Coalition Formation of Vehicular Users for Bandwidth Sharing in Vehicle-to-Roadside Communications

Transcription

1 Coalition Formation of Vehicular Users for Bandwidth Sharing in Vehicle-to-Roadside Communications Dusit Niyato, Ping Wang, Walid Saad, and Are Hørungnes School of Computer Engineering, Nanyang Technological University (NTU), Singapore UNIK - University Graduate Center, University of Oslo, Keller, Norway Abstract In vehicular-to-roadside (V2R) communications of vehicular network, the bandwidth from roadside unit (RSU) can be shared among vehicular users to improve the utilization and reduce the cost. In this paper, the coalition game is formulated to model the situation in which multiple vehicular users can cooperate to share the bandwidth to RSU. First, the rational coalition formation is considered in which each vehicular user has self-interest to form coalitions so that the individual utility is maximized. The dynamic model of such a formation is presented to obtain the stable coalition collection. The distributed algorithm based on split and merge mechanisms is also proposed. Then, the optimal coalition formation is considered in which the coalitions is formed so that the social welfare of all vehicular users is maximized. The performance evaluation shows that the optimal coalition formation yields higher utility than that of rational coalition due to the group-interest of all vehicular users. Also, both optimal and rational coalition formations achieve much higher utility than that without bandwidth sharing. Keywords-Intelligent Transportation Systems (ITS), vehicleto-roadside (V2R) communications, coalition game. I. INTRODUCTION Vehicle-to-roadside (V2R) communication based on wireless technologies will be the key to support various intelligent transportation system (ITS) applications. In V2R communications, vehicular users reserve bandwidth of wireless roadside units (RSUs) such as base stations or access points from the network service provider. This bandwidth is used to transfer the data (e.g., road traffic information and infotainment data) of vehicular users [1], [2]. To efficiently utilize the reserved bandwidth from RSUs, multiple vehicular users can share this bandwidth to reduce the cost. However, depending on the mobility pattern and wireless channel, quality-of-service (QoS) performance of vehicular users may be degraded if the bandwidth is poorly shared (e.g., many users access bandwidth at the same RSU). Therefore, rational vehicular users will seek for the best cooperation among each other to minimize the cost of bandwidth reservation while QoS performance is maintained at the target level. The problem of bandwidth sharing for vehicular users in V2R-based vehicular network can be formulated as the coalition game to obtain stable or optimal coalition structure. We first consider the case that the vehicular user has selfinterest to maximize individual payoff (i.e., utility). This is referred to as the rational coalition formation in which the dynamic model based on Markov chain can be developed to analyze the behavior of the vehicular users. Specifically, the stable coalition collection (i.e., the formation in which none of the vehicular users can improve the payoff by changing coalition formation decision) can be obtained analytically. Also, given the incomplete information of the network, distributed learning algorithm based on split and merge mechanisms is also presented to implement the rational coalition formation. In addition, we consider the optimal coalition formation in which all vehicular users make decision of coalition formation so that the social welfare (i.e., total utility) is maximized. With analytical model and distributed algorithm, the performance evaluation shows that the coalition formation of vehicular users in vehicular network can improve the utilization of the bandwidth while the QoS requirements can be met. The rest of this paper is organized as follows. The related work are reviewed in Section II. Section III describes the system model and assumptions. The coalition game formulation of vehicular users is presented in Section IV. Section V presents the performance evaluation results. The summary is given in Section VI. II. RELATED