On the Impact of Soft Vertical Handoff on Optimal Voice Admission Control in PCF-Based WLANs Loosely Coupled to 3G Networks

Transcription

1 356 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 8, NO. 3, MARCH 2009 On the Impat of Soft Vertial Handoff on Optimal Voie Admission Control in PCF-Based WLANs Loosely Coupled to 3G Netorks Raha Ben Ali, Student Member, IEEE, and Samuel Pierre, Senior Member, IEEE Abstrat Soft vertial handoff (VHO) and admission ontrol are usually onsidered as to independent mehanisms ensuring respetively paket-level QoS and all-level QoS for voie alls in loosely oupled 3G/WLAN netorks. In this paper, e evaluate the impat of the soft VHO on the bloking performane of the optimal voie admission ontrol in different mobility environments here the WLAN operates the Point Coordination Funtion (PCF). For this purpose, e propose an aurate analytial mobility model for the soft VHO region. Then, based on the proposed model, e derive and analyze the bloking and dropping probability expressions of the optimal voie admission ontrol algorithm in the 3G netork loosely oupled to the PCF-based WLAN. Results sho us that a resoure-effiient soft handoff (RESHO) performs signifiantly better than a statithreshold soft handoff (STSHO) partiularly in WLAN mobility environments. In fat, the 3G ne all bloking probability redution gained by using RESHO ompared to STSHO is largely inreased hen mobile station (MS) veloities have lo mean and high variability hih typially haraterizes the WLAN mobility environment. Besides, results sho us that RESHO redues all bloking and dropping probabilities. We believe that the provided model and the presented results ould help design effiient MS ontrolled soft VHO algorithms for emergent loosely oupled 3G/WLAN netorks. Index Terms Loosely oupled UMTS/WLAN, soft vertial handoff, mobility model, voie optimal CAC, bloking probability. I. INTRODUCTION WIRELESS loal area netork (WLAN) tehnology have been idely deployed in reent years and is still experiening a tremendous groth due to its flexibility, highrate aess and lo ost. On the other hand, voie over Internet protool (VoIP) has beome one of the fastest groing Internet appliations beause of its high bandidth effiieny and lo ost. In the near future, voie over WLAN (VoWLAN) hih an be vieed as the onvergene of these to tehnologies is expeted to have a dramati popularity. Hoever, voie on today s popular WLAN systems still suffers from several quality of servie (QoS) issues due to its real-time onstraints hih make it an ative researh area. In fat, a lot of reent Manusript reeived Deember, 2007; revised February 22, 2008; aepted April, The assoiate editor oordinating the revie of this paper and approving it for publiation as D. I. Kim. The authors are ith the Mobile Computing and Netorking Researh Laboratory (LARIM), Department of Computer Engineering, Eole Polytehnique de Montreal, P.O. Box 6079, Station Centre-ville, Montreal, Quebe, H3C 3A7, Canada ( {raha.benali, samuel.pierre}@polymtl.a). This ork as supported in part by the NSERC-Erisson Industrial Chair. Digital Objet Identifier 0.09/TWC /09$ IEEE researh [] [3] has been arried out in order to resolve paket-level QoS issues of VoWLAN by providing effiient admission ontrol mehanisms and fast handoff proedures. Partiularly, the primary operation mode of WLAN alled the Distributed Coordination Funtion (DCF) makes the voie traffi; hih is very sensitive to delay and jitter; interfere ith other types of traffi resulting in a violation of its delay bound or in a large delay variane [4]. In our ork, e onsider a voie-entri WLAN provider implementing the optional Point Coordination Funtion (PCF) hih, as opposed to the very popular DCF, provides strit voie QoS by designing a simple admission ontrol based on [5]. te that PCF also provides better seurity ompared to DCF sine denial of servie attaks by outside users an be better ontrolled by a PCF based WLAN. Dealing ith VoWLAN QoS issues related to user mobility, several WLAN smart sanning and fast handoff algorithms minimizing the servie disruption during handovers have been presented in [3]. For instane, authors in [6] used a neighbor graph ahe mehanism to redue sanning lateny don to a level that meets the voie QoS delay onstraints during WLAN handoff. Hoever, even though reent proposed solutions an provide suitable fast handoff for VoWLAN users as stated in [3], the small overage of WLAN netorks is a major limitation for providing the required ubiquity to mobile voie over IP users. Hopefully this limitation an be bypassed by integrating the WLAN ith a large overage overlaying 3G ellular netork. In fat, ith the inreasing popularity of multi-mode mobile stations (MS) equipped ith both ellular and WLAN interfaes, suh as reent PDAs and smartphones, voie users ill profit from both the lo ost of high-speed WLAN aess in hotspot areas and the ubiquitous overage of 3G ellular netorks. For this purpose, e assume that 3G ellular netorks and WLAN netorks are integrated in order to provide voie all ontinuity. Furthermore, onsidering the benefits of implementation flexibility and independent deployment of 3G and WLAN netorks, e assume a loose oupling sheme that integrates these heterogeneous ireless netorks using an external IP-based netork. Hoever, it is knon that an IP-based loosely oupled integration indues high handoff latenies espeially for handoffs involving IP address and loation updates suh as in vertial handoffs using either SCTP or Mobile IP protools. Therefore, a soft handoff mehanism is needed in order to ensure the paket-level QoS of voie alls during the disruption period Authorized liensed use limited to: ECOLE POLYTECHNIQUE DE MONTREAL. Donloaded on vember 6, 2009 at 0:07 from IEEE Xplore. Restritions apply.

