Sei sulla pagina 1di 19

This article has been accepted for inclusion in a future issue of this journal.

Content is final as presented, with the exception of pagination.

Sinan Kurt and bulent Tavli

Path-Loss Modeling
for Wireless Sensor
Networks
A review of models and comparative evaluations.

P
ropagation models are used to abstract the actual propaga- WSN OVERVIEW
tion characteristics of electromagnetic waves utilized for The idea of WSNs was born as a vision, mainly due to the needs
conveying information in a compact form (i.e., a model of the military, more than five decades ago. In fact, pioneering
with a small number of parameters). The correct modeling WSN deployments were used during the Vietnam conflict [2].
of propagation and path loss is of paramount importance in wire- Indeed, the mythical air-dropped sensors scenario was real-
less sensor network (WSN) system design and analysis [1]. Most ized in the jungles of Southeast Asia. However, the technology
of the important performance metrics commonly employed for of that era was both expensive and bulky, and it took more than
WSNs, such as energy dissipation, route optimization, reliability, four decades (at the dawn of the 21st century) for the enabling
and connectivity, are affected by the utilized propagation model. platforms (e.g., Mica motes) to become affordable, small sized,
However, in many studies on WSNs, overly simplistic and unre- versatile, and capable enough to perform basic communication,
alistic propagation models are used. One of the reasons for the sensing, and computation functionalities. There has been an
utilization of such impractical propagation models is the lack of explosive growth in research and development efforts on WSNs
awareness of experimentally available WSN-specific propaga- both in academia and industry for more than a decade. Current-
tion and path-loss models. In this article, necessary succint ly, the vision of WSNs is transforming into a mature technology
background information is given on general wireless propagation with applications in such areas as telecommunications, precision
modeling, and salient WSN-specific constraints on path-loss agriculture, logistics, smart grids, security, surveillance, habitat
modeling are summarized. Building upon the provided back- monitoring, and health. Because WSN research is becoming
ground, an overview of the experimentally verified propagation more and more practical and application oriented rather than
models for WSNs is presented, and quantitative comparisons of purely theoretical and abstract, the level of realism in all aspects
propagation models employed in WSN research under various of WSN research is getting more important. The utilization of
scenarios and frequency bands are provided. accurate path-loss models is, therefore, of utmost importance in
WSN research.
Digital Object Identifier 10.1109/MAP.2016.2630035
A typical WSN is composed of a base station and a plurality of
Date of publication: 6 January 2017 sensor nodes spatially distributed over a target region to remotely

2 1045-9243/17©2017Ieee february 2017 IEEE AntEnnAs & ProPAgAtIon MAgAzInE


This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

monitor the desired phenomena [3]. It is possible that multiple developed for cellular systems directly in WSNs? What are the
base stations and/or actuators are also utilized in WSNs. Most differences? We believe that providing the necessary background
of the research activities on WSNs (e.g., routing, medium access to be able answer these questions is of the utmost importance. To
control, position estimation, synchronization, lifetime optimiza- the best of our knowledge, except for a technical report on ter-
tion, modulation, error correction, and security) are implicitly restrial wireless communications that provides a limited overview
related to propagation media and accurate propagation models of models for wireless communications [33], there is no review
of signals among sensor nodes [4]. The accuracy and precision of paper on propagation models for terrestrial WSNs.
the employed propagation model affect the coverage area, the A focused review/tutorial on WSN propagation models will
selected transmission power, and the lifetime of a network [5]. be beneficial for most WSN researchers and specifically for
The realistic reconstruction of WSN topologies is possible if researchers who are not experts on propagation mechanisms,
the path-loss values among the sensor nodes are computed via approximations, and models. In fact, the targeted audience for
the proper propagation models [6]. The design of WSN routing this article includes researchers working on WSNs who do not
protocols by employing simplistic path-loss models would not have the necessary background on radio propagation modeling
perform as expected under practical applications, and the char- in a terrestrial WSN environment. In this article, we have three
acterizations of these protocols under such assumptions would be main targets, itemized as follows.
questionable at best. The correct modeling of the path loss is of ■ A brief introduction to propagation models for wireless
the utmost importance in determining the signal-to-noise ratio, communications is given, and specific theoretical propaga-
which, in turn, is the determining factor in transmission power tion models commonly used for outdoor WSN applications
control [7], [8]. The WSN lifetime is significantly affected by the are explained to provide a background on propagation
optimal assignment of transmission power levels and the extent of mechanisms and to assess the applicability of well-known
retransmissions due to packet failures, which are all dependent on propagation models used for traditional wireless communi-
the correct modeling of the path loss. In general, the path loss is cations to WSNs.
not deterministic. Instead, it is stochastic. Furthermore, it can be ■ Available outdoor WSN-specific large-scale propagation
different over time and may change rapidly according to the used model alternatives reported to be more accurate when com-
frequency/scenario. Therefore, path-loss models generally include pared to pure theoretical (analytical) models are presented
statistical descriptors (e.g., mean values and standard deviations). in a systematic fashion to provide awareness of inaccuracies
In traditional wireless communications (e.g., cellular com- of the commonly utilized models.
munications), well-established propagation models exist that ■ Quantitative comparisons of propagation models are pre-
are well suited to specific applications that are experimentally sented under various scenarios and frequencies to iden-
corrected and adopted to be used with high accuracy [9]–[17]. tify the extent of modeling errors and guide researchers in
Nevertheless, some of the models for the WSN case are used adopting the appropriate path-loss models.
without awareness of the assumptions made in that model, and Note that, in this article, we consider only outdoor WSN
consequently, an unsuitable model could be used [18], which applications with stationary near-ground sensor nodes in the
can lead to flawed, if not unrealistic, design and analysis of 868-MHz and 2.4- and 3.5-GHz bands. Therefore, our focus is
WSN protocols and architectures. on outdoor propagation modeling. However, WSN applications
In the wireless communications literature, basic propagation are not limited to the cases we consider in this article. In fact,
mechanisms and models for both indoor and outdoor environ- WSN application domains are much wider (e.g., intravehicular,
ments are investigated, and various empirical and statistical intervehicular, indoor, and underground WSN deployments),
models are presented for specific application areas [19]–[21]. and our treatment of WSN path-loss modeling does not include
However, these models are not proposed for considering WSN all WSN application scenarios and domains. For example, there
applications scenarios. There are several survey/review papers is a rich literature on near-ground propagation modeling for
providing a limited discussion on WSN propagation models, but indoor scenarios [34]–[37].
the main targets of these papers are network behavior, network
topology, and system-level modeling [22]; general characteris- PATH LOSS IN WIRELESS COMMUNICATION
tics and applications [18], [23]; WSN simulation, emulation, or Propagation models are mainly focused on estimating the aver-
debugging tools [24]–[26]; and cross-layer approaches for WSN age signal strength drop at different transmitter–receiver (T–R)
problems [27]–[30]. The importance of a physical layer for signal separations and the signal strength variability in close proxim-
propagation effects, signal detection, modulation, power efficien- ity to the specific T–R separation [38]. The variation of signal
cy, frequency selection, and generation is emphasized in [3]. The strength is due to the changes in the propagation path between
different aspects of signal propagation are investigated, and the a transmitter and a receiver. The relative movements of the
estimation of link quality metrics is presented in [31]. Further- transmitter and/or receiver lead to signal strength variations.
more, there are survey papers for specific WSN application areas, Furthermore, even if both the transmitter and the receiver are
such as underground WSNs [32]. Nevertheless, when it comes to static, the signal strength can also change due to moving scatter-
available propagation models to be used in terrestrial WSN appli- ers or shadowing objects that affect the propagation environment.
cations, a gap seems to exist. For example, the following questions With respect to spatial and time scales, propagation models can
are left unanswered. Why can’t we use the propagation models be grouped in two classes, i.e., large scale and small scale [39].

IEEE AntEnnAs & ProPAgAtIon MAgAzInE february 2017 3


This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

Large-scale propagation models characterize the mean signal of these mechanisms comes from the fact that they result in
strength due to large distance T–R movements. Large-scale multiple copies of the transmitted signals, each having different
propagation effects are a power spread of radiated power in the amplitudes and phases, when they reach the receiver. The com-
propagation channel and shadowing, which is caused by obstacles posite signal obtained at the receiver is the sum of the copies
on the path between the transmitter and the receiver. Large- arriving from different paths. This may improve or degrade the
scale propagation models are very useful in coverage estimations. LOS signal depending on the phases of each signal received
On the other hand, small-scale (or fading) models try to char- from different paths, and the amplitudes of the signals define
acterize rapid fluctuations in the signal strength over relatively the level of improvement or degradation.
short spatial changes (a few wavelengths) and/or short time dura- Reflection occurs when a propagating wave is incident on
tions (i.e., spatial and/or time scale is/are short) [40]. This article an object that has large dimensions compared to the wave-
mainly focuses on the large-scale propagation models. length (i.e., a relatively smooth surface). The characteristics of
The free-space propagation model must be understood as a a reflecting surface are frequency dependent (i.e., the smooth-
basis on which to comprehend more advanced models. The free- ness of the surface changes with the frequency of interest). For
space propagation model can be used when there are no obstacles example, stucco may be considered as smooth for low frequen-
that can create paths except the line-of-sight (LOS) path from the cies, but it is rough at millimeter-wave frequencies. Generally
transmitter to the receiver (e.g., satellite communications) [41]. speaking, below a few gigahertz, building surfaces and walls
More precisely stated, the free-space propagation model is accu- are examples of reflecting objects, ignoring some microstruc-
rate when there are no objects at all besides the transmitter and tures, such as windowsills. However, if the transmitted wave is
the receiver because the existence of any other object could result incident on a rough surface or on a large number of obstacles
in reflected paths. This model is applicable only when the receiver with small dimensions compared to the wavelength, then such
is in the far field of the transmitter antenna (e.g., the transmitter interaction is known as scattering because wavefronts spread
receiver distance is larger than the Fraunhofer distance) [19]. The out or diffuse. It should be noted that scattering causes reradi-
Fraunhofer distance, d f , is given by ation in all directions, which could help to fill coverage holes,
2
especially at millimeter-wave frequencies. If the maximum
d f = 2D , (1) protuberance of a surface is larger than the critical roughness
m
level, then the reflection coefficient of the surface should be
where D is the largest dimension of the transmitter antenna and multiplied by a scattering loss factor. The Rayleigh criterion
m is the wavelength. The model is based on the Friis transmis- is used to define the critical roughness of a surface to model
sion equation [42], in which spherical propagation is considered the scattering effect [43]. Using the angle of incidence, i i, the
and the total received power is equal to the power density times critical roughness level is given by
the effective area of receiver antenna
m
hc = . (5)
P 8 cos i i
Pr = tot 2 A e, (2)
4rd
Diffraction occurs at the edge of an obstacle that blocks the
where Ptot = Pt G t, Pt is the transmitter power, G t is the trans- T–R path (the edge of an obstacle can be sharp or rounded, e.g.,
mitter antenna gain, and d is the T–R distance. The effective diffraction around a sphere [44] or a cylinder [45]). Even if there
area of receiver antenna, A e, is related to the physical size of the is no LOS T–R path, by means of diffracted waves, a receiver
antenna and given by can get the transmitted waves. Treetops, building edges, and
2
hills are examples of edges that result in diffraction. As the oper-
A e = m G r, (3) ation frequency increases, the diffraction loss increases. Indeed,
4r
the effects of diffraction are dependent on various factors, such
where G r is the receiver antenna gain and m is the wavelength. as the correlation length of the height variations [46]. Neverthe-
Substituting (3) into (2), we get the free-space propagation less, the diffraction loss increases as the blockage height of the
model as obstacle on the T–R path increases.
When modeling wave propagation for wireless communica-
Pt G t G r m 2 tions, the aforementioned mechanisms should be taken into
Pr = . (4)
(4r) 2 d 2
account, depending on the positions of the transmitter and the
For the sake of completeness, system losses (L) can also receiver, the presence of obstacles, and the environmental char-
be incorporated into the propagation model (e.g., absorption acteristics. Furthermore, understanding these phenomena helps
[19]). Equation (4) shows that the received power falls off with to analyze the trends observed in experimental data.
the square of the T–R distance. Besides clear LOS propaga- Considering the overall effects, the free-space propagation
tion, obstacles may exist (e.g., reflecting surfaces) on the path model alone can be highly misleading for most of the radio
between the transmitter and the receiver that necessitate the channels for terrestrial communication and specifically for
inclusion of other propagation mechanisms affecting signal WSNs. Before investigating propagation model alternatives for
propagation. Basic propagation mechanisms can be grouped WSNs, three well-known wireless communication path-loss
into reflection, scattering, and diffraction. The importance models are presented in the following sections.