WORK In [3], different wireless technologies were adopted in V2R communications to support safety and non-safety ITS applications. The coverage and capacity requirements were analyzed for digital broadcasting systems, Universal Mobile Telecommunications System (UMTS), and dedicated short range communication (DSRC) systems. The vehicular users can cooperate by relaying packet of other user so that the throughput of data transmission can be improved [4]. In [5], the proxy-based vehicles to RSU access (PVR) was proposed to support collaborative and opportunistic packet forwarding in V2R communications. The efficient protocol was also introduced. The problem of bandwidth allocation in V2R communications was studied based the auction theory [6]. The amount of bandwidth is determined by the bids from different vehicles. Noncooperative game model was formulated and solved to obtain Nash equilibrium for the bidding strategy. However, none of the work in the vehicular network literature considered the cooperation among vehicular users for bandwidth sharing of V2R communications. Coalition game was applied to model cooperation behavior of the wireless users. For example, in [7], coalition game for cooperative sensing was proposed to obtain the stable group of secondary users to detect the primary users. In [8], cooperation among service provider to share the resources (e.g., spectrum, and base station) is analyzed using transferable payoff coalition game model. It was shown that the

2 2 optimal cooperation can be obtained as the solution of convex optimization. However, the cooperation behavior of vehicular users to share the bandwidth in vehicular network with V2R communications has never been studied. III. SYSTEM MODEL AND ASSUMPTIONS We consider a vehicular network with V2R communications. In the service area, the set of roadside units (RSUs) is denoted by R. There are N vehicular users communicating with RSUs in this service area. An example of service area with 4 users (specifically 4 vehicles) is shown in Fig. 1. To download 1 data from the infotainment server (e.g., video streaming), bandwidth of RSU k R can be reserved with the cost of C k per unit of bandwidth per unit of time. The set of RSUs whose bandwidth is reserved by vehicular user i is denoted by R i R. Vehicular users can form a coalition to share the bandwidth and the cost will be equally divided among them. In the same coalition, vehicular users can access the bandwidth at different RSUs simultaneously. However, if these vehicular users are at the same RSU, the bandwidth will be randomly assigned to one of users with equal probability. For example, in Fig. 1, users 1 and 2 are in the same coalition, i.e., S = {1, 2} to share bandwidth at RSU1. User 1 will use the bandwidth if user 2 is not at RSU1, and vice versa. However, if both users 1 and 2 are at RSU1, each of them will use the bandwidth with probability 0.5. From this example, users 1 and 2 can reduce the cost of bandwidth reservation at RSU1 by half. However, if users 1 and 2 have high chance to be at the same RSU1, the QoS performance may be degraded due to the bandwidth sharing. As a result, they may deviate from coalition S. This conflict situation will be modeled as the coalition game in this paper. RSU 1 RSU 2 RSU 3 Vehicle 1 RSU 4 RSU 5 RSU 6 Vehicle 2 not influence or control the decision of each individual user. This decision depends specifically on the received benefit (i.e., utility) of the user. A. Utility of Vehicular User For downlink V2R communications, there is the buffer to store data packet to be delivered to vehicular user i through the transmission of RSU k R i. To obtain performance measures (e.g., average number of packets in queue, packet loss probability, average packet delay, and throughput), the queueing model similar to that in [9] can be used given the mobility model of vehicular user, traffic arrival parameters, channel quality, and the coalition of the users. We consider wireless transmission with adaptive modulation and coding (e.g., as in DSRC [10]). Let p r (k) denote the probability of using transmission rate r at RSU k for r = 0,..., r max with totally r max modes. Given the channel quality (i.e., average SNR), this probability determines