2 BEN ALI and PIERRE: ON THE IMPACT OF SOFT VERTICAL HANDOFF ON OPTIMAL VOICE ADMISSION CONTROL 357 of the handoff proess. In general, the soft handoff provides simultaneous onnetivity to more than one base station (BS) / aess point (AP) in order to seamlessly transfer the all beteen them. Considering the introdution of the soft handoff mehanism in these integrated netorks, standard optimal admission ontrol mehanisms have to be adapted in order to provide effiient ireless bandidth utilization hile maintaining the required paket-level and all-level QoS for the voie servie. In the literature, effiient all admission ontrol (CAC) in loosely oupled 3G/WLAN netorks is analyzed in [7] and soft handoff is introdued in [8], [9]. Hoever, to the best of our knoledge no researh ork as onduted to evaluate the impat of the soft handoff on the voie all bloking performane of optimal admission ontrol. The rest of this paper is organized as follos. In Setion II, e propose our advaned analytial model taking into aount both mobility and soft handoff harateristis in 3G/WLAN integrated netorks. Then in Setion III, e present the resoure-effiient soft handoff algorithm for ensuring the paket-level QoS and e analyze its related parameters. In Setion IV, e provide our adaptation of the optimal admission ontrol algorithm to a 3G/WLAN integrated netork in order to ensure the all-level QoS by maximizing the number of aepted ne alls hile maintaining a hard onstraint on a very lo all dropping rate. In Setion V, e validate the analytial model using omputer simulation and e present the bloking performane results of soft vertial handoff algorithms under different mobility shemes. Finally, Setion VI onludes the paper. II. SYSTEM MODEL In our system model, e onsider a loosely oupled 3G/WLAN system in hih several WLAN ells overlay eah 3G ell. Within this system, loation dependant mobility and traffi distribution are adequately modeled, voie apaity analysis is briefly disussed, bloking probabilities are estimated and soft handoff parameters are analyzed. Sine many symbols are used, Table I summarizes the important ones. A. Multi-region mobility model In general, mobility an be modeled using either analytial or simulation models. Simulation is able to provide realisti mobility models sine it uses a large amount of detail to build the trajetory of the mobile user by periodially traking its loation in small time steps. Hoever suh models are generally intratable hen onsidering large netork overage topologies ith large number of users. Therefore, analytial mobility models based on random variables and stohasti proesses used to model ell residene time are more suitable in this ase. In our ork e develop a ne analytial mobility model in 3G/WLAN overage to study the impat of soft handoff on resoure utilization. We propose a loation dependant mobility model ithin a 3G ell hih is omposed of three types of regions based on the number of overlaying overages (single or double) and the number of oupied resoures (uni-asting or bi-asting) illustrated in Figure 2 : several WLAN ore regions, their orrespondent WLAN border regions designed for soft handoff and finally one single overage 3G region. These regions are defined as folloing : A WLAN ore region, noted RW, is haraterized by the double overage of both WLAN and 3G netorks but only resoures of one overage are alloated at a time to a multi-mode MS. We assume a netork seletion strategy that gives resoure alloation preferene to the lo-ost WLAN rather than the generally highly loaded 3G. Due to the lo mobility in WLAN environments (indoor environments in general) most users stay ithin this region for a short time, hile fe users stay for a very long time. It as shon in [0] that heavytailed Pareto distributions provide the best fit to aptured realisti WLAN residene times. Besides, it as shon that usual exponential distributions are no more valid for modeling WLAN residene time. Therefore in order to apture the Pareto effet ithout loosing the Markov property needed in queuing analysis, e use a hyperexponential distribution ith a mean rate η and a mobility variability parameter α to model the residene time in this RW region illustrated in Figure. Thus, the probability density funtion (pdf) of this residene time is expressed as folloing: f RW (t) = α +α α η e α η t + +α η α e η α t () A soft handoff WLAN border region, noted RWC (resp. RCW), is haraterized by the double overage of both WLAN and 3G netorks and in hih dupliate resoures are alloated on both netorks at the same time to the multi-mode MS moving out of (resp. into) the WLAN ore region. Partiularly, it defines the soft vertial handoff (VHO) region in hih the multi-mode MS maintains to simultaneous onnetions by bi-asting the ongoing voie all on both WLAN AP and 3G BS. This bi-asting in this WLAN border region provides the required paketlevel QoS for voie alls during the handoff disruption period. The all-level QoS an be signifiantly improved by reovering all droppings indued by WLAN horizontal handoff (HHO) failures using the soft upard VHO, i.e. the soft handoff that overflos the all from WLAN to 3G. Depending on the mobility management arhiteture implemented by 3G and WLAN operators, the soft handoff may involve Mobile IP bi-asting or SCTP multi-homing mehanisms. Sine e are interested in the handoff part that involves the dupliate resoures defining the soft handoff, the handoff delay represents the delay of the ireless signalling proess immediately after the hannel reservation on the next AP/BS added to the delay of the IP address onfiguration and the loation update on the orrespondant node (CN). We model this handoff delay by the sum of a onstant added to a non negative random variable that depends on IP address onfiguration, CN loation and Internet lateny. Sine a onstant distribution does not keep the Markovian property [] required for our further queuing analysis, e approximate it using the Erlang distribution ith a large number of r stages. Therefore, e model the total residene time in a soft VHO region using the sum of r Authorized liensed use limited to: ECOLE POLYTECHNIQUE DE MONTREAL. Donloaded on vember 6, 2009 at 0:07 from IEEE Xplore. Restritions apply.

3 358 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 8, NO. 3, MARCH 2009 TABLE I SUMMARY OF IMPORTANT SYMBOLS Symbol Definition Traffi parameters k Index representing a WLAN ell if k= and a 3G ell if k= λ k Mean ne all arrival rate to ell k μ Mean all holding time μk ne (μk ho ) Mean hannel holding time for a ne (handoff) all in ell k η (η ) WLAN (3G ell) ore region residene time mean parameter α (α ) WLAN (3G ell) ore region residene time variability parameter σ k (K k ) Mean of the variable part (onstant part) of the residene time in the ell k soft handoff border region η rt WLAN residene time mean parameter WRT (CRT) WLAN (3G ell) residene time P Probability of a user moving out of urrent 3G ell to overlaying WLAN ell ν k (ν k ) Horizontal (Vertial) handoff rate in (to) k ell CAC parameters g k (g k ) Limited frational guard hannel LFGC for ne (vertial handoff) alls in (to) k ell β k (.,.) (β k (.,.)) Probability aeptane funtion for ne alls (vertial handoff alls) in ell k k Voie apaity of ell k Pd,max k (Pk d ) (Maximum alloed) Call dropping probability for alls in the separated k netork Pd,max (P d ) (Maximum alloed) Call dropping probability for alls in the integrated netork ε Resolution of the estimated handoff rates and steady-state probabilities QoS metris Pb k Probability of bloking ne alls requesting admission to the k netork Pf k (Pfv k ) Probability of failing the horizontal (vertial) handoff in (to) the k netork P ne,k h (P ho,k h ) Probability that a ne (handoff) all hand off from the ell k Ph k Probability that a all (either ne all or handoff all) hand off from the ell k Reeived signal strength parameters RSS 0 Reeived signal strength measured at time 0s RSS in Reeived signal strength threshold that triggers the move into the WLAN ell after moving out from it RSS add Reeived signal strength threshold that triggers the stati-threshold soft handoff to neighboring WLAN ell or 3G overlaying ell RSS out Reeived signal strength threshold that triggers the move out from the WLAN ell n Slope index of the reeived signal strength