Performance analysis of distributed cluster-based MAC protocol for multiuser MIMO wireless networks

Transcription

1 RESEARCH Open Access Performance analyss of dstrbuted cluster-based MAC protocol for multuser MIMO wreless networks Azadeh Ettefagh *, Marc Kuhn, Celal Eşl and Armn Wttneben Abstract It s known that multuser multple-nput multple-output (MIMO) communcaton can enhance the performance of wreless networks. It can substantally ncrease the spectral effcency of wreless networks by utlsng multuser nterference rather than avodng t. Ths paradgm shft has most mpact on the medum access control (MAC) protocol because most exstng MAC protocols are desgned to reduce the nterference. In ths artcle, we propose a novel cluster-based carrer sense multple access wth collson avodance (CB-CSMA/CA) scheme. The proposed scheme enables multuser MIMO transmssons n wreless local area networks (WLANs) by utlsng the multuser nterference cancellaton capablty of the physcal layer. In ths artcle we focus on the performance analyss of CB- CSMA/CA. We nvestgate saturaton throughput applyng optmum backoff parameters and n the presence of synchronsaton errors. Furthermore, we study the mpact of dfferent clusterng methods on non-saturaton throughput. We show that CB-CSMA/CA mproves throughput sgnfcantly compared to the CSMA/CA scheme used n the IEEE system. It s a promsng approach for a varety of network confguratons ncludng typcal nfrastructure WLANs as well as many other wreless cooperatve networks. Keywords: Cluster-based CSMA/CA, Cooperatve wreless networks, MIMO, WLAN, Infrastructure network, Clusterng I. Introducton Multuser multple-nput multple-output (MIMO) communcaton s an effectve approach to mprove the performance of a wreless system by realsng dstrbuted spatal multplexng and/or dversty gans. However, the current WLAN MIMO standard, IEEE n, supports only pont-to-pont MIMO lnks [1]. The IEEE n specfes the medum access control (MAC) and physcal (PHY) layers to enhance the data rate of wreless local area networks (WLANs) usng MIMO technques. The man channel access method of the IEEE MAC protocol,.e. the dstrbuted coordnaton functon (DCF), s based on carrer sense multple access wth collson avodance (CSMA/CA) [2]. Whle multuser MIMO technques can provde spatal multplexng gansevennnetworkswthsngle-antenna statons (STAs), DCF prevents ths by a collson avodance mechansm. Therefore, n order to realse dstrbuted spatal multplexng gans, the nterference avodance * Correspondence: ettefagh@nar.ee.ethz.ch Communcaton Technology Laboratory, ETH Zurch, 8092, Swtzerland scheme should be modfed to facltate controlled nterference. Contrbuton of ths work Frstwebreflydevseanovelcluster-based MAC as extenson of CSMA/CA to utlse the multuser nterference mtgaton capablty of PHY. The proposed scheme s smple and can be appled to dfferent nfrastructure or ad hoc networks wth or wthout relays. Accordng to the proposed cluster-based CSMA/CA (CB-CSMA/CA) scheme, nodes n a network, ncludng statons (STAs) and access pont (AP) or relays, are allocated to dfferent clusters. The nodes belongng to the same cluster are allowed to transmt at the same tme, provded that there are enough degrees of freedom [3] avalable to effcently decode the desred stream at the destnaton. In ths artcle, we focus on the MAC performance analyss. We nvestgate the performance bounds of CB- CSMA/CA for a representatve applcaton: a network whch conssts of a multple-antenna AP and several 2011 Ettefagh et al; lcensee Sprnger. Ths s an Open Access artcle dstrbuted under the terms of the Creatve Commons Attrbuton Lcense ( whch permts unrestrcted use, dstrbuton, and reproducton n any medum, provded the orgnal work s properly cted.

2 Page 2 of 14 sngle-antenna STAs. We present comprehensve performance results of the consdered applcaton and of a reference system, whch s based on the IEEE n standard. As t wll be explaned n Secton III, n a network wth an AP whch has four antennas the uplnk throughput of CB-CSMA/CA s about 2.5 tmes hgher than that of IEEE n. In order to fnd out how the proposed protocol performs n realstc scenaros we nvestgate the throughput n dfferent stuatons. We estmate performance bounds of the CB-CSMA/CA protocol under dfferent condtons. We study the maxmum throughput, whch can be obtaned by properly adjustng the backoff parameters. Then we focus on performance analyss n realstc scenaros where for example the protocol fals to meet ts basc requrements due to synchronsaton errors. We denote ths problem by the term event-synchronsaton error. We evaluate the mpact of event-synchronsaton errors on the network throughput under saturaton condton analytcally and by smulatons. Furthermore, we evaluate the non-saturaton throughput and nvestgate mpact of dfferent clusterng methods on throughput. In addton to the IEEE reference, we compare the results wth another multuser protocol whch has been proposed n [4]. Related Work Several papers consder the MAC enhancement to support concurrent multuser transmssons [4-8] or to resolve collsons [9,10]. However, as t wll be explaned n more detal n the followng paragraphs, the proposed CB-CSMA/CA dffers from the exstng methods n several ways, ncludng the backoff procedure. Park et al. [5] have proposed a MAC protocol mtgatng nterference by usng multple antennas at the recevers. In [8], the authors proposed a MAC protocol whch consders the spatal correlaton between the sgnal and nterference to decde whether multple lnks should transmt smultaneously or not. In ther proposal, lnks are allowed to contend for the channel sequentally whle transmttng data packets smultaneously. The throughput of the multpacket recepton n a WLAN has been nvestgated n [4], where the authors consder an uplnk scenaro wth a multple-antenna AP. In ther scenaro smultaneous multuser transmssons up to a certan number can be decoded by the AP and hence are not consdered as collson. However, n ther work, contrary to CB-CSMA/CA, there s no systematc way to explot multuser transmssons. A MAC protocol wth antenna arrays s suggested for ad hoc networks n [6]. The antenna array s used to null the nterference from neghbours. In ths scheme, bandwdth s dvded nto two orthogonal channels: a data and a control channel. The nodes use a CSMA/ CA-based scheme on the control channel to acqure the channel and learn about ongong transmssons by montorng ths channel. Smlarly, concurrent transmssons from dfferent users may be allowed by applyng the busy-tone medum access [7]. In ths method, out of band busy-tones on control channels are used to notfy nodes about the number of actual streams and hence, avod collsons. In [9], the authors suggest a MAC protocol wth the ablty to resolve