Connection Admission Control for Multi-Service Integrated Cellular/WLAN System

Transcription

1 Connection Admiion Control for Multi-Service Integrated Cellular/WLAN Sytem Enrique Steven-Navarro Student Member IEEE A. Hamed Mohenian-Rad Student Member IEEE and Vincent W.S. Wong Senior Member IEEE Abtract The complementary characteritic of wirele cellular network and wirele local area network WLAN make them uitable to jointly offer eamle wirele acce ervice to mobile uer. In an integrated cellular/wlan ytem the quality of ervice QoS requirement for different ervice e.g. voice real-time video require admiion control to limit the number of connection in each acce network. In thi paper we firt develop an analytical model to facilitate the evaluation of different admiion control policie in a multi-ervice integrated cellular/wlan ytem. We then formulate two different revenue maximization problem. Each problem take different QoS requirement into account. By olving the equivalent cot minimization problem we evaluate the ytem performance when different combination of cutoff priority and fractional guard channel admiion control policie are being ued. Reult how that uing cutoff priority policy in both wirele acce network can achieve the optimal olution for the two optimization problem under a wide range of network condition. Index Term Cellular/WLAN interworking admiion control handoff management heterogeneou wirele network multi-dimenional Markov chain. I. INTRODUCTION Recent tudie have hown that wirele wide area network uch a third generation 3G wirele cellular ytem can be integrated with wirele local area network WLAN to offer Internet acce and IP multimedia ervice to mobile uer. In an integrated cellular/wlan ytem WLAN are uually deployed in denely populated area and wirele cellular network are ued to provide wide area network coverage. Variou interworking architecture have been propoed in the literature [] [4]. Uer carrying mobile device equipped with multiple interface can etablih connection with different available acce network. A the uer move within the coverage area they are able to witch connection among network according to roaming agreement. The IEEE ha alo et up the 8. media independent handover working group to tandardize inter-operability between 8 and non-8 network e.g. 3G cellular ytem [5]. The 3G Partnerhip Project 3GPP and 3GPP are alo aiming to extend their 3G packet data and IP multimedia ervice to the WLAN environment. Different Manucript received June 7; revied November 6 7 January 3 8; accepted February 8. Thi work wa upported by Bell Canada the Natural Science and Engineering Reearch Council NSERC of Canada and the Programa de Mejoramiento del Profeorado PROMEP from Mexico. Part of thi paper wa preented at the IEEE Global Telecommunication Conference Globecom Wahington DC November 7. The review of thi paper wa coordinated by Prof. Qian Zhang. The author are with the Department of Electrical and Computer Engineering Univerity of Britih Columbia Vancouver BC V6T Z4 Canada {enrique hamed vincentw}@ece.ubc.ca. level of integration have been propoed ranging from common billing and cutomer care to eamle mobility and eion continuity [6] [7]. The proce of witching connection among network i called handoff or handover. A handoff i called horizontal if it i between two network which ue the ame acce technology e.g. between two WLAN or between two neighboring cell in a wirele cellular network. On the other hand a handoff i called vertical if it i between two network which ue different acce technologie e.g. from a cell in a wirele cellular network to a WLAN or vice vera. In order to guarantee the quality of ervice QoS of different IP multimedia application e.g. voice real-time video it i neceary to limit the number of connection admitted to a network. Thu a proper connection admiion control policy i required in each network. An admiion control policy can either accept the connection requet and allocate the reource accordingly or reject the connection requet. In general higher priority i given to the requet from the handoff uer a oppoed to the new uer ince from the uer point of view having a connection abruptly terminated i more annoying than being blocked occaionally on new connection attempt. Some of the admiion control policie for wirele cellular network include the cutoff priority CP [8] and the fractional guard channel FG cheme [9]. The CP policy reerve a fixed number of channel for connection requet from handoff uer. The connection requet from new uer are blocked if there i no unreerved channel available. On the other hand the FG policy reerve channel for handoff requet by blocking the connection requet from new uer with a probability which i proportional to the current occupancy. Both CP and FG policie manage to limit the maximum number of connection in each network according to the QoS requirement of the exiting connection. We now ummarize ome of the related work on the integrated cellular/wlan ytem. In [] an admiion policy for 3G cellular ytem with complementary WLAN i propoed. In [] three different load haring cheme for integrated univeral mobile telecommunication ytem UMTS/WLAN ytem with buffering capabilitie are propoed. In [] an integrated cellular/wlan ytem with reource haring and admiion control capabilitie i propoed. The CP admiion control policy i ued and the network performance i evaluated in term of the blocking probabilitie of new and handoff connection. All the work in [] [] only conider ingle-ervice cla. That i they aume that all arriving connection requet the ame amount of bandwidth.

2 Some recent work in cellular/wlan interworking aim to differentiate ervice requirement. In [3] the WLAN-firt admiion control cheme i propoed where both voice and data connection requet within the overlapped coverage area are tranferred to the WLAN. In [4] a randomized guard channel admiion control policy i propoed in which a random number of channel i reerved for voice handoff. Some channel are alo elected randomly to be excluively ued by new voice connection. The remaining bandwidth i then hared by all data connection. Although there have been variou model and admiion control policie propoed in the literature our work i motivated by three particular apect: the conideration of everal WLAN deployed inide the cell of a wirele cellular network; the upport of multiple ervice clae with different bandwidth requirement; and 3 the effect of uing a combination of different admiion control policie in wirele acce network. Our work aim to incorporate thee important apect in an optimization-baed deign for connection admiion control in integrated cellular/wlan ytem. In thi paper we develop an analytical model to facilitate the evaluation of different combination of connection admiion control policie in a multi-ervice integrated cellular/wlan ytem. In our model everal WLAN are deployed within the ame coverage area of the wirele cellular