Sei sulla pagina 1di 15

Computer Networks 55 (2011) 37593773

Contents lists available at ScienceDirect

Computer Networks
journal homepage: www.elsevier.com/locate/comnet

Improvement of handover prediction in mobile WiMAX by using two thresholds


Zdenek Becvar , Pavel Mach, Boris Simak
Czech Technical University in Prague, Faculty of Electrical Engineering, Department of Telecommunication Engineering, Technicka 2, 16627 Prague, Czech Republic

a r t i c l e

i n f o

a b s t r a c t
One of the most important challenges in mobile wireless networks is to provide full mobility together with minimum degradation of quality of service. This can be ensured by handover prediction. Handover prediction means a determination of the next station that will serve a mobile station. This paper proposes a prediction technique based on monitoring the signal quality between the mobile station and all base stations in its neighborhood. The proposed technique utilizes two different thresholds for selection of the most likely target base station. Further, the potential improvement of the prediction efciency via techniques originally proposed for minimizing the number of redundant handovers is analyzed. The efciency of the proposed prediction technique is evaluated and compared with other prediction techniques based on channel characteristics in scenarios according to IEEE 802.16m. The proposed technique achieves a prediction hit rate of up to 93%. 2011 Elsevier B.V. All rights reserved.

Article history: Available online 8 May 2011 Keywords: Channel characteristics Handover Mobility Two thresholds prediction WiMAX

1. Introduction Mobile wireless networks should enable full mobility for all users simultaneously while guaranteeing the requested Quality of Service (QoS). The QoS is signicantly inuenced by the mobility of users. As users move, the Base Station (BS) to which they are connected (denoted as serving BS) has to be updated accordingly. This process is known as a handover. To perform a handover properly, a Mobile Station (MS) continuously scans its neighborhood and monitors the channel parameters, for example the signal strength or packet delay, of all available BSs. If some of the monitored signal parameters of the serving BS drop below a predened level or below the level of a neighboring BS, the MS should perform a handover to provide the required QoS. The mobile WiMAX denes three types of handover: hard handover, Macro Diversity Handover (MDHO), and Fast Base Station Switching (FBSS). Hard handover is man Corresponding author. Tel.: +420 2 2435 5994; fax: +420 2 2333 9810.
E-mail addresses: zdenek.becvar@fel.cvut.cz (Z. Becvar), machp2@ fel.cvut.cz (P. Mach), simak@fel.cvut.cz (B. Simak). 1389-1286/$ - see front matter 2011 Elsevier B.V. All rights reserved. doi:10.1016/j.comnet.2011.03.020

datory in WiMAX systems and the other two types of handover are optional. During the hard handover process, the MS closes all connections with the serving BS. Subsequently, it initiates establishment of new connections to a new BS, which is denoted as the target BS. After all connections with the serving BS are closed the MS is disconnected from the network until new connections to the target BS are set up. This short time break, known as handover interruption, handover delay, or handover latency [1], should be minimized since it decreases QoS [2]. Handover interruption occurs if a hard handover is utilized, that is, when the MS always communicates with just one BS. However, the MS can be simultaneously connected to more than one BS in the case of MDHO or FBSS. The list of BSs involved in such a communication is usually called the diversity set in WiMAX [3]. To ensure optimum network performance, the size of the diversity set (the number of BSs in the diversity set) needs to be optimized according to the network conditions and signal quality [4]. The handover interruption as well as the optimization of soft handover can be solved by prediction of the target BS. This paper proposes a new prediction technique that allows high prediction efciency while neither modication

3760

Z. Becvar et al. / Computer Networks 55 (2011) 37593773

of management signaling nor additional demands on users and network are required. The assessment of the prediction efciency is in line with the recommendation on evaluation of networks according to IEEE 802.16m standard [1]. The rest of the paper is structured as follows. The second section provides an overview of the state of the art related to handover prediction. Section 3 introduces the principle of handover in WiMAX and techniques for elimination of redundant handovers. The subsequent section describes the proposed prediction technique. The fth section denes simulation scenarios and parameters considered for assessment of prediction efciency. The results of simulations are presented in Section 6. This section also analyzes possible efciency improvement of the proposed technique and the impact of the proposal on the elimination of redundant handovers. The last section gives our conclusions and future work plans.

2. Related works The advantage of handover prediction is twofold: minimization of interruption during hard handover and achievement of the optimal diversity set size in the case of soft handover. If a proper and efcient handover prediction is performed, the number of redundant handovers (sometimes termed unnecessary handovers in the literature) can be reduced. Redundant handovers are caused by the so-called ping-pong effect, when the MS is continuously being switched between two neighboring BSs since it is moving along the edge of cells boundaries [5]. Another purpose of the utilization of handover prediction is to optimize an admission control as presented in [6] and [7]. The utilization of handover prediction for reservation of resources for admission control is also presented in [8]. The paper proposes two admission control schemes to optimize the utilization of dedicated bandwidth. Prediction for resource allocation is investigated in [9 and [10]. In [9], the authors investigate two approaches to handover prediction: cell and user. The cell approach predicts the number of users in the cell whereas the user approach utilizes mobility prediction to determine information on the next handover. The paper summarizes the advantages of both approaches and their suitability for utilization in different scenarios. An extension of the previous paper is presented in [10]. The authors propose a new resource allocation mechanism showing that the user approach performs better at providing a reduction in handover failures. On the other hand, the cell approach together with the proposed resource allocation mechanism improves cell blocking probability. A modication of the handover procedure for mobile WiMAX exploiting the prediction of target BS and thus signicantly reducing handover interruption is proposed in [11]. The paper shows that the prediction can reduce the downlink handover interruption by up to 90% compared to a conventional IEEE 802.16e handover. The prediction of target BS can be based on several approaches utilizing: handover history, users movement trajectory, and radio channel characteristics.

The rst approach stores information related to the previous handovers of all MSs in the network. This method requires all updates of the serving BS of all MSs in the network to be monitored. This means that identication of the serving and target BSs must be stored in the memory if a MS performs the handover. The prediction is based on the ratio of handovers among individual pairs of BSs performed in the past. An analysis of the efciency of handover prediction utilizing handover history is presented in [12]. This paper further investigates the impact of the number of neighboring BSs on the prediction efciency. According to the results, the maximum efciency is only approximately 45% for three neighboring BSs in Manhattan-like street deployment. The second approach to handover prediction is to determine the next positions of the MS based on its movement in the past as addressed, for example, in [13]. Three general assumptions must be fullled for highly efcient prediction of the MSs movement [14]: (i) exact knowledge of current and previous positions of the MS, (ii) knowledge of the users prole, and (iii) knowledge of the prole of the area in which the prediction is performed. Knowledge of the exact position implies a utilization of localization equipment such as GPS (Global Positioning System) [15]. The second assumption requires the acquisition of user information such as areas of interest, favorite places, time schedule, etc. Consequently, each user has to ll in this information and keep it updated. To meet the last assumption, it is necessary to have a proper geographic map of areas in which the prediction is being executed. This information can be acquired by a network provider and also must be kept up to date. Fulllment of all the above mentioned assumptions imposes very high demands on the user, mobile equipment, and network. Therefore, the prediction is convenient for neither users nor operators. In [16], the authors analyze the effectiveness of the prediction to reduce power consumption in ad-hoc networks. This goal is achieved by delaying the communication until a MS becomes closer to the target BS. The prediction is also based on the MSs movement history with all its drawbacks. Prediction of the users position is further exploited for example in [13] and [17]. Both papers present advanced algorithms for prediction of the users location. However, the papers do not eliminate general weaknesses of this kind of prediction and assume there is knowledge of the exact position of the user. The last prediction approach, based on the channel (or network) characteristics, exploits information which is usually exchanged among MSs and core network (represented by BSs) during normal operation (e.g., for handover purposes). Hence, no additional requirements are implied for either the MSs or the BSs. The efciency of techniques using the channel characteristics is signicantly higher than the efciency of techniques based purely on the handover history (see [18] and [12]). On the other hand, these techniques are usually outperformed by position based prediction. Prediction utilizing channel characteristics is investigated in [19]. The authors generally describe the new network evaluation method as a criterion for handover decision with user preference and the usual estimation standards. Handover prediction is performed