model d(.) Distane beteen the MS and the AP funtion of time v i (α i ) Veloity (Angle beteen WLAN ell diameter and movement diretion) of the MS in the WLAN during time interval i i add Ideal time interval in hih the resoure-effiient soft handoff algorithm trigger itself θ Time interval duration for averaging the RSS and estimating the RSS slope under urrent veloity Other parameters m Number of WLAN ells overlaying one single 3G ell r Number of stages of the Erlang distribution used to approximate a onstant distribution P ov Probability that no other AP an be disovered before reahing the out of range RSS on the urrent AP. Erlang distributions ith rate K for soft upard VHO (resp. K for soft donard VHO) and an exponential distribution ith a mean rate σ (resp. σ ). r /K is the onstant part of the handoff delay and /σ is the mean parameter of its variable part. In a further setion, e ill distinguish beteen a theoretial resoure-effiient and a pratial less resoure-effiient soft handoff regions. The first one is optimally dimensioned for the strit lateny of the handoff proedure ith a parameter /σ.the seond one is dimensioned for a higher value /σ.the RWC residene time pdf is expressed as folloing: f RWC (t) =(t K )e σ (t K ) (2) A 3G ell ore region, noted RC, is haraterized by a single 3G overage. The residene time in this region is modeled, as in RW, using a hyper-exponential distribution ith a mean rate η and a mobility variability parameter α. We use this distribution to overome the no more valid exponentially distributed CRT assumption and to apture the CRT high variability in next generation ellular systems as notied in [2]. te that the RC residene time pdf is similar to the RW one. B. Voie apaity model Depending on the aess tehnology used, there are several analytial methods in the literature that bound the maximal R Fig.. Fig. 2. P - Exp Exp R Erlg Exp -Pf* R 3G/WLAN multi-region ell residene model P - 3G ell BS (de-b) -P*f WLAN ell BS (AP) Bi-mode MS Exp Exp P - P - Double overage WLAN border region (Soft VHO region) 3G ellular single overage region Double overage WLAN ore region Multi-region mobility ithin a 3G ell overlaying WLAN number of voie alls, i.e. the number of hannels, that a ell either 3G or WLAN an aommodate ith the required paket-level QoS. Authorized liensed use limited to: ECOLE POLYTECHNIQUE DE MONTREAL. Donloaded on vember 6, 2009 at 0:07 from IEEE Xplore. Restritions apply.

4 BEN ALI and PIERRE: ON THE IMPACT OF SOFT VERTICAL HANDOFF ON OPTIMAL VOICE ADMISSION CONTROL 359 ) 3G ell voie apaity model: We assume that the 3G netork is based on the popular UMTS standard hih uses the ide band ode-division multiple aess (W-CDMA) ellular system. Assuming different types of alls, the W- CDMA soft apaity is modeled using the very popular load expression in [3]. Using the expression in [3] that bounds the interferene, e an estimate the 3G ell voie apaity in term of maximal number of voie alls hih is a funtion of the atual number of alls admitted in eah other lass. In order to simplify our analysis e assume that voie alls are preemptively prioritized over other types of alls. Therefore, the voie apaity region an be easily evaluated independently from other types of alls. 2) WLAN ell voie apaity model: Even though the DCFbased QoS differentiation mehanism introdued in the reent IEEE 802.e standard [4] an provide transmission priority to voie traffi it an not guarantee its delay bound sine it annot preempt ongoing transmissions of other lo-priority traffi using the distributed operation mode. Hoever, despite the quite loer resoure effiieny of the PCF operation mode hen no voie traffi is present due to the unneessary MS polling, this entralized aess method an provide strit delay guarantees for voie alls, as opposed to the DCF. The algorithm proposed in [5] provides an analytial expression giving the number of nodes that a PCF-based AP an aommodate in order to satisfy a given delay onstraint at eah of the nodes. In the paper, it as shon that an AP onfigured ith Contention Free Period repetition interval CFPri of 25 ms an ensure the strit QoS of 8 voie alls using the G.723. ode. In our ork, it is possible to limit the maximal number of voie alls to 5 so that a minimal Contention Period (CP) ill be guaranteed for data traffi using the DCF as an example of a restrited aess resoure sharing sheme [5]. In fat, if the MS has no ative voie sessions it ill be removed from the polling list and a longer CP ill be reserved for data traffi. C. The admission ontrol poliy Usually hen e an derive the voie apaity for the 3G ell [3] (resp. for the PCF-based WLAN ell [5]) in terms of maximal number of voie alls supported ithout violating the paket-level QoS, a simple admission ontrol algorithm an be designed. Hoever, sine it is ommonly onsidered that the dropping of an ongoing all has muh more negative impat on users pereption than the bloking of a nely initiated all, e have to prioritize handoff alls over ne ones in aessing the ireless bandidth resoures. An aeptane probability funtion β k (i, j) (resp. β k (i, j)) similar to the one used in [6] is designed for admitting nely initiated voie alls (resp. VHO alls) in the loosely oupled 3G/WLAN netork. Where k is either for WLAN or for ellular 3G. We assume that sine in a loosely oupled 3G/WLAN netork the 3G and the WLAN bandidth resoures belong usually to to distint administrative domains, VHO alls hih are ontrolled and triggered at the MS level are proessed as ne alls. And therefore the priority to aess the voie apaity for VHO alls and for ne alls are equals (β k (i, j)=β k(i, j)). This aeptane funtion is tightly related to the the limited frational guard hannel (LFGC) [6] hih defines the noninteger number of hannels g k exlusively reserved for HHO alls. Up to k g k hannels an be oupied by ne alls andupto k g k an be oupied by VHO alls. te that g k = g k sine β k (i, j) =β k(i, j). In fat, β k (i, j) is equal to ifi + j < k g k, g k g k if i + j = k g k and 0 if i+ j > k g k. g k have to be hosen as the optimal solution of the CAC problem presented further. D. Expression of hannel holding times (CHT) Let λ (resp. λ ) denote the ne all arrival rate to a WLAN ell (resp. to a 3G ell) and let /μ denote the mean voie all holding time, i.e. its mean duration. It is ommonly assumed that ne alls arrive folloing a Poisson proess and that the all holding time is exponentially distributed. For sake of notation simpliity, e note WRT (resp. CRT) the random variable representing the WLAN ell residene time (resp. the 3G ell residene time). Besides, e note RW, RWC and RC the random variables representing the residene times in their respetive regions. Sine WRT is the sum of to non-negative independent ontinuous random variables RW and RWC, i.e. WRT = RW + RWC, its pdf is the onvolution of the RW s pdf and the RW s pdf : f WRT (t)=(f RW f RWC )(t)= t 0 f RW (x) f RWC (t x)dx (3) The Laplae transform of WRT s pdf is easier to develop sine it is the produt of Laplae transforms of RW s pdf and RWC s pdf. Thus applying the Laplae transform to Equations and 2 e obtain : F RW (s) = (α2 α + )s + α η (4) (α s + η )(s + α η ) K σ F RWC (s) = ( K + s )r ( σ + s ) (5) F WRT (s) = F RW (s)f RWC (s) = (α2 α + )s + α η (α s + η )(s + α η ) ( K σ K + s )r σ + s (6) Assuming a hannelized PCF-based WLAN bandidth for voie alls as previously desribed and a onservative maximal number of voie alls analytially predetermined and remaining onstant during the evaluation period, e use Theorem 2 in [7] to express the Laplae transform of CHT s pdf of nely initiated alls and the Laplae transform of CHT s pdf of handoff alls in the WLAN ell as the folloing : F ne,wlan CHT (s) = μ s + μ + η WRTs2 ( FWRT (s + μ)) (7) s + μ F ho,wlan CHT (s) = μ s + μ + s s + μ F WRT(s + μ) (8) Assuming M WRT (.) is the moment generating funtion of WRT s pdf, e an express η WRT, i.e. the WLAN ell residene rate in the previous Equations, as the folloing : η WRT = E(WRT) = M WRT (0) = F WRT (0) σ η K = (9) K σ + rη σ + K η In fat, it is knon that dm WRT(t) dt t=0 = df WRT( t) dt t=0 = F WRT for a ontinuous non-negative random variable WRT. Using (0) Authorized liensed use limited to: ECOLE POLYTECHNIQUE DE MONTREAL. Donloaded on vember 6, 2009 at 0:07 from IEEE Xplore. Restritions apply.