collsons. Accordng to ths protocol, when collsons occur, the collded packets have to be buffered. In the subsequent tme slots, a set of nodes behave as non-regeneratve relays and forward the sgnal they have receved durng the collson slot one by one. Ths method has been extended and appled to a largescale wreless sensor network by dvdng the network nto several clusters [10]. In [10] each cluster has a clusterhead, and the nodes n a cluster only communcate wth the clusterhead. Collsons wthn a cluster are resolved as n [9], whle the clusterhead acts lke a base staton. Contrary to [9] and [10], n whch the authors focus on the dversty technques, we focus on utlsng the spatal multplexng gan n CSMA/CA-based systems. In [11], we have presented the basc dea of CB- CSMA/CA and appled t to an ad hoc scenaro where several amplfy-and-forward relays performed multuser nterference cancellaton. Ths has been done by choosng approprate relay gan allocatons. However, n [11] we have only consdered a smple scenaro where all STAs where perfectly event-synchronsed and always had packets ready to be transmtted. The CB-CSMA/CA scheme has some dstnct advantages compared to exstng proposals: t requres nether sequental contenton per node for the data transmsson as t s needed n [5] and [8], nor a control channel as n [6,7], or [9]. Besdes, by groupng nodes nto clusters, the collson probablty of the CB-CSMA/CA depends on the number of contendng clusters rather than the total number of contendng statons. As t has been shown n [11], the CB-CSMA/CA collson probablty s consderably reduced compared to the standard IEEE MAC. Furthermore, n contrast to [4] where there s a multpacket transmsson only f accdentally more than one STA transmt n a tme slot, the CB- CSMA/CA s desgned such that each transmsson attempt contans as many data packets as possble. Outlne In Secton II, we present the proposed CB-CSMA/CA scheme. In Secton III we descrbe a representatve applcaton whch s consdered throughout ths artcle as well as reference systems. Furthermore, we brefly present the model on whch our analyss s based. We

3 Page 3 of 14 nvestgate the CB-CSMA/CA performance bounds n Secton IV n whch we analyse throughput for optmum backoff wndow parameters, asynchronous case, where STAs n a cluster transmt ndependently, and nonsaturaton scenaros where dfferent clusterng methods are appled. Conclusons are drawn n Secton V. II. Cluster-based CSMA/CA protocol Assume a network where multple users can transmt ndependent data streams smultaneously and the destnaton nodes can effcently decode the desred sgnal out of multple streams. Ths s possble f there are enough degrees of freedom avalable at the destnatons, e.g. each destnaton has enough antennas, or approprate cooperaton exsts among nodes. In such a network, to enhance the spectral effcency, multple nodes should be able to transmt smultaneously. In order to do so, we classfy the nodes n a network nto clusters. The nodes belongng to the same cluster access the channel at the same tme. Smlarly, nodes belongng to the same cluster may receve smultaneous streams provded that the ntended stream at each destnaton can be decoded. Consequently, the spectral effcency for the gven system mproves. Smlar to multple transmssons from a cluster, a cluster may receve several smultaneous streams f the multuser nterference can be cancelled, ether drectly at the destnaton, or for example by settng proper gan factors at the transmtters or at the amplfy-and-forward (AF) relays. Hence, from the MAC layer pont of vew, the clusters replace the ndvdual nodes, and the nodes belongng to the same cluster look lke a sngle entty. Consequently, the probablty of collson n the network can be substantally reduced. We consder three types of clusters: source-, destnaton- and relay-clusters. In a system wth V clusters n total, we denote the vth cluster by C v,wherev Î {1, 2,..., V }. A cluster whch generates the data packets s called the source-cluster and denoted by C sv. The sze of the source-cluster s lmted by the maxmum number of concurrent streams whch can be effcently decoded. A cluster whch ncludes the target recevers s a destnaton-cluster and denoted by C dv. Even though formng a destnaton-cluster s not always necessary, for smplcty we assume that the destnatons of a gven sourcecluster belong to a sngle destnaton-cluster. It should be noted that a source-cluster and ts destnaton-cluster may nterchange ther functons at dfferent tme slots. A relay-cluster, denoted by C rv, s a cluster of relays whch receves and forwards data to other clusters wthout prvate beneft. Clusters can be formed based on a certan applcaton or a gven structure. For example, n an nfrastructure network wth a multple-antenna AP, the AP forms clusters of STAs such that t can decode multple streams n the uplnk (for example by successve nterference cancellaton), or separates them n downlnk transmssons (for example by performng zero-forcng). Snce n the nfrastructure mode all STAs assocate wth the AP, t s aware of each new STA enterng ts network or any STA leavng t. The AP can cluster nodes such that neghboured STAs belong to a sngle cluster or as t wll be explaned n Secton IV-C adaptvely allocate STAs whch have packets ready to be transmtted to one cluster. The maxmum sze of the clusters s determned by the number of antennas at the AP. The AP tself consttutes a sngle-member cluster C r. Fgure 1 depcts a scenaro where the AP has four antennas and each source- and destnaton-cluster conssts of four sngle-antenna STAs. Throughout ths artcle we consder a wreless network where STAs belong to the same basc servce set (BSS). a Furthermore, for practcal reasons, we consder only half-duplex nodes whch are not able to transmt and receve smultaneously. We focus on the data transmsson phase and assume that clusters are defned a pror. A. Bascs of CB-CSMA/CA Accordng to the CB-CSMA/CA, the nodes n a cluster behave smlar to a sngle node,.e. they access the channel at the same tme and transmt smultaneously. Therefore, assumng the DCF access method s used, we need to consder two major modfcatons to the current backoff procedure: () havngthesamentalbackoff duraton for all cluster members and () updatng ths value at the same tme. The frst requrement can be acheved for example by havng the same random generator seed n each cluster, so that the same pseudo-random numbers are generated. In nfrastructure networks the AP nforms STAs n a cluster about the ntal value C s1 C s2 S7 S2 S4 S1 S3 S5 S6 S8 C D3 1 r D1 AP D7 D4 D2 C d C d2 D5 D6 D8 Fgure 1 An nfrastructure network wth four antennas at the AP and 8 source-destnatons pars. AP acts as relay for transmssons from sources to destnatons.