network. To accommodate the behavior of different admiion control algorithm we introduce the concept of policy function. They are defined baed on the ervice category e.g. voice data the type of the connection requet i.e. new requet or handoff requet and the admiion control policy being ued. Different from our previou work in [5] here we define different policy function for each type of handoff requet i.e. horizontal/vertical. The contribution of our work are a follow: Our model take into account variou important ytem parameter including the level of mobility and the arrival rate of connection requet from the uer the capacity and the coverage area of each wirele network the admiion control policie and the QoS requirement in term of the blocking and dropping probabilitie. Flexible policy function are defined for each ervice category and for each type of connection requet. We ue the cutoff priority [8] and the fractional guard channel [9] admiion control policie and determine the correponding policy function. Thee function alo allow u to evaluate different policy combination. 3 We formulate two different revenue maximization problem for multi-ervice integrated cellular/wlan ytem. Each problem take a ditinct et of QoS requirement into conideration. 4 We evaluate the performance of the cellular/wlan ytem uing different policy combination under variou level of mobility and arrival rate of connection requet. Reult how that uing cutoff priority policy in both wirele acce network can achieve the optimal olution for both optimization problem under a wide range of network condition. Fig.. An integrated cellular/wlan ytem. The ret of thi paper i organized a follow. The analytical model for the integrated cellular/wlan ytem i decribed in Section II. The optimization-baed admiion control problem are formulated in Section III. Numerical reult are preented in Section IV. Concluion are given in Section V. II. CELLULAR/WLAN SYSTEM MODEL Conider an integrated cellular/wlan ytem where one or more WLAN may be deployed inide each cell of the cellular network a hown in Fig.. There are two pecific coverage area to be conidered: the cellular-only coverage area and the dual cellular/wlan coverage area. In thi context coverage correpond to ervice availability. In general the dual cellular/wlan coverage area are deployed in pecific area where the demand for data ervice i much higher than the ret of the cellular-only area [4]. Horizontal and vertical handoff can occur due to mobility of the uer under different coverage area. In thi ection we preent a model for the multi-ervice integrated cellular/wlan ytem. Given the admiion control policie we determine the probabilitie of blocking connection requet from the new uer and the probabilitie of dropping connection requet from the handoff uer. A. Traffic and Mobility Model We firt introduce the notation. Let M c denote the et of all cell in a wirele cellular network A c i denote the et of cell adjacent to cell i Wi c denote the et of WLAN inide the coverage area of cell i A w k denote the et of WLAN adjacent to WLAN k and Dk w denote the et containing the overlaying cell of WLAN k i.e. a dual cellular/wlan coverage area. A an example from Fig. we have: M c = { 3 4} A c = { 3 4} W c = {5 6} A w 5 = {6} and D5 w = {}. Let S denote the et of multimedia ervice available to the mobile uer. Each ervice S require b c baic bandwidth unit BBU [6] and b w BBU to guarantee it QoS requirement in the cellular network and the WLAN repectively. A an example a BBU in the wirele cellular network can be 3 kbp and a BBU in the WLAN can be 64 kbp. The new connection requet for ervice arrive at cell i and WLAN k according to independent Poion procee with rate λ c i

3 3 and λ w k repectively. The duration of the connection uing ervice i defined a connection time t. We aume that t i an exponentially ditributed random variable with mean /v. Since the exponential ditribution i memoryle the reidual i.e. remaining connection time t R i alo exponentially ditributed with mean /v. To model the mobility we define the inter-boundary time imilar to [7] a the time interval between any two conecutive acce network boundary croing by a mobile uer. The inter-boundary time depend on the ize of the cell and the mobility pattern of the uer. If an inter-boundary time tart at the moment of entering cell i then we denote it by t c b i. If an inter-boundary time tart at the moment of entering WLAN k then we denote it by t w b k. We aume that both t c b i and t w b k are exponentially ditributed random variable with mean /ηi c and /ηk w repectively. Fig. how tc b i between boundary croing point A and B and t w b k between boundary croing point B and C. The channel holding time in cell i i defined a the time that a connected mobile uer continue to ue b c BBU of reource in the network. The aociation holding time in WLAN k i defined a the time that a connected uer continue to aociate with the acce point. For ervice type the channel holding time in cell i and the aociation holding time in WLAN k are obtained a mint R t c b i and mint R t w b k repectively. Since t R t c b i and have exponential ditribution for all S i M c and t w b k k Wi c the holding time are alo exponentially ditributed with parameter µ c i = v +ηi c and µw k = v +ηk w repectively. A mobile uer who i holding a connection of ervice type in cell i may terminate it connection at the end of it holding time and leave the integrated cellular/wlan ytem with probability qi c = υ /υ + ηi c. It may alo move within the ytem and continue in an adjacent cell or an underlaying WLAN with probability qi c. We have: q c i = ηc i υ + η c i j A c i q cc ij q cw ik where qij cc denote the probability of attempting a horizontal handoff from cell i to neighboring cell j and qik cw denote the probability of attempting a vertical handoff from cell i to WLAN k which i inide the coverage area of cell i. Similarly a mobile uer who i holding a connection of ervice type in WLAN k may terminate it connection at the end of it aociation holding time and leave the integrated ytem with probability qk w = υ /υ +ηk w. It may alo move within the ytem and continue in an adjacent WLAN or an overlaying cell with probability qk w. We have: q w k = ηw k υ + η w k l A w k q ww kl i D w k q wc ki where qkl ww denote the probability of attempting a horizontal handoff from WLAN k to adjacent WLAN l and qki wc denote the probability of attempting a vertical handoff from WLAN k to it overlaying cell i. B. Multi-Dimenional Birth-Death Procee m c i Each cell i i aumed to have a capacity of Ci c BBU. Let denote the number of connection uing multimedia ervice type in cell i. The capacity contraint require that m c i b c Ci c i M c. 3 S From 3 there can be at mot Ci c/bc connection of ervice type in cell i at any time. We define m c i = m c i m c i... m c i S a the occupancy vector in cell i. For each cell i M c the admiion control