the departure process of the queueing model [11]. When there are n users from the same coalition at the same RSU k, this transmission probability becomes p r (k, n) = p r(k) (1) n for r = 1,..., r max and p 0 (k, n) = 1 r max r =1 p r (k, n). With bandwidth sharing in the coalition S, the utility of vehicular user given the queueing performance (i.e., throughput) Q i (S) can be defined as follows: φ i (S) = Q i (S) k R i C k (S) P i (Q(S), q thr ) (2) for i S and k R i. C k (S) = C k Ŝi,k is the cost of reserving bandwidth at RSU k per user. Ŝ i,k S is the subcoalition whose member shares bandwidth with user i at RSU k. P i (Q(S), q thr ) is the penalty cost function due to performance degradation below threshold q thr. This penalty cost function can be defined as follows: { wpen, Q(S) < q P i (Q(S), q thr ) = thr (3) 0, otherwise. where w pen is penalty weight. Note that for the packet loss probability and average delay, the utility function can be defined similar to that in (2) and (3). RSU 7 RSU 8 RSU 9 Vehicle 4 Vehicle 3 Fig. 1. System model for the vehicular network with coalition formation among vehicular users to share the bandwidth. To facilitate the coalition formation among vehicular users, we assume that there is the coordinator at the application server. This coordinator provides information of active vehicular users. Users perform coalition oining or partitioning through this coordinator. Nonetheless, this coordinator does 1 The model presented in this paper is also applicable to an uplink communications. IV. GAME FORMULATION OF VEHICULAR USER COALITION FORMATION Obective of vehicular users is to form the coalition among each other to share the bandwidth such that the they gain benefit (i.e., utility) higher than that without sharing. In this section, first the coalition game model for rational coalition formation with self-interested vehicular user is presented. In this case, the user is rational to form coalition such that the individual utility is maximized given the decision of other users. The distributed algorithm implementation of ration coalition formation based on the concept of merge and split is presented. Then, the optimal coalition formation is also introduced to maximize the social welfare (i.e., total utility of all users).

3 3 A. Rational Coalition Formulation Coalition game model is formulated whose rational players are the vehicular users. The set of players is denoted by N = {1,..., N}. Coalition denoted by S is a subset of N, and N is the grand coalition (i.e., coalition of all players). The strategy of player is to form the coalition, and payoff is the utility as defined in (2). Characteristic function v(s) determines the value of coalition S which can be defined as follows: v(s) = i S φ i (S) (4) and v( ) = 0. Collection is a group of coalitions of all users. The coalition collection is defined as ω = {S 1,..., S,..., S s } where S S = for, s =1 S = N, and s is the total number of coalitions in collection ω. The total number of possible collections for N users is denoted by D N where n 1 ( n 1 D n = =1 ) D (5) for n 1 and D 0 = 1. We consider non-transferable utility coalition game in which the utility of vehicular user defined as function of queueing performance cannot be arbitrarily apportioned among the users in a coalition. The solution of coalition game is the stable collection which can be defined based on internal and external stability as follows. Internal stability: If coalition S is internal stable, there is no user can improve its payoff by leaving its coalition, i.e., φ i (S) v({i}) = φ i ({i}). External stability: If coalition S is external stable, there is no other coalition R can improve the payoff by oining coalition S, i.e., v(r) > v(s R) v(s) for R, S N and R S =. The collection is stable if the conditions of internal and external stability hold. The probabilities of vehicular user to oin or leave any coalition are determined from the received utility given the decision. The decision of user i can be made among the following choices. Partitioning: User will individually decide to form a singleton coalition if the following condition is satisfied φ i ({i}) > φ i ( ) or φ i ( {i}) > φ i ( ) i, i i (6) where is a coalition which user i