4 Page 4 of 14 of the contenton wndow (CW) and, as explaned n Secton II-C, nforms STAs n case they need to update t. As a second requrement, the STAs wthn a cluster should be event-synchronous. b Ths s necessary snce accordng to the IEEE , the CW doubles after any unsuccessful transmsson up to a maxmum value [2]. Thus, after a transmsson, STAs n a cluster should all know whether they need to ncrease the CW or not, as explaned n the next subsecton. In ths way, each node n a cluster contends only wth other clusters and not wth other nodes n the same cluster. We assume that the clusters are constructed n a way that the destnatons are able to decode the ntended message. Consequently, smultaneous transmssons of the STAs wthn a cluster are resolved and do not lead to collsons. The clusters access the channel one after the other n a smlar way as sngle nodes act n the current CSMA/ CA systems. As t wll be shown n the next secton, the event-synchronsaton error leads to a throughput loss but the CB-CSMA/CA s relatvely robust to such errors. In the worst case scenaro, when all cluster members are not synchronous anymore, the CB-CSMA/ CA performance gets close to that of the standard CSMA/CA. c In a sngle-nput sngle-output (SISO) WLAN system, the header of each transmtted frame can be deally decoded by any node whch s n the communcaton range of the transmtter. Accordngly nodes wthn a WLAN learn about an ongong transmsson by decodng the headers whch nclude the duraton feld. However, the stuaton changes n a CB-CSMA/CA system, where several frames mght be transmtted n parallel. The system s structured n such a way that each destnaton s able to decode ts ntended packet s header as well as ts payload. However, other nodes n the network may not be able to do so. In order to solve ths problem, a cluster common preamble should be sent pror to the data transmsson. The common preamble, whch s sent at the lowest rate, could have a smlar format as that of the legacy preamble of the PLCP header used n the IEEE n mxed format [1]. d It addtonally ncludes the address of the source-cluster and ts LENGTH feld should be set to the maxmum length of all concurrent packets. The maxmum value can be ether set to the value specfed by the standard for all transmssons or adaptvely set by the AP accordng to the actual traffc and cluster structure for each cluster. In the latter case the AP nforms the STAs about ths value once t forms the cluster. B. PHY layer requrements The exstng IEEE n supports only MIMO pontto-pont lnks. However advanced sgnal processng technques have enabled dstrbuted multuser MIMO transmssons. In order to effcently decode an ntended data packet n the presence of multuser transmssons, the nterference from other nodes has to be cancelled out. Ths can be done smlar to pont-to-pont MIMO networks by successve or parallel nterference cancelaton. The man dfference s that here ndvdual nodes partcpate n each transmsson and form some sort of a vrtual antenna array. In the current WLAN, the recevers estmate the channels by recevng the tranng sequence n the PLCP preamble of each frame. In CB-CSMA/CA, each transmsson attempt may consst of several smultaneous streams. Therefore, orthogonal tranng sequences are allocated to dfferent STAs wthn a source-cluster. The orthogonal tranng sequences can be desgned as proposed n the IEEE n for MIMO transmssons, where for each transmt chan, the tranng feld s cyclcally shfted [1]. It should be noted that n OFDM systems, as long as the sum of the maxmum delay offset and the channel delay spread s shorter than the cyclc prefx, both the multuser nterference and the nter-symbol nterference can be mtgated. Snce we focus on ndoor scenaros the maxmum channel delay spread and dstances between nodes are qute small. As, n ths artcle, our man goal s to analyse the performance of MAC layer, we assume an deal PHY layer wth perfect tme and frequency synchronsaton and no channel or decodng errors. C. Collson detecton In WLANs, packets may fal due to collsons and channel errors. The cause of falure, however, cannot be dstngushed n current WLAN systems [12,13]. Hence, n both cases, the backoff nterval ncreases exponentally up to a maxmum value whch s specfed by the standard. However, n the event of a channel error, the bnary exponental backoff may reduce effcency and cause unfarness. An nherent advantage of our proposed CB- CSMA/CA s ts ablty to dfferentate between these losses n practcal SNR regmes. As explaned n the next paragraph, ths feature can be optonally used to enhance the performance. We assume that all nodes n a BSS are n the rado range of each other. At each transmsson attempt a common preamble s transmtted by all members of the source-cluster. The common preamble s ncluded n the frame header. It s short and transmtted at the lowest data rate. Therefore, t s assumed that t s errorfree. However, as dfferent clusters send dfferent common preambles, the common preambles collde f more than one cluster transmts n a tme slot. Accordngly, upon recepton of the frame, collsons can be postdetected as explaned n the followng paragraph.

