《通信专业英语》ADVANCES IN SMART ANTENNAS

ADVANCES IN SMART ANTENNAS COOPERATIVE WIRELESS COMMUNICATIONS A〔R0 SS-LAYER APPROACH PEI LIU, ZHIFENG TAO, ZINAN LIN, ELZA ERKIP, AND SHIVENDRA PANWAR, POLYTECHNIC UNIVERSITY ABSTRACT deployed. Even when MIMO technology is feasi ble, wireless engineers are runni Denise and her husband Mitch are at opposite roadblock: the inefficient way the electromagnet Rate2 Ra ends of a living room at a crowded party. Denise ic spectrum has been allocated to different class- tries to attract Mitch's attention and shouts some- es of users, mainly for historical or regulatory Celine, in the middle of the room, who overhears trum are grossly underused, the popular unli Denise and notices their predicament, repeats to censed bands are very crowded. Given this Mitch the part she hears: Go home. This time, all limitation, for unlicensed bands, the issue of Mitch hears is the word home. Mitch finally fig nterference from having too many users has ures our that his wife wants to go home. " This become as important as how much bandwidth analogy from everyday life vividly portrays the can be squeezed from it. essential element of cooperative wireless com This article outlines one way to address these munications, namely, utilizing information over- problems by using the notion of cooperation heard by neighboring nodes to provide robust between wireless nodes In cooperative commu In cooperative communication between a source and its desti- nications, multiple nodes in a wireless network nation. Cooperative communication exhibits var- work together to form a virtual antenna array communications ious forms at different protocol layers and Using cooperation, it is possible to exploit the introduces many opportunities for cross-layer spatial diversity of the traditional MIMO tech multiple nodes in design and optimization, some of which will be niques without each node necessarily having mul- explored in detail in this article wireless network tiple antennas. Multihorabling intermediate form of cooperation by er INTRODUCTION nodes to forward the message from source to work together to destination. However, cooperative communica form a virtual The burgeoning demand for mobile data net- tion techniques described in this article are fun- works has highlighted some constraints on its damentally different in that the relaying nodes future growth. Wireless links have always had can forward the information fully or in part antenna array orders of magnitude less bandwidth than their Also the destination receives multiple versions of wireline counterparts. Mobile users have always the message from the source, and one or more chafed at this limitation, which essentially forces relays and combines these to obtain a more reli- it is possible to them to use applications in a manner reminis- able estimate of the transmitted signal as well as cent of wireline networks of decades past, albeit higher data rates. The main advantages of coop- exploit the spatial freeing them from a desktop. Newer technolo- erative communications are gies such as multiple-input multiple-output Higher spatial div esistance to both diversity of the (MIMO) systems are starting to increase the small scale and shadow fading traditionol mimo number of bits per second per hertz of band Higher throughput/lower delay: higher achiev- width through spatial multiplexing, and to able data rates, fewer retransmissions, and techniques without improve the robustness/range of the wireless link lower transmission delay or a given data rate through space-time coding Reduced interference/lower transmitted and beamforming. However, all these improve- equency reuse In a ments come at the cost of multiple rF front llular/WLAN deployment multiple antennas nds at both the transmitter and the receiver Adaptability to network conditions: oppor Furthermore the size of the mobile devices may tunistic use and redistribution of network limit the number of antennas that can be energy and bandwidth The past few years have seen tremend interest in cooperative communications, mostly The work is partially supported by the National Science at the physical layer. However, significant Foundation under grant no. 0520054, and the Wireless research challenges still exist. some of which we Internet Center for Advanced Technology(WICAT), an outline in this article NSF Industry/University Cooperative Research Center. The goal of this article is to provide new 1536-1284/06/S20.002006IEEE IEEE Wireless Communications august 2006

84 1536-1284/06/$20.00 © 2006 IEEE IEEE Wireless Communications • August 2006 Rate2 Rate Rate4 Rate Rate1 R1 R2 ADVANCES IN SMART ANTENNAS INTRODUCTION The burgeoning demand for mobile data net￾works has highlighted some constraints on its future growth. Wireless links have always had orders of magnitude less bandwidth than their wireline counterparts. Mobile users have always chafed at this limitation, which essentially forces them to use applications in a manner reminis￾cent of wireline networks of decades past, albeit freeing them from a desktop. Newer technolo￾gies such as multiple-input multiple-output (MIMO) systems are starting to increase the number of bits per second per hertz of band￾width through spatial multiplexing, and to improve the robustness/range of the wireless link for a given data rate through space-time coding and beamforming. However, all these improve￾ments come at the cost of multiple RF front ends at both the transmitter and the receiver. Furthermore, the size of the mobile devices may limit the number of antennas that can be deployed. Even when MIMO technology is feasi￾ble, wireless engineers are running into another roadblock: the inefficient way the electromagnet￾ic spectrum has been allocated to different class￾es of users, mainly for historical or regulatory reasons. Thus, while large portions of the spec￾trum are grossly underused, the popular unli￾censed bands are very crowded. Given this limitation, for unlicensed bands, the issue of interference from having too many users has become as important as how much bandwidth can be squeezed from it. This article outlines one way to address these problems by using the notion of cooperation between wireless nodes. In cooperative commu￾nications, multiple nodes in a wireless network work together to form a virtual antenna array. Using cooperation, it is possible to exploit the spatial diversity of the traditional MIMO tech￾niques without each node necessarily having mul￾tiple antennas. Multihop networks use some form of cooperation by enabling intermediate nodes to forward the message from source to destination. However, cooperative communica￾tion techniques described in this article are fun￾damentally different in that the relaying nodes can forward the information fully or in part. Also the destination receives multiple versions of the message from the source, and one or more relays and combines these to obtain a more reli￾able estimate of the transmitted signal as well as higher data rates. The main advantages of coop￾erative communications are: • Higher spatial diversity: resistance to both small scale and shadow fading • Higher throughput/lower delay: higher achiev￾able data rates, fewer retransmissions, and lower transmission delay • Reduced interference/lower transmitted power: better frequency reuse in a cellular/WLAN deployment • Adaptability to network conditions: oppor￾tunistic use and redistribution of network energy and bandwidth The past few years have seen tremendous interest in cooperative communications, mostly at the physical layer. However, significant research challenges still exist, some of which we outline in this article. The goal of this article is to provide new PEI LIU, ZHIFENG TAO, ZINAN LIN, ELZA ERKIP, AND SHIVENDRA PANWAR, POLYTECHNIC UNIVERSITY ABSTRACT “Denise and her husband Mitch are at opposite ends of a living room at a crowded party. Denise tries to attract Mitch’s attention and shouts some￾thing at him. All Mitch can hear is the word ‘Let’s.’ Celine, in the middle of the room, who overhears Denise and notices their predicament, repeats to Mitch the part she hears: ‘Go home.’ This time, all Mitch hears is the word ‘home.’ Mitch finally fig￾ures out that his wife wants to go home.” This analogy from everyday life vividly portrays the essential element of cooperative wireless com￾munications, namely, utilizing information over￾heard by neighboring nodes to provide robust communication between a source and its desti￾nation. Cooperative communication exhibits var￾ious forms at different protocol layers and introduces many opportunities for cross-layer design and optimization, some of which will be explored in detail in this article. COOPERATIVE WIRELESS COMMUNICATIONS: A CROSS-LAYER APPROACH The work is partially supported by the National Science Foundation under grant no. 0520054, and the Wireless Internet Center for Advanced Technology (WICAT), an NSF Industry/University Cooperative Research Center. In cooperative communications, multiple nodes in a wireless network work together to form a virtual antenna array. Using cooperation, it is possible to exploit the spatial diversity of the traditional MIMO techniques without each node having multiple antennas. ERKIP LAYOUT 8/3/06 1:24 PM Page 84