is a member. In general, given original coalition S, users in this coalition can collectively form (i.e., be partitioned into) multiple new coalitions whose set is denoted by M if the following condition is satisfied φ i ( ) > φ i( ) i =. (7) M Vehicular users agree to form multiple coalitions if the payoffs of all users is higher than that in one coalition. Joining: Otherwise, vehicular user i in a singleton coalition (i.e., {i}) will individually oin coalition S if the following condition is satisfied φ i ({i} S ) > φ i ({i}), and φ i ({i} S ) > φ i (S ) (8) for i S. In particular, to oin any coalition, not only the payoff of user i must be higher, but also the payoff of other users i in the target coalition S must be higher. In general, multiple coalitions S in set M can collectively oin each other become new coalition, if the following condition is satisfied φ i ( ) > φ i( ) and φ i ( ) > φ i (S(i) S(i) ). (9) for i and i S(i) where = M S(i). In particular, multiple coalitions will oin each other if all users in all candidate coalitions receive higher payoff. Depended on the current coalition collection at time t denoted by ω(t) = {S 1 (t),..., (t),..., S s (t)}, the new coalition of user i is formed to maximize immediate payoff (i.e., myopic policy) as follows:, φ i( ) > φ i( ) and φ i ( ) > φ i( (t)) (t + 1) =, φ i( ) > φ i( ) and (10) φ i ( ) > φ i( (t)) (t), otherwise. To obtain the stable coalition collection, the dynamic model based on discrete-time Markov chain of coalition game can be formulated [12]. The state space of this Markov chain is defined as follows: = {(ω x ); x = {1,..., D N }} (11) where ω x represents the coalition collection of all vehicular users and D N can be obtained from (5). The probability transition matrix of this Markov chain is denoted by P whose element is P ω,ω. This element P ω,ω represents the probability that the coalition collection (i.e., state) of all users changes from ω to ω during a certain time interval. This time interval can correspond to the transmission slot of the RSU. Let C ω,ω denote the set of candidate users which can make decision and result in the change of collection from ω to ω. The transition probability can be obtained as follows: { i Cω,ω P ω,ω = γβ i(ω ω)(1 γ) N C ω,ω, ω ω 0, otherwise (12) where C ω,ω is the cardinality of set C ω,ω. ω ω is the reachability condition. In particular, if collection ω is reachable from ω given the decision of all users defined in (10), then the condition ω ω is true and false otherwise. γ is the probability that the user makes decision in a time interval. β i (ω ω) is the best-reply rule of user i. That is, β i (ω ω) is the probability that the decision of user i changes the current coalition from ω to ω. This best-reply rule can be defined as follows: { ˆβ, β i (ω φi ( ω) = ω ) > φ i ( ω) (13) ɛ, otherwise

4 4 where 0 < ˆβ 1 is the constant, ɛ is the small probability of irrational decision, e.g., ɛ = 0.01, and ˆβ ɛ. The irrational decision can be made by the user due to the lack of information or due to the exploration in learning process. In this case, the state transition probability P ω,ω is determined from the product of transition probabilities of users which do and do not make decisions in a time interval. With nonzero probability of irrational decision (i.e., ɛ > 0), the stationary probability of the Markov chain can be obtained by solving π T P = π T and π T 1 = 1 where 1 is a vector of ones. π = [ ] T π ω1 π ωx π ωdn is a vector of stationary probabilities. π ω is the probability that the collection ω will be formed. The average individual utility of user i can be obtained from D N U i = π ωx φ i ( ) for ω x. (14) x=1 Then, let the probability of irrational decision approach zero (i.e., ɛ 0). In the Markov chain defined by state space in (11) and transition probability in (12), there could be the ergodic set E of states if P ω,ω = 0 for ω E and ω / E, and no nonempty proper subset of E has this property. Also, singleton ergodic sets are referred to as the absorbing states. Specifically, ω is absorbing if P ω,ω = 1. Ergodic set is important for the coalition formation since once all vehicular users reach the state in ergodic set, they will remain in ergodic set forever. In addition, users