5 360 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 8, NO. 3, MARCH 2009 Equations 7 and 8, e formulate the mean CHT, i.e. the average WLAN oupany time, of both types of alls: μ ne μ ho = (F ne,wlan CHT ) (0) = μ σ η K K σ +rη σ +K η ( F WRT (μ)) μ 2 (0) = (F ho,wlan CHT ) (0) = F WRT(μ) () μ Similar Equations an be easily formulated for a 3G ell. k k-gk i (number of ne alls at layer k) (i+,j) k(i,j) k (i+) nk k k+ k k+ k(i,j) k k(i,j) k (i,j-) (i,j) (i,j+) j hk k(i,j) k i nk (i-,j) (j+) hk j (number of horizontal & vertial handoff alls at layer k) E. Expression of soft handoff probabilities The soft handoff probability is defined as the probability that a all needs at least one more soft handoff during its remaining lifetime. We define the soft handoff as a all s request of a ne hannel from the next AP/BS a period of time, alled bi-asting period, before the all frees the oupied hannel in the urrent AP/BS. Contrastingly, a hard handoff is defined as a all s request of a ne hannel from the next AP/BS immediately after the all frees the oupied hannel in the urrent AP/BS. In our model, the soft handoff is triggered after the elapsed residene time in the WLAN ore region RW, by ontrast to the hard handoff that is triggered later i.e. after the elapsed residene time in the hole region WRT=RW+RWC. Therefore, it is possible to reformulate the handoff probability expressions in Equation 2 and 6 in [] to our problem. By replaing F WRT by F RW and η WRT by η e have the folloing expressions for soft handoff probabilities of nely initiated alls and handoff alls in the WLAN ell : P ne,wlan h = η μ ( F RW (μ)) (2) P ho,wlan h = F RW (μ) (3) te that the expressions developed for the CHT in Equation and for the handoff probability in Equation 3 take into aount the soft handoff region that e introdued in our model. Hoever, the system performane in terms of bloking probabilities still have to be analyzed. F. Estimating bloking probabilities We model the voie apaity oupation in eah WLAN (resp. 3G) ell using a bi-dimensional Markov hain illustrated in Figure 3 although the servie time, i.e. the CHT, is not exponentially distributed sine it is stated in [] that the sum of hyper-exponential and Erlang distributions keeps the Markovian property. We an easily derive the Markov hain balane equations from Figure 3, then using the Gauss-Seidel approah on these equations e iteratively estimate state probabilities until the onvergene to their respetive steadystate probabilities. We use this approah sine no losed-form expressions exist for these steady-state probabilities. Folloing the admission poliy desribed in Subsetion II-C and using the estimated steady-state probabilities, e express the various bloking probabilities : Fig. 3. k-g k Bi-dimensional Markov hain for voie oupation model in ell k ) The ne all bloking probability Pb = i+ j=0 β (i + j)p i, j,k. 2) The HHO failure probability : Pf = ( P ov ) i+ j= P i, j, + P ov here P ov is the WLAN out of overage probability. 3) The VHO failure probability : Pfv = P b sine upard VHO alls are admitted as nely initiated alls in the overlaying 3G ell due to the assumed loose oupling beteen 3G and WLAN. 4) The all dropping probability in the separate netork (See Appendix): Pd = P h P f P h ( P f ). Similar equations an be formulated for the 3G ell. Considering the all ontinuity using VHO in the ase of WLAN HHO failures, e develop more suitable expression for this dropping probability under the loosely oupled 3G/WLAN sheme : P d = P d P d ( +( P h k ) P b Pf ) (4) We provide our proof of this WLAN/3G dropping probability expression in the appendix. G. Estimating handoff rates In general, e use handoff rates as already given input parameters to the Gauss-Seidel algorithm that estimates the steady-state probabilities. Hoever, in order to have more auray in our performane analysis, e estimate these handoff rates by applying the flo onservation fixed point equations that an be easily dedued from Figure 4. Starting ith null handoff rates e iteratively evaluate these fixed-point equations and for eah iteration e estimate the bloking probabilities using the Gauss-Seidel approah. We stop iterating on the fixed-point equations hen all handoff rates onverge to their respetive steady values ith a given resolution ε. From Figure 4, e have the folloing fixed-point equations respetively for WLAN HHO rate ν, donard VHO rate Authorized liensed use limited to: ECOLE POLYTECHNIQUE DE MONTREAL. Donloaded on vember 6, 2009 at 0:07 from IEEE Xplore. Restritions apply.