5 Page 5 of 14 If the common preamble cannot be decoded, t can be assumed that a collson has occurred. However, f the common preambles collde, the nformaton about nvolved clusters may not be obtaned. Hence, a short packet, called contenton wndow update request (CWUR) packet, should be broadcasted to all clusters. In nfrastructure networks ths s performed by the AP. Upon recepton of CWUR all source-clusters enter the next backoff process. However, only the source-clusters whchhavebeennvolvednthependngtransmsson ncrease ther CW unless the maxmum CW value has already been reached. In the latter case they keep the maxmum CW for the upcomng transmsson. Fgure 2 shows the uplnk transmsson n an nfrastructure network, where the AP has two antennas and STA 1 Fgure 2 CB-CSMA/CA access mechansm n an nfrastructure network wth a multple-antenna AP as the recever. In(a) an example of a successful transmsson s shown, where STA 1 and STA 2 n C s1 have a shorter backoff than that of STAs n C s2 and they transmt frst. In (b) both clusters begn to transmt at the same tme and hence collson occurs. a b and STA 2 belong to cluster C s1, whle STA 3 s a member of cluster C s2.infgure2(a)anexampleofasuccessful transmsson s shown, where STAs n C s1 have a shorter backoff than STAs n C s2, hence they transmt frst. In Fgure 2(b) both clusters begn to transmt at the same tme and hence a collson occurs. In ths fgure superscrpt c above the data ndcates the source-cluster from whch the data packet s orgnated whle subscrpt j below the data or ACK specfes the respectve STA. III. Outlne of representatve applcatons and the analytcal model In ths secton frst we ntroduce a CB-CSMA/CA applcaton whch s consdered throughout ths artcle. Furthermore, we explan the reference applcaton, operatng based on the standard, and another multuser transmsson protocol. Then we brefly revew the exstng model for calculatng the throughput of a CSMA/ CA-based network and pont out the modfcatons whch have to be consdered for CB-CSMA/CA applcatons. A. Applcatons The proposed CB-CSMA/CA has wde applcablty to dfferent cooperatve wreless networks. Approprate applcatons nclude cooperatve scenaros lke multuser zero-forcng relayng (MUZFR) [14] and two-way relayng [15], whch requre multple STAs to transmt smultaneously and are not supported by the current WLAN MAC. Besdes, any multuser WLAN scenaro that s supported by the standard MAC, s more effcent under the CB-CSMA/CA f multuser nterference can be cancelled. In ths artcle we focus on nfrastructure networks where the AP s equpped wth multple antennas. For theanalyssweonlyconsderuplnktransmssons. Hence, the AP acts as the recever and all other STAs transmt to the AP. The ad hoc case where several relays assst the communcaton between sources and destnatons s explaned n [11]. 1) CB-CSMA/CA applcaton In ths artcle we use a representatve applcaton to quantfy the performance of the CB-CSMA/CA scheme. Ths applcaton s an uplnk scenaro n an nfrastructure network wth a multple-antenna AP. The system conssts of an AP wth N a antennas and n STAs, each equpped wth a sngle antenna. Hence, at each transmsson attempt N a STAs should be able to transmt n parallel. In order to do so, sources are grouped nto clusters and operate based on CB-CSMA/CA as explaned n Secton II. Although we only focus on uplnk n ths artcle, multple transmssons n downlnk can be supported by the CB-CSMA/CA protocol n a smlar way.

6 Page 6 of 14 2) Reference 1 IEEE MAC Protocol As a frst reference system, we consder the above-mentoned nfrastructure network whch operates based on the IEEE DCF. Therefore, n ths reference system, at each tme nstant at most one node s allowed to access the channel. 3) Reference 2 a multuser MIMO protocol In addton to the standard CSMA/CA, we compare the results wth those of a multpacket recepton protocol descrbed n [4]. In [4] and [16] the authors have proposed a multpacket recepton (MPR) protocol and nvestgated ts performance for varous types of networks and parameters. In [4], uplnk transmssons n an nfrastructure network are consdered, where the AP has N a antennas. STAs compete for the channel accordng to the DCF request to send/clear to send (RTS/CTS) mechansm. However, f accdentally more than one STA transmts n a tme slot and the AP can decode the RTS frames, t sends the CTS frame to all senders. Ths can be done as long as the number of concurrent transmssons are less or equal to N a. Afterwards the transmtters send ther data packets smultaneously to the AP. The suggested MAC closely follows the standard MAC, however, some modfcatons are requred. For example, as the AP does not have any aprorknowledge about the transmtters and ther channels, t should apply blnd technques to decode the RTS frames. Furthermore, t has to allocate orthogonal tranng sequences to the transmtters once t sends the CTS frames [4]. B. Throughput analyss Throughput s defned by the average payload bts whch are transmtted successfully n a tme slot dvded by the duraton of the tme slot,.e. T slot.wemanlyconsder saturaton throughput where there s always a packet n the buffer of each staton ready for transmsson. However, n Secton IV-C non-saturaton throughput and mpact of the presence of packets n buffers on the throughput are studed. In ths artcle we use Banch s model [17] as the bass for calculatng the collson and transmsson probabltes. In [17] an deal channel wth no channel errors, hdden nodes or capture s assumed. All nodes operate n saturaton condton. Furthermore, no retransmsson lmt s consdered. The model has been valdated n [17] for IEEE b parameters. In [18] the authors have consdered the IEEE a parameters and showed that the results usng the orgnal model are