Energy Efficient Relay Selection for Cooperative Relaying in Wireless Multimedia Networks

Transcription

1 Energy Efficient Relay Selection for Cooperative Relaying in Wireless Multimeia Networks Zhengguo Sheng, Jun Fan, Chi Harol Liu, Victor C. M. Leung, Xue Liu*, an Kin K. Leung Orange Labs, France Telecom, Beijing Dept. of Software Service Engineering, Beijing Institute of Technology, China Dept. of Electrical an Computer Engineering, University of British Columbia, Canaa * School of Computer Science, McGill University, Canaa Department of Electrical an Electronic Engineering, Imperial College, UK Abstract In existing wireless networks, supporting multimeia services is becoming more popular an important. In general, wireless multimeia networks shoul require energy efficiency an reliable transmission while keeping satisfactory quality of services. In this respect, cooperative communications have been consiere as an efficient approach to aress these emans by offering significant iversity gains over single antenna systems without increasing requirements on raio resources. In this paper, we propose a power-allocation metho to optimize the ecoean-forwar (DF) cooperative transmission for source an relay noes as a means to reuce the total power consumption, while maintaining the require quality of services, an investigate funamental characteristics of cooperative transmission in terms of power efficiency. Moreover, for a network with multiple cooperative noes, we also propose energy efficient relay-selection rule to offer fairness at each noe an implement it into a practical routing protocol. Our performance analysis is supplemente by simulation results to illustrate the significant energy savings of the propose methos. Inex Terms Relaying, energy efficiency, reliability, QoS, multimeia I. INTRODUCTION Future wireless networks are expecte to support the mixture of real-time applications, such as monitoring [] an multimeia streams [], [3], an non-real-time ata applications, such as web browsing, messaging an file transfers. Compare with wire environments, the associate communication channels an traffic patterns in mobile wireless networks are more unpreictable. Hence all of these applications impose stringent an iversifie Quality-of-Service (QoS) requirements, which cannot be satisfactorily aresse through the traitional communication system. Recently, the availability of low-cost an high processing capability harware which is capable of elivering multimeia content from the environment has fostere the evelopment of wireless multimeia networks [4] [7], i.e., networks of resource-constraine wireless evices that can retrieve an eliver multimeia content such as voice an vieo streams, still images, messaging an file transfer. As a Copyright (c) 3 IEEE. Personal use of this material is permitte. However, permission to use this material for any other purposes must be obtaine from the IEEE by sening a request to pubs-permissions@ieee.org. This work was supporte in part by the Canaian Natural Sciences an Engineering Research (NSERC), the NSERC DIVA Strategic Research Network, an various inustry partners, by the National Natural Science Founation of China (Grants no. 6379). result, it is preicte that wireless multimeia networks shoul maintain both transmission reliability an energy efficiency while keeping satisfactory quality of services. Towars this goal, multiple-input multiple-output (MIMO) has receive significant attention, which can provie spatial iversity an hence represents a powerful technique for interference mitigation an reuction [8], [9]. Although MIMO systems can show their huge benefit in cellular base scenarios, they may face challenges when it comes to their eployment in mobile evices. In particular, the typically small-size of wireless evices makes it impractical to eploy multiple antennas. To overcome this rawback, the concept of cooperative communication mechanisms have been propose as an effective way of exploiting spatial iversity to improve the quality of wireless links [] [4]. The key iea is to have multiple wireless evices in ifferent locations cooperatively share their antenna resources an ai each other s wireless transmission effectively to form virtual an istribute antenna arrays. In cooperative communication, the term cooperation refers to a noe s willingness to share its own resource (e.g., energy, transmission opportunity) for the benefit of other noes. It is thus important to unerstan how much resource must be consume to reap the benefits of the cooperative communication. In our previous work [5], we have shown that it is avantageous to employ cooperative transmission in a network with multiple, mutually cooperative noes, which can significantly reuce the total power consumption while maintaining a given level of quality of service (QoS). However, there is no clear answer about whether cooperative communication requires more (or less) overall resources than conventional, non-cooperative communication to achieve the same level of wireless link quality? If so, how much resource we can best save when employing cooperative communication? What is the impact of each noe s willingness to cooperate on energy efficiency when selfish an unselfish natures are impose to iniviuals? What are the applicable scenarios an how to incorporate the propose solution into practical routing protocol to support multimeia services? This paper is our answers to these questions, with particular focus on the energy consumption issues in cooperative communication to support multimeia services with stringent QoS requirements.

2 S Fig.. R An example of a wireless cooperative link D We consier in this paper the ecoe-an-forwar (DF) cooperative communication. Fig. illustrates the concepts of DF, where the transmission of a source (S) to a estination (D) is aie by a relay (R). While there are variants of cooperative communication, epening on how the relay cooperates to the source s transmission, R s action in DF is to overhear the packet transmitte by S, ecoe it, an retransmit it to D, improving the reception quality of the (combine) signal at D. This cooperative scheme lens itself to a relatively easy implementation in harware an software. Moreover, it is shown in [] that such a cooperative scheme can achieve full secon-orer iversity an therefore provies significant improvement to reception reliability. Specifically, we explore the energy consumption aspect of DF cooperative communications (CC) from various angles. First, we look at how much transmit powers are require for the source an the relay respectively in the cooperative transmission for a given requirement on the link quality. This result is then use to investigate how much an in which case the power can be save by using the cooperative communication, compare to conventional an (non-cooperative) irect communication. Base on these analytical results, we propose the strategy for a resource allocation problem in networks of multiple cooperative noes, namely the energyefficient relay-selection rule for each packet transmission. In orer to stuy the achievable energy savings ue to cooperative communications at a funamental level, we assume throughout the paper that a preefine QoS requirement in terms of the transmission rate an the outage probability is given. The following summarizes our contributions an key results: We erive a close-form solution for the optimal transmission power require by each source an relay noe in DF cooperative communication uner a Rayleigh faing channel moel to achieve the given QoS requirements. Uner the optimal power allocation, our analysis shows that the require transmission power of the relay is always smaller than that of the source, a result which lays a founation to encourage the cooperative behaviors as this means that the helping party (relay) only nees to spen relatively small amount of energy than the one seeking help from others (source). We analyze the power consumption in the optimal cooperation scheme an compare its performance with both irect transmission an conventional cooperation where both source an relay power are consiere as ientical. Specifically, we efine the term of power efficiency an investigate the best relay location to achieve the minimum power consumption as well as erive the boun performance compare with conventional cooperation. By utilizing the optimal results, we propose aaptive relay selection rule that will help select appropriate relays for the fairness an maximal energy saving of each noe in a multi-noe environment an analyze noe s willingness to cooperate when selfish an unselfish natures are impose to iniviuals as well as incorporate them into the implementation of a practical machine-tomachine (MM) routing protocol. Simulation results are supplemente to illustrate the significant energy savings of the propose relays selection rules in proviing reliable services. This paper is organize as follows. The literature review of relate work is given in Section II. The system moel an optimal power consumption are introuce an analyze in Section III. The analysis of the optimal cooperation scheme is presente in Section IV. The energy-efficient relay-selection rules an their practice in the MM routing protocol are propose in Section V an VI, respectively. Simulation results are provie in Section VII. Finally, concluing remarks are given in Section VIII. II. RELATED WORK Cooperative iversity has largely been consiere by physical layer researchers as a low-cost way to achieve spatial iversity. The key feature of cooperative transmission is to encourage multiple single-antenna users/sensors to share their antennas cooperatively. There have been intensive stuies on the physical layer techniques of cooperative communication, we refer the intereste reaer to some state-of-art works [], [] for a preliminary unerstaning of cooperative transmission at physical layer. In general, there are two common approaches to cooperative iversity: amplify-an-forwar (AF) an ecoe-an-forwar (DF) [6]. The first scheme can be viewe as repetition coing from two separate transmitters, except that the relay transmitter amplifies its own receiver noise. While for the secon scheme, the relay fully ecoes an retransmits the receive signal to the estination (an possibly transmits ecoing errors). The estination can employ a variety of combining techniques to achieve iversity gain from cooperation. Due to the enhance performance an relatively easy implementation in harware an software, we explore the ecoe-an-forwar scheme [] in this work. A. Funamental unerstaning of cooperative benefits More recent works in the literature shows that cooperative communication can significantly improve the overall quality of the wireless transmission. Min et al. in [7] consier energy efficient relay selection for two-way relay channel using analog

