Interference aware vertical handoff decision algorithm for

Computer Networks 54 (2010) 726–740

Contents lists available at ScienceDirect

Computer Networks

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

Interference aware vertical handoff decision algorithm for qualityof service support in wireless heterogeneous networks

Celal Çeken a,*, Serhan Yarkan b, Hüseyin Arslan b

a University of Kocaeli, Technical Education Faculty, Electronics and Computer Education Department, 41380 Kocaeli, Turkeyb Department of Electrical Engineering, University of South Florida, 4202 E. Fowler Avenue, ENB–118, Tampa, FL 33620, United States

a r t i c l e i n f o a b s t r a c t

Article history:Received 9 January 2009Received in revised form 8 September 2009Accepted 25 September 2009Available online 12 October 2009Responsible Editor: Ing. W. Kellerer

Keywords:Vertical handoffHeterogeneous networksDecision makingFuzzy logicInterference

1389-1286/$ - see front matter � 2009 Elsevier B.Vdoi:10.1016/j.comnet.2009.09.018

* Corresponding author. Tel.: +90 2623032240.E-mail addresses: [email protected] (C. Çe

usf.edu (S. Yarkan), [email protected] (H. Arslan).

Next generation wireless networks concept aims at collaboration of various radio accesstechnologies in order to provide quality of service (QoS) supported and cost efficient con-nections at anywhere and anytime. Since the next generation wireless systems areexpected to be of heterogeneous topology, traditional handoff (horizontal handoff/hand-over) mechanisms are not sufficient to meet the requirements of these types of networks.More intelligent vertical handoff algorithms which consider user profiles, applicationrequirements, and network conditions must be employed in order to provide enhancedperformance results for both user and network. Moreover, frequency reuse of one (FRO)seems to be the strongest candidate of deployment options for next generation wirelessnetworks; therefore, interference conditions gains a significant attention in vertical hand-off decision making process. In this study, a fuzzy logic-based handoff decision algorithm isintroduced for wireless heterogeneous networks. The parameters; data rate, received sig-nal strength indicator (RSSI), and mobile speed are considered as inputs of the proposedfuzzy-based system in order to decide handoff initialization process and select the bestcandidate access point around a smart mobile terminal. Also, in contrast to the traditionalfuzzy-based algorithms, the method proposed takes ambient interference power, which isreferred to as interference rate, as another input to the decision process. The results showthat the performance is significantly enhanced for both user and network by the methodproposed.

� 2009 Elsevier B.V. All rights reserved.

1. Introduction

As wireless communication evolves, many different sig-naling schemes and methods emerge and are standardized.Despite the obvious dominance of some specific types ofnetworks in daily life such as cellular mobile networks,emergence of new standards forces different types of net-works to coexist. Therefore, the concept ‘‘next generationnetworks” needs to allow for heterogeneous structureand to aim at collaboration of these various wireless tech-

. All rights reserved.

ken), syarkan@mail.

nologies in order to provide quality of service (QoS)supported and cost efficient connections at anywhere andanytime.

Even though heterogeneous network structure is a verybig concern by itself, terminals within the networks bringabout some other issues for the next generation wirelessnetworks. In this regard, it can be said that the next gener-ation wireless networks need to include terminals that areable to: (i) be aware of the existence of heterogeneous net-works around, and (ii) manage seamless transitions be-tween existing networks when necessary. Considering (i)and (ii) together, one can deduce that recently emergingtechnology called cognitive radio (CR) can be a remedyfor both, since CRs are able to be aware of, learn about,and adapt to the changing conditions in radio environment

http://dx.doi.org/10.1016/j.comnet.2009.09.018

mailto:[email protected]

mailto:syarkan@mail. usf.edu

mailto:syarkan@mail. usf.edu

mailto:[email protected]

http://www.sciencedirect.com/science/journal/13891286

http://www.elsevier.com/locate/comnet

C. Çeken et al. / Computer Networks 54 (2010) 726–740 727

[1]. Here, note that the term ‘‘radio environment” hasmany aspects such as physical propagation environment,radio frequency spectrum, available networks and termi-nals around, and so on. Because these aspects are highlydynamic, capabilities of CRs become crucial from the per-spective of both individual terminals and networks includ-ing them. Along with CR, cooperative networks conceptshould also be considered in the context of next generationwireless systems. The cooperative networks concept as-sumes that all of the wireless technologies such as cellularnetworks, wireless local/metropolitan area networks, wire-less personal area networks, short range communications,and digital video/audio broadcasting can coexist in a heter-ogeneous wireless-access infrastructure and cooperate inan optimal way in order to provide high speed and reliableconnectivity anywhere and anytime [2].

In addition to the considerations related to the capabil-ities of both terminals and networks, next generation wire-less networks should also provide high data ratetransmission despite their heterogeneous topology andcomplex structure. Moreover, they are desired to performas close to wired networks as possible over wireless med-ium in terms of cost efficiency and of supporting highlysophisticated services that imply seamless transmissionof different traffic types such as voice, data, and video.

Handoff is described as a process of transferring anongoing call or data session from one access point to an-other in wireless networks. When all of the aforemen-tioned aspects are contemplated, it is not difficult toconclude that handoff will be one of the vital mechanismsfor next generation wireless networks. Traditional handoffprocess, which is called horizontal handoff, takes place toprovide an uninterrupted service when a user moves be-tween two adjacent cells. Generally, horizontal handoffprocess is initialized when the link quality conditionparameters such as received signal strength indicator(RSSI), signal-to-noise ratio (SNR), and so on drop belowa specified handoff threshold. Because the next generationwireless systems involve a heterogeneous topology, tradi-tional handoff mechanisms will not be sufficient. There-fore, a new type of handoff, which is known as ‘‘verticalhandoff,” is introduced. Vertical handoff is defined as aprocess which transfers a user connection from one tech-nology to another such as a transfer from Global Servicefor Mobile (GSM) to WLAN or to WiMAX. Vertical handoffrequires more intelligent algorithms which evaluate moreparameters such as interference power, monetary cost,QoS, remaining energy, and so on in addition to alreadyexisting link quality condition quantifiers. Among all ofthe parameters, interference must be treated in a separateplace, since every wireless system is interference limited.In traditional cellular-based systems, harmful impact ofinterference is tried to avoid/minimize by reusing theavailable frequencies in distant cells, which is called ‘‘fre-quency reuse.” In the literature, frequency reuse is quanti-fied by a factor called ‘‘frequency reuse factor” whichrepresents the distinct number of frequency sets (or equiv-alently, number of cells) in a cluster. In this regard, mini-mum frequency reuse factor can be one and it is called‘‘frequency reuse of one (FRO)” or universal frequency re-use. FRO implies that each cell is allowed to use the entire