very close to the results obtaned from a more accurate model ntroduced n [18]. As shown n [17], analyss of the Markov chan model leads to the followng equatons whch have to be solved numercally to obtan the transmsson and condtonal collson probablty (collson probablty gven that a STA transmts) n a network wth n competng statons: P col =1 (1 τ) n 1, (1) 2(1 2p) τ = (1 2p)(W +1) + pw(1 (2p) m ), (2) where W and m can be calculated from the mnmum and maxmum contenton wndow szes, denoted by CW mn and CW max, respectvely, as follows: W = CW mn + 1 and CW max =2 m W -1 For the CB-CSMA/CA applcaton we need to adapt the basc model. For scenaros where all members of clusters are perfectly event-synchronsed we just need to replace the number of STAs n the orgnal model by the number of clusters, N c. However, as t wll be shown n the next secton, further changes are requred for cases where one or more STAs are not event-synchronous anymore. For the reference model and the MPR we can drectly apply the equatons gven n [17] and [4], respectvely, to calculate transmsson and collson probablty. Due to space lmt we do not repeat those equatons here and refer the nterested reader to [17] and [4]. Saturaton throughput for N a = 4 for both CB-CSMA/ CA and the standard CSMA/CA are depcted n Fgure 3. Alnkrateof19.5Mb/s(PHYrateof26Mb/sanda codng rate of 3/4) s assumed for CB-CSMA/CA applcaton. Ths s one of the IEEE n possble data rates. Snce there are four antennas at the AP, n the reference system at each transmsson attempt there s a Aggregate throughput [Mb/s] CB CSMA/CA, Model CB CSMA/CA, Smulaton Reference (CSMA/CA), Model Reference (CSMA/CA), Smulaton Number of STAs Fgure 3 Aggregate throughput for CB-CSMA/CA and the reference system vs. number of STAs n the network. AP has four antennas whle each STA has a sngle antenna. Throughput results obtaned from the analytcal model as well as smulatons are plotted.

7 Page 7 of sngle-nput multple-output (SIMO) lnk. Compared to a SISO lnk, for the reference system the maxmum spectral effcency gan of log(4) = 2 b/s/hz (or 40 Mb/s for a system wth 20 MHz bandwdth) can be expected [19]. Here as an extreme case (n favour of the reference system), we have ncreased the data rate of the reference system compared to the CB-CSMA/CA applcaton to the next hgher rate of IEEE n (58.5 Mb/s whch can be acheved by a PHY rate of 78 Mb/s and codng rate of 3/4). For both systems control packets are transmtted at the lowest data rate,.e. 6.5 Mb/s. As t can be observed n ths fgure, CB-CSMA/CA can mprove the throughput sgnfcantly. The results are verfed by computer smulatons. The smulaton results are also plotted n the same fgure. As t s expected smulated results follow the analytcal results closely and the results become more accurate as the network sze grows. The parameters whch are used for numercal nvestgatons throughout ths artcle are gven n Table 1. IV. Performance bounds In ths secton we analyse throughput upper and lower bounds of the CB-CSMA/CA applcaton under dfferent condtons. Frst we calculate the maxmum saturaton throughput and compare the results wth the throughput of the reference system and that of the MPR protocol. Then the performance of the CB-CSMA/CA protocol n non-deal condtons, e.g. where STAs n a cluster cannot transmt smultaneously or some of them may not have packets to transmt, s consdered. The former s acheved by takng the mpact of event-synchronsaton error on the network throughput nto consderaton. The latter s done by consderng nonsaturaton throughput and mpact of dfferent clusterng methods on the network throughput. In practce, throughput of a CB-CSMA/CA applcaton may vary between the calculated bounds. Table 1 The analyss parameters Parameters CB-CSMA/CA MPR SIFS 16 μs 16 μs DIFS 34 μs 34 μs Propagaton delay δ 1 μs 1 μs CW mn CW max Slot tme s 9 μs 9 μs Payload sze L 1024 byte 1024 byte Basc rate 6.5 Mb/s 6.5 Mb/s MAC Header (Data) 40 byte 34 byte MAC Header (ACK, CTS) 14 byte 20 byte RTS MAC Header N/A byte 20 byte CWUR MAC Header 8 byte N/A byte A. Maxmum saturaton throughput Frst we estmate the upper bound of the throughput by varyng the transmsson probablty τ and fndng the maxmum throughput for a network wth n STAs. In order to do so we can ether calculate throughput results for dfferent values of τ and fnd the τ* whch maxmses the throughput or fnd τ* as explaned n [13] by applyng the followng equaton: τ = n(n 1)ω n 2 (3) ω(n 1) where ω = T c /s wth T c beng the tme duraton when the channel s sensed busy due to collson and s beng the duraton of a tme slot whch s 9 μs accordng to IEEE n [1]. The maxmum throughput for CB-CSMA/CA and MPR versus the number of antennas at the AP are plotted n Fgure 4. The throughput of the reference system s also plotted for comparson. In the CB-CSMA/ CA applcaton the cluster sze ncreases wth the number of antennas at the AP, N a. In the MPR applcaton the maxmum number of smultaneous transmssons whch can be supported by the AP s ncreased wth N a. Whle a constant rate per lnk of 58.5 Mb/s s assumed for all CB-CSMA/CA and MPR setups, lnk data rate of the reference system has been ncreased logarthmcally wth the number of antennas at the AP. e Ths s done snce n the CB-CSMA/CA applcaton, the antennas at the AP are used to cancel the multuser nterference but n the reference system we have SIMO lnks where at each tme nstant the antennas at the AP are only used for a sngle transmsson. In all setups number of STAs s set to n = 60. Maxmum throughput [Mb/s] CB CSMA/CA, Constant header CB CSMA/CA, Varable header MPR Reference Number of antennas at the AP Fgure 4 Maxmum throughput for CB-CSMA/CA, MPR and CSMA/CA versus number of antennas at the AP for n =60 STAs.