3 3 network coing. Yinman et al. in [8] examine the symbolerror-rate (SER) performance of ecoe-an-forwar (DF) cooperative communications with multiple Dual-Hop relays over Nakagami-m faing Channels an show that SER performance is significantly improve with channel conitions or faing parameters, because of the increase iversity orer. Vahi et al. in [9] also consier in spectrum sharing systems the error performance of cognitive (seconary) user s communication can be significantly improve by implementing the partial relay selection using DF without affecting the performance of license (primary) users. Meanwhile, Meixia et al. in [] aress an optimization problem involving transmission moe selection (irect or cooperative transmission), relay selection as well as subcarrier assignment to maximize throughput in cooperative OFDMA networks. It shows that the propose algorithm can enhance throughput performance by more than 75% compare to irect transmission. Mohame et al. in [] propose an energy efficient routing protocol for cooperative networks by employing relay clusters along a non-cooperative path an reveal that the propose cooperative transmission protocol can save up to 4% of energy compare with the isjoint-paths an the one-path scheme using only irect transmission. There are also existing works on the analysis of elay an network capacity of wireless networks by using the concept of cooperation. Mao et al. in [] measure queuing elay in two-user cooperation system an the propose scheuling policy is proven to greatly reuce the elay unbalance between users. Also Sanquan et al. in [3] show that the connectivity of wireless networks can be significantly improve through collaboration. Furthermore, the collaborative networks require less power than non-collaborative networks in orer to maintain connectivity of the whole network. B. Cross-layer approach to achieve energy efficiency Although various cooperative transmission schemes have been evelope to increase the banwith efficiency, most of existing literatures focus on physical layer techniques an there is lack of unerstaning of cooperative benefits at upper layers, e.g., routing an applications. In this paper, we take a cross layer approach to tackle cooperative efficiency from physical layer up to network an application layers. Specifically, by proposing efficient relay selection rules, we evaluate the performance of cooperative iversity in wireless networks to satisfy the stringent requirements impose by multimeia services. In the latest work of relay selection, Ritesh et al. in [4] consier to employ a varying subset (an number) of relay noes to cooperatively beamform at any given time to achieve energy efficiency. Ramin et al. in [5] propose an optimization problem to fin a number of relay along the path with the minimum en-to-en outage probability from source to estination. The propose solution requires an optimization over all the paths connecting source-estination subject to a fixe total power constraint. Salama et al. in [3] investigate the performance of the best-relay selection an show that the best-relay selection not only reuces the amount of require resources but also can maintain a full iversity orer. Sung- Rae et al. in [6] also propose a best relay selection scheme to ensure minimum outage probability given a Poisson fiel of relay noes an the presence of path loss an faing, an argue that relays geographically closing to the source an estination are preferre to others. Different to the above literatures, our contribution in this paper is that we consier power efficiency as a main object in relay selection to sure QoS requirements. Moreover, the propose relay selection rules can help achieve the global energy efficiency as well as reliability in wireless multimeia networks. Hence these results will potentially have a broa impact across a range of inustry areas, incluing MM an home automation, etc. III. SYSTEM MODEL AND OPTIMAL DF COOPERATION We consier a cooperative network in Fig. : source noe (S), estination noe (D), an a relay noe (R), which overhears S s transmission to D, retransmits or relays the receive signal to D, improving the reception quality of the (combine) signal at D. Our scheme employs two transmission slots: In the first time slot, the source broacasts its ata to the relay an the estination. In the secon time slot, the relay transmits the signal it receive in the previous time slot, if the SNR excees a threshol; otherwise, the source retransmits the signal. We thus implicitly assume a mini-slot at the beginning of the secon slot uring which ACKs are sent error-free from the relay to the source. Two time slots are use to transmit an relay a given ata signal to avoi RF capture effects when simultaneously transmitting an receiving in the same frequency ban. As a result, the estination receives two inepenent copies of the same packets transmitte through ifferent wireless channels. Diversity gain can be achieve by combining the ata copies using one of a variety of combining techniques, e.g., the Maximum Ratio Combining (MRC) where the receive signals are weighte with respect to their SNR an then summe together. A etaile system assumptions are summarize as follows: The channel moel incorporates both path loss an Rayleigh faing. The channel gain a s, between the noes s an is moelle as a s, = h s, / α/ s,, where h s, captures the channel faing characteristics, s, is the istance between the noes s an an α is the path-loss exponent. The channel faing parameter h s, is assume to be complex Gaussian with zero mean an unit variance, an inepenent an ientically istribute (i.i..) across times slots, packets an across links. b is efine as a esire ata rate in the unit of bit/s/hz, p an q enote the source an relay power, respectively. Aitive white Gaussian noise is also assume with the same variance σn at both relay an estination. Outage probability is efine as the probability that a given ata rate cannot be supporte because of channel variations.

