42647 0.43943 0.45053 0.45994 0.46784 0.47441 0.47982 0.48422 0.48778 0.49061 0.49286 0.49461 0.49598 0.49702 0.49781 0.49841 0.49886 0.06 0.02392 0.06356 0.10257 0.14058 0.17724 0.21226 0.24537 0.27637 0.30511 0.33147 0.35543 0.37698 0.39617 0.41308 0.42785 0.44062 0.45154 0.46080 0.46856 0.47500 0.48030 0.48461 0.48809 0.49086 0.49305 0.49477 0.49609 0.49711 0.49788 0.49846 0.49889 0.07 0.02790 0.06749 0.10642 0.14431 0.18082 0.21566 0.24857 0.27935 0.30785 0.33398 0.35769 0.37900 0.39796 0.41466 0.42922 0.44179 0.45254 0.46164 0.46926 0.47558 0.48077 0.48500 0.48840 0.49111 0.49324 0.49492 0.49621 0.49720 0.49795 0.49851 0.49893 0.08 0.03188 0.07142 0.11026 0.14803 0.18439 0.21904 0.25175 0.28230 0.31057 0.33646 0.35993 0.38100 0.39973 0.41621 0.43056 0.44295 0.45352 0.46246 0.46995 0.47615 0.48124 0.48537 0.48870 0.49134 0.49343 0.49506 0.49632 0.49728 0.49801 0.49856 0.49896 0.09 0.03586 0.07535 0.11409 0.15173 0.18793 0.22240 0.25490 0.28524 0.31327 0.33891 0.36214 0.38298 0.40147 0.41774 0.43189 0.44408 0.45449 0.46327 0.47062 0.47670 0.48169 0.48574 0.48899 0.49158 0.49361 0.49520 0.49643 0.49736 0.49807 0.49861 0.49900
21
can simply calculate a histogram or frequency table before generating the chi-square test. However, the value of the chi-square test statistic are dependent on how the data is binned. The chi-square test is dened for the hypothesis: H0 : The data follow a specied distribution. Ha : The data do not follow the specied distribution. Test Statistic: For the chi-square goodness-of-t computation, the data are divided into k bins and the test statistic is dened as in
k
X2 =
i=1
(Oi Ei )2 , Ei
(3.2)
where Oi is the observed frequency for bin i and Ei is the expected frequency for bin i. Put simply if the computed test statistic X 2 is large, then the observed and expected values are not close and the model is a poor t to the data.
3.3.1 Deciding what value of X 2 is critical for accepting or rejecting a hypothesis Before understanding how to reject or accept a hypothesis we need to understand two terms which help in this process. Degrees of Freedom: The number of independent pieces of information that go into the estimate of a parameter is called the degrees of freedom (df ) [27]. In general, the degrees of freedom of an estimate is equal to the number of independent scores that go into the estimate minus the number of parameters estimated as intermediate steps or constraints in the estimation of the parameter itself. For example, if the variance, 2, is to be estimated from a random sample of N independent scores, then the degrees of freedom is equal to the number of independent scores (N ) minus the number of parameters estimated as intermediate steps (one, mean) and is therefore equal to N 1. In chi-square test statistic with k data bins, degrees of freedom can be calculated as df = No. of data bins (k) No. of constraints 22 (3.3)
For chi-square test statistic with k data bins the only constraint is No. of observed data points = No. of expected data points Therefore, df for a chi-square test statistic with k data bins is equal to k 1. Level of Signicance or p-value: The p-value or calculated probability is the estimated probability of rejecting the null hypothesis (H0 ) of a study question when that hypothesis is true [27]. The term signicance level is used to refer to a pre-chosen probability and the term p-value is used to indicate a probability that one calculates after a given study. The smaller the p-value, the more strongly the test conrms the null hypothesis. A pvalue of .05 or less conrms the null hypothesis at the 5% level that is, the statistical assumptions used imply that only 5% of the time would the supposed statistical process produce a nding this extreme if the null hypothesis were false. A p < 0.05 is considered statistically signicant and p < 0.001 is considered statistically highly signicant. (3.4)
3.3.2 Finding p-value for Chi-square Test By using (3.2) and (3.3) we can get values for chi-square test statistic X 2 and degrees of freedom df . The nal step is to refer to a professionally prepared table (Table 3.2) of the probabilities of X 2 values. These tables come in a variety of sizes, depending upon how many subdivisions (columns) are present, and how high the degrees of freedom go. The table lists the degrees of freedom as the headings to the rows. Across the top are probability gures - the probability of the Chi-Square. The interior of the table consists of the sum of the X 2 values themselves. How to use Table 3.2: 1. Find the degrees of freedom from the data and look in the left-hand column of the table.
23
2. Scan across the row of X 2 values beside the df number for two values which bracket the calculated X 2 number. This means that one of the gures will be larger, and the other will be smaller. If the table were subdivided into enough columns, one might have found the exact calculated value on the table. Generally, one has to be satised with nding the bracketing numbers. 3. Look up at the top of the table to see which probabilities correspond to the bracketing X 2 values. If the exact X 2 value is not found on this table, its probability would have fallen somewhere between these two.
3.4 Curve Fitting Field data is often accompanied by noise. Even though all control parameters (independent variables) remain constant, the resultant outcomes (dependent variables) vary. A process of quantitatively estimating the trend of the outcomes, also known as regression or curve tting, therefore becomes necessary. The curve tting process ts equations of approximating curves to the raw eld data. Nevertheless, for a given set of data, the tting curves of a given type are generally not unique. Thus, a curve with a minimal deviation from all data points is desired. This besttting curve can be obtained by the method of least squares [26]. For Log-distance Path Loss Model discussed in section 2.5.1 a linear least squares curve tting is used. The least-squares line uses a straight line equation as shown below
Y = P1 X + P2 ,
(3.5)
to approximate the given set of data, (x1 ,y1), (x2,y2), ..., (xn ,yn ), where n >=2. The best tting curve f (x) has the least square error, i.e.,
n n
=
i=1
[yi f (xi )]2 =

i=1
[yi (P1 xi + P2 )]2 .
(3.6)
24
Table 3.2: Critical values of chi-square distribution with degrees of freedom.