Z. Becvar et al. / Computer Networks 55 (2011) 37593773

3761

according to a weighted combination of several network parameters such as bit rate, latency, or power consumption. In [18], the authors evaluate several ltering methods for handover prediction. The authors compare the efciency of handover prediction for Grey [20], Kalman [21], Fourier [22], and Particle [23] ltering of RSSI (Received Signal Strength Indication) values. The prediction is based on the mutual relation of RSSIs of the target and serving BSs. The prediction is performed if the difference between the two RSSI levels falls into a predened interval. The results show the best performance (roughly 80% successful handover prediction) for Grey ltering. Grey ltering is also analyzed in [24]. The paper evaluates and proves the positive impact of Grey prediction on the reduction of redundant handovers. Chan presents QoS adaptive prediction combined with the second type of prediction in [25]. The maximum prediction efciency of the proposed technique is roughly 75%. The results are achieved for regular movement trajectories followed by employees in their workplaces. In [26], the authors compare several handover prediction techniques like the handover history, mobility pattern, movement extrapolation, or distance. The paper shows that the best performance (highest ratio of correct predictions) can be achieved by prediction based on a mobility pattern or movement extrapolation for a road mobility model or a random waypoint mobility model [26] respectively. The prediction efciency is approximately 60% in both cases. The goal of this paper is to design a prediction technique which can achieve efciency as high as that obtained by techniques based on knowledge of the users position while eliminating their drawbacks additional demands on users, mobile equipments, and network. Hence, prediction using channel characteristics is considered as the basis of the proposed technique. The improvement of the prediction efciency is achieved by utilization of two mutually independent thresholds for making a decision on the selection of the predicted target BS. Furthermore, the rise in prediction efciency obtained by techniques originally used for reduction of the number of redundant handovers (avoiding the ping-pong effect) is also investigated in this paper. Three techniques are considered: Hysteresis Margin (HM) [3], [27], windowing (also known as signal averaging) [27], and Handover Delay Timer (HDT) [28]. All assumptions for the proposed prediction are based only on parameters and metrics that can be obtained during conventional activities of networks according to the IEEE 802.16e and IEEE 802.16j standards [29].

collects information on neighboring BSs with the aim of nding a suitable target BS. Based on the results of the scanning process the possible target BS is selected in the frame of the cell reselection phase. If all the conditions for handover are met, the handover decision and initiation phase is performed. Then, the MS starts synchronization with the downlink of the target BS. As soon as the synchronization is nished, the MS initiates the network re-entry consisting of three substages: ranging, re-authorization, and re-registration. After successful accomplishment of all three substages, the MS can start with normal operation, that is, it can exchange data with the new serving BS. The above described procedure explains the principle of hard handover. In the case of MDHO or FBSS, the principle is similar; however the MS is connected to more BSs simultaneously. When the MDHO or FBSS is supported, a list of BSs which are involved in the handover procedure (i.e., the diversity set) is maintained by the MS and BSs. This set is updated via MAC (Medium Access Control) management messages (see [3] and/or [29] for more information). The diversity set is dened for each MS in the network. In the case of MDHO, the MS continuously monitors all BSs in the diversity set and selects an anchor BS. The MS is synchronized, authorized, and registered to the anchor BS. Furthermore, the MS performs ranging and monitors a downlink channel of the anchor BS for control information. The MS communicates simultaneously (including user trafc) with the anchor BS and with all active BSs in the diversity set. Unlike the MDHO, the MS communicates only with the anchor BS for all types of uplink and downlink trafc including management messages while FBSS is utilized. The anchor BS can be changed on a frame to frame basis depending on a BS selection scheme. This means that the MS can receive individual frames from different BSs out of all BSs in the diversity set. 3.2. Elimination of redundant handovers In general, the hard handover is of low complexity and easy to implement in mobile networks. On the other hand, it results in more signicant degradation of QoS (see e.g., [2]). Moreover, any type of handover is interconnected with the generation of additional management overhead. To avoid both negative phenomena, the elimination of so-called redundant handovers has to be ensured. The redundant handover represents the case when handover is executed but not nished before the time when the next handover decision takes place. Also, handovers repeated frequently between two adjacent cells in a short time interval (i.e., ping-pong effect) should be considered as redundant handovers since the MS cannot take advantage of the connection to the new BS. Redundant handovers are usually caused either by fading effects or by movements of users along the edges of cells. Several techniques can be utilized for minimization of the number of redundant handovers. Standard IEEE 802.16e denes HM and Time-To-Trigger (TTT) [3]. Other commonly used techniques are, for example, windowing or HDT extending conventional TTT. All methods are based on delaying the handover for a predened time interval. The utilization of these methods for prediction purposes

3. Handover in mobile WiMAX 3.1. Principle of handover The handover procedure can be split into several stages according to [3]: network topology advertisement, scanning of the MSs neighborhood, cell reselection, handover decision and initiation, and network re-entry. The rst two stages are performed before the handover process begins. During both, the MS searches its neighborhood and

3762

Z. Becvar et al. / Computer Networks 55 (2011) 37593773

does not increase the management overhead since the parameters are already incorporated in MAC management messages in WiMAX. Thus, these parameters are distributed within the network regardless of whether the prediction is used or not. In the case of HM, the handover decision and initiation are based on a comparison of one or several signal parameters of the serving and target BSs. The handover is initiated if the signal parameter of the target BS exceeds the signal parameter of the serving BS plus HM, as dened by the next equation:

4. Proposal on Two Thresholds Prediction The principle of the proposed technique is explained in Fig. 1. It exploits reports on signal (channel) quality continuously obtained by the network in the frame of the MSs scanning procedure. Our improvement is based on the definition of two independent thresholds as depicted in Fig. 1a. One threshold is related to the signal level received by the MS from the serving BS while the second threshold is related to the signal level measured by the MS from the potential target BS. Hence the prediction is entitled Two Thresholds Prediction (TTP). Fig. 1a depicts RSSI evolution between the MS and several BSs during the MSs movement along a straight line (red dashed trajectory in Fig. 1b). The speed of the MS is 15 m/s and the observation time is 100 s; that is, the distance covered by the MS within one observation cycle is 1500 m. Each curve in Fig. 1a represents the set of RSSIs received by the MS from all BSs obtained within several movements of the MS along the red line. Minor uctuation of RSSI is caused by variations in channel parameters among all runs of the MS (for more details on simulation parameters see Section 5). The following two thresholds are dened: HO_ThrSerX,Y and HO_ThrTarX,Y. The rst represents a typical RSSI level of the serving BSx at the moment of the initiation of the MSs handover to the target BSy (see Fig. 1a, where the serving BSx is BS4 and the target BSy is BS2). The second threshold corresponds to a typical RSSI level of the predicted target BSy at the moment of initiation of the MSs handover from the serving BSx. In practice, both threshold levels are usually very close in most cases; nevertheless they are not equal. In principle, the level of HO_ThrTarX,Y is slightly higher than HO_ThrSerX,Y since the handover decision is generally made if the target BS can provide higher connection quality than the serving BS in WiMAX. Moreover, both thresholds are also unequal due to non-stationary signal levels. The monitoring and evaluation of the signal evolution from all neighboring BSs is performed in the following manner. If the RSSI level from the serving BS decreases and draws near to HO_ThrSerX,Y, the probability of a handover from BSx to BSy increases (in Fig. 1a

STar > SSer HM; i i

where SSer and STar represent the signal quality parameters t t of the serving and target BSs respectively. If windowing is applied, the handover decision is made if the average value of the observed signal parameter from the target BS drops below the average level of the same parameter at the serving BS:

PWS

Tar i1 Si

PWS >

Ser i1 Si

WS

WS

where WS corresponds to the window size, that is, the number of samples over which the average value is calculated. Implementation of the HDT is based on the insertion of a short delay between the time when the handover conditions are rst met and the time when handover initiation is executed. This method is based on TTT. In the case of TTT the signal is continuously monitored for each frame during a short interval (up to 255 ms in WiMAX), whereas the HDT evaluates only several signal samples measured during longer periodic intervals. The handover is performed if:

SSer < STer jt 2 t HO ; t HO HDT; t t

where HDT represents the duration of the handover delay timer and tHO is the time instant when the handover conditions are fullled.

Fig. 1. Denition of handover threshold (a) based on movement of MS along the same direction (b).

Z. Becvar et al. / Computer Networks 55 (2011) 37593773

3763

HO_ThrSerX,Y is HO_ThrSer4,2). If the RSSI from one of the neighboring BSs increases and draws near to HO_ThrTarX,Y, the probability of a handover from BSx to BSy also rises (in Fig. 1a HO_ThrTarX,Y is HO_ThrTar4,2). If the RSSI of the serving BS drops below HO_ThrSerX,Y and simultaneously the RSSI of a neighboring BS exceeds HO_ThrThrX,Y, the prediction result is the expectation of a handover of the MS from BSx to BSy. This means that BSy is labeled as the predicted target BS. Both thresholds for each pair of BSx and BSy (thresholds HO_ThrSerX,Y and thresholds HO_ThrThrX,Y) are averaged out in order to derive one specic value related to the serving BS threshold as well as one value related to the target BS threshold. The mean values of the typical thresholds for the handover are calculated as an average of several previous signal levels leading to the handover initiation. Determination of the sufcient number of RSSI samples is the object of investigation addressed further in this paper. The mean thresholds can be described by the following equations:

where HOOT represents the interval between the two sigZone nal levels in which the BS is determined as predicted target BS in competitive proposals. If two thresholds are considered, the uctuation of only one signal level (e.g. RSSIMS,BSX) does not necessarily lead to selection of BSy as the predicted target BS since the condition related to the second RSSI (e.g. RSSIMS,BSY) is still not fullled. All neighboring BSs of BSx (including BSy) are included in the so-called Neighbor Set of BSx denoted in this paper as NSx. Considering the NS, the probability of handover from the BSx to the BSy (labeled as Px,y) can be formulated by the following equation:

P x;y

ax;y ; BSy 2 NSx ;


0; BSy R NSx ;

Av gHO Thr SerX;Y

1 HOBSX;BSY 1 HOBSX;BSY

HOBSX;BSY

X
i1

i RSSIMS;BSX ;

HO

HOBSX;BSY

Av gHO Thr TarX;Y

X
i1

i RSSIMS;BSY ;

HO

where ax,y represents the exact probability value of the handover from BSx to BSy. It is in the range 0 6 ax,y 6 1. The value of ax,y depends not only on the number of BSs in the NS but also on the layout of the area where the prediction is analyzed and monitored (e.g., the layout of streets, deployment of buildings and transmitters, etc.). Px,y depends on the probability of fulllment of (6) and (7). As both conditions are independent, Px,y can be rewritten as:

Px;y PAv gHO Thr SerX;Y HOZone > RSSIMS;BSX PAv gHO ThrTarX;Y HOZone < RSSIMS;BSY n PBSSer PBSTar n: 10
The probability of the case when the MS will not perform the handover to the predicted BSy Px;y despite fulllment  of conditions (6) and (7) by BSy is expressed by the n function. According to (6) and (7), Px,y is a function of parameter HOZone and actual values of RSSIMS,BSX and RSSIMS,BSY. If ax,y represents the value of probability Px,y, the probability of the handover from BSx to all BSs excluding BSy P x;y is  1 ax,y. For example, this probability corresponds to the case when the MS randomly turns away from the anticipated direction. Then the probability of handover from BSx to BSy could be generally described by a function formulated in the following way:

where HOBSX,BSY represents the number of handovers that occur between the current serving BS and the potenHOi tial target BS during the observed time interval; RSSIMS;BSX HOi and RSSIMS;BSY are RSSIs received at the MS from BSx and BSy respectively at the time instant of the handover decision; and index i species the individual handover event. It is not appropriate to perform the target BS prediction only when the typical thresholds are reached since the prediction would be made too late for exploitation of the prediction results in advance of the handover execution. Therefore, the target BS should be selected when the RSSIs between the MS and the serving and target BSs are within intervals of HOZone dened by the following formulas:

HO Thr SerX;Y HOZone > RSSIMS;BSX ; HO Thr TarX;Y HOZone < RSSIMS;BSY ;

6 7

Px;y f RSSIMS;BSX ; RSSIMS;BSY ; HOZone ; n:

11

where HOZone represents the interval when the BSy is marked as the predicted target BS (see Fig. 1a) and RSSIMS,BSX and RSSIMS,BSY correspond to the signal level currently received by the MS from the serving and target BSs respectively. The utilization of two thresholds instead of conventional prediction with one threshold enables higher efciency to be achieved since it can reduce the ratio of incorrect target BS prediction. In the case of conventional prediction (dened by (8)), the prediction efciency can be inuenced by signal level uctuation, for example due to shadowing, fast fading, and so on. If only one signal level (RSSIMS,BSX or RSSIMS,BSY) is affected by fast fading or shadowing, (8) can be fullled even if the user does not change location.