8 Page 8 of 14 For the CB-CSMA/CA applcaton two types of tranng sequences have been consdered. As explaned n Secton II, tme-orthogonal tranng sequences per cluster-members can be used to make channel estmaton possble at the AP. In ths way, the header length ncreases as the cluster sze becomes larger. In Fgure 4 the results for ths case are denoted as varable header. However, tranng sequences may be orthogonalsed n other domans such as frequency. In ths way duraton of the header does not change wth the cluster sze, cf., CB-CSMA/CA constant header curvenfgure4.in ths fgure constant headers are assumed for the MPR applcaton. As we can observe n Fgure 4, CB-CSMA/CA gans more than other applcatons from ncreasng N a.however, note that for very large cluster szes, overhead ncreases as long as orthogonal tranng sequences n tme are appled. It should be noted that the results n Fgure 4 are the upperbound of throughput. In practce where the CW parameters are fxed, the throughput of the consdered applcatons are below the values shown here. B. Impact of event-synchronsaton errors on CB-CSMA/CA performance So far we have assumed perfect event-synchronsaton n all clusters. Consequently, each tme when a cluster accesses the medum, all of ts members transmt concurrently. However, n practce members of a cluster may become event-asynchronous when for example one or more of them cannot receve or decode the CWUR packet. Ths may happen f any of the members are n a deep fade. In ths secton we study the mpact of eventsynchronsaton errors on the throughput performance of the CB-CSMA/CA. However, to quantfy the MAC layer performance agan PHY channels are assumed to be perfect. An event-synchronsaton error can orgnate from dfferent types of errors, ncludng decodng errors on control frames and hdden nodes. In ths chapter we assume that event-synchronsaton errors occur due to decodng errors. Hence, the orgnal backoff model can stll be appled. In the presence of an event-synchronsaton error a subset of STAs n a cluster may be slent whle the others are transmttng. Consequently, the number of parallel streams s not the same as the cluster sze anymore. In CB-CSMA/CA even f all STAs n all clusters become event-asynchronous, the condtonal collson probablty s reduced compared to the CSMA/CA case. Ths s due to the fact that smultaneous transmssons from the same cluster can stll be decoded. It should be noted that n CB-CSMA/CA the preambles are defned n such a way that multple streams orgnated from a sngle cluster can be decoded. On the other hand, f for example two STAs, each from one cluster, begn to transmt smultaneously, the tranng sequences may not be orthogonal any more and hence the recever cannot estmate the channel and decode the packets. Ths happens even f the recever has multple antennas. Therefore, t s assumed that collson occurs f at least two statons whch belong to dfferent clusters begn to transmt at the same tme regardless of the number of parallel streams. We begn the analyss by consderng a symmetrc scenaro where all clusters suffer from synchronsaton errors n the same way. Then we extend the scenaro to a general case n whch each STA has an eventsynchronsaton probablty of P se. For all scenaros, we nvestgate the throughput equatons as a functon of P se. Agan we consder the uplnk of an nfrastructure network where the AP has four antennas and each sourcecluster contans four sngle-antenna STAs. We assume that all clusters can support a lnk data rate of 58.5 Mb/ s regardless of the number of parallel streams. Ths couldbethecasenthehghsnrregmewhereths rate could be supported on ndvdual lnks. Frst we consder a smple symmetrc scenaro, where all N c clusters have the same number of asynchronous STAs. Assumng each cluster has a total number of N nc STAs from whch, k STAs are event-asynchronous. In ths way, n the worst case scenaro all STAs are asynchronous,.e. k = N nc and each cluster sees k(n c -1)= N nc (N c - 1) contendng unts at each transmsson attempt. 1) Worst case scenaro, k = N nc Frst we assume that all STAs n clusters encounter synchronsaton errors. We study an extreme case where P se s assumed to be equal to 1, (.e. all members of clusters are wth probablty one asynchronous). Takng the MAC layer nto account, ths s the worst case whch can happen and hence the throughput values n ths case gve the lower bound of the CB-CSMA/CA throughput. In ths case, a collson happens f one or more STAs n a cluster transmt whle any other STA from any other cluster transmts at the same tme. It should be noted that even n ths case smultaneous transmssons of multple STAs wthn a cluster can be resolved and hence do not lead to collsons. The collson probablty gven that at least one STA transmts can be calculated by takng nto account the collson probablty n an equvalent CSMA/CA, wth n = N c N nc STAs, and then subtractng the probablty that any