Probability of exceeding the critical value df 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 0.995 0.01 0.072 0.207 0.412 0.676 0.989 1.344 1.735 2.156 2.603 3.074 3.565 4.075 4.601 5.142 5.697 6.265 6.844 7.434 8.034 8.643 9.26 9.886 10.52 11.16 11.808 12.461 13.121 13.787 0.99 0.02 0.115 0.297 0.554 0.872 1.239 1.646 2.088 2.558 3.053 3.571 4.107 4.66 5.229 5.812 6.408 7.015 7.633 8.26 8.897 9.542 10.196 10.856 11.524 12.198 12.879 13.565 14.256 14.953 0.975 0.001 0.051 0.216 0.484 0.831 1.237 1.69 2.18 2.7 3.247 3.816 4.404 5.009 5.629 6.262 6.908 7.564 8.231 8.907 9.591 10.283 10.982 11.689 12.401 13.12 13.844 14.573 15.308 16.047 16.791 0.95 0.004 0.103 0.352 0.711 1.145 1.635 2.167 2.733 3.325 3.94 4.575 5.226 5.892 6.571 7.261 7.962 8.672 9.39 10.117 10.851 11.591 12.338 13.091 13.848 14.611 15.379 16.151 16.928 17.708 18.493 0.9 0.016 0.211 0.584 1.064 1.61 2.204 2.833 3.49 4.168 4.865 5.578 6.304 7.042 7.79 8.547 9.312 10.085 10.865 11.651 12.443 13.24 14.041 14.848 15.659 16.473 17.292 18.114 18.939 19.768 20.599 0.1 2.706 4.605 6.251 7.779 9.236 10.645 12.017 13.362 14.684 15.987 17.275 18.549 19.812 21.064 22.307 23.542 24.769 25.989 27.204 28.412 29.615 30.813 32.007 33.196 34.382 35.563 36.741 37.916 39.087 40.256 0.05 3.841 5.991 7.815 9.488 11.07 12.592 14.067 15.507 16.919 18.307 19.675 21.026 22.362 23.685 24.996 26.296 27.587 28.869 30.144 31.41 32.671 33.924 35.172 36.415 37.652 38.885 40.113 41.337 42.557 43.773 0.025 5.024 7.378 9.348 11.143 12.833 14.449 16.013 17.535 19.023 20.483 21.92 23.337 24.736 26.119 27.488 28.845 30.191 31.526 32.852 34.17 35.479 36.781 38.076 39.364 40.646 41.923 43.195 44.461 45.722 46.979 0.01 6.635 9.21 11.345 13.277 15.086 16.812 18.475 20.09 21.666 23.209 24.725 26.217 27.688 29.141 30.578 32 33.409 34.805 36.191 37.566 38.932 40.289 41.638 42.98 44.314 45.642 46.963 48.278 49.588 50.892 0.005 7.879 10.597 12.838 14.86 16.75 18.548 20.278 21.955 23.589 25.188 26.757 28.3 29.819 31.319 32.801 34.267 35.718 37.156 38.582 39.997 41.401 42.796 44.181 45.559 46.928 48.29 49.645 50.993 52.336 53.672
25
The MATLAB curve tting tool box is used to nd the linear t in the discussions to follow in Chapter 5.
26
CHAPTER 4
EXPERIMENTAL SETUP 4.1 Introduction This chapter outlines the experimental procedure involved, the hardware and software used, the propagation environment, dierent scenarios considered, and the research design used for generating an indoor propagation model. The experimental procedure involves identifying dierent measurement scenarios, measurement of signal levels in each scenario, followed by numerical analysis of the data collected.
4.2 Propagation Environment The power received by a mobile receiver is inuenced by the characteristics of the propagation environment. If a generalization study for indoor propagation is to be made, it is essential that we identify the features that inuence the propagation characteristics in dierent buildings. In this section, a brief description is given of the buildings where the measurements were made. Experiments are conducted in two dierent buildings. In the rst building experiments are conducted for four dierent scenarios. The College of Engineering, University of North Texas, Research Park was chosen for conducting the rst set of experiments. The facility is a two oor building. Figure 4.1 gives an overview of the rst oor, Figure 4.2 gives an overview of the second oor, and Figure 4.3 gives a detailed view of the part of the second oor where the experiments will be conducted. The second set of experiments are conducted in General Academic Building (GAB) in the University of North Texas main campus. Signal measurements are conducted for only two scenarios in this building.
27
Figure 4.1: College of Engineering, rst oor overview. 28
Figure 4.2: College of Engineering, second oor overview. 29
Figure 4.3: College of Engineering, second oor detail.
30
Figure 4.4: Building One Closed Corridor. 4.2.1 Description of Measurement Scenarios for Building One Four dierent scenarios are considered for measurements. The scenarios used will help in developing signal loss equations, by which a generalization for propagation in an indoor environment at 2.4 GHz can be obtained. The scenarios are described as follows: Closed Corridor: A closed corridor on the second oor of building one is used for signal measurements. This corridor is closed on both sides with walls. This corridor is 9 high and 87 wide. Signal measurements are taken at every ve feet interval in the middle of the corridor. Figure 4.4 gives a blue print of the corridor and also shows the dierent locations at which measurements are taken. Open corridor: An open corridor on the second oor of building one is used for signal 31
Figure 4.5: Building One Open Corridor. measurements. The corridor is open on one side and closed with a wall on the other side. This corridor is 125 high and 147 wide. Signal measurements are taken at every ve feet interval in the middle of the corridor. Figure 4.5 gives a blue print of the corridor and also shows the dierent locations at which measurements are taken. Class room: A lecture room with furniture and computers is considered for signal measurements. This room is 40 X 30X 9. Signal measurements are taken at every ve feet interval diagonally across the room. Figure 4.6 gives a blue print of the room and also shows the dierent locations at which measurements are taken. Computer lab: A computer lab with furniture and computers is considered for signal measurements. This room is 45 X 30X 9. Signal measurements are taken at every ve feet interval diagonally across the room. Figure 4.7 gives a blue print of the room 32
Figure 4.6: Building One Classroom.
Figure 4.7: Building One Computer Lab.
33
and also shows the dierent locations at which measurements are taken.
4.2.2 Description of Measurement Scenarios for Building Two Two scenarios in building two that are similar to building one scenarios closed corridor and classroom are considered for measurements. The scenarios used will help in comparing the results from building one. The scenarios are described as follows: Closed Corridor: A closed corridor on the third oor of building two is used for signal measurements. This corridor is closed on both sides with walls. This corridor is 89.25 high and 86 wide. Signal measurements are taken at every ve feet interval in the middle of the corridor. Figure 4.8 gives a blue print of the corridor and also shows the dierent locations at which measurements are taken. Class room: A lecture room with furniture and computers is considered for signal measurements. This room is 30 X 24X 89.25. Signal measurements are taken at every ve feet interval diagonally across the room. Figure 4.9 gives a blue print of the room and also shows the dierent locations at which measurements are taken. In all of the above described scenarios the measurements of the two access points are done alternately. A laptop with the wireless client adapter is moved away from the access point at xed distance intervals and the signal strength at each interval is measured by rotating the laptop along its axis.
4.3 Hardware and Software Description The measurements are done using two separate access points. The access points are from dierent manufacturers. One is LinkSys [28], the other is D-Link [29]. Both access points are IEEE 802.11b compliant. The access points operate at 2.4 GHz and provide a bandwidth
34
Figure 4.8: Building Two, Closed Corridor.
Figure 4.9: Building Two, Classroom.
35
of 11 Mbps. A laptop attached with a wireless client adapter is used to measure the signal strength. The client adapter used is manufactured by LinkSys. The signal measurements were done using the software NetStumbler [30] which is a tool for Windows that allows one to measure the signal level of WLANs using 802.11a, 802.11b or 802.11g.
4.4 Data Acquisition Using NetStumbler measurements were taken for the above described scenarios. In each scenario the signal strength was measured for both (LinkSys and D-Link) access points at regular increments of distance. At each interval signal measurements were taken by rotating the laptop twice along its axis.
4.5 Numerical Analysis Numerical analysis is performed to generate a path loss model on the collected data in each scenario. The data is subjected to following steps of numerical analysis. 1. Use Curve tting to evaluate path loss exponent n, in Log-distance Path Loss Model: A scatter plot for signal levels at dierent distance intervals is drawn with distance ratio d/d0 as X-axis and path loss P L as Y -axis on a loglog scale. As discussed in Section 2.5.3, a linear equation of the form (3.5) is generated using MATLAB curve tting tool. The slope of the linear equation generated is equal to 10n where n is the path loss exponent, which indicates the rate at which path loss increases with distance d. Now that we know n (path loss exponent) for a scenario, we will be able to relate distance d with path loss P L as shown in (2.1). 2. Evaluate Standard deviation , for Log-Normal Distribution:
36
Before we evaluate Standard deviation for Log-Normal Distribution we need to prove that the shadowing eect over the range of distance intervals is a normal distribution. This is achieved using Normal Plot, Standard Normal Distribution, and Chi-square goodness-of-t test. Loss dierential can be dened as dierence between mean path loss at each distance interval compared to the measured data at each interval. As discussed in Section 2.10.1 if the scatter plot of Loss dierential is a straight line we can say the distribution is normal. Once the normality of the distribution is established, mean and standard deviation are calculated. 3. Standard Normal Distribution & Chi-Square Goodness-of-Fit Test: Section 3.2 describes standardization of a distribution by converting raw scores into Z-scores using (3.1). Standardizing Loss dierential will result in a standard normal distribution with zero mean and a standard deviation of one. Since the shape of the distribution is not eected by standardization, Chi-square goodness-of-t test is calculated to further verify numerically, if the distribution is normal.
37
CHAPTER 5
NUMERICAL ANALYSIS 5.1 Introduction This chapter describes the numerical analysis for scenarios in Building One as described in Chapter 4. In order to extract useful information from the raw measurement data, data processing is necessary and includes the following steps (Figure 5.1): 1. Calculate the mean signal levels. 2. Using least squares method to calculate curve tting. 3. Use curve tting to evaluate path loss exponent n, in Log-distance Path Loss Model. 4. Using Normal Plot, verify normal distribution. 5. Evaluate standard deviation , for log-normal distribution, and determine standard normal distribution. 6. Calculate Chi-square Goodness-of-Fit Test. 7. Compare with two-ray model. The close-in reference distance d0 , the path loss exponent n, and the standard deviation , statistically describe the path loss model for a specic T-R separation. Chi-square Goodnessof-Fit Test and standard normal distribution are used to quantitatively evaluate the quality of Normal Distribution.
38
Figure 5.1: Numerical Analysis. 39
Figure 5.2: Building One, closed corridor loss for AP1 (D-Link). 5.2 Analysis of Results
5.2.1 Scenario 1: Building One Closed Corridor Figure 5.2 and Figure 5.3 gives the loss (in dBm) versus T-R (Transmitter-Receiver) separation (in meters) for AP1 (D-Link) and AP2 (LinkSys) respectively. The experiment was conducted in a closed corridor at The College of Engineering, University of North Texas, Research Park (Figure 4.4). As seen from Figure 5.2 and Figure 5.3 loss increases as one goes further away from the access point. The laptop used to measure the signal strength is moved at dierent distance intervals and at each distance interval the laptop is rotated twice along its axis. Due to this rotation during measurements, the observed signal strength and loss variation is more than when the laptop is not rotated. 1. Calculate mean signal levels. Mean values of loss at each distance interval is calculated. Figure 5.4 shows a plot between mean loss for both the access points (D-Link and LinkSys) versus T-R separation. As seen from the gure, signal strength deteriorates with increasing T-R
40
Figure 5.3: Building One, closed corridor loss for AP2 (LinkSys). separation. In the remaining part of the numerical analysis an equation is determined which describes the relationship between loss and T-R separation in a closed corridor. 2. Using least squares method to calculate curve tting. MATLAB curve tting tool which is a graphical user interface tool is used to visually explore data and ts as scatter plots. A linear t is calculated using least squares method. Figures 5.5 and 5.6 show the linear curve tting for the data collected using D-Link (AP1) and LinkSys (AP2) access points. Listed below are the output results from curve tting. Closed corridor curve tting for AP1 (D-Link) Linear model Poly1: f(x) = p1 * x + p2 Coecients (with 95 percent condence bounds): p1 = 15.72 (15.68, 15.77) p2 = 0 (xed at bound) 41
Figure 5.4: Building One, closed corridor mean loss for AP1 and AP2. Closed corridor curve tting for AP2 (LinkSys) Linear model Poly1: f(x) = p1 * x + p2 Coecients (with 95 percent condence bounds): p1 = 15.8 (15.76, 15.85) p2 = 0 (xed at bound) 3. Use curve tting to evaluate path loss exponent n, in Log-distance Path Loss Model. Curve tting gives a linear equation of the form (3.5) which when compared to (2.2) helps in determining the value for path loss exponent n. From Figures 5.5 and 5.6 and output results listed above we get the slope of the curve to be 15.72 for D-Link and 15.8 for LinkSys. From (2.2), the slope of the tted curve is 10n, i.e., calculated path loss exponents (n) for D-link and LinkSys access points in closed corridor is 1.572 and 1.58. The next step is to evaluate Log normal shadowing. Before evaluating Log normal shadowing, normal probability plot is used to verify if the distribution is normal. 42
Figure 5.5: Building One, closed corridor curve tting for AP1 (D-Link).
Figure 5.6: Building One, closed corridor curve tting for AP2 (LinkSys).
43
Figure 5.7: Building One, closed corridor normal probability plot for AP1 (D-Link). 4. Using Normal Plot, verify normal distribution. In this part of the analysis we calculate the loss dierential, which is a measure of variation or scatter of loss compared to the mean value of loss calculated above. Normal probability plot is used to verify if the variation of loss (loss dierential) is distributed normally. The normal plots from Figure 5.7 and Figure 5.8 indicate that the assumption of normality is not unreasonable. Both plots pass the fat pencil test. 5. Evaluate standard deviation , for log-normal distribution, and determine standard normal distribution. Mean , and standard deviation of loss dierential is calculated for both the access points. Mean is calculated to be 0 and standard deviation for D-Link access point is 3.9849 and for LinkSys standard deviation is 4.022. Using (3.1) loss dierential is standardized. Figure 5.9 and Figure 5.10 show the plots for standard normal distributions for AP1 and AP2. 6. Calculate Chi-square Goodness-of-Fit Test. As discussed in section 3.3 to evaluate goodness of t we need to calculate test statistic 44
Figure 5.8: Building One, closed corridor normal probability plot for AP2 (LinkSys).
Figure 5.9: Building One, closed corridor standard normal distribution for AP1 (D-Link).
45
Observed 0 0 1 6 23 91 277 613 1147 1692 2019 1854 1215 654 277 100 25 13 2 0 0 10009
Chi Statistic 0.010009 0.10009 0.0616 0.24974 0.045981 0.025404 0.0091947 2.8824 3.3688 1.8087 0.93984 6.399 0.014095 9.52E-03 0.0091947 0.60252 0.037395 13.424 1.9043 0.10009 0.010009 -
46
Observed 1 0 1 8 22 92 260 611 1139 1710 2043 1859 1217 637 268 94 27 13 5 1 1 10009
Chi Statistic 97.92 0.10009 0.0616 1.9706 0.17501 0.0030725 1.2418 3.1535 4.2656 0.83609 2.2782 7.0183 0.031045 0.57923 0.40334 0.023251 0.36143 13.424 22.803 8.0911 97.92 -
47
Figure 5.10: Building One, closed corridor standard normal distribution for AP2 (LinkSys). X 2 and degrees of freedom df . Table 5.1 and 5.2 are used for calculating Chi-square test statistic X 2 for AP1 and AP2. To nd the expected value for a particular bin, the area under standard normal curve for the bin is multiplied by total observations. Chi-square statistic is calculated using (3.2). To evaluate if the distribution is normal the bins that are used to calculate X 2 are from -3.25 to 3.25, a total of 13 bins. As discussed in section 3.3, the degrees of freedom df = 13 - 1 = 12. From Table 5.1 X 2 for AP1 (D-Link)= 16.152, and from Table 5.2 X 2 for AP2 (LinkSys)= 20.3699 By using Table 3.2 we can identify to p-value associated with X 2 and df values. We now have our Chi-square statistic X 2 for AP1 and AP2 (16.152 and 20.3699), our predetermined level of signicance (0.05), and our degrees of freedom (df =12). Finding the Chi-square distribution table with 12 degree of freedom and reading along the row we nd our value of X 2 for AP1 = 16.152, lies between 6.304 and 18.549. The corresponding probability is 0.9 < p < 0.1. This is below the conventionally accepted signicance level of 0.05 or 5%, so the null hypothesis that loss dierential is normal is veried for AP1 (D-Link). 48
Figure 5.11: Comparison of closed corridor mean loss with two-ray model. Finding the Chi-square distribution with 12 degree of freedom and reading along the row we nd our value of X 2 for AP2 = 20.3699, lies between 18.549 and 21.026. The corresponding probability is 0.10 < p < 0.05. This is below the conventionally accepted signicance level of 0.05 or 5%, so the null hypothesis that loss dierential is normal is veried for AP2 (LinkSys). 7. Compare with two-ray model. A graphical comparison is made between measured data and the two-ray model. Figure 5.11 and Figure 5.12 show the comparison between the theoretical two-ray model and measured data. It can be seen from these plots that signal deteriorates at a faster rate in two-ray model than in actual measurements.
49
Figure 5.12: Comparison of closed corridor AP1 and AP2 loss with mean signal and two-ray model. 5.2.2 Scenario 2: Building One Open Corridor Figure 5.13 and Figure 5.14 gives the loss (in dBm) versus T-R (Transmitter-Receiver) separation (in meters) for AP1 (D-Link) and AP2 (LinkSys) respectively. The experiment was conducted in an open corridor at The College of Engineering, University of North Texas, Research Park (Figure 4.5). As seen from Figure 5.13 and Figure 5.14 loss increases with distance relative to the access point. 1. Calculate mean signal levels. Mean values of loss at each distance interval is calculated. Figure 5.15 shows a plot between mean loss for both the access points (D-Link and LinkSys) versus T-R separation. As seen from the gure, signal strength deteriorates with increasing T-R separation. In the remaining part of the numerical analysis an equation is determined which describes the relationship between loss and T-R separation in an open corridor. 2. Using least squares method to calculate curve tting. Using MATLAB curve tting tool, a linear t is calculated using least squares method. 50
Figure 5.13: Building One, open corridor loss for AP1 (D-Link).
Figure 5.14: Building One, open corridor loss for AP2 (LinkSys).
51
Figure 5.15: Building One, open corridor mean loss for AP1 and AP2. Figures 5.16 and 5.17 show the linear curve tting for the data collected using D-Link (AP1) and LinkSys (AP2) access points. Listed below are the output results from curve tting. Open corridor curve tting for AP1 (D-Link). Linear model Poly1: f(x) = p1 * x + p2 Coecients (with 95 percent condence bounds): p1 = 16.88 (16.84, 16.92) p2 = 0 (xed at bound) Open corridor curve tting for AP2 (LinkSys). Linear model Poly1: f(x) = p1 * x + p2 Coecients (with 95 percent condence bounds):
52
Figure 5.16: Building One, open corridor curve tting for AP1 (D-Link). p1 = 16.3 (16.26, 16.34) p2 = 0 (xed at bound) 3. Use curve tting to evaluate path loss exponent n, in Log-distance Path Loss Model. From Figures 5.16 and 5.17 and output results listed above we get the slope of the tted curve to be 16.88 for D-Link and 16.3 for LinkSys. From (2.2), the slope of the tted curve is 10n, i.e., calculated path loss exponents (n) for D-link and LinkSys access points in open corridor is 1.688 and 1.63. 4. Using Normal Plot, verify normal distribution. As discussed in Scenario 1 (Closed corridor), normal probability plot is used to verify if the variation of loss (loss dierential) is distributed normally. The normal plots from Figure 5.18 and Figure 5.19 indicate that the assumption of normality is not unreasonable. Both plots pass the fat pencil test. 5. Evaluate standard deviation , for log-normal distribution, and determine standard normal distribution.
53
Figure 5.17: Building One, open corridor curve tting for AP2 (LinkSys).
Figure 5.18: Building One, open corridor normal probability plot for AP1 (D-Link).
54
Figure 5.19: Building One, open corridor normal probability plot for AP2 (LinkSys). Mean , and standard deviation of loss dierential is calculated for both the access points. Mean is calculated to be 0 and standard deviation for D-Link access point is 3.5773 and for LinkSys standard deviation is 3.2642. Using (3.1) loss dierential is standardized. Figure 5.20 and Figure 5.21 show the plots for standard normal distributions for AP1 and AP2. 6. Calculate Chi-square Goodness-of-Fit Test As discussed in section 3.3 to evaluate goodness of t we need to calculate test statistic X 2 and degrees of freedom df . Table 5.3 and 5.4 are used for calculating Chi-square test statistic X 2 for AP1 and AP2. To nd the expected value for a particular bin, the area under standard normal curve for the bin is multiplied by total observations. Chi-square test statistic is calculated using (3.2). To evaluate if the distribution is normal, the bins that are used to calculate X 2 are from -3.25 to 3.25, a total of 13 bins. As discussed in section 3.3, degrees of freedom df = 13 - 1 = 12. From Table 5.3 X 2 for AP1 (D-Link)= 20.0012, and from Table 5.4 X 2 for AP2 (LinkSys)= 37.6687 By using Table 3.2 we can identify to p-value associated with
55
Figure 5.20: Building One, open corridor standard normal distribution for AP1 (D-Link).
Figure 5.21: Building One, open corridor standard normal distribution for AP2 (LinkSys).
56
Table 5.3: Open corridor X 2 table for AP1 (D-Link).
Observed 0 0 1 12 26 81 233 582 1023 1538 1884 1683 1098 599 250 88 22 12 1 0 0 9133
Chi Statistic 0.009133 0.09133 0.11613 12.709 0.74864 0.13971 1.7708 0.48486 6.0697 2.0528 3.6414 4.8297 0.042991 3.03E-06 0.069954 0.15056 0.00012994 12.709 0.11613 0.09133 0.009133 -
57
Table 5.4: Open corridor X 2 table for AP2 (LinkSys).
Observed 0 1 1 6 21 76 260 556 1061 1544 1821 1759 1020 617 263 89 33 5 0 0 0 9133
Chi Statistic 0.009133 9.0406 0.11613 0.52687 0.040829 0.84257 0.13155 3.0927 1.7436 1.6449 0.18024 16.814 6.5225 0.53831 0.30344 0.24685 5.567 0.063841 0.71237 0.09133 0.009133 -
58
X 2 and df values. We now have our chi square statistic X 2 for AP1 and AP2 (20.0012 and 37.6687), our predetermined level of signicance (0.05), and our degrees of freedom (df = 12). Finding the Chi-square distribution with 12 degree of freedom and reading along the row we nd our value of X 2 for AP1 = 20.0012, lies between 18.549 and 21.026. The corresponding probability is 0.10 < p < 0.05. This is below the conventionally accepted signicance level of 0.05 or 5%, so the null hypothesis that loss dierential is normal is veried for AP1 (D-Link). Finding the Chi-square distribution with 12 degree of freedom and reading along the row we nd our value of X 2 for AP2 = 37.6687, lies beyond 28.3. So the corresponding probability p-value is less that 0.005. This is not an acceptable signicance level. 7. Compare with two-ray model. A graphical comparison is made between measured data and two-ray model. Figure 5.22 and Figure 5.23 show the comparison between the theoretical two-ray model and measured data. It can be seen from these plots that signal strength variation is similar in two-ray model and open corridor measurements.
5.2.3 Scenario 3: Building One Classroom Figure 5.24 and Figure 5.25 gives the loss (in dBm) versus T-R (Transmitter-Receiver) separation (in meters) for AP1 (D-Link) and AP2 (LinkSys) respectively. The experiment was conducted in an classroom at The College of Engineering, University of North Texas, Research Park (Figure 4.6). As seen from Figure 5.24 and Figure 5.25 loss increases as one goes further away from the access point. 1. Calculate mean signal levels.
59
Figure 5.22: Comparison of open corridor mean loss with two-ray model.
Figure 5.23: Comparison of open corridor AP1 and AP2 loss with mean signal and two-ray model.
60
Figure 5.24: Building One, classroom loss for AP1 (D-Link).
Figure 5.25: Building One, classroom loss for AP2 (LinkSys).
61
Figure 5.26: Building One, classroom mean loss for AP1 and AP2. Mean values of loss at each distance interval is calculated. Figure 5.26 shows a plot between mean loss for both the access points (D-Link and LinkSys) versus T-R separation. As seen from the gure, signal strength deteriorates with increasing T-R separation. In the remaining part of the numerical analysis an equation is determined which describes the relationship between loss and T-R separation in a classroom. 2. Using least squares method to calculate curve tting. Figures 5.27 and 5.28 show the linear curve tting for the data collected using D-Link (AP1) and LinkSys (AP2) access points. Listed below are the output results from curve tting. Classroom curve tting for AP1 (D-Link). Linear model Poly1: f(x) = p1 * x + p2 Coecients (with 95 percent condence bounds): p1 = 12.58 (12.44, 12.72) 62
Figure 5.27: Building One, classroom curve tting for AP1 (D-Link). p2 = 0 (xed at bound) Classroom curve tting for AP2 (LinkSys). Linear model Poly1: f(x) = p1 * x + p2 Coecients (with 95 percent condence bounds): p1 = 12.63 (12.49, 12.78) p2 = 0 (xed at bound) 3. Use curve tting to evaluate path loss exponent n, in Log-distance Path Loss Model. From Figures 5.27 and 5.28 and output results listed above we get the slope of the tted curve to be 12.58 for D-Link and 12.63 for LinkSys. From (2.2), the slope of the tted curve is 10n, i.e., calculated path loss exponents (n) for D-link and LinkSys access points in a classroom is 1.258 and 1.263. 4. Using Normal Plot, verify normal distribution.
63
Figure 5.28: Building One, classroom curve tting for AP2 (LinkSys). As discussed in Scenario 1 (Closed corridor), normal probability plot is used to verify if the variation of loss (loss dierential) is distributed normally. The normal plots from Figure 5.29 and Figure 5.30 indicate that the assumption of normality is not unreasonable. Both plots pass the fat pencil test. 5. Evaluate standard deviation , for log-normal distribution, and determine standard normal distribution. Mean , and standard deviation of loss dierential is calculated for both the access points. Mean is calculated to be 0 and standard deviation for D-Link access point is 3.7607 and for LinkSys standard deviation is 4.053. Using (3.1) loss dierential is standardized. Figure 5.31 and Figure 5.32 show the plots for standard normal distributions for AP1 and AP2. 6. Calculate Chi-square Goodness-of-Fit Test. Table 5.5 and 5.6 are used for calculating Chi-square test statistic X 2 for AP1 and AP2. To nd the expected value for a particular bin, the area under standard normal curve for the bin is multiplied by total observations. Chi-square test statistic is calculated 64
Figure 5.29: Building One, classroom normal probability plot for AP1 (D-Link).
Figure 5.30: Building One, classroom normal probability plot for AP2 (LinkSys).
65
Figure 5.31: Building One, classroom standard normal distribution for AP1 (D-Link).
Figure 5.32: Building One, classroom standard normal distribution for AP2 (LinkSys).
66
Observed 0 2 2 6 15 50 142 320 590 926 1103 987 605 324 137 44 16 2 3 2 0 5276
Chi Statistic 0.005276 71.868 6.1314 4.5336 0.42519 0.030684 0.16067 1.9622 3.6519 0.021606 3.6253 4.6502 1.7352 1.41E+00 0.66166 0.46777 0.87032 0.13037 16.281 71.868 0.005276 -
67
Observed 0 0 0 1 15 46 149 333 631 916 1064 981 580 351 125 58 20 6 0 0 0 5276
Chi Statistic 0.005276 0.05276 0.41153 0.96757 0.42519 0.15806 0.031258 0.49273 0.083032 0.033279 0.48385 3.8368 5.3214 0.070573 3.2531 1.7441 4.2284 4.5336 0.41153 0.05276 0.005276 -
68
using (3.2). To evaluate if the distribution is normal, the bins that are used to calculate X 2 are from -3.25 to 3.25, a total of 13 bins. As discussed in section 3.3, degrees of freedom df = 13 - 1 = 12. From Table 5.5 X 2 for AP1 (D-Link)= 19.6687, and from Table 5.6 X 2 for AP2 (LinkSys)= 20.1618 By using Table 3.2 we can identify to p-value associated with X 2 and df values. We now have our chi-square test statistic X 2 for AP1 and AP2 (19.6687 and 20.1618), our predetermined level of signicance (0.05), and our degrees of freedom (df =12). Finding the Chi-square distribution with 12 degree of freedom and reading along the row we nd our value of X 2 for AP1 = 19.6687, lies between 18.549 and 21.026. The corresponding probability is 0.10 < p < 0.05. This is below the conventionally accepted signicance level of 0.05 or 5%, so the null hypothesis that loss dierential is normal is veried for AP1 (D-Link). Finding the Chi square distribution table with 12 degree of freedom and reading along the row we nd our value of X 2 for AP2 = 20.1618, lies between 18.549 and 21.026. The corresponding probability is 0.10 < p < 0.05. This is below the conventionally accepted signicance level of 0.05 or 5%, so the null hypothesis that loss dierential is normal is veried for AP2 (LinkSys). 7. Compare with two-ray model. A graphical comparison is made between measured data and two-ray model. Figure 5.33 and Figure 5.34 show the comparison between the theoretical two-ray model and measured data. It can be seen from these plots that signal strength deteriorates much higher in two-ray model when compared to classroom measurements.
69
Figure 5.33: Comparison of classroom mean loss with two ray-model.
70
Figure 5.35: Building One, computer lab loss for AP1 (D-Link). 5.2.4 Scenario 4: Building One Computer Lab Figure 5.35 and Figure 5.36 gives the loss (in dBm) versus T-R (Transmitter-Receiver) separation (in meters) for AP1 (D-Link) and AP2 (LinkSys) respectively. The experiment was conducted in an computer lab at The College of Engineering, University of North Texas, Research Park (Figure 4.7). As seen from Figure 5.35 and Figure 5.36 loss increases as one goes further away from the access point. 1. Calculate mean signal levels. Mean values of loss at each distance interval is calculated. Figure 5.37 shows a plot between mean loss for both the access points (D-Link and LinkSys) versus T-R separation. As seen from the gure, signal strength deteriorates with increasing T-R separation. In the remaining part of the numerical analysis an equation is determined which describes the relationship between loss and T-R separation in a computer lab. 2. Using least squares method to calculate curve tting. Using MATLAB curve tting tool, a linear t is calculated using least squares method. Figures 5.38 and 5.39 show the linear curve tting for the data collected using D-Link 71
Figure 5.36: Building One, computer lab loss for AP2 (LinkSys).
Figure 5.37: Building One, computer lab mean loss for AP1 and AP2.
72
(AP1) and LinkSys (AP2) access points. Listed below are the output results from curve tting. Computer lab curve tting for AP1(D-Link) Linear model Poly1: f(x) = p1 * x + p2 Coecients (with 95 percent condence bounds): p1 = 14.47 (14.29, 14.65) p2 = 0 (xed at bound) Computer lab curve tting for AP2(LinkSys) Linear model Poly1: f(x) = p1 * x + p2 Coecients (with 95 percent condence bounds): p1 = 14.28 (14.09, 14.47) p2 = 0 (xed at bound) 3. Use curve tting to evaluate path loss exponent n, in Log-distance Path Loss Model. From Figures 5.38 and 5.39 and output results listed above we get the slope of the tted curve to be 14.47 for D-Link and 14.28 for LinkSys. From (2.2), the slope of the tted curve is 10n, i.e. calculated path loss exponents (n) for D-link and LinkSys access points in a computer lab is 1.447 and 1.428. 4. Using Normal Plot, verify normal distribution. As discussed in Scenario 1 (Closed corridor), normal probability plot is used to verify if the variation of loss (loss dierential) is distributed normally. The normal plots 73
Figure 5.38: Building One, computer lab curve tting for AP1 (D-Link).
Figure 5.39: Building One, computer lab curve tting for AP2 (LinkSys).
74
Figure 5.40: Building One, computer lab normal probability plot for AP1 (D-Link). from Figure 5.40 and Figure 5.41 indicate that the assumption of normality is not unreasonable. Both plots pass the fat pencil test. 5. Evaluate standard deviation , for log-normal distribution, and determine standard normal distribution. Mean , and standard deviation of loss dierential is calculated for both the access points. Mean is calculated to be 0 and standard deviation for D-Link access point is 3.7049 and for LinkSys standard deviation is 3.8460. Using (3.1) loss dierential is standardized. Figure 5.42 and Figure 5.43 show the plots for standard normal distributions for AP1 and AP2. 6. Calculate Chi-square Goodness-of-Fit Test. As discussed in section 3.3 to evaluate goodness of t we need to calculate test statistic X 2 and degrees of freedom df . Table 5.7 and 5.8 are used for calculating Chi-square test statistic X 2 for AP1 and AP2. To nd the expected value for a particular bin, the area under standard normal curve for the bin is multiplied by total observations. Chi-square test statistic is calculated using (3.2). To evaluate if the distribution is 75
Figure 5.41: Building One, computer lab normal probability plot for AP2 (LinkSys).
Figure 5.42: Building One, computer lab standard normal distribution for AP1 (D-Link).
76
Table 5.7: Computer lab X 2 table for AP1 (D-Link).
Observed 0 0 0 1 10 19 74 165 308 452 558 496 293 175 63 32 5 4 2 0 0 2657
Chi Statistic 0.002657 0.02657 0.20725 0.068934 2.047 1.2603 2.43E-05 0.49365 0.56183 0.31483 2.1362 2.1944 2.516 3.01E-03 1.6235 2.251 0.30034 5.6139 15.508 0.02657 0.002657 -
77
Table 5.8: Computer lab X 2 table for AP2 (LinkSys).
Observed 0 0 0 0 6 18 83 167 309 468 537 496 287 170 69 33 12 1 0 0 1 2657
Chi Statistic 0.002657 0.02657 0.20725 1.2993 0.023188 1.754 1.1056 0.30371 0.48133 0.032983 0.29663 2.1944 3.6897 0.10488 0.33232 2.8972 4.9384 0.068934 0.20725 0.02657 374.37 -
