A SECURITY-ENABLED WIRELESS TOKEN CLUSTER MAC PROTOCOL WITH INTELLIGENT TOKEN POLICY

Transcription

1 A SECURITY-ENABLED WIRELESS TOKEN CLUSTER MAC PROTOCOL WITH INTELLIGENT TOKEN POLICY Tao Jin * Chunxiao Chigan ** College of Computer and Information Science and Department of Electrical and Computer Engineering Northeastern University, MA Michigan Technological University, MI taojin@ccs.neu.edu cchigan@mtu.edu ABSTRACT Most existing wireless MAC protocols are based on the assumption that all hosts cooperate with each other and are compliant to predefined rules, which is not always true in wireless network. Therefore, wireless MACs are often vulnerable to MAC misbehaviors from which selfish hosts could gain unfair channel access at the expense of others performance by deviating from protocols. Recently, researches on MAC misbehaviors have drawn considerable attention. The detection of misbehaviors is quite challenging due to the intrinsic random feature of the contension-based wireless MAC design. Besides, the contention-based random access protocols cannot support bounded delay or guaranteed channel throughput, two important aspects of QoS. In this paper, we propose STC- MAC, a Seurity-enalbed Wireless Token Cluster MAC Protocol. Thus, the reserved channel access and bounded delay could be achieved with the help of the dynamic infrastructure. With an intelligent multi-token awarding policy, the proposed approach takes advantage of space reuse in wireless network to further improve the channel utilization. In addition, an effective detection and penalty scheme is designed to coutermeasure against possible selfish MAC misbehaviors. I. INTRODUCTION Most existing wireless MAC, such as CSMA/CA, function well only when all the hosts cooperate with each other and abide by the predefined rules. Whereas, such assumption is not always true, especially in the infrastructureless wireless ad hoc networks wherein no trust can be preassumed due to the lack of the central controller and diverse interests among hosts. Thus sometimes it is beneficial for individual host to deviate from the protocol in order to gain higher chance of channel access. Besides, with the increase of programmability of wireless network card, it is getting much easier to manipulate the network parameters to invoke misbehaviors in MAC. MAC misbehaviors could be categorized into two categories: the malicious behavior and the selfish behavior. Malicious hosts are aimed at disrupting normal network operation regardless their own benefit. For example, with a typical denial-of-service (DoS) attack, malicious hosts keep sending junk messages to jam the communication channel. On the contrary, selfish hosts desist from degrading the network performance obtained by others, if such attempts cannot improve its own performance. For example, selfish hosts intentionally select smaller backoff value to increase its chance of channel access. Due to different motivations of selfish hosts and malicious hosts, the countermeasures are different. In this paper, we thrust to design a distributed MAC protocol that countermeasure against the selfish MAC misbehaviors while ensure high channel utilization and QoS in terms of throughput and bounded delay for wireless ad hoc networks. 1.1 Related Works Many solutions have been proposed to prevent or detect selfish MAC misbehaviors in popular MAC protocols. The essence of the selfish misbehavior in MAC layer is to manipulate protocol parameters to gain unfair share of channel access. Possible selfish behaviors mainly include: a. select smaller random backoff value by using smaller contention window or avoiding doubling window size after failure transmission. b. Selectively scramble frames sent by other hosts in order to increase their contention window to increase their own chance of channel access. c. Reserve channel for a long duration to transmit a large amount of data. The selfish behavior in case a is considered to be the most difficult one to detect, since the randomness in the backoff value selection makes it very difficult to distinguish the selfish misbehavior, from a well-behaved host s selection of smaller value by chance. Maxim, et al. [1] designed a system to sequentially test the extent to which the MAC parameters have been manipulated in hot spots by comparing with some averaged values. As the author pointed out, this method could be deceived by cheating peers in some cases. Both [2] and [3] proposed modifications to to alleviate the randomness in backoff selection, in order to simplify the detection of MAC misbehavior. However, neither of them further improves the network performance in terms of throughput, delay, etc. In [4], the authors studied a simple DoS attack model to indicate the significance of fairness to mitigate * This work effects was conducted of such attacks. while the Some author game was with theoretic solutions department also of Electrical have been & Computer proposed Engineering to design at Michigan protocols which Tech. ** Contact are resilient author: to Dr. selfish Chunxiao hosts, Chigan. due to the intrinsic /07/$ IEEE 1 of 7