policie for connection requet from new horizontal handoff and vertical handoff uer for ervice type S can be modeled by policy m c i m c i function βn c mc i m c i βc hh i mm c i and βc vh i mm c i repectively. The policy function βn c mc i m c i determine the probability of not accepting a connection requet from a new uer for ervice type in cell i. Similarly the policy function βhh c mc i m c i and βvh c i m m i c determine the probability of not accepting a connection requet from a horizontal handoff uer and from a vertical handoff uer for ervice type in cell i repectively. Since handoff requet i.e. either horizontal or vertical have higher priority than new requet it i neceary that βhh c i m m c i βc n i m m c i and βc vh i m m c i βc n i m m c i for all i M c and S. Note that the probability of not accepting connection requet depend on the pecific admiion control policy being ued. In fact many admiion control policie including CP and FG can be mathematically modeled in the form of their correponding policy function. We will dicu policy function in detail in Section III. m c i An occupancy vector m c i i feaible if mc i for all S and the contraint in 3 i atified. We denote the et of all feaible m c i vector by Θc i. The occupancy of cell i evolve according to a multi-dimenional birth-death proce [8]. A birth event happen when a connection requet to cell i from a handoff or a new uer i accepted. A death event occur when a uer either terminate it connection or leave cell i. The multi-dimenional birth-death proce ha S dimenion where denote the cardinality of the et. The th dimenion model the channel occupancy evolvement due to the change in the number of connection uing ervice type. m c i Let Pi cmc m c i denote the probability of being in tate mc m c i in the S -dimenional birth-death proce correponding to cell i. We have Bn c i P c i m m i c i βn c mc i m i c i 4 m c i Θc i B c hh i m c i Θc i B c vh i m c i Θc i m c i m c i Pi c m m i c i βhh c mc i m i c i 5 Pi c m m i c i βvh c i m m i c i 6 where Bn c i denote the probability of blocking connection requet for ervice in cell i from new uer and Bhh c i and Bvh c i denote the probabilitie of dropping connection requet for ervice in cell i from horizontal handoff and vertical handoff uer repectively. To model the capacity in IEEE 8. WLAN it i reaonable to aume that there are only packet tranmiion between the acce point and the mobile device but not among the device. For each WLAN k the media acce

4 4 i controlled either in a centralized manner uing the point coordination function PCF or in a decentralized manner uing the ditributed coordination function DCF. We how in Appendix I that in either cae the capacity contraint can be modeled a: m w k b w Ck w k Wi c i M c 7 S where Ck w i the effective data rate in WLAN k in BBU m w k i the number of connection uing ervice type in WLAN k and m w k = mw k m w k... m w k S i the occupancy vector in WLAN k. If PCF i being ued then the effective data rate i cloe to the nominal data rate. On the other hand if DCF i being ued which i widely deployed in current WLAN the effective data rate Ck w i ignificantly le than the nominal rate. There are a few approximate analytical model that can obtain the capacity of the WLAN under certain aumption [] []. However finding an accurate value i not an eay tak. Neverthele we can ue either 8.-baed imulation or tet-bed meaurement to etimate Ck w. In thi paper we ue n- [] imulation to etimate Ck w. Conider an arbitrary cell i M c. For each WLAN k Wi c the admiion control policie for connection requet from new horizontal and vertical handoff uer for ervice type S can be modeled by policy function βn w mw k m w k βhh w k m m w k and βw vh k m m w k repectively. An occupancy vector m w k i feaible if mw k for all S and the contraint in 7 i atified. We denote the et of all feaible m w k vector by Θw k. The occupancy of WLAN k evolve according to an S -dimenional birth-death proce independent of other WLAN. Let Pk wmw m w k denote the probability of being in tate m w k in the S -dimenional birthdeath proce correponding to WLAN k. Let Bn w k denote the probability of blocking connection requet for ervice type in WLAN k for new uer. On the other hand Bhh w k and Bvh w k denote the probabilitie of dropping connection requet for ervice type in WLAN k for horizontal handoff and vertical handoff uer repectively. We have Bn w k Pk w m m w k βn w k m m w k 8 m w k Θw k B w hh k m w k Θw k B w vh k m w k Θw k m w k Pk w m m w k βhh w mw k m w k 9 Pk w m m w k βvh w mw k m w k. Let φ c i denote the birth rate of ervice type in the birthdeath proce correponding to cell i. Similarly let φ w k denote the birth rate of ervice type in the proce correponding to WLAN k. We have: φ c i = λ c i βn c mc i m i c i h cc ji βhh c mc i m i c i j A c i vki wc + τki wc βvh c mc i m i c i IEEE 8.a upport and 54 Mbp nominal data rate. IEEE 8.b alo upport 5.5 and Mbp data rate [9]. φ w k = λ w k βn w mw k m w k i D w k l A w k lk βhh w mw k m w k h ww vik cw + τik cw βvh w mw k m w k where h cc ij denote the horizontal handoff rate of ervice offered to cell i from it adjacent cell j vki wc denote the vertical handoff rate of ervice offered to cell i from it underlying WLAN k τki wc denote the rate of all handoff traffic of ervice that i not accepted in WLAN k and hence i tranferred to cell i h ww lk denote the horizontal handoff rate of ervice offered to WLAN k from it adjacent WLAN l vik cw denote the vertical handoff rate of ervice offered to WLAN k from it overlaying cell i and τik cw denote the rate of all handoff traffic of ervice that i not accepted in cell i and hence i tranferred to WLAN k. We have: qji cc h cc ji = λ c j B c n j l W c j x A c j h cc xj B c hh j q cc ji vlj wc + τlj wc Bvh c j q cc ji 3 v wc ki = λ w k B w n k q wc ki j D w k l A w k h ww lk Bhh w k qki wc vjk cw + τjk cw Bvh w k q wc ki 4 τki wc = vik cw Bvh w k h ww lk Bhh w k 5 l A w k h ww lk = λ w l Bn w l q ww lk i D w l y A w l h ww yl Bhh w l q ww lk vil cw + τil cw Bvh w l q ww lk 6 vik cw = λ c i Bn c i R ik qik cw q cw ik τ cw ik = l W c i j A c i v wc li R ik + τ wc li v wc ki B c vh i j A c i h cc ji R ik Bhh c i B c vh i h cc ji Bhh c i q cw ik 7 R ik 8 where R ik denote the coverage factor between WLAN k and cell i i.e. the ratio between the radio coverage area of WLAN k and the radio coverage area of cell i. Note that R ik for all i M c and k Wi c. New connection arrival rate a well a the horizontal and vertical handoff rate are hown in Fig. where the ubcript i omitted for the ake of clarity. Let ϕ c i denote the death rate of ervice type in the birthdeath proce correponding to cell i. Recall that a death event occur when a uer either terminate it connection or leave cell i. Similarly let ϕ w k denote the death rate of ervice type in the proce correponding to WLAN k. We have: ϕ c i = m c i µ c i 9