6 BEN ALI and PIERRE: ON THE IMPACT OF SOFT VERTICAL HANDOFF ON OPTIMAL VOICE ADMISSION CONTROL 36 (-Pb ) Pf (-Pb ) Pf (-Pf ) C (-P - +Pfv P - ) (-Pb ) Pb Pb Pb Pf (-Pfv ) Pb Pb Pf (-Pfv ) 3G ell... (-Pb ) Pb (-Pf ) (-P - +Pfv P - ) (-Pb ) Pb Pf (-Pfv ) Pf (-Pfv ) Pb Pb Pb Pb 3G ell... (-Pb ) Pb (-Pf ) (-P - +Pfv P - ) (ubiquitous 3G overage) (-Pb ) (-Pfv ) m P - (-Pf ) WLAN Pf Pfv Pf Pfv Pf Pfv Pf Pfv Ne Calls (-Pb ) (-Pfv (-Pb ) ) (-Pb ) (-Pfv ) (-Pfv ) m P - m P - m P - HHO Calls UVHO Calls Pf Pf Pf Pf DVHO Calls (-Pf ) WLAN (-Pf ) (m WLANs (-Pf ) WLAN (-Pf ) WLAN (-Pf ) (m WLANs Bloked Calls overlaying one 3G ell) overlaying one 3G ell) Dropped Calls Fig. 4. Flo onservation in 3G and WLAN ells for voie all traffi ν, 3G HHO rate ν and upard VHO rate ν : ν = λ ( Pb )Pne,lan h + ν ( Pf )Pho,lan h + ν ( P fv )Pho,lan h (5) ν = λ ( Pb )Pne,3g h + ν ( P + P Pfv ) ( Pf )Pho,3g h + ν ( Pfv )Pho,lan h (6) ν = (/m)p ν (7) ν = mp f ν (8) START L 0 U C g 0 Iter 0 U-L> AND Iter<Imax Compute Pb,Pb, Pd -,Pd Pd - <Pd - max Iter= Pb0 Pb0 Pb Pb Iter= L 0 U C g 0 Iter 0 U-L> AND Iter<Imax Compute Pb,Pb, Pd -,Pd Pd <Pd max H. Voie admission ontrol as an optimization problem We kno from Subsetion II-F that the bloking probabilities is funtion of aeptane probability funtions β k (.). Therefore, e need to find optimal admission parameters that give optimal QoS performane for the 3G/WLAN integrated netork. For this purpose, e redefine the optimal CAC problem for a 3G/WLAN integrated netork here ne alls and handoff alls an vertially handoff beteen the to netorks. We assume that due to the high aess ost and the high load of 3G ells, the most suitable netork seletion strategy is a WLAN-first strategy similar to the one analyzed in [7]. This strategy admits nely initiated alls and keeps handoff alls of multi-mode MS in the WLAN netork before overfloing them to the overlaying 3G netork. We notied that for the ommon voie servie most of the mobile operators try to distinguish from others by providing the loest dropped alls. Therefore, the WLAN CAC has to minimize the ne all bloking probability on WLAN hile maintaining a hard onstraint on a lo all dropping probability hih is redefined in Equation 4 to take into aount the upard VHO to a 3G ell in ase of WLAN HHO failures (inluding out of WLAN overage P ov ). Therefore, the algorithm that estimates the optimal admission parameters is the one that resolves a modified version of the ell knon MINBLOCK optimization problem [6] adapted to the overlaying WLAN/3G arhiteture. The CAC optimization problem is formulated as folloing : min g P b (g ) subj. to : Pd (g )=Pd P d ( +( Ph ) P b Pf ) Pd,max (9) In Figure 5 e developed an iterative searh algorithm for the optimal admission parameters based on a bi-setion searh Fig. 5. U g g L+U) Iter Iter L g g L+U)/2 Pb0 -Pb < AND Pb0 -Pb < L g g L+U) MinB algorithm for 3G/WLAN integrated netorks Iter END Iter U g g L+U)/2 approah. Partiularly, the algorithm searhes for the optimal non integer limited frational guard hannels (LFGC) g for WLAN ells and g for for 3G ells. III. THE SOFT HANDOFF ALGORITHM Sine the soft handoff onsumes dupliate resoures to seamlessly handoff the all, in order to optimize resoure utilization espeially on the high-ost and highly loaded 3G ells, the soft handoff duration have to be minimized to its strit minimum. Standard implementations of soft handoff algorithms are simple and straightforard. Hoever, a resoureeffiient one is usually omplex and CPU-onsuming sine it is hallenging to ompute aurate predition estimates of the right time to trigger the soft handoff in a highly varying ireless mobile environment. A. The standard stati-threshold soft handoff (STSHO) While maintaining a voie all over the W-CDMA interfae, the multi-mode MS monitors the reeived signal strength (RSS) on neighboring ells (3G and WLAN) and hen it reahes a pre-defined RSS threshold (RSS in for the WLAN), it Authorized liensed use limited to: ECOLE POLYTECHNIQUE DE MONTREAL. Donloaded on vember 6, 2009 at 0:07 from IEEE Xplore. Restritions apply.