4 february 2017 IEEE AntEnnAs & ProPAgAtIon MAgAzInE


This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

TWO-RAY (GROUND-REFLECTION) MODEL


Considering direct-path and ground-reflected waves together, the Transmitter
two-ray model produces a useful formulation based on geometric LOS Wave
optics (see Figure 1). Earth is assumed to be flat in the model. To d1 Receiver
calculate the power at the receiver, the vector sum of the LOS
E-field, E LOS, and the reflected E-field, E C , can be used. ht Reflected Wave
The reflection coefficient of the surface, C, is given by the d2 hr
following equations for incidence from free space to the ground θi
for parallel and perpendicular polarized waves, respectively:
d
- f r sin i i + f r - cosi 2i
C< = , (6)
f r sin i i + f r - cos i 2i FIGURE 1. a two-ray model illustration: (a) the ground plain
(path loss), (b) park (path loss), and (c) university yard (path loss).
sin i i - f r - cos i 2i
C= = , (7)
sin i i + f r - cos i 2i
LOG-DISTANCE PATH-LOSS MODEL
where f r is the relative permittivity of the ground that ranges By merging theoretical results and field measurements, it is
from 4 to 25 according to the ground properties. For small val- found that the average received signal power exhibits a logarith-
ues of i i independent of f r , the magnitude of C approaches 1. mic decrease with the T–R distance. In its general form, the log-
For large values of d, the magnitudes of both E C and ELOS will distance model is expressed as
be equal as d 1 /d 2 approaches unity. Under these assumptions,
the phase difference will be the important factor. PL (avg) = PL (d 0) + 10n log ( d ) , (11)
d0
Under the assumption that d is large enough, the received
power related to the square of the electric field is calculated where PL (avg) is in decibels, n is the path-loss rate, and PL (d 0)
as is the path loss at a reference distance (d 0) in the proximity to the
transmitter. d 0 is usually taken as 1 m, 100 m, or 1 km according
h 2t h r2
Pr = Pt G t G r . (8) to the application ranges of the model. PL (d 0) is sometimes set
d4
to the free-space path loss at d 0 or, alternatively, can be deter-
Therefore, the path loss (in dB) can be expressed by the two-ray mined by the measurement at d 0 . The loss rate n depends on
model as the environment, and it is in the range of 2–6 [50]. For the free-
space and the two-ray models, the n values are 2 and 4, respec-
PL = 40 log d - 10 (log G t + log G r + 2 log h t + 2 log h r) . (9) tively. The value of n increases as the frequency gets higher and
decreases with increasing antenna heights [12]. Generally, cases
Note that, for low d values, the assumptions of the two-ray with n values higher than 4 are observed for indoor or shadowed
model are not valid. The threshold distance after which the 1/d 4 outdoor environments. For example, n = 4.96 is reported for
power decay occurs (d b) can be obtained as European fixed forest environments [51].

4h t h r
db = . (10) LOG-NORMAL SHADOWING
m
Both the two-ray and log-distance models are deterministic
Although the two-ray model is an elegant theoretical model, models, and they give radially symmetric path-loss values.
it does not cover all mechanisms involved in electromagnetic However, there are random variations in the path loss due to
wave propagation. For example, if the antennas are very near environmental clutter, resulting from object blockage in the
the ground, then the path-loss estimation accuracy of the two- propagation path or from changes in reflecting and scattering
ray model deteriorates significantly [47]. To improve the theo- objects. As a result, the path loss at different locations having
retical models, empirical approaches based on the curve fitting the same T–R distance can differ considerably. Furthermore,
of the measured data can be used. In fact, empirical models the measured power along a straight line from the transmitter
are generally more advantageous because they can capture the does not exactly fit the log-distance characteristics. Therefore,
effects of many propagation mechanisms, even the hidden ones the nondeterministic characteristics of the path loss should be
(i.e., the ones that are not explicitly accounted for). The accuracy captured by using statistical models. The most common model
of an empirical model is very much dependent on the assump- for the characterization of nondeterministic effects is log-normal
tions on the actual path-loss mechanisms. For example, path- shadowing [52]. The path loss in decibels is given as
loss models with multiple breakpoints are more accurate than
one-slope models, as the actual measurements exhibit multiple PL (d) = PL (d 0) + 10n log ( d ) + X v , (12)
d0
breakpoints [48]. Furthermore, empirical model parameters are
different at LOS and nonline-of-sight (NLOS) scenarios; yet, in where X v is a zero-mean Gaussian distributed random vari-
both cases, the path loss is a probabilistic function of the dis- able with standard deviation v (both in dB). In practical appli-
tance and antenna heights [49]. cations, n and v values are computed by using the linear

IEEE AntEnnAs & ProPAgAtIon MAgAzInE february 2017 5


This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

regression of the measured data. The difference between the In the following section, we present propagation models
measured data and the estimated path loss is minimized. The used for WSN applications.
objective of the minimization is usually chosen as mean-square
error minimization. WSN PROPAGATION MODEL ALTERNATIVES
One of the most widely used propagation models for WSNs is
WSN CONSTRAINTS the free-space propagation model given in (4), which can be
Wireless communication systems, such as satellite, cellular, further simplified as
ultrahigh-frequency, and very-high-frequency band radios, usu-
P0
ally have well-established theoretical and experimental propaga- Pr (d) = , (13)
d
( )2
tion models [15]. Most of these models are improved in time by d0
incorporating correction factors obtained by the results of exten-
sive field measurements [14], [53]. Utilizing one or more of these where P0 is a reference power measured at a reference distance
mature models in WSNs directly is not a feasible option due the from the transmitter (d 0) . The simplicity of the model makes it
fact that WSNs have their own set of constraints that are not popular. This model is used to account for the energy dissipation
compatible with the assumptions made in most of the traditional of routing protocol simulations focusing on the network lifetime
wireless communication systems. In other words, WSN-specific and energy efficiency [59]–[62].
propagation models require consideration of the constraints
imposed by the nature of WSN use cases and scenarios, item- SIMPLIFIED TWO-SLOPE, TWO-RAY MODEL
ized in the following. An improvement upon the free-space model is the two-slope,
■ Low antenna heights: There are many WSN applications for two-ray model, which is among the most popular models used
both military and civilian purposes. In some of these WSN in WSN research [63]. A threshold (d b) is used for the T–R dis-
deployments, the antennas are not near the ground (e.g., tance, and, depending on the distance, either the d 2 path loss
military WSN scenarios where the base station is airborne ^ d 1 d b h or the d 4 path loss ^ d $ d b h is utilized. Although the
and civilian scenarios, such as smart cities, infrastructure assumptions and values used in [63] can be argued, the model
monitoring, and solar panel WSNs). However, in a signifi- proposed is much more realistic for outdoor WSN applications
cant portion of WSN deployment scenarios, a large num- than the free-space model. Therefore, this model is also widely
ber of small-form-factor sensor nodes are assumed to be used in WSN research [61]. Nevertheless, this model does not
placed on the ground. Obviously, the antennas used in these account for many important parameters, such as the effects of
devices are standing not more than a few meters above the surface reflectivity, antenna patterns, and heights.
ground [54]. Although it is possible that the base station To the best of our knowledge, there is no other WSN
and/or some of the sensor nodes can be positioned at a high- propagation model that has, arguably, attained the level of
er elevation, most of the sensors are directly on the ground. consensus achieved by the aforementioned two-slope, two-ray
Therefore, propagation models that assume antenna heights model. Unfortunately, the commonly used one is too simplis-
(both for the transmitter and the receiver) higher than a few tic to cover even the effects that can be modeled with slight
meters are not directly applicable to such WSN deployment modifications. Recently, there have been various field experi-
scenarios. ments leading to many useful WSN propagation models using
■ Low transmission power: The transmission power level in sensor platforms commonly employed in WSNs. Some of
WSN platforms is comparatively lower than many other them attempted to determine the effects of parameters such
wireless communication systems (e.g., the typical maximum as antenna heights and T–R distances [64]. The common attri-
transmission level in WSNs is in the order of 100 mW) butes of the WSN-specific path-loss models in the literature
[55], [56]. Therefore, the transmission ranges, typically, are are itemized as follows.
significantly lower than the transmission ranges of most of 1) The average radio signal strength drop correlates with the
the low-data-rate wireless communications systems (e.g., a transmission distance.
terrestrial trunked radio [57]). 2) The elevation of antennas reduces the reflection and obstruc-
■ Stationary network topology: Although mobile sensor nodes tion effects.
or mobile base stations are used in some WSN scenarios, in 3) There are random variations in the path loss.
most WSN scenarios, the sensor nodes are assumed to be The following sections present the alternative propagation
stationary (not moving) [3]; therefore, the channel models models used by researchers for WSN applications and shown to
that capture the effects of the relative mobility between be more accurate than these two basic models.
the transmitter and the receiver are not necessary in static
WSNs (e.g., the impact of the Doppler effect is negligible). ONE-SLOPE LOG-NORMAL MODEL
■ Directivity of antennas: The use of omnidirectional anten- The one-slope log-normal model employs (12), which can be
nas is the usual assumption in WSNs. However, the assump- used as an analytical model by utilizing certain predetermined
tion of a perfect omnidirectional radiation pattern is not values for PL (d 0), n, and v. For example, in [65], these param-
realistic for WSNs, which consist of inexpensive tiny sensor eters are taken as PL (d 0) = 55 dB, d 0 = 1 m, n = 4, and
platforms [6], [58]. v = 4. Although these values can be used as references, for the