5 5 ϕ w k = m w k µ w k. Given the policy function βn c i βhh c βc i vh βw i n k βhh w k βvh w k and the network parameter λ c i λ w k ηi c ηw k µc i µ w k qik cw qij cc qki wc qkl ww Ci c Cw k υ b c and b w for all i j M c k l Wi c and S we can olve the global balance equation from the birth-death procee and obtain the correponding blocking and dropping probabilitie Bn c i Bhh c Bc i vh Bw i n k Bhh w k and Bvh w k for all i M c k Wi c and S. To compute the birth rate in - we need to olve the et of fixed-point equation given by the handoff rate 3-8. It can be accomplihed by uing the iterative fixed-point algorithm of repeated ubtitution [3]. The fixed-point algorithm i decribed in Appendix II. III. CONNECTION ADMISSION CONTROL In thi ection we introduce the concept of policy function and derive the correponding function for the admiion control policie. We alo define the policy combination and formulate the optimization problem. A. Policy Function For each cell i M c connection admiion control include the policy function βn c i m m c i βc hh i m m c i and βc vh i m m c i for new horizontal and vertical handoff connection requet repectively. The policy function are able to model the behavior of the admiion control policie. They determine the probability of not accepting a connection for each type of requet and ervice given the occupancy vector m c i and according to a pecific policy. Thu network deigner can ue them either to evaluate admiion control policie already propoed or to examine new one. To determine the correponding policy function the deigner decide how each type of connection requet i being treated i.e. accept or reject baed on the current number of connection of each ervice i.e. the occupancy vector. A policy can be modeled if it can be repreented a a function of the occupancy vector. A mentioned in Section II priority i uually given to handoff connection requet over new connection requet i.e. βhh c i m m c i βc n i m m c i and βc vh i m m c i βc n i m m c i. In the cae of an integrated cellular/wlan ytem conideration have to be made on how to et βhh c mc i m c i and βvh c mc i m c i baed on the difference between horizontal and vertical handoff connection requet. Note that the vertical handoff deciion proce alway occur before the connection requet. Such deciion proce i more elaborate than the one for the horizontal handoff which i uually baed on the received ignal trength RSS from the bae tation. The vertical handoff deciion beide RSS need to conider additional parameter uch a acce cot power conumption and QoS factor [4]. Intereted reader may refer to [5] [6] and the reference therein. For the cope of thi work the admiion control policy i invoked once the vertical handoff deciion ha been made. From the network operator point of view we can divide the treatment of the vertical handoff connection requet into two cae: If the connection requet i from a uer who i a ubcriber of the network on which admiion i requeted then the operator may et the policy function βhh c i m m c i = βc vh i m m c i. That i the uer in cellular network receive the ame QoS in term of probability of dropping connection when viiting a WLAN. If the connection requet i from a uer who i not a ubcriber of the network on which admiion i requeted i.e. a roaming uer then the operator may et the policy function βhh c i m m c i < βc vh i m m c i. That i the horizontal handoff uer from neighboring cell have priority over the vertical handoff uer from WLAN. By uing the policy function we can extend CP and FG a two example of admiion control policie from wirele cellular network to the integrated cellular/wlan ytem. Recall from Section I that the CP policy reerve a fixed number of available channel i.e. BBU for handoff requet. Uing the notation of policy function a connection requet to cell i for ervice type i rejected by the CP policy with the following probability for new uer: βn c mc i m i c i = if S m c i bc T c i otherwie and the following probabilitie for horizontal and vertical handoff uer: if m c βhh c i m m i c i i = bc Cc i b c S βvh c mc i m i c i = otherwie if S otherwie m c i bc V c i 3 where S mc i bc denote the current occupancy the integer parameter Ti c in i ued to tune the threhold to give priority to handoff requet and the parameter Vi c in 3 i ued to tune the threhold to give priority to horizontal handoff requet. Note that for all i M c and S we have Ti c Ci c bc and Vi c Ci c bc. In and 3 cae i conidered. Note that by etting Vi c = Ci c bc cae reduce to cae i.e. βhh c mc i m c i = βc vh mc i m c i. In the FG policy connection requet from either new uer or vertical handoff uer are rejected with probabilitie which are proportional to the current occupancy. A connection requet to cell i M c for ervice type i rejected by FG with the following probability for new uer: if βn c mc i m i c i = Sm c i bc T i c min ΦTi c 4 otherwie m c i bc T c i ΦT c i = S C c i bc + T c i 5 and the following probabilitie for horizontal and vertical handoff uer: if m c βhh c i m m i c i i = bc Cc i b c S 6 otherwie

6 6 if m c βvh c mc i m i c i bc V c i = S min ΦVi c otherwie. i 7 Note that 5 i equal to zero when the occupancy i at it threhold i.e. S mc i bc = T c i and i greater than or equal to one when there i not enough bandwidth to allocate i.e. S mc i bc Cc i bc +. The probability of not accepting new connection requet increae linearly from zero to one a the occupancy increae from Ti c to Ci c bc +. For WLAN k connection requet from new horizontal and vertical handoff uer for ervice type are not accepted with probabilitie βn w mw k m w k βw hh mw k m w k and βw vh mw k m w k repectively. Such policy function can be defined imilarly according to CP or FG policie. We need to replace the upercript c with w and the ubcript i with k in -7. B. Policy Combination and Optimization Problem Given the CP and FG admiion policie four different policy combination can be conidered: Wirele cellular ytem and WLAN both ue CP policy i.e. CP c -CP w. Wirele cellular ytem ue CP policy and WLAN ue FG policy i.e. CP c -F G w. 3 Wirele cellular ytem ue FG policy and WLAN ue CP policy i.e. F G c -CP w. 4 Wirele cellular ytem and WLAN both ue FG policy i.e. F G c -F G w. For any combination different parameter Ti c Tk w Vi c and Vk w for all i M c k Wi c and S can lead to different performance. The quetion are: Which of the four poible combined policie hould be ued? How hould the correponding parameter be choen? To anwer thee quetion we conider the following two optimization problem: Optimization Problem : Given the policy function and the network parameter maximize a linear objective function of the accepted traffic for connection requet from new a well a horizontal and vertical handoff uer: maximize Ti c T k w V i c V k w S i M c [α cni λ ci B cni + α c hh i ψ c hh i Bhh c i + α c vh ψc i vh i [α wnk λ wk B wnk + αhh w k ψhh w k Bvh c i B w hhk + α w vhk ψ w vhk B w vh k ]] 8 where ψhh c and ψw i hh k denote the aggregated