9 Page 9 of 14 other node from the same cluster transmts whle other clusters are slent: P col =1 (1 τ) N cn nc 1 [ Nnc 1 ( ) ] N nc 1 τ (1 τ) N nc 1 =1 (1 τ) N nc(n c 1). P col can also be drectly obtaned by takng nto account the total number of contenders,.e. the number of contendng STAs whch belong to other clusters: (4) P col =1 (1 τ) (N c 1)N nc. (5) As expected, for dentcal parameters, Equatons 4 and 5 lead to the same result. Snce all members of clusters are asynchronous, the aggregate throughput of all clusters s defned by: ( ) N k k c m=1 τ m m (1 τ) (k m) ml S = (1 P col )(1 P e ), T slot where k = N nc and T slot s the duraton of the tme wndow whch can be obtaned usng: (6) T slot = P d σ + P s (1 P e )T s + (1 P d P s )T c + P e P s T e, (7) where P d, P s and P e are the probabltes that the slot s dle, contans a successful transmsson and channel error, respectvely. T s, T c and T e are the average tme requred for successful data transmsson, collson and channel error. P s and P d can be calculated from the followng equatons: P s = N c P t (1 P col ), P d = (1 τ) N (8) ncn c, where P t s the probablty that at least one STA n a cluster transmts n a slot: P t = k m=1 (k m )τ m (1 τ) (k m). Throughout ths artcle we focus on MAC layer features and we set P e to 0. For the CB-CSMA/CA applcaton T s and T c can be obtaned from: T s = T Data +SIFS+δ + T ACK + δ +DIFS, T c = T Data + δ +SIFS+T CWUR + δ +DIFS, where T Data and T ACK are the duraton of the data (ncludng the common preamble) and the ACK transmsson, respectvely, and T CWUR s the duraton of the CWUR frame. The duraton of the data and a control frames n an OFDM-based WLAN can be obtaned as follows [20]: T Data = T PLCPP + T PLCPSIG DataMACHeader+(16+6)/8+L + T sym, BpS(M) (9) T Control = T PLCPP + T PLCPSIG MACHeader + (16 + 6)/8 + BpS(M T sym, ) (10) where T PLCPP, T PLCPSIG and T sym are the duratons of the PHY layer convergence protocol preamble (PLCP), PLCP SIGNAL and one OFDM symbol, respectvely. The number of bytes per OFDM symbol for a gven modulaton M and the payload sze are denoted by BpS (M) and L, respectvely. As t s expected by settng N nc to 1 and N c to be equal to the number of STAs, n, (6) reduces to the throughput equaton n the standard case,.e. Eq. (13) n [17]. The normalsed throughput of the worst case scenaro, where all STAs encounter synchronsaton errors wth probablty one, s plotted n Fgure 5. Normalsed throughput s a untless parameter whch s defned by the tme used for payload transmsson dvded by the total tme needed for that transmsson,.e. T slot.we also depct the throughput of a reference system based on the standard CSMA/CA. It s assumed that both systems transmt at the same data rate. Ths s a sensble assumpton snce even for CB-CSMA/CA wth hgh probablty only one node transmts at each transmsson attempt. In ths secton we, furthermore, focus on the hgh SNR regme where CB-CSMA/CA can also operate atthehghestphydataratedefnedbythestandard regardless of the number of parallel streams. Normalsed throughput CB CSMA/CA, frst setup: worst case Reference (CSMA/CA) CB CSMA/CA, second setup Number of STAs Fgure 5 Comparson of normalsed throughput results obtaned usng the model wth those from smulatons, P se =1. Sold lnes show the analytcal results and the symbols show the smulated results. For CB-CSMA/CA smulatons have been performed for a second setup where the mpact of orthogonal tranng sequence has been consdered when calculatng the collson probablty.

10 Page 10 of 14 For the CB-CSMA/CA applcatons, t has been assumed that multple transmssons from more than one cluster always lead to collsons. Ths s the worst case as n practce dfferent frames from dfferent clusters may stll have orthogonal tranng sequences and hence the AP can decode those frames as long as the number of streams does not exceed the number of antennas at the AP. It can be observed that n the worst case scenaro, when all STAs are event-asynchronous, the CB-CSMA/ CA throughput s reduced to that of the reference system. It should be noted that evennthsscenarothe collson probablty of the CB-CSMA/CA s smaller than that of the standard CSMA/CA, however, the headers n CB-CSMA/CA are longer than that n the reference system due to the common preamble and the orthogonal tranng sequences. For large network sze, ths leads to slghtly smaller throughput n CB-CSMA/ CA as compared to that of the standard CSMA/CA, cf., Fgure 5. In order to valdate the results obtaned from the model, we have smulated the same scenaro n MATLAB. The normalsed throughput obtaned from smulatons are compared wth those obtaned from the analytcal model n Fgure 5. As t can be observed, results from the model match the smulaton. The model becomes more accurate as the network sze ncreases. For the CB-CSMA/CA applcaton, n addton to the worst case scenaro, a second set of the results s shown n Fgure 5. In ths case, we have assumed that there are totally N nc = N a groups of orthogonal tranng sequences and collsons occur only when ether more than N a STAs transmt at the same tme or two or more STAs wth non-orthogonal tranng sequences transmt concurrently. The smulated results for ths case are also depcted n Fgure 5. As t has been expected the second setup has hgher throughput than the frst one. 2) Asymmetrc scenaro In practce dfferent clusters may have dfferent channel condtons and hence we extend the study to cover asymmetrc scenaros. In ths secton, we consder an asymmetrc scenaro where only one cluster suffers from synchronsaton errors. In ths case we need to dstngush between two dfferent categores of clusters: () Category whch has only one cluster,.e. C.EachSTA n C has a non-zero P se. () The second category ncludes all other clusters wth P se =0.Wedenotethe clusters n ths category by C j where j. Snce the clusters n the latter category are all perfectly event-synchronsed, the members of C face collsons only when any STA of C begns to transmt whle any other cluster transmts at the same tme. Therefore, the collson probablty of C when k of ts members are asynchronous