1=1.0.d2=0.5,d=0.5 (a) S transmits directly to D 102 S transmits R relays for S 10- N/2 coded bits N/2 Average total received SNR at the destination(dB) a Figure 1. a) Cooperative system for an isolated link; b)time division in cooperative coding c) two user cooperative coding perfor- mance for d,=l, d2=0.5 and d=0.5, (13, 15, 15, 17)convolutional code, 100-byte frame size cross-layer research directions in order to illus- or ad hoc systems; only one copy of the signal, trate the feasibility and performance of coopera- whether it comes from the mobile directly or tive wireless networking. We first describe the from a relay, is processed at the destination notion of physical-layer cooperation and cooper- Hence, cooperative relaying is substantially dif ative diversity. However, in order to realize a ferent than traditional multihop or infrastructure rative network, research at the physi- based methods. cal layer should be coupled with higher layers of This notion of cooperation dates back to the the protocol stack, in particular, the MAC sub- relay channel model in information theory exten ayer and the network layer. We describe how sively studied in the 1970s by Cover and El MAC SUK Cooperation can be integrated with Gamal [2], but we owe the recent popularity to the MAC sublayer for dramatic improvements in 3-5, which showed the benefits of cooperative throughput and interference. We also outline relaying in a wireless environment. In order to some of the challenges in extending the notion illustrate the idea of cooperation and coopera of cooperative diversity to the network lay tive diversity at the physical layer, we consider the cooperative coding scheme used in [6, 7.Let MOTIVATION FOR us consider an isolated source s who wants to communicate with a destination D with the help COOPERATIVE COMMUNICATION of a cooperative relay R, as illustrated in Fig. 1 Here, di denotes the distances between the In this section we introduce the basic concepts nodes underlying cooperative communications. Coop- For direct transmission (i.e, if the relay R is erative techniques utilize the broadcast nature of not utilized), each channel block, or packet, con- wireless signals by observing that a source signal tains B data bits and r parity bits for forward intended for a particular destination can be error correction(FEC), leading to a total of N "overheard"at neighboring nodes. These nodes, B+r coded bits, as shown at the top of Fig 1b called relays, partners, or helpers, process the sig- For ease of exposition, we have r2 B. We assume nals they overhear and transmit towards the des- that cyclic redundancy check(CrC) is employed tination. The relay operations can consist of for error detection. In order to cooperate, S repetition of the overheard signal (obtained, for divides its channel block into two and only trans- example, by decoding and then re-encoding the mits in the first half, as shown at the bottom of information or by simply amplifying the received Fig 1b. Hence, in the cooperative mode S ends signal and then forwarding), or can involve more up sending only half of its coded bits. These bits sophisticated strategies such as forwarding only are received both by the destination and by the part of the information, compressing the over- relay R. The relay observes a higher coding rate heard signal, and then forwarding. We refer the and thus a weaker FEC. Nevertheless, it attempts reader to [1] for a detailed overview of relaying to decode the underlying B data bits. If R has methods. The destination combines the signals the correct information(which can be checked oming from the source and the relays, enabling using the CRC) -encodes and sends the higher transmission rates and robustness against remaining N/2 parity bits in the second half of hannel variations due to fading. We note that Ss time slot. Otherwise, R informs S that there he spatial diversity arising from cooperation is was a failure in decoding, and s continues trans not exploited in current cellular, wireless LAN, mission. Therefore, when R decodes correctly, IEEE Wireless Communications august 2006

IEEE Wireless Communications • August 2006 85 cross-layer research directions in order to illus￾trate the feasibility and performance of coopera￾tive wireless networking. We first describe the notion of physical-layer cooperation and cooper￾ative diversity. However, in order to realize a fully cooperative network, research at the physi￾cal layer should be coupled with higher layers of the protocol stack, in particular, the MAC sub￾layer and the network layer. We describe how physical-layer cooperation can be integrated with the MAC sublayer for dramatic improvements in throughput and interference. We also outline some of the challenges in extending the notion of cooperative diversity to the network layer. MOTIVATION FOR COOPERATIVE COMMUNICATION In this section we introduce the basic concepts underlying cooperative communications. Coop￾erative techniques utilize the broadcast nature of wireless signals by observing that a source signal intended for a particular destination can be “overheard” at neighboring nodes. These nodes, called relays, partners, or helpers, process the sig￾nals they overhear and transmit towards the des￾tination. The relay operations can consist of repetition of the overheard signal (obtained, for example, by decoding and then re-encoding the information or by simply amplifying the received signal and then forwarding), or can involve more sophisticated strategies such as forwarding only part of the information, compressing the over￾heard signal, and then forwarding. We refer the reader to [1] for a detailed overview of relaying methods. The destination combines the signals coming from the source and the relays, enabling higher transmission rates and robustness against channel variations due to fading. We note that the spatial diversity arising from cooperation is not exploited in current cellular, wireless LAN, or ad hoc systems; only one copy of the signal, whether it comes from the mobile directly or from a relay, is processed at the destination. Hence, cooperative relaying is substantially dif￾ferent than traditional multihop or infrastructure based methods. This notion of cooperation dates back to the relay channel model in information theory exten￾sively studied in the 1970s by Cover and El Gamal [2], but we owe the recent popularity to [3–5], which showed the benefits of cooperative relaying in a wireless environment. In order to illustrate the idea of cooperation and coopera￾tive diversity at the physical layer, we consider the cooperative coding scheme used in [6, 7]. Let us consider an isolated source S who wants to communicate with a destination D with the help of a cooperative relay R, as illustrated in Fig. 1a. Here, di denotes the distances between the nodes. For direct transmission (i.e., if the relay R is not utilized), each channel block, or packet, con￾tains B data bits and r parity bits for forward error correction (FEC), leading to a total of N = B + r coded bits, as shown at the top of Fig. 1b. For ease of exposition, we have r ≥ B. We assume that cyclic redundancy check (CRC) is employed for error detection. In order to cooperate, S divides its channel block into two and only trans￾mits in the first half, as shown at the bottom of Fig. 1b. Hence, in the cooperative mode S ends up sending only half of its coded bits. These bits are received both by the destination and by the relay R. The relay observes a higher coding rate and thus a weaker FEC. Nevertheless, it attempts to decode the underlying B data bits. If R has the correct information (which can be checked using the CRC), it re-encodes and sends the remaining N/2 parity bits in the second half of S’s time slot. Otherwise, R informs S that there was a failure in decoding, and S continues trans￾mission. Therefore, when R decodes correctly, ■ Figure 1. a) Cooperative system for an isolated link; b) time division in cooperative coding; c) two user cooperative coding perfor￾mance for d1 = 1, d2 = 0.5 and d3 = 0.5, (13, 15, 15, 17) convolutional code, 100-byte frame size. S transmits directly to D N coded bits S d2 d3 d1 D R (a) (b) Average total received SNR at the destination (dB) 0 10–2 10–3 Frame error rate10–1 100 5 10 (c) d1 = 1.0, d2 = 0.5, d3 = 0.5 15 20 25 Direct transmission Cooperative coding S transmits R relays for S N/2 coded bits N/2 coded bits ERKIP LAYOUT 8/3/06 1:24 PM Page 85

Rates S transmits directly to D Time T2 s transmits 1 traffic to s R2 Time t. Time T2 a) a Figure 2. a) Cooperation in a network; b)illustration of the delay and throughput improvement achieved by cooperation in the time the destination will receive half the coded bits HIGHER SPATIAL DIVERSITY from S and the remaining ones from R, thus cre- ating spatial diversity. The question is how often As a simple example, Fig 2a shows a small net this happens and how it affects the overall error work of four mobile nodes. If the channel quali- ty between mobile nodes S and D degrades Figure Ic illustrates simulation results for severely (e. g, due to shadow or small-scale fad frame error rate(FER) versus the total tran ng), a direct transmission between these two mit signal-to-noise ratio(SNR) for the scenario nodes may experience an intolerable error rate where the relay is located halfway between the which in turn leads to retransmissions Alterna source and destinati on (i.e, di=1.0, d2=0.5, tively, S can exploit spatial diversity by having and d3=0.5). Note that direct transmission relay RI overhear the transmissions and then and cooperative coding use the same total forward the packet to D as discussed above. The power and bandwidth(we consider a low-mobil- source S may resort to yet another terminal R2 ity environment). Hence, along with path loss, for help in forwarding the information, or use RI we assume all links experience independent and R2 simultaneously (8 Similar ideas apply to slow Rayleigh fading that stays constant for the larger networks as well. Therefore, compared duration of each packet. The nodes use convo- with direct transmission, the cooperative lutional coding and each node has the same approach enjoys a higher successful transmission average power constraint We observe from the probability. We note here that cooperative com- figure that for an error rate of 10-3 we obtain munications has the ability to adapt and to miti- about 18 dB improvement in SNR with cooper- gate the effects of shadow fading better than ation. Also, the FER for cooperative coding MIMO since, unlike MIMO, antenna elements decreases at a much faster rate than direct of a cooperative virtual antenna array are sepa ransmission; in fact, cooperation is able to rated in space and experience different shadow achieve two full levels of diversity similar to a fading MIMO system with two transmit antennas and one receive antenna HIGHER THROUGHPUT- LOWER DELAY The above example considers one particular At the physical layer, rate adaptation is achieved cooperative scheme to obtain diversity, yet it through adaptive modulation and adaptive chan- shows the potential of cooperation at the physi- nel coding. Many MAC protocols have intro- cal layer. Indeed, there is a rich literature on duced rate adaptation to combat adverse chann physical-layer cooperation that investigates many conditions. For instance, when a high channel aspects, such as cooperative protocols for two error rate is encountered due to a low average more users, performance bounds for cooperative SNR, the wireless LAN standard IEEE 802.11 systems, resource allocation for cooperation, and switches to a lower transmission rate so as to partner-choice strategies Using a cross-layer guarantee a certain error rate. The power of approach between physical and MAC layers, this cooperation is evident when it is applied in con article investigates how these gains can be junction with any rate adaptation algorithm. In attained in a wireless network Fig. 2a, specifically, if Rate2 and Rates are higher than Rate I such that the total transmission ti for the two-hop case through R2 is smaller than BENEFITS OF COOPERATIVE NETWORKING that of the direct transmission,cooperation read From the perspective of the network, coopera- ily outperforms the legacy direct transmission, in tion can benefit not only the nodes involved, but terms of both throughput and delay perceived by the whole network in many different aspects. the source S. Furthermore, for relays such as R1 For illustration purposes, we choose to explain and R2, it turns out that their own individual only a few potential benefits below self-interest can be best served by helping others IEEE Wireless Communications August 2006