spectrum available. Although frequency reuse is a veryeffective method in combating interference, it comes atthe expense of inefficient spectrum usage and of expensivedesign processes. In contrast to traditional systems, FROseems to be the strongest candidate in order to avoidexpensive planning process and to overcome the problemof under utilized resource use in next generation wirelessnetworks. It must be noted that FRO causes significantinterference levels especially in the vicinity of cell bordersin return. This renders interference one of the most criticalparameters in handoff process for next generation wirelessnetworks.

In the light of discussions given above, it is clear thatinterference, data rate, and mobility constitute the threeprominent aspects of the next generation wireless net-works. In this study, a new smart mobile terminal (SMT)is proposed. The proposed SMT is assumed to have cogni-tive capabilities such as sensing the environment periodi-cally for available radio access technologies (RATs),evaluating their working conditions using its fuzzy logic-based algorithm, triggering handoff process if necessary,and deciding the best access point (AP) to camp on. Thedecision is based on interference rate, data rate, and RSSIdue to the following reasons: Interference rate quantifieshow severe the ambient co-channel interference power le-vel, data rate takes into account the available transmissionrate for applications carried out, whereas RSSI roughlyhelps to evaluate the mobility. The proposed SMT is mod-eled and simulated using OPNET Modeler Software for per-formance evaluation. Besides, the fuzzy logic-basedhandoff algorithm incorporated in SMT is implemented inMATLAB Software. The contributions of this study can besummarized as follows:

� Considering the fact that most of the wireless communi-cation systems are interference limited, in contrast tothe most of the fuzzy-based algorithms, the decisionmechanism in the method proposed takes into accountinterference rates from different base stations as inputto its fuzzy logic system in order to make a more reliablehandoff.

� A new adaptive multi-criteria handoff decision system,which has the ability to adapt its structure accordingto the application requirements and network conditions,is proposed.

� A new cognitive smart terminal, which senses the envi-ronment for available APs and changes its workingparameters such as frequency band, bandwidth, modu-lation scheme, medium access control (MAC) protocoland so on in order to camp on an appropriate AP, isdeveloped.

The remainder of the paper is organized as follows: Sec-tion 2 presents related works to the vertical handoff in theliterature. Section 3 provides the proposed models for SMT,handoff, and base station (BS). Section 4 includes exampleheterogeneous network scenarios which have overlappingRATs with different working parameters as well as pro-posed SMT, followed by performance evaluation. The paperis concluded with Section 5 providing final remarks and adiscussion.

728 C. Çeken et al. / Computer Networks 54 (2010) 726–740

2. Related works

Although the vertical handoff concept is relatively new,several studies can still be found in the literature. In [3],the authors discuss different factors and metrics whichare considered when triggering handoff. Besides, they de-scribe a vertical handoff decision function (VHDF) whichenables devices to assign weights to different network fac-tors such as monetary cost, QoS, power requirements, per-sonal preference, and so on.

A novel fuzzy logic-based handoff decision algorithmfor the mobile subsystem of tactical communications sys-tems is introduced in [4]. Handoff decision metrics used in[4] are: RSSI, the ratio of the used capacity to the totalcapacity for the access points, and relative directionsand speeds of the mobiles to APs. The authors comparetheir algorithm with the RSSI-based handoff decision algo-rithm as well. Note that in [4, Eq. (3)], ambient interfer-ence power is embedded into the parameter of capacity,rather than being used as a direct input to the decisionprocess.

In [5], the author presents a review on the proposedvertical handoff management, and focuses on the decisionmaking algorithms in vertical handoff. The article [6] pre-sents a tutorial on the design and performance issues forvertical handoff in an envisioned multi-network fourth-generation environment. In [7], the authors give a fuzzy lo-gic based vertical handoff scheme involving some keyparameters and the solution of the wireless network selec-tion problem using a fuzzy multiple attribute decisionmaking (FMADM) algorithm.

It is considered that adaptation is crucial for next gener-ation wireless networks from every aspect, such as handoffmanagement and scheduling, since FRO seems to be one ofthe strongest deployment candidates [8–10]. Becauseinterference is a very dynamic phenomenon, success ofthe adaptation of next generation wireless networks de-pends on being aware of the factors affecting it [11,12].Therefore, the traditional fuzzy-based algorithms might

Fig. 1. The SMT cross-lay

not be able to meet the requirements of next generationwireless networks unless they take into account interfer-ence in their decision procedures. To the best knowledgeof authors, none of the fuzzy-based handoff algorithmsconsiders ambient interference power level as a direct in-put to their decision mechanisms.

3. The proposed models and algorithms for verticalhandoff

Vertical handoff is a process issue compared to horizon-tal handoff as explained earlier. The cognitive smart termi-nal proposed in this study is in complete charge ofmanaging the handoff process as well as its other func-tions. It scans the environment periodically for availableAPs, obtains the operating parameters, combines and pro-cesses all necessary parameters using its fuzzy logic-basedclassifier, initializes handoff process, and chooses the bestcandidate AP. The following subsections include the pro-posed models and algorithms implemented using OPNETModeler simulation tool and MATLAB software.