7 362 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 8, NO. 3, MARCH 2009 triggers the soft handoff to the next ell hih ill be preferably a WLAN ell if available. Hoever, hile maintaining the all over the IEEE 802. interfae, it is more onvenient to monitor the RSS on the urrent WLAN ell rather than neighboring WLAN ell, sine sithing hannels to measure signals on neighboring WLAN ells during voie ativity may degrade the voie QoS. Usually hen the urrent RSS dereases belo a ertain level noted RSS add, a soft handoff is triggered preferably to a neighboring WLAN ell, otherise to an overlaying 3G ell. As in [8], after using an RSS averaging mehanism to smooth the shadoing flutuations, the RSS level an be modeled as follos : v 3 d 0 3 v 2 v vt d d 2 2 RSS(iθ) =R 0 0nlogd(iθ)dBm (20) d 2 (iθ) =v 2 i (iθ)2 + d0 2 2d 0v i iθ os(α) (2) We note n the RSS slope index (typially 4 for miro-ells in the ity). We assume the veloity model presented in Figure 6, here v is the MS veloity intensity and α theangleof MS movement diretion. We also assume time intervals ith length θ short enough to have both v and α onstants during eah instane i of these time intervals. B. The resoure-effiient soft handoff (RESHO) In order to minimize the use of highly expensive and highly loaded 3G ells, e propose using a resoure-effiient soft handoff (RESHO) algorithm for voie multi-mode MS that estimates and adjusts the optimal RSS threshold of the VHO trigger depending on MS veloity and handoff signaling lateny. We assume that the MS veloity and diretion remain onstant just before entering and inside the soft handoff region. This assumption is based on the fat that the time spent inside a RESHO region overoming the handoff lateny is short enough regarding the high WLAN residene time. te that some more or less resoure-effiient RESHO algorithms are presented in the literature [8], [9] but not evaluated as in our ork. A possible design of a RESHO algorithm is the folloing. At the start of eah time interval i, theresho algorithm estimates the RSS variation slope using the Equation 2 and the previously measured RSS values. Then it uses this estimation to predit the right time interval, noted i add, for hih a soft handoff have to be triggered. i add an be approximated by the latest interval during hih the soft handoff have to be triggered in order to establish the end-toend voie all on the next AP/BS before the RSS falls belo the out of range RSS out hih is imminent to free the all on the previous AP. This handoff lateny of establishing the same end-to-end voie all on a different AP/BS an be signaled from a history table by the ireless aess gateay, i.e. the equipment that loosely interonnets the to administrative domains of the respetive ore netorks of 3G and WLAN. The detailed design of this algorithm is out of the sope of this paper. C. Residene times for RESHO vs. STSHO In order to quantify the performane gain ahieved by using a RESHO mehanism as e desribed it, e express the mean residene times in the ore WLAN region t RW and the hole Fig. 6. RSS add RSS in RSS out WLAN RSS thresholds for soft vertial handoff triggering WLAN region t WRT for the RESHO vs. STSHO algorithms. Reall that the STSHO is based on a pre-planned RSS add = ρrss out onfigured for all MSs. We assume that eah MS enters the WLAN ell at time t = 0 and leaves it at time t = t WRT ith a onstant veloity v and a onstant diretion α. We also assume that a hysteresis loop mehanism is implemented in order to prevent handoff ping-pong effets. Therefore e distinguish different RSS thresholds for moving in (RSS in )and out of (RSS out ) the WLAN as illustrated in Figure 6. RSS(0) = R 0 0nlogd 0 = RSS in (22) RSS(t RW ) = R 0 0nlogd(t RW )=RSS add (23) RSS(t WRT ) = R 0 0nlogd(t WRT )=RSS out (24) v η(x) = (25) d 0 osα + d0 2 os2 α d0 2 +(e R 0 x 0n ) 2 η WRT = = η(rss out ) (26) t WRT Then, for STSHO and RESHO e have : η = = η(rss add ) (27) t RW σ = t sig = t WRT (r/k ) t RW (28) η = t WRT (r/k ) tsig (29) σ = tsig (30) For different veloities and a given onfigured RSS threshold RSS add, it is possible to use the previous equations to ompute η and σ for the STSHO algorithm and η for the RESHO algorithm. /σ is the mean handoff signaling lateny hih is assumed independent from the MS veloity. IV. PERFORMANCE RESULTS In this Setion e present the performane results provided by the proposed models and algorithms. Table II presents the values used for the input parameters. Authorized liensed use limited to: ECOLE POLYTECHNIQUE DE MONTREAL. Donloaded on vember 6, 2009 at 0:07 from IEEE Xplore. Restritions apply.

8 BEN ALI and PIERRE: ON THE IMPACT OF SOFT VERTICAL HANDOFF ON OPTIMAL VOICE ADMISSION CONTROL 363 TABLE II SYSTEM INPUT PARAMETERS 0 0 Parameter Value Parameter Value n 3.3 Pd,max (P d,max ) % RSS out -85 dbm 3 RSS in -80 dbm 5 RSS add -60 dbm (μ) 4mn R dbm K (K ) 60 ms d 0 95 m P 5% K 60 mse (η ) 2mn m 7 (σ ) 600 ms P ov 20% ε P b (analyti) P d (analyti) P b (analyti) P (analyti) d P (simulated) b P (simulated) d P b (simulated) P d (simulated) A. Analytial model validation In order to validate our bloking probability analysis, e have developed our on disrete event simulator using C++. We assumed a 3G ubiquitous overage area divided into several hexagonal ells, eah ell has six neighbors ells. In order to over as many ells as possible and eliminate the edge effet not present in a ubiquitous overage, e used the ell knon rap around model hih inludes in our ase 49 ells. By rapping around the 49 ells, the handoff traffi hih flos out of the overage ill flo into the area again. This prevents from losing alls. We assumed a WLAN area ith limited overage overlaying eah 3G ell. Exatly m = 7 WLAN ells overlay one single 3G ell ith a planned out of overage probability P ov. In eah ell (either 3G or WLAN), ne all arrivals are generated aording to the Poisson distribution ith their respetive mean rates. The handoff all arrivals are triggered in the simulator, rather than generated, hen their respetive generated CHT durations elapse. Our presented simulation results are statistial values that are averaged over the required number of runs to ahieve a 95% onfidene level of less than 0.00 of onfidene interval idth. The purpose of the simulation is tofold. First, sine no appropriate analytial model haraterizing the handoff traffi exists in the literature and therefore simulation is the only ay to estimate it aurately. Seond, even though simulations onsume exessive CPU yles, it is a poerful ay to hek if all the bloking and dropping probabilities estimated using the analytial model are still valid in the real simulated system. From Figure 7 e notie that the 3G ell ne all bloking and ongoing all dropping probabilities estimated using our proposed analytial model are almost equal to the ones estimated using simulation. Similarly, from Figure 7 e observe that the analysis results agrees ith the simulation results for the WLAN ell. Furthermore, from these Figures e also observe that the minb algorithm adjusts the optimal 3G ell LFGC g (respetively the optimal WLAN ell LFGC g ) depending on the ne all arrival rate suh that it minimizes the ne all bloking rate hile maintaining a hard QoS onstraint on a all dropping rate not exeeding 0 2.We notie that due to the loer handoff traffi intensity in WLAN ells the LFGC starts to inrease only at a higher ne all arrival rate of 2.75 alls/mn for WLAN vs..75 alls/mn for 3G λ or λ (all/mn) Fig. 7. Validation of the proposed analytial model for 3G and WLAN B. Resoure-effiient soft handoff performane From results presented in Figures 8, 0 and 9 e obviously notie that under all mobility shemes in the WLAN area (MS veloity ith different mean from 0.5 m/s to 4 m/s and different variability from to 6) the RESHO algorithm outperforms the STSHO algorithm in terms of optimal CAC performane metris. This is due to the fat that in average, dupliate apaity resoures for the soft handoff are oupied for shorter time hen using the RESHO algorithm. te that the all dropping probability of alls initiated in the 3G netork is not presented on the results sine it is onstantly equal to the QoS threshold (P d,max = 0 2 ) that is due to the operation of the 3G netork in a high load region. Surprisingly in Figure 0 inreasing the mean veloity in the WLAN area inreases the bloking probability of ne alls on WLAN ells hih seems ounter-intuitive sine loer hannel oupation duration leaves more available apaity for ne alls. Hoever, this fat an be explained by the inreased WLAN horizontal soft handoff traffi hih is re-injeted in the WLAN ells ith more dupliate resoures oupied and thus inreasing the WLAN load more than a higher veloity dereases it. Furthermore, e notie that inreasing the veloity variability α in the WLAN area dereases the ne all bloking probability on WLAN netork. In fat, sine a larger veloity variability α indiates that more users staying ithin the WLAN for shorter time (fast users) and less users staying ithin the WLAN for longer time (slo users), a large fration of voie alls initiated in the WLAN area an be arried by the less oupied WLAN hih onsequently relieves the 3G ell from the overfloing traffi. More interestingly, it is observed that the RESHO algorithm provides muh higher performane gain than the standard STSHO algorithm for loer MS veloities. In fat, as notied in Figures 8 and 9 the redution ratios for both 3G ne all bloking probability and WLAN initiated all dropping probability are signifiantly inreased hen the MS veloity is lo. This is due to the fat that hen the MS veloity dereases, the deviation of the residene time in the stati STSHO soft handoff region from the residene time in the Authorized liensed use limited to: ECOLE POLYTECHNIQUE DE MONTREAL. Donloaded on vember 6, 2009 at 0:07 from IEEE Xplore. Restritions apply.