2 rationality of game model[5]-[8]. However, most solutions are based on the assumption that every host is within the range of others, which is not realistic. Also, since game theoretic solutions assume all hosts are selfish, the performance in such system is substantially suboptimal for networks consisting of well-behaved hosts. 1.2 Challenges to Design Effective Countermeasures in MAC Some features shared by existing popular wireless MAC make it very challenging to design effective countermeasures against selfish misbehaviors: a. The contention based random access mechanisms make it difficult to distinguish selfish misbehaviors from malfunction of well-behaved hosts. b. The non-identical channel condition perceived by different nodes in the air is one of the significant factors causing misdiagnosis of selfish behaviors. c. Most existing contention-based MAC protocols have no constraints on when the host should release the channel. Once one host succeed in the channel competition, it could reserve the medium for unfairly long time, which is typically called Capture Effect Besides, most existing distributed contention-based MAC protocols (e.g., CSMA/CA), are simple and naïve, and QoS cannot be supported because neither bounded delay nor channel throughput can be ensured in such MAC protocols. Therefore, our goal in this work is to design a new wireless MAC protocol featured as: a. less random channel access b. effective channel utilization c. be resilient to selfish behaviors d. QoS supported in terms of bounded transmission delay and throughput Fig 1. cluster architecture Intuitively, token-based medium access control could be a good candidate to achieve the above features. Although some solutions of token based wireless MAC have been proposed [11] [16], they are subject to low channel utilization and vulnerable to selfish behaviors. In this paper, we propose a multi-cluster wireless network architecture to achieve intelligent token-based channel access. With this novel MAC framework, (1) multiple tokens can be supported to increase channel utilization; (2) and the intelligent token passing policy is introduced to countermeasure possible selfish misbehaviors. The rest of the paper is organized as follows: in section II, we present the basic protocol sketch. Possible selfish behaviors will be analyzed in section III and the detection and handling approach will be presented thereafter. Simulation and numerical results will be given in section IV. The conclusion is given in section V. II. BASELINE STC-MAC PROTOCOL OVERVIEW 2.1 Token-Based Multi-Cluster Network Architecture The lack of centralized control and random access scheme complicates the detection of selfish MAC misbehavirors. Also, most exisiting distributed MAC protocols have lower efficiency compared with the centralized approach. In this paper, we introduce loosely centralized control to wireless ad hoc networks to improve channel efficiency as well as to alleviate the selfish MAC misbehaviors. Figure 1 illustrates the basic architecture of our proposed multi-cluster wireless network. All nodes are grouped into clusters, each of which is managed by a Cluster Head (CH), the elected temporary base station for all the nodes within its transmission range or the subset of them. The Cluster Head is responsible for scheduling the channel access for all its cluster members. We introduce the token based channel access control, wherein only the cluster members seizing valid tokens could initiate transmission. Hence, the channel is better utilized due to the elimination of retransmissions caused by collisions. Besides, the properly defined Token Holding Time (THT), the maximum time that each host could transmit data with, can guarantee the bounded delay between two consecutive transmissions for each individual cluster member. Nevertheless, the cluster architecture is a loose infrastructure because of the dynamic change of topology in ad hoc networks. Therefore, the effective clustering algorithm and cluster recovery scheme is required to handle the mobility feature of ad hoc networks. Some clustering algorithms have been proposed[17]-[19]. Since the focus of this paper is to design a MAC protocol to better utilize channel resource, support QoS and countermeasure selfish behaviors, we assume the cluster architecture discussed in the rest of the paper is formed following the lowest-id cluster head selection rule [17]. 2.2 Channel Access Scheme With the existence of dynamic cluster-based centralized control, it is easy to achieve effective wireless channel access in terms of both fairness and QoS. The detailed 2 of 7