4 4 A QoS is ecie by the target outage probability η an transmission ata rate b, which is require by multimeia service applications. A. Direct transmission We start with irect transmission, the receive signal at a estination is moele as y [n] = a s, x s [n] + n [n] () where x s [n] is the signal transmitte by a source s, n [,..., N] is the inex of the transmitting packet an n [n] is aitive white Gaussian noise, with variance σ n, at the receiver. The channel capacity between the source s an the estination is I s, = log( + p a s, ) () where p = E b /N is efine as the normalize transmission power. Since for Rayleigh faing, a s, is exponentially istribute with parameter α s,. The outage probability satisfies ( ) ɛ out = Pr[I s, < b] = exp (b ) α s, p ( α b ) s, p for large p. Here b is a esire ata rate in the unit of bit/s/hz, which is efine by QoS requirements. We then have the normalize transmission power for irect transmission ( p D = α b ) s,. (4) B. DF cooperative transmission ɛ out Let s,, s,r an r, be the respective istances among the source, relay an estination noe. During the first time slot, the estination an relay receive y, [n] = h s, x s [n] + α/ s, n [n] from the source noe, where x s [n] is the information transmitte by the source an n [n] is white noise. During the secon time slot, the estination noe receives y, [n] = h s, α/ s, h r, α/ r, x s [n] + n [n], x r [n] + n [n], if hs,r < f(p) α/ s,r if hs,r f(p) α/ s,r where f(p) = ( b )/p can be erive from irect transmission an is analogous to (3). In this protocol, the relay transmits only if the SNR excees a threshol; otherwise, the source retransmits in the secon time slot. Assuming that the relay noe can perform perfect ecoing when the receive SNR excees a threshol, the channel capacity of this cooperative link can be shown as { I s, = log( + p a s, ), a s,r < f(p) log( + p a s, + q a r, ), a s,r f(p) (6) (3) (5) where p an q are the normalize source an relay power, respectively. Therefore, the outage event is given by I s, < b an the outage probability becomes ɛ out = Pr[I s, < b] = Pr[ a s,r < f(p)]pr[ a s, < f(p)] +Pr[ a s,r f(p)]pr[ a s, + q p a r, By computing the limit, we obtain from (7) < f(p)]. (7) f ɛout = f Pr[ a s,r < f] f Pr[ a s, < f] }{{}}{{} T T + Pr[ a s,r f] }{{} f Pr[ a s, + q p a r, < f] T3 }{{} T4 where f = f(p), T α s,r, T α s, /, T3, which can be erive from irect transmission an are analogous to (3). Moreover, accoring to the results of Fact an of Appenix I in [], we can erive the cumulative istribution function of the sum of two inepenent exponential ranom variables a s, an a r,, an have the approximate value T4 p q α s, α r, /. Since f(p) = ( b )/p, we obtain a close-form expression for the outage probability between the source an the estination using cooperative transmission (8) ɛ out C = α s,( α s,r + p q α r,) (b ) p. (9) C. Optimal DF transmission power A meaningful optimization problem is to minimize the total transmission power consumption of a cooperative link given that a target Quality-of-service (QoS) is satisfie an can be formulate as min p + q () s.t. ɛ out C (p, q) η p an q enote the source an relay power, respectively, an ɛ out C (p, q) is the outage probability efine by (9), A QoS is ecie by the target outage probability η an transmission ata rate b, which is require by multimeia service applications. Theorem : The optimal transmission power to minimize the total power consumption of DF cooperation given that a target QoS is satisfie, is given by p a + b a + 8ab = +, q = ap () p b where a = µα s, α r, η, b = µα s, α s,r η, µ = ( b ) an η is the outage constraint. Proof : See Appenix A.

5 5 Property : The optimal relay power q is always smaller than the optimal source power p with p > q () p The result follows () an has q = which [+ +8 α s,r /α r, ] is always smaller than p. In general, we fin that the optimal DF cooperation saves the relay power as it moves closer to the estination. Moreover, the optimal cooperation can help reuce the total transmission power, which will be further analyze in the following. IV. ANALYSIS OF OPTIMAL DF COOPERATION In this section, we provie a comprehensive analysis of the optimal DF cooperative transmission. First, we iscuss the best relay location for optimal DF which can achieve the maximum energy saving compare with irect transmission. Then, we analyze its avantage in energy saving, by comparing it with a conventional cooperative scheme. A. Power efficiency factor We introuce a power efficiency factor β that represents the ratio of the total transmission power of cooperative transmission to that of irect transmission: β = p + q p D. (3) Clearly, small values of β are always preferable. Accoring to (4) an (), the power efficiency factor for optimal DF is efine by β = p + q p D = m+ 4 ( α s,r + α r, K α s, m α s,r ), (4) where K = (( b + ) ɛ out ) is the QoS factor, γ = α r, α s,r an m = γ+ γ +8γ. B. Best relay location for optimal DF Result 3: For any relay r which is non-collinear with the source s an estination, we can always fin a mapping relay r on s which achieves a lower total power consumption, given the same target QoS. Proof : As can be seen from Figure, given a relay r which is outsie the line s, we can fin a point r on s as the mapping relay where rr is perpenicular to s. Clearly, we have s,r < s,r, r, < r, an hence, a < a, b < b. From (), we efine f(a, b) = p + q. Since f(a,b) a > an f(a,b) b >, we can obtain p +q = f(a, b) > f(a, b) > f(a, b ) = p + q, which completes the proof. Result 4: For path loss α =, the best relay location that minimizes β for the optimal DF cooperation is at the estination. Proof : From Result 3, we can fin that the relay location which minimizes β for the optimal DF cooperation is surely s r sr, r, sr, r, Fig.. r s, The position of a mapping relay. on the line s, namely s,r + r, = s,. Bring this result into (4), we can obtain the ratio for α = : β = ( ( ) ) m + γ + K 4 m r, s, where γ = α r,, m = γ+ γ +8γ α s,r. Since β, it is easy to observe that the minimum value can be obtaine as K when r, s, =. Note that this result is ifferent from that of cooperative schemes with ientical power [6], [7], where for α > the best relay location for DF cooperation is prove to be the miway between source an estination. C. Iniviual power behaviors Most of recent literature on single relay selections are base on the ientical power assumption for both source an relay, e.g., [5], [6], [8]. To better evaluate the performance of the optimal DF cooperation an compare its performance with existing solutions, we consier this equal power scenario as the conventional cooperative scheme where the source an relay noes always use the ientical transmission power, i.e., p = q = p con. Hence the minimum total power consumption can be erive from (9) as follows: p con = α s, (α s,r + α ) r, )(b ɛ out. (5) C Then, the power efficiency factor is efine by: α s,r + α r, (b + ) ɛ out β = p con p D = α s,. (6) Theorem 5: Given a relay r, the power efficiency of optimal DF (4) is always lower than that of conventional cooperative scheme (6), an we have the boun performance of optimal DF as: β < β < β. (7) ) The lower boun is obtaine when the relay approaches to the estination, where we have the optimal source power p = p con an the performance gain can reach to its maximum with the relay power q own to. ) The upper boun is obtaine when the relay closes to the source noe, where we erive p = q = p con, which means that the optimal cooperation uses the same amount of power as the conventional cooperation.