86 IEEE Wireless Communications • August 2006 the destination will receive half the coded bits from S and the remaining ones from R, thus cre￾ating spatial diversity. The question is how often this happens and how it affects the overall error performance. Figure 1c illustrates simulation results for frame error rate (FER) versus the total trans￾mit signal-to-noise ratio (SNR) for the scenario where the relay is located halfway between the source and destination (i.e., d1 = 1.0, d2 = 0.5, and d3 = 0.5). Note that direct transmission and cooperative coding use the same total power and bandwidth (we consider a low-mobil￾ity environment). Hence, along with path loss, we assume all links experience independent slow Rayleigh fading that stays constant for the duration of each packet. The nodes use convo￾lutional coding and each node has the same average power constraint. We observe from the figure that for an error rate of 10–3 we obtain about 18 dB improvement in SNR with cooper￾ation. Also, the FER for cooperative coding decreases at a much faster rate than direct transmission; in fact, cooperation is able to achieve two full levels of diversity similar to a MIMO system with two transmit antennas and one receive antenna. The above example considers one particular cooperative scheme to obtain diversity, yet it shows the potential of cooperation at the physi￾cal layer. Indeed, there is a rich literature on physical-layer cooperation that investigates many aspects, such as cooperative protocols for two or more users, performance bounds for cooperative systems, resource allocation for cooperation, and partner-choice strategies. Using a cross-layer approach between physical and MAC layers, this article investigates how these gains can be attained in a wireless network. BENEFITS OF COOPERATIVE NETWORKING From the perspective of the network, coopera￾tion can benefit not only the nodes involved, but the whole network in many different aspects. For illustration purposes, we choose to explain only a few potential benefits below. HIGHER SPATIAL DIVERSITY As a simple example, Fig. 2a shows a small net￾work of four mobile nodes. If the channel quali￾ty between mobile nodes S and D degrades severely (e.g., due to shadow or small-scale fad￾ing), a direct transmission between these two nodes may experience an intolerable error rate, which in turn leads to retransmissions. Alterna￾tively, S can exploit spatial diversity by having a relay R1 overhear the transmissions and then forward the packet to D as discussed above. The source S may resort to yet another terminal R2 for help in forwarding the information, or use R1 and R2 simultaneously [8]. Similar ideas apply to larger networks as well. Therefore, compared with direct transmission, the cooperative approach enjoys a higher successful transmission probability. We note here that cooperative com￾munications has the ability to adapt and to miti￾gate the effects of shadow fading better than MIMO since, unlike MIMO, antenna elements of a cooperative virtual antenna array are sepa￾rated in space and experience different shadow fading. HIGHER THROUGHPUT-LOWER DELAY At the physical layer, rate adaptation is achieved through adaptive modulation and adaptive chan￾nel coding. Many MAC protocols have intro￾duced rate adaptation to combat adverse channel conditions. For instance, when a high channel￾error rate is encountered due to a low average SNR, the wireless LAN standard IEEE 802.11 switches to a lower transmission rate so as to guarantee a certain error rate. The power of cooperation is evident when it is applied in con￾junction with any rate adaptation algorithm. In Fig. 2a, specifically, if Rate2 and Rate3 are higher than Rate1 such that the total transmission time for the two-hop case through R2 is smaller than that of the direct transmission, cooperation read￾ily outperforms the legacy direct transmission, in terms of both throughput and delay perceived by the source S. Furthermore, for relays such as R1 and R2, it turns out that their own individual self-interest can be best served by helping others. ■ Figure 2. a) Cooperation in a network; b) illustration of the delay and throughput improvement achieved by cooperation in the time domain. (b) S transmits directly to D Time T1 R1 transmits its own traffic to D Time T2 (a) Rate2 Rate3 Rate4 Rate5 Rate1 R1 R2 D S S transmits Time T3 Time T4 R1 transmits its own traffic to D R1 relays for S to D Time T2 ERKIP LAYOUT 8/3/06 1:24 PM Page 86

As further illustrated in Fig. 2b, the intermediate signaling message have to be introduced to the As wireless network node Ri that cooperates enjoys the benefit of MAC layer, and information on channel condi lower channel-access delay, which in turn can be tions for related wireless links should be made deployments become translated into higher throughput. It is worth- available to the upper layers so that the coop while to note that Fig. 2b also draws a rough tion can be fully enabled. Another example of a ever more dense analogy with the cooperative scheme discussed cross-layer approach to cooperation, which above(Fig. 1b)and illustrates that rate adapta- involves interaction between the application a reduction of sir tion can further imprive the benefits of coopera- layer and the phisical iaer, is provided in inktor ill directly lead to a boost in network LOWER POWER CONSUMPTION AND LOWER COOPMAC: A COOPERATIVE INTERFERENCE /EXTENDED COVERAGE MEDIUM ACCESS CONTROL capacity. Indeed, the The diversity error rate. and throughput gai problem of dense obtained through cooperation can be traded in As described above, cooperation at the physical for power savings at the terminals. Alternatively, layer uses the broadcast nature of the wireless deployment has cooperation leads to an extended coverage area medium and overheard information to improve when the performance metric(error rate, the performance. Unfortunately, conventional already been hroughput, etc. )is fixed wireless medium access control (MAC)proto The advantage of cooperation also leads to cols have long treated this feature as a problem reported for IEEE reduced interference when the network is rather than something that can be exploited. The 802.1b/ deployed in a cellular fashion to reuse a limited methodology of cooperation, however, embraces bandwidth. With the improvement of through- this concept, and thus creates a new paradigm networks which put, we can reduce the average channel time for MAC protocol design in wireless network. used by each station to transfer a certain amount We present a new MAC protocol called have only thre of traffic over the network. Therefore, the sig- Coop MACI for IEEE 802.11 wireless LANs nal-to-interference ratio (SIR) between proximal which exploits both the broadcast nature of the cells using the same channel can be reduced, and wireless channel and cooperative diversity. As a more uniform coverage can be achieved. As we demonstrate, the CoopMAC protocol fully wireless network deployments become ever more capitalizes on the notion of cooperation, and dense, a reduction of SIR will directly lead to a realizes some of the key benefits previously higl boost in network capacity. Indeed, the problem lighted, such as higher throughput, lower delay, of dense deployment has already been reported better coverage, and reduced interference In the for IEEE 802.11 b/g networks, which have only end, we briefly discuss a preliminary CoopMAC three nonoverlapping channels. mplementatio Zhao and valenti also consider a MAC pro ADAPTABILITY TO tocol [11] for exploiting cooperative diversity, NETWORK CONDITIONS but it is based upon a conceptual generalization of the hybrid automatic repeat request scheme The cooperative communication paradigm allows(hybrid-ARQ), instead of the widely deployed wireless terminals to seamlessly adapt to chang- 802.11 protocol. Recently, there have been ing channel and interference conditions. The attempts to explore the benefits of virtual MIMO choice of relays, cooperation strategy, and the at the network level, by pursuing a cross-layer amount of resources available for cooperation approach spanning the physical, MAC, and net can be opportunistically decided. For example, working(e.g, routing) layers [ 12]. However, the in Fig. 2a, if the source S has some informati about the current channel gains, packet-loss that multiple nodes can be perfectly synchro- rates, traffic conditions, interference, or remain- nized. Although the protocol mechanism pro ing battery energy of nodes in the network, it posed herein bears some resemblance to that may choose to transmit its information directly described in[13), the two protocols address fun to its destination D, using Ri or R2 or both in a damentally different issues in two distinct prob- mission mode results in better performance(in the main focus of [13], while cooperative diversi- terms of error rates, throughput, or power). This ty is incorporated in the protocol introduced gy or bandwidth at a particular node can be uti lized by other nodes in the network in a manner COoPMAC PROTOCOL DESCRIPTION that will benefit everyone, including the relay node itself cO When a source node has a new MAC proto- data unit(MPDU) to send, it can either Although originating from physical-layer transmit directly to t the destination. or use an ooperation, all the aforementioned benefit intermediate helper for relaying, whichever con- cannot be fully realized until proper mechanisms sumes less total air time. The air time is cor have been incorporated at higher protocol layers pared using cached information on the feasible (e.g, MAC, network) and the necessary informa- data rates between the three nodes. The feasible tion is made available from the lower layer(e. g, data rate is the largest data rate that guarantees PHY). Indeed, a cross-layer approach has to be a predetermined average error rate threshold for followed to reap all the benefits of in average channel snr As we illustrate via the cooperative MAC proto- Beyond its normal function, a request to A preliminary version col described in the following section, an addi- send(RTS) message is also used by CoopMAc the Coop MAC protoco tional three-way handshake procedure and a new to notify the node that has been selected for was described in /101 IEEE Wireless Communications August 2006 87