78
Figure 5.43: Building One, computer lab standard normal distribution for AP2 (LinkSys). normal, the bins that are used to calculate X 2 are from -3.25 to 3.25, a total of 13 bins. As discussed in section 3.3, degrees of freedom df = 13 - 1 = 12. From Table 5.7 X 2 for AP1 (D-Link)= 15.7022, and from Table 5.8 X 2 for AP2 (LinkSys)= 18.1544 By using Table 3.2 we can identify to p-value associated with X 2 and df values. We now have our chi-square test statistic X 2 for AP1 and AP2 (15.7022 and 18.1544), our predetermined level of signicance (0.05), and our degrees of freedom (df =12). Finding the Chi-square distribution with 12 degree of freedom and reading along the row we nd our value of X 2 for AP1 = 15.7022, lies between 6.304 and 18.549. The corresponding probability is 0.9 < p < 0.1. This is below the conventionally accepted signicance level of 0.05 or 5%, so the null hypothesis that loss dierential is normal is veried for AP1 (D-Link). Finding the Chi-square distribution with 12 degree of freedom and reading along the row we nd our value of X 2 for AP2 = 18.1544, lies between 6.304 and 18.549. The corresponding probability is 0.9 < p < 0.1. This is below the conventionally accepted
79
Figure 5.44: Comparison of computer lab mean loss with two ray model. signicance level of 0.05 or 5%, so the null hypothesis that loss dierential is normal is veried for AP2 (LinkSys). 7. Compare with two-ray model. A graphical comparison is made between measured data and two-ray model. Figure 5.44 and Figure 5.45 show the comparison between the theoretical two-ray model and measured data. It can be seen from these plots that signal strength deteriorates much higher in two-ray model when compared to computer lab measurements.
5.3 Summary of Results In this chapter a numerical analysis is performed on measurements conducted in Building One. Figure 5.46 lists the experimental results for all the scenarios from Building One. For Building One Closed Corridor, Path loss exponent (n) is 1.572 for AP1 and 1.58 for AP2, Standard deviation () is 3.9849 for AP1 and 4.022 for AP2. Now the Chi-square test statistic (X 2 ) values for both AP1 and AP2 (16.152 for AP1 and 20.3699 for AP2) are acceptable values to prove that the observed fading is a normal distribution at 5% signicance
80
Figure 5.45: Comparison of computer lab AP1 and AP2 loss with mean signal and two ray model. level. So, for Building One Closed Corridor Alternative Hypotheses is accepted for the rst research question. For Building One Open Corridor, Path loss exponent (n) is 1.688 for AP1 and 1.63 for AP2, Standard deviation () is 3.5773 for AP1 and 3.2642 for AP2. Now the Chi-square test statistic (X 2 ) value for AP1 (20.0012) is acceptable value to prove that the observed fading is a normal distribution at 5% signicance level but for AP2 Chi-square test statistic (X 2 ) value of 37.6687 is not an acceptable value to prove that the observed fading is a normal distribution at 5% signicance level. This is an exception in this study. So, for Building One Open Corridor Alternative Hypotheses is accepted for the rst research question. For Building One Classroom, Path loss exponent (n) is 1.258 for AP1 and 1.263 for AP2, Standard deviation () is 3.7607 for AP1 and 4.053 for AP2. Now the Chi-square test statistic (X 2 ) values for both AP1 and AP2 (19.6687 for AP1 and 20.1618 for AP2) are acceptable values to prove that the observed fading is a normal distribution at 5% signicance level. So, for Building One Classroom Alternative Hypotheses is accepted for the rst research question.
81
Figure 5.46: Results at a glance. For Building One Computer lab, Path loss exponent (n) is 1.447 for AP1 and 1.428 for AP2, Standard deviation () is 3.7049 for AP1 and 3.846 for AP2. Now the Chi-square test statistic (X 2 ) values for both AP1 and AP2 (15.7022 for AP1 and 18.1544 for AP2) are acceptable values to prove that the observed fading is a normal distribution at 5% signicance level. So, for Building One Computer lab Alternative Hypotheses is accepted for the rst research question. Varying values for Path loss exponent (n) and Standard deviation () for dierent scenarios signies that the path loss and multi path fading vary from scenario to scenario. So, for Building One Alternative Hypotheses is accepted for the second research question.
82
CHAPTER 6
COMPARISON WITH SIMILAR SCENARIOS 6.1 Introduction This chapter describes the numerical analysis for scenarios in Building Two as described in Chapter 4 and compares the results with similar scenarios in Building One.
6.2 Scenario 1: Building Two Closed Corridor Figure 6.1 and Figure 6.2 give the loss (in dBm) versus T-R (Transmitter-Receiver) separation (in meters) for AP1 (D-Link) and AP2 (LinkSys) respectively. The experiment was conducted in a closed corridor at General Academic Building (GAB) in the University of North Texas main campus. As seen from Figure 6.1 and Figure 6.2 loss increases as one goes further away from the access point. 1. Calculate mean signal levels. Mean values of loss at each distance interval is calculated. Figure 6.3 shows a plot between mean loss for both the access points (D-Link and LinkSys) versus T-R separation. As seen from the gure, signal strength deteriorates with increasing T-R separation. In the remaining part of the numerical analysis an equation is determined which describes the relationship between loss and T-R separation in a closed corridor for Building Two. 2. Using least squares method to calculate curve tting. Using MATLAB curve tting tool a linear t is calculated using least squares method. Figures 6.4 and 6.5 show the linear curve tting for the data collected using D-Link
83
Figure 6.1: Building Two, closed corridor loss for AP1 (D-Link).
Figure 6.2: Building Two, closed corridor loss for AP2 (LinkSys).
84
Figure 6.3: Building Two, closed corridor mean loss for AP1 and AP2. (AP1) and LinkSys (AP2) access points. Listed below are the output results from curve tting. Closed corridor curve tting for AP1 (D-Link). Linear model Poly1: f(x) = p1 * x + p2 Coecients (with 95 percent condence bounds): p1 = 16.66 (16.56, 16.76) p2 = 0 (xed at bound) Closed corridor curve tting for AP2 (LinkSys). Linear model Poly1: f(x) = p1 * x + p2 Coecients (with 95 percent condence bounds): p1 = 16.48 (16.38, 16.58) 85
Figure 6.4: Building Two, closed corridor curve tting for AP1 (D-Link). p2 = 0 (xed at bound) 3. Use curve tting to evaluate path loss exponent n, in Log-distance Path Loss Model. From Figures 6.4 and 6.5 and output results listed above we get the slope of the tted curve to be 16.66 for D-Link and 16.48 for LinkSys. From (2.2), the slope of the tted curve is 10n, i.e. calculated path loss exponents (n) for D-link and LinkSys access points in a closed corridor for Building Two is 1.666 and 1.648. From Section 5.2.1 the calculated path loss exponents for D-Link and LinkSys access points in a closed corridor for Building One are 1.572 and 1.58. It can be said that signal strength deteriorates at a faster rate in Building Two. Even though the width of closed corridors in both Building One and Building Two are almost the same, the dierence in path loss exponents can be attributed to dierences in material used for construction and variation of multi-path caused by height of corridors. 4. Using Normal Plot, verify normal distribution. Normal probability plot is used to verify if the variation of loss (loss dierential) is distributed normally. The normal plots from Figure 6.6 and Figure 6.7 indicate that 86
Figure 6.5: Building Two, closed corridor curve tting for AP2 (LinkSys). the assumption of normality is not unreasonable. Both plots pass the fat pencil test. 5. Evaluate standard deviation , for log-normal distribution, and determine standard normal distribution. Mean , and standard deviation of loss dierential is calculated for both the access points. Mean is calculated to be 0 and standard deviation for D-Link access point is 3.9275 and for LinkSys standard deviation is 3.7699. Using (3.1) loss dierential is standardized. Figure 6.8 and Figure 6.9 show the plots for standard normal distributions for AP1 and AP2. 6. Calculate Chi-square Goodness-of-Fit Test. As discussed in section 3.3 to evaluate goodness of t we need to calculate test statistic X 2 and degrees of freedom df . Table 6.1 and 6.2 are used for calculating Chi-square test statistic X 2 for AP1 and AP2. To evaluate if the distribution is normal, the bins that are used to calculate X 2 are from -3.25 to 3.25, a total of 13 bins. As discussed in section 3.3, degrees of freedom df = 13 - 1 = 12. From Table 6.1 X 2 for AP1 (D-Link)= 11.7578, and from Table 6.2 X 2 for AP2 87
Figure 6.6: Building Two, closed corridor normal probability plot for AP1 (D-Link).
Figure 6.7: Building Two, closed corridor normal probability plot for AP2 (LinkSys).
88
Figure 6.8: Building Two, closed corridor standard normal distribution for AP1 (D-Link).
Figure 6.9: Building Two, closed corridor standard normal distribution for AP2 (LinkSys).
89
Observed 0 0 0 3 14 49 140 309 582 908 1069 918 618 331 138 59 13 4 1 0 0 5156
Chi Statistic 0.005156 0.05156 0.40217 0.090894 0.20925 0.037265 0.0862 2.519 2.7961 0.061169 2.5693 0.33704 0.053237 1.53E-01 0.2121 2.6943 0.030046 0.86726 0.88869 0.05156 0.005156 -
90
Observed 0 0 0 4 16 48 148 302 577 899 1069 927 609 355 136 50 12 4 0 0 0 5156
Chi Statistic 0.005156 0.05156 0.40217 0.86726 1.0519 0.0023232 0.14002 3.8722 3.5057 0.0027646 2.5693 0.7752 0.34939 0.83584 0.39374 0.11416 0.012268 0.86726 0.40217 0.05156 0.005156 -
91
Table 6.3: Closed corridor results from Building One and Building Two.
(LinkSys)= 13.6248. By using Table 3.2 we can identify to p-value associated with X 2 and df values. We now have our chi-square test statistic X 2 for AP1 and AP2 (11.7578 and 13.6248), our predetermined level of signicance (0.05), and our degrees of freedom (df =12). From Table 3.2 we can say X 2 for AP1 = 11.7578, lies between 6.304 and 18.549. The corresponding probability is 0.9 < p < 0.1. This is below the conventionally accepted signicance level of 0.05 or 5%, so the null hypothesis that loss dierential is normal is veried for AP1 (D-Link). Now for AP2 our value of X 2 = 13.6248, lies between 6.304 and 18.549. The corresponding probability is 0.9 < p < 0.1. This is below the conventionally accepted signicance level of 0.05 or 5%, so the null hypothesis that loss dierential is normal is veried for AP2 (LinkSys). Table 6.3 lists the results for AP1 and AP2 from Building One and Building Two. The dierence in path loss exponents for a closed corridor in Building One and Building Two is about 5%. So, it can be said that path loss models from Building Two validate the path loss models developed in Building One. 7. Compare with two-ray model.
92
Figure 6.10: Comparison of closed corridor mean loss with two-ray model. A graphical comparison is made between measured data and two-ray model. Figure 6.10 and Figure 6.11 show the comparison between the theoretical two-ray model and measured data. It can be seen from these plots that signal strength deteriorates at a faster rate in two-ray model than in actual measurements.
6.3 Scenario 2: Building Two Classroom Figure 6.12 and Figure 6.13 gives the loss (in dBm) versus T-R (Transmitter-Receiver) separation (in meters) for AP1 (D-Link) and AP2 (LinkSys) respectively. The experiment was conducted in an classroom at General Academic Building (GAB) in the University of North Texas main campus. As seen from Figure 6.12 and Figure 6.13 loss increases with the distance from the access point. 1. Calculate mean signal levels. Mean values of loss at each distance interval is calculated. Figure 6.14 shows a plot between mean loss for both the access points (D-Link and LinkSys) versus T-R separation. As seen from the gure, signal strength deteriorates with increasing T-R
93
Figure 6.11: Comparison of closed corridor AP1 and AP2 loss with mean signal and two-ray model.
Figure 6.12: Building Two, classroom loss for AP1 (D-Link).
94
Figure 6.13: Building Two, classroom loss for AP2 (LinkSys). separation. In the remaining part of the numerical analysis an equation is determined which describes the relationship between loss and T-R separation in a classroom for Building Two. 2. Using least squares method to calculate curve tting. Using MATLAB curve tting tool a linear t is calculated using least squares method. Figures 6.15 and 6.16 show the linear curve tting for the data collected using D-Link (AP1) and LinkSys (AP2) access points. Listed below are the output results from curve tting. Classroom curve tting for AP1 (D-Link). Linear model Poly1: f(x) = p1 * x + p2 Coecients (with 95 percent condence bounds): p1 = 13 (12.74, 13.26) p2 = 0 (xed at bound) 95
Figure 6.14: Building Two, classroom mean loss for AP1 and AP2. Classroom curve tting for AP2 (LinkSys). Linear model Poly1: f(x) = p1 * x + p2 Coecients (with 95 percent condence bounds): p1 = 12.98 (12.72, 13.25) p2 = 0 (xed at bound) 3. Use curve tting to evaluate path loss exponent n, in Log-distance Path Loss Model. From Figures 6.15 and 6.16 and output results listed above we get the slope of the tted curve to be 13.0 for D-Link and 12.98 for LinkSys. From (2.2), the slope of the tted curve is 10n, i.e. calculated path loss exponents (n) for D-link and LinkSys access points in a classroom is 1.3 and 1.298. From Section 5.2.1 the calculated path loss exponents for D-Link and LinkSys access points in a classroom for Building One are 1.258 and 1.263. It can be said that signal strength deteriorates at a faster rate in Building Two. Even though the width of dimensions of classrooms in both Building 96
Figure 6.15: Building Two, classroom curve tting for AP1 (D-Link).
Figure 6.16: Building Two, classroom curve tting for AP2 (LinkSys).
97
Figure 6.17: Building Two, classroom normal probability plot for AP1 (D-Link). One and Building Two are almost the same, the dierence in path loss exponents can be attributed to dierences in material used for construction and types of furniture in the two rooms. 4. Using Normal Plot, verify normal distribution. Normal probability plot is used to verify if the variation of loss (loss dierential) is distributed normally. The normal plots from Figure 6.17 and Figure 6.18 indicate that the assumption of normality is not unreasonable. Both plots pass the fat pencil test. 5. Evaluate standard deviation , for log-normal distribution, and determine standard normal distribution. Mean , and standard deviation of loss dierential is calculated for both the access points. Mean is calculated to be 0 and standard deviation for D-Link access point is 4.127 and for LinkSys standard deviation is 4.2860. Using (3.1) loss dierential is standardized. Figure 6.19 and Figure 6.20 show the plots for standard normal distributions for AP1 and AP2. 6. Calculate Chi-square Goodness-of-Fit Test. 98
Figure 6.18: Building Two, classroom normal probability plot for AP2 (LinkSys).
Figure 6.19: Building Two, classroom standard normal distribution for AP1 (D-Link).
99
Observed 0 0 0 0 6 21 59 134 256 339 446 408 225 132 56 21 5 2 1 1 0 2112
Chi Statistic 0.002112 0.02112 0.16474 1.0328 0.16854 0.11136 0.00076798 0.14802 0.00095691 2.4226 2.026 4.1454 3.6421 3.08E-01 0.13218 0.11136 0.0011124 0.90585 4.2351 45.37 0.002112 -
100
Observed 0 0 1 2 7 17 62 123 238 374 419 413 233 133 62 18 9 1 0 0 0 2112
Chi Statistic 0.002112 0.02112 4.2351 0.90585 0.73005 0.32664 0.17555 1.7406 1.1994 0.070658 0.010215 5.2733 1.9823 0.22061 0.17555 0.11918 3.0353 0.0010397 0.16474 0.02112 0.002112 -
101
Figure 6.20: Building Two, classroom standard normal distribution for AP2 (LinkSys). As discussed in section 3.3 to evaluate goodness of t we need to calculate test statistic X 2 and degrees of freedom df . Table 6.4 and 6.5 are used for calculating Chi-square test statistic X 2 for AP1 and AP2. To nd the expected value for a particular bin, the area under standard normal curve for the bin is multiplied by total observations. Chi-square test statistic is calculated using (3.2). To evaluate if the distribution is normal, the bins that are used to calculate X 2 are from -3.25 to 3.25, a total of 13 bins. As discussed in section 3.3, degrees of freedom df = 13 - 1 = 12. From Table 6.4 X 2 for AP1 (D-Link)= 13.2181, and from Table 5.6 X 2 for AP2 (LinkSys)= 15.0594. By using Table 3.2 we can identify to p-value associated with X 2 and df values. We now have our chi square statistic X 2 for AP1 and AP2 (13.2181 and 15.0594), our predetermined level of signicance (0.05), and our degrees of freedom (df =12). From Table 3.2 we can say X 2 for AP1 = 13.2181, lies between 6.304 and 18.549. The corresponding probability is 0.9 < p < 0.1. This is below the conventionally accepted signicance level of 0.05 or 5%, so the null hypothesis that loss dierential is normal is veried for AP1 (D-Link). 102
Table 6.6: Classroom results from Building One and Building Two.
Now for AP2 our value of X 2 = 15.0594, lies between 6.304 and 18.549. The corresponding probability is 0.9 < p < 0.1. This is below the conventionally accepted signicance level of 0.05 or 5%, so the null hypothesis that loss dierential is normal is veried for AP2 (LinkSys). Table 6.6 lists the results for AP1 and AP2 from Building One and Building Two. The dierence in path loss exponents for a classroom in Building One and Building Two is less than 5%. So, it can be said that path loss models from Building Two validate the path loss models developed in Building One. 7. Compare with two-ray model. A graphical comparison is made between measured data and two-ray model. Figure 6.21 and Figure 6.22 show the comparison between the theoretical two-ray model and measured data. It can be seen from these plots that signal strength deteriorates much higher in two-ray model when compared to classroom measurements.
6.4 Summary Tables 6.3 and 6.6 compare Path loss exponent (n), mean () and Standard deviation () for Closed Corridor and Classroom scenarios in Building One and Building Two. It can be 103
Figure 6.21: Comparison of classroom mean loss with two-ray model.
104
seen that the Path loss exponent (n), mean () and Standard deviation () of the respective scenarios are approximately equal. So the path loss models developed in chapter 5 can be used in generation of Propagation models for indoor environment.
6.5 Conclusion In this study indoor propagation models were developed for an Open Corridor, a Closed Corridor, a Classroom and a Computer lab. In Chapter 2 a brief description of dierent types of propagation models (Empirical & Deterministic models) was given. Chapter 2 also gives a description of Log-distance Path Loss Model and Log-Normal Shadowing (Empirical model), and Two-Ray Model (Deterministic model). These models are later used in determining path loss equations for dierent scenarios. In Chapter 3 statistical tools used to support the ndings of the study are described. A detailed description of Normal Plot, Standard Normal Distribution, and Chi-square Goodnessof-Fit Test is given. The chapter also discusses Curve Fitting which was used to calculate the Path Loss exponent (n). In Chapter 4 a detailed description of the measurement scenarios, the experimental setup, the software, and hardware used for measurements was given. In Chapter 5 a numerical analysis of measurements in each scenario was conducted and the study determined equations that describe path loss for each scenario. This chapter concludes by accepting Alternative Hypotheses for both the Research Questions. Chapter 6 does a comparative study of the measurement scenarios in Building One and Building Two. The chapter concludes that there is no dierence in path loss models calculated in Building One and Building Two.
105
6.6 Future Work We conclude by outlining possible directions for future research: 1. A further extension to this study could be developing an algorithm for Optimal Access Point Selection and Channel Assignment. The path loss equations, and multipath fading calculated in this study can be used to simulate propagation in indoor environments. By combining these two studies a GUI based model can be developed for ecient positioning of Access Points in an indoor environment. A user can specify the maximum number of access points, and the algorithm would nd the optimal placement of access points in a given service area. 2. IEEE 802.11b and 802.11g use the unlicensed ISM frequency. Many electric appliances such as microwave ovens, cordless phones, and Bluetooth devices use the same frequency band. The Propagation models can further be extended to consider interferences between these electric devices and wireless LANs.
106
BIBLIOGRAPHY [1] T.S. Rappaport. Wireless Communications Principles and Practice. Prentice Hall, 1996. [2] William Stallings. Wireless Communications and Networks. Prentice Hall, 2002. [3] IEEE. Wireless LAN medium access control (MAC) and physical layer (PHY) specications. IEEE Standard 802.11, 1997. [4] G.A. Halls. HIPERLAN: the high performance radio local area network standard. Electronics and Communication Engineering Journal, 6:289296, December 1994. [5] IEEE. Wireless LAN medium access control (MAC) and physical layer (PHY) specications - Amendment 4: Further higher-speed physical layer extension in the 2.4 GHz band. IEEE Standard 802.11g, 2003. [6] IEEE. Wireless LAN medium access control (MAC) and physical layer (PHY) specications - Amendment 2: Higher speed physical layer (PHY) extension in the 2.4 GHz band. IEEE Standard 802.11b, 1999. [7] SIG. Bluetooth specication version 1.1. http://www.bluetooth.com/dev/specications.asp, February 2001. [8] IEEE. Wireless LAN medium access control (MAC) and physical layer (PHY) specications - Amendment 1: High-speed physical layer in the 5 GHz band. IEEE Standard 802.11a, 1999. [9] D. Dobkin. Indoor propagation issues for wireless LANs. RF Design; www.rfdesign.com, September 2002. [10] H.J. Zepernick and T.A. Wysocki. Multipath channel parameters for the indoor radio at 2.4 GHz ISM band. IEEE Veh. Technol. Conf., 1:190193, May 1999.
107
[11] A.F. Agelet, A. Formella, J.M.H. Rabanos, I.F. de Vicente, and P.F. Fontan. Ecient ray-tracing acceleration techniques for radio propagation modeling. IEEE Veh. Technol. Conf., 49:20892104, November 2000. [12] M. Iskander, Z. Yun, and Z. Zhang. Outdoor/indoor propagation modeling for wireless communications systems. IEEE Antennas and Propagation Society International Symposium, 2:150153, July 2001. [13] S.P.T. Kumar, B. Farhang-Boroujeny, S. Uysal, and C.S. Ng. Microwave indoor radio propagation measurements and modeling at 5 GHz for future wireless LAN systems. Microwave Conference, 1999 Asia Pacic, 3:606609, 30 November-3 December. 1999. [14] K.A. Remley, H.R. Anderson, and A. Weisshar. Improving the accuracy of ray-tracing techniques for indoor propagation modeling. IEEE Veh. Technol. Conf., 49:2350 2358, November 2000. [15] W.K. Tam and V.N. Tran. Multi-ray propagation model for indoor wireless communications. Electronics and Communication Engineering Journal, 32:135137, January 1996. [16] J. Tarng and T. Liu. Eective models in evaluating radio coverage on single oors of multioor building. IEEE Veh. Technol. Conf., 48:782789, 1999. [17] A. Neskovic, N. Neskovic, and G. Paunovic. Modern approaches in modeling of IEEE Communications Surveys,
mobile radio systems propagation environment. http://www.comsoc.org/pubs/surveys, 2000.
[18] T. Holt, I. C. Pahlavan, and J. F. Lee. Ray tracing algorithm for indoor radio propagation modeling. 3rd IEEE Int. Symp. on Personal, Indoor and Mobile Radio Communications, October 1992.
108
[19] J. W. McKown and R. L. Hamilton. Ray tracing as a design tool for radio networks. IEEE Network, 6:2730, November 1991. [20] S. Y. Seidel and T. S. Rappaport. A ray tracing technique to predict path loss and delay spread inside buildings. IEEE Global Telecommunications Conf., 2:649653, December 1992. [21] R. Valenzuela. A ray tracing approach to predicting indoor wireless transmission. IEEE Veh. Technol. Conf., 43rd:214218, May 1993. [22] IEEE. Wireless LAN medium access control (MAC) and physical layer (PHY) specications Amendment 6: Medium Access Control (MAC) security enhancements. IEEE Standard 802.11i, 2004. [23] J. C. Chen, M. C. Jiang, and Y. W. Liu;. Wireless LAN security and IEEE 802.11i. Wireless Communications, IEEE, 12:2736, February 2005. [24] Cyrus Peikari and Seth Fogie. Maximum Wireless Security. Sams, 2002. [25] Statit Software:Process Management Tools and Applications. http://www.statit.com, November 1999. [26] The MathWorks Inc. MATLAB HELP. http://www.mathworks.com. [27] StatsDirect Statistical Software. Statistical help. http://www.statsdirect.com. [28] LinkSys A Division of Cisco Systems, Inc. http://www.LinkSys.com/. [29] D-Link Corporation/D-Link Systems, Inc. http://www.dlink.com/. [30] M. Milner. Network stumbler version 0.4.0. NetStumbler; www.netstumbler.com, 20012004. Statit Bulletin.
109
ABSTRACT
Title of thesis:
WLAN SIGNAL CHARACTERISTICS IN AN INDOOR ENVIRONMENT AN ANALYTIC MODEL AND EXPERIMENTS
Damayanti Gupta, Master of Science 2005 ,
Thesis directed by:
Professor Ashok Agrawala, Department of Computer Science
In today's environment, in which WLAN technology is being deployed extensively, in order to improve the effectiveness of such deployments it is necessary to have a detailed understanding of WLAN signal characteristics. Radio signal attenuation and path losses depend on the environment and have been recognized to be difficult to calculate and predict. Past studies of the signal propagation, in both an indoor and in an outdoor environment have used several models with varying degrees of success and/ or complexity. I present here a simple analytic model for an indoor environment, and use it for determining the signal strength in a 3-D environment with one transmitter. The model is experimentally verified and is shown to yield a good match with the measurements. Several consequences of the model are studied and contour plots mapping signal strengths are generated. Signal strengths in the presence of an obstruction in the field of the transmitter are studied.
WLAN SIGNAL CHARACTERISTICS IN AN INDOOR ENVIRONMENT - AN ANALYTIC MODEL AND EXPERIMENTS
by
Damayanti Gupta
Thesis submitted to the Faculty of the Graduate School of the University of Maryland, College Park in partial fulfillment of the requirements for the degree of Master of Science 2005
Advisory Committee: Prof. Ashok K. Agrawala, Chair Dr Samrat Bhattacharjee Dr. Amitabh Varshney
Acknowledgements
I would like to acknowledge significant help provided by Ms Rashi Narain, in conducting the experiments.
ii
Table of Contents
Acknowledgements ...... ii Table of Contents ... iii Chapter 1. Introduction ....................................................................................................... 1 1.1 Motivation................................................................................................................. 1 1.2 My Work................................................................................................................... 2 1.3 Thesis Outline ........................................................................................................... 2 Chapter 2. Related Work..................................................................................................... 4 Chapter 3. Background ....................................................................................................... 8 3.1 Wave Propagation and the Free Space Model .......................................................... 8 3.2 Indoor Environment .................................................................................................. 9 3.2.1 Reflection........................................................................................................... 9 3.2.2 Absorption.......................................................................................................... 9 3.2.3 Diffraction........................................................................................................ 10 3.2.4 Scattering ......................................................................................................... 10 3.2.5 Refraction......................................................................................................... 10 3.3 The Multipath Effect............................................................................................... 11 Chapter 4. Theoretical Model ........................................................................................... 12 4.1 Basic Premise.......................................................................................................... 12 4.2 The Environment .................................................................................................... 12 4.3 Paths Traveled by the Waves.................................................................................. 13 4.4 Magnitude Calculation............................................................................................ 16 Chapter 5. Experiments..................................................................................................... 18 5.1 Experiment 1........................................................................................................... 19 5.1.1 Experimental Setup.......................................................................................... 19 5.1.2 Results.............................................................................................................. 22 5.1.3 Model Parameter Adjustment .......................................................................... 23 5.1. Both Sets of Data ............................................................................................... 26 5.1.5 Sources of Error ............................................................................................... 27 5.2 Experiment 2........................................................................................................... 30 5.2.1 Experimental Setup.......................................................................................... 30 5.2.2 Results.............................................................................................................. 30 5.2.3 Location Jitter .................................................................................................. 33 5.3 Re-Parameterized Model ........................................................................................ 35 5.4 Model Verification.................................................................................................. 36 5.4.1 Experimental Setup.......................................................................................... 36 5.4.2 Results.............................................................................................................. 36 5.5 Conclusions............................................................................................................. 38 Chapter 6. Model-Based Analysis .................................................................................... 39 6.1 One Axis Variations................................................................................................ 39 6.2 Contour Graphs....................................................................................................... 46 Chapter 7. Obstructions .................................................................................................... 50 7.1 Theory and Models ................................................................................................. 50 7.2 Conclusions............................................................................................................. 54
iii
Chapter 8. Concluding Remarks and Future Work........................................................... 55 References......................................................................................................................... 57
iv
Chapter 1. Introduction
1.1 Motivation
Wireless LAN technology, in particular 802.11b and 802.11g technology, has been gaining prominence in its use and deployment. In this technology, wireless communication in the 2.4 GHz band is used for digital communication at the rates of 11 Mbps for 802.11b and up to 54 Mbps for 802.11g. The current technology supports access to the wireless network through the deployment of Access Points (APs), which can be accessed within a radius of a few hundred feet (both indoors and outdoors). The range and the quality of the connection depend on the environment, the walls, and other obstructions, reflecting surfaces, etc., that are there in the path from the transmitter to the receiver. In order to not only improve the digital communication in such environments, but also to use the availability of these signals for other purposes such as determining locations, we need to understand the signal characteristics in detail. In this thesis I present the results of one such study in which I use an analytic model of signal propagation to determine signal strength at various locations in a 3-D region in the shape of a rectangular box. The models used in this study were empirically verified by making signal strength measurements.
The study presented here is part of a larger project which is aimed at exploiting the detailed knowledge of the signal characteristics to better manage WiFi deployments and in improving the quality of digital communications in such environments.
1.2 My Work
RF signal propagation follows well known laws of physics. An RF wave emitted by an antenna is attenuated due to distance in a direct line of sight (LoS) path. Also, the receiver may receive multiple such waves in an indoor environment due to reflection and other such phenomena [15]. The signal strength at the receiver depends on all such component waves; their amplitudes as well as their phases. In this study I take a simplistic view, taking into account the direct LoS signal, and signals received after a single reflection off of each of the available walls. I consider a rectangular box corresponding to a corridor in a building and develop an analytic model of the signal strength at various locations in this region when one transmitting antenna is deployed at a specified location. For verification of the model I measured the actual signal strengths in a corridor in the A.V. Williams building, University of Maryland, College Park, using the APs that are already deployed there and measuring the signal strength using a laptop with a network access card. The results of this study are presented in this thesis.
1.3 Thesis Outline