6 6 3) When the relay goes to infinite or stayin the mile of source an estination, we have β 7 β = 3. Proof : See Appenix B. Result 6: Compare with the transmission power of the conventional cooperative scheme, the optimal source power p is boune by: p con < p <.3p con. (8) Proof : We first erive the upper boun of p p con. It is observe that if (3) can has the maximum value, then its inner term + γ +8γ +γ shoul be maximum. Note that h(γ) = + γ +8γ +γ is a convex function for γ > since h(γ) γ. Hence there will be only one maximum for γ >. Taking the first orer of h(γ), we have the optimal γ to get the maximum h(γ) as γ =. Replacing it in (3), we can obtain the upper boun. From (3), the lower boun performance is achieve when the relay noe closes to the source or the relay noe closes to the estination, i.e., γ = or γ =. In general, Result 6 tells us that the optimal source actually spens more power than that in the conventional cooperation. However, from (3) an Theorem 5, a careful reaer might notice that the optimal cooperation can help significantly reuce the relay power, especially when the relay approaches to the estination. In other wors, slightly increasing the source power can help significantly reuce the relay power an thereafter saving the total energy. D. Average Power Efficiency of DF In this section, we further investigate how much transmission power can be save by using cooperative transmission. We assume that relay caniates are ranomly locate in space accoring to a Poisson point process with ensity λ. A source-estination pair locate at ( s,, ) an ( s,, ), respectively, will choose the best relay noe to achieve the minimum total transmission power among all available relay caniates, where the best relay is the one that results in the best efficiency factor efine in (3). A network with a higher ensity of relay noes can provie better choices for relay selection. We let R be a ranom variable of the selecte relay istance to the estination an r enote the istance between the closest relay an the estination. The probability istribution function of r is given by Pr[R < r] = Pr[R r] = Pr[N r = ] = e λπr (9) where N r is the number of relays within istance r from the estination. The probability ensity function (pf) of the selecte relay istance is Rayleigh istribute f(r) = λπre λπr, r. () We note that relays with the same istance r to the estination may not lea to the same β, since the source-to-relay istances may be ifferent, an hence the optimal p. But we can use the probability istribution function () to boun E [β] as follows. Theorem 7: The average power efficiency of the optimal DF cooperation relative to irect transmission for α = is lower boune by E [β] > e ρ K + K s, 4 + (r s, s, ) f(r)r () where K = (( b + ) ɛ out ), ρ = πλ s, /4. Proof : See Appenix C. It is worth noting that targeting a smaller outage probability can lea to better power efficiency. For any other pathloss exponent α >, the best relay location can only be characterize numerically through (4) an the corresponing average power efficiency is shown by simulation results in Section VII. Comparison with the conventional cooperative scheme: we consier a conventional cooperative scheme where the source an relay noes always use the ientical transmission power, i.e., p = q. Theorem 8 [7]: The average power efficiency of the conventional cooperative scheme (p = q) relative to irect transmission is lower boune by E [β ] > K ( ρ ) α 4 Γ ( α ) () where ρ = πλ s, /4, α is the path-loss exponent an Γ(x) = e t t x t is the gamma function. From the above result, we can conclue that targeting a smaller outage probability, a longer istance or a larger path-loss exponent can lea to better power efficiency, which means that cooperative transmission can better cope with a harsh network environment. It is also worth noting that such conclusion is still vali for the optimal cooperation scheme (which oes not assume p = q). Moreover, the optimal cooperative can achieve much better power efficiency as shown in Section VII-A. V. ENERGY-EFFICIENT COOPERATIVE RELAYING FOR RELIABLE COMMUNICATION In this section, we consier a more general network setting where multiple noes co-exist an cooperate with each other by acting as relays for the transmissions of each other. First, consiering each noe s cooperative state, we propose a weighte aaptive relay selection mechanism for single-hop transmission. Then, we apply the propose relay selection mechanism into a practical MM protocol [9], to facilitate the routing in elivering multimeia services. A. System assumptions Our interest is to fin a strategy that etermines which noe to select as the relay for the maximal power efficiency of each noe in this multi-noe environment. It is worth noting