IEEE Wireless Communications • August 2006 87 As further illustrated in Fig. 2b, the intermediate node R1 that cooperates enjoys the benefit of lower channel-access delay, which in turn can be translated into higher throughput. It is worth￾while to note that Fig. 2b also draws a rough analogy with the cooperative scheme discussed above (Fig. 1b) and illustrates that rate adapta￾tion can further improve the benefits of coopera￾tion in a network setting. LOWER POWER CONSUMPTION AND LOWER INTERFERENCE/EXTENDED COVERAGE The diversity, error rate, and throughput gains obtained through cooperation can be traded in for power savings at the terminals. Alternatively, cooperation leads to an extended coverage area when the performance metric (error rate, throughput, etc.) is fixed. The advantage of cooperation also leads to reduced interference when the network is deployed in a cellular fashion to reuse a limited bandwidth. With the improvement of through￾put, we can reduce the average channel time used by each station to transfer a certain amount of traffic over the network. Therefore, the sig￾nal-to-interference ratio (SIR) between proximal cells using the same channel can be reduced, and a more uniform coverage can be achieved. As wireless network deployments become ever more dense, a reduction of SIR will directly lead to a boost in network capacity. Indeed, the problem of dense deployment has already been reported for IEEE 802.11 b/g networks, which have only three nonoverlapping channels. ADAPTABILITY TO NETWORK CONDITIONS The cooperative communication paradigm allows wireless terminals to seamlessly adapt to chang￾ing channel and interference conditions. The choice of relays, cooperation strategy, and the amount of resources available for cooperation can be opportunistically decided. For example, in Fig. 2a, if the source S has some information about the current channel gains, packet-loss rates, traffic conditions, interference, or remain￾ing battery energy of nodes in the network, it may choose to transmit its information directly to its destination D, using R1 or R2 or both in a cooperative fashion, depending on which trans￾mission mode results in better performance (in terms of error rates, throughput, or power). This way, a surplus of resources such as battery ener￾gy or bandwidth at a particular node can be uti￾lized by other nodes in the network in a manner that will benefit everyone, including the relay node itself. Although originating from physical-layer cooperation, all the aforementioned benefits cannot be fully realized until proper mechanisms have been incorporated at higher protocol layers (e.g., MAC, network) and the necessary informa￾tion is made available from the lower layer (e.g., PHY). Indeed, a cross-layer approach has to be followed to reap all the benefits of cooperation. As we illustrate via the cooperative MAC proto￾col described in the following section, an addi￾tional three-way handshake procedure and a new signaling message have to be introduced to the MAC layer, and information on channel condi￾tions for related wireless links should be made available to the upper layers so that the coopera￾tion can be fully enabled. Another example of a cross-layer approach to cooperation, which involves interaction between the application layer and the physical layer, is provided in [9] for transmission of video signals over wireless links. COOPMAC: A COOPERATIVE MEDIUM ACCESS CONTROL As described above, cooperation at the physical layer uses the broadcast nature of the wireless medium and overheard information to improve the performance. Unfortunately, conventional wireless medium access control (MAC) proto￾cols have long treated this feature as a problem, rather than something that can be exploited. The methodology of cooperation, however, embraces this concept, and thus creates a new paradigm for MAC protocol design in wireless network. We present a new MAC protocol called CoopMAC1 for IEEE 802.11 wireless LANs, which exploits both the broadcast nature of the wireless channel and cooperative diversity. As we demonstrate, the CoopMAC protocol fully capitalizes on the notion of cooperation, and realizes some of the key benefits previously high￾lighted, such as higher throughput, lower delay, better coverage, and reduced interference. In the end, we briefly discuss a preliminary CoopMAC implementation. Zhao and Valenti also consider a MAC pro￾tocol [11] for exploiting cooperative diversity, but it is based upon a conceptual generalization of the hybrid automatic repeat request scheme (hybrid-ARQ), instead of the widely deployed 802.11 protocol. Recently, there have been attempts to explore the benefits of virtual MIMO at the network level, by pursuing a cross-layer approach spanning the physical, MAC, and net￾working (e.g., routing) layers [12]. However, the proposed scheme is based on the assumption that multiple nodes can be perfectly synchro￾nized. Although the protocol mechanism pro￾posed herein bears some resemblance to that described in [13], the two protocols address fun￾damentally different issues in two distinct prob￾lem spaces. More specifically, rate adaptation is the main focus of [13], while cooperative diversi￾ty is incorporated in the protocol introduced here. COOPMAC PROTOCOL DESCRIPTION •When a source node has a new MAC proto￾col data unit (MPDU) to send, it can either transmit directly to the destination, or use an intermediate helper for relaying, whichever con￾sumes less total air time. The air time is com￾pared using cached information on the feasible data rates between the three nodes. The feasible data rate is the largest data rate that guarantees a predetermined average error rate threshold for an average channel SNR. •Beyond its normal function, a request to send (RTS) message is also used by CoopMAC to notify the node that has been selected for 1 A preliminary version of the CoopMAC protocol was described in [10]. As wireless network deployments become ever more dense, a reduction of SIR will directly lead to a boost in network capacity. Indeed, the problem of dense deployment has already been reported for IEEE 802.11 b/g networks, which have only three nonoverlapping channels. ERKIP LAYOUT 8/3/06 1:24 PM Page 87

60% 4-4-44-44444在A4△ 50% 日日日日日日日日 510152025303 Number of stations Number of stations Figure 3. Network capacity comparison: a)saturation capacity; b) network capacity gain with respect to 802. 11g ooperation. Moreover, Coop MAC introduces a combining, not supported by any existing wire which is used by the helper to indicate its avail- generation wireless baseband chip. Given the ability after it receives the rtS message from constraint of using existing hardware, we have the source. If the destination hears the HTs developed a backward compatible mode of to resage, it issues a clear to send(CTS)message CoopMAC, which does not perform receiver serve channel time for a two-hop transmis combining and therefore only requires a driver sion. Otherwise, it still sends out the CTs, but or firmware upgrade nly reserves channel time for a direct transmis- Without diversity combining: If no combining capability is supported at the destination, the If both HTS and Cts are received at the packet should be transmitted on both the first source, the data packet should be transmitted to and second hop at the highest physical layer rate the relay first, and then forwarded to the desti- that the respective link can sustain nation by the relay. If the source does not receive With diversity combining: When receiver an HTS, it should then initiate a direct transmis- combining is enabled, the relay now can forward sion to the destination packets at a rate equal to or greater than the A normal ACK is used to acknowledge a one that it adopts in Coop MAC where combi orrect reception, regardless of whether the ing is not possible. More specifically, the trans- packet is forwarded by the relay, or is directly mission rate between the source and relay is transmitted from the source If ne cessa retransmission is attempted, again in a coopera- bility at the relay. Although the destination can- tive fashion not fully decode the packet after the first-hop It is crucial that each node obtains and con- transmission, this received signal will be stored stantly updates its information about the avail- If the relay can successfully receive the packet, it ability of potential relays. The CoopMac then forwards the packet to the destination. The protocol deals with this issue mainly through transmission rate on the second hop is the high maintaining a table called the Coop Table in its est one that meets a predetermined average management plane. Each entry in the CoopT- error rate at the destination, once the destin able corresponds to a potential relay, and con- tion combines the source and relay signals. tains such information as the ID(e. g, 48-bit The diversity combining capability allows MAC address) of the potential relay, the latest CoopMAC to leverage both the spatial diversity time at which a packet from that potential relay and the coding gain, thereby resulting in ever is overheard by the source, and the data rate better performance than the protocol without used for direct transmission between the poten- receiver combining. Using the coded coopera tial relay and destination, and between the cur- tion framework described above, the helper pro- rent node and the potential relay. A set of vides different coded bits than the source protocols have been defined in Coop MAC to leading to a better error performance than rep properly establish, manage, and update the table tition coding It is worthwhile to note that although the Due to the broadcast nature of the channel, protocol architecture and signaling mechanism the destination will receive the signals transmit- defined above are applicable both with and with ted by both the source and the relay. If the desti- out diversity combining at the receiver, the nation is capable of combining these two copies relay-selection scheme may not yield an optimal to decode the original information, then cooper- choice for CoopMAC with receiver combining ative diversity can be fully leveraged. Receiver any longer, because it does not take the possible IEEE Wireless Comm