In Chapter 2, I present some past work and related reading. In Chapter 3, I discuss the basic physical concepts behind this problem. In Chapter 4, I delineate my theoretical model and some results. Some empirical studies conducted for verification of the model are presented in Chapter 5. Some theoretical implications of this model are presented in
Chapter 6. In Chapter 7, I examine the effect of the presence of obstructions in the path of the RF radiation. I conclude in Chapter 8 and discuss future directions.
Chapter 2. Related Work
The area of modeling radio wave propagation has been studied by many researchers and from a variety of perspectives. When such studies were conducted for supporting the design of a system, the aim usually was to come up with empirical models which could be used in practice to engineer the systems. Many of these studies have been conducted to support the design and implementation of cellular telephone systems [e.g. 17]. Recently some studies have been conducted looking at the signal propagation in an indoor environment also.
In [1], a review of popular propagation models for the wireless communication channel is undertaken. Macrocell (typically a large outdoor area), microcell (a small outdoor area), and indoor environments are considered. In the first case, LoS conditions are usually not satisfied. The signal propagates by reflection, diffraction and refraction. Since there are so many factors, creating a model is difficult, but a few have been proposed. The microcell model is easier to formulate, both from empirical results, and by physical analysis. For example, an empirical model is proposed in [2]. A ray-optic theoretical model is proposed in [3].
Several studies have indicated that indoor propagation models can be difficult to formulate and use. Field strength measurements [4] show that fluctuations can be very high (up to 80dB). Parameters used in empirical models require more compensation to fit experimental data than for the outdoor situations. This phenomenon is attributed to the
fact that at a particular point, the signal strength is determined by a much larger number of indirect components.
There are two types of models to fit behavior experimental/statistical1 models, and theoretical models2. Many experimental models are based on measurements. A model is formulated to fit the data. For example [5] proposes:
PL(d) = PL(d0) + 10*n*log(d/d0) + A
(2.1)
Where PL(d) is path loss in dB at distance d, PL(d0) is the known path loss at reference distance d0, n is the exponent depending on propagation environment, A is uncertainty in model. Parameter n is very sensitive to propagation environment type of construction material, type of interior, relative location within building etc. n ranges between 1.2 (waveguide effect3) to 6. These models have a significant error rate.
A few studies have used analytic models. In the ray-tracing model (e.g. [6]), all possible signal paths from the transmitter to the receiver are calculated. Predictions can be based on free-space transmissions, reflections, diffraction, diffuse wall scattering, and transmission through various materials. At any point, the sum of all the components is taken to get the signal strength. This method can also be used to predict time dispersion
In these the experiment is conducted, and mathematical formulae are inferred from the results. The model is created using principals of physics.
Waveguides are typically tubular structures. Their characteristics are such that waves traveling in them are forced to follow their structure. This is an efficient mechanism for signal transmission, it reduces loss.
(signal fluctuation across time at a particular location). This model needs a very detailed analysis and can be computationally very expensive. Note that the approach I have taken in this thesis is similar to the one in this paper except that I do not use explicit ray tracing techniques.
In the finite-difference time-domain (FDTD) models, a numerical solution of Maxwells equations is undertaken [7]. Maxwells equations are approximated by a set of finitedifference equations. A numerical solution is obtained by finite differencing. This is also a computationally demanding method. It is only suitable for small areas. For larger sizes, ray-tracing models are considered better.
The theoretical models are computationally very expensive, and the empirical models are not very accurate. These limitations can be overcome by an artificial neural network model. The ETF- Artificial Neural Network(ANN) model [4, 8] is based onmultilayer perceptron feedforward neural networks. For a particular environment, the neural network has to be trained with measured data. While the training is expensive, it only needs to be done once. Experiments show that the accuracy of this model is comparable to the accuracy of the ray-tracing and FDTD models.
Other phenomena affecting signal transmission have been studied. E.g. in [4], it is seen that a receiver moving slowly within the indoor environment experiences Ricean or Rayleigh fading4. Faster moving receivers experience Doppler shifts. The presence of human beings in the experimental area affects the results also [9]. The users own body
4
These are mathematical equations modeling the fading.
(very close to the receiver) causes a drop in signal strength. Other people in the vicinity not only cause attenuation, but fluctuation also. These irregularities are hard to predict.
An interesting application of measurement and modeling of signal features is the field of indoor geolocation. (e.g. [10], [11]). Geolocation is the tracking of mobile human beings and objects within a building. An algorithm is introduced, (e.g. [11],[12]), signal measurements are made, and the locations are calculated. The measurements made may include [10] arrival distance of the signal, angle of arrival, signal strength, phase, time of arrival and so on. This is a complex scenario, so a variety of location finding algorithms have emerged e.g. [13].
We note that all these studies have looked at the RF propagation in different environments and have studied ways of predicting the signal characteristics taking into account the specific environment under study. In my study I consider a simple model of propagation in a 3D environment.
Chapter 3. Background
3.1 Wave Propagation and the Free Space Model