7 7 that relay selection affects the overall energy transmission, since the optimal DF power epens upon the location of the selecte relay an the channel conitions. When multiple relays are available, we expect the overall energy consumption to ecrease. Our setup consists of a set of noes N = {,..., n}, where each noe i N transmits a number of packets over time, each time with some arbitrary estination noe in the network. For simplicity, we assume all packets have the same constant length with the same QoS constraints, though it is straight-forwar to erive relay selection rules in a more general setup. We also assume that time is ivie into iscrete time slots. TDMA is use to provie collision-free transmissions from the sources an the relays. We enote by p i,j (t) an q i,j (t) the transmit power of a source noe i an a relay noe j, respectively, when i woul use cooperative transmission with j as the relay to some estination at time t. We assume that the source an the relay use the optimal transmission powers given by () for each packet transmission. When noe i uses irect transmission at time t, we enote its transmit power as p D (t). Energy consumption of a noe E i (t : t ) uring a time interval [t : t ] is the sum of noe i s transmit power either as a source or a relay over all t [t, t ] (we assume a noe consumes zero-power at t if it is neither a source or a relay at t). We use R i (t) to enote the set of the noes (except noe i) which can achieve energy saving compare to irection (β < ), for source noe i s transmission to its estination at time t, i.e., R i (t) = {j N {i} p i,j (t) + q i,j (t) < p D (t)}. Result 9: For any time interval of [t, t ], the total energy consumption of the network i N E i(t ) is minimize if each i is assigne a relay noe at each time by the Min-Total- Energy-Selection rule, i.e., r i (t) = arg min j Ri(t) p i,j (t) + q i,j (t). Proof : Since we are only intereste in the total energy consumption, we can scheule the whole transmission into several rouns an each noe can only transmit no more than one packet in each roun. Since any assignment r is injective in each roun, for any two noes i an k, S i S k =, an i N S i N, where S i is a set of source noes whose relay is i. Therefore, the total energy consumption in each roun i N E i = i N (p i,r i + j S i q j,i ) can be re-written as i N p i,r i + i N j S i q j,i = i N p i,r i + j N q j,r j = i N (p i,r i + q i,ri ), which is minimize if each iniviual term p i,ri + q i,ri is minimum. It is worth noting that this result is obtainable from the optimal power allocation metho in (). It is myopic in nature since the selection is base only on the projecte power consumptions of itself an other potential relay noes for the upcoming transmission at each t, but not on the past energy consumptions of itself or other noes. However, it is easy to see that, though simple, the Min-Total-Energy- Selection rule is optimal in the sense that it minimizes the total energy consumption of the network. In other wors, the relay assignments that yiels the minimum total energy consumption can be simply obtaine by having each source noe select a relay noe such that the combine transmission power for the source an the relay is minimum. From the iniviual noes perspective, however, the relay selection can lea to the situation that some noes en up with higher energy consumption than woul be the case when all noes employ irect transmission. This is especially true if some unfortunate noes are heavily selecte as relays an hence consume more energy in relaying than that save from its own transmission as a source. We now consier how to hanle such unfairness issue in CC. B. Weighte aaptive relay selection Our objective is to let each noe act as a relay only when it has save more energy than that it has lost from cooperative transmission in the past an meanwhile best achieve the energy efficiency ue to cooperative transmissions. ) Payoff function: To represent how much energy saving the cooperative transmission can yiel in comparison to irect transmission, we begin by introucing the notion of the payoffs of the noes. The payoff function, u i (t), of noe i at time t is efine as: p D i (t) p i,j(t) if j s.t., r i (t) = j u i (t) = q j,i (t) if i = r j (t) for some source j otherwise. (3) The above represents how much energy a noe i locally saves (or loses) compare to irect transmission at time t, where p D i (t) p i,j(t) enotes the power save from i s cooperative transmission using some relay j at time t, an q j,i (t) the power spent in i s transmission as a relay for some other noe j at time t. In all other cases (if i oes not transmit either as a source or a relay, or if i uses irect transmission), the payoff is. The initial u i (t) can be any arbitrary value, but for simplicity, we assume u i (t) = for all i N. Then the cumulative payoff over a time interval [t : t ] is efine as u i (t : t ) = t τ=t u i (τ), which represents the overall energy savings of a noe uring the time interval. ) Cooperation inex: A binary cooperation inex variable C i (t) is maintaine for each noe i an upate at each time t (hence the term aaptive ) such that if u i ( : t ) C i (t) = ( if u i ( : t ) = for selfish noe ) if u i ( : t ) <. (4) This C i (t) value is use in the ecision as to whether noe i can act as a relay for other noes (when C i (t) =, i.e., in Details on the impact of selfish nature will be iscusse in Section VII-B. We set C i () = in orer to enable the initial cooperative conition when all noes s payoffs are zero. If C i () = for all i, no noe woul cooperate to other noes.

8 8 cooperative moe) or i shoul not be selecte as relay for any other noe (when C i (t) = ). 3) Fairness factor: Recognizing that some noes may benefit more from the larger cooperative transmission opportunities than the others ue to ifference in the amount of ata an to potentially unfair meium access protocol, we introuce the fairness factor to bring the balance (or fairness ) of the amount of payoffs that iniviual noes collect. For example, two relay caniates with ifferent positive payoffs may not be given the same priority to be selecte as relay, since one relay caniate may have gain significant payoff from the past energy consumption, compare with another one with marginal positive payoff. In such a case, the noe with higher payoff shoul give more opportunity to be selecte as relay. How much importance will be given to the fairness term reflecting the payoff an how much to the power consumption term epens on how fast the function w(u) ecays as the payoff value u increases. In our case, we employ a power-law function w(u) = u k (5) where parameter k is a positive constant an can be use to traeoff fairness for energy consumption. In our simulation stuy, we fin that w(u) = u 6 strikes a goo balance. Consiering all above rules, we propose a Weighte Aaptive Relay Selection approach to bring the balance (or fairness) of the amount of payoffs that iniviual noes collect: Weighte Aaptive Relay Selection: A relay is selecte for source i at time t such that r i (t) = arg min {w (u j( : t )) (p i,j (t)+q i,j (t))}, j R i(t),c j(t)= (6) where w(u) = u 6 is a non-increasing function of the payoff value u. The constraint C j (t) = guarantees that a noe whose cumulative payoff is negative will cease to act as a relay, an will be potentially available as a relay when its payoff becomes positive. Here, along with the power consumption factor (p i,j (t) + q i,j (t)), the weight function w(u j ( : t )) is introuce in the relay selection criteria, such that the noes with larger payoffs (i.e., smaller weight) will have a higher chance to get selecte as the relay for each packet transmission. More specifically, among relays which have the same total power consumption, preference will be given to the ones with higher cumulative payoff. VI. PRACTICAL IMPLEMENTATION OF RELAY SELECTION RULE IN SUPPORTING MULTIMEDIA SERVICES In low power an lossy networks (LLNs), mobile evices typically operate with constraine memory, processing power an energy, an their interconnections are typically characterize as unreliable links with high loss rates. RPL [9] is a routing protocol esigne for LLNs, which is a e-facto MM stanar to support multimeia services provie by upper layer protocol, i.e., Constraine Application Protocol (CoAP). RPL is a istance vector routing protocol, in which noes construct a estination oriente Acyclic Graph (DODAG) by Relay table No. Cost C B-C B Fig. 3. A exchange Routing table Relay table No. Cost B C-B An example of builing up relay tables. exchanging istance vectors an root to a controller. Through broacasting routing constraints, the root noe (i.e., central control point) filters out the noes that o not meet the constraints an select the optimum path accoring to the metrics (e.g., hop count, energy cost, latency). In the stable state, each noe has ientifie a stable set of parents an forware packets along the path towars the root of the DODAG. However, the current solution cannot well support multimeia services in wireless networks, for example transmitting images in a multi-hop fashion in a harsh outoor environment to monitor emergency accients may expect high packet loss. Moreover, because of the hierarchical transmission structure, the noes closing to the root may experience more traffic an energy consumption, thus being vulnerable to the energy epletion. To aress these issues an improve the reliability of wireless routing in low power an lossy networks, we incorporate CC into the RPL protocol an propose the cooperationaie routing protocol for lossy networks. Specifically, in the stage of topology formation, each noe shoul maintain two tables: the routing table, a list of parents towar the root; an the relay table, a set of caniate noes that can be serve as the relay between the noe itself an its parents. Each noe buils up its routing table through DIO message. Neighboring noes perioically exchange routing tables to check if they have the same parent. If so, each of them will be selecte as a caniate relay for the other an ae to the corresponing relay table. In this way, the relay table constructs a relay link between both sies where cooperative transmission can be performe. An example of this process is shown in Fig. 3, where noe B an C serve as the caniate relay for each other because they share a common parent noe A. Finally, when a noe transmit its packet towar the root, the next hop is etermine by the routing table. If its relay table is not empty, the noe itself will select one relay from the caniates accoring to the weighte aaptive relay selection rule in V-B an perform the optimal DF cooperative transmission. Therefore, enhance reception reliability an reuce energy consumption are expecte uring the transmission of each hop. VII. SIMULATION RESULTS In this section, we provie the numerical an simulation results obtaine using MATLAB. C