88 IEEE Wireless Communications • August 2006 cooperation. Moreover, CoopMAC introduces a new message called helper-ready to send (HTS), which is used by the helper to indicate its avail￾ability after it receives the RTS message from the source. If the destination hears the HTS message, it issues a clear to send (CTS) message to reserve channel time for a two-hop transmis￾sion. Otherwise, it still sends out the CTS, but only reserves channel time for a direct transmis￾sion. •If both HTS and CTS are received at the source, the data packet should be transmitted to the relay first, and then forwarded to the desti￾nation by the relay. If the source does not receive an HTS, it should then initiate a direct transmis￾sion to the destination. •A normal ACK is used to acknowledge a correct reception, regardless of whether the packet is forwarded by the relay, or is directly transmitted from the source. If necessary, retransmission is attempted, again in a coopera￾tive fashion. It is crucial that each node obtains and con￾stantly updates its information about the avail￾ability of potential relays. The CoopMAC protocol deals with this issue mainly through maintaining a table called the CoopTable in its management plane. Each entry in the CoopT￾able corresponds to a potential relay, and con￾tains such information as the ID (e.g., 48-bit MAC address) of the potential relay, the latest time at which a packet from that potential relay is overheard by the source, and the data rate used for direct transmission between the poten￾tial relay and destination, and between the cur￾rent node and the potential relay. A set of protocols have been defined in CoopMAC to properly establish, manage, and update the table in a timely manner. Due to the broadcast nature of the channel, the destination will receive the signals transmit￾ted by both the source and the relay. If the desti￾nation is capable of combining these two copies to decode the original information, then cooper￾ative diversity can be fully leveraged. Receiver combining, not supported by any existing wire￾less hardware, can be implemented in the next￾generation wireless baseband chip. Given the constraint of using existing hardware, we have developed a backward compatible mode of CoopMAC, which does not perform receiver combining and therefore only requires a driver or firmware upgrade. Without diversity combining: If no combining capability is supported at the destination, the packet should be transmitted on both the first and second hop at the highest physical layer rate that the respective link can sustain. With diversity combining: When receiver combining is enabled, the relay now can forward packets at a rate equal to or greater than the one that it adopts in CoopMAC where combin￾ing is not possible. More specifically, the trans￾mission rate between the source and relay is chosen so as to guarantee a desired error proba￾bility at the relay. Although the destination can￾not fully decode the packet after the first-hop transmission, this received signal will be stored. If the relay can successfully receive the packet, it then forwards the packet to the destination. The transmission rate on the second hop is the high￾est one that meets a predetermined average error rate at the destination, once the destina￾tion combines the source and relay signals. The diversity combining capability allows CoopMAC to leverage both the spatial diversity and the coding gain, thereby resulting in even better performance than the protocol without receiver combining. Using the coded coopera￾tion framework described above, the helper pro￾vides different coded bits than the source, leading to a better error performance than repe￾tition coding. It is worthwhile to note that although the protocol architecture and signaling mechanism defined above are applicable both with and with￾out diversity combining at the receiver, the relay-selection scheme may not yield an optimal choice for CoopMAC with receiver combining any longer, because it does not take the possible ■ Figure 3. Network capacity comparison: a) saturation capacity; b) network capacity gain with respect to 802.11g. Number of stations 0 5 7 8 Capacity (Mb/s) 9 10 11 12 13 14 10 15 20 (a) (b) 25 30 35 40 Number of stations 5 0% 10% Capacity gain (percentage) 20% 30% 40% 50% 60% 10 15 20 25 30 35 40 CoopMAC with receiver combining CoopMAC without receiver combining 802.11g CoopMAC without receiver combining CoopMAC with receiver combining ERKIP LAYOUT 8/3/06 1:24 PM Page 88

= D CoopMAC without receiver combining MAc with receiver combini 0.04 5% v25 25% 0.02 5% 0 0152025303540 15 a Figure 4 Channel access delay comparison: a)mean channel access delay; b)improvement of mean channel access delay with respect to 802. 11g g rate increase on the second hop into considera- bytes. The MSDU size of 1500 bytes has been tion. In addition, the relay has to be aware of chosen in the simulation because the data pack the average link quality (e. g, average SNr) ets usually assume such a length in wireless between the source and destination so that it can LANS, as widely reported in recent traffic pat second hop. The information can be easily con- MAC-level condition, where each station alway 9 properly select a higher transmission rate on the tern research [14]. Saturation here refers to veyed to the relay by sandwiching a"shim"link has packets to transmit at any time instant. Note, quality field between the legacy MAC header however, that the MAC-level saturation does not and the MAC payload in the data packets that necessarily imply that the physical wireless chan- the source transmits nel is always occupied, as all the stations have to Although CoopMAC bears some superficial perform backoff according to the random-access resemblance to conventional ad hoc routing pro- MAC protocol. trated First and foremost, forwarding in CoopMAC is CoopMAC can achieve a much higher network an essential means to accomplish the goal of capacity than the legacy IEEE 802.11g. Between leveraging cooperative diversity. Secondly, all the two versions of CoopMAC, the one with the associated operations occur in the MAc receiver-combining capability can deliver more layer, which enjoys a shorter response time and throughput, as was anticipated above. more convenient access to the physical layer Another highly desirable feature of Coop information, as compared to the traditional net- MAC that Figs. 3a and 3b reveal is that both the work layer routing. In addition, no channel con- network capacity and the capacity gain for Coop tention is needed when the relay forwards the MAC with respect to 802. 11g increase as the packets to the destination, thereby leading to a number of nodes in the network grows. This shorter delay for the relay and a more efficient improvement primarily stems from the increas- channel utilization as well. Last but not the least, ing availability of relays as the network becomes no routing protocol that we are aware of has more populated adopted the receiver-combining technique to tap For a wide variety of network sizes, Fig 4 into the potential of cooperative diversity portrays the simulation results for the average hannel access delay, which essentially is the THROUGHPUT DELAY, AND duration from the time a packet becomes the ENERGY EFFICIENCY OF COOPMAC head-of-line(hol) packet until the time the packet is successfully transmitted. The corre- To evaluate the performance of CoopMAC, we sponding delay improvement over 802. 11g i have developed an event-driven custom simula- shown in Fig 4b. It is evident that data packets tor using the programming language C to faith- in CoopMAC experience significantly less delay fully model all the critical MAC and physical than in legacy IEEE 802.11g layer features of IEEE 802.11 and CoopMAC. It is also worthwhile to note that the same The parameters in the performance evaluation trend in throughput and delay improvement can assume the default values specified for an ieee be observed for networks operating in a medi 802.11g network operating in a typical office um-to-low-traffic regime In addition, even more environment with low user mobility. mprovement can be achieved when a larger Figures 3-5 depict the simulation results for a frame size is used. Due to space limitations, we saturated network with a payload size of 1500 will not present additional results in this article IEEE Wireless Communications August 2006