rate of horizontal handoff traffic of ervice type that arrive at cell i and WLAN k for all i M c and k Wi c repectively. ψc vh i and ψvh w k denote the aggregated rate of vertical handoff and tranferred traffic of ervice type that arrive at cell i and + WLAN k for all i M c and k Wi c repectively. We have: ψ c hh i ψ w hh k ψ c vh i j A c i l A w k ψ w vh k i D w k h cc ji 9 h ww lk 3 v wc ki v cw ik i D w k τ wc ki 3 τ cw ik. 3 Note that the blocking and dropping probabilitie Bn c Bc i hh i Bvh c Bw i n Bw k hh and Bw k vh k depend on the policy function βn c βc i hh βc i vh βw i n βw k hh and βw k vh k wherea the policy function depend on Ti c Tk w Vi c and Vk w. In 8 the contant αhh c i αhh w k αvh c i and αvh w k denote the revenue of accepting a connection requet for ervice type from a horizontal and from a vertical handoff uer in cell i and WLAN k repectively. Similarly αn c and αw i n k denote the revenue of accepting a connection requet for ervice type from a new uer in cell i and WLAN k repectively. In general it i reaonable to et αn c i αhh c i and αn c i αvh c i for all i M c and S to enure that a higher priority i given to accepting connection requet from handoff uer of any type rather than new uer. We can alo aign different revenue for different ervice. It i ueful when the ervice are offered with different ervice fee. By removing the contant term problem 8 can be reduced to the following equivalent blocking cot minimization problem: [ minimize α Ti c T k w V i c V k w n c λc i i Bn c + αc i hh ψc i hh Bc i hh i S i M c + αvh c i ψvh c i Bvh c i [ αn w k λ w k Bn w k ]] + αhh w ψw k hh Bw k hh +αw k vh ψw k vh Bw k vh k 33 where αhh c i can be interpreted a the amount of revenue lot due to dropping a connection requet for ervice type from a horizontal handoff uer in cell i. It can alo be interpreted a the cot of dropping. The ret of the parameter can be interpreted in a imilar way i.e. cot of blocking. Thu the objective of problem 33 i to minimize all penalty cot i.e. revenue lo incurred in the integrated cellular/wlan ytem when connection requet from new horizontal and vertical handoff uer are blocked and dropped repectively. For the ret of the paper we refer to thi cot a the cot of blocking connection. Optimization Problem : Given the policy function and the network parameter maximize a linear function of accepted traffic for connection requet from new uer ubject to the contraint on dropping probabilitie for connection requet

7 7 from handoff uer: maximize Ti c T k w V i c V k w S i M c [ αn c λc i i Bn c i α w n k λ w k B w n k ] ubject to B c hh i Γ c hh i i M c S B w hh i Γ w hh k k W c i S B c vh i Γc vh i i M c S B w vh i Γw vh k k W c i S 34 where Γ c hh i and Γ w hh k are the maximum dropping probabilitie tolerated for horizontal handoff uer of ervice type in cell i and WLAN k repectively. Γ c vh i and Γ w vh k are the maximum dropping probabilitie tolerated for vertical handoff uer of ervice type in cell i and WLAN k repectively. Compared to the objective function in 8 we can ee that the one in 34 doe not include the revenue obtained from handoff uer. Intead new contraint are introduced to guarantee the QoS requirement for thoe uer. By removing the contant term problem 34 can be reduced to the following equivalent blocking cot minimization problem: [ minimize α Ti c T k w V i c V k w n c i λ c i Bn c i S i M c ] αn w λw k k Bn w k ubject to B c hh i Γc hh i i M c S B w hh i Γw hh k k W c i S B c vh i Γc vh i i M c S B w vh i Γw vh k k W c i S. 35 In problem 35 the parameter αn c i can be interpreted a the cot of blocking a connection requet for ervice type from a new uer in cell i. Thu the objective of problem 35 i to minimize all penalty cot incurred in the integrated cellular/wlan ytem when connection requet from new uer are blocked ubject to QoS contraint for connection requet from handoff uer. In other word the QoS i guaranteed for connection from uer already accepted in the ytem while aiming to minimize the penalty of rejecting new uer. The problem in 33 and 35 are combinatorial optimization problem. They can be olved by uing either the exhautive earch or other meta-heuritic algorithm [7]. IV. NUMERICAL RESULTS AND DISCUSSIONS We evaluate the performance of an integrated cellular/wlan ytem coniting of a wirele cellular network with M c = 3 cell and Wi c = WLAN. The cell and WLAN are enumerated a follow: M c = { 3} W c = {4 5} W c = {6 7} and W3 c = {8 9}. Thu inide the coverage of cell there are two dual cellular/wlan coverage area given by WLAN 4 and WLAN 5 repectively. Network throughput Mbp Service connection Service connection Fig.. Aggregate network throughput in a WLAN when the number of connection for ervice type varie from to and the number of connection for ervice type varie from to 6. In each cell i M c the network capacity i Mbp and the BBU i et to 3 kbp baed on the 3GPP upported multimedia bearer ervice [8]. Thi implie that the capacity in 3 of each cell i Ci c = 6 BBU. We aume that two multimedia ervice are offered i.e. S = { }. The firt ervice i.e. = i voice connection requiring 3 kbp. The econd ervice i.e. = i video connection requiring 64 kbp. The value are et according to the multimedia codec for 3GPP [9]. Thu the QoS proviioning in cell i require that b c = BBU and b c = BBU. In each WLAN k Wi c we need to determine the effective data rate in order to et the network capacity Ck w in 7. In thi paper we ue n- [] imulation to etimate Ck w. The IEEE 8.b [9] i conidered and the nominal data rate i et to Mbp. Two group of contant bit rate CBR ource are being imulated: the firt group with 64 kbp data rate repreenting ervice for voice connection and the econd group with 8 kbp data rate repreenting ervice for video connection. Note that the data rate of the ervice in WLAN are reaonably aumed to be larger than the rate in the wirele cellular ytem in order to benefit from the additional capacity. The meaured aggregate throughput when the number of connection of ervice increae from to and the number of connection of ervice increae from to 6 i hown in Fig.. We can ee that when the number of connection i low the aggregate throughput increae linearly with repect to the increae in the number of ervice and ervice connection. However a the number of uer pae certain threhold the network become aturated and the throughput doe not increae further. The throughput at the aturation point i indeed the effective capacity that the WLAN can upport. From the reult in Fig. we ee that the effective data rate of Ck w i 7.4 Mbp. For the WLAN each BBU i equivalent to 64 kbp. The QoS proviioning in WLAN k require that b w = BBU and b w = BBU. Thu the capacity of each WLAN in 7 i Ck w = 6 BBU. In our tudy we conider different traffic pattern by aigning different value to parameter λ c i and λ w k and auming that λ c i < λ w k. The coverage factor R jk i.5. The connection