will stop evolving once the absorbing state is reached. The absorbing state is referred to as the stable coalition collection ω. In particular, no user has an incentive to change its decision given the prevailing coalition collection. PROPERTY 1: The coalition game (N, v) of bandwidth sharing among vehicular users has at least one absorbing state. Proof: We use Theorem 1 in [12] to prove this property. Specifically, the Markov chain of coalition formation process with best-reply rule possesses at least on absorbing state if for all coalitions S the condition i S v({i}) v(s) holds. Based on the value of the coalition for bandwidth sharing defined in (4), it can be shown that i S v({i}) = i S φ i({i}) = v(s). Therefore, there exists at least one absorbing state or equivalently the coalition game of bandwidth sharing has at least one stable coalition collection ω. B. Distributed Algorithm To obtain the stable coalition collection for vehicular users sharing the bandwidth from RSUs, distributed algorithm based on split and merge mechanisms [7] is presented in Algorithm 1. This algorithm is based on the reinforcement learning in which the user gradually learns the benefit of different coalition. The decision is made by vehicular user based on the knowledge κ i ( ) to maximize the payoff. Note that 0 rand() 1 is random number generator. C. Optimal Coalition Formation Vehicular users can cooperate to form optimal coalition so that the social welfare (i.e., total utility) is maximized, for Algorithm 1 Distributed coalition formation algorithm for bandwidth sharing in vehicular network. 1: t = 0 and vehicular users are partitioned into {S 1 (t),..., S s (t)} for s =1 S = N 2: loop 3: Vehicular user i observes the queueing performance Q i ( (t)) and cost k R i C k ( (t)) given the current coalition (t) 4: Compute utility φ i ( (t)) 5: Update the knowledge of the utility, i.e., κ i ( (t)) = α i φ i ( (t)) + (1 α i )κ i ( (t)) where 0 < α i < 1 is the learning rate Merge mechanism 6: if κ i ( M S (t)) > κ i ( (t)) for i t M S (t), t where M t is a set of coalitions to be merged at time t then 7: Merge coalitions S (t) for M t 8: else 9: if rand() ɛ then 10: Merge coalitions S (t) for M t 11: end if 12: end if Split mechanism 13: if κ i (S ) > κ i ( (t)) for i (t), M t where M t is a set of coalitions split from (t) at time t then 14: Split coalition (t) into S for M 15: else 16: if rand() ɛ then 17: Split coalition (t) into S for M 18: end if 19: end if 20: t = t : end loop example, when all users are belonged to the same public transportation service provider. The optimal coalition collection is defined as follows: ω = {S 1,..., S,..., S s } = arg max ω i N φ i ( ). (15) Again, to obtain the optimal collection, the Markov chain model can be formulated. The state space is the same as that defined in (11). The state transition P ω,ω can be defined similarly to that in (12). However, the best-reply rule β i ( ) becomes { ˆβ, β i (ω V (ω ω) = ) > V (ω) (16) ɛ, otherwise where V (ω) = φ i ( (17) ω i ) is the total utility of all vehicular users. Again, the stationary probability π ωx can be obtained. The total utility of all users

5 5 can be obtained from D N U total = A. Parameter Setting x=1 π ωx v(s ). (18) S ω x V. PERFORMANCE EVALUATION We consider the V2R-based vehicular network as shown in Fig. 1. There are 9 RSUs, and 4 vehicular users. Vehicular users visit the RSUs randomly as follows R 1 = {RSU1,..., RSU6}, R 2 = {RSU4,..., RSU9}, R 3 = R 4 = {RSU1,..., RSU4, RSU6,..., RSU9}. The average channel quality of the connection between vehicular user and RSU is 14dB. Utility function of the user is defined based on the queue throughput Q i (S). The threshold of minimum throughput is q thr = 1.2 packets/time slot. w pen = 5 for penalty cost function. The probability of irrational decision is ɛ = The parameters of best-reply rule are ˆβ = γ = 0.9. The distance between RSUs is 500 meters, and the average speed of vehicular user is 50km/h. B. Numerical Results ω 10 {1,2},{3},{4} ω 12 {1,2},{3,4} {1,3},{2},{4} ω 2 ω {1,2,4},{3} 7 ω 1 {1,4},{2},{3} {1},{2},{3},{4} {1,3},{2,4} ω 3 Fig. 2. ω 13 {2,3},{1},{4} ω 11 {2,4},{1},{3} {3,4},{1},{2} ω 9 {1,2,3},{4} ω 14 ω 15 {1,3,4},{2} ω {1,4},{2,3} 4 {1},{2,3,4} State transition of the coalition among 4 vehicular users. ω 8 ω 6 ω 5 {1,2,3,4} Fig. 2 shows the state transition diagram of the coalition formation among 4 vehicular users. There are totally 15 states with different collections. Users can form the coalition, where in the first stage there will be three coalitions (e.g., {1, 2}, {3}, {4} as state w 10 in Fig. 2). Then in the second stage, each individual user can oin other coalition and the number of coalitions becomes 2 (e.g., {1, 2, 3}, {4}). In this second stages, some coalition can be partitioned, while some can oin each other and become grand coalition (i.e., {1, 2, 3, 4} as state ω 8 in Fig. 2). Fig. 3 shows the stationary probability of the coalition collection for both optimal and rational formations. Note that the label of coalition state is shown in Fig. 2. As expected, there is a set of coalition states (i.e., ergodic states). These sets of states obtained for optimal formation and rational formation are different, and these sets of states are also different when the mobility parameters (e.g., speed of vehicular user) are varied. When the speed of vehicular user is uniform (i.e., 50km/h in average), the optimal coalition collection is ω = {{1, 2}, {3, 4} since vehicular users 1 and 2 as well as users 3 and 4 have the smallest chance to share the bandwidth Fig. 3. Stationary probability Stationary probability Coalition state (speed of vehicular user at RSU5 is 30km/h) Coalition state (speed of vehicular user at RSU5 is 50km/h) Stationary probability of coalition state. at the same RSU. Therefore, the utility is the highest and this coalition collection is preferred by all users. However, as the speed of vehicular user decreases (e.g., due to traffic am) at RSU5, it is beneficial for users to form different coalition to gain the highest total utility. In this case, the optimal coalition becomes {{1, 4}, {2, 3}}. Due to traffic am at RSU5, users 1 and 2 have high chance to be at RSU5. Therefore, users 1 and 2 will establish the new coalitions with users 4 and 3, respectively to maximize total utility. Note that the stationary probability of the suboptimal coalition can be non-zero due to the irrational decision of the users. For rational coalition formation, the set of coalition collections with non-zero stationary probability is generally different from that of optimal formation, since the decision of vehicular user depends on the individual utility. When the speed of vehicular user is uniform, there are two stable coalition collections with non-zero stationary probability, i.e., ω = {{1, 2, 4}, {3}} and ω = {{1, 2, 3}, {4}}. In this case, the coalition with three users in which users 3 and 4 are not in the same coalition have non-zero probability. Since users 3 and 4 have the same sets of RSUs to be visited, either of them can improve individual utility by oining the different coalition. However, when the speed of vehicular user at RSU5 decreases, users 1 and 2 are willing to oin different coalition as long as the size of coalition is 2 or 3. Therefore, the coalition collections with the size of 2 or 3 have nonzero stationary probability. We observe that in this situation, coalition collections {{1, 3}, {2, 4}} and {{1, 4}, {2, 3}} have the highest stationary probability since users 1 and 2 are in the different coalition. From the stationary probability shown in Fig. 3, we can observe that without coalition or with the grand coalition, these collections are neither stable nor optimal. In this case, the total utility is not maximized, and the rational user will have the other coalition whose individual utility can be improved. Fig. 4 shows the total utility as the speed of vehicular user at RSU 5 is varied. As expected, optimal coalition formation can