IEEE Wireless Communications • August 2006 89 rate increase on the second hop into considera￾tion. In addition, the relay has to be aware of the average link quality (e.g., average SNR) between the source and destination so that it can properly select a higher transmission rate on the second hop. The information can be easily con￾veyed to the relay by sandwiching a “shim” link quality field between the legacy MAC header and the MAC payload in the data packets that the source transmits. Although CoopMAC bears some superficial resemblance to conventional ad hoc routing pro￾tocols, they are in essence completely different. First and foremost, forwarding in CoopMAC is an essential means to accomplish the goal of leveraging cooperative diversity. Secondly, all the associated operations occur in the MAC layer, which enjoys a shorter response time and more convenient access to the physical layer information, as compared to the traditional net￾work layer routing. In addition, no channel con￾tention is needed when the relay forwards the packets to the destination, thereby leading to a shorter delay for the relay and a more efficient channel utilization as well. Last but not the least, no routing protocol that we are aware of has adopted the receiver-combining technique to tap into the potential of cooperative diversity. THROUGHPUT, DELAY, AND ENERGY EFFICIENCY OF COOPMAC To evaluate the performance of CoopMAC, we have developed an event-driven custom simula￾tor using the programming language C to faith￾fully model all the critical MAC and physical layer features of IEEE 802.11 and CoopMAC. The parameters in the performance evaluation assume the default values specified for an IEEE 802.11g network operating in a typical office environment with low user mobility. Figures 3–5 depict the simulation results for a saturated network with a payload size of 1500 bytes. The MSDU size of 1500 bytes has been chosen in the simulation because the data pack￾ets usually assume such a length in wireless LANs, as widely reported in recent traffic pat￾tern research [14]. Saturation here refers to a MAC-level condition, where each station always has packets to transmit at any time instant. Note, however, that the MAC-level saturation does not necessarily imply that the physical wireless chan￾nel is always occupied, as all the stations have to perform backoff according to the random-access MAC protocol. As demonstrated in Fig. 3, both flavors of CoopMAC can achieve a much higher network capacity than the legacy IEEE 802.11g. Between the two versions of CoopMAC, the one with receiver-combining capability can deliver more throughput, as was anticipated above. Another highly desirable feature of Coop￾MAC that Figs. 3a and 3b reveal is that both the network capacity and the capacity gain for Coop￾MAC with respect to 802.11g increase as the number of nodes in the network grows. This improvement primarily stems from the increas￾ing availability of relays as the network becomes more populated. For a wide variety of network sizes, Fig. 4 portrays the simulation results for the average channel access delay, which essentially is the duration from the time a packet becomes the head-of-line (HOL) packet until the time the packet is successfully transmitted. The corre￾sponding delay improvement over 802.11g is shown in Fig. 4b. It is evident that data packets in CoopMAC experience significantly less delay than in legacy IEEE 802.11g. It is also worthwhile to note that the same trend in throughput and delay improvement can be observed for networks operating in a medi￾um-to-low-traffic regime. In addition, even more improvement can be achieved when a larger frame size is used. Due to space limitations, we will not present additional results in this article ■ Figure 4. Channel access delay comparison: a) mean channel access delay; b) improvement of mean channel access delay with respect to 802.11g. Number of stations 5 0 Avera 0.01 ge channel access delay (s) 0.02 0.03 0.04 0.05 0.06 10 15 20 25 (a) (b 30 35 40 Number of stations 5 0% 5% Improvement of average channel access delay (percentage) 10% 15% 20% 25% 30% 35% 40% 45% 50% 10 15 20 25 30 35 40 802.11g CoopMAC without receiver combining CoopMAC with receiver combining CoopMAC with receiver combining CoopMAC without receiver combining ERKIP LAYOUT 8/3/06 1:24 PM Page 89

x10 without receiver combining -ECoo 40% 豆 688880607a 05爸 0510152025303540 510152025303540 Number of stations umber of stations a Figure 5. Energy efficiency comparison: a) average energy consumption per bit per user; b)average user energy efficiency gain with spect to 802.11g for the nonsaturation condition or for larger signal-to-interference-plus-noise ratio (SINR) frame sizes for each point in a cell by randomly choosing the In addition to conventional measures like locations of six interfering nodes in the six proxi throughput and delay, we have also evaluated mal cells and assuming path loss as well as the ene rgy efficiency of CoopMAC, since power Rayleigh fading. The maximum feasible data ways a key concern for wireles rate is estimated based on the sinr and the networks. Figure 5a depicts the energy consump- error rate threshold requirement tion per user in terms of total amount of energy the interference for 802.1 needed to successfully deliver a bit for each user ig MAC and CoopMAC in a multicell environ (i. e, joules/bit/user), which includes the energy ment with a frequency reuse factor of three. All consumed in transmission, reception and chan- three systems are under the same traffic load in nel sensing Figure 5b shows the percentage all cells From these figures, another advantage mprovement with respect to 802.11g. We of cooperation becomes apparent. CoopMAC bserve that as the number of nodes increases, without receiver combining decreases the aver- the improvement in per user energy efficiency age interference by 21.5 percent, while receiver achieved by CoopMAC also grows. This is pri- combining enables another 12.5 percent reduc marily due to the fact that although CoopMAc tion Since both versions of CoopMAC are more requires nodes to receive and retransmit traffic efficient in terms of throughput, the transmission for each other, it also enables them to spend less time for the same amount of traffic using the time listening to the medium. Ultimately, this Coop MAC protocol is less than that of the lega saving outweighs the new energy expense, and cy system, therefore reducing total energy radiat- leads to an increase in energy efficiency. We ed to the network. Due to the lower background refer the readers to [15 for details of the ener- interference, the sustainable regions for all four gy-consumption model rates supported by IEEE 802.lIg are extended [15 INTERFERENCE REDUCTION IN A DENSE NETWORK IMPLEMENTATION The deployment of wireless networks has grown In order to further validate the design of Coop- increasingly dense, leading to concerns that MAC and demonstrate the feasibility of an incre deployments may become interference-limited. mental deployment, we have made efforts to For instance, there are 11 channels defined in implement the CoopMAC protocol using off the 2.4 GHz spectrum for operation of ieee the-shelf IEEE 802.1 Ib network interface cards 802.11 WLANS in the United States. However, (NICs)on a Linux platform [ 15. Since no exist in order to avoid interference between adjacent ing hardware can perform the receiver-combin- cells, only three mutually nonoverlapping chan- ing function, only the Coop MAC protocol nels can be used at the same time without diversity combining can be implemented In the following discussion, we focus only on In fact, due to the constraint in accessing the co-channel interference for a cellular deploy- firmware on the chip, we had to take an emula- ment of IEEE 802.11 with a reuse factor of three. tion approach at the driver level for the Coop- Note that while a node is transmitting packet MAC version without receiver combining, which a particular cell, there will be six proximal cells unfortunately incurs additional protocol over in which parallel transmissions generate co-chan- head. Nevertheless, as demonstrated in the nel interference. Our simulation calculates the experiment, CoopMAC can still reduce the aver IEEE Wireless Communications. Ar

90 IEEE Wireless Communications • August 2006 for the nonsaturation condition or for larger frame sizes. In addition to conventional measures like throughput and delay, we have also evaluated the energy efficiency of CoopMAC, since power conservation is always a key concern for wireless networks. Figure 5a depicts the energy consump￾tion per user in terms of total amount of energy needed to successfully deliver a bit for each user (i.e., joules/bit/user), which includes the energy consumed in transmission, reception and chan￾nel sensing. Figure 5b shows the percentage improvement with respect to 802.11g. We observe that as the number of nodes increases, the improvement in per user energy efficiency achieved by CoopMAC also grows. This is pri￾marily due to the fact that although CoopMAC requires nodes to receive and retransmit traffic for each other, it also enables them to spend less time listening to the medium. Ultimately, this saving outweighs the new energy expense, and leads to an increase in energy efficiency. We refer the readers to [15] for details of the ener￾gy-consumption model. INTERFERENCE REDUCTION IN A DENSE NETWORK The deployment of wireless networks has grown increasingly dense, leading to concerns that deployments may become interference-limited. For instance, there are 11 channels defined in the 2.4 GHz spectrum for operation of IEEE 802.11 WLANs in the United States. However, in order to avoid interference between adjacent cells, only three mutually nonoverlapping chan￾nels can be used at the same time. In the following discussion, we focus only on co-channel interference for a cellular deploy￾ment of IEEE 802.11 with a reuse factor of three. Note that while a node is transmitting packets in a particular cell, there will be six proximal cells in which parallel transmissions generate co-chan￾nel interference. Our simulation calculates the signal-to-interference-plus-noise ratio (SINR) for each point in a cell by randomly choosing the locations of six interfering nodes in the six proxi￾mal cells and assuming path loss as well as Rayleigh fading. The maximum feasible data rate is estimated based on the SINR and the error rate threshold requirement. Figure 6 compares the interference for 802.1 ig MAC and CoopMAC in a multicell environ￾ment with a frequency reuse factor of three. All three systems are under the same traffic load in all cells. From these figures, another advantage of cooperation becomes apparent. CoopMAC without receiver combining decreases the aver￾age interference by 21.5 percent, while receiver combining enables another 12.5 percent reduc￾tion. Since both versions of CoopMAC are more efficient in terms of throughput, the transmission time for the same amount of traffic using the CoopMAC protocol is less than that of the lega￾cy system, therefore reducing total energy radiat￾ed to the network. Due to the lower background interference, the sustainable regions for all four rates supported by IEEE 802.llg are extended [15]. IMPLEMENTATION In order to further validate the design of Coop￾MAC and demonstrate the feasibility of an incre￾mental deployment, we have made efforts to implement the CoopMAC protocol using off￾the-shelf IEEE 802.1 lb network interface cards (NICs) on a Linux platform [15]. Since no exist￾ing hardware can perform the receiver-combin￾ing function, only the CoopMAC protocol without diversity combining can be implemented. In fact, due to the constraint in accessing the firmware on the chip, we had to take an emula￾tion approach at the driver level for the Coop￾MAC version without receiver combining, which unfortunately incurs additional protocol over￾head. Nevertheless, as demonstrated in the experiment, CoopMAC can still reduce the aver- ■ Figure 5. Energy efficiency comparison: a) average energy consumption per bit per user; b) average user energy efficiency gain with respect to 802.11g. Number of stations (a) 0 5 0 0.5 Average user energy consumption per bit (J/b/user) 1 1.5 2 2.5 3 3.5 4 4.5 x10-6 10 15 20 25 30 35 40 Number of stations (b) 5 0% 10% Improvement in average user energy consumption per bit (percentage) 20% 30% 40% 50% 10 15 20 25 30 35 40 CoopMAC without receiver combining CoopMAC with receiver combining 802.11g CoopMAC without receiver combining CoopMAC with receiver combining ERKIP LAYOUT 8/3/06 1:24 PM Page 90