The function n is directly proportional to the probability that the MS will not change direction signicantly enough to perform the handover to a different BS in the next time interval s. This probability is labeled as P SD . Thus, n can be MS formulated as:

n % PSD s; MS

12

sis a function of speed and distance as expressed by the next formula:

DistMS;cell

v MS

13

RSSIMS;BSX RSSIMS;BSY < HOOT ; Zone

where vMS is the average velocity of the MS and DistMS,cell is the distance of the MS from the place where the handover should be executed (cell edge). The distance is measured along the MSs trajectory. Then Eq. (11) can be rewritten as follows:

3764

Z. Becvar et al. / Computer Networks 55 (2011) 37593773

Px;y f RSSIMS;BSX ; RSSIMS;BSY ; HOZone ; v MS ; DistMS;cell ; PSD MS   Dist MS;cell PBSSer PBSTar PSD : MS

mMS

14
The previous formula can be expressed in another way as the probability of successful prediction of the target BSy if the MS is moving out of the coverage area of the serving BSx. Considering the above mentioned facts, the probability of successful prediction increases with either a decrease in DistMS,cell or a decrease in the difference between the current values of RSSIMS,BSx and AvgHO_ThrSerX,Y as well as RSSIMS,BSy and AvgHO_ThrTarX,Y. On the other hand, a drop in P SD (caused by lowering mMS) can reduce the probability MS of successful handover prediction. The impact of the last parameter in (14), HOZone, is investigated later in this paper. Since the MS can arrive in an area where more than one potential target BS fullls the conditions for the prediction, some mechanisms for the selection of the single most likely target BS should be dened. This mechanism is based on the calculation of the minimum difference between both thresholds AvgHO_ThrSerX,Y and AvgHO_ThrTarX,Y and the current RSSIMS,BSX and RSSIMS,BSY respectively. This is done for all possible target BSs with RSSI values in the range dened by (6) and (7). These stations are listed in ListOfTargetBS. The selection of the target BS that will be labeled as the predicted target BS is done according to the results of the next equation:

Report metric, and therefore one of those bits can be used to identify the new metric. This method retains the backward compatibility with former WiMAX standards. The above described TTP assumes that the scanning results of the MSs neighborhood are stored, which can be done in either the MSs or the BSs. Therefore, the prediction can be made by either the MSs or the BSs. Nevertheless, several aspects should be considered before the entity responsible for prediction is selected. Firstly, a lot of data have to be kept in the stations memory. Furthermore, if the MS is responsible for the prediction, all information on all handovers in the network has to be delivered to this MS. This signicantly increases management overhead. The MS can only exploit information on handovers performed by itself, which dramatically prolongs the time required to gather enough information to ensure high efciency of the prediction. If the BSs are responsible for the prediction, no additional information has to be exchanged among MSs and BSs since BSs receive all information via scanning reports provided by MSs. Thus, BSs can easily determine appropriate thresholds. Then, the prediction of a target BS is performed by BSs based on evaluated thresholds and actual reports provided by MSs. All of the above mentioned aspects indicate that prediction by the BSs is distinctly more efcient. Hence, prediction performed by BSs is considered in the rest of the paper. 5. Simulation scenario The prediction of the target BS is evaluated for the case when the reporting of RSSI is assumed. The simulation parameters are summarized in Table 1. The simulations are performed in a developed MATLAB simulator. All BSs are deployed in a symmetric manner (see Fig. 2) at the same height and transmit at the same power level according to the recommendations on evaluation of an IEEE 802.16 m network [1]. At the beginning of the simulation, the positions of all 48 MSs are randomly generated. A Probabilistic Random Waypoint Mobility Model (PRWMM)

DiffBSX;BSY jAv gHO Thr SerX;Y RSSIMS;BSX j jAv gHO Thr TarX;Y RSSIMS;BSY j: 15

The differences between RSSI levels and thresholds for each potential target BS are compared afterwards and the BS with the minimum DiffBSX,BSY is selected as the predicted target BS (see next formula).

PredictedTargetBS fYgjListOfTargetBSY minListOfTargetBS: 16

As the prediction of more than one target BS can be protable, all BSs fullling conditions (6) and (7) can be denoted as predicted target BSs as well if required, for example, if the most likely target BS cannot accept the MS due to overloading. Other signal parameters such as CINR (Carrier to Interference and Noise Ration), SNR (Signal to Noise Ratio), delay, or other parameters expressing the quality of channel between a MS and BSs can be used for the prediction instead of RSSI. The only limitation on the parameter utilized for the prediction is that it has to correspond to the metric used in scanning reports sent by the MS to the serving BS (message MOB_SNC-REP; see more details in [3]). If the parameter is not listed among the parameters conventionally used in mobile WiMAX, two messages related to scanning must be modied: the scanning response (MOB_ SNC-RSP) and scanning report (MOB_SNC-REP). The modication of both messages is very simple since only inclusion of a new parameter in the eld Report metric is necessary. There are four reserved bits in the eld

Table 1 Simulation parameters and scenario denition for channel characteristics. Parameter Number of BSs Number of MSs BS transmitting power [dBm] BS height [m] MS height [m] MS speed [m/s] Frequency band [GHz] Frame duration [ms] Scanning reporting period [s] Simulation duration [s] HOZone[dB] Hysteresis margin [dB] HDT [s] Window size [samples] Path loss model Channel variation Shadowing Mobility model Size of simulated area [m] Value 15 48 46 32 2 15 2.5 10 1 10800 016 4, 8, 12, 16, or 20 1/2/3/4/5 5, 8, 10, 15, or 20 Urban Macrocell [1] r = 0, 0.3, 0.6, or 0.8 Standard deviation 8 dB [1] PRWMM 2330 2100 (4.89 km2)

Z. Becvar et al. / Computer Networks 55 (2011) 37593773

3765

[30] is considered for the MSs movement since it provides a higher level of movement randomness than other mobility models. Thus the obtained results correspond to the worst case scenario. Signal strengths among all MSs and BSs are calculated using the urban macrocell path loss model dened in [1]. Channel variation is represented by low signal level uctuation [31]. The path loss with channel variation is equal to

PLCV i PLi CV randi Si ;

17

where i indicates a step of the simulation, PL is the macrocell path loss dened in [1], CVrand is a random level of uctuation, the exact value of uctuation is randomly selected according to lognormal distribution with l = 0 and r = 0, 0.3, 0.6, or 0.8 (depending on the specic scenario) [31], and S is the sign (positive or negative) of CVrand. Si is expressed by the following formula:

Not predicted handover occurs if the handover is carried out despite the fact that no target BS has been predicted since the time when the MS performed the previous handover. This situation takes place especially at the beginning of the simulation as not enough data have been collected to perform successful prediction of the target BS (typical thresholds cannot be set up precisely as either there is still no information on the previous handovers at all or the information gathered is insufcient). The ratio of not predicted handovers (NPR) is calculated in the following way:

NPR

NP HO NHO

0 6 NPR 6 1;

21

PSi 1

a;

Si1 1; 18

1 a; Si1 1;  1 a; Si1 1; PSi 1 a; Si1 1:

where NPHO is the total number of not predicted handovers. An error in prediction (wrong prediction) occurs if the predicted target BS differs from the real target BS of the MS. The ratio of wrong predictions (WPR) can be expressed by the following equation:

WPR

The signal parameters are evaluated in each scanning reporting period (i.e., 1 s). The performance of the proposed prediction scheme is evaluated by means of three parameters: the ratio of successfully predicted handovers HR (Hit Ratio), the ratio of not predicted handovers (NPR), and the ratio of wrong prediction (WPR). The handover prediction is assumed to be successful if the MS executes the handover to the predicted target BS. The number of successfully predicted handovers (SPHO) is used for the calculation of the prediction hit ratio according to the following formula:

WPHO NHO

0 6 WPR 6 1;

22

where WPHO is the total number of incorrectly predicted handovers. 6. Evaluation of proposed technique efciency Several sets of simulations are performed considering different types of scenarios. The rst one analyzes the impact of the HOZone parameter on the prediction efciency if no other technique for efciency improvement is considered (i.e. HM, HDT, or windowing). Next, three sets of simulations investigate the impact of all the individual techniques on prediction efciency. The last one determines the optimum setting of all three techniques in cooperating mode to obtain maximum prediction efciency. The results acquired by simulations are separated into ve subsections according to the investigated techniques. 6.1. Impact of HOZone on prediction efciency The results of the simulation are shown in Figs. 3 and 4. These gures present the dependence of the prediction efciency on the simulation time. Both gures contain the results of HR, NPR, and WPR. It is evident that if the WPR decreases together with the NPR, the overall ratio of successful prediction (HR) increases proportionally since the relation among those parameters is: HR = 1 NPR WPR. Figs. 3 and 4 enable determination of the minimum time interval for collection of RSSI information to ensure sufciently high prediction efciency. This is slightly over 1000 s (at that time, roughly 2000 handovers have already been performed within the simulation). This corresponds to roughly 10 handovers between each pair of neighboring BSs). The individual gures differ in diverse parameter settings of the channel model. While Fig. 3 assumes that the handover prediction is performed only according to RSSI evolution, Fig. 4 also takes into account another factor,

HR

SP HO NHO

0 6 HR 6 1;

19

where NHO represents the total number of all handovers in the network. It can be calculated as:

NHO

NBS N BS X X BSX1 BSY1

HOBSX;BSY

BSX BSY;

20

where NBS is an overall number of BSs in the network.

Fig. 2. Deployment of BSs in the simulation.

3766

Z. Becvar et al. / Computer Networks 55 (2011) 37593773

Fig. 3. Results of handover prediction based on the RSSI evolution, without channel variation.

Fig. 4. Results of handover prediction based on the RSSI evolution, with channel variation (r = 0.8).

i.e., channel variation with r = 0.8 [1], [31] and shadowing with a standard deviation of 8 dB [1]. Windowing, HDT, and HM are disabled in Figs. 3 and 4. The results of the prediction hit ratio acquired during 3 h (10800 s) of data monitoring are summarized in Fig. 5. This gure also considers other levels of channel variation, that is, r = 0.3 and r = 0.6. The maximum HR (approximately 71%) is achieved if the channel variation is not considered. The higher channel variation degrades maximum HR to 45, 41, or 37% for r = 0.3, 0.6, or 0.8 respectively. Additionally, Fig. 5 demonstrates signicant dependence of the prediction and its effectiveness on the HOZone. The prediction efciency is under 10% if HOZone is equal to 0 dB. The optimal value of HOZone observed in Fig. 5 is 4 dB for all levels of channel variation. At this HOZone level, the prediction mechanism shows the highest ratio of successfully predicted target BSs (71%) and low NPR (14%) as well as WPR (15%) if no channel variation is introduced. Fig. 6 indicates that the channel variation increases the NPR to slightly over 30% for all levels of r at HOZone = 4 dB. The impact of different channel variation levels on the NPR is negligible.

The WPR also rises to 25, 28, or 31% for r = 0.3, 0.6, or 0.8 respectively at HOZone = 4 dB (see Fig. 7). In the other simulations focused on a comparison of the impact of windowing, MH, and HDT, the channel variation with r = 0.8 will be taken into account (denoted as CV on). The results of the scenario without consideration of the channel variation are also depicted in the following gures to enable comparison of results (denoted as CV off). The detailed behaviors of the prediction ratios over the simulation time, as depicted in Figs. 3 and 4, are not presented in the following subsections. The overall results are illustrated in the form of separate results for HR, NPR, and WPR so that they are more transparent. 6.2. Impact of windowing on prediction efciency In order to suppress the negative impact of the channel variation on the prediction and to increase its efciency, the windowing technique can be used. Fig. 8 demonstrates a distinguishable increase in the HR by windowing even if channel variation (r = 0.8) is assumed. The Window

Z. Becvar et al. / Computer Networks 55 (2011) 37593773

3767

Fig. 5. Target BS prediction hit ratio over HOZone.

case with no windowing. In the rest of the paper, if the gain is for example 10%, it means an increase of, for example, 50% to 60%. An additional gain of 6% is introduced when the WS is further increased to eight samples. In this case, the impact of the channel variation is eliminated and the HR for the channel variation with r = 0.8 and WS = 8 is equal to the scenario with no channel variation and no windowing. In both cases, maximum HR is 71%. Another gain of 2% and 5% is observed by increasing WS to 10 and 15 samples respectively. However, the results show only a marginal increase in the HR for WS = 20. Therefore, a further increase in WS is useless since it brings no additional improvement of the prediction efciency. Analogical conclusions can be derived from Figs. 9 and 10. Fig. 9 shows merely a slight decrease in the NPR for WS above 15 samples. Similarly, Fig. 10 presents only a marginal decrease in the WPR for values where WS rises above 15 samples. The NPR decreases while the level of HOZone rises since a larger HOZone allows the prediction to be performed earlier. However, earlier prediction increases the WPR as well. This is due to the higher probability of the MS turning away from the anticipated direction when prediction takes place earlier (see Eq. (14)). 6.3. Impact of hysteresis margin on prediction efciency Another way to suppress the negative impact of channel variation is to use HM. The results of this investigation are presented in Figs. 1113 with HM = 4, 8, 12, 16, and 20 dB. The scenario with HM = 4 dB improves the HR by 41% (at HOZone = 4 dB) in comparison to the scenario where HM is not considered. Note that the HR for HM = 4 dB at HOZone = 4 dB and the channel variation with r = 0.8 is higher in comparison to the scenario with no channel variation (by roughly 7%). Additional gain is reached by increasing HM up to 12 dB at HOZone = 4 dB. The prediction efciency reaches approximately 90% at this HM level. If HM is set to either 16 dB or 20 dB no signicant gain in the HR is acquired at HOZone = 4 dB in comparison to HM = 12 dB. The HR at HM above 12 dB reaches its maximum at a higher

Fig. 6. Ratio of not predicted handovers over HOZone.