6 6 Fig. 4. Total utility Distributed algorithm Without coalition Speed of vehicular users at RSU5 (km/h) Total utility under different speed of vehicular user at RSU5. achieve the highest total utility compared with rational formation and without coalition. Fig. 5 shows the total utility under different channel quality at RSU 5. Again, optimal coalition formation achieves the highest total utility. For both cases, distributed algorithm based on split and merge mechanisms achieve the total utility close to that of rational coalition formation. Fig. 5. Total utility Distributed algorithm Without coalition Channel quality at RSU5 (db) Total utility under different channel quality at RSU5. VI. SUMMARY In vehicle-to-roadside communications, the vehicular user can share the bandwidth for the connection to different roadside units. In this paper, the coalition game model has been formulated to analyze the behavior of vehicular users to share the bandwidth. The coalition is formed according to the mobility of the vehicular users which can reduce the cost of bandwidth reservation while the QoS requirements are also met. First, we have considered the rational coalition formation in which the vehicular user has self-interest to make decision to maximize individual utility. The dynamic model of coalition game based on Markov chain is formulated. The stable coalition collection can be obtained based on the concept of absorbing state. The distributed algorithm with split and merge mechanisms has been presented. In addition, we have considered the optimal coalition formation in which the vehicular users are concerned to maximize social welfare (i.e., total utility). For the future work, the optimal selection of RSUs to reserve the bandwidth will be proposed. Also, the effect of cache buffer to the data transfer between vehicular user and roadside unit will be studied under coalition. REFERENCES [1] J. Zhao, Y. Zhang, and G. Cao, Data Pouring and Buffering on the Road: A New Data Dissemination Paradigm for Vehicular Ad Hoc Networks, IEEE Transactions on Vehicular Technology, vol. 56, no. 6, pp , November [2] J. Lee, Design of a Network Coverage Analyzer for Roadside-to- Vehicle Telematics Networks, in Proceedings of ACIS International Conference on Software Engineering, Artificial Intelligence, Networking, and Parallel/Distributed Computing, pp , August [3] P. Belanovic, D. Valerio, A. Paier, T. Zemen, F. Ricciato, and C. Mecklenbrauker, On Wireless Links for Vehicle-to-Infrastructure Communications, IEEE Transactions on Vehicular Technology, to appear (available online at IEEE Xplore). [4] K. Yang, S. Ou, H.-H. Chen, and J. He, A Multihop Peer-Communication Protocol With Fairness Guarantee for IEEE Based Vehicular Networks, IEEE Transactions on Vehicular Technology, vol. 56, no. 6, part 1, pp , November [5] M.-F. Jhang and W. Liao, On Cooperative and Opportunistic Channel Access for Vehicle to Roadside (V2R) Communications, in Proceedings of IEEE Global Telecommunications Conference (GLOBECOM), November-December [6] K. Akkaraitsakul and E. Hossain, An Auction Mechanism for Channel Access in Vehicle-to-Roadside Communications, in Proceedings of IEEE International Conference on Communications (ICC) Workshops, pp. 1-5, June S. Yoon, D. T. Ha, H. Q. Ngo, and C. Qiao, MoPADS: A Mobility Profile Aided File Downloading Service in Vehicular Networks, IEEE Transactions on Vehicular Technology, to appear (available online at IEEE Xplore). [7] W. Saad, Z. Han, M. Debbah, A. Hørungnes and T. Basar, Coalitional Games for Distributed Collaborative Spectrum Sensing in Cognitive Radio Networks, in Proceedings of IEEE International Conference on Computer Communications (INFOCOM), April [8] A. Aram, C. Singh, S. Sarkar, and A. Kumar, Cooperative Profit Sharing in Coalition Based Resource Allocation in Wireless Networks, in Proceedings of IEEE International Conference on Computer Communications (INFOCOM), pp , April [9] D. Niyato and P. Wang, Optimization of the mobile router and traffic sources in vehicular delay tolerant network, IEEE Transactions on Vehicular Technology, to appear (available online at IEEE Xplore). [10] B. Shrestha, D. Niyato, Z. Han, and E. Hossain, Wireless Access in Vehicular Environments Using BitTorrent and Bargaining, in Proceedings of IEEE Global Telecommunications Conference (GLOBECOM), November-December [11] Q. Liu, S. Zhou, and G. B. Giannakis, Cross-layer Combining of Adaptive Modulation and Coding with Truncated ARQ over Wireless Links, IEEE Transactions on Wireless Communications, vol. 3, no. 5, pp , September [12] T. Arnold and U. Schwalbe, Dynamic coalition formation and the core, Journal of Economic Behavior & Organization, vol. 49, no. 3, pp , November 2002.