To further validate the design of CoopMAC and demonstrate the feasibility of we hove made efforts to implement a Figure 6. Interference(W): frequency reuse factor =3, traffic= 500 p/s, transmission power=luW. a) the CoopMAC protocol using of the-shelf ieee age file-transfer times significantly below the ad hoc networks. Indeed, the notions of re original value, and a more significant perfor- and routing protocols may change when 802 1 b network mance improvement can be achieved when the ation is fully integrated in the link layer. Cooper entire CoopMAC without receiver combining is ative partners should be carefully selected along interface cards completely implemented in firmware In addi- the route so that optimality at both the link and tion,an even higher performance gain would be path levels can be accomplished, while spatial (NICs)on a LinUx possible if the CoopMAC with receiver combi reuse in ad hoc networks is not compromised form ng can be realized in a baseband chip Similar subtle cross-layer design issues abound in ad hoc networks, and the implications of node CONCLUSION AND FUTURE WORK cooperation, including cooperative routing alge rithms and the scalability of network capacity y introducing collaboration from nodes that with the number of node les in a network. deserve otherwise do not directly participate in trans- further investigat mission, cooperative communication introduces a new paradigm for wireless communication. It REFERENCES enables a tremendous improvements in robust- (11G. Kramer, M. Gastpar, and P. Gupta, "Ce ness, throughput, and delay; a significant reduc- tion in interference: and an extension o Trans. Info. Theory, vol. 51, Sept coverage range. To fully leverage the concept (2)T Cover and A EGamal, "Capacity Theorems for the Relay Channel, "IEEE Trans. Info. Theory, vol. IT-25 from physical layer to networking protocols- To illustrate the necessity of a cross-layer (414 Sendonaris, E. Erkip, and B. Aazhang, "Us d have explored coop at both layers 1 and 2 of the OSI protocol stack, mance Analysis, "IEEE Trans. Commun., vol. 51, Now IEEE 802.11 networks which we call CoopMAC (5)J N Laneman, D Tse, and G W Wornell, "Cooperative In particular, the Coop MAC protocol has an ks: Efficient Protocols and Outage Behavior, "IEEE Trans. Info. Theory, vol. 50. option to enable the capability of diversity com- Dec.2004,pp.3062-80 bining at the receiver, where two versions of the [6 A Stefanov and E. Erkip, "Cooperative Coding for Wire- same data are jointly decoded to recover the ss Networks, "IEEE Trans. Commun. vol 52, Sept 1470-76. original packet. As verified by extensive simula- [7] T. E. Hunter and A. Nosratinia, "Diversity through tions, the CoopMAC protocol, both with and Coded Cooperation. "IEEE Trans. Wireless Commun. without receiver-combining capability, ca vol.5,Feb.2006,pp.283-89 achieve substantial performance improvements [8]J. N. Laneman and G. w. wornel ributed Space- erative dive without incurring appreciable additional com- plexity in system implementation. Compared oct2003,pp.241525. with the noncombining version, the CoopMAC [91H. Shutoy,Y Wang, and E.Erkip, "Cooperative Source protocol with receiver-combining capability Proc. IEEE Int'l Conf Image Processing, Atlanta, GA, pushes cooperation to an even higher level and eaps additional benefits 10]P and S. Panwar, "A Cooperative MAC Pr To further exploit cooperation gains at the tocol for wireless Local Area Networks, Proc. IEEE ICC network layer for highly adaptive and scalable ad [11]B Zhao and M. C. valenti, "Practical Relay Networks hoc networks, many research challenges remain alization of Hybrid-ARQ. "IEEE JSAC, voL. 23, Given the increasing number of cooperating nodes listening to each transmission, packet for- 112) G. Jakilari et al. A Framework for Distribute warding can now be done in a more opportunis- works. "Proc. IEEE INFOCOM tic way than has been traditionally considered Apr.2006. IEEE Wireless Communications August 2006

IEEE Wireless Communications • August 2006 91 age file-transfer times significantly below the original value, and a more significant perfor￾mance improvement can be achieved when the entire CoopMAC without receiver combining is completely implemented in firmware. In addi￾tion, an even higher performance gain would be possible if the CoopMAC with receiver combin￾ing can be realized in a baseband chip. CONCLUSION AND FUTURE WORK By introducing collaboration from nodes that otherwise do not directly participate in trans￾mission, cooperative communication introduces a new paradigm for wireless communication. It enables a tremendous improvements in robust￾ness, throughput, and delay; a significant reduc￾tion in interference; and an extension of coverage range. To fully leverage the concept of cooperation, the entire protocol stack — from physical layer to networking protocols — should be carefully reengineered or even redesigned. To illustrate the necessity of a cross-layer design approach, we have explored cooperation at both layers 1 and 2 of the OSI protocol stack, and have proposed a new MAC protocol for IEEE 802.11 networks which we call CoopMAC. In particular, the CoopMAC protocol has an option to enable the capability of diversity com￾bining at the receiver, where two versions of the same data are jointly decoded to recover the original packet. As verified by extensive simula￾tions, the CoopMAC protocol, both with and without receiver-combining capability, can achieve substantial performance improvements, without incurring appreciable additional com￾plexity in system implementation. Compared with the noncombining version, the CoopMAC protocol with receiver-combining capability pushes cooperation to an even higher level and reaps additional benefits. To further exploit cooperation gains at the network layer for highly adaptive and scalable ad hoc networks, many research challenges remain. Given the increasing number of cooperating nodes listening to each transmission, packet for￾warding can now be done in a more opportunis￾tic way than has been traditionally considered in ad hoc networks. Indeed, the notions of routing and routing protocols may change when cooper￾ation is fully integrated in the link layer. Cooper￾ative partners should be carefully selected along the route so that optimality at both the link and path levels can be accomplished, while spatial reuse in ad hoc networks is not compromised. Similar subtle cross-layer design issues abound in ad hoc networks, and the implications of node cooperation, including cooperative routing algo￾rithms and the scalability of network capacity with the number of nodes in a network, deserve further investigation. REFERENCES [1] G. Kramer, M. Gastpar, and P. Gupta, “Cooperative Strategies and Capacity Theorems for Relay Networks,” IEEE Trans. Info. Theory, vol. 51, Sept. 2005, pp. 3037–63. [2] T. Cover and A. E. Gamal, “Capacity Theorems for the Relay Channel,” IEEE Trans. Info. Theory, vol. IT-25, Sept. 1979, pp. 572–84. [3] A. Sendonaris, E. Erkip, and B. Aazhang, “User Cooper￾ation Diversity Part I: System Description,” IEEE Trans. Commun., vol. 51, Nov. 2003, pp. 1927–38. [4] A. Sendonaris, E. Erkip, and B. Aazhang, “User Cooper￾ation Diversity Part II: Implementation Aspects and Per￾formance Analysis,” IEEE Trans. Commun., vol. 51, Nov. 2003, pp. 1939–48. [5] J. N. Laneman, D. Tse, and G. W. Wornell, “Cooperative Diversity in Wireless Networks: Efficient Protocols and Outage Behavior,” IEEE Trans. Info. Theory, vol. 50, Dec. 2004, pp. 3062–80. [6] A. Stefanov and E. Erkip, “Cooperative Coding for Wire￾less Networks,” IEEE Trans. Commun., vol. 52, Sept. 2004, pp. 1470–76. [7] T. E. Hunter and A. Nosratinia, “Diversity through Coded Cooperation,” IEEE Trans. Wireless Commun., vol. 5, Feb. 2006, pp. 283–89. [8] J. N. Laneman and G. W. Wornell, “Distributed Space￾Time-Coded Protocols for Exploiting Cooperative Diver￾sity in Wireless Networks,” IEEE Trans. Info. Theory, vol. 49, Oct. 2003, pp. 2415–25. [9] H. Shutoy, Y. Wang, and E. Erkip, “Cooperative Source and Channel Coding for Wireless Video Transmission,” Proc. IEEE Int’l. Conf. Image Processing, Atlanta, GA, Oct. 2006 (to appear). [10] P. Liu, Z.Tao, and S. Panwar, “A Cooperative MAC Pro￾tocol for Wireless Local Area Networks,” Proc. IEEE ICC, June 2005. [11] B. Zhao and M. C. Valenti, “Practical Relay Networks: A Generalization of Hybrid-ARQ,” IEEE JSAC, vol. 23, Jan. 2005, pp. 7–18. [12] G. Jakilari et al., “A Framework for Distributed Spa￾tiotemporal Communications in Mobile Ad Hoc Net￾works,” Proc. IEEE INFOCOM 2006, Barcelona, Spain, Apr. 2006. ■ Figure 6. Interference (W): frequency reuse factor = 3, traffic = 500 p/s, transmission power = l µW: a) 802.11g; b) CoopMAC without receiver combining; and c) CoopMAC with receiver combining. x10-7 2 1.8 1.6 1.4 1.2 1 (a) (b) (c) To further validate the design of CoopMAC and demonstrate the feasibility of an incremental deployment, we have made efforts to implement the CoopMAC protocol using off-the-shelf IEEE 802.1 lb network interface cards (NICs) on a Linux platform ERKIP LAYOUT 8/3/06 1:24 PM Page 91