3 channel access operations will be explained in the following subsections Channel Acess within Cluster Every host periodically broadcasts its connectivity list, so the cluster head could maintain a complete connectivity table for the entire cluster, as shown in figure 2. The cluster head will award the token in the order of node id. In our work, we propose two types of token configurations, single-token cluster and multi-token cluster. CH CH CH CH CH CH CH Fig 2. STC-MAC network cluster and node connectivity table Token Awarding A. Single-Token Cluster In this case, at a time, only one token could exist in the cluster. Hence, the cluster head simply awards the token from the lowest-id node in the connectivity table, node 1 in figure 2, to the highest-id node. Each node will release the token back to cluster head after the transmission is completed. Figure 3 is the basic packet format of the token frame. CID TokenOwner k CH {seq_no, time} maxtht Fig 3. basic packet format of token frame in single-token cluster CID ID assigned to this cluster TokenOwner ID of current token owner k CH {seq_no, time} token verification field to indicate this is generated by clutser head maxtht maximum possible THT for token owner s transmission In this case, only one transmission exists in a cluster, and thus collisions could be minimized since the channel has been reserved before transmission. Also, the bounded delay could be guaranteed since the token owner is supposed to release the token within THT. The single-token cluster concept is similar to the wireless token ring protocol (WTRP), proposed by Ergen, et al.[11]. In their work, they have indicated via simulations that the token based wireless medium access control outperforms in terms of higher throughput, fair bandwidth allocation and predictable medium access latency. B. Multi-Token Cluster The open wireless medium introduces a new feature, the partial connectivity. In the wired token ring, all ring members are connected to the same wire, so one host could hear any other nodes in the same ring. Such full connectivity enables that one and only one node that seizes the token could access the channel at a time, otherwise, the collision might occur. Since the transmission ranges and distances of hosts in wireless token ring/cluster may vary a lot, not all the nodes in one ring are necessarily connected to each other. Therefore, the baseline token policy could lead to low channel utilization. For example, in figure 2, node 1 owns the token at current time, and it communicates with 2. Node 4 is out of the range of node 2, thus whoever node 4 communicates with, its transmission will not interfere node 1 s transmission. However, with the singletoken policy, node 4 has to defer its transmission although it has data to send. Thus, it is feasible and beneficial to have multiple tokens co-exist in one cluster. In our work, we propose the multi-token wireless cluster and introduce the concepts of Primary Token Owner (PTO) and Secondary Token Owner (STO). The cluster head polls the cluster members in the order of node id. Whether a prospective token owner needs the token is determined by 1) whether it has data to transmit; 2) whether its intended receiver is ready to receive. If the prospective node has no data to transmit, it directly acknowledges the cluster head with no data message. Otherwise, we use an explicit method to acknowledge the polling frame. The prospective token owner sends requestto-send (RTS) message to its intended receiver, which replies with clear-to-send (CTS) message if it is ready to receive data. Thus, upon hearing the CTS message, the cluster head could tell that the Primary Token should be awarded to the prospective token owner. Otherwise, the cluster head looks up the next node in its connectivity table and polls the next one. After the cluster head confirms the Primary Token Owner (PTO) and Primary Receiver (PR), it could intelligently select multiple secondary token owners to further utilize channel resources by allowing simultaneous transmissions, as well as to exclude possible interference across different token owners. The secondary token owner selection algorithm is shown in Algorithm 1. Algorithm 1: Secondary Token Owner Selection Algorithm 1) all nodes that are connected to the primary receiver are covered nodes 2) the cluster head selects one node from uncovered nodes to be the secondary token owner. 3) all nodes that are connected to any possible receiver of the secondary token owner selected from step 2) are covered nodes 3 of 7