8 8 Cot of blocking connection Arrival rate ervice λ Arrival rate ervice λ Cot of blocking connection Cot of dropping a handoff requet α h Fig. 3. Cot of blocking connection veru the arrival rate of new connection requet of ervice type λ and the arrival rate of new connection requet of ervice type λ for optimization problem. duration have mean /υ = /υ = 6 minute. The interboundary time in cell i ha mean /ηi c = minute and the inter-boundary time in WLAN k ha mean /ηk w = 4 minute. For all i M c and all k Wi c the previou value of connection duration and inter-boundary time imply that qi c = qi c =.5 and qk w = qk w =.4 which correpond to a mobility level of 75% for the uer in the wirele cellular network and a lower mobility level of 6% for the uer in the WLAN. For the iterative fixed-point algorithm decribed in Appendix II we ue ɛ = 9. We firt conider cae from Section III-A where the operator offer the ame QoS in term of dropping probabilitie to horizontal and vertical handoff uer. Thu βhh c i m m c i = βvh c i m m c c i. In addition Vi = Ci c bc and Vk w = Ck w bw for all i M c k Wi c and S. Reult for cae will be preented in Section IV-C. To olve the optimization problem 33 and 35 and to obtain the integer olution Ti c and Tk w for all i M c k Wi c and S we ued the exhautive earch. For each of the minimization problem 33 and 35 we conider all four policy combination e.g. CP c -CP w F G c -CP w and obtain the optimal value for each cae. A. Reult for Optimization Problem Fig. 3 how the optimal value obtained from each combined policy for the cot minimization verion of the firt optimization problem i.e. problem 33. We aume that for all i M c k W c i and S αc n i = α w n k = α c hh i = α w hh k = and α c vh i = α w vh k =. In addition we et λ = λ c i and λ w k = σλ c i new connection requet per minute. Since more traffic i generated in a dual cellular/wlan coverage area [4] we aume that σ = 6. When traffic i low i.e. λ <.5 and λ <.5 new connection requet per minute the four policy combination operate quite cloe due to the ufficient network capacity. However a the arrival of connection requet from both ervice increae the performance of thee four combination differ. The lowet blocking cot i achieved by CP c -CP w. Fig. 4. Cot of blocking connection veru the cot of dropping a connection requet from a handoff uer α h for all S for optimization problem. Handoff dropping probability B c hh B c hh B w hh 4 B w hh Arrival rate ervice λ Fig. 5. Handoff dropping probabilitie in cell and WLAN 4 for CP c CP w veru the arrival rate of new connection requet of ervice type λ for optimization problem. Note that Bhh c = B c i vh and B w i hh = B w k vh k for all i M c k Wi c and S. In Fig. 4 we invetigate the cot of blocking connection requet from handoff uer in problem 33. In thi figure α h = αhh c i = αhh w = αc k vh i = αvh w for all i M c k k Wi c and S and α h increae from to 5. Thu the penalty cot i.e. revenue lo incurred by the network operator increae when a handoff connection requet of any type i.e. horizontal or vertical i dropped. Notice that here we have: βhh c i m m c i = βc vh i m m c i and βw hh k m m w k = βvh w mw k m w k. Reult how that the relative ranking of the four combination remain the ame and that CP c -CP w provide the lowet blocking cot. When α h = CP c - CP w provide 4.7% 35.% and 46.6% lower blocking cot than F G c -CP w CP c -F G w and F G c -F G w repectively. Fig. 5 how the dropping probabilitie for handoff connection requet in cell and WLAN 4 for CP c -CP w. The arrival rate of new connection requet for ervice i increaed

9 9 Cot of blocking connection c Mobility in cell i q i Cot of blocking connection Arrival rate ervice λ Arrival rate ervice λ Fig. 6. Cot of blocking connection veru mobility in cell i for all i M c for optimization problem. Fig. 8. Cot of blocking connection veru the arrival rate of new connection requet of ervice type λ and the arrival rate of new connection requet of ervice type λ for optimization problem. Cot of blocking connection blocked and dropped. The increae of the blocking cot can be oberved in Fig. 6. In Fig. 7 the level of mobility in WLAN k for all k Wi c and S given by i.e. qk w i increaed from. to.5 ince the mobility in WLAN i lower than in the cell while in cell i for all i M c and S the level of mobility i fixed at qi c = qi c =.5. There i an increae in the cot of blocking connection a the level of mobility increae. On the other hand when the mobility decreae the blocking cot decreae becaue mot of the connection terminate inide their current cell or WLAN. In both cae the ue of CP in both network achieve the lowet blocking cot w Mobility in WLAN k q k Fig. 7. Cot of blocking connection veru mobility in WLAN k for all k Wi c for optimization problem. while the arrival rate of requet for ervice i fixed at.5 new connection requet per minute. In thi cenario the optimal value are Ti c = 54 Ti c = 54 Ti w = and Ti w = 9 for CP c -CP w for all i M c and k Wi c and correpond to λ =.5 and λ =.5 new connection requet per minute. In Fig. 6 and 7 the level of mobility in the cell of the wirele cellular network and WLAN i increaed repectively. The traffic i et to λ = λ = new connection requet per minute. In Fig. 6 the level of mobility in cell i for all i M c and S given by i.e. qi c i increaed from. to.7 while the mobility in WLAN k for all k Wi c and S i fixed at qk w = qk w =.4. The mobility i increaed by reducing the probability of terminating a connection from ervice in cell i. Thu more horizontal and vertical handoff requet arrive at the adjacent cell and WLAN. Due to the fixed capacity and traffic an increae in connection requet from handoff uer tranlate into more connection being B. Reult for Optimization Problem Fig. 8 how the optimal value obtained from each combined policy for the cot minimization verion of the econd optimization problem i.e. problem 35 with QoS contraint on the dropping probabilitie for connection requet from handoff uer: Γ c hh i = Γ w hh k =. Γ c vh i = Γ w vh k =. Γ c hh i = Γ w hh k =.5 and Γ c vh i = Γ w vh k =.5. We aume that for all i M c k Wi c and S αn c = αw i n k =. A the arrival of new connection requet from both ervice increae the performance of thee four combination differ. The lowet blocking cot i achieved by CP c -CP w which i followed cloely by CP c -F G w. On the other hand the minimum blocking cot of F G c -CP w and F G c -F G w appear to be cloe. Due to the contraint in the handoff dropping probabilitie and lower capacity of the cellular network when compared to WLAN the performance of the wirele cellular network dominate the performance of the integrated cellular/wlan ytem cauing the policy combination uing the ame policy in the wirele cellular network e.g. CP c -CP w and CP c -F G w to have imilar blocking cot; hence cauing policy combination CP c -F G w and F G c -CP w to have different ranking than in the firt problem. However the lowet blocking cot i achieved by CP c -CP w.