3.1. Smart terminal process model

The proposed smart terminal process model has a cross-layer design and is developed using OPNET Modeler soft-ware. It includes physical, MAC, and some upper layerfunctions. It has a carrier sense multiple access/collisionavoidance (CSMA/CA) MAC module for the wireless fidelity(WiFi) capability and a GSM module to handle GSM opera-tions. Besides, it has a fuzzy logic-based smart handoffdecision unit which is in charge of managing all of thehandoff operations.

Fig. 1 outlines the state transition diagram of the SMTprocess model. The process starts with the Init state. Thisstate performs a delay until the other processes in thesimulation are initialized and loads the control variables.Then the process enters the Spectrum Scan and Handoff

er process model.


Decision states which are responsible for scanning theenvironment for available APs and managing the entirehandoff process exploiting the proposed fuzzy logic-basedhandoff algorithms. The WiFi Mode and GSM Mode statesstand for WiFi and GSM functionalities, respectively.

During the spectrum sensing phase, SMT listens towireless medium for any handoff broadcast packet whichmight be sent by potential APs for a specified time span.All of the GSM APs has a broadcast control channel (BCCH)which is the first channel of allocated spectrum and is usedfor broadcasting network information periodically for pos-sible handoff process in addition to its other functions. Thedetailed information about GSM technology can be foundin [13]. The WiFi APs broadcast a handoff informationpacket periodically for this purpose as well. During the lis-tening period, the SMT changes its working parameterssuch as frequency, modulation, data rate, and bandwidthin order to adapt to any possible AP and to receive handoffbroadcast packet.

When any AP is available, SMT receives the handoffbroadcast packet and extracts the network workingparameters. It then invokes fuzzy-based handoff decisionalgorithm which takes these parameters as inputs; pro-cesses them; and produces an output called AP candidacyvalue (APCV). APCV is generally defined by a real numberin order to quantify the strength of the candidacy level ofthe AP found. For instance, APCV can be designed to varybetween one and ten where one denotes the weakest,whereas ten represents the strongest candidacy level ofquantification. Subsequently, all the aforementioned net-work parameters along with APCV are stored in the hand-off decision table (HDT) for further usage.

All of these steps are repeated until the scan process isterminated. In each turn, SMT listens to the environmentfor potential APs, receives the handoff broadcast packetof the AP found, calculates the APCV using its adaptive fuz-zy inference system, and stores all of the pieces of informa-tion required in the HDT.

The sequence diagram of the proposed handoff decisionalgorithm is outlined in Fig. 2. This schema is repeated inevery 10 s throughout the simulation run time.

As soon as the scan process is completed, APCV of eachavailable AP is compared with that of current APs. If thedifference between the compared values is equal to orgreater than the handoff resolution (HR)1, that is a valuedetermined by user, then the second condition, i.e., mo-bile speed, is evaluated. The mobile speed 10 km/h is se-lected as a threshold value. Any speed value below thisthreshold is regarded as walking speed and in this case,either any GSM or WiFi AP can be chosen as a servingnode. Otherwise, only GSM network can be preferred,since the WiFi AP might serve SMT only for a very shortduration. When these conditions are satisfied, handoffprocess is initialized.

1 Handoff resolution (HR) value is used to introduce a hysteresis to theproposed vertical handoff algorithm. Mobile terminal takes into accountthe HR value in order to decide whether any handoff process is required ornot. If the candidacy level of any potential AP is greater with an amount ofspecified HR value than that of the current AP, then the handoff process isinitialized.

3.2. Proposed handoff decision algorithm

Sophisticated handoff decision algorithms should con-sider more than one criteria and a methodology to com-bine and process them. Different decision algorithmshave been proposed in the literature for vertical handoffas mentioned earlier. Artificial intelligence-based systemssuch as fuzzy logic and artificial neural networks are goodcandidates for pattern classifiers due to their non-linearityand generalization capability [4,14]. Therefore, in the pro-posed handoff decision system a fuzzy logic-based ap-proach has been adopted.

Vertical handoff decision algorithm should initializehandoff process considering available network interfaces(link capacity, power consumption, link cost, and so on),system information (remaining battery), and user/applica-tion requirements (cost, QoS parameters, and so on). Theblock diagram of the proposed handoff decision system isgiven in Fig. 3.

The algorithm combines the user/application require-ments and network capabilities, and produces an outputwhich is utilized to make handoff decision and to choosethe best candidate AP.2 In the proposed handoff system,there are three inputs (data rate, interference rate, and RSSI)for fuzzy inference system. Membership functions of theseinputs are given in Figs. 4–6, respectively. In the figures,the horizontal axis indicates the crisp values of the afore-mentioned handoff parameters, whereas the vertical axis(i.e. l values) stands for the membership value of relatedparameter. The crisp inputs are converted into the fuzzy var-iable by means of these membership functions. Trim andtrapezoid shapes are chosen as fuzzy membership functionsdue to their capability of achieving better performance espe-cially in real time applications.

The data rate (DR) input has the ability to change itsstructure according to the application requirements aswell. For instance, if the DR requirement of an applicationis 9.6 Kbps (GSM data transfer), then the membershipfunction is similar to the one given in Fig. 4a. On the otherhand, when the application needs more bandwidth, e.g.,25 Kbps (GPRS Class 6 traffic), then it dynamically changesits structure to adapt the new working condition as seenfrom Fig. 4b.

The interference rate parameter is also obtained by eachAP and sent to the SMT in order to be considered in handoffdecision process. In the proposed algorithm, interferencerate refers to a special fuzzy logic variable whose value isdetermined by the ambient CCI power level. In GSM, Thespecification regarding the CCI level (GSM 05.05) recom-mends that the carrier-to interference ratio (C/I) is at least9 dB in order to meet the bit-error rate (BER) requirement.This indicates that for a reference signal level of�101 dB m, which is the standard value for a GSM mobilestation, CCI power level should be less than �110 dB m.However, in a typical GSM deployment, it is reported that

2 It is worth mentioning here that data rate and interference ratiocalculations in Fig. 3 can be considered as parameters related to the user/application requirements, since there are certain values for these param-eters mandated by the application (e.g., data transmission or voiceservices).