4 4) if there exists any uncovered node, go to step 2), otherwise, secondary token owner selection is done. The cluster head generates a new token with the format as figure 4 and broadcasts the token to the whole cluster. CID PTO PR STO1 STO2 k CH {seq_no, time} THT Fig 4. basic token format in multi-token cluster The THT required by the primary token owner also can be obtained from RTS/CTS message exachanges, and no secondary transmissions could exceed the THT required by the primary transmission. All token owners, both the primary and secondary, initiate their own transmission upon receiving the valid token, if they have data to transmit. They are supposed to finish their transmission within THT. Figure 5 depicts the basic work flow of the multi-token awarding operations. token with updated sequence number after a certain time, i.e. predefined token expiration time. Another case of cluster failure is that the cluster head is down or leaves the cluster. Since each node broadcasts its list of neighbors, the cluster configuration is dynamically updated. In case the cluster head is down, a new cluster will be reconstructed soon according to clutering rules. 2.3 Multi-Cluster Resolution To reduce the inter-cluster interference, separate orthogonal channels are assigned to neighboring clusters. The spatial reuse of channel resources could eliminate the inter-cluster interference. The channel is selected at the cluster formation phase by CH and distributed to all cluster members. With the increase of nodes, the clusters might outnumber the available channels. In this case, as long as no channel is shared between adjacent clusters, the intercluster interference still could be effectively eliminated. III. DETECTION AND HANDLING OF SELFISH MISBEHAVIORS IN STC-MAC Fig 5. work flow of token passing in multi-token cluster According to the multi-token awarding protocol designed above, even if all the token owners have data to transmit, there will be no collision among them because STO is connected neither to the primary receiver, nor any possible secondary receivers. As such, more nodes could have chance to access the channel and better resource utilization is achieved Token Releasing We have assumed that no secondary transmission exceeds the THT selected by the primary token owner, so the primary transmission is always the last to complete. Hence, we use the Acknowledgement packet, which is generated by the primary receiver and sent to the primary token owner, to implicate token releasing as this is the sign of the completion of primary transmission. In this way, the control overhead could be reduced Token Recovery In case the token is lost, e.g. acknowledgement from PR to PTO is corrupted, the cluster head will generate a new As we discussed in earlier sections, the proposed tokenbased cluster architecture can effectively support QoS and improve channel utilization. However, similar to most existing wireless MAC protocols, our baseline STC-MAC protocol functions well only when all the hosts abide by the legitimate protocol operations. In the following section, we will first analyze the possible selfish MAC misbehaviors in wireless token cluster networks. Later on, the security enhanced design is proposed to address the challenges of selfish nodes. 3.1 Possible Selfish Misbehavior Since our work focuses on designing effective countermeasure against selfish MAC misbehaviors, we made an assumption that misbehaved nodes will desist from deviating from legitimate protocol operations if they cannot gain interest from such deviation. As presented in our baseline STC-MAC protocol, the fairness and bounded delay is guaranteed by the fact that all cluster members abide by the agreed THT value, and the correct order of token awarding, which is the spirit of token based protocol design. Based on the above analysis, possible selfish misbehaviors include: Selfish hosts forge token to access the channel, competing with the legal token owner. Selfish Primary Token Owner doesn t release token on time. Here Primary Token Owner could be both ordinary cluster members and the cluster head. Cluster head intentionally dismiss some nodes in the round-robin token awarding procedure in order to achieve shorter delay for itself or its colluders. 4 of 7

5 Primary token owner and primary receiver form a selfish collusion to reserve channel for quite a long duration. We do not take into account the case that secondary token owners and receivers deviate from the protocol by not releasing token on time. In case this happened, the acknowledgement frame for primary transmission will be collided by secondary transmission at the cluster head, because the cluster head could hear every cluster member. In this case, the cluster head has to wait for token expiration time to generate a new token, which probably offsets the extra throughput achieved by selfish secondary hosts. 3.2 Selfish Misbehavior Detection and Handling Approach Token Verification We modify the basic token format into figure 6. The cluster head uses its private key to sign the list of PTO, PR, STOs, and seq_no and time is also attached into this digital signature. The token verification could be classified into two types: Ensure the token is awarded by the cluster head In this case, upon receiving the token frame, the cluster member decrypts the signature and compares the time field with its local time. If the difference is larger than the predefined synchronous time skew, this will be taken as an invalid token. Ensure the sender has been awarded with valid token The token owners will attach the token frame into its data frame. Thus, every time when one node receives data from another one, it first decrypts the signature in token frame. If the cluster member receives