10 .6 Handoff dropping probability B c hh B c hh B w hh 4 Γ c hh =Γ w hh Γ c hh =Γ w hh Cot of blocking connection B w hh Arrival rate ervice λ Mobility in cell i q i c Fig. 9. Handoff dropping probabilitie in cell and WLAN 4 for CP c CP w veru the arrival rate of new connection requet of ervice type λ for optimization problem. Note that Bhh c = B c i vh and B w i hh = B w k vh k for all i M c k Wi c and S. Fig. 9 how the probability of dropping connection requet from handoff uer of ervice and in cell and WLAN 4 when the arrival rate of new connection requet from ervice increae. The arrival rate of new connection requet from ervice i fixed at.5. The reult correpond to the combination CP c -CP w which provide the bet performance in term of handoff dropping probabilitie. In thi cenario the optimal value are: Ti c = 6 Ti c = 53 Ti w = and Ti w = for all i M c and k Wi c and correpond to the traffic λ = and λ = new connection requet per minute. Note that in both acce network the dropping probabilitie for ervice are higher due to the fact that each connection requet two time the BBU than ervice. Alo for both ervice the dropping probabilitie are higher in the cell than in the WLAN due to the lower capacity in the wirele cellular network compared to the WLAN which generate value cloely approaching the QoS contraint. In Fig. and the level of mobility in the cell of the wirele cellular network and WLAN i increaed repectively. The traffic i et to λ = λ =.5 new connection requet per minute. In Fig. the level of mobility in cell i for all i M c and S i increaed from. to.7 while the mobility in WLAN k for all k Wi c and S i fixed at qk w = qk w =.4. In Fig. the level of mobility in WLAN k for all k Wi c and S i increaed from. to.5 while the mobility in cell i for all i M c and S i fixed at qi c = qi c =.5. In both cae due to the contraint in the handoff probabilitie the cot of blocking of the four combination are cloe when qi c <.4 and qk w <.4. That i only 4% or le of the uer perform handoff. Thi behavior i different from the firt optimization problem. Finally in both mobility cae the ue of CP in the two network achieve the lowet blocking cot. Fig.. Cot of blocking connection veru mobility in cell i for all i M c for optimization problem. Cot of blocking connection Fig.. k Wi c w Mobility in WLAN k q k Cot of blocking connection veru mobility in WLAN k for all for optimization problem. C. Reult for Handoff Differentiation In thi ubection we conider cae from Section III-A where βhh c mc i m c i βc vh mc i m c i. Fig. how an example of thi cae in which we et βvh c mc i m c c i a 3 with Vi = 58 in cell i for all i M c and S and βvh w k m m w k with Vk w = in WLAN k for all k Wi c and S. Fig. how the probability of dropping connection requet from horizontal and vertical handoff uer of ervice and in cell and WLAN 4 when the arrival rate of new connection requet from ervice i increaed. The arrival rate of new connection requet from ervice i fixed at.5. The arrow depict the difference in probability of dropping connection requet between each type of handoff requet. A an example we can ee that in cell the probability of dropping a horizontal handoff i 8% and 84% lower than the probability of dropping a vertical handoff for ervice and repectively.

11 Handoff dropping probability % 97% 8% 94% B c vh B c vh B c hh B c hh B w vh 4 B w vh 4 B w hh 4 B w hh Arrival rate ervice λ Fig.. Horizontal and vertical handoff dropping probabilitie in cell and WLAN 4 for CP c CP w veru the arrival rate of new connection requet of ervice type λ. V. CONCLUSIONS In thi paper we developed an analytical model to facilitate the performance evaluation and parameter adjutment of different admiion control policie in a multi-ervice integrated cellular/wlan ytem. Our model take into account the mobility and the rate of connection requet of the uer the capacity and the coverage area of each network the admiion control policie the cot from blocking connection requet for each ervice and the QoS requirement in term of blocking and dropping probabilitie. Our work aim to incorporate thee important apect in an optimization-baed deign for connection admiion control in integrated cellular/wlan ytem. Given the model we alo formulate two different revenue maximization problem to adjut the admiion control parameter. The firt problem aim to maximize the network revenue in term of the accepted connection requet from new and handoff uer. The econd problem aim to maximize the network revenue in term of the accepted connection requet from new uer ubject to QoS contraint on the handoff dropping probabilitie. We evaluate four different combination of admiion control policie by extending the cutoff priority and the fractional guard channel admiion control policie with policy function. Reult how that under a wide range of connection requet rate and variou uer mobility level uing the cutoff priority policy in both acce network achieve the bet performance for both deign objective. APPENDIX I IEEE 8. WLAN CAPACITY MODEL The IEEE 8. tandard [9] define two media acce method: the point coordination function PCF and the ditributed coordination function DCF. We conider both cae: a WLAN k i controlled uing PCF. Since the media acce control i centralized in thi cae there i no interference among the tranmiion of different uer. A a reult the uer can fully utilize the available effective capacity. Knowing that the aggregate required data rate to upport connection from all different ervice i equal to S mw k b w contraint 7 i directly reulted. Note that Ck w in thi cae i almot the ame a the WLAN nominal capacity. b WLAN k i controlled uing DCF with carrier ene multiple acce with colliion avoidance CSMA/CA. We aume that the neighboring WLAN operate on different frequency channel o that they do not interfere with each other. Let U denote the et of connected uer of ervice. Alo let f u denote the fraction of time that uer u i active i.e. it tranmit/receive data from the acce point. Since DCF i ditributed uer compete to acce the channel and interfere in each other tranmiion. According to the protocol interference model [3] it i neceary to have f u. 36 S u U To erve uer u U it i required that: b w = f u C w k 37 where f u Ck w denote the data rate achieved by uer u. From 37 we have: f u = bw C w = S u U S k Ck w m w k b w 38 S u U where we ued the fact that U = m w k. Replacing 38 in 36 inequality 7 i obtained. In practice Ck w i ignificantly lower than the nominal capacity when DCF i being ued. In thi paper the effective capacity Ck w i obtained uing imulation a explained in Section IV. APPENDIX II ITERATIVE FIXED-POINT ALGORITHM To compute the blocking and dropping probabilitie for connection requet from new and handoff uer of ervice the following iterative fixed-point algorithm i ued: Input: Specify ɛ >. Input: Set Bn c i = Bhh c i = Bvh c i = and Bn w = Bw k hh k = Bhh w k = for all i M c k Wi c and S. while Bn c + Bc i hh + Bc i vh > ɛ and Bw i n k + Bhh w + Bw k vh > ɛ do k end Solve the ytem of equation of handoff rate given by 3-8. Compute the birth rate φ c i and φ w k given by -. Calculate the blocking and dropping probabilitie Bn c i Bhh c i Bvh c i Bn w k Bhh w k and B w vh k by olving the global balance equation and uing 4-6 and 8-. Update Bn c = Bc i n Bc i hh = Bc i hh i Bvh c i = Bvh c i Bn w = Bw k n Bw k hh k = Bhh w k and Bvh w = Bw k vh k. In thi paper the function B i defined a j B j B j.