9 364 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 8, NO. 3, MARCH G ne all bloking probability redution ratio for mobility variability α = 6 x 0 3 P b (Standard Soft Handoff, Veloity Variability in WLAN α =) 3G ne all bloking probability redution ratio for mobility variability α =2 P b (Optimal Soft Handoff, Veloity Variability in WLAN α =) 25 3G ne all bloking probability redution ratio for mobility variability α =6 5 P b (Standard Soft Handoff, Veloity Variability in WLAN α =2) P (Optimal Soft Handoff, Veloity Variability in WLAN α =2) b RESHO P b redution ratio over STSHO Probability of bloking ne alls in WLAN ell P b (Standard Soft Handoff, Veloity Variability in WLAN α =6) P b (Optimal Soft Handoff, Veloity Variability in WLAN α =6) MS mean veloity (v) in the WLAN area (m/s) MS mean veloity in WLAN area (m/s) Fig. 0. Bloking probability of ne alls in WLAN for STSHO vs. RESHO Fig G ne all bloking probability redution ratio WLAN initiated all dropping probability redution ratio for mobility variability α = WLAN initiated all dropping probability redution ratio for mobility variability α = 2 WLAN initiated all dropping probability redution ratio for mobility variability α = 6 diffiult to implement. In fat, hen the MS mean veloity remains lo, more samples an be used to predit a more aurate RESHO residene time. RESHO P d redution ratio over STSHO (%) Fig MS mean veloity (v) in the WLAN area (m/s) WLAN initiated all dropping probability redution ratio dynami RESHO soft handoff region inreases. Besides, hen dupliate resoures are onsumed for less time the bloking probabilities are highly redued. In addition, besides a lo MS veloity induing a better performane gain, hen the veloity variability is high (α = 6) the performane gain for the 3G ne all bloking probability is even higher. This is due to the fat that the inrease of veloity variability indues more temporal statistial multiplexing in the WLAN apaity oupation induing less WLAN bloking and less WLAN handoff failures and onsequently less overfloing upard VHO alls traffi and less 3G ne all bloking probability. Sine a lo-mean high-variane veloity profile typially haraterizes the WLAN mobility environment [0], it may push WLAN designers to implement the RESHO algorithm if its expeted performane gain estimated by our proposed model is appreiated. Furthermore, the signifiant performane gain provided by the RESHO algorithm in partiularly lo mobility environments make the algorithm a someho less V. CONCLUSION In this paper e presented a novel analytial model in order to evaluate the ne all bloking probability and the all dropping probability of the optimal voie admission ontrol algorithm redefined in loosely oupled 3G/WLAN netorks. This model uses more aurate residene time distributions and takes into aount the soft handoff region for eah 3G/WLAN ell for hih dupliate resoures are alloated. Results sho that the delimitation of the soft handoff region using a more or less effiient soft handoff algorithm has a big impat on the CAC performane metris. Partiularly, it is observed that the ne all bloking probability redution gained by using a resoure-effiient soft handoff (RESHO) algorithm ompared to a stati-threshold soft handoff (STSHO) algorithm is largely inreased hen MS veloities have lo mean and high variability; hih typially haraterizes the WLAN environment. Our proposed model, validated by simulations, an help netork designers to figure out if the gained performane from a hallenging RESHO algorithm in different mobility environments is orth to implement it. APPENDIX Derivation of the 3G/WLAN dropping probabilities Assuming that all dropping an only be avoided by suessful obligatory handoffs, i.e. HHO and UVHO, the expression of dropping probability of a all initiated in the 3G, noted Pd, is already knon in ellular netorks : P d = i= (P h )i ( P f )i P f = P h P f P h ( P f ) (3) Where Ph is the 3G handoff probability and P f is the 3G handoff failure probability. Hoever, the expression of dropping probability of a all initiated in the overlaying WLAN, noted Pd, need to be developed. let P = Pd the probability that a voie all that is not initially bloked Authorized liensed use limited to: ECOLE POLYTECHNIQUE DE MONTREAL. Donloaded on vember 6, 2009 at 0:07 from IEEE Xplore. Restritions apply.