6 february 2017 IEEE AntEnnAs & ProPAgAtIon MAgAzInE


This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

sake of precision and accuracy, it is better to utilize parameter grass park, a huge ground plain (quasi-ideal flat ground), and
values accurately characterizing the path loss for the specific a university yard surrounded by four-story buildings, are (2.09,
scenario [i.e., PL (d 0) , n, and v cannot be determined accu- 4.01), (2.34, 3.73), and (2.76, 4.00), respectively. As in the case
rately without taking into account scenario-dependent inputs of the estimation of the parameters of the one-slope model, the
like antenna heights and operating frequency]. estimation of the parameters of the two-slope model is accom-
In [66] and [67], one-slope models based on experiments con- plished by least-squares curve fitting. To validate the accuracy
ducted at 868-MHz and 2.4-GHz bands, respectively, are pre- of the one-slope and two-slope models constructed based on
sented. For the measurements at 868 MHz, the transmitter and the experimental data, simulations with 100 nodes are per-
receiver antenna heights are fixed at 13 cm. The transmitter uses formed. The metric used for validation is the average number
a constant transmit power level, and the distance between the of neighbors per node. The simulation results reveal that the
transmitter and the receiver is varied to obtain signal strength two-slope model achieves significantly higher accuracy than
values as a function of the T–R distance. In fact, measurements the one-slope model.
are taken at 0.5-m increments up to 3 m, at 1-m increments In [67], the measurement results obtained at 2.4 GHz by the
between 3 and 10 m, and at 2-m increments for distances methodology presented in the “One-Slope Log-Normal Model”
beyond 10 m. The farthest distance measurements are termi- section are used to estimate the parameters of the two-slope
nated is the point where the receive power is below −85 dBm. model through the least-squares method. The results of [67]
For spatial averaging, measurements are repeated at five differ- also confirm that the model given in (14) has superior estima-
ent points within a 50-cm diameter area centered at each of the tion accuracy over that of the one-slope model for three differ-
original measurement points. For temporal averaging, five time ent scenarios. In fact, the estimation accuracy of the path-loss
samples are taken at each position. Spatial and temporal averag- models is verified through the independently measured data at
ing is performed on raw data to eliminate fluctuations. By using two sites (i.e., the measurement data used to extract the model
the least-squares curve fitting, the parameters of the one-slope parameters are not used in the verification). The estimation
model are determined from the measured and processed data. errors of the one-slope and two-slope models are found to be in
For the measurements at 2.4 GHz, three pairs of antenna the 5.17–20.45- and 0.28–6.66-dB intervals, respectively. Add-
heights are used (i.e., h t /h r are 1 m/1 m, 3 cm/1 m, and 3 cm/2 m) ing different sets of antenna heights for the measurements, it is
[67]. Actually, two transmitter antenna heights of 1 and 3 cm observed that the breakpoint of the two-slope model is in accor-
and two receiver antenna heights of 1 and 2 m are employed. dance with the value of the breakpoint estimated by the two-ray
The measurements are taken in 1-m increments up to 10 m and model given in (10). Although the break-point values change
in 2-m increments between 10 and 100 m. At each original mea- slightly from scenario to scenario, it is shown that the dominant
surement point, data are collected in ten extra locations (within parameters affecting the value of the breakpoint are h t , h r, and
a 10-cm distance from the original point) for spatial averaging. m. As the antenna heights decrease, the breakpoint moves closer
Twenty samples are taken at each point for temporal averaging. to the transmitter antenna.
Linear regression employing the least-squares method is used to
obtain the parameters of the one-slope model. TWO-RAY MODEL
A propagation model based on the two-ray ground path-loss
TWO-SLOPE LOG-NORMAL MODEL model to simulate the physical layer characteristics of IEEE
In addition to the one-slope model, in [66] and [67], the param- Standard 802.15.4 is proposed in [68]. The proposed model
eters of two-slope models at 868-MHz and 2.4-GHz bands, takes the antenna gains and heights into account (8). The mea-
respectively, are also extracted from the field measurements surement results reported in [56] (at 1-m antenna heights of
described in the “One-Slope Log-Normal Model” section. In [66], the transmitter and the receiver) are used for verification. The
the field measurements at three different outdoor environments model is simulated by using the Tiny Operating System Simula-
are reported. The proposed two-slope model is expressed as tor (TOSSIM) and compared with two available TOSSIM mod-
els: 1) a simple model that only assumes all nodes are connected
PL (d) = )
PL (d 01) + 10n 1 log (d) + X v1 if d # d b
, (14) and 2) a lossy model, in which link connectivity is decided by a
PL (d 02) + 10n 2 log (d) + X v2 if d 2 d b
random number. The results of the comparative analysis reveal
where the model given in (12) is used with different slope that there are significant differences between the two-ray
regions separated at a breakpoint distance d b . It is shown ground path-loss model and the other models.
that the two-slope model fits better to the measured data
than the one-slope model. Three important observations on VARIABLE SOIL-REFLECTIVITY MODEL
path-loss modeling are reported in [66]. First, d b changes with Subterranean layers can greatly impact near-ground propagation
the environment even when h t and h r are fixed, which can [69]. In [70], a variable soil-reflectivity (VSR) model is proposed
be explained by the fact that Fresnel zone interception with based on the two-ray model by taking the reflective index of
the ground changes for different environments. Second, the soil and the directivity of the transmitter and receiver antennas
maximum coverage estimations with the two-slope model are into account. The proposed model targets outdoor applica-
more restrictive. Third, the path-loss exponents (n 1, n 2) com- tions where sensor nodes are placed at a given height above the
puted for three different environments, i.e., a slightly curved ground and the soil surface is smooth, which is generally the

IEEE AntEnnAs & ProPAgAtIon MAgAzInE february 2017 7


This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

case for WSN-monitoring applications on agricultural crop fields polynomials. The fourth component is the antenna radiation
[71]. The proposed VSR model is expressed as impact, which is embedded into (17) through parameters K 1
(which expresses the changes in the strength of the signal in the
c 1 + R D + 2RD cos (d) m,
2 2
P0
Pr = (15) direct path) and K 2 (which expresses the changes in the strength
d 2 (r/d) (r/d)
of the reflected signal). Both directivity parameters, K 1 and K 2,
where P0 is the power at a reference distance; R is the reflectiv- are critically dependant on the specifics of the antennas used
ity of the surface; D is the directivity of the antennas; d and r and obtained from the antenna radiation patterns determined
are the LOS and reflected-wave path lengths, respectively; and through measurements. In the experiments, a Tmote Sky sensor
d is the phase difference between these two paths. The model node is mounted vertically at a height of 70 cm from the ground
is verified by the field measurements at 1.4-m antenna heights as the transmitter and the signal strength values are measured
for both the transmitter and the receiver. Comparisons of the on a circle with a radius of 12 m in 15° polar angle separations.
proposed model with the free-space and two-ray models are These signal strength values are normalized, and the radiation
performed for different reflection coefficients. pattern is obtained. Indeed, K 1 and K 2 are determined by these
sets of normalized values.
FREE-SPACE OUTDOOR MODEL The proposed model is verified by the measurement at
In [72], a free-space outdoor model (FOM) is proposed where antenna heights of 0.12, 0.7, 1.5, and 1.97 m with Tmote Sky
the ground-reflection path, antenna pattern irregularity, and platforms equipped with 2.4-GHz CC2420 radios. The FOM
uncertainty in path loss are taken into consideration. FOM path- is a similar approach with the log-normal path-loss model. To
loss formulation can be expressed as calculate the average path loss, the two-ray model is used rather
than the curve fitting of measurement values used in the log-
Pr (d) = Pt c m m c K 21 + K 22 C 2 + 2K 2 C cos ( 2r DL) m, (16)
2
normal model.
4rd m
In another study [73], instead of utilizing a propagation
Pr (d) = Pr (d) + X v ( Pr), (17) model, a probability density function that characterizes the
relationship between the T–R distance and the reception rate in
where Pr is the mean value for the received signal; X v models WSNs is adopted. The measurements are performed at 433 and
the uncertainty in path loss; C is the ground-reflection coef- 916 MHz with antenna heights of 0, 0.08, 0.30, 0.90, and 1.50 m.
ficient; K 1 and K 2 are coefficients to reflect antenna pattern One of the important findings of [73] is that the existence of very
irregularities for LOS and reflected paths, respectively; and low reception rates with low T–R distances and high reception
DL is the difference between these two paths. Actually, (16) rates with large T–R distances is observed. Although this study
and (17) are constructed by synthesizing four components. The does not focus on propagation modeling, the observations made
first two components are free-space propagation and ground can be explained by using the characteristics of the two-ray
reflection, which are both analytical models. Indeed, the basic model (i.e., the nulls of the two-ray model).
form of (16) is created by using these two components, analyti- A summary of the aforementioned models is given in Table 1.
cally. The third component is the receive signal strength (RSS) Except for the free-space and two-ray models, all path mod-
uncertainty model that is given in (17) (i.e., X v ( Pr) ). To determine els in Table 1 include statistical descriptors to account for the
the RSS distribution, X v ( Pr), measurements are taken at different random variations in the path loss. More precisely stated, log-
antenna heights and distances. It is found that the RSS distribu- normal shadowing for the large-scale fading is incorporated into
tion follows the Gaussian distribution. Moreover, the standard the propagation model by employing a Gaussian random vari-
deviation of the Gaussian distribution is dependent on the mean able characterized by its mean and standard deviation. Utiliz-
signal strength value [i.e., v = f (Pr)]. More specifically, the rela- ing multiple Gaussian random variables to account for multiple
tion between the standard deviation and the mean value of the log-normal shadowing regimes provides better path-loss estima-
distribution can be expressed as the division of two third-order tion, as exemplified by the two-slope model where both slope

TABLE 1. A SUMMARY OF PATH-LOSS MODELS.