The starting point in this study is the wave propagation in free space. The signal at distance x from the source, J (x) , can be written as [18]J ( x) = ( J 0 x
D
)e i ( 2
fxc )
(3.1.1)
where x = distance from source J0 = signal amplitude at source f = frequency(2.4 GHz) D = Exponent value contributing to decay of signal as a function of distance x c = speed of light in free space
We note that for free space generally the value of D should be two. However, in several empirical studies researchers have proposed a number of other values for D, varying from 1.2 to 4, which reflect the specific environment that they were considering. I will start with the value of D as 2. Further, the signal J (x) is treated as a complex variable and this aspect has to be taken into account when I manipulate this variable.
The basic premise of this study involves a transmitter/antenna/AP(access point) called T (this transmits the radio signal), and the receiver R, which is the recipient of this signal.
3.2 Indoor Environment

When a radio wave encounters objects in its path it gets affected by them. The most significant factors affecting the propagation of radio waves are reflection, absorption, diffraction, scattering, and refraction [15]. Let us consider these next.
3.2.1 Reflection Reflection occurs when the radio wave is incident on a surface which has much larger dimensions than its wavelength. In an outdoor environment, these would occur off the surface of the earth, buildings etc. In an indoor environment, it would be walls, people and other obstructions. The reflected waves follow the typical laws of reflection in that the incident angle is equal to the angle of reflection and that the wave undergoes a phase change of 180 degrees. Depending on the material of the reflecting surface a portion of the signal may also be absorbed. Perfectly conducting materials are perfect reflectors, while poorer conductors involve more absorption of the signal before reflection. The conductivity of a material further depends on its dielectric properties.
3.2.2 Absorption As the radio wave passes through a material, a portion of its energy is absorbed. The amount of absorption depends on the characteristics of the material.
3.2.3 Diffraction This occurs when the radio path between the transmitter and the receiver is obstructed by a surface that has sharp irregularities (edges). The radio signal impinging on the edge results in secondary waves which propagate in all directions around the edge(including behind the obstacle.) This phenomenon is responsible for providing a path between the T and the R even when there is no direct or reflected path. The diffraction depends on the geometry of the object, as well as on the amplitude, phase, and polarization of the incident wave at the point of diffraction.[15]
3.2.4 Scattering When there are objects of comparable dimensions to the wavelength of the radiation in the medium of transmission, scattering of the signal occurs off of these objects. If the number of such objects per unit volume is large, the effects of scattering can be appreciable. Scattering is particularly prevalent when there are rough and irregular surfaces present.
3.2.5 Refraction In an outdoor environment, the atmosphere has a refractive index, which curves the path of the signal. This changes the geometry of the situation. In an indoor environment, this effect is not significant. However, the signal does pass through objects, within which
10
refraction will occur. So the signal will come out at a different position than expected. Thus, obstructions change the path of the radiation.
3.3 The Multipath Effect

As a consequence of the factors noted in section 3.2, the receiver R may get multiple waves following different paths, from the source T to it. Each of these waves will have its own amplitude and phase, and will arrive at R with its own delay. Note that the signal received at R is the superposition of all these component waves obtained by adding the signal values treated as complex numbers. I can then determine the signal strength as the amplitude of the complex number.
11
Chapter 4. Theoretical Model

4.1 Basic Premise
Given a rectangular 3-D space whose dimensions are known, and with a radio wave transmitter and receiver of known coordinates, I calculate the field strength at every point in this space by taking into account the direct LoS ray and the reflected waves reaching the receiver. I only consider the waves reaching the receiver after one reflection.
4.2 The Environment

In this study I consider a rectangular 3-D space to model a corridor in a building (this space is now called the box) in which one AP has been installed at a fixed location. The dimensions of the corridor and the location of the AP are known. I proceed to calculate the signal strength measured at a location in the box by taking into account the direct LoS path and 6 reflected waves.
I assume that the corridor walls are parallel to the axes of the coordinate system and that the origin is at one corner, x-axis is along the length, y-axis is along the width and z-axis vertical in this box. Further, The corridor has dimensions A, B, C. The access point coordinates are x1, y1, z1. The receiver coordinates are x2, y2, z2. The walls have reflection coefficients of rc1.
12
The floor has a reflection coefficient of rc2. The ceiling has a reflection coefficient of rc3. The original amplitude, as emitted by T, is given by J0.
4.3 Paths Traveled by the Waves