Fig. 2. Sequence diagram of the proposed handoff decision algorithm.


C/I is around 14–15 dB [15]. In this sense, interference rateparameter is designed in such a way that the power leveldifference between the desired signal and interference islower than �14 dB is assigned to be a low fuzzy logic value,whereas a difference that is greater than 14 dB is selectedto be a high fuzzy logic variable. The corresponding mem-bership function for the interference rate parameter can beseen in Fig. 5.

The RSSI input of the fuzzy system has also the abilityto change its structure according to the network require-ments. The RSSI membership function for GSM and WiFinetworks are different as shown in Fig. 6a and b,respectively.

As stated earlier, according to the inputs of availableAPs the fuzzy inference system produces an output valuebetween one and ten which describes the candidacy levelof related AP. Any handoff initialization process is decidedupon this value. One of the most crucial parts of this studyis the new adaptive fuzzy inference system which is devel-oped in order to make handoff decision. A fuzzy logic sys-tem consists of three main parts: Fuzzifier, InferenceEngine, and Defuzzifier. Fuzzifier converts a crisp inputinto a fuzzy variable where physical quantities are repre-sented by linguistic variables with appropriate member-ship functions. These linguistic variables are then used inrule base of Fuzzy Inference Engine. Since there are three

Fig. 3. Block diagram of the proposed fuzzy logic-based handoff decision system.

Fig. 4. Fuzzy membership function for two different data rates (DR).

Fig. 5. Fuzzy membership function for interference rate (IR).


input variables each has three levels (i.e., low, medium,and high), there are 27 rules used for producing a newset of fuzzy linguistic variables. Some of the fuzzy rulesin the rule base are tabulated in Table 1. For instance, Rule

1 corresponds to the following IF–THEN structure: if thepotential AP supports low data rate, it is in a bad interfer-ence condition, and its RSSI is weak, then the APCV of theAP is 1, which means it is not a strong candidate. On theother hand, Rule 21 outputs a greater APCV value, i.e. 10,which implies the APs candidacy level is quite high.

Defuzzifier is responsible for converting this fuzzy en-gine output into a number called APCV. The output of thefuzzy system, APCV, is then combined with the mobilespeed parameter to make handoff decision.

The analytic model of the fuzzy inference system is asfollows [4,16]. Three dimensional pattern vector (input ofthe fuzzifier) for candidate access points is:

PVc ¼ ½DRc; IRc; RSc�; ð1Þ

where DR is data rate, IR is interference rate, and RS is RSSIvalue of available AP. Three dimensional fuzzy pattern vec-tor (output of fuzzifier and input of inference engine) forcandidate access points:

Fig. 6. Fuzzy membership functions for two different RSSIs (RS).

Table 1Example fuzzy rules.

Rule IF THEN

1 (DataRate is low) and (InterferenceRate is low) and (RSSI is weak) (APCV is 1)2 (DataRate is low) and (InterferenceRate is low) and (RSSI is medium) (APCV is 2)3 (DataRate is low) and (InterferenceRate is low) and (RSSI is strong) (APCV is 1). . . . . . . . .

21 (DataRate is high) and (InterferenceRate is low) and (RSSI is strong) (APCV is 10)


PVF ¼ ½PF1; PF2; PF3�: ð2Þ

Since product inference rule is utilized in the fuzzy infer-ence engine then, for a new pattern vector, contributionof each rule in the fuzzy rule base is:

Cr ¼Y3

i¼1

lFiðPiÞ; ð3Þ

where lFiðPiÞ is the membership value of the Pi for fuzzy

set Fi, and obtained from the aforementioned membershipfunctions. There are 27 rules in the proposed system and acenter average defuzzifier is used. Hence, the output of thedefuzzifier (i.e., APCV) becomes:

Ma ¼

P27l¼1yl Q3

i¼1lFiðPiÞ

� �

P27l¼1

Q3i¼1

lFiðPiÞ

� � ;

where, yl is the output of the rule l.

3.3. GSM base station

In GSM, each BS periodically broadcasts handoff deci-sion packet using its BCCH in order to convey networkinformation to any terminal in the vicinity for a possiblehandoff. In addition to this, in the proposed method, BSis in charge of both determining its interference rate andconveying this information to the SMT with the aid of itshandoff broadcast packet. Fig. 7 shows the proposed GSMprocess model realized using OPNET Modeler.

As in the former process model, the process starts withthe init state as well, then enters the idle state, and waitsthere until a specific interrupt arrives. The fromRx state

machine delivers arriving packets to the next state ma-chine, which can be either bwRequest or data state ma-chine depending on the packet formats. If the bwRequeststate machine is selected, then connection establishment/termination requests are handled. In case the data statemachine is selected, then the received data is deliveredto its destination. The interference state machine deter-mines the interference rate of the BS and inserts this infor-mation into the related field of handoff broadcast packet.Finally, the handoff broadcast packet is created, providedwith all of the required information, and then broadcastedto the environment in the handoff state machine.

4. Numerical results and discussions

4.1. Assumptions

In this study, two different simulation scenarios areevaluated in order to investigate performance of the devel-oped models and algorithms. The first one, which is illus-trated in Fig. 8 and referred as ‘‘Scenario 1” in theremainder of the paper, has a higher interference rate,whereas the second one, which is given in Fig. 9 and re-ferred as ‘‘Scenario 2” in the remainder of the paper, hasa lower interference rate. Having these two scenarios inhand, the impact of interference can be extracted andinterpreted from the simulation results. Three wirelessnetworks are considered in the simulation scenarios. Oneof them is a WiFi network and the rest are GSM networkseach of which has specific working parameters. GSM–WiFipair is selected due to the following reasons: Both GSM andWiFi are very well-established and vastly deployed stan-dards that are currently being used; therefore, they consti-

Fig. 8. Vertical Handoff example with higher interference rate (referredas Scenario 1).