Antenna Reflection Antenna Random


Model Type Height Coefficient Pattern Variations Distance
free space [42] analytical No No No No yes
Two ray [19] analytical yes yes No No yes
One slope [49], [66], [67] empirical yes yes No yes yes
Two slope [66], [67] empirical yes yes No yes yes
VSr [70] analytical yes yes No No yes
fOM [72] analytical yes yes yes yes yes

8 february 2017 IEEE AntEnnAs & ProPAgAtIon MAgAzInE


This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

regimes have Gaussian random variables with different mean model. However, two-slope models are available only for 868-
and standard deviation values. and 2,400-MHz data [66], [67]. Only the one-slope model for
The antenna height dependence of the path loss is account- the 3,500-MHz data is available [49]. Despite this shortcoming,
ed for in all path-loss models in Table 1 except in the free- the models presented in [49] are useful in assessing the differ-
space model. Two-ray and VSR models take the antenna ence between the LOS and NLOS cases with different antenna
heights as inputs and incorporate their effects on the path loss heights. For the NLOS case, a direct path was obstructed by two
through mathematical formulations. One-slope, two-slope, rows of shrubs and small trees. The parameters of the one-slope
and FOM models do not incorporate antenna heights into the and two-slope models at different frequencies, sites, and antenna
path loss through mathematical relationships. Instead, the heights are presented in Tables 2 and 3, respectively.
measurements are performed at specific antenna heights, so The path-loss models we utilized in comparisons can be
the effects of the antenna heights are dictated by the experi- grouped into the following two classes:
mental setups themselves. 1) analytical models (free-space and two-ray models)
2) measurement-based models (one-slope and two-slope
QUANTITATIVE COMPARISONS OF PATH-LOSS MODELS models and generic log-normal shadowing).
FOR WSNs Analytical models can be conveniently used in the analysis of
Up to this point, we have introduced the fundamentals of path- WSNs without any scenario-specific parameters except for the
loss modeling in WSNs and presented the path-loss models antenna heights. However, the price paid for the convenience
developed for WSNs. In this section, we provide quantitative is the loss of accuracy. Measurement-based models necessitate
comparisons of various path-loss models for WSNs under dif- that the scenario-specific data are known, which is the price to
ferent scenarios and operating frequencies. Qualitative com- be paid for the accuracy.
parisons and assessments are valuable in evaluating the path-loss As elaborated in the sections “Variable Soil-Reflectivity
models. However, without quantitative comparisons through Model” and “Free-Space Outdoor Model,” the VSR and FOM
numerical evaluations, the comprehension of the subject will not models are based on the two-ray model. In fact, the difference
be complete. with these path-loss models is the incorporation of the reflec-
tion coefficient of the ground and antenna patterns into the
MATERIAL AND METHODS two-ray model. Therefore, these two models are not included in
Most of the WSN platforms available on the market use radio the figures. In the comparisons, for the two-ray model, instead
frequency transceivers working at 868 (Europe), 915 (United of a −1 reflection coefficient assumption, which is the usual
States), and 2,400 MHz in unlicensed industrial, scientific, and assumption, the reflection coefficient is calculated using (6) and
medical frequency bands [55], [74]–[77]. Licensed higher-fre- (7) with the typical relative permittivity of the ground taken as
quency bands are also envisioned to be used for WSNs through 15 [19]. To calculate the received power, instead of using the
the concept of cognitive radio, where secondary users can approximated two-ray model given in (16), direct-wave and
utilize licensed bands without creating interference for the pri- reflected-wave powers are calculated for all distances by taking
mary users [78]. For example, the 3,500-MHz frequency band the reflection coefficient and the antenna height parameters
licensed for Worldwide Interoperability for Microwave Access into account.
can be used for WSNs as secondary users. In Figures 2–8, the path-loss values for 868- and 2,400-
To assess the performances of propagation models to be used MHz bands are presented. Note that the two-slope model
for WSN applications, we present comparative performance is the base model used to compare the other models for the
evaluations by utilizing the measurement data available for 868- and 2,400-MHz cases in this study. Indeed, the path-loss
868-, 2,400-, and 3,500-MHz frequency bands. We adopt the estimation performance of the two-slope model is confirmed
measurement configuration parameters presented in [49], [66], by the measurement results represented in [66] and [67]. As
and [67] for numerical analysis. Indeed, the models obtained by shown in Tables 2 and 3, the v values of the two-slope model
curve fitting the measured data will serve as the base model in for all of the scenarios are smaller than those of the one-slope
evaluating the path-loss models. Curve fitting is done by mini- model, which indicates a better fit for the measurement results.
mizing the error between the measured data and the models In Figure 9, we present the path-loss values for the 3,500-MHz
in (12) (i.e., the one-slope model) and (14) (i.e., the two-slope band. The effects of LOS and NLOS propagation are clearly
model) through the least-squares method. In fact, the number of illustrated in Figure 9 [Figure 9(a) and (c) is for LOS cases, and
parameters and the level of detail are correlated with the accu- Figure 9(b) and (d) is for NLOS cases].
racy of the path-loss model. Being best fit for measured data, the
two-slope model is the most elegant model considered in this COMPARATIVE EVALUATIONS OF PATH-LOSS MODELS
study. The validation of the one-slope and two-slope models in The first comparison set is given for the three scenarios of
[66] shows that the utilization of the two-slope model enables [66] in Figure 2. Mica2 motes operating at 868 MHz are
much more realistic network topology reconstruction than the employed in measurements, and the antennas are 13 cm
one-slope model. Moreover, the verification of the one-slope and above the ground. The measurements are performed in three
two-slope models in [67] reveals that the two-slope model leads different scenarios/environments: 1) a fairly ideal flat ground
to a much lower path-loss estimation error than the one-slope (ground plain), 2) an open field surrounded by four-story

IEEE AntEnnAs & ProPAgAtIon MAgAzInE february 2017 9


This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

TABLE 2. THE ONE-SLOPE PATH-LOSS MODEL PARAMETERS [49], [66], [67].

Frequency Site h t /h r PL (d 0) n v

868 MHz Park 13 cm/13 cm 31 3.69 1.42


868 MHz Ground plain 13 cm/13 cm 31 3.12 1.83
868 MHz university yard 13 cm/13 cm 31 3.57 3.27
2,400 MHz Plaza 1 m/1 m 42.86 1.86 3.25
2,400 MHz Plaza 3 cm/1 m 51.51 2.21 4.34
2,400 MHz Plaza 3 cm/2 m 53.81 1.86 4.62
2,400 MHz Sidewalk 1 m/1 m 40.36 1.99 2.45
2,400 MHz Sidewalk 3 cm/1 m 50.91 2.47 5.74
2,400 MHz Sidewalk 3 cm/2 m 52.49 2.14 5.23
2,400 MHz Grassland 1 m/1 m 40.51 1.90 3.02
2,400 MHz Grassland 3 cm/1 m 51.08 2.48 5.19
2,400 MHz Grassland 3 cm/2 m 53.56 2.08 5.59
3,500 MHz LOS 0.3 m/2.1 m 43.3 2.0 1.3
3,500 MHz NLOS 0.3 m/2.1 m 43.3 2.8 4.3
3,500 MHz LOS 2.1 m/2.1 m 43.3 2.1 1.3
3,500 MHz NLOS 2.1 m/2.1 m 43.3 2.6 4.1

TABLE 3. THE TWO-SLOPE PATH-LOSS MODEL PARAMETERS [66], [67].

Frequency Site h t /h r PL (d 01) PL (d 02) n1 n2 v1 v2 db


868 MHz Park 13 cm/13 cm 33 23 2.09 4.01 0.28 0.67 0.95
868 MHz Ground plain 13 cm/13 cm 33 23 2.34 3.73 0.60 0.42 6.2
868 MHz university yard 13 cm/13 cm 33 23 2.76 4.00 2.98 1.82 3.2
2,400 MHz Plaza 1 m/1 m 67.68 65.61 1.52 3.74 2.49 1.85 32
2,400 MHz Plaza 3 cm/1 m 57.29 60.26 1.00 2.91 1.99 2.94 5
2,400 MHz Plaza 3 cm/2 m 60.54 61.36 1.06 2.87 1.75 2.25 7
2,400 MHz Sidewalk 1 m/1 m 69.29 67.84 1.55 3.34 3.09 1.55 38
2,400 MHz Sidewalk 3 cm/1 m 55.35 57.36 1.01 3.31 2.02 3.47 4
2,400 MHz Sidewalk 3 cm/2 m 58.87 61.44 1.05 3.29 1.69 2.01 7
2,400 MHz Grassland 1 m/1 m 65.03 66.09 1.26 3.93 2.98 1.51 38
2,400 MHz Grassland 3 cm/1 m 55.69 57.57 0.65 3.62 1.37 1.90 5
2,400 MHz Grassland 3 cm/2 m 62.36 64.04 1.14 3.43 1.39 1.42 9

buildings (university yard), and 3) a slightly sloped grassland models, the free-space and two-ray models, are also presented.
(park). These scenarios are useful in observing the effects of It is possible to use the two-ray model with both parallel polar-
various propagation mechanisms on the path loss. In fact, the ization [using (6)] and perpendicular polarization [using (7)].
ground plain scenario is useful mainly in observing the effects Hence, the path-loss curves for both polarizations are avail-
of reflection. The park scenario is useful in observing the able in the figures. The result for the log-normal shadowing
effects of scattering in addition to reflection. The university path-loss model with generic values given in [65] is included.
yard scenario adds the multipath effects from surrounding This model is similar to the one-slope model with parameters
buildings to the other scenarios. PL (d 0) = 55 dB, n = 4, and v = 4. However, the effects of
Figure 2 presents the path-loss curves as functions of the dis- the antenna heights or the operating frequency are not taken
tance between the transmitter and the receiver. Along with the into account. Hence, it is a generic WSN path-loss model, and
one-slope and two-slope path-loss models, two pure analytical the inclusion of this model into the comparisons will reveal

10 february 2017 IEEE AntEnnAs & ProPAgAtIon MAgAzInE


This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

–20 –20
–40 –40
Received Power (dBm)

Received Power (dBm)


–60 –60
–80 –80
–100 –100
–120 –120
–140 –140
–160 –160
–180 –180
0 1 2 3
10 10 10 10 100 101 102 103
Distance (m) Distance (m)
(a)
Ground Plain Park University Yard
–20
–40 FIGURE 3. a scenario comparison for the two-slope model at
Received Power (dBm)

868 MHz.
–60
–80
A holistic view of the path-loss characteristics of the models
–100
by considering the three environments in Figure 2 reveals that
–120 there are three distinct groups of path-loss models. The first
–140 group is formed by two of the pure theoretical models (i.e., the
–160 free-space and two-ray parallel-polarization models), which con-
sistently underestimate the path loss in all scenarios. The two-
–180
slope, one-slope, and two-ray perpendicular-polarization models
100 101 102 103 form the second group, where the intragroup differences are
Distance (m) significantly lower than the intergroup path-loss differences as
(b) a general trend. The third group, in fact, has only one member,
which is the log-normal shadowing model in [65], with scenario-
–20
independent parameters that overestimate the path loss in all
–40 environments.
Received Power (dBm)

–60 Taking the two-slope model as the reference, the two-ray


–80 model with perpendicular polarization is the most realistic
theoretical model. However, if the electric field polariza-
–100
tion is chosen as the parallel polarization, then the two-ray
–120
model leads to significant path-loss estimation errors (e.g.,
–140 as high as 30 dB). Note that the one-slope and log-normal
–160 shadowing models are based on the same path-loss formula
–180 (12). However, in the one-slope model, the parameters in
the formula are chosen by the curve fitting of the measure-
100 101 102 103
ment data; yet, in the log-normal shadowing model, the
Distance (m)
parameters are fixed (i.e., the generic parameter values are
(c)
employed). The one-slope model gives path-loss values in
Two-Ray (Perpendicular Polarized) close proximity to the two-slope model, whereas the log-
Two-Ray (Parallel Polarized) normal shadowing model leads to significant path-loss esti-
Log-Normal Shadowing mation errors. Therefore, the utilization of the environment
Free Space One Slope Two Slope and scenario-independent path-loss models can result in
large path-loss estimation errors.
FIGURE 2. The path loss at 868 MHz (with transmitter and In Figure 3, the path-loss curves obtained by using the
receiver antenna heights of 13 cm): (a) the ground plain, two-slope model for the three scenarios considered are
(b) park, and (c) university yard.
presented as functions of the transmitter and receiver sepa-
the suitability of employing a generic path-loss model in WSNs ration. This figure is used to analyze whether the path-loss
without considering the effects of the specific propagation envi- parameters measured at one site can be employed at another
ronment and the application scenario. site. The answer we get from Figure 3 is that the level of

IEEE AntEnnAs & ProPAgAtIon MAgAzInE february 2017 11


This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

–20
–20
–40

Received Power (dBm)


–40
Received Power (dBm)

–60
–60
–80
–80
–100
–100
–120
–120
–140
–140
–160
–160
–180
–180
100 101 102 103
100 101 102 103
Distance (m)
Distance (m)
(a)
(a)

–20
–20
–40
–40

Received Power (dBm)


Received Power (dBm)

–60
–60
–80
–80
–100
–100
–120
–120
–140
–140
–160
–160
–180
–180
100 101 102 103
100 101 102 103
Distance (m)
Distance (m)
(b)
(b)

–20 –20
–40 –40
Received Power (dBm)
Received Power (dBm)

–60 –60
–80 –80
–100 –100
–120 –120
–140 –140
–160 –160
–180 –180

100 101 102 103 100 101 102 103


Distance (m) Distance (m)
(c) (c)

Two-Ray (Perpendicular Polarized) Free Space One Slope


Two-Ray (Parallel Polarized) Two Slope Two Slope
Log-Normal Shadowing

FIGURE 4. The analytic path loss at 2.4 GHz for different FIGURE 5. The measurement-based path loss at 2.4 GHz
transmitter and receiver antenna heights (grassland for different transmitter and receiver antenna heights
scenario). a measurement-based two-slope model is also (grassland scenario). The two-slope model can be taken as
included as the reference. (a) Path loss 1 m/1 m, (b) path loss the reference. (a) Path loss 1 m/1 m, (b) path loss 3 cm/1 m,
3 cm/1 m, and (c) path loss 3 cm/2 m. and (c) path loss 3 cm/2 m.