signature without seq_no updated, the token is taken as invalid. Otherwise, it can be ensured that the transmission is valid as long as the sender belongs to the list of PTO, PR and STOs in token. CID k CH {PTO,PRCV,STO1,STO2,..., Seq_no, time} THT Fig 6. security-enabled token format Out-of-Order Token Awarding The cluster head periodically broadcasts the complete connectivity table of its cluster, and each cluster member could therefore know the order of token awarding. If some node is dismissed, all cluster members could tell, and the new cluster head will be elected if such misbehavior is occurred for more than threshold times Illegal Token Holding Time As shown in figure 5, before the token is awarded to the prospective primary token owner, RTS/CTS handshake operation is required, which broadcasts the expected THT value selected by prospective token owners based on current number of nodes (NoN) and the amount of data to transmit. The formula of maximum THT computation could be defined in the protocol. As such, the cluster head could verify the validity of the expected THT before awarding the token, and set the THT field with legal value if the expected THT is larger than the maximum possible THT under current network environment. Of course the selfish token owner can cheat the cluster head with a legal THT in the RTS/CTS phase, but reserve channel for longer than legal duration. Well behaved primary receiver In this case, the primary receiver could get the legal THT value from the broadcasted token frame in the token awarding procedure. If the primary token owner keeps transmitting after THT, the primary receiver could choose to discard its frame. Collusion of selfish token owner and receiver In this case, the observation taken by the 3 rd party is required. We allow the cluster head to monitor the actual token holding time. We denote the time when CTS is heard by the cluster head as T_cts; and the time when Ack is heard as T_ack. If T=T_ack-T_cts > α*tht, where 1<α<1+, the cluster head knows that the channel has been reserved for illegally long time, and this primary token owner will be dismissed in next token awarding rotation. CTS indicates that the connection between the primary token owner and the receiver has been successfully established; Ack is the sign of completion of transmission. The different between these two values could be estimated as the actual THT. is introduced because the overhead of token frame transmission is taken into account. IV. SIMULATION In this section, we provide some numerical results to illustrate the performance of the proposed token-based wireless cluster MAC protocol. We first compare the performance of the single-token cluster and the multi-token cluster STC-MAC in terms of system throughput and delay. Then we will present a sample selfish MAC misbehavior, and investigate the efficiency of our proposed misbehavior detection and handling scheme. Fig 7. sample topology used in simulation In our simulation, we randomly generate a cluster with 10 hosts, each of which has the same transmission range. 5 of 7

6 Node 1 is the cluster head. All other nodes are not necessarily connected to each other. Cluster head polls each node one by one to check out whether it has data to transmit, the polled node generates packet with probability P, which changes from 0.1 to 1 with increment of 0.1 each round in our simulation. And the packet size is randomly Figure 10 illustrates that in the cluster without selfish behavior detection, well-behaved hosts experience longer delay than they do in the normal cluster. Figure 11 shows that node 3 gains extra throughput by hurting all other well-behaved nodes. Both figures 10 and 11 show that with the proposed detection and penalty approach, the QoS of Fig 8. averge medium access delay vs. packet generation probability Fig 10. individual medium access delay with or without selfish attack Fig 9. system throughput vs. packet generation probability selected from (here we use time slots to measure the packet size). STOs are randomly selected according to algorithm 1. For each cluster topology, we run the simulation for time slots. Figures 8 and 9 show the performance of single-token cluster and multi-token cluster. Every data point is averaged over 10 random topologies. Both figures 8 and 9 illustrate that by enabling simultaneous transmissions, multi-token cluster outperforms single-token cluster in terms of both system throughput and delay. We also design a simulation to test the efficiency of our proposed countermeasure against the selfish misbehavior. In this simulation, we use a fixed topology as figure 7 wherein node 3 is a selfish host and access the channel for 2*THT every time when it seizes token. The penalty scheme used in the simulation is that cluster head will dismiss node 3 for 2 rounds once its selfish behavior is detected. Fig 11. individual throughput with or without selfish attack the well-behaved nodes could be ensured and they all achieve similar performance as