Fig. 7. Ratio of wrongly predicted target BSs over HOZone.

Size (WS) is set to 5, 8, 10, 15, and 20 samples. The utilization of the windowing technique with WS = 5 increases HR from 37% to 65% (28% increase in HR) in comparison to the

Fig. 8. Target BS prediction hit ratio over HOZone for a set of WSs.

3768

Z. Becvar et al. / Computer Networks 55 (2011) 37593773

Fig. 9. Ratio of not predicted handovers over HOZone for a set of WSs.

Fig. 12. Ratio of not predicted handovers over HOZone for a set of HMs.

Fig. 10. Ratio of wrongly predicted target BSs over HOZone for a set of WSs.

Fig. 13. Ratio of wrongly predicted target BSs over HOZone for a set of HMs.

In addition, Figs. 12 and 13 show that an increase in the HM above 12 dB leads neither to signicant improvement in the reduction of the NPR handovers nor to considerable minimization of the WPR. The NPR is even slightly higher for HM = 16 dB and HM = 20 dB than for HM = 12 dB. Another reason for the utilization of a lower level of HM is its impact on the network throughput. A higher value of HM can decrease the network throughput since the MS communicates with the BS, which does not provide the best signal quality. The lower signal quality can result in the utilization of a more robust modulation and coding scheme (MCS) for communication between the MS and the BS (see [3]). 6.4. Impact of handover delay timer on prediction efciency The third means of mitigating the negative impact of the channel variation on prediction efciency is the application of HDT. The results of simulation are depicted in Figs. 1416 with a channel model including channel variation with r = 0.8 and utilization of HDT = 1, 2, 3, 4, and 5 s. As Fig. 14 indicates, HDT cannot fully eliminate the negative impact of the channel variation. If the HDT is set to

Fig. 11. Target BS prediction hit ratio over HOZone for a set of HMs.

HOZone. The maximum for HM = 16 dB and HM = 20 dB is achieved at HOZone = 6 dB (91%) and HOZone = 8 dB (92%) respectively.

Z. Becvar et al. / Computer Networks 55 (2011) 37593773

3769

1 s, the HR is improved by only 2% at HOZone = 4 dB in comparison to the scenario without HDT. A further increase in HDT duration (up to HDT = 3 s) brings rises in the HR of 2% per 1 s of HDT duration. With additional prolongation of HDT (above 3 s) the improvement in the HR is negligible (less than 1% per 1 s of HDT duration). The HDT increases the NPR up to roughly 15% (depending on HOZone) in comparison to the scenario with no HDT, HM, or windowing. The duration of HDT has only a minor impact on the NPR (see Fig. 15). On the other hand, the WPR is improved by approximately 15% (see Fig. 16) by implementation of HDT. However, the WPR is inuenced only marginally by the duration of HDT. 6.5. Maximization of handover prediction efciency This section presents the results when all three methods are used together in order to improve overall prediction efciency. A very high number of combinations of all techniques can be specied. Therefore, only the combinations which potentially offer the best results (according

Fig. 16. Ratio of wrongly predicted target BSs over HOZone for a set of HDTs.

Fig. 14. Target BS prediction hit ratio over HOZone for a set of HDTs.

to results taken from previous subsections) are assumed. If the system performs similarly for two different values, the lower one is selected as optimal since a lower negative impact on the throughput can be assumed [32]. To nd an optimal setting of all parameters, the values of individual methods showing the best performance of the prediction are summarized in Table 2. The complete list of all investigated scenarios is introduced in Table 3. Scenarios are dened with respect to the impact of each particular technique on their mutual cooperation. Values of HM higher than 12 dB are not considered in the following scenarios as these values noticeably decrease the MSs throughput (see [32]). Note that the minimal value for WS is 1 sample, that for HM is 0 dB, and that for HM is 0 s. In total, 12 scenarios are dened for the investigation of maximum prediction efciency. All results are distributed in two gures due to the higher clarity of plotted curves (the rst one contains the results of scenarios A-F; the second one presents scenarios G-L). As can be observed from Fig. 17, the highest HR can be achieved by scenario C (HR is 93% at HOZone = 6 dB). This scenario corresponds to the case when the optimum parameters of HM and WS are set up (see Table 2) while HDT is disabled. Furthermore, an increase in optimal HOZone with HDT is noticeable from Fig. 17 (compare, e.g., scenarios A, B, and C). However, the maximum HR decreases with increases in the duration of HDT (compare scenarios A, B, and C or D, E, and F). Another conclusion can be obtained by comparing scenario C with F, B with E, A with D, or G with H. These scenarios show higher improvement in the HR with

Table 2 Best performing parameters of particular techniques. WS [samples] 15 at HOZone = 4 dB 10 at HOZone = 4 dB 20 at HOZone = 4 dB HM [dB] 12 at HOZone = 4 dB 16 at HOZone = 6 dB 20 at HOZone = 8 dB HDT [seconds] 1 at HOZone = 10 dB 2 at HOZone = 4 dB 3 at HOZone = 4 dB

Fig. 15. Ratio of not predicted handovers over HOZonefor a set of HDTs.

3770 Table 3 List of all simulation scenarios. Scenario A B C D E F G H I J K L WS [samples] 15 15 15 1 1 1 10 2 2 15 8 3 HM [dB] 12 12 12 12 12 12 12 12 1 1 4 2

Z. Becvar et al. / Computer Networks 55 (2011) 37593773

HDT [seconds] 3 1 0 3 1 0 1 1 1 1 0 0

WS only as long as the HDT is turned off. Otherwise, it is better to utilize a lower WS. Scenarios I and J in Fig. 17 demonstrate that utilization of low HM together with enabled HDT leads to a HR under 50%. Moreover, HR is nearly constant for HOZone higher than 4 dB. Additionally, scenarios K and L achieve very high HR (89% and 82%) in spite of fact that both utilize very low values of HM and WS simultaneously with HDT equal to zero. Both scenarios are able to outperform the conventional ltering techniques (such as Grey, Fourier, Kalman, or Particle) presented in [18]. All combinations of techniques improve prediction efciency for all HOZone (compare all scenarios with the green dashed-dotted line) and only very low values of HM and WS allow the impact of channel variation to be eliminated fully. Scenarios K and F are also evaluated for disabled channel variation. The results show insignicant impact of the channel variation on the HR. The improvement in the HR is only approximately 3% for scenario F and 0.5 % for scenario K at HOZone = 4 dB. Consequently, the level of channel variation does not inuence the prediction efciency as it is eliminated by HM and windowing. Hence, a combination of those techniques can be considered more effective in comparison to the ltering techniques investigated in [18] since it provides higher prediction efciency. More-