9 9 Normalize Tx power of iniviual noe (B) Direct transmission Conventional cooperation power Optimal source power Optimal relay power Normalize relay position from source Fig. 4. Iniviual power behaviors for α = : Normalize relay position from the source is the ratio of istance between source-relay to the istance between source-estination. if relay is close to the source, if relay is close to the estination. Normalize total Tx power (B) S Y D X Fig. 5. Optimal total power behavior for α = : The best relay location to achieve minimum total power consumption is at the estination. Fig. 6. Average power efficiency Conventional cooperation for α= Conventional cooperation for α=3 Conventional cooperation for α=4 Optimal cooperation for α= Optimal cooperation for α=3 Optimal cooperation for α= Density of relay noes Average power efficiency of the optimal CC an conventional CC. A. Power consumption of optimal DF for single sourceestination We first evaluate the power consumption of CC for a single source-estination pair. Here we set the QoS constraints of the bit rate b = bps/hz, an the target outage probability ɛ out =., an the source an estination noe are place at the coorinates (m, ) an ( m, ) respectively in a -imensional plane. Fig. 4 shows the numerical results for the iniviual powers of the source an the relay in optimal CC as the location of the relay is varie along the line between the source an the estination. Here the x-axis represents the relative location of the relay w.r.t. those of the source an estination, an y axis is the transmit power in B. It can be seen the source can reuce at least 7 B of its power compare to the irect transmission. Moreover, the relay s power is always smaller than the source s, an monotonically ecreases as its location gets closer to the estination. A careful reaer might notice that the optimal source actually spens more power than that in the conventional cooperation. Therefore, slightly increasing the source power can help significantly reuce the relay power an thereafter saving the total energy. Fig. 5 shows the optimal total power consumption as the location of relay is varie in the -imensional plane. The result also confirms Result 4 that the best relay location to achieve minimum total power consumption is at the estination. In Fig. 6, we plot the power efficiency factor of optimal CC an conventional CC, for ifferent path-loss exponent an ensity of the potential relays in the x-axis. We assume that relay caniates are ranomly locate in space accoring to a Poisson point process with ensity λ. The results are average over simulating packet transmissions, an the relay with the smallest p + q is use. The results are consistent with what our analysis preicts: the optimal cooperation consumes less energy than the conventional cooperation (β < β ) with an average improvement larger than %. It also shows that power efficiency improves as more relays are available. This is because it is easier to fin a well-positione relay in a ense network. Finally, we can observe that cooperative transmission achieves better power efficiency with larger pathloss exponents, thus being more helpful in a harsher raio environment. B. Performance of relay selection for multi-noe cooperation In this scenario, we place N (varie between 5 an 5) noes at uniformly ranom locations in a m m region (the eges of the region are wrappe (toroi) to eliminate ege effects). The transmission range of each noe is 4m. Throughout the simulation, we set the path-loss exponent α = 3, the ata rate b = bps/hz an the targete ɛ out =.. A total of 5 N packets are transmitte, an at each time t, a packet is transmitte by a ranomly selecte source an a ranomly selecte estination. The initial payoff value of every noe (u i ()) is set. Fig. 7 shows the average energy consumption per noe, normalize by the minimum value in the ata set (i.e.,

10 Jain's fairness inex Normalize payoff function Normalize average total energy consumption per noe Direct transmission Ranom relay selection Minimum total energy selection Weighte aaptive relay selection Total number of noes Fig Average total energy consumption per noe Minimum total energy selection Weighte aaptive relay selection Total number of noes Fig. 8. Jain s fairness on payoff value. Minimum total energy selection Weighte aaptive relay selection Inices of noes Fig. 9. Normalize payoff value per noe. Minimum total energy selection with 5 noes) for ifferent relay selection methos. The propose relay selection scheme outperforms the irect transmission or the ranom relay selection. However, since the Minimum total energy selection is the optimal solution in this case, the weighte aaptive relay selection performs a bit worse (this is compensate by fairness results below). Furthermore, as the number of noes increases, the average energy consumption of the propose relay selection scheme ecreases, this is because it is easier to fin a well-positione relay an thus save more power. Fig. 8 shows the fairness in terms of how much energy is save for iniviual noes using each relay selection methos, where the y-axis represent Jain s fairness inex of noes cumulative payoffs. 3 It is clear that the weighte aaptive relay selection scheme achieves the best fairness compare with the Minimum total energy selection. Moreover, the inex curve shows a non-ecreasing tenency towar increase total number of noes, this is so because relay selection can be better balance with an increasing number of relay choices in a ense network. As another example to highlight the fairness, we show in Fig. 9 the energy consumptions of iniviual noes at the en of simulation in 5-noe network. It is clear that the weighte aaptive relay selection achieves the best fairness in this example it is the only scheme that ensures that all noes have positive payoff whereas Min-Total-Energy- Selection results in negative payoff for some noe. We aitionally conuct a set of simulations to measure the impact of each noe s willingness to cooperate when its payoff is zero when our aaptive relay selection rule is use. To see this, we slightly change the rule (V-B) for a subset of noes, an ivie the noes into two groups: U = {i C i (t) = if u i (t) = } ( Unselfish group ), an S = {i C i (t) = if u i (t) = } ( Selfish group ); the rule remains the same as (V-B) for both group when u i (t), an run the simulations using aaptive [5] an weighte aaptive relay selection. We expect that the cooperative behaviors of the noes ten to strengthen over time if more noes are in the first group of cooperative noes. In Fig., we show the proportion of the noes with C i (t) = in the y-axis as the time progresses in x-axis. Different curve represents ifferent ratio of T :T, where T = U an T = S with noes in the network. The result is rather surprising: in all cases, the proportion of noes in the cooperative states converges to, even when only one noe cooperates initially to others out of 99. Also, convergence spee is faster with the weighte relay selection. What this result inicates is quite interesting: the cooperative behavior of iniviuals is vital in cooperative communication, an the cooperation among the noes can emerge even faster when combine with some policing mechanism for ensuring fair allocation of resources. It will be an interesting future research topic to formally analyze the emergence of the cooperation in the cooperative 3 Jain s fairness inex is efine by ( U i ) /(N Ui ). The result ranges from (worst case) to (best case). The larger the inex is, the better fairness N that we can achieve.