10 BEN ALI and PIERRE: ON THE IMPACT OF SOFT VERTICAL HANDOFF ON OPTIMAL VOICE ADMISSION CONTROL 365 by a WLAN ill be normally ompleted ithout any handoff failure. P is the sum of the probabilities of the folloing events : The all is ompleted normally in the first WLAN ell; The all is ompleted normally in a WLAN ell after one or more WLAN-to-WLAN HHOs hih are all suessful; The all is ompleted normally in a 3G ell after one or more WLAN-to-WLAN HHOs, one WLAN-to-3G UVHO, and eventual subsequent 3G-to-3G HHOs hih are all suessful. We assume that an ongoing all moving to the WLAN from the overlaying 3G ell (ith a small probability P as an optional ontinuity to the previous event) an not be dropped sine the overlaying 3G ell ont free its reserved hannel before ensuring the suessful soft DVHO to the WLAN here it is assumed to be ompleted normally. Thus : P =( P h )+( P h ) i=(p h )i ( P f )i +( P h ) [ i i= j= (P h ) j ( P f ) j P f ( P b )(P h )i j ( P f )i j ] P = ( P h ) P h ( P f ) + ( P h )P f P h ( P f ) [ P h ( P f )][ P h ( P f )] ( P b ) ( P f ) (32) P d = P = P h P f (P h P f + P b P h P b ) ( P h + P h P f )( P h + P h P f ) = Pd P d ( +( Ph ) P b Pf ) (33) If WLAN-to-3G UVHO alls are admitted as 3G-to-3G HHO alls hih an be done in a tightly oupled 3G/WLAN, i.e. by replaing Pb ith Pf in the last Equation, then Pd = Pd P d /P h. Reall that P d and P d are respetively the dropping probabilities in the separate WLAN and 3G netorks respetively. REFERENCES [] TOC Vie, Solutions to performane problems in VoIP over a 802. ireless LAN," IEEE Trans. Veh. Tehnol., vol. 54, no., pp , [2] X. Shen, Y. B. Lin, A. C. Pang, and J. Pan, Guest editorial: voie over ireless loal area netork," IEEE Wireless Commun. [see also IEEE Personal Commun.], vol. 3, no., pp. 4-5, [3] S. Pak, J. Choi, T. Kon, and Y. Choi, Fast-handoff support in IEEE 802. ireless netorks," IEEE Commun. Surveys Tutorials, vol. 9, no., pp. 2-2, [4] L.Cai,Y.Xiao,X.S.Shen,andJ.W.Mark, VoIPoverWLAN:voie apaity, admission ontrol, QoS, and MAC," Int. J. Commun. Syst., vol. 9, pp , [5] B. Sikdar, An analyti model for the delay in IEEE 802. PCF MAC based ireless netorks," IEEE Trans. Wireless Commun., vol. 6, no. 4, pp., [6] C. S. Li, Y. C. Tseng, and H. C. Chao, A neighbor ahing mehanism for handoff in IEEE 802. ireless netorks," in Pro. Int. Conf. Multimedia Ubiquitous Engineering 2007 (MUE 07), pp [7] W. Song, H. Jiang, and W. Zhuang, Performane analysis of the WLAN-first sheme in ellular/wlan interorking," IEEE Trans. Wireless Commun., vol. 6, no. 5, pp , [8] A. H. Zahran, B. Liang, and A. Saleh, Signal threshold adaptation for vertial handoff in heterogeneous ireless netorks," Mobile Netorks Appliations, vol., no. 4, pp , [9] W. Guan, X. Ling, X. S. Shen, and D. Zhao, Handoff trigger table for integrated 3G/WLAN netorks," in Pro. International Conf. Commun. Mobile Computing, pp , [0] S. Thajhayapong and J. M. Peha, Mobility patterns in miroellular ireless netork, IEEE Trans. Mobile Computing, vol. 5, no., pp , Jan [] Y. Fang, Modeling and performane analysis for ireless mobile netorks: a ne analytial approah," IEEE/ACM Trans. Netorking (TON), vol. 3, no. 5, pp , [2] H. Zeng and I. Chlamta, Adaptive guard hannel alloation and bloking probability estimation in PCS netorks," Computer Netorks, vol. 43, no. 2, pp , [3] H. Holma and A. Toskala, WCDMA for UMTS: Radio Aess for Third Generation Mobile Communiations. John Wiley and Sons, [4] IEEE 802. Working Group, IEEE Std 802.e ireless LAN medium aess ontrol and physial layer speifiations, medium aess ontrol quality of servie enhanements," IEEE Standards, [5] M. Naghshineh and A. S. Aampora, QoS provisioning in miroellular netorks supporting multiple lasses of traffi," Wireless Netorks, vol. 2, no. 3, pp , 996. [6] R. Ramjee, D. Tosley, and R. Nagarajan, On optimal all admission ontrol in ellular netorks," Wireless Netorks, vol. 3, no., pp. 29-4, 997. [7] Y. C. Fang and I. Y. B. Lin, Channel oupany times and handoff rate for mobile omputing andpcs netorks," IEEE Trans. Computers, vol. 47, no. 6, pp , 998. [8] H. Bing, C. He, and L. Jiang, Performane analysis of vertial handover in a UMTS-WLAN integrated netork," in Pro. 4th IEEE Personal, Indoor and Mobile Radio Commun. (PIMRC 2003), vol.. Raha Ben Ali (S 02) reeived the M.S. degree (2003) and the Ph.D. degree (2008) in Computer Engineering from Eole Polytehnique de Montreal, Qu ebe, Canada. He as a omputer netork arhitet and engineer at Tele-University of Quebe from 2006 to Previous to that, he orked as a Graduate Researh Assistant ith the Mobile Computing and Netorking Researh Laboratory (LARIM) and ith Erisson Researh Canada. The urrent fous of his researh interests is on radio resoure management, all admission ontrol, heterogenous mobile netorks interorking, performane evaluation and quality of servie (QoS) provisioning for next-generation netorks. Samuel Pierre (SM 97) reeived the B.Eng. degree in Civil Engineering in 98 from Eole Polytehnique de Montreal, Qu ebe, Canada, the B.S. and M.S. degrees in Mathematis and Computer Siene in 984 and 985, respetively, from the UQAM, Montreal, the M.S. degree in Eonomis in 987 from the University of Montreal, and the Ph.D. degree in Eletrial Engineering in 99 from Eole Polytehnique de Montreal. From 987 to 998, he as a Professor at the University of Qu ebe at Trois-Rivi eres prior to joining the T el e-universit e of Qu ebe, an Adjunt Professor at Universit e Laval, Ste-Foy, Qu ebe, an Invited Professor at the Siss Federal Institute of Tehnology, Lausanne, Sitzerland, and the the Universit e Paris 7, Frane. He is urrently a Professor of Computer Engineering at Eole Polytehnique de Montreal, here he is Diretor of the Mobile Computing and Netorking Researh Laboratory (LARIM), Chairholder of the NSERC/Erisson Chair in Next Generations and Mobile Netorking Systems, and the Diretor of the Mobile Computing and Netorking Researh Group (GRIM). His researh interests inlude ireline and ireless netorks, mobile omputing, artifiial intelligene, and telelearning. Dr. Pierre is a Fello of the Engineering Institute of Canada (EIC). He is a Regional Editor of the JOURNAL OF COMPUTER SCIENCE, an Assoiate Editor of IEEE COMMUNICATIONS LETTERS, IEEE CANADIAN JOURNAL OF ELECTRICAL AND COMPUTER ENGINEERING and IEEE CANADIAN REVIEW, and serves on the editorial board of TELEMATICS AND INFORMATICS, edited by Elsevier Siene. Authorized liensed use limited to: ECOLE POLYTECHNIQUE DE MONTREAL. Donloaded on vember 6, 2009 at 0:07 from IEEE Xplore. Restritions apply.