they do in normal cluster. V. CONCLUSION Based on our analysis of the flaws of existing wireless MAC protocols, and the related work on wireless MAC misbehaviors, we adopted a concept of the virtual centralized approach to design an enhanced distributed MAC protocol, STC-MAC, to simplify the security issue as well as to improve the performance of wireless MAC. With our proposed token-based wireless cluster MAC protocol, the feature of the token-based medium access control ensures the channel access throughput since the channel is reserved before transmission by token awarding. Thus, the channel utilization is increased by eliminating retransmissions caused by collisions. Also, the bandwidth is fairely assigned to individual hosts since each host takes a turn to access the channel. On the other hand, the cluster 6 of 7

7 architecture provides a dynamic infrastructure platform, which simplifies the medium access control problem in distributed networks. Based on the analysis of the partial connectivity feature in the wireless medium, we proposed an intelligent token awarding policy, multi-token cluster, to better utilize the channel resource. We also analyzed the possible selfish MAC misbehaviors in wireless token cluster networks, the corresponding security scheme is proposed. Our simulation has evidenced the efficiency of our detection and penalty scheme against selfish attacks in the wirless token cluster networks. REFERENCES [1] Raya M, Hubaux JP, and Aad I, DOMINO: A System to Detect Greedy Behavior in IEEE Hotspots, in Proceedings of ACM MobiSys, Boston - MA, 2004 [2] Kyasanur P, and Vaidya NH, Selfish MAC Layer Misbehavior in Wireless Networks, IEEE Transactions on Mobile Computing, Volume 4, Issue 5, Sept.-Oct. 2005, pp [3] Cárdenas AA, Radosavac S, and Baras JS, Detection and prevention of MAC layer misbehavior in ad hoc networks, in Proceedings of ACM workshop on Security of Ad Hoc and Sensor Networks (SASN), Washington DC, Oct [4] Gupta V, Krishnamurthy S, and Faloutsos M, Denial of Service attacks at the MAC layer in wireless ad hoc networks, In Proceedings of Military Communication Conference (MILCOM), Anaheim, CA, Oct [5] MacKenzie AB, and Wicker SB, Selfish Users in Aloha: A Game-Theoretic Approach, in Proceedings of IEEE Vehicular Technology Conference, Oct [6] Cagalj M, Ganeriwal S, Aad I, and Hubaux JP, On Selfish Behavior in CSMA/CA Ad Hoc Networks, in Proceedings of IEEE Infocom Conference 2005, Vol. 4, March 2005, pp [7] MacKenzie AB, and Wicker SB, Stability of Multipacket Slotted Aloha with Selfish Users and Perfect Information, in Proceedings of Infocom 2003, San Francisco, CA, April [8] MacKenzie AB, and Wicker SB, Game Theory and design of self-configuring, adaptive wireless networks, IEEE Communication Magazine,, Nov. 2001, pp [9] Buttyán L, and Hubaux JP, Stimulating Cooperation in Self- Organizing Mobile Ad Hoc Networks, Technical Report No. DSC/2001/046, Swiss Federal Institute of Technology, Lausanne, August [10] Buchegger S, and Boudec JYL, Performance Analysis of the CONFIDANT Protocol: Cooperation Of Nodes - Fairness In Distributed Ad-hoc NeTworks, in Proceedings of IEEE/ACM Workshop on Mobile Ad Hoc Networking and Computing (MobiHOC), Lausanne, CH, June [11] Ergen M, Lee D, Sengupta R, and Varaiya P, WTRP - wireless token ring protocol, IEEE Transactions on Vehicular Technology, Volume 53, Issue 6, Nov pp [12] Lee D, Attias R, Puri A, Sengupta R, Tripakis S, and Varaiya P, A wireless token ring protocol for intelligent transportation systems, in Proceedings of IEEE Intelligent Transportation Systems, Aug. 2001, pp [13] Deng Z, Lu Y, Wang C, and Wang W, EWTRP: enhanced wireless token ring protocol for small-scale wireless ad hoc networks, in Proceedings of 2004 International Conference on Communications, Circuits and Systems (ICCCAS 2004), Volume 1, Issue, June 2004, pp [14] Ergen M, Lee D, Sengupta R, and Varaiya P, Wireless token ring protocol-performance comparison with IEEE , in Proceedings of Eighth IEEE International Symposium on Computers and Communication (ISCC 2003), 30 June-3 July 2003, pp [15] Taheri SA, and Scaglione A, Token enabled multiple access (TEMA) for packet transmission in high bit rate wireless local area networks, in Proceedings of IEEE International Conference on Communications (ICC), 2002, pp [16] Miura S, Nakamura H, Kamienoo M, and Araki K, Radio link control method by using radio token in high speed wireless LAN system, in Proceedings of IEEE Global Telecommunications Conference (GLOBECOM 98), Volume 3, Issue, 1998, pp [17] Gerla M, and Tsai JTC, Multicluster mobile multimedia radio networks, ACM-Baltzer Journal of Wireless Networks, 1(3): , [18] Baker DJ, Flynn JA, Ephremides A, The Design and Simulation of a Mobile Radio Network with Distributed Control, IEEE Joumal on Selected Areas in Communications, pp , [19] Safwat A, Hassanein HS, and Mouftah H, Power-Aware Fair Infrastructure Formation for Wireless Mobile Ad Hoc Communications, in Proceedings of IEEE Globecom 2001, Volume 5, pp of 7