11 Proportion of noes with positive payoff Aaptive relay selection (T:T=:99) Aaptive relay selection (T:T=5:5) Weighte aaptive selection (T:T=:99) Weighte aaptive selection (T:T=5:5) No. of packets transmitte Fig.. Proportion of noes with positive payoff communication networks, possibly using the concepts an analytical tools in evolutionary game theories an population ynamics [3]. C. Performance of relay selection rule in MM routing In this scenario, we consier a gri network topology for multipoint-to-point simulation, where the esignate root locates at the center of a m m region, with surrouning noes in the peripheral area. As shown in Fig. (a), accoring to the number of hops to the root, the surrouning noes can be ivie into two categories base on the RPL routing protocol, namely the rank noes 5 an the rank noes 6 3. Note that the soli lines are next-hop links an the ashe lines are relay links, as iscusse in Section VI. Throughout the simulation, we set the path-loss exponent α = 3, the ata rate b = bps/hz an the targete ɛ out =.. The initial payoff value of every noe (u i ()) is set. The transmission range is 45m. A total of packets are transmitte to the root from ranom selecte source using RPL routing. Fig. (b) shows the normalize total energy consumption of each noe. Obviously, each noe consumes less energy in the weighte aaptive relay approach compare to the irection transmission. Furthermore, it is worth noting that the performance of lower rank noes (close to the root) with cooperation is close to the performance of higher rank noes without cooperation, which shows that the lower rank noes with heavy traffic actually benefit more from cooperative transmission. Moreover, we can observe that the propose relay selection scheme can fairly istribute the energy consumption among noes with the same rank. This is so because our propose scheme prioritize the selection of relays with larger payoff value, while in turn the relaying transmission reuces its cumulative payoff, thus balancing the opportunity of being selecte among all noes. As a ifferent example, we consier a ranom network topology as shown in Fig. (c), where the root stays at (4, 4) with 9 ranomly locate noes. Without changing the parameters, the routing path is generate by RPL with next-hop links (soli lines) an relay links (ashe lines). To analyze the simulation result, we ivie the 9 noes into two groups: the cooperation group (noe, 3, 4, 6, 7, 9, ) an irect transmission group (noe 5, 8). Fig. () shows the comparison of normalize total energy consumption of cooperation-aie routing with that of using irection transmission. Overall, the noes in cooperation group can significantly reuce their energy consumption because of the help from relaying; for noe 5 an 8, because there is no potential relays nearby, the energy consumption is the same with irect transmission. Furthermore, it is worth noting that the noes especially with heavy traffic loa (e.g., noe an 7) successfully gain a satisfactory level of benefit from cooperation, which is consistent with our results in Fig. (b). VIII. CONCLUSIONS We have shown in this paper that it is avantageous to allocate non-uniform powers to various cooperative transmitters in wireless multimeia networks, which can significantly reuce the total power consumption while maintaining a given level of quality of service (QoS). Specifically, we have propose an optimal power-allocation metho for the ecoe-an-forwar (DF) wireless cooperative networks an investigate its power efficiency. Our analysis shows how the propose DF cooperation outperforms the conventional cooperation an irect transmission. We have also introuce the aaptive relayselection rule that can serve as an effective tool to achieve a esirable traeoff between fairness an energy consumption at each noe, an emonstrate the avantages in practical routing protocols. To evelop further robust cooperative schemes to cope with new emans in future wireless multimeia networks, we plan to explore the performance gain of the cooperative relay-selection methos an propose aitional robust relayselection mechanisms. For example, the aitional mechanisms must be able to consier network scenarios where noes can have ifferent traffic loas, while maintaining a satisfactory egree of fairness. We also plan to consier the life time or the remaining energy of each noe as input parameters to the fairness measure. APPENDIX A Accoring to the Kuhn-Tucker conition (p.44: KKT conitions for convex problems [3]), the inequality constraint in () can be converte to the equality constraint an have the target outage probability Then we obtain α s,( α s,r + p q α r,) (b ) p = η. (7) q = f(p) = ap p b (8) where a = µ α s, α r, /η, b = µα s, α s,r/η, µ = ( b ) an η is the outage constraint.

12 meters meters Rank Rank Root meters (a) A gri network scenario Root Normalize total transmission power consumption Direct transmission Weighte aaptive relay transmission Rank Inices of noes Rank (b) Normalize total transmission power consumption in the gri network scenario. Normalize total transmission power consumption Direct transmission Weighte aaptive relay transmission meters (c) A ranom network scenario. Fig Inices of noes () Normalize total transmission power consumption in the ranom network scenario Simulation results Substituting (8) into p + q, an minimizing wrt p, we have the solution p = a + b a + 8ab ±. (9) To be a vali solution for q, the solution must satisfy p > b. So, we have a unique solution given by p = a + b a + 8ab +. (3) Using this result in (8) leas to (). APPENDIX B From (), we can obtain the power ratio of the optimal source power to the optimal relay power as follows: p q = ( ) >. (3) γ From () an (5), we have the power ratio of the optimal source power to the conventional source power as follows: p α r, + α r, + 8α r, α s,r + α s,r = p con ( α s,r + α r, ) ( = + + ) γ + 8γ, (3) + γ

13 3 Conventional cooperation Power efficiency b= b' K min= K Fig.. S cut b r a D Integration Optimal cooperation Integration from geometric point of view where γ = α r,. Bring (3) into (3), we can erive: α s,r β β = + + γ + 8γ + γ γ Note that (33) is a function of γ. Then, we have: Since g(γ) γ (33) β β = g(γ), γ >. (34) >, we can get the boun of g(γ) as: g min (γ) = lim g(γ) = γ, g max (γ) = lim g(γ) =. (35) γ Therefore, we have the result in (7). Aitional comments to the boun performance are explaine as follows: ) Lower boun: This is obtaine by putting γ = in (3) an (3). It is worth noting that the optimal relay power highly epens on the relay location an has q, if r,. (36) In that case, the whole receiver sie can actually be treate as a MIMO antenna system with the relay an estination combine together. ) Upper boun: This is obtaine by putting γ = in (3) an (3). 3) The result β (3) an (3). β = 7 3 is obtaine by putting γ = in infinity. r is the selecte relay istance to the estination. Since the minimum β at point a is smaller than the minimum β at point b, when β increases to point b, both β an β have the same power efficiency on the same cut. It is worth noting that the two circles on the same cut are the set of relay locations that achieve the same power efficiency for the conventional cooperation an the optimal cooperation, respectively. Therefore, we have the expecte power efficiency > > E[β] = s, s, βf(r)r βf(r)r + s, K f(r)r + K = e ρ K + K s, β f(r)r s, 4 + (r s, 4 + (r s, s, ) s, ) f(r)r f(r)r. (37) Zhengguo Sheng is a Research Associate at The university of British Columbia, Canaa. His current research projects cover machine-to-machine (MM), mobile clou computing, vehicular communications an wireless sensor networks. He was previously with France Telecom Orange Labs as the senior researcher an project manager in MM an Internetof-Things, as well as the coorinator of Orange an Asia telco on NFC-SWP partnership. He is also the winner of Orange Outstaning Researcher Awar an CEO Retention bonus recipient,. He also worke as a research intern with IBM T. J. Watson research centre USA an US Army Research labs. Before joining Orange Labs, he receive his Ph.D an M.S. egrees with istinction at Imperial College Lonon in an 7, respectively, an B.Sc. egree from University of Electronic Science an Technology of China (UESTC) in 6. He has publishe more than technical papers in international journals an conference proceeings, co-eite 3 books an stanarization contributions in OMA an OneMM. He is a member of the eitorial boars of ELSEVIER Journals of Computer Communications (COMCOM), co-organizer of IEEE WiVeC4, an technical program committee member of Qshine 4, ContextDD 4. Jun Fan receive his Master Degree from the Department of Computer Science at Beijing Institute of Technology, China. His research interests inclue cooperative communication, vehicular a hoc networks an the Internet-of-Things (IoT). He is currently a software engineer at Baiu Corp. APPENDIX C Proof of Theorem 7: Fig. illustrates the integration metho of E[β], the integration is performe from the best relay location at the bottom with the minimum value K to