Let us look at the different paths taken by the rays to travel from transmitter to receiver as in Fig 4.1 and Fig. 4.2 LoS (line of sight) path between T and R called T-R. Let this distance be d1.
Clearly this distance can be calculated as ________________________________
d1 =
(x2 x1)2 + (y2- y1)2 + (z2 z1) 2
(4.3.1)
R also receives 6 different rays from reflections. These are paths traveled by rays from T to each separate plane in the box and then reflected back to R.
In order to consider the reflections let us examine one pair of reflections. They are off of 2 planes the y = 0 plane, and the y = B plane. Consider a ray that starts from T, then impinges on surface y = 0, undergoes reflection, and then reaches R , the receiver. The diagram(Fig. 4.1) shows this.
13
Fig. 4.1 Reflection from y = 0 plane
The point of reflection is S(x, y, z). As we note from the equation for the signal strength, the main factor to consider for the reflected wave is the total distance traveled and the loss at the point of reflection. A phase change of also has to be added. In order to
calculate the distance traveled I draw the image of the transmitter on the other side of the x-axis.
For the reflection, the distance traveled by the ray is T-S + S-R. Looking at the diagram, from simple geometry, it can be seen that this is equal to I1-R. So d, distance traveled by the ray reflecting off of y = 0 plane, is x2
d = ( x2 x1) 2 + ( y 2 + y1) 2 + ( z 2 z1) 2
(4.3.2)
14
Now for the y = B plane. Refer to diagram Fig. 4.2.
Fig. 4.2 Reflection from y = B plane
The point of reflection is S(x,B,z). So, d, the distance traveled by the ray reflecting off of y = B plane, is I1-R, which is
d = ( x 2 x1) 2 + ( y 2 2 B + y1) 2 + ( z 2 z1) 2
(4.3.3)
The other reflected distances (from the other planes) can be calculated analogously.
15
4.4 Magnitude Calculation

As noted in Chapter 3, the signal strength at any point, with attenuation due to distance, is given by
J ( x) = J 0 x e
D
2 fx ) c
(4.4.1)
Where x = distance from source J0 = signal amplitude at source f = frequency(2.4 GHz) D = exponent value contributing to decay of signal As a function of distance x c = speed of light in free space
By Eulers equation, this can be rewritten as

J ( x ) = J 0 x D [Cos( 2 fx 2 fx ) + iSin( )] c c
(4.4.2)
For the LoS path, using eqn (4.3.1), the amplitude at R due to this ray is given by
J ( R ) = J (T )d 1 e
D
2 fd 1 ) c
Or
J ( R) = J (T )d1 D [Cos ( 2 fd1 2 fd1 ) + iSin( )] c c
(4.4.3)
16
Consider the case of the reflections.
The 2 rays travel distances of d and d respectively (as in (4.3.2) and (4.3.3)). The amplitude at R due to the reflected ray having traveled distance d is
J ( R) = J (T )rc.d
2 fd + ) c
(4.4.4)
This can also be written as(for a reflection off a wall which has reflection coefficient rc1)
J ( R ) = J (T ) rc1d
[Cos(
2 fd 2 fd + ) + iSin( + )] c c
(4.4.5)
So we have seven rays arriving at R. We calculate the real and imaginary parts of each ray using the sine and cosine components .So we have r1.r7 real contributions, and im1.im7 imaginary contributions.
Real = r1 + r2 + r3 + r4 + r5 + r6 + r7 Img = im1 + im2 + im3 + im4 + im5 + im6 + im7

The total magnitude is calculated as
(4.4.8)
____________
Magnitude = Real2 + Img2 (4.4.9)
In this way we can calculate the magnitude of the signal received at any arbitrary point.
17
Chapter 5. Experiments
In order to verify the applicability of the model developed in Chapters 3 and 4, I conducted some experiments by measuring signal strengths in a corridor in the AV Williams building. I used the AP installed in that corridor for the UMDnet and measured the signal strength by placing a laptop with a network interface card at measured locations. I used a software package called Horus5 to take the measurements. At each location, 300 readings of the signal strength were taken in 100 seconds. From these readings, I obtained the average signal strength as well as the standard deviation at each location.
Note that while the model used is deterministic, in practice there are noise factors that affect the signal strength. These include the movement of objects around the receiver during the measurement, and also receiver and antenna noise. These noise factors will give rise to variability in the results. I expect the model to capture the average behavior of the measurements. However, I will indicate the standard deviation of the measurements in my results.
When making the model calculations I used the following parameter values which are based on some published results: The decay exponent D = 2. This is as indicated for Eqn. 3.1.1.(Though I will modify this and indicate if I have done so)
Developed by Dr. Moustafa A. Youssef, University of Maryland at College Park
18
The walls have reflection coefficients of rc1 = 0.5. The floor has a reflection coefficients of rc2 = 0.3. The ceiling has a reflection coefficient of rc3 = 0.1.
(The reflection coefficients are as used as in [16].) In a later section, I will look at the impact of model parameter values and how better values can be selected for my experiments.
There are some other factors that the model does not directly account for. These include antenna and receiver gain, system losses, absorption, and noise factors. I expect this to contribute to a linear shift, or difference between the model and experimental results. I will experimentally determine the magnitude of this shift, and will shift the value of the model plot accordingly.
5.1 Experiment 1
The experiments were conducted in a corridor on the 4th floor of the AVW Building.
5.1.1 Experimental Setup The corridor layout is shown in Fig. 5.1.
19
185 cm
192 cm
AP
(1357, 180,230)
1923 cm
185 cm x-axis
y-axis Fig . 5.1 Room Layout
The corridor I used for the measurements had length 1923 cm and width 185 cm. The height of the room was 248 cm. The x- and y- axes were chosen as indicated in Fig. 5.1. The z-axis was chosen to be perpendicular to the floor. The 192 cm and 185 cm correspond to gaps in the walls for the corridors leading out of this space. (I chose measurement locations such that no reflected rays affecting the data points would be incident on those areas). The AP is marked in the diagram and had coordinates of (1357, 180, 230). The experimental setup was as in Fig. 5.2.
20
AP
Receiver positions First position of receiver
31 cm 75 cm
(75, 480, 83)
480 cm
Fig. 5.2 Experimental setup
In Fig. 5.2, the x- and y- axes are as marked. The first position of the receiver was the green dot, and its position coordinates are shown. The subsequent positions were the red dots. All measurements were taken by placing the laptop on a trolley so that the NIC was at a height of 83 cm.
At the first point, the receiver coordinates were (480, 75, 83). I moved the cart to increment the x-position (by 31 cm each time), and measured the magnitude at each point.
21
5.1.2 Results The value of the signal strength at each point was measured. The experiment was repeated . So there are two sets of measurements. The results for the first set of experiments are plotted in Fig. 5.3.
Fig. 5.3 Signal values from Model and from Experiment
The errorbars show measured signal strength values within one sigma bound of the mean value. The rms error between experiment and model is 6.23 dBm.
I include the results from the second set of measurements(Fig. 5.4). This set of measurements was taken immediately after the first one. I take multiple sets to observe the variability of results for the same experiment repeated with the same setup.
22
The rms error between experiment and model is 6.08 dBm. The 5th data point has an unexpected dip in value.
It is noted that there is attenuation in the signal with increasing separation from the antenna, as expected. Also, data points exhibit dips and peaks in the signal strength, and this can be attributed to variations introduced by the multipath effect.
5.1.3 Model Parameter Adjustment I shall examine if a better fit to experimental data can be obtained by adjusting the values of parameters used in the model.
23
Decay exponent, D
From first principles, the decay exponent should have value 2. However, it has been hypothesized in the literature [15] that a decay exponent other than 2 should be used. I shall vary the value of D until a good fit with experiment is observed.
Reflection coefficients rc1, rc2, rc3 .
In the model plots that I have drawn so far, I used reflection coefficients of 0.5, 0.3, and 0.1 as indicated in [16]. However, in our setup, these may consist of different values. So I should vary the values of these coefficients until a good fit with experiment is obtained.
I proceed to vary all the parameters together until a best fit is obtained. I vary D from 0.2 to 2.0, and each of the rc values from 0.1 to 0.9.
For the first set, I get a best fit at D = 0.6, rc1 = 0.1, rc2 = 0.1 and rc3 = 0.1. I shift the graph by 32.87 dBm. The rms error is 2.26 dBm.(Fig.5.5)
24
Fig 5.5 Signal values from Model and from Experiment
It is seen that values predicted by the model now lie entirely within the range of experimentally measured values.
For the second set, I again vary the D and rcs as previously. The best fit is obtained with at D = 0.8, rc1 = 0.1, rc2 = 0.1 and rc3 = 0.1. The graph hasis shifted by 29 dBm. The rms error is 3.82 dBm.(Fig.5.6)
25
Again, it is seen that values predicted by the model lie almost entirely within the range of experimentally measured values.
5.1. Both Sets of Data I plot both datasets together to get Fig.5.7. I will use the average of the parameter values used so far(D = 0.7, rc1 = 0.1, rc2 = 0.1 and rc3 = 0.1, shift = 31 dBm).
26
Fig. 5.7 Signal values from 2 Datasets and from Model
It is seen that (except for a few deviant points) data from the different datasets are relatively close together. Clearly there are other factors at work which impact the actual measured values. Let us look at some of them next.
5.1.5 Sources of Error There are errors in my measurement process which have not been accounted for. Errors manifest as a result of receiver and antenna noise. Also there is error due
to people walking around and obstructing the rays. These will manifest as the variance in the measurements made.
27
Positional errors
These are the errors in the measurement of room dimensions, and antenna and receiver coordinates. I shall consider only the last item, as the receiver position is constantly changed, and there is more chance of error.
I recalculate values generated by the re-parameterized model with the optimal D , rcs, ( and shift, as obtained from the last two sets of the experiment), and with the introduction of a jitter value in the receiver coordinates ( Fig. 5.8). I allow a jitter value of +- 7cm in each coordinate direction (y- and z- coordinates).
Fig. 5.8 Values with jitter in receiver y- and z- coordinate values

[AKA1]
The values in the errorbars in the plot of the model indicate the range of values obtained by jittering the receiver coordinate values. It is seen that variability is introduced due to
28
position jitter. The range of model values still liealmost entirely within the range of experimentally observed values when the R position error is included in the graph.
I repeat this for the second set of measurements as shown in Fig. 5.9.
Fig. 5.9 Values with jitter in receiver y- and z-coordinate values
Some variability is introduced due by position jitter. Again, the model values still lie almost entirely within the range of experimental values.
It is seen that location error introduces some variability into the model results.
29
5.2 Experiment 2
The experiment was repeated in the same corridor with different receiver locations.
5.2.1 Experimental Setup The receiver position was again varied only on the x-axis. As before, antenna coordinates were (1357, 180, 230). The first position of the receiver was at (345, 83, 91). The receiver was moved by 31 cm each time, and I had 40 data points. Two sets of measurement were made.
5.2.2 Results I plot the results. I directly optimize the values of D, rcs, and shift (Fig. 5.10).The results from the first set of experiments are plotted in Fig. 5.10 For the first set, the best fit is obtained with at D = 0.8, rc1 = 0.3, rc2 = 0.3 and rc3 = 0.1. The graph is shifted by 26.2 dBm. The rms error is 2.63 dBm.
30
There is excellent agreement between experiment and the adjusted model.
I took another set of measurements, as plotted in Fig. 5.11. the best fit is obtained with at D = 0.8, rc1 = 0.3, rc2 = 0.1 and rc3 = 0.1. The graph is shifted by 26.2 dBm. The rms error is 2.14 dBm.
31
Again, there is an excellent match between theory and experiment.
I plot both datasets together in Fig. 5.12. For the parameter values, I use the average of the last two sets and use D= 0.8, rc1 = 0.3, rc2 = 0.2 and rc3 = 0.1, shift = 26.2 dBm. I get Fig. 5.12.
32
Fig. 5.12 Signal values from 2 Datasets and From Model
It is again seen that mostly, data from the different datasets are very close together.
5.2.3 Location Jitter I consider errors in the measurement of R as before. I introduce a jitter(of +- 7cm) in the receiver coordinate positions, and regenerate the model graph .(I use the average values of D , rcs, and shift as obtained from the previous two sets of this experiment.)
For Set 1, I get Fig. 5.13
33
It is seen that the model values lie almost entirely within the range of experimental values.
I repeat this procedure for Set 2 to get Fig. 5.14.
34
Again, it is seen that the model values lie almost entirely within the range of experimental values.
5.3 Re-Parameterized Model

So far, I adjusted some parameter values in my model to get a good fit to the data of Experiments 1 and 2. Over the two experiments, the average value of the decay exponent D is 0.75. The values of the reflection coefficients are 0.2, 0.15 and 0.1.The average value of the linear shift is 28.5 dBm. I shall use these values from now on.
35
5.4 Model Verification

I shall verify my re-parameterized model against another experiment.
5.4.1 Experimental Setup This was conducted in the same corridor as in the previous two experiments. The initial position of the receiver was at (348, 105, 98) with increments of 31 cm along the x-axis happening each time. 22 measurements were taken.
5.4.2 Results For the first set, I plot the experimental and model results in Fig.5.15. ( The model uses the parameters as explained earlier.
36
The rms error is 3.6 dBm. The values predicted by this model lie somewhat within the range of values as measured by experiment.
Another set of measurements was taken as shown in Fig. 5.16.
37
Fig. 5.16 Signal values from Model and Experiment
The rms error is 2.7 dBm. The values predicted by this model are largely within the range of values as measured by experiment.
5.5 Conclusions
It is seen that there is a good match of predictions of my model with experimental results. While the match is not perfect, it is well within experimental errors. Therefore I conclude that the model with adjusted parameters is useful in determining the signal strength in an indoor corridor.
38
Chapter 6. Model-Based Analysis
In this chapter I undertake further investigation into the effects of the analytic model. First, this gives us insights into the working of the model. Next, it helps us understand the salient features of the 3-D signal strength topology. Since the model results match the measurements well, we can use my model to predict signal strengths in an indoor environment. Further, it gives us ideas into future improvements of the model. Note that I use the re-parameterized model of Chapter 5. (The value of the decay exponent D is set at 0.75, the reflection coefficients at 0.2, 0.15 and 0.1, and the linear shift at 28.5 dBm.)
6.1 One Axis Variations

For Fig. 6.1, I use the data (receiver coordinates, antenna location, room coordinates) from Experiment 1 in Section 5. However, I extend the range of the readings for receiver x-values continuing beyond that of the x-position of the antenna. Also, a finer-grained set of x-positions are considered. I model the magnitude due to the LoS or direct line-ofsight ray, and also the composite magnitude containing all 7 rays in the model.
39
Fig. 6.1 Signal strength as predicted by model
Note that the AP is at x = 1357 cm. The model shows that the signal strength (both LoS and composite components) increase with proximity to the antenna, as expected. The LoS component is smooth, whereas the composite magnitude shows a lot of fluctuations due to the interference caused by the reflected waves. I evaluated signal strengths for this configuration for some other values of y and z, and found thata similar variability is observed.
To understand the signal characteristics, I also consider the signal strengths for variations along the other axes. I vary the y-coordinate smoothly keeping the x and z constant at x = 650 cm and z = 83 cm(Fig. 6.2)
40
Fig. 6.2 Signal strength V. y-coordinate
There is a repetitive variation across the y-axis. There are drops of ~ 2 dBm over distances of ~ 20 cm.
Next, x and y are kept constant at x = 950 cm, and y = 160 cm. Varying the z-coordinate, I get the results shown in Fig. 6.3.
41
Fig. 6.3 Signal Strength v. z-coordinate
The magnitude of the LoS signal is mostly greater than that of the composite signal. This can be attributed to the geometry of the situation again.. The gross trend of the LoS signal gradually increases with increase in z-coordinate (increasing proximity to AP which is at z = 230 cm There are many jagged variations across the z-axis. There are changes of ~ 3 dBm over 20 cm.
Going back to Fig.6.1, I investigate the composite magnitude variations further. I zoom in on a portion of the graph to get Fig. 6.4.
42
Fig. 6.4 Zoomed in signal strength
There are sharp, repetitive variations over short distances. The signal strength is sensitive to position on x-axis and a change in position of ~2 cm can cause a fluctuation of ~ 4 dBm.
Next let us look at the reflected waves separately. It is interesting to see the constituent signal strengths, and to see how they add up.
I split up the magnitudes due to each pair of reflected rays (off of 2 facing walls). The results from the 3 sets of reflections are shown in Fig. 6.5 (a),(b) and (c).
43
Fig. 6.5(a) Constituent Signals
44
Fig. 6.5(b) Constituent Signals
Fig. 6.5(c) Constituent Signals
45
In all cases, the reflection signal strengths (due to one pair of reflections) are lower than the LoS magnitude. This is due to absorption losses at the point of reflection as well as the increased distance traveled by the wave. Clearly the components are complex numbers and have to be added up resulting in the composites as shown in the figures indicating a considerable interference amongst the different rays. The composite signal always shows the most variation across the graph.
Note that using the model we can examine the contributions of each of the seven components of the composite signal. However, what is seen by any receiver is the composite signal only.
6.2 Contour Graphs