12 REFERENCES [] A. Salkintzi Interworking Technique and Architecture for WLAN/3G Integration Toward 4G Mobile Data Network IEEE Wirele Commun. Mag. vol. no. 3 pp. 5 6 June 4. [] M. Buddhikot G. Chandranmenon S. Han Y. Lee S. Miller and L. Salgarelli Integration of 8. and Third-Generation Wirele Data Network in Proc. of IEEE INFOCOM 3 San Francico CA April 3. [3] V. Varma S. Rameh K. Wong and J. Friedhoffer Mobility Management in Integrated UMTS/WLAN Network in Proc. of IEEE ICC 3 Anchorage AK May 3. [4] A. Doufexi E. Tameh A. Nix S. Armour and A. Molina Hotpot Wirele LAN to Enhance the Performance of 3G and Beyond Cellular Network IEEE Commun. Soc. Mag. vol. 4 no. 7 pp July 3. [5] IEEE 8. Working Group [6] 3GPP Requeriment on 3GPP ytem to Wirele Local Area Network interworking TS.34 v8.. March 7. [7] 3GPP 3GPP-WLAN interworking S.R87-A v. February 6. [8] Y. Fang and Y. Zhang Call Admiion Control Scheme and Performance Analyi in Wirele Mobile Network IEEE Tran. Veh. Technol. vol. 5 no. pp March. [9] R. Ramjee D. Towley and R. Nagarajan On Optimal Call Admiion Control in Cellular Network Wirele Network vol. 3 no. pp. 9 4 March 997. [] S. Tang and W. Li Performance Analyi of the 3G Network with Complementary WLAN in Proc. of IEEE GLOBECOM 5 St. Loui MO November 5. [] D. Chen X. Wang and A. Elhakeem Load Sharing with Buffering over Heterogeneou Network in Proc. of IEEE VTC 5 Fall Dalla TX September 5. [] E. Steven-Navarro and V. W. S. Wong Reource Sharing in an Integrated Wirele Cellular/WLAN Sytem in Proc. of Canadian Conference on Electrical and Computer Engineering CCECE 7 Vancouver Canada April 7. [3] W. Song H. Jiang and W. Zhuang Performance Analyi of the WLAN-Firt Scheme in Cellular/WLAN Interworking IEEE Tran. Wirele Commun. vol. 6 no. 5 pp May 7. [4] W. Song Y. Cheng and W. Zhuang Improving Voice and Data Service in Cellular/WLAN Integrated Network by Admiion Control IEEE Tran. Wirele Commun. vol. 6 no. pp November 7. [5] E. Steven-Navarro A. H. Mohenian-Rad and V. W. S. Wong On Optimal Admiion Control for Multi-Service Cellular/WLAN Interworking in Proc. of IEEE GLOBECOM 7 Wahington DC November 7. [6] D. Deniz and N. Mohamed Performance of CAC Strategie for Multimeda Traffic in Wirele Network IEEE J. Select. Area Commun. vol. no. pp December 3. [7] I. Akyildiz and W. Wang A Dynamic Location Management Scheme for Next-generation Multitier PCS Sytem IEEE Tran. Wirele Commun. vol. no. pp January. [8] G. Bolch S. Greiner H. de Meer and K. Trivedi Queueing Network and Markov Chain: Modeling and Performance Evaluation with Computer Science Application. Wiley & Son 998. [9] IEEE 8. Working Group Part : Wirele LAN medium acce control MAC and phyical layer PHY pecification 999. [] G. Bianchi Performance Analyi of the IEEE 8. Ditributed Coordination Function IEEE J. Select. Area Commun. vol. 8 no. 3 pp March. [] Y. Cheng X. Ling W. Song L. Cai W. Zhuang and X. Shen A Cro- Layer Approach for WLAN Voice Capacity Planning IEEE J. Select. Area Commun. vol. 5 no. 4 pp May 7. [] The Network Simulator - n- [3] K. Ro Multiervice Lo Model for Broadband Telecommunication Network. Springer 995. [4] E. Steven-Navarro and V. W. S. Wong Comparion between Vertical Handoff Deciion Algorithm for Heterogeneou Wirele Network in Proc. of IEEE VTC 6 Spring Melbourne Autralia May 6. [5] J. McNair and F. Zhu Vertical Handoff in Fourth-generation Multinetwork Environment IEEE Wirele Commun. Mag. vol. no. 3 pp. 8 5 June 4. [6] E. Steven-Navarro Y. Lin and V. W. S. Wong An MDP-baed Vertical Handoff Deciion Algorithm for Heterogeneou Wirele Network IEEE Tran. Veh. Technol. in pre 8. [7] C. Blum and A. Roli Metaheuritic in Combinatorial Optimization: Overview and Conceptual Comparion ACM Computing Survey vol. 35 no. 3 pp [8] 3GPP Circuit Bearer Service Supported by a Public Land Mobile Network TS. v7.. June 7. [9] Codec for Circuit Switched Multimedia Telephony Service TS 6. v7.. June 6. [3] K. Jain J. Padhye V. Padmanabhan and L. Qiu Impact of Interference on Multi-hop Wirele Network Performance in Proc. of ACM MobiCom 3 San Diego CA September 3. Enrique Steven-Navarro S 99 received the B.Sc. degree from Univeridad Autonoma de San Lui Potoi UASLP San Lui Potoi Mexico in and and the M.Sc. degree from Intituto Tecnologico y de Etudio Superiore de Monterrey ITESM Monterrey Mexico in both in electrical engineering. He i currently working toward the Ph.D. degree with the Department of Electrical and Computer Engineering at the Univerity of Britih Columbia UBC Vancouver BC Canada. From to 3 he wa a Project Manager at Q-Voz IVR Outourcing in Monterrey Mexico. In 3 he became a Lecturer with the Faculty of Science UASLP. Hi reearch interet are in mobility and reource management and admiion control for heterogeneou wirele network. Amir-Hamed Mohenian-Rad S 4 i a Ph.D. candidate in electrical and computer engineering at the Univerity of Britih Columbia UBC Vancouver BC. He received hi B.Sc. degree from Amir-Kabir Univerity of Technology Tehran Iran and hi M.Sc. degree from Sharif Univerity of Technology Tehran Iran 4 both in electrical engineering. From March to July 7 he wa a viiting cholar at Princeton Univerity Princeton NJ. Mr. Mohenian-Rad ha been granted the UBC Graduate Fellowhip a well a the Pacific Century Graduate Scholarhip from the Britih Columbia Provincial Government. He currently erve a TPC member for IEEE International Conference on Communication ICC 9 and IEEE Conumer Communication and Networking Conference CCNC 9. Hi reearch interet are in the area of optimization theory and it application in computer communication and wirele networking. Vincent W.S. Wong SM 7 received the B.Sc. degree from the Univerity of Manitoba Winnipeg MB Canada in 994 the M.A.Sc. degree from the Univerity of Waterloo Waterloo ON Canada in 996 and the Ph.D. degree from the Univerity of Britih Columbia UBC Vancouver BC Canada in. From to he worked a a ytem engineer at PMC-Sierra Inc. He joined the Department of Electrical and Computer Engineering at UBC in and i currently an Aociate Profeor. Hi reearch interet are in reource and mobility management for wirele meh network wirele enor network and heterogeneou wirele network. Dr. Wong i an aociate editor of the IEEE Tranaction on Vehicular Technology and an editor of KICS/IEEE Journal of Communication and Network. He erve a TPC member in variou conference including IEEE Globecom 8 ICC 9 and Infocom 9. He i a enior member of the IEEE and a member of the ACM.