14 4 Chi Harol Liu [S 5, M ] is an Associate Professor an Department Hea of Software Service Engineering at Beijing Institute of Technology, China. He hols a Ph.D. egree from Imperial College, U.K., an a B.Eng. egree from Tsinghua University, China. Before moving to acaemia, he joine IBM Research - China as a staff researcher an project manager, after working as a Postoctoral researcher at Deutsche Telekom Laboratories, Germany, an a visiting scholar at IBM T.J. Watson Research Center, USA. His current research interests inclue the Internet-of-Things (IoT), big ata analytics, mobile computing, an wireless a-hoc, sensor an mesh networks. He receives the Distinguishe Young Scholar Awar in 3, IBM First Plateau Invention Achievement Awar in, IBM First Patent Application Awar in, an interviewe by EEWeb.com as the Feature Engineer in. He has publishe more than 3 prestigious conference an journal papers, an owne a few EU/US/China patents. He serves as the eitor for KSII Trans. on Internet an Information Systems, an the book eitor for two books publishe by Taylor & Francis Group, USA. He also has serve as the General Chair of IEEE SECON3 workshop on IoT Networking an Control, IEEE WCNC workshop on IoT Enabling Technologies, an ACM UbiComp Workshop on Networking an Object Memories for IoT. He is a member of IEEE an ACM. Xue Liu is an Associate Professor in the School of Computer Science at McGill University. He worke as an Associate Professor an hel the Samuel R. Thompson Chair in the Department of Computer Science & Engineering at University of Nebraska- Lincoln in. He obtaine his Ph.D. in Computer Science from the University of Illinois at Urbana- Champaign in 6. He obtaine his B.S. egree in Mathematics an M.S. egree in Automatic Control both from Tsinghua University, China. He ha also worke at Hewlett-Packar Labs in Palo Alto, California an IBM T. J. Watson Research Center in Hawthorne, New York. He has been grante US patents an file other US patents, an publishe more than 5 research papers in major International acaemic journals an peerreviewe conference proceeings, incluing the Best Paper Awar from IEEE Transactions on Inustrial Informatics for year 8, an the First Place Best Paper Awar from the Fourth ACM Conference on Wireless Network Security (ACM WiSec ) in. He serve chair or technical program committee positions for 7+ acaemic conferences, is a reviewer for 3+ journals an 8+ conferences/workshops. Dr. Liu serve as associate eitor for IEEE Transactions on Parallel an Distribute Systems, an IEEE Communications Surveys an Tutorial. He also serve on various national an international grant review panels. He currently serves as the chair of The Action Group on Hybri Dynamic Systems of IEEE Technical Committee on Computational Aspects of Control Systems Design. Victor C. M. Leung [S 75, M 89, SM 97, F 3] receive the B.A.Sc. (Hons.) egree in electrical engineering from the University of British Columbia (UBC) in 977, an was aware the APEBC Gol Meal as the hea of the grauating class in the Faculty of Applie Science. He attene grauate school at UBC on a Natural Sciences an Engineering Research Council Postgrauate Scholarship an complete the Ph.D. egree in electrical engineering in 98. From 98 to 987, Dr. Leung was a Senior Member of Technical Staff an satellite system specialist at MPR Teltech Lt., Canaa. In 988, he was a Lecturer in the Department of Electronics at the Chinese University of Hong Kong. He returne to UBC as a faculty member in 989, an currently hols the positions of Professor an TELUS Mobility Research Chair in Avance Telecommunications Engineering in the Department of Electrical an Computer Engineering. Dr. Leung has co-authore more than 7 technical papers in international journals an conference proceeings, 9 book chapters, an co-eite 8 book titles. Several of his papers ha been selecte for best paper awars. His research interests are in the areas wireless networks an mobile systems. Dr. Leung is a registere professional engineer in the Province of British Columbia, Canaa. He is a Fellow of IEEE, the Royal Society of Canaa, the Engineering Institute of Canaa, an the Canaian Acaemy of Engineering. He was a Distinguishe Lecturer of the IEEE Communications Society. He is a member of the eitorial boars of the IEEE Wireless Communications Letters, Computer Communications, an several other journals, an has previously serve on the eitorial boars of the IEEE Journal on Selecte Areas in Communications - Wireless Communications Series, IEEE Transactions on Wireless Communications, IEEE Transactions on Vehicular Technology, IEEE Transactions on Computers, an Journal of Communications an Networks. He has guest-eite many journal special issues, an contribute to the organizing committees an technical program committees of numerous conferences an workshops. He receive the IEEE Vancouver Section Centennial Awar an UBC Killam Research Prize. Kin K. Leung receive his B.S. egree (with firstclass honors) from the Chinese University of Hong Kong in 98, an his M.S. an Ph.D. egrees in computer science from University of California, Los Angeles, in 98 an 985, respectively. He starte his career at AT&T Bell Labs in 986 an worke at its successor companies, AT&T Labs an Bell Labs of Lucent Technologies, until 4. Since then, he has been the Tanaka Chair Professor in Internet Technology at Imperial College in Lonon. His research interests inclue network resource allocation, MAC protocol, TCP/IP protocol, istribute optimization algorithms, mobility management, network architecture, real-time applications an teletraffic issues for broaban wireless networks, wireless sensor an ahoc networks. He is also intereste in a wie variety of wireless technologies, incluing IEEE 8., 8.6, an 3G an future generation cellular networks.