s defned by: P (k) col =1 (1 τ (k) j ) N c 1, (11) where superscrpt (k) ndcates dependency on k. P (k) = colj 1 (1 τ (k) j ) Nc 2 (1 τ (k) ) k+1 w.p. ( Nnc k )P k (1 se Pse)Nnc k, k 1 =0:Nnc 1 (1 τ (k) j ) Nc 2 (1 τ (k) ) Nnc w.p. ( Nnc Nnc )P Nnc se, k = Nnc (12) On the other hand, for any C j n the second category, a collson happens f t transmts and at the same tme any other cluster from the same category or any of the STAs n C begn to transmt. Accordngly, for ths category collsons occur wth dfferent probabltes, dependng on the number of clusters as well as the number of asynchronous STAs wthn the C, see (12) above. In ths scenaro agan we can use (2) to calculate τ. For each k, Equatons 2, 11 and 12 can be solved numercally. For each transmsson attempt of C, there are dfferent numbers of parallel streams. The number of parallel streams depends on the number of asynchronous STAs wthn C and whether they transmt by chance at the same tme or not. To calculate the throughput of cluster C we frst need to determne the probablty that k STAs out of N nc STAs n C are asynchronous,.e. P (k) r = ( Nnc k ) P k se (1 P se ) N nc k. (13) For a gven k, thestasnc are dvded nto two groups: ()thefrstgroupconsstsofn nc k synchronsed STAs, these nodes access the channel at the same tme and act as a sngle unt. The transmsson probablty of ths group s denoted by P ta. () the second group conssts of k asynchronous members, these STAs access the channel ndvdually, however, some of them may by chance transmt at the same tme. We denote the transmsson probablty of ths group by P tb. For a gven k we have: P (k) = τ (k) ta, k P (k) tb = m=1 ( ) k m (τ (k) =1 (1 τ (k) ) k. ) m (1 τ (k) ) k m The probablty that only STAs n C transmt s: P (k) s = [ P (k) ta (1 P (k) tb )+P(k) tb (1 P(k) (1 P (k) ) col [ ] = 1 (1 P (k) )(1 P (k) ) tb ta )+P (k) P (k) ta ta tb ] (14) (1 P (k) col ). (15)

11 Page 11 of 14 The probablty that only one of the clusters from the second category transmts s: P (k) s j =(N c 1)τ (k) j (1 P (k) col j ). (16) Hence, throughput of cluster C and aggregate throughput of all other clusters are, respectvely, expressed as: Nnc [ k=0 P(k) r P (k) (1 P (k) )(N k)l ta tb nc S = T slot ] +P (k) (1 P(k) )ml + P (k) P (k) (m + N nc k)l tb ta (1 P (k) ), col S j = ta T slot Nnc k=0 P(k) r P (k) sj tb N ncj L, T slot (17) a where T slot s equal to: [ T slot = P d σ + ( + P c T c, Nnc k=0 P r (k) P (k) )+( s Nnc k=0 ] P r (k) P (k) ) T sj s (18) where P d and P c,.e. probablty that collsons occur n a slot, are, respectvely, gven by: Nnc P d = P r (k) (1 τ (k) j ) Nc 1 (1 P ta )(1 P tb ) k=0 Nnc Nnc P c =1 P d P r (k) P (k) P s r (k) P (k). sj k=0 k=0 (19) b The throughput of each cluster category s shown n Fgure 6(a). As the number of clusters ncreases, the throughput of C mproves wth ncrease n P se.when P se ncreases, wth hgh probablty there are more asynchronous STAs n C. However, as smultaneous transmssons from C can be resolved and hence are not consdered as collsons, ths only ncreases the collson probablty of other clusters. Consequently, C may transmt wth hgher probablty and thus beneft from longer backoff duratons at other clusters. For the same reason the throughput of any other cluster degrades when P se ncreases. The aggregate throughput versus the number of clusters for dfferent values of P se has been depcted n Fgure 6(b). As t s expected, an event-synchronsaton error leads to throughput loss and the aggregate throughput decreases when P se ncreases. However, the throughput values for dfferent P se get close to each other for a large number of clusters. The mpact of a sngle cluster C almost dsappears once the number of clusters becomes very large. 3) General scenaro In ths secton, we wll consder a general and more realstc case, where members of each cluster may suffer from synchronsaton errors wth a certan probablty. Fgure 6 Throughput when each STA n C suffers from eventsynchronsaton error wth probablty P se. ItsassumedthateachclusterC has N nc STAs out of whch k are asynchronous. For each cluster, the collson probablty depends on the number of total unts,. e. asynchronous and synchronous STAs, n other clusers. Consequently, for 0 k N nc where {0,..., N nc } we have: P (k1,k2,...,knc ) col =1 Nc (1 τ (k1,k2,...,knc ) j )ñj, j=1,j (20) k j for k j = N ncj ñ j = { 1+kj for k j N ncj For any number of asynchronous STAs n each cluster the transmsson probablty can be obtaned by solvng the set of equatons gven n (20) and (2). Accordngly, the transmsson probablty of synchronous and asynchronous STAs n each C can be calculated as follows: P (k1,k2,...,knc ) ta = τ (k1,k2,...,knc ) P (k1,k2,...,knc ) tb = k m=1 ( ) k m (τ (k1,k2,...,knc )m ) (1 τ (k1,k2,...,knc )k m ). (21)

12 Page 12 of 14 Here and n the followng, for smplcty we denote the superscrpt (k 1, k 2,..., k nc ) by ( k). The probablty that k out of N nc STAs are asynchronous can be obtaned from (13), however, here k maytakedfferentvaluesn dfferent clusters. In ths scenaro, the probablty that only members of one cluster transmt also depends on ( k) and can be obtaned as follows: N nc1 N nc2 P s =... k 1=0 k 2=0 N ncnc k nc =0 P r (k1) 1 P r (k2) 2...P r (knc ) nc (22) [ ] P ( k) t a (1 P ( k) t b )+P ( k) t b (1 P ( k) t a )+P ( k) t a P ( k) t b (1 P ( k) col ). The throughput of each cluster s gven by: N nc1 N nc2 S =... k 1=0 k 2=0 N ncnc k nc =0 P r (k1) 1 P r (k2) 2...P r (knc ) nc [P ( k) t a (1 P ( k) t b )(N nc k )L +(1 P ( k) k