I look at some contour graphs to gain a better understanding of the situation. I look at a yz plane parallel to the plane of the APs in Fig. 6.6. x is held at 500 cm, and the y, z coordinates are varied across their entire range. As before, the AP is at (1357, 180, 230).
46
Fig. 6.6 Contour graph on y-z plane
There is a total variation of about 3 dBm across the plane. The distribution is due to the geometry of the situation.
I look at a y-z plane which is closer to the AP, with x = 1300(Fig. 6.7).
47
Fig. 6.7 Contour graph on y-z plane
The curves swirl towards the upper right corner, with successively higher signal strengths. This is because the AP is located in that region. There is a total variation of 6 dBm across the plane.
I look at a variation (Fig. 6.8) across the x-y plane(x is varied from 300 to 1600 cm, y from 0 to 185 cm), z is held constant at 92 cm. (This corresponds to values used in Expt. 2).
48
Fig. 6.8 Contour graph on x-y plane
Signal strengths increase with proximity to the AP, as in the experiment. There is a variation of 7 dBm across the plane.
Contour graphs are a useful tool in visualizing signal strength configurations in a given environment. Further, they can be used to predict the presence of dead spots or regions of very low or non-appreciable signal strengths.
49
Chapter 7. Obstructions
The signal strength configuration changes in the presence of an obstruction in the path of the waves. Obstructions can include walls, ceilings, humans (possibly moving), and so on. Thus, studying this gives us better estimates of signal strength configurations. Further, an interesting application can be the prediction of the presence of obstructions (and their locations) by the measurement of signal strengths.
7.1 Theory and Models

Rays passing through an obstruction are attenuated [15]. (Note that this applies to rays in their LoS path between the transmitter and receiver, as well as to reflected rays.) The amount of absorption depends on the dielectric properties of the material. There are also diffraction and refraction effects which alter the signal strength configuration, but I will not consider that for now.
I modify my model to check if the rays from T to R fall within the coordinates of the obstruction. In those cases, eqn. 4.4.1 is modified as
J ( x) = J 0 .x e
D
2 fx ) c
(7.1.1)
where is the transmission coefficient, the fraction of the magnitude transmitted through the material.
50
An attenuated signal configuration is expected in the presence of an obstruction. I plot some graphs to explore this.
First, I take a particular configuration, and plot signal strengths in the absence of (Fig. 7.1), and in the presence of(Fig. 7.2), an obstruction. The obstruction is set to have a transmission coefficient of 0.1.
Fig. 7.1 Signal strength configuration with no obstruction
51
Fig. 7.2 Signal strength configuration with obstruction
It is seen that there is signal strength attenuation in Fig. 7.2 as compared to Fig. 7.1.
I move the obstruction closer to the AP in the model to get Fig. 7.3.
52
For Fig. 7.4, I move the AP along the y-axis (in the model.)
53
The various non-symmetrical effects observed are because of the interference of the LoS and reflected rays.
7.2 Conclusions
There are very specific signal strength configurations that result depending on the presence and position of an obstruction. My model can be modified to include this effect to predict the resulting configuration with a good degree of accuracy.
54
Chapter 8. Concluding Remarks and Future Work

In this study I developed an analytic model for determining the signal strength at various locations in a 3-D rectangular space containing one Access Point. In my model I used the standard wave propagation equation and took into account multipath effects due to reflected waves received from six sides of the rectangular box. My measurements, conducted to verify the model, indicate that the model with single reflections shows sufficient accuracy and can be used effectively.
Using the model, I calculated signal strengths along several axes and on several planes. All of these calculations show significant variability in the signal strength. Of note is the fact that the signal strength can change by several dBms as the location changes by a few cms. Such changes in the signal strength may not have any serious consequences for data transmission. However, when the signal strength measurements are used for other purposes, such as determining the location, the consequences of such variability can be very far reaching. Also, as most WiFi systems use the signal strength to decide which AP to associate with, a significant change in the measured signal strength can trigger a handover where one may not be necessary.
I looked at models of signal strength attenuation in the presence of an obstruction in the path of the RF waves. These results indicate that the changes in the signal strength depend significantly on the location and type of the obstruction. This knowledge can be used for determining the location of an obstruction, such as a person, in the corridor.
55
The work presented in this thesis is the first step in developing detailed and workable analytic models for indoor space. Many additional practical details will have to be included in the enhanced models. This work can be extended in several directions. For example the model can be enhanced to:
Include other effects such as refraction, scattering, and fading. Incorporate more detailed effects due to people walking by, and other obstructions in the room.
Study how the signal strength changes while an obstruction moves through the room.
Study the effects of multiple reflections. Include effects due to the characteristics of the AP. I have made an approximation of the AP as a fixed point, radiating rays uniformly in all directions. This is not true in real life, but depends on the specific antenna involved. The distribution pattern of the antenna can also be reflected in the model.
Study in greater detail the causes of discrepancies between the model and experiment.
56
References
(1) Modern Approaches in Modeling of Mobile Radio Systems Propagation Environment, IEEE Communications Surveys and Tutorials: Aleksandar Neskovic, Natasa Neskovic, and George Paunovic, University of Belgrade (2) Short Distance Attenuation Measurements at 900MHz and 1.8GHz Using Low Antenna Heights for Microcells, IEEE JSAC, vol. 7, no. 1, Jan. 1989: P. Harley (3) Radio Propagation Characteristics for Line-of-Sight Microcellular and Personal Communication", IEEE Trans. Ant. and Prop., vol. 41, no. 10, Oct. 1993.: H. H. Xia et al. (4) Artificial Neural Network Indoor Radio Propagation Model (band 900MHz), Master's Thesis, Faculty of Electrical Engineering -- Belgrade, 1997: A. Neskovic (5) Short Distance Attenuation Measurements at 900MHz and 1.8GHz Using Low Antenna Heights for Microcells", IEEE JSAC, vol. 7, no. 1, Jan. 1989.: P. Harley (6) Finite Conductivity Uniform GTD Versus Knife-edge Diffraction in Prediction of Propagation Path Loss," IEEE Trans. Ant. Prop., vol. AP-32, Jan. 1984: R. J. Luebbers (7) Wireless Information Networks, New York: John Wiley & Sons, Inc., 1995: K. Pahlavan and A. H. Levesque (8) Indoor Electric Field Level Prediction Model Based on the Artificial Neural Networks," IEEE COMML, vol. 4, no. 6, June 2000 : A. Neskovic, N. Neskovic, D. Paunovic (9) Influence of the Human Body on Indoor Radio Communications at 450MHz -- Measurements and Analysis, Proc. 43th Annual Conf. ETRAN, Budva -Yugoslavia, June 1996: A. Neskovic, N. Neskovic, and D. Paunovic (10) Indoor geolocation science and technology, Communications Magazine, IEEE , Volume: 40 , Issue: 2 , Feb. 2002:K. Pahlavan, Li Xinrong, J.P. Makela (11) Radio propagation modeling for indoor geolocation applications, Personal, Indoor and Mobile Radio Communications, 1998. The Ninth IEEE International Symposium on , Volume: 1 , 8-11 Sept. 1998 : P. Krishnamurthy, K. Pahlavan, J. Beneat
57
(12) Wideband radio propagation modeling for indoor geolocation applications, Communications Magazine, IEEE , Volume: 36 , Issue: April 4, 1998 : , K. Pahlavan, P. Krishnamurthy, A. Beneat (13) A comparison of wireless geolocation algorithms in the indoor environment, Wireless Communications and Networking Conference, 2004. WCNC. 2004 IEEE , Volume: 1 , 21-25 March 2004: K. Pahlavan, M. Kanaan (14) Radio propagation models : http://people.deas.harvard.edu/~jones/es151/prop_models/propagation.html #harley (15) Wireless Communications Principles and Practice: Theodore S. Rappaport (16) Indoor Radio Channel Propagation Modelling by Ray Tracing Techniques, PhD Thesis: David I. Laurenson (17) Radio Propagation in Cellular Networks, N. Blaunstein (18) Physics, Part I, Resnick and Halliday
58
Greg Durgin, Neal Patwari, Theodore S. Rappaport Mobile and Portable Radio Research Group at Virginia Polytechnic Institute and State University Bradley Department of Electrical and Computer Engineering Blacksburg, VA 24061-0350 540231-2967 Fax: 540231-2968 http:==www.mprg.ee.vt.edu
Abstract For radio propagation prediction, recent simulations involving ray tracing o er unprecedented accuracy 1 , 11 , 13 , 14 . These techniques surpass statistical channel models and provide a bounty of additional information including RMS delay spread, angle of arrival, and overall wideband channel impulse response. In particular, three dimensional ray tracing produces an accurate, deterministic channel model for wireless system design. This paper presents a new 3D ray tracing technique of unprecedented speed and accuracy. Simulation results are compared to microcell measurements at 1900 MHz. Three main areas of error in a ray tracing simulation are propagation modeling errors, database errors, and kinematic errors. Propagation modeling errors come from the geometrical optics models used to describe radio wave behavior. Database errors stem from the limitations of a nite, numerical description of the world. The ability of a ray tracing algorithm to nd and interpret radiation paths determines the kinematic errors of a simulation. This paper focuses on a method to launch and interpret rays that model the wavefronts of radio propagation; we present a new method that eliminates many of the kinematic errors associated with a ray launching scheme. This paper concentrates on three major ray launching concepts. It rst establishes methods for geodesic ray launching, originally proposed by Seidel 13 , 12 . The following section reviews the principles and inconsistencies of interpreting ray information using the reception sphere model found in the literature 11 . Finally, the paper introduces a new method for interpreting ray information using a special weighting function to construct ray-traced wavefronts. The distributed wavefront model has several important advantages over the reception sphere models used in previous literature i.e. 13 , 15 . This paper provides fundamental principles to construct an extremely accurate three dimensional ray launcher for propagation prediction, unlike any previously proposed. An environment database for 3D ray launching consists of at polygons that represent the surfaces of buildings and terrain. Flat polygons are easy to describe numerically and
This work is sponsored by the National Science Foundation and the MPRG Industrial A liates Program.
An Advanced 3D Ray Launching Method for Wireless Propagation Prediction
allow simple re ection calculations. The well-de ned edges also lend themselves to di raction calculations 6 . A polygon database may be constructed easily using computeraided design CAD software. Figure 1 shows an example of a ray-traced environment in AutoCADR 5 .
I. Introduction
Fig. 1. An example of rays traced through an urban environment Rosslyn, VA.
A key issue for implementing widespread, site-speci c propagation tools is the formal speci cation of environment databases. These databases must contain all of the information relevant to a propagation simulation such as geometry, material properties, and surface characteristics. For this reason, current research is working to develop the Real World Database Format RWDF, a standard method for constructing a site-speci c propagation database. An alternative to ray launching that avoids the geometrical complexities of a ray launching algorithm is the method of images. The method of images uses image theory to place arti cial sources in the environment that model reections from the at planes of a database 3 , 7 . Because image theory determines exact radiation paths, the method of images introduces no errors into the radiation paths it nds. However, this method only works for re ected modes of propagation, since di raction introduces in nite degrees of freedom in the direction of a ray path. The primary drawback to image theory algorithms is their heavy dependence on the number of elements in the environment database. For single re ection, each
III. Ray Tracing Techniques
II. Propagation and Database Modeling
additional surface in the database can double the number of images available to a receiver. Therefore, image theory should not be used to render complicated, threedimensional cityscapes, which require multiple re ections from hundreds of surfaces. Image theory is best suited for simpli ed, two-dimensional city blocks. A versatile ray tracing technique, and the one explored in the following sections, launches rays from a transmitter and re ects them through the environment. For this method, the launched rays that pass arbitrarily close to a receiver establish the actual radiation paths. Although this paper concentrates on re ection, ray launching techniques are extendible to di raction. The principle advantage of this method is its ability to quickly render a complicated 3D scene since the algorithm has only a linear dependence on the number of database elements. The overall processing time can be reduced further by incorporating spatial discrimination, such as a bounding volume hierarchy 2 .
IV. Geodesic Ray Launching
Fig. 2. An icosahedron a is tessellated to produce a geodesic sphere b. Rays launch from the vertices of this structure.
approximates uniformity, there are some discrepancies in angular separation among the launched rays. For instance, each ray emanating from one of the twelve vertices of the icosahedron has only ve neighboring rays rather than the usual six see Figure 3. This type of aberration becomes insigni cant for heavily tessellated spheres 12 .
In a ray launching scheme, rays emanate from a unit sphere centered on the transmitter location. Launch points around this sphere follow a regular, computer-generated geometry. There are two desirable characteristics of a ray launch geometry: Large-Scale Uniformity - The launch points must distribute evenly around the sphere so that all regions of space are illuminated equally by rays. Large-scale uniformity delivers unbiased ray coverage in three dimensions. Small-Scale Uniformity - The local pattern of rays impinging on a wavefront should be a predictable, uniform pattern. This corresponds to equal angles be- Fig. 3. Examples of ve-neighbor aberrations on a geodesic sphere tween a ray and its neighbors, which assists the interwith tessellation frequency N = 8. Tessellation frequency is the number of triangles along the edge of an icosahedron. pretation of wavefront information. Ray shooting from the vertices of regular polyhedrons is the only way to exactly satisfy the two uniformity criteria 15 . Another aberration occurs in the angular separation beSince no regular polyhedron has more than twenty vertices, tween rays. There is an average discrepancy of twenty perhigh precision ray tracing must use other geometries 12 . cent between the smallest and largest angular separations on a sphere 15 . Angular separation plays an important A. Geodesic Geometry part in a ray tracing calculation. It determines the distance The geodesic sphere arises by tessellating the faces of to the closest ray on the same wavefront - vital informaa regular polyhedron and extrapolating the intersection tion for interpreting simulation results. Regardless of the points to the surface of a sphere 4 . Figure 2 depicts the tessellation frequency, there will be a twenty percent variageodesic facets and vertices that result when the sides of an tion between the minimum and maximum separation on a icosahedron are subdivided into smaller equilateral trian- geodesic sphere. High tessellation frequency, however, does gles. The geodesic vertices provide ray launch points with mitigate the variations in localized angular separation 12 . equivalent angular separation around the entire sphere 13 . Characterizing the angular separation for each ray should Moreover, each ray will have exactly six neighbor rays that then use the maximum separation angle with respect to its surround the original in a predictable hexagonal pattern. six neighbors, rather than a single separation value for all The receiver schemes presented in this paper exploit this geodesic rays. regularity when calculating power levels. Despite aberrations, an average value for angular separation is a useful parameter in a simulation. The angular B. Geodesic Aberrations resolution of a geodesic sphere is a function of the tesselUniformity allows for simple interpretation of a ray lation frequency N , the number of triangles on the edge launch. The reception sphere model takes advantage of of an icosahedron face. Assuming that the 20N 2 total the uniformity of a geodesic launcher to collect rays us- geodesic facets are congruent equilateral triangles, Equaing a simple circular area. Because the geodesic sphere tion 1 nds the average radial separation by equating
their total area with the 4 steradians of a unit sphere. = 1 Equation 1 is an excellent approximation to the average angular spreading of rays. It gives a measure of angular resolution for a three dimensional ray tracing simulation. Once rays are traced through a scene, a program must interpret the results by measuring voltage or power levels at arbitrary points in space. A common method to interpret the traced ray information is the reception sphere model. This method assumes uniform ray launching from a geodesic sphere in an environment composed of at surfaces, such as buildings or planar segments of terrain. The simple assumptions and easy implementation of the reception sphere model make it a useful ray tracing algorithm. A. Reception Sphere Mechanics The reception sphere model surrounds a receiver point with a sphere of varying size. Rays that intersect this sphere contribute to the total received power. One of the traits of the reception sphere model is its simple implementation since the sphere-ray intersection tests are easy to describe mathematically. The size of the reception sphere depends on the characteristics of the incoming ray. Because rays spread out as they leave the source, the reception sphere must increase size accordingly. Figure 4 shows a two dimensional example of tting reception sphere sizes with incoming rays to guarantee the collection of exactly one ray from each wavefront. If the sphere is too large, more than one ray from the same wavefront adds to the total power. An undersized sphere can miss a wavefront altogether.
V. The Reception Sphere
4 p N 5 3
1:205 radians = N
69:0
N
1 one ray from a wavefront is p3 the distance between rays 11 , 13 . This radius sweeps out a circular area across the wavefront where sometimes two rays fall within the sphere, registering additional voltage and power.
Rx
Minimum Size
Rx
Double Count
Fig. 5. Reception sphere double counting on a uniform wavefront. The pattern represents the ideal impingement of 3D rays launched from a geodesic sphere.
Fig. 6. Double count errors white regions along an ideal wavefront.
The double counting errors occur frequently across a measured wavefront. Figure 6 highlights the double counting on a map of an ideal wavefront. The probability of a randomly placed receiver experiencing a double count comes from the geometry of Figure 6. For the ideal case, double countings occur with a probability of 32p3 , 1 or Sphere for LOS Wave 20:9 of the time. This probability is independent of tesselSphere for Reflected Wave lation frequency and worsens with increased geodesic and r re ecting surface aberrations. itte m ns Incoherent accumulation of this additional power will reTra sult in a +3 dB error. However, the arrival angles and the path lengths of two rays from the same wavefront are idenX Receiver tical. Since their amplitudes and phases are equal, a coherent addition will result in twice as much predicted voltage Reflecting Surface and, subsequently, a +6 dB power error. Assuming random location of a receiver along a geodesic wavefront, a coherent Fig. 4. Reception sphere sizes for line-of-sight LOS and re ected voltage measurement with reception spheres will increase wavefronts. The spheres are sized to guarantee the reception of the mean power by at least 1.25 dB and introduce an adone and only one ray from each wavefront. ditional 2.4 dB of standard deviation error. Simulations that do not use localized angular separation to calculate B. Ray Double Counting reception sphere sizes exhibit even worse statistics 16 . In two dimensions, reception spheres work perfectly; VI. Distributed Wavefronts a third dimension adds complexity and ambiguity. FigThe method of distributed wavefronts remedies the probure 5 shows the ideal impingement of geodesic-launched rays onto a spherical wavefront. The minimum radius for lems inherent with the reception sphere model while maina reception sphere to guarantee the collection of at least taining the speed and simplicity of ray launching. It re-
moves the double count errors that the previous model introduced. Instead of counting hit-or-miss rays, the weighting of a nearby ray is a function of proximity to the receiver. Thus, the total received power comes from a three dimensional weighting of the nearby rays. The problem of eld strength prediction reduces to matching a weighting function to the speci c launch geometry.
A. Weighting Functions
B. Distributed Wavefront Geometry The curve in Figure 7 weights power or eld contributions as a function of arc length from receiver to ray, shown as d in Figure 8. Equations for total distance traveled by the ray, R, and the normalized distance along a wavefront, x, are given by 2 and 3.
R = L2 + t2 + 2Lt cos x= d t = 1 tan,1 L +sin R t cos
p
2
3 The localized uniformity of a geodesic launch make it an excellent candidate for the use of weighting functions. The L - ray length prior to intersection t - length from receiver to intersection uniform pattern of impinging rays see Figure 5 ts nicely - receiver's specular deviation angle from ray into a weighting scheme. A radially symmetric weighting d - arc length from receiver to ray function must be found that, when placed on all of the R - arc length from ray to nearest neighboring ray nearby geodesic ray points of Figure 5 and summed, equals The normalized distance x, a ratio of arc lengths, allows a constant for any position across the wavefront. the distributed wavefront the We used a Monte Carlo approach to nd valid weighting function given by Table I model to use 0 same weighting in the range x 1. functions. This method uses a tabulated weighting function for points in the Figure 5 pattern to calculate how the weighted sums vary across the potential wavefront. Brute Adjacent Ray force iterations increase or decrease the tabulated function Primary Ray values until the curve converges to a useful weighting funcTx tion, as in Figure 7. Table I is an example of a smooth d function that will weight the surrounding rays so that the Rx total is nearly invariant of the position across the wave L front.
Distributed Wavefront Weighting Function
Contribution to Field Strength f(x)
Reflecting Surface
0.8 0.6 0.4 0.2 0 0
Fig. 8. Geometry for a distributed wavefront 2D projection.
0.2 0.4 0.6 0.8 Normalized Separation Distance x
Fig. 7. A plot of the tabulated radial weighting function. TABLE I