Cooperative partners [13]H Zhu and G. Cao, " rDCF: A Relay-Enabled Medium ELZA ERKIP [ 93, M 96, SM,o5](e fireless Ad Hoc Network should be carefully i4 he e . ozoussefa fid h e wareless sihe ra teri ing degr rech selected along the Turkey, in iggr ing from Middle route so that 2004,http://www.csumd.edu/moustafa/papers/cs-ir.associatep6-199gshewaseringofRiceUniversityshe of electrical and computer engineenng [15]P Liu et al, "CoopMAC: A Cooperative MAC for Wire. optimality at both lessLanS,"tech.rep.,http://catt.polyedu/catt/pan d the 2004 C the link and path BIOGRAPHIES levels can be PEL LIU[S01](pliu@ photon poly. edu) completed his B.S. accomplished, while wireless networks SHIVENDRA S. PANWAR [S 82, M85, SM,] hoc networks is not ZHIFENG TAo (Soo] Jeff. tao@ photon poly. edu) received a compromised from Xi'an Jiaotong University, P R China, in 2000. Since Amherst, in 1983 and 1986, respectively. He joined the Electrical and Comp communications cations(CATT). He spent the summer of 1987 as a visiting scientist at the IBM T J Watson Research Center. Yorktown ZINAN LIN[500) (l ino3@utopia- poly.edu)received a BE ries, Holmdel, N. His research interests include the perfo cludes video systems over peer-to-peer networks, switch (NTU), Singapore, in 2001. From 199 p8t02 000 she wa ireless networks. He has served as secre awarded a research scholarship by NTU and worked as ry of the Technical Affairs Council of the IEEE Communica nt in the Center for tions Society(1992-1993)and is a member of the Technical School of Electronic and Electrical Engineering, NTU. Since 2001, she has been Ph. D. candidate in the Department of of two books, Network Management and Control, Vol. ll ty, Brooklyn, NY. Her general 994 and 1996, respectively), and co-author of TCP/IP Essen- ials: A Lab-Based Approach(Camb University Press, especially channel coding, diversity techniques, and CDMA Society Leonard G. Abraham Prize in the Field of Commun cations System IEEE Wireless Comm

92 IEEE Wireless Communications • August 2006 [13] H. Zhu and G. Cao, “rDCF: A Relay-Enabled Medium Access Control Protocol for Wireless Ad Hoc Networks,” Proc. IEEE INFOCOM, Mar. 2005. [14] J. Yeo, M. Youssef, and A. Agrawala, “Characterizing the IEEE 802.11 Traffic: The Wireless Side,” Univ. of Maryland, College Park, res. rep. CS-TR-4570, Mar. 2004, http://www.csumd.edu/moustafa/papers/CS-IR- 4570.pdf [15] P. Liu et al., “CoopMAC: A Cooperative MAC for Wire￾less LANs,” Tech. rep., http//catt.poly.edu/CATT/pan war.html BIOGRAPHIES PEL LIU [S’01] (pliu@photon.poly.edu) completed his B.S. and M.S. degrees in electrical engineering at Xi’an Jiaotong University, China, in 1997 and 2000, respectively. He is a Ph.D. candidate in the Department of Electrical and Com￾puter Engineering of Polytechnic University, Brooklyn, NY. His research interests are in wireless communications and wireless networks. ZHIFENG TAO [S’00] (jeff.tao@ photon.poly.edu) received a B.E. degree in communication and information engineering from Xi’an Jiaotong University, P. R. China, in 2000. Since then, he has been a Ph.D. candidate in the Department of Electrical and Computer Engineering at Polytechnic Univer￾sity. He also received his M.S. degree in telecommunication networking from Polytechnic University in May 2002. His current research interests include wireless networking, medium access control, quality of service, and cooperative communications. ZINAN LIN [S’00] (zlin03@utopia.poly.edu) received a B.E degree in information engineering from Zhejiang Universi￾ty, Hangzhou, China, in 1998, and an M.E. degree in elec￾trical engineering from Nanyang Technological University (NTU), Singapore, in 2001. From 1998 to 2000 she was awarded a research scholarship by NTU and worked as a research assistant in the Center for Signal Processing, School of Electronic and Electrical Engineering, NTU. Since 2001, she has been Ph.D. candidate in the Department of Electrical and Computer Engineering, Polytechnic Universi￾ty, Brooklyn, NY. Her general research interests include wireless communications and digital signal processing, especially channel coding, diversity techniques, and CDMA sequence design. ELZA ERKIP [S’93, M’96, SM’05] (e1za@poly.edu) received M.S. and Ph.D. degrees in electrical engineering from Stan￾ford University in 1993 and 1996, respectively, and a B.S. degree in electrical and electronic engineering from Middle East Technical University, Turkey, in 1990. She joined Poly￾technic University in Spring 2000, where she is currently an associate professor of electrical and computer engineering. During 1996–1999 she was with the Department of Electri￾cal and Computer Engineering of Rice University. She received the 2004 Communications Society Stephen O. Rice Paper Prize in the field of communications theory, the NSF CAREER award in 2001, and the IBM Faculty Partnership Award in 2000. She is the Technical Program Co-Chair of the 2006 Communication Theory Workshop. She is also an Associate Editor of IEEE Transactions on Communications and a Publications Editor for IEEE Transactions on Informa￾tion Theory. Her research interests are in wireless commu￾nications, information theory, and communication theory. SHIVENDRA S. PANWAR [S’82, M’85, SM’00] (panwar@catt.poly.edu) received a B.Tech. degree in electri￾cal engineering from the Indian Institute of Technology, Kan￾pur, in 1981, and M.S. and Ph.D. degrees in electrical and computer engineering from the University of Massachusetts, Amherst, in 1983 and 1986, respectively. He joined the Department of Electrical Engineering at the Polytechnic Insti￾tute of New York, Brooklyn, NY (now Polytechnic University), where he is a professor in the Electrical and Computer Engi￾neering Department. Currently he is the director of the New York State Center for Advanced Technology in Telecommuni￾cations (CATT). He spent the summer of 1987 as a visiting scientist at the IBM T. J. Watson Research Center, Yorktown Heights, NY, and was a consultant to AT& T Bell Laborato￾ries, Holmdel, NJ. His research interests include the perfor￾mance analysis and design of networks. Current work includes video systems over peer-to-peer networks, switch performance, and wireless networks. He has served as Secre￾tary of the Technical Affairs Council of the IEEE Communica￾tions Society (1992–1993) and is a member of the Technical Committee on Computer Communications. He is a co-editor of two books, Network Management and Control, Vol. II, and Multimedia Communications and Video Coding (Plenum, 1994 and 1996, respectively), and co-author of TCP/IP Essen￾tials: A Lab-Based Approach (Cambridge University Press, 2004). He is a co-recipient of the 2004 IEEE Communications Society Leonard G. Abraham Prize in the Field of Communi￾cations Systems. Cooperative partners should be carefully selected along the route so that optimality at both the link and path levels can be accomplished, while spatial reuse in ad hoc networks is not compromised. ERKIP LAYOUT 8/3/06 1:24 PM Page 92