over, these techniques are already implemented and largely used in mobile networks. Thereupon, neither hardware modication of equipment nor implementation of new ltering procedures in the software of MSs is required for the implementation of the proposed TTP prediction. Fig. 18 illustrates the NPR over HOZone for all scenarios. As can be observed, NPR always decreases with rises in HOZone. All scenarios with disabled HDT show only a negligible NPR value for HOZone higher than 4 dB (it is nearly 0%). The scenarios with enabled HDT indicate gradual decreases in NPR over HOZone. NPR becomes higher while the duration of HDT increases. The impact of the channel variation on the NPR is insignicant in scenarios F and K. Exactly opposite behavior in comparison to Fig. 18 can be observed in Fig. 19. This gure represents the dependence of WPR on HOZone. All scenarios perform signicantly better than the scenario with all techniques turned off. The WPR is between 4% and 17% at HOZone = 4 dB for all scenarios. All scenarios with a higher value of HM (HM = 12 dB) show WPR under 9%. Therefore, lower values of HM lead to higher WPR. Fig. 19 also shows decreases in WPR with increases in HM. Turning off the channel variation slightly decreases the WPR in scenario F (up to 3% at HOZone = 4 dB). However, the channel variation does not inuence the WPR in scenario K. A comparison of the results of the proposed technique with other competitive proposals regarding target BS prediction is presented in Table 4. The proposed technique TTP, which uses two thresholds, shows approximately 10% higher efciency compared to the ltering techniques dened in [18] which utilize only one threshold. The results clearly show that only the trajectory prediction technique proposed by Samaan in [14] reaches comparable efciency. However the trajectory prediction according to [14] requires a huge amount of additional information from the network as well as a lot of inputs related to the users neighborhood and behavior. Therefore, this kind of prediction is not convenient or practical for utilization in real networks. In contrast, the TTP requires no additional

Fig. 17. Target BS prediction hit ratio over HOZone for a set of combinations of HDT, HM, and WS: (a) Scenarios A-F, (b) Scenarios G-L.

Z. Becvar et al. / Computer Networks 55 (2011) 37593773

3771

Fig. 18. Ratio of not predicted handovers over HOZone for a set of combinations of HDT, HM, and WS: (a) Scenarios A-F, (b) Scenarios G-L.

Fig. 19. Ratio of wrongly predicted target BSs over HOZone for a set of combinations of HDT, HM, and WS: (a) Scenarios A-F, (b) Scenarios G-L.

Table 4 Comparison of efciency of proposed TTP with other prediction techniques. Prediction technique Mobility extrapolation; Kwon [26] Mobility pattern prediction; Chan [25] Grey/Fourier/Kalman/Particle ltering; Bellavista [18] Trajectory prediction; Samaan [14] Proposed technique TTP Maximum prediction efciency 65% 75% 7881% 93% 93%

information on either users behavior or neighborhood knowledge. Further, the efciency of elimination of redundant handovers by TTP (for scenarios K and F) and ltering techniques investigated in [18] is presented in Fig. 20 to prove that our proposal is able to cope sufciently with this problem. Two scenarios (K and F) are selected since they both achieve very high efciency of handover prediction. Scenario K could be regarded as the scenario with the lowest

efciency of elimination of redundant handovers among the scenarios reaching high prediction efciency due to its low HM and WS. Therefore, scenario K can be considered as one of the worst case scenarios from the point of view of elimination of redundant handovers. On the other hand, scenario F promises higher elimination of redundant handovers since HM is set to a higher level. Comparison of TTP with other proposals is not included since not enough information is mentioned in these papers for simulations to be performed. All handovers that lead to the ping-pong effect are considered as redundant handovers. The values on the x axis represent the time interval after the handover initiation when another handover initiation is determined to be useless and leads to a redundant handover. For example, 5 s on the x axis means that if the handover is initiated until 5 s after the previous handover of the same MS, the rst handover is considered redundant. The y axis expresses the ratio of the amount of redundant handovers to the count of redundant handovers when no technique for their elimination is used.

3772

Z. Becvar et al. / Computer Networks 55 (2011) 37593773

Acknowledgement This work has been performed in the framework of the FP7 project ROCKET IST-215282 STP, which is funded by the European Community. The Authors would like to acknowledge the contributions of their colleagues from ROCKET Consortium (http://www.ict-rocket.eu).

References
[1] IEEE 802.16m-08/004r5: IEEE 802.16m Evaluation Methodology Document, 2009. [2] Z. Becvar, P. Mach, R. Bestak, Impact of Handover on VoIP Speech Quality in WiMAX Networks, in: International Conference on Networks (ICN 2009), Cancun, Mexico, 2009, pp. 281286, doi:10.1109/ICN.2009.11. [3] IEEE 802.16e-2005: Air Interface for Fixed and Mobile Broadband Wireless Access Systems, Amendment 2: Physical and Medium Access Control Layers for Combined Fixed and Mobile Operation in Licensed Bands, and Corrigendum 1, New York, 2006. [4] M. Schinnenburg, I. Forkel, B. Haverkamp, Realization and optimization of soft and softer handover in UMTS networks, in: Fifth European Personal Mobile Communications Conference, 2003, pp. 603607. [5] M.M. Zonoozi, P. Dassanayake, Optimum hysteresis level and signal averaging time selection for minimizing unnecessary handovers, in: Australian Telecommunication Networks and Application Conference (ATNAC96), Melbourne, 1996. [6] V. Pla, J.M. Gimnez-Guzmn, J. Martnez-Bauset, V. Casares-Giner, Optimal admission control using handover prediction in mobile cellular networks, in: Second International Working Conference on Performance Modeling and Evaluation of Heterogeneous Networks, 2004. [7] J. Martinez-Bauset, J.M. Gimenez-Guzman, V. Pla, Optimal admission control in multimedia mobile networks with handover prediction, Wireless Communications 15 (5) (2008) 3844. DOI: 10.1109/ MWC.2008.4653130. [8] L.-L. Lu, J.-L.C. Wu, Handoff prediction by mobility characteristics in wireless broadband networks, in: Sixth IEEE International Symposium on a World of Wireless Mobile and Multimedia Networks, 2005, pp. 469471, DOI: 10.1109/WOWMOM.2005.49. [9] L. Perato, K. Al Agha, Handover prediction: user approach versus cell approach, in: Fourth International Workshop on Mobile and Wireless Communications Network, 2002. [10] L. Perato, K. Al Agha, Handover Prediction for cellular systems in multi-services context, in: Fifth International Symposium on Wireless Personal Multimedia Communications, 2002. [11] Z. Becvar, P. Mach, Fast predicted handover in IEEE 802.16 networks, European Transaction on Telecommunications 22 (2) (2011) 6880. DOI: 10.1002/ett.1445. [12] Z. Becvar, Efciency of handover prediction based on handover history, Journal of Convergence Information Technology 4 (4) (2009) 4147. DOI: 10.4156/jcit.vol4.issue4.7. [13] W.S. Soh, H.S. Kim, QOS provisioning in cellular networks based on mobility prediction techniques, IEEE Communications Magazine 41 (1) (2003) 8692. [14] N. Samaan, A. Karmouch, A mobility prediction architecture based on contextual knowledge and spatial conceptual maps, IEEE Transaction on Mobile Computing 4 (6) (2005) 537551. DOI: 10.1109/TMC.2005.74. [15] GPS System Description, Global Positioning System Precise Positioning Service Performance Standard, 2007. [16] S. Chakraborty, Y. Dong, D.K.Y. Yau, J.C.S. Lui, On the effectiveness of movement prediction to reduce energy consumption in wireless communication, IEEE Transactions on Mobile Computing 5 (2) (2006) 157169. DOI: 10.1109/TMC.2006.24. [17] B. Liang, Z.J. Haas, Predictive distance-based mobility management for multidimensional PCS networks, IEEE/ACM Transactions on Networking 11 (5) (2003) 718732. DOI: 10.1109/TNET.2003. 815301. [18] P. Bellavista, A. Corradi, C. Giannelli, Evaluating ltering strategies for decentralized handover prediction in the wireless internet, in: 11th IEEE Symposium on Computers and Communications (ISCC06), 2006, doi:10.1109/ISCC.2006.70.