Tabulated values for a weighting function.
C. Intersection Formulation The distributed wavefront method accumulates eld strength from rays if there is an unobstructed path between the receiver location and the ray's source either the transmitter or a surface intersection. Equation 4 shows the total electric eld phasor calculated from the source points of traced rays.
~ ET =
~R x ~i x ~ Ei
N X i=1
~ f xi V~ R ; ~ i ej i Ei x x
4
0.000 0.025 0.050 0.075 0.100 0.125 0.150 0.175 0.200 0.225 0.250 0.275 0.300 0.325
f x
1.000000 0.997451 0.994310 0.989887 0.983723 0.975419 0.964567 0.950793 0.933756 0.913197 0.888966 0.861056 0.829511 0.794412
0.350 0.375 0.400 0.425 0.450 0.475 0.500 0.525 0.550 0.575 0.600 0.625 0.650 0.675
f x
0.755992 0.714609 0.670687 0.624725 0.577259 0.528856 0.480105 0.431592 0.383894 0.337574 0.293157 0.251151 0.211982 0.176035
0.700 0.725 0.750 0.775 0.800 0.825 0.850 0.875 0.900 0.925 0.950 0.975 1.000
f x
0.143627 0.114993 0.090079 0.068779 0.051053 0.036723 0.025471 0.016905 0.010621 0.006190 0.003215 0.001356 0.000000
- receiver location - location of ith ray source - electric eld vector associated with the ray from the ith source i - phase of the electric eld N - total ray sources in environment f xi - distributed wavefront weighting V~ ; y - visibility function value 1 if LOS, 0 if not x~
~ For this coherent summation, Ei and i are functions of the total distance traveled, R, and the attenuations and phase shifts from surface re ections. Similar terms for di raction and scattering can also be introduced into 4. The environment database determines the visibility function, V~ ; ~, which has a value of 1 if a clear path exists xy
between two vectors and 0 if the path is obstructed. A key characteristic of Equation 4 is that the source points, ~ i , x do not depend on receiver location. For this reason, a computer only needs to trace a scene once for a xed transmitter. Characteristics for subsequent receivers only require recalculation of the visibility functions and eld values for the new locations. Dividing ray tracing into separate tracing and linking stages saves calculation time for repeated simulation of a xed transmitter.
VII. Results and Measurements
Power Delay Profile at Receiver #14

55 60 Measured Predicted
Received Power (dBm)
65 70 75 80 85 0
Time delay relative to trigger (ns) The method of distributed wavefronts was tested on a region of the Virginia Tech campus shown in Figure 9. Ex- Fig. 10. Predicted and measured power delay pro les 0 dBm transmit power at 1900 MHz. Total received power is -55.7 dBm tensive wideband measurements were performed at this site measured, compared to -56.8 dBm predicted. by Saldanha in 10 . These measurements were made at 1900 MHz, using a spread spectrum channel sounder with 10ns resolution 8 . The transmitter and receiver were at References heights above ground of 3 and 1.5 meters, respectively. 1 H.L. Bertoni, W. Honcharenko, L.R. Maciel, Howard H. Xia.
100
200
300
400
500
2 3 4 5 Fig. 9. Rays traced for a Virginia Tech campus database at receiver location 14. 6 7 8 9 10 11 12 13 14 15 16
Figure 10 shows ray traced results using the distributed wavefront model compared to the measurements for an arbitrary receiver location. The ray tracing predicts the position and amplitudes of multipath components very well, with slight perturbations due to the accuracy of the computer database. As expected, a few multipath components were omitted in the prediction since only re ected modes of propagation were simulated in a simpli ed environment. Distributed wavefront ray launching proves to be an accurate model for nding a channel impulse response.
VIII. Conclusions
Geodesic spheres and distributed wavefront methods increase the accuracy of 3D ray tracing for propagation prediction without sacri cing simplicity and speed. The distributed wavefront method corrects ray double counts in the reception sphere model. The method also lends itself to spatial interpolation of discrete 3D antenna patterns. A link budget for the intersection points also reveals an important principle in ray tracing: subsequent simulations in an environment with a xed transmitter do not require additional ray tracing. Future work will incorporate di raction, material properties, and polarization e ects.
UHF Propagation Prediction for Wireless Personal Communications". Proceedings of the IEEE, vol 82, no 9, pp. 1333-1359, Sep 1994. J. Goldsmith and J. Salmon. Automatic Creation of Object Hierarchies for Ray Tracing". IEEE Computer Graphics and Applications, May 1987. C.M.P. Ho, T.S. Rappaport, M.P. Koushik. Antenna E ects on Indoor Obstructed Wireless Channels and a Deterministic Image-Based Wide-Band Propagation Model for In-Building Personal Communication Systems". International Journal of Wireless Information Networks, vol 1, no 1, 1994. H. Kenner. Geodesic Math and How to Use it. Berkeley, CA: University of California Press Ltd., 1976. J.C. Liberti and T.S. Rappaport. Analysis of CDMA Cellular Radio Systems Employing Adaptive Antennas in Multipath Environments". IEEE VTS 46th Vehicular Technology Conference, vol 2, pp. 1076-1080, Atlanta GA, May 1996. R.J. Luebbers. Propagation Prediction for Hilly Terrain Using GTD Wedge Di raction". IEEE Transactions on Antennas and Propagation, vol AP-32, no 9, pp. 951-955, Sep 1984. H. Meng, P. Koushik, T.S. Rappaport. Comparative Study of Indoor and Outdoor Site-Speci c Propagation Prediction Models. MPRG-TR-93-07, Virginia Tech, Feb 1993, 76 pages. W.G. Newhall, K. Saldanha, T.S. Rappaport. Using RF Channel Sounding Measurements to Determine Delay Spread and Path Loss". RF Design, pp. 82-88, Jan 1996. T.S. Rappaport. Wireless Communications: Principles and Practice. New Jersey: Prentice-Hall Inc., 1996. K. Saldanha and J.H. Reed. Performance Evaluation of DECT in Di erent Radio Environments. MPRG-TR-96-28, Virginia Tech, Aug 1996, 163 pages. K.R. Schaubach, N.J. Davis, T.S. Rappaport. A Ray Tracing Method for Predicting Path Loss and Delay Spread in Microcellular Environments". IEEE 42nd Vehicular Technology Conference, vol 2, pp. 932-935, Denver CO, May 1992. S.Y. Seidel and T.S. Rappaport. Site-Speci c Propagation for Wireless In-Building Personal Communication System Design. MPRG-TR-93-4, Virginia Tech, Feb 1993, 231 pages. S.Y. Seidel and T.S. Rappaport. Site-Speci c Propagation Prediction for Wireless In-Building Personal Communication System Design". IEEE Transactions on Vehicular Technology, vol 43, no 4, pp. 879-891, Nov 1994. S.Y. Tan and H.S. Tan. Propagation Model for Microcellular Communication Applied to Path Loss Measurements in Ottawa City Streets". IEEE Transactions on Vehicular Technologies, vol 44, no 2, pp. 313-316, May 1995. S.Y. Tan and H.S. Tan. Modelling and Measurements of Channel Impulse Response for an Indoor Wireless Communication System". IEE Proceedings on Microwave, Antennas, and Propagation, vol 142, no 6, pp. 405-410, Oct 1995. S.Y. Tan and H.S. Tan. Improved Three-Dimensional Ray Tracing Technique for Microcellular Propagation Models." IEE Electronics Letters, vol 31, no 17, pp. 1503-1505, 17 August 1995.

Indoor Model

Caricato da

Informazioni sul documento

Titolo originale

Copyright

Formati disponibili

Condividi questo documento

Condividi o incorpora il documento

Opzioni di condivisione

Hai trovato utile questo documento?

Questo contenuto è inappropriato?

Copyright:

Formati disponibili

Indoor Model

Caricato da

Copyright:

Formati disponibili

Czech Technical University in Prague

Faculty of Electrical Engineering Department of Electromagnetic Field

Novel Approach to Indoor Electromagnetic Wave Propagation Modelling

July 2003 Prague, Czech Republic

Czech Technical University in Prague

Novel Approach to Indoor Electromagnetic Wave Propagation Modelling

July 2003 Prague, Czech Republic

Preface and Genesis of the Motif Model

Martin Klepal Prague, July 2003

4. Motif Model ....................................................................................................... 19 - Novel Approach to Electromagnetic Wave Propagation Modelling

Abbreviations and Acronyms

FDTD - Finite Different Time Domain

RMS SNR TLM UHF URSI UTD Wi-Fi

TETRA - Trans European Trunked Radio System

WLAN - Wireless Local Area Network

2. Aims of the Dissertation

Aims of the Dissertation

3. Current Approaches to Indoor Propagation Modelling

Current Approaches to Indoor Propagation Modelling

General Outline of Electromagnetic Wave Propagation Inside Buildings

Electromagnetic Wave Indoor Propagation Behaviour

3. Current Approaches to Indoor Propagation Modelling

Waveguiding Effect of Corridors

Direct and Reflected Waves

Waveguiding Effect of Corridor Contribution of Evanescent Waves

3. Current Approaches to Indoor Propagation Modelling

Models for Indoor Propagation Prediction

3. Current Approaches to Indoor Propagation Modelling

Transmission Line Matrix Hybrid

ParFlow Maxwell Equations

Empirical Relations One-Slope Model

Fig. 3.2. Principal indoor models

COST231 One-Slope Model

d LOSM = L0 ( d 0 ) + n10 log d0

4d0 L0 ( d0 ) = FSL ( d0 ) = 20 log

LMWM = L1 + 20 log(d ) + LWalls + LFloors

a) Most of the ray power interacts

b) Only a fraction of the ray power interacts

3. Current Approaches to Indoor Propagation Modelling

Incoming Ray Transmitter

Obstacle Transmitting Ray

Incoming Ray Transmitter

Obstacle Edge Transmitter

First Fresnel Zone Cross Section

First Fresnel Zone Cross Section

First Fresnel Zone Cross Section

b) Fig. 3.5. First Fresnel zone interaction with obstacle

Dominant Path Model

3. Current Approaches to Indoor Propagation Modelling

3. Current Approaches to Indoor Propagation Modelling

Ray-Optical Method of Moment Hybrid Model

Ray-Optical Multi-Wall Hybrid Model

Summary and Motivation for Developing a New Model

b) Deterministic models Fig. 3.8. Basic approaches

4. Motif Model Novel Approach to Electromagnetic Wave Propagation Modelling

Delineation of New Model Ideas

Wave-Tracing Technique Ideas

Outline of the Principles of the Motif Model

Fig. 4.1. Example of site-specific description of a building, based on grid elements

Fig. 4.2. Overall probability radiation pattern of a cluster of obstacles (Motif)

Fig. 4.4. Example of the ray propagation process

Incident Plane Wave

Incident Plane Wave