12 february 2017 IEEE AntEnnAs & ProPAgAtIon MAgAzInE


This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

–20 –20
–40 –40

Received Power (dBm)


Received Power (dBm)

–60 –60
–80 –80
–100 –100
–120 –120
–140 –140
–160 –160
–180 –180

100 101 102 103 100 101 102 103


Distance (m) Distance (m)
(a) (a)

–20 –20
–40 –40

Received Power (dBm)


Received Power (dBm)

–60 –60
–80 –80
–100 –100
–120 –120
–140 –140
–160 –160
–180 –180

100 101 102 103 100 101 102 103


Distance (m) Distance (m)
(b) (b)

–20 –20
–40 –40
Received Power (dBm)
Received Power (dBm)

–60 –60
–80 –80
–100 –100
–120 –120
–140 –140
–160 –160
–180 –180

100 101 102 103 100 101 102 103


Distance (m) Distance (m)
(c) (c)

Two-Ray (Perpendicular Polarized) Free Space Two-Ray (Perpendicular Polarized) Free Space
Two-Ray (Parallel Polarized) Two Slope Two-Ray (Parallel Polarized) Two Slope

FIGURE 6. The analytic path loss at 2.4 GHz for different FIGURE 7. The analytic path loss at 2.4 GHz for different
transmitter and receiver antenna heights (plaza scenario). transmitter and receiver antenna heights (sidewalk scenario).
The measurement-based two-slope model is also included as The measurement-based two-slope model is also included as
the reference. (a) Path loss 1 m/1 m, (b) path loss 3 cm/1 m, the reference. (a) Path loss 1 m/1 m, (b) path loss 3 cm/1 m,
and (c) path loss 3 cm/2 m. and (c) path loss 3 cm/2 m.

IEEE AntEnnAs & ProPAgAtIon MAgAzInE february 2017 13


This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

where the transmitter and the receiver always have a direct


–20 LOS path). For example, the path-loss values of the three
cases are within 5- and 10-dB neighborhoods at 10 and 100
–40
m, respectively.
Received Power (dBm)

–60
To observe the effects of the operating frequency on the
–80 path-loss models, the rest of this section presents a compara-
–100 tive analysis using the three scenarios of [67] conducted at
–120 2.4 GHz. The measurement results for three different trans-
mitter and receiver antenna heights (i.e., h t /h r are 1 m/1 m,
–140
3 cm/1 m, and 3 cm/2 m) are presented. The three environ-
–160 ments used for the measurements are a large plaza, a straight
–180 sidewalk, and an open grassland. The plaza is a rectangular-
100 101 102 103
shaped site (an area of 115 m 2 # 100 m 2 ) bordered by trees
Distance (m) and several buildings. The second site is a straight sidewalk
(a) along a suburban road, and the other side is a large plot of
lowland with scattered trees. The sidewalk is a fairly flat ter-
–20 rain paved with pitch and bricks. The grassland is a flat land
–40 consisting of soil and sand.
Received Power (dBm)

In Figures 4 and 5, the path-loss values obtained in the


–60
grassland scenario for both the theoretical and measurement-
–80 based models are presented as functions of the transmitter
–100 and receiver distance. Because plotting all models in the same
–120 figure makes the curves indistinguishable, we opt to present the
–140
theoretical models in Figure 4 and the experimental models in
Figure 5. The two-slope model is plotted in both figures as the
–160
reference. The two-ray model with perpendicular polarization
–180 is in close proximity to the two-slope model, provided that the
100 101 102 103 transmission distance is larger than 10 m. The best path-loss
Distance (m) estimation performance of the two-ray model is attained when
(b) both the transmitter and receiver antenna heights are 1 m. The
measurements at 868 MHz are carried out when both transmit-
–20 ter and receiver antennas are on the ground; conversely, in the
–40 2.4-GHz case, the antennas are never simultaneously at the
Received Power (dBm)

–60 ground level (such measurement data sets are lacking in the
WSN literature). Therefore, the two-ray model with perpen-
–80
dicular polarization results in lower path-loss estimation errors
–100
than the one-slope model at 2.4 GHz, in general, unlike the 868-
–120 MHz case. The two-ray parallel-polarization, free-space, and
–140 log-normal shadowing models with generic values produce high
–160 errors, so they are not suitable to be used in this scenario.
Figures 6 and 7 present the path-loss values for only the
–180
two-slope model, the free-space model, and both polarizations
100 101 102 103 of the two-ray model in the plaza and sidewalk scenarios. We
Distance (m) do not present the path-loss curves for the one-slope model
(c) and the log-normal shadowing model with generic parameters
Plaza because they do not exhibit significant differences compared to
Sidewalk their characteristics in Figures 4 and 5. The two-ray model with
Grassland
perpendicular polarization follows the two-slope model more
closely in both the plaza and sidewalk scenarios than it does in
FIGURE 8. a 2.4-GHz scenario comparison for the two-slope the grassland scenario as long as the transmission distance is
model: (a) path loss (1 m, 1 m), (b) path loss (3 cm, 1 m), and
larger than 10 m.
(c) path loss (3 cm, 2 m).
When we consider Figures 4–7 jointly, we can make
the following observations. The two-ray model is a good
precision required is the determining factor, provided that choice when both the transmitter and receiver antennas
the obstruction characteristics and densities do not differ sig- are 1 m high. Therefore, if there are no measurements
nificantly between the sites (as in the case of our scenarios, available and the antennas are high enough from the

14 february 2017 IEEE AntEnnAs & ProPAgAtIon MAgAzInE


This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

–10 –10
–20 –20
–30 –30
Received Power (dBm)

Received Power (dBm)


–40 –40
–50 –50
–60 –60
–70 –70
–80 –80
–90 –90
–100 –100
–110 –110
100 101 102 103 100 101 102 103
Distance (m) Distance (m)
(a) (b)

–10 –10
–20 –20
–30 –30
Received Power (dBm)

Received Power (dBm)


–40 –40
–50 –50
–60 –60
–70 –70
–80 –80
–90 –90
–100 –100
–110 –110
100 101 102 103 100 101 102 103
Distance (m) Distance (m)
(c) (d)

Two-Ray (Parallel Polarized) Free Space One Slope

FIGURE 9. a 3.5-GHz scenario comparison for LOS–NLOS: (a) LOS path loss (0.3 m, 2.1 m), (b) NLOS path loss (0.3 m, 2.1 m),
(c) LOS path loss (2.1 m, 2.1 m), and (d) NLOS path loss (2.1 m, 2.1 m).

ground (e.g., at least a few wavelengths), then the two-ray and (b), the path-loss values are in the 10- and 20-dB neigh-
model that includes the ground-reflection coefficient with borhood of each other, respectively.
correct polarization, antenna gains, and heights gives The better performance of the analytical models with increas-
a fairly good estimate of the actual path loss. However, ing antenna heights from the ground is also observed in the 3,500-
if one of the antennas is very close to the ground, then MHz case (Figure 9). The differences between the analytical
the accuracy of the two-ray model deteriorates. Another models (the two-ray and free-space models) and the empirical
observation is that the breakpoints in the two-slope mod- model (the one-slope model) reduce as the transmitter antenna
els are mostly antenna height dependent. height is increased from 0.3 m [Figure 9(a) and (b)] to 2.1 m
In Figure 8, the path-loss values obtained with the two- [Figure 9(c) and (d)] for both the LOS and NLOS cases. Further-
slope model for different scenarios at the same antenna more, the difference between the analytical models and the empiri-
heights are presented [e.g., in Figure 8(a), the path loss cal model is higher in the NLOS cases compared to the LOS cases.
obtained with 1-m-high transmitter and receiver antennas for For easy navigation of the results from the figures, the main
the plaza, grassland, and sidewalk scenarios are presented]. As findings and observations on the figures presented in the “Com-
in the 868-MHz case, the option of utilizing the parameters parative Evaluations of Path-Loss Models” section are summa-
for one site to estimate another site’s path-loss characteristics rized as follows.
is dependent on the level of precision required. However, if 1) Figure 2 shows the path loss at the 868-MHz band for the
the antennas are very close to the ground, then the extent analytical and empirical models. Both models with unsuitable
of the path-loss difference is higher compared to the cases parameters can lead to large errors when both antennas are
where antenna heights are high. For example, in Figure 8(a) on the ground.

IEEE AntEnnAs & ProPAgAtIon MAgAzInE february 2017 15


This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

2) Figure 3 shows the differences between the two-slope model breaches because a certain portion of the area to be monitored
at the 868-MHz band for different scenarios. The level of is left wide open for intruders.
precision is the determining factor in utilizing an accurate We assumed each node transmits at 0 dBm and the receiver
model for one site as the path-loss model on another site. sensitivity is −101 dBm (i.e., if the received power is below
3) Figures 4 and 5 show the path loss at 2.4 GHz in the grass- −101 dBm on a link between two nodes, then those nodes can-
land scenario for the analytical and empirical models, not communicate) in our experiments [80]. For fair comparison,
respectively. The two-ray model with perpendicular polar- we apply these parameters to all three bands (i.e., 868 MHz
ization is the analytical model with the best performance, and 2.4 and 3.5 GHz). We assumed a disk-shaped network
especially for higher antenna heights [i.e., (1 m, 1 m) case]. deployment area (the network radius is denoted as R net ) with
4) Figures 6 and 7 show the analytical path-loss models for the 100 nodes. The sensor nodes are deployed by using a random
plaza and sidewalk scenarios at the 2.4-GHz band, respec- uniform distribution, and the base station is positioned at the
tively. The free-space two-ray parallel-polarization models center of the disk. The results are averaged for 100 independent
can lead to large errors. The breakpoint distance is mostly node deployments and path-loss assignments to the links for
antenna height dependent rather than scenario dependent. each path-loss model.
5) Figure 8 shows the differences between the two-slope model In Table 4, the average source-to-sink path length in terms
at the 2.4-GHz band for different scenarios. The path-loss of hops (N H) , the average number of neighbors per node (N N),
estimation error in the scenario mismatch is higher for lower and the average value of the maximum transmission range
antenna heights. (d max) are presented for various scenarios using two-slope,
6) Figure 9 shows the path loss at 3.5 GHz for the LOS and one-slope, free-space, two-ray perpendicular-polarization, and
NLOS cases with different antenna heights. The theoretical two-ray parallel-polarization path-loss models. In Table 5,
model estimation errors are higher in the NLOS cases. the average percentage of disconnected nodes (p discon) is pre-
sented. In the scenarios presented in Table 4, the network
IMPACT OF PATH-LOSS MODELS ON WSN PERFORMANCE radius values are chosen in such a way that p discon is below
ASSESSMENT 1% because, when p discon is high (e.g., 50%), the N H values
In the preceding section, we presented a comparative evaluation do not reflect the actual network performance (e.g., in high
of the path-loss values for different path-loss models. In this sec- p discon networks, only the nodes closer to the base station can
tion, we present an analysis of the impact of path-loss models on establish paths to the base station, and the N H values of these
WSN performance. In the WSN scenario we considered, sensor paths are too low, so the reliability of the N H metric is lost
nodes are required to send data to the base station via minimum in high p discon network topologies). Nevertheless, we present
hop routing either directly or through other sensor nodes act- only the p discon metric without presenting the N H and N N met-
ing as relays (i.e., multihop routing). Minimum hop routing is a rics in Table 5 for the scenarios with p discon higher than 90.0%.
well-known and widely utilized routing approach in WSNs [5], In each scenario, the network radius values are also given.
[79]. In minimum hop routing, each sensor node sends its data We first investigate the effects of LOS and NLOS propaga-
toward the base station by utilizing the minimum hop route. tion environments for the 3.5-GHz band on the WSN perfor-
The network performance metrics we considered are the mance metrics. For the NLOS case, R net = 600 m, whereas for
average hop count per node, the average number of neighbor the LOS case, R net = 900 m. These values are limiting in the
nodes per node, and the average number of disconnected sense that p discon is starting to surpass 1% if the R net values are
nodes in the network. Certain nodes get disconnected because chosen slightly larger. In the NLOS case, all analytical models
no path can be found to the sink. The average number of fail to predict the N H , N N , and d max values with significant
hops is a direct indicator of the transmission and reception errors because these models lack the ability to incorporate the
energy dissipation in the network. For example, in a five-hop effects of obstacles in the propagation path. However, in the
path, each data packet is transmitted and received five times, LOS case, the estimated values of the metrics obtained with
whereas in a single-hop path, one transmission and one recep- the two-ray model with perpendicular polarization are very
tion is enough (i.e., the energy dissipation for communication close to those obtained with the one-slope model. Yet, the use
in a five-hop route is five times the energy dissipation in a of the free-space and two-ray parallel-polarization model still
single-hop path given a constant transmission power level for leads to significant errors. The network disconnection fails to
all nodes). Furthermore, the packet delay also increases with be modeled accurately by all of the analytical models for the
an increasing hop count. Therefore, the increase in the hop NLOS case, whereas the two-ray model with perpendicular
count is undesirable in WSNs. The reliability of a WSN is high- polarization gives a p discon value very close to the one-slope
er when the average number of neighbors is higher because, model for the LOS case.
in case of node failures, the creation of alternative routes Second, we investigate the effects of antenna height varia-
depends on the availability of relay nodes, which are more tions for the 2.4-GHz band on the WSN performance met-
ample if the average number of neighbors is high. Network rics. For all antenna heights, the estimations of N H , N N , d max,
disconnection has severe consequences in WSNs. For example, and p discon by the two-slope and one-slope models differ
in a border-monitoring application, the loss of communica- significantly. The two-ray model with perpendicular polariza-
tion with a subset of sensor nodes creates significant security tion gives remarkably close results to the two-slope model.

16 february 2017 IEEE AntEnnAs & ProPAgAtIon MAgAzInE


This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

TABLE 4. THE AVERAGE NUMBER OF HOPS AND NEIGHBORS FOR DIFFERENT


SCENARIOS AND PATH-LOSS MODELS.

Two-Ray Two-Ray
Perpendicular Parallel
Two Slope One Slope Free Space Polarized Polarized
Rnet NH NN dmax NH NN dmax NH NN dmax NH NN dmax NH NN dmax
NLOS, 600.0 — — — 2.7 8.3 172.5 1.0 77.8 765.0 1.1 55.4 575.0 1.0 96.2 1,032.0
2.1–2.1 m,
3,500 MHz
LOS, 900.0 — — — 1.6 29.1 570.8 1.3 46.5 765.0 1.6 29.9 575.0 1.0 68.5 1,032.0
2.1–2.1 m,
3,500 MHz
Plaza, 800.0 2.7 10.8 285.8 1.0 86.0 1,460.9 1.0 84.7 1,116.1 2.8 10.7 285.0 1.0 99.0 1,582.0
1–1 m,
2,400 MHz
Plaza, 400.0 2.8 9.3 230.8 1.8 19.7 195.4 1.0 99.0 1,116.1 5.3 4.6 91.0 1.0 99.0 1,585.0
3 cm–1 m,
2,400 MHz
Plaza, 600.0 3.5 7.3 178.8 1.6 32.5 432.1 1.0 98.5 1,116.1 4.4 3.4 115.0 1.0 99.0 1,583.0
3 cm–2 m
2,400 MHz
Ground 400.0 3.1 9.2 131.0 2.0 18.2 185.2 1.0 99.0 3,462.0 3.5 7.7 110.0 1.0 99.0 4,384.0
plain,
868 MHz
Park, 100.0 2.6 11.7 37.2 2.2 14.8 41.9 1.0 99.0 3,462.0 1.0 72.4 110.0 1.0 99.0 4,384.0
868 MHz
university 100.0 3.1 8.8 31.9 2.1 14.7 41.9 1.0 99.0 3,462.0 1.0 72.4 110.0 1.0 99.0 4,384.0
yard,
868 MHz

Especially at higher antenna heights (i.e., both antennas are estimating the metrics leads to significant errors (worse than
at 1-m height), the two-ray model with perpendicular polar- the two-ray model with perpendicular polarization).
ization produces results in close proximity to those of the In summary, the use of the free-space and two-ray paral-
two-slope model. Both the free-space and two-ray parallel- lel-polarization models is definitely not suitable in path-loss
polarization models fail to give accurate estimations of N H , estimation. The two-ray model with perpendicular polariza-
N N , d max , and p discon . tion gives satisfactory results if the propagation path is not
Third, we investigate the effects of different propagation NLOS. The one-slope model produces mixed results for dif-
environments with very low antenna heights for the 868-MHz ferent scenarios. Therefore, the use of the two-slope model is
band on the WSN performance metrics. In all environments, the best option.
both the free-space and two-ray parallel-polarization models
fail to provide accurate estimations of any metrics. In fact, CONCLUSIONS AND OPEN ISSUES
huge differences between the estimations using both the Path-loss modeling is of the utmost importance for the reliable
free-space and two-ray parallel-polarization models compared and accurate analysis of WSNs; yet, in many studies on WSNs,
with the two-slope model exist in all scenarios. The two-ray the use of inappropriate and simplistic path-loss models is
model with perpendicular polarization produces results close employed, leading to unrealistic assessments of the problems
to the two-ray model only for the ground plain because, in all under investigation. Our starting point in the creation of this
868-MHz environments, the ground plain is closest to being article is to provide the necessary but succinct background
an LOS scenario. However, in the park and university yard on the basics of the path loss in wireless communications and
scenarios, the two-ray model with perpendicular polarization to briefly introduce the mechanisms of the free-space, two-
also fails to accurately model the path loss. The one-slope ray, log-distance, and shadowing models. We summarize the
model gives the best results (compared with the two-slope constraints imposed by the WSN applications and scenarios
model) for the park and university yard scenarios. For the on propagation modeling and explore the path-loss models
ground plain scenario, the use of the one-slope model in used in WSN research. To give a more tangible picture of the

IEEE AntEnnAs & ProPAgAtIon MAgAzInE february 2017 17


This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

TABLE 5. THE AVERAGE PERCENTAGE OF DISCONNECTED NODES FOR


DIFFERENT SCENARIOS AND PATH-LOSS MODELS.

Two-Ray Two-Ray
Perpendicular Parallel
Two Slope One Slope Free Space Polarized Polarized
Rnet pdiscon dmax pdiscon dmax pdiscon dmax pdiscon dmax pdiscon dmax
NLOS, 2.1–2.1 m, 1,500.0 — — 94.5 172.5 0.0 765.0 0.0 575.0 0.0 1,032.0
3,500 MHz
LOS, 2.1–2.1 m, 4,500.0 — — 96.0 570.8 83.3 765.0 96.4 575.0 21.5 1,032.0
3,500 MHz
Plaza, 1–1 m, 2,000.0 94.2 285.8 0.0 1,460.9 0.0 1,116.1 94.8 285.0 0.0 1,582.0
2,400 MHz
Plaza, 3 cm–1 m, 1,100.0 96.5 230.8 20.1 195.4 0.0 1,116.1 98.4 91.0 0.0 1,585.0
2,400 MHz
Plaza, 3 cm–2 m, 1,200.0 92.4 178.8 0.0 432.1 0.0 1,116.1 97.9 115.0 0.0 1,583.0
2,400 MHz
Ground plain, 900.0 94.0 131.0 34.9 185.2 0.0 3,462.0 96.1 110.0 0.0 4,384.0
868 MHz
Park, 868 MHz 300.0 96.5 37.2 94.5 41.9 0.0 3,462.0 0.0 110.0 0.0 4,384.0
university yard, 300.0 97.5 31.9 92.7 41.9 0.0 3,462.0 0.0 110.0 0.0 4,384.0
868 MHz

differences between the path-loss models, we provide quantita- different scenario, such as different frequencies or
tive comparisons of these models under various scenarios. The antenna heights) would produce significant path-loss
highlights of this article are as follows. estimation errors.
1) Before using a particular path-loss model, its applicability to 5) If it is not possible to conduct experiments to determine the
the specific WSN case under consideration should be two-slope model’s parameters for a particular scenario and if
checked. An awareness of the assumptions made for the there are no data reported in literature that closely represent
chosen model and the suitability for the application scenario the scenario, then the two-ray model with correct polariza-
must be established. While employing any path-loss model, tion should be employed.
the incorporation of salient scenario-dependent parameters, 6) To apply the two-ray model with higher accuracy, both h t
such as antenna height, wavelength, antenna polarization, and h r should be included in the utilized model, and it
and ground reflectivity, should not be ignored. Adding should be ensured that the antenna heights are larger than at
parameters that are crucial for the application improves the least a few wavelengths.
accuracy of the used model. 7) The utilization of the free-space, two-ray incorrect-polar-
2) Most of the observations on measured data can be ization, and generic models (e.g., the log-normal shadow-
explained by suitable theoretical models. Specifically, for ing model in [65] with scenario-independent parameters)
the unobstructed outdoor WSN case, this model can be should be avoided so that significant path-loss estimation
the two-ray model. Especially high path-loss regions with errors are not made.
small T–R distances can be explained by nulls in the There are several open issues to be explored in path-loss
model that result from the destructive addition of ground- modeling for WSNs.
reflected waves. However, the performance of the two-ray 1) There are no field measurements available to characterize
model deteriorates significantly for the NLOS propaga- the path loss for WSNs at 868 MHz with different transmit-
tion paths. ter and receiver antenna heights (e.g., h t /h r are 1 m/1 m,
3) The most accurate currently available model to character- 3 cm/1 m, and 3 cm/2 m).
ize the path loss in WSNs is the two-slope model. But keep 2) Path-loss measurements for WSNs operating at 2.4 GHz
in mind that the parameters of the model should be with both transmitter and receiver antennas placed on the
assigned by curve fitting the measured data obtained from ground (e.g., 3 cm/3 cm) are missing in the literature.
field experiments. 3) Experimental evaluations with different antenna polariza-
4) Utilizing the two-slope model with parameters that do tions in WSNs are also an open research avenue.
not match the characteristics of the deployment consid- 4) The characterization of the path loss in WSNs with mobile
ered (e.g., by utilizing the parameters obtained for a base stations and/or mobile nodes with low antenna heights

18 february 2017 IEEE AntEnnAs & ProPAgAtIon MAgAzInE


This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

through direct experimentation is another insufficiently Conf. Systems, Signals, and Image Processing, 2009, pp. 1–4.
[5] K. Akkaya and M. Younis, “A survey on routing protocols for wireless sensor
explored topic. networks,” Ad Hoc Netw., vol. 3, no. 3, pp. 325–349, May 2005.
5) Unlike the other areas of wireless communications, there is [6] G. Zhou, T. He, S. Krishnamurthy, and J. A. Stankovic, “Impact of radio
no standard path-loss model for WSNs even after a more irregularity on wireless sensor networks,” in Proc. Int. Conf. Mobile Systems,
Applications, and Services, 2004, pp. 125–138.
than decade-long intensive worldwide research and develop-
[7] H. Cotuk, K. Bicakci, B. Tavli, and E. Uzun, “The impact of transmission
ment effort, which hinders the comparability of research power control strategies on lifetime of wireless sensor networks,” IEEE Trans.
results. Therefore, a set of path-loss models for certain WSN Comput., vol. 63, no. 11, pp. 2866–2879, 2014.
[8] H. U. Yildiz, B. Tavli, and H. Yanikomeroglu, “Transmission power control for
environments/scenarios, agreed upon by international
link level handshaking in wireless sensor networks,” IEEE Sensors J., vol. 16, no. 2,
authorities like the International Telecommunications pp. 561–576, Jan. 2016.
Union, would be invaluable. [9] Y. Okumura, E. Ohmori, T. Kawano, and K. Fukua, “Field strength and its
variability in UHF and VHF land mobile radio service,” Rev. Electr. Commun.
As discussed in the “WSN Overview” section, the utiliza-
Lab., vol. 16, nos. 9–10, pp. 825–873, Sept.–Oct. 1968.
tion of accurate path-loss models in simulations and analysis of [10] M. Hata, “Empirical formula for propagation loss in land mobile radio
WSNs is necessary for realistic performance assessment and services,” IEEE Trans. Veh. Technol., vol. 29, no. 3, pp. 317–325, Aug. 1980.
characterization. As an example scenario, let’s consider a WSN [11] COST Telecommunications, “COST Action 231: Digital mobile radio
towards future generation systems—Final report,” European Commission,
deployed over farmland for monitoring the soil conditions. Brussels, Belgium, Rep. EUR 18957, 1999.
The antennas of the nodes of such a WSN should be over the [12] V. Erceg, L. Greenstein, S. Tjandra, S. Parkoff, A. Gupta, B. Kulic, A.
ground. Therefore, the path losses among the nodes cannot Julius, and R. Bianchi, “An empirically based path loss model for wireless
channels in suburban environments,” IEEE J. Sel. Areas Commun., vol. 17,
be estimated accurately by analytical models (i.e., free space no. 7, pp. 1205–1211, July 1999.
or two ray). Hence, assessing the WSN performance metrics [13] V. Erceg, K. V. S. Hari, M. S. Smith, D. S. Baum, K. P. Sheikh, C. Tap-
based on huge path-loss estimation errors would result in highly penden, J. M. Costa, C. Bushue, A. Sarajedini, R. Schwartz, D. Branlund,
T. Kaitz, and D. Trinkwon, “Channel models for fixed wireless applications,”
ineffective (if not nonfunctional) WSN deployments. Further- IEEE 802.16 Broadband Wireless Access Working Group, Tech. Rep. IEEE
more, maintaining a connected network topology is a vital 802.16.3c-01/29r4, 2001.
functionality in WSNs. If the path loss among the sensor nodes [14] International Telecommunications Union, “Guidelines for evaluation of
radio interface technologies for IMT-advanced,” ITU, Geneva, Switzerland,
is overestimated, then the number of sensor nodes required Rep. M.2135-1, 2009.
to be deployed will also be overestimated. On the other hand, [15] V. Abhayawardhana, I. Wassell, D. Crosby, M. Sellars, and M. Brown,
if the path-loss values are underestimated, then the network “Comparison of empirical propagation path loss models for fixed wireless access
systems,” in Proc. IEEE Vehicular Technology Conf., 2005, pp. 73–77.
connectivity cannot be established for some sensor nodes. As [16] A. Molisch, Wireless Communications. Hoboken, NJ: Wiley, 2005.
demonstrated in the “Impact of Path-Loss Models on WSN [17] International Telecommunications Union, “Propagation data and prediction
Performance Assessment” section, the utilization of inappropri- methods for the planning of short-range outdoor radiocommunication systems
and radio local area networks in the frequency range 300 MHz to 100 GHz,”
ate path-loss models results in huge performance assessment ITU, Geneva, Switzerland, Rep. ITU-R P.1411-8, 2015.
errors in the average path length (in terms of hops), the number [18] D. Kotz, C. Newport, R. S. Gray, J. Liu, Y. Yuan, and C. Elliott, “Experimen-
of neighbors, and the connectedness of the network. As exempli- tal evaluation of wireless simulation assumptions,” in Proc. ACM Int. Symp. Mod-
eling, Analysis and Simulation of Wireless and Mobile Systems, 2004, pp. 78–83.
fied by the preceding examples, the accurate analysis of WSN
[19] T. Rappaport, Wireless Communications: Principles and Practice, 2nd ed.
architectures and deployments is not possible without employ- Englewood Cliffs, NJ: Prentice Hall, 2001.
ing proper radio propagation models for target scenarios. [20] T. K. Sarkar, Z. Ji, K. Kim, A. Medouri, and M. Salazar-Palma, “A survey
of various propagation models for mobile communication,” IEEE Antennas
Propag. Mag., vol. 45, no. 3, pp. 51–82, June 2003.
AUTHOR INFORMATION [21] M. Iskander and Z. Yun, “Propagation prediction models for wireless communication
Sinan Kurt (skurt@etu.edu.tr) received his Ph.D. degree in systems,” IEEE Trans. Microw. Theory Techn., vol. 50, no. 3, pp. 662–672, Mar. 2002.
[22] J. K. Jacoub, R. Liscano, and J. S. Bradbury, “A survey of modeling tech-
electrical and electronics engineering from the TOBB University
niques for wireless sensor networks,” in Proc. Int. Conf. Sensor Technologies and
of Economics and Technology, Ankara, Turkey. He is a lead Applications, 2011, pp. 103–109.
design engineer at ASELSAN Inc., Ankara. His research inter- [23] C. F. Garcia-Hernandez, P. H. Ibarguengoytia-Gonzalez, J. Garcia-Hernandez,
and J. A. Perez-Diaz, “Wireless sensor networks and applications: A survey,” Int. J.
ests are in wireless communications, electromagnetics, circuit
Comput. Sci. Netw. Security, vol. 7, no. 3, pp. 264–273, Mar. 2007.
design, and optimization. [24] M. Korkalainen and M. Sallinen, “A survey of RF-propagation simulation
Bulent Tavli (btavli@etu.edu.tr) is a professor in the Depart- tools for wireless sensor networks,” in Proc. Int. Conf. Sensor Technologies and
ment of Electrical and Electronics Engineering, TOBB Universi- Applications, 2010, pp. 342–347.
[25] M. Jevtic, N. Zogovic, and G. Dimic, “Evaluation of wireless sen-
ty of Economics and Technology, Ankara, Turkey. His current sor network simulators,” in Proc. Telecommunications Forum, 2009, pp.
research areas are wireless communications, networking, optimi- 1303–1306.
zation, embedded systems, information security, and smart grid. [26] A. K. Dwivedi and O. P. Vyas, “An exploratory study of experimental tools
for wireless sensor networks,” Wireless Sensor Netw., vol. 3, no. 7, pp. 215–240,
July 2011.
REFERENCES [27] L. D. Mendes and J. J. Rodrigues, “A survey on cross-layer solutions for
[1] S. Kurt and B. Tavli, “Propagation model alternatives for outdoor wireless wireless sensor networks,” J. Netw. Comput. Appl., vol. 34, no. 2, pp. 523–534,
sensor networks,” in Proc. Int. Federation for Information Processing Wireless Mar. 2011.
Days, 2013, pp. 1–3. [28] S. De, C. Qiao, D. Pados, M. Chatterjee, and S. J. Philip, “An integrated
[2] J. T. Correll, “Igloo white,” Air Force Mag., vol. 87, no. 11, pp. 56–61, Nov. 2004. cross-layer study of wireless CDMA sensor networks,” IEEE J. Sel. Areas
[3] I. F. Akyildiz, W. Su, Y. Sankarasubramaniam, and E. Cayirci, “A survey on Commun., vol. 22, no. 7, pp. 1271–1285, Sept. 2004.
sensor networks,” IEEE Commun. Mag., vol. 40, no. 8, pp. 102–114, Aug. 2002. [29] R. Madan, S. Cui, S. Lall, and A. Goldsmith, “Cross-layer design for life-
[4] M. Malajner, K. Benkic, P. Planinsic, and Z. Cucej, “The accuracy of propa- time maximization in interference-limited wireless sensor networks,” IEEE
gation models for distance measurement between WSN nodes,” in Proc. Int. Trans. Wireless Commun., vol. 5, no. 11, pp. 3142–3152, Nov. 2006.

IEEE AntEnnAs & ProPAgAtIon MAgAzInE february 2017 19


This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

[30] S. Cui, R. Madan, A. Goldsmith, and S. Lall, “Cross-layer energy and delay [58] T. Stoyanova, F. Kerasiotis, A. Prayati, and G. Papadopoulos, “Evaluation
optimization in small-scale sensor networks,” IEEE Trans. Wireless Commun., of impact factors on RSS accuracy for localization and tracking applications,” in
vol. 6, no. 10, pp. 3688–3699, Oct. 2007. Proc. ACM Int. Workshop on Mobility Management and Wireless Access, 2007,
[31] N. Baccour, A. Koubâa, L. Mottola, M. A. Z. Niga, H. Youssef, C. A. Boano, pp. 9–16.
and M. Alves, “Radio link quality estimation in wireless sensor networks: A sur- [59] A. Wang, W. B. Heinzelman, A. Sinha, and A. P. Chandrakasan, “Energy-
vey,” ACM Trans. Sensor Netw., vol. 8, no. 4, pp. 34:1–34:33, Sept. 2012. scalable protocols for battery-operated microsensor networks,” J. VLSI Signal
[32] I. F. Akyildiz, Z. Suna, and M. C. Vuran, “Signal propagation techniques Process. Syst. Signal Image Video Technol., vol. 29, no. 3, pp. 223–237, Nov.
for wireless underground communication networks,” Phys. Commun., vol. 2, 2001.
no. 3, pp. 167–183, Sept. 2009. [60] J. N. Al-Karaki, R. Ul-Mustafa, and A. E. Kamal, “Data aggregation and
[33] P. Barsocchi, “Channel models for terrestrial wireless communications: routing in wireless sensor networks: Optimal and heuristic algorithms,” Comput.
A survey,” National Research Council, Information Science and Technologies Netw., vol. 53, no. 7, pp. 945–960, May 2009.
Institute, Pisa, Italy, Tech. Rep. 2006-TR-16, 2006. [61] Z. Cheng, M. Perillo, and W. B. Heinzelman, “General network lifetime
[34] F. Dagefu, D. Liao, and K. Sarabandi, “An efficient model for near-ground and cost models for evaluating sensor network deployment strategies,” IEEE
wave propagation in the presence of building walls/indoor obstacles,” in Proc. Trans. Mobile Comput., vol. 7, no. 4, pp. 484–497, Apr. 2008.
IEEE Antennas and Propagation Society Int. Symp., 2009, pp. 1–4. [62] Y. Yun and Y. Xia, “Maximizing the lifetime of wireless sensor networks
[35] F. Dagefu and K. Sarabandi, “Simulation and measurement of near-ground with mobile sink in delay-tolerant applications,” IEEE Trans. Mobile Comput.,
wave propagation for indoor scenarios,” in Proc. IEEE Antennas and Propaga- vol. 9, no. 9, pp. 1308–1318, Sept. 2010.
tion Society Int. Symp., 2010, pp. 1–4. [63] W. B. Heinzelman, A. Chandrakasan, and H. Balakrishnan, “An application
[36] F. Dagefu and K. Sarabandi, “Analysis and modeling of near-ground wave specific protocol architecture for wireless microsensor networks,” IEEE Trans.
propagation in the presence of building walls,” IEEE Trans. Antennas Propag., Wireless Commun., vol. 1, no. 4, pp. 660–670, Oct. 2002.
vol. 59, no. 6, pp. 2368–2378, June 2011. [64] E. F. Lee, C. Wang, and L. Xiao, “A study of radio signal behaviors in
[37] M. Sheikhsofla and K. Sarabandi, “Indoor wave propagation modeling at complex environments,” Department of Computer Science, Michigan State
low-VHF band,” in Proc. IEEE–APS Topical Conf. Antennas and Propagation in University, East Lansing, Tech. Rep. MSU-CSE-06-6, 2006.
Wireless Communications, 2014, pp. 778–781. [65] M. Zuniga and B. Krishnamachari, “Analyzing the transitional region in low
[38] L. Boithais, Radio Wave Propagation. New York: McGraw-Hill, 1987. power wireless links,” in Proc. IEEE Int. Conf. Sensing, Communication, and
[39] C. Haslett, Essentials of Radio Wave Propagation. Cambridge, U.K.: Networking, 2004, pp. 517–526.
Cambridge Univ. Press, 2008. [66] A. Martinez-Sala, J. M. Molina-Garcia-Pardo, E. Egea-Lopez, J. Vales-
[40] A. Goldsmith, Wireless Communications. Cambridge, U.K.: Cambridge Alonso, L. Juan-Llacer, and J. Garcia-Haro, “An accurate radio channel
Univ. Press, 2005. model for wireless sensor networks simulation,” J. Commun. Netw., vol. 7,
[41] C. A. Balanis, Antenna Theory: Analysis and Design. Hoboken, NJ: Wiley, 2005. no. 4, pp. 401–407, Dec. 2005.
[42] H. T. Friis, “A note on a simple transmission formula,” Proc. IRE, vol. 34, [67] D. Wang, L. Song, X. Kong, and Z. Zhang, “Near-ground path loss
no. 5, pp. 254–256, May 1946. measurements and modeling for wireless sensor networks at 2.4 GHz,”
[43] M. Born and E. Wolf, Principles of Optics. Cambridge, U.K.: Cambridge Int. J. Distributed Sensor Netw., vol. 8, no. 8, p. 969712, 2012.
Univ. Press, 1999. [68] C. Suh, J. E. Joung, and Y. B. Ko, “New RF models of the TinyOS simula-
[44] C. V. Raman and K. S. Krishnan, “On the diffraction of light by spherical tor for IEEE 802.15.4 standard,” in Proc. IEEE Wireless Communications and
obstacles,” Proc. Phys. Soc. Lond., vol. 38, no. 1, p. 350, Aug. 1925. Networking Conf., 2007, pp. 2236–2240.
[45] J. B. Keller, “Geometrical theory of diffraction,” J. Opt. Soc. Amer., [69] R. Belmonte, S. Fast, and J. Schuster, “comparison of near earth propaga-
vol. 52, no. 2, pp. 116–130, Feb. 1962. tion over layered media,” in Proc. IEEE Military Communications Conf., 2008,
[46] D. Liao and K. Sarabandi, “An approximate numerical model for simulation pp. 1–6.
of long-distance near-ground radiowave propagation over random terrain pro- [70] J. C. Giacomin, L. H. A. Correia, T. Heimfarth, G. M. Pereira, V. F. Silva,
files,” in Proc. IEEE Military Communications Conf., 2007, pp. 1–7. and J. L. P. D. Santana, “Radio channel model of wireless sensor networks oper-
[47] R. Foran, T. Welch, and M. Walker, “Very near ground radio frequency ating in 2.4 GHz ISM band,” INFOCOMP, vol. 9, no. 1, pp. 98–106, Mar. 2010.
propagation measurements and analysis for military applications,” in Proc. IEEE [71] N. Wang, N. Zhang, and M. Wang, “Wireless sensors in agriculture and
Military Communications Conf., 1999, pp. 336–340. food industry–recent development and future perspective,” Comput. Electron.
[48] R. C. Bernhardt, “The effect of path loss models on the simulated perfor- Agri., vol. 50, no. 1, pp. 1–14, Jan. 2006.
mance of portable radio systems,” in Proc. IEEE Global Telecommunications [72] T. Stoyanova, F. Kerasiotis, A. Prayati, and G. Papadopoulos, “A practical
Conf. and Exhibition, 1989, pp. 1356–1360. RF propagation model for wireless network sensors,” in Proc. Int. Conf. Sensor
[49] M. Rodriguez, R. Feick, H. Carrasco, R. Valenzuela, M. Derpich, and L. Technologies and Applications, 2009, pp. 194–199.
Ahumada, “Wireless access channels with near-ground level antennas,” IEEE [73] A. Cerpa, J. Wong, L. Kuang, M. Potkonjak, and D. Estrin, “Statistical
Trans. Wireless Commun., vol. 11, no. 6, pp. 2204–2211, June 2012. model of lossy links in wireless sensor networks,” in Proc. Association for Com-
[50] T. Yadav and P. P. Bhattacharya, “Signal strength and system operating mar- puting Machinery/IEEE Int. Symp. Information Processing in Sensor Networks,
gin estimation for vehicular ad-hoc networks in Rayleigh fading environment,” 2005, pp. 81–88.
Int. J. Comput. Sci. Mobile Comput., vol. 2, no. 3, pp. 41–45, Mar. 2013. [74] J. Polastre, R. Szewczyk, and D. Culler, “Telos: Enabling ultra-low power
[51] A. Palaios, Y. Labou, and P. Mahonen, “A study on the forest radio propa- wireless research,” in Proc. Association for Computing Machinery/IEEE Int.
gation characteristics in European mixed forest environment,” in Proc. IEEE Symp. Information Processing in Sensor Networks, 2005, pp. 364–369.
Military Communications Conf., 2014, pp. 376–381. [75] K. Bilinska, M. Filo, and R. Krystowski. (2007). Mica, Mica2, MicaZ.
[52] J. Salo, L. Vuokko, and P. Vainikainen, “Why is shadow fading lognormal?” [Online]. Available: http://wwwpub.zih.tu-dresden.de/dargie/wsn/slides/
in Proc. Int. Symp. Wireless Personal Multimedia Communications, 2005, students/MICA.ppt
pp. 522–526. [76] M. Maurya and S. R. N. Shukla, “Current wireless sensor nodes (motes):
[53] E. M. Failli, “Digital land mobile radio communications—Final report,” Performance metrics and constraints,” Int. J. Adv. Res. Electron. Commun.
Office for Official Publications of the European Communities, Luxembourg, Eng., vol. 2, no. 1, pp. 45–48, Jan. 2012.
Tech. Rep. COST 207, 1989. [77] A. Seema and M. Reisslein, “Towards efficient wireless video sensor net-
[54] J. F. Janek and J. J. Evans, “Predicting ground effects of omnidirec- works: A survey of existing node architectures and proposal for a Flexi-WVSNP
tional antennas in wireless sensor networks,” Wireless Sensor Netw., vol. 2, design,” Commun. Surveys Tuts., vol. 13, no. 3, pp. 462–486, 2011.
no. 12, pp. 879–890, Dec. 2010. [78] D. Cavalcanti, S. Das, J. Wang, and K. Challapali, “Cognitive radio based
[55] B. Tavli, K. Bicakci, R. Zilan, and J. M. Barceló-Ordinas, “A survey of visual wireless sensor networks,” in Proc. Int. Conf. Computer Communications and
sensor network platforms,” Multimedia Tools Appl., vol. 60, no. 3, pp. 689–726, Networks, 2008, pp. 1–6.
Oct. 2012. [79] H. U. Yildiz, M. Temiz, and B. Tavli, “Impact of limiting hop count on
[56] G. Anastasi, A. Falchi, A. Passarella, M. Conti, and E. Gregori, “Perfor- the lifetime of wireless sensor networks,” IEEE Commun. Lett., vol. 19, no. 4,
mance measurements of motes sensor networks,” in Proc. ACM Int. Symp. pp. 569–572, Apr. 2015.
Modeling, Analysis, and Simulation of Wireless and Mobile Systems, 2004, [80] Texas Instruments, “CC1000 single chip very low power RF transceiver,”
pp. 174–181. Texas Instruments, Dallas, TX, Tech. Rep. SWRS048A, 2009.
[57] J. Dunlop, D. Girma, and J. Irvine, Digital Mobile Communications and the
Tetra System. Hoboken, NJ: Wiley, 2013.

20 february 2017 IEEE AntEnnAs & ProPAgAtIon MAgAzInE

Potrebbero piacerti anche