Fig. 20. Efciency of elimination of redundant handovers.

As can be observed in Fig. 20, the performance of TTP scenario K is close to the performance of the ltering methods evaluated in [18] from the point of view of efciency of redundant handover elimination. In contrast, TTP scenario F eliminates nearly all redundant handovers and thus outperforms ltering methods. However, both techniques (TTP as well as the ltering one) show very high efciency since more than 98.6% of redundant handovers are eliminated even if the interval for redundant handovers lasts 10 s. 7. Conclusions and future work The paper proposes a new technique for handover prediction based on channel characteristics. The proposed TTP technique differs from other prediction techniques in the denition of two thresholds derived from the signal levels among the MS and the neighboring BSs at the time of the handover initiation. One threshold is related to the serving BS and the second is related to the potential target BS. As the results indicate, the prediction hit ratio can be positively inuenced by use of HM, HDT, and windowing techniques. The maximum prediction hit ratio reaches 93% if individual techniques are combined, and it depends on the level of HOZone. The best prediction performance is usually achieved for HOZone equal to 4 dB. The proposed technique enables signicantly higher prediction efciency to be obtained in comparison to signal ltering methods. The proposed solution outperforms a similar technique with only one threshold by roughly 10%. The proposed solution requires exchange of no additional information except information sent conventionally among MSs and BSs during the normal operation of a WiMAX network. Additionally, no inputs from users or special knowledge or capabilities of networks are required. Hence no modications to the conventional WiMAX MAC layer or hardware are required. Future work will tackle issues of utilization of the prediction results for the purpose of reducing interruptions in handovers. Further, the same principle of prediction will be analyzed and evaluated in a scenario for vertical handovers.

Z. Becvar et al. / Computer Networks 55 (2011) 37593773 [19] Q. Zhou, W. Lv, S. Hu, J. Wang, A prediction-based handover decision for wireless networks, in: IEEE International Conference on Mobile Adhoc and Sensor Systems, 2006. [20] Y. Lin, S. Liu, A historical introduction to grey systems theory, in: IEEE International Conference on Systems, Man and Cybernetics, 2004, pp. 24032408, doi:10.1109/ICSMC.2004.1400689. [21] G. Welch, G. Bishop, An Introduction to the Kalman Filter, UNCChapel Hill, TR 95-041, 2004. [22] P. Bloomeld, Fourier Analysis of Time Series: An Introduction, second ed., John Wiley & Sons, 2000. [23] R. van der Merwe, A. Doucet, J.F.G. De Freitas, E. Wan, The unscented particle lter, Advances in Neural Information Processing Systems (NIPS13) (2000). [24] S.S.C. Rezaei, B.H. Khalaj, Grey Prediction Based Handoff Algorithm, World Academy of Science, Engineering and Technology 2, 2005, <http://www.waset.org/journals/waset/v2/v2-15.pdf>, Accessed: May 25, 2009. [25] J. Chan, S. Zhou, A. Seneviratne, A QoS adaptive mobility prediction scheme for wireless networks, in: IEEE Global Telecommunications Conference, 1998, pp. 14141419, doi:10.1109/GLOCOM. 1998. 776573. [26] H. Kwon, M.-J.Yang, A.-S. Park, S. Venkatesan, Handover prediction strategy for 3G-WLAN overlay networks, in: IEEE Network Operations and Management Symposium, 2008, pp. 819822, doi:10.1109/NOMS.2008.4575222. [27] M. Zonoozi, P. Dassanayake, M. Faulkner, Optimum hysteresis level, signal averaging time and handover delay, in: 47th IEEE Vehicular Technology Conference 1997, pp. 310313, doi:10.1109/VETEC. 1997.596370. [28] C. Hoymann et al., Advanced radio resource management algorithms for relay-based networks, IST-27675 STP FIREWORKS project deliverable 2D2, pp.121125, 2007, available at: <http:// reworks.intranet.gr/deliverables/Fireworks_2D2.pdf>, cited August, 25th, 2009. [29] IEEE 802.16j: Air Interface for Broadband Wireless Access Systems Amendment 1: Multihop Relay Specication, 2009. [30] T. Camp, J. Boleng, V. Davies, A survey of mobility models for ad hoc network research, Wireless Communications & Mobile Computing 2 (5) (2002). DOI: 10.1002/WCM.72. [31] A.F. Molisch, Wireless Communications, John Wiley & Sons, 2005. [32] K. Sambale et al., Mechanisms for increasing the efciency of MAC/ PHY protocols, ICT-215282 STP ROCKET project deliverable 5D1, pp.8691, 2009, available at: <http://www.ict-rocket.eu/documents/ Deliverables/ROCKET_5D1RWTHb.pdf>, cited February 25th, 2010.

3773

Pavel Mach received his M.Sc. and Ph.D. degree in Telecommunication engineering from Czech Technical University, Prague, Czech Republic in 2006 and 2010 respectively. During his study he joined research groups at Sintronics and R&D centers focusing on wireless mobile technologies. He has been actively involved in several national and international projects. He participated in EU FP projects FIREWORKS, ROCKET and he currently participates in EU FP7 project FREEDOM. His research interests include MAN/LAN networks based on relay architectures. He is dealing with aspects relating to radio resource management in emerging wireless technologies and focuses on cross layer optimization processes.

Boris imk is currently professor and the dean of Faculty of Electrical Engineering, Czech Technical University in Prague. He is actively involved in research of digital signal processing. He is the technical director of R&D centre for mobile communication at CTU in Prague and the member of executive board of Sitronics Centre at CTU in Prague. He participated on several national and international projects.

Zdenek Becvar received MSc in telecommunication at the Czech Technical University in Prague, Faculty of Electrical Engineering in 2005 and Ph.D. degree in 2010. In 2008, he became a researcher at the Department of Telecommunication Engineering. His current research interests include MAC procedures in wireless networks (LTE, LTE-A, WiMAX) with focus on radio resource management and mobility support. He participated in several ICT FP6 and FP7 projects. Furthermore, he was actively involved in research activities of Vodafone and Sitronics R&D centres at CTU in Prague and in several national research projects.

Potrebbero piacerti anche