Fig. 7. Proposed process model for APs.

Fig. 9. Vertical Handoff example with lower interference rate (referred asScenario 2).


tute an appropriate case study for vertical handoff perfor-mance evaluations. In addition, GSM is a cellular standardin which handoff is one of the default network facilities,whereas WiFi does not have the concept handoff in thefirst place. From this perspective, handoff becomes a morechallenging problem compared to other possible networkpairs, such as GSM–CDMA pair, since CDMA contains thehandoff concept as well. Also, in spite of the fact that manyGSM deployments are not of FRO mode, GSM specificationallows FRO mode of operation. Note that it is always de-sired to see how the proposed method performs in theworst case scenario. Since the method proposed makes adecision in regards to CCI present in the environment,studying a scenario which poses the harshest interferencecondition would be a reasonable approach. Because FROintroduces significant interference power levels especiallyin the vicinity of cell borders, a deployment that is basedon FRO would be an appropriate scenario in order to eval-uate the performance of the method proposed. Therefore, aGSM deployment of FRO is considered in the simulatedscenarios. This way, it is also possible to get a clue aboutthe performance of the method proposed in next genera-tion wireless networks which are expected to deploy FRO.

Beside the networks, there are two mobile terminalswhich are served by GSM networks individually along witha smart terminal (SMT1) which has the capability of gener-ating time sensitive voice traffic (13 Kb/s), GSM data traffic(9.6 Kb/s), and Class 6 GPRS data traffic (25 Kb/s). In bothscenarios, MTi denotes the ith mobile terminal served bybase station labeled with BSi. These mobile nodes are ofcrucial importance for the method proposed, because theygenerate wireless traffic which causes interference toSMT1 and affects the decision regarding the handoff. It isworth mentioning that since interference spilled overSMT1 is a function of time and space, directions of motionof mobile terminals are important in terms of performanceevaluation of the method proposed. Note that SMT1 movesalong the trajectory shown in Figs. 8 and 9 of a specifiedspeed during the simulation run time, senses the environ-ment continuously for candidate APs, and has the afore-mentioned adaptive fuzzy logic-based handoff decision


system as well as the capability of performing both WiFiand GSM functionalities.

In this sequel, one might wonder about the impact oftraffic loads of APs present in the environment on the per-formance of the method proposed. In both of the scenariosconsidered in this study, it is assumed that one of the GSMbase stations can provide less data rate compared to theother. This assumption is actually the common conse-quence of the following two items: Either (I) applicationsof the user of interest require different bandwidths, suchas 9.6 Kbps for voice application and 25 Kbps for datatransmission, or (II) regardless of the application require-ments of the user of interest, GSM base stations can offerdifferent data rates for a newcomer due to their differenttraffic loads. In other words, one of the GSM base stationscan offer less data rate, because it cannot offer higher datarate due to its high traffic load present. From this point ofview, results for the scenarios considered in this study al-ready include the impact of traffic loads of APs present inthe environment.

Diameter of the cluster which constructs the overallnetwork topology is chosen to be four kilometers. The sim-ulation was run for 3600 s and other relevant parametersare tabulated in Table 2.

In addition to the scenario assumptions given above,due to its importance, interference assumptions need tobe explained further for the sake of completeness. Consid-ering the fact that dynamic condition of the mobile radiopropagation environment changes very rapidly, the re-ceived signal at a receiver can be defined for mobile radiosby the following two processes:

rðtÞ ¼ mðtÞsðtÞ; ð4Þ

where mðtÞ is the fast fading (i.e., short-term multipathfading) component and sðtÞ is the local mean (i.e., long-term shadow fading) of the received signal [17,18]. Gener-ally, mðtÞ and sðtÞ are identified in terms of probability dis-tributions, since both are random processes. In theliterature, Rayleigh and Rice distributions are frequentlyassigned the envelope of mðtÞ for non-line-of-sight (NLOS)and line-of-sight (LOS) cases, respectively. Field measure-ments reveal that shadowing process follows a log-normaldistribution with a standard deviation r which is defined,

Table 2Simulation parameters.

Parameter Value

Message sizea 13 Kb/s, 25 Kb/s, 9.6 Kb/sData rate WiFi = 1 Mb/s, GSM = 270, 833 b/sFrequency band WiFi = 2.4GHz, GSM = 890–935 MHz (with

FRO)Handoff

resolution3

Handoff threshold WiFi = �80 dB mTransmitter

powerWiFi = 100 mW, GSM = 1.5 W

Speed of mobile 10 km/hArea size 4 km � 4 kmScan period 10 s

a Generated using exponential distribution function exp (Mean).

again, in linear scale [18]. Hence, the following formalmodel is adopted for representing sðtÞ:

sðtÞ ¼ eðrgðtÞÞ; ð5Þ

where r is the standard deviation of the shadowing andgðtÞ is a normal process with Nð0;1Þ.3 It is also importantto note that mðtÞ and sðtÞ are statistically independent pro-cesses [18,19].

In regard to shadowing processsðtÞ, empirical studiesshow that there is a correlation in shadowing with respectto distance between two points in the space. Spatial corre-lation of shadowing is reported to be of an exponentiallydecaying form [20]. Considering practical purposes, thereare several models proposed in the literature. For instance,[21] defines the normalized relationship between shadow-ing and its spatial correlation as:

qðDdÞ ¼ e�Dd

dqlnð2Þ

� �; ð6Þ

where dq ¼ 5 m for indoor and dq ¼ 20 m for outdoorenvironments exemplifying practical scenarios.

Bearing in mind that small-scale and shadowing pro-cesses are mutually independent of each other, the pro-cess that represents the interfering signal can be formedby generating each process individually. In the simula-tions a Rayleigh fading envelope ðjmðtÞjÞ with a classicalJakes’ Doppler spectrum is adopted for emulating the mul-tipath component in (4) for OPNET platform [22]. Sinceshadowing process is described by a random variablewhose logarithm is normally distributed with Nð0;r2Þ,where r is the standard deviation of the shadowing,Gauss–Markov assumption can be used for approximatingit in order to capture the impact of the correlation given inas follows:

x½i� ¼ ax½i� 1� þffiffiffiffiffiffiffiffiffiffiffiffiffiffi1� a2p

n½i�; ð7Þ

where x½i� is the process observed, a is the correlation coef-ficient, and n½i� is assumed to be additive white Gaussiannoise (AWGN) with Nð0;1Þ. Here, in order for the desiredcorrelation properties to be captured, the following isnecessary:

a ¼ bðvDt=dqÞ: ð8Þ

In this sequel, it is worth mentioning that the selection of bdepends on the environment and there are several valuesfor several different environments such as b ¼ 0:3 fordq ¼ 100 m, whereas b ¼ 0:82 for dq ¼ 10 m [23].

4.2. Simulations and performance analysis

Before proceeding to the performance results and re-lated discussions, it is appropriate to evaluate how wellthe Gauss–Markov assumption can capture the shadowingprocess introduced in Section 4.1. In order to verify thevalidity of Gauss–Markov assumption, the settings givenfor a suburban area in [23] are adopted with the following

3 In this sequel, it must be stated that (5) actually contains a constantrepresenting the impac of path loss for the received signal at the instant t.However, the impact of path loss can be neglected under appropriateassumptions, such as pedestrian speed of motion.

Fig. 10. Correlation coefficient of shadowing process versus displacement for simulated and theoretical values.


variables: b ¼ 0:82;v ¼ 3 m=s;Dt ¼ 1 s, and dq ¼ 100 m.4

Within the decorrelation distance, each value of the shad-owing process is obtained with the aid of (7) for the afore-mentioned parameter set. The process is initiated by aninitial value which comes from the normal process n½��withNð0;1Þ similar to that in (5). Fig. 10 plots the correlationcoefficients versus displacement for both the theoreticaland the simulated cases. It is clear that simulated valuesfollow the theoretical ones very closely implying the suit-ability of the Gauss–Markov assumption for shadowingprocess generation. Note that the logarithm of the shadow-ing process is a normal process; therefore, it can com-pletely be described in terms of its mean and variance. If(7) is analyzed, one can easily verify that the generatedprocess meets the requirement of the shadowing processas well.

The first performance simulation set focuses on the casein which SMT1 moves along with the trajectory as illus-trated in Figs. 8 and 9 of a specified speed. The simulationof the example scenarios has been run for different appli-cations and working conditions, i.e., voice transfer and datatransfer (9.6 Kb/s and 25 Kb/s), in order to evaluate theperformance of the proposed approach comparatively.The performance metrics examined are; APCV of eachavailable AP, number of handoff(s), and average end-to-end (EED) delay between SMT1 and APs.

Fig. 11 illustrates the outputs of adaptive fuzzy logic-based handoff decision algorithm for a voice transfer appli-

4 It is known that WLAN is designed for very low mobility scenarioswhich can be considered of pedestrian speed classes in ITU-R StandardChannel Models. Therefore, considering a mobile which is faster than anaverage pedestrian speed is not reasonable, since a high-speed user handedover a WLAN base station cannot maintain the connection. Note also thatsince a high-speed user cannot be accommodated by WLAN, it cannot behanded over any GSM base station stemming from the fact that handoffmandates an existing connection by its very definition.

cation in Scenario 1. APCV of each available AP is computedonce in every 10 s during the spectrum sensing process bythe proposed fuzzy logic-based algorithm employed inSMT1. As can be seen from the figure, the APCV of the WiFihot spot is greater than those of the others when the sim-ulation starts. At 300 ms, WiFi is not able to provide higherdata rates anymore causing a decrease in the output of theproposed fuzzy system. Furthermore, because SMT1 movesout of the coverage area of WiFi, a rapid attenuation in RSSis experienced. Hence, the APCV is reduced once more be-tween simulation times 1550 and 2000 s. When producingAPCV, in case two candidate GSM BSs are of interest, theinterference rate becomes the determinant input stem-ming from the fact that GSM BSs have approximately sim-ilar RSSI and DR values. The APCV of the candidate GSM BSsvaries between 5 and 9.25 as can be seen from Fig. 11. Notethat, for the GSM BSs, behavior of the interference ratelooks similar to each other due to the motion of both mo-bile terminals in the vicinity of cell borders. These randommovements cause random fluctuations in the interferencerates observed as well. As a consequence, determiningthe candidate BS depends on the instantaneous interfer-ence rate values.

Fig. 12 presents the number of handoffs as a function ofHR value for Scenario 1. As expected, the number of hand-offs increases when SMT has a lower HR value. Hence, thelower the HR, the higher the number of handoffs. In orderto determine an optimum HR value, user preferences,application requirements, and/or network conditions needto be taken into account.

Fig. 13 illustrates the outputs of adaptive fuzzy logic-based handoff decision algorithm for Scenario 2 along withthe same procedure carried out in Scenario 1. Note that inScenario 2, the behavior of APCV for WiFi behaves exactlythe same as in Scenario 1. When producing APCV, in casetwo candidate GSM BSs are considered, the interference

Fig. 11. APCV output of the proposed handoff decision algorithm for voice transfer application in Scenario 1.

Fig. 12. Number of handoffs versus HR for Scenario 1.


rate is still the determinant input because GSM BSs haveapproximately similar RSSI and DR values. In general, APCVof GSM1 is greater than that of GSM2 most of the timesduring the simulation. Especially after 1000 s of the simu-lation run time, the dominance of the APCV value of GSM1is obvious. This stems from the motion of MT2 towardsGSM2 implying a diminishing interference power observedby GSM1. Therefore, the probability of selecting GSM1 asserving BS increases in time.

Fig. 14 presents the number of handoffs as a function ofHR value for Scenario 2. The number of handoff increaseswhen SMT has a lower HR value, as in Scenario 1. As com-pared to the one given in Fig. 12, the number of handoff isquite less in Fig. 14 as expected, since GSM1 has lowerinterference rates most of the times during the simulation.

In order to evaluate the performance of our proposedalgorithms comparatively, Fig. 15 can be examined.Fig. 15 illustrates the measured GSM and WiFi RSSI resultsas a function of the simulation run time. If any RSSI-basedhorizontal handoff algorithm considered, which is alterna-tive to the proposed multi-criteria approach, the followingtwo conclusions can be drawn from the performance fig-ure: (i) Since the RSSI values of the GSM are greater thanthose of WiFi along with the simulation run time; theSMT always prefers the GSM base station to camp on.However, this truth is not indicated in our study, it meansadditional monetary cost, since the WiFi is less expensivethan GSM network. (ii) According to our scenario (i.e., Sce-nario 2), the supported bandwidth by the WiFi is greaterthan GSM from simulation starting time to the 300 s. Dur-

Fig. 13. APCV output of the proposed handoff decision algorithm for voice transfer application in Scenario 2.

Fig. 14. Number of handoffs versus HR for Scenario 2.


ing this time proposed scheme chooses the WiFi AP,whereas the RSSI-based traditional algorithms choosesthe GSM BS due to its greater RSSI values, which can beraised unexpected results in terms of QoS parameters.

In Fig. 16, average EED results (between SMT1 and APs)of different application traffics are presented as a functionof the simulation run time. The EED results are obtained forScenario 2 which represents the better interference rateconditions. Note that although all of the settings used forScenario2 are the same as those for Scenario1, the mobilitybehavior of MTs in Scenario 2 is different than that in Sce-nario 1. A different mobility behavior with the same set-tings, as will be discussed here, will affect the results dueto changing interference conditions; therefore, it will form

a different simulation environment from the perspective ofdecision process proposed. In the light of this, SMT1 firstlycamps on WiFi hot spot since it has a sufficient data andlower interference rate along with a high RSSI. After300 s, the data rate supported by the WiFi AP dramaticallydecreases. Even though both RSSI and interference rate areappropriate, fuzzy-based handoff decision scheme decidesto change the AP considering inadequacy in the DR offered.As can be seen from the Fig. 9, there are two alternativeAPs. For data transfer application with 9.6 Kb/s and voicetransfer application GSM1 is selected as the new AP sinceboth the DR and RSSI value meet the QoS requirementsalong with a relatively lower interference rate. The latencyresults of voice transfer are lower than those of data trans-

Fig. 15. RSSI-based performance evaluation chart.

Fig. 16. EED results (SMT1-APs) for different application traffics.


fer with 9.6 Kb/s. This is not surprising, because one slot isguaranteed for voice transfer application, whereas even

reserving one slot might not be possible for data transferdue to dynamic network load conditions.


When the data transfer application with 25 Kb/s is run-ning on SMT1, a conflict between data and interferencerate arises. On the one hand, GSM1 observes a lower inter-ference rate compared to GSM2. On the other hand, GSM2can provide a higher DR compared to GSM1. Results showthat GSM2 AP is selected after the decrease in DR offeredby WiFi at 300 s of the simulation run time. This stemsfrom the fact that fuzzy logic system weights the dataand interference rate inputs differently because of QoSrequirements of the application running on SMT1. As canbe seen from Fig. 16, the average delay for this applicationis lower than those for the other applications, since GPRSClass 6 traffic type can be assigned up to four slots.

5. Conclusions

Vertical handoff is defined as a process which transfersa user connection from one technology to another. It is ex-pected that the next generation wireless systems will in-clude several different network technologies cooperatingwith each other; therefore, vertical handoff mechanismsshould be investigated in detail. In general, handoff mech-anisms allow for many parameters including user profiles,application requirements, and network conditions. In thiswork, an adaptive fuzzy-based handoff decision systemwhich combines data and interference rate, RSSI, and speedparameters is proposed in order to satisfy both user andnetwork requirements for next generation wireless hetero-geneous networks.

Simulation results show that the proposed fuzzy logic-based vertical handoff decision algorithm is able to deter-mine the most appropriate access network under differentdynamic working conditions. Moreover, this study revealsthat interference plays a crucial role in making verticalhandoff for next generation wireless networks.

Acknowledgements

This research has been supported by a grant from TheScientific and Technological Research Council of Türkiye(TÜB_ITAK).

References

[1] J. Mitola III, Cognitive radio an integrated agent architecture forsoftware defined radio, Ph.D. Dissertation, KTH Royal Institute ofTechnology, Stockholm, Sweden, May 8, 2000. <http://www.it.kth.se/jmitola/MitolaDissertation8Integrated.pdf>.

[2] L. Berlemann, G. Dimitrakopoulos, K. Moessner, J. Hoffmeyer,Cognitive radio and management of spectrum and radio resourcesin reconfigurable networks, Working Group 6 White Paper, 2005,Wireless World Research Forum.

[3] A. Hasswa, N. Nasser, H. Hassanein, Generic vertical handoff decisionfunction for heterogeneous wireless, in: Proceedings of the SecondIFIP International Conference on Wireless and OpticalCommunications Networks (WOCN 2005), Mar. 6–8, 2005, pp.239–243.

[4] T. Onel, C. Ersoy, E. Cayırcı, G. Parr, A multicriteria handoff decisionscheme for the next generation tactical communications systems,Computer Networks 46 (5) (2004) 695–708.

[5] A. Ezzouhairi, A. Quintero, S. Pierre, A fuzzy decision making strategyfor vertical handoffs, in: IEEE CCECE Conference, 2008, pp. 583–588.

[6] J. McNair, F. Zhu, Vertical handoffs in fourth-generationmultinetwork environments, IEEE Wireless Communications 11 (3)(2004) 8–15 (See also IEEE Personal Communications), June.

[7] Y. Nkansah-Gyekye, J.I. Agbinya, Vertical handoff between WWANand WLAN, in: Proceedings of the International Conference onNetworking, International Conference on Systems and InternationalConference on Mobile Communications and Learning Technologies(ICNICONSMCL ’06), Morne, Mauritius, Apr. 23–29, 2006, pp. 132–137.

[8] M.C. Necker, Towards frequency reuse 1 cellular FDM/TDM systems,in: Proceedings of the 9th ACM International Symposium onModeling Analysis and Simulation of Wireless and Mobile Systems(MSWiM ’06), ACM Press, Terromolinos, Spain, 2006, pp. 338–346.

[9] M.C. Necker, Local Interference Coordination in Cellular OFDMANetworks, VDE/ITG 12. Mobilfunktagung, May 30–31, 2007,Osnabrück, Germany.

[10] R. Bachl, P. Gunreben, S. Das, S. Tatesh, The long term evolutiontowards a new 3GPP air interface standard, Bell Labs TechnicalJournal 11 (4) (2007) 25–51.

[11] S. Yarkan, A. Maaref, K.H. Teo, H. Arslan, Impact of Mobility on theBehavior of Interference in Cellular Wireless Networks, GlobalTelecommunications Conference, 2008. IEEE GLOBECOM 2008, pp.1-5.

[12] S. Yarkan, H. Arslan, K.H. Teo, Identification of shadowed fast fadinginterference in cellular mobile radio systems, IEEE CommunicationsLetters, in press.

[13] J. Eberspaecher, C. Bettstetter, H.-J. Vhogel, GSM: Switching, Servicesand Protocols, John Wiley & Sons, Inc., New York, USA, 2001.

[14] J.-Z. Sun, A review of vertical handoff algorithms for cross-domainmobility, in: Proceedings of the International Conference onWireless Communications, Networking and Mobile Computing(WiCom 2007), Sept. 21–25, 2007, pp. 3156–3159.

[15] Bernhard Walke, Peter Seidenberg, Marc Peter Althoff, UMTS: TheFundamentals, John Wiley & Sons, Ltd., 2003.

[16] L.-X. Wang, Adaptive Fuzzy Systems and Control: Design andStability Analysis, Prentice-Hall, Inc., Upper Saddle River, NJ, USA,1994.

[17] H. Suzuki, A statistical model for urban radio propagation, IEEETransactions on Communications COM-25 (7) (1977) 673–680.

[18] G.L. Stuber, Principles of Mobile Communications, fourth ed., KluwerAcademic Publishers, 1996. ISBN: 0–7923–9732–0.

[19] B. Sklar, Rayleigh fading channels in mobile digital communicationsystems part I: Characterization, IEEE Communications Magazine 35(7) (1997) 90–100.

[20] M. Gudmundson, Analysis of handover algorithms [microcellularradio], in: 41st IEEE Vehicular Technology Conference, Gateway tothe Future Technology in Motion, St. Louis, Missouri, USA, May 19–22, 1991, pp. 537–542.

[21] UMTS, Universal Mobile Telecommunications System (UMTS);Selection procedures for the choice of radio transmissiontechnologies of the UMTS, Universal Mobile TelecommunicationsSystem (UMTS), 650 Route des Lucioles – Sophia AntipolisValbonne–France, Technical Report TR 101 112 V3.2.0 (1998–2004), Apr. 1998, UMTS 30.03 version 3.2.0.

[22] J. Dorleus, R. Holweck, Z. Ren, H. Li, H.-L. Cui, J. Medina, Modeling andsimulation of fading and pathloss in opnet for range communications,in: Proceedings of the IEEE Radio and Wireless Symposium, LongBeach, California, USA, Jan. 9–11, 2007, pp. 407–410.

[23] M. Gudmundson, Correlation model for shadow fading in mobileradio systems, Electronics Letters 27 (23) (1991) 2145–2146.

Celal Çeken received the M.Sc. and Ph.D.degrees from University of Kocaeli, Turkey in2001 and 2004, respectively. His activeresearch interests include wireless communi-cations, broadband networks, ATM networks,high-speed communication protocols, Wire-less Sensor Networks and cognitive radio.

http://www.it.kth.se/jmitola/MitolaDissertation8Integrated.pdf

http://www.it.kth.se/jmitola/MitolaDissertation8Integrated.pdf


Serhan Yarkan received his B.S. and M.S.degrees in computer science from IstanbulUniversity, Turkey, in 2001 and 2003,respectively. He is currently pursuing hisPh.D. degree in electrical engineering at theUniversity of South Florida. His researchinterests are cognitive radio, wireless propa-gation channel modeling, and cross-layeradaptation and optimization.

Hüseyin Arslan received his Ph.D. degree in1998 from Southern Methodist University,Dallas, Texas. From January 1998 to August2002 he was with the research group ofEricsson Inc., North Carolina, where he wasinvolved with several project related to 2Gand 3G wireless cellular communication sys-tems. Since August 2002 he has been with theElectrical Engineering Department of theUniversity of South Florida. He has alsoworked for Anritsu Company, Morgan Hill,California, as a visiting professor during the

summers of 2005 and 2006, and as a part-time consultant since August2005. His research interests are related to advanced signal processing

techniques at the physical layer, with cross-layer design for networkingadaptivity and QoScontrol. He is interested in many forms of wirelesstechnologies including cellular, wireless PAN/LAN/MANs, fixed wireless-access, and specialized wireless data networks like wireless sensorsnetworks and wireless telemetry. His current research interests are inUWB, OFDM based wireless technologies with emphasis on WiMAX, andcognitive and software defined radio. He has served as technical programcommittee member, and session and symposium organizer for severalIEEE conferences. He is an editorial board member for Wireless Com-munication and Mobile Computing Journal, and was technical programco-chair of IEEE Wireless and Microwave Conference 2004.

Documents

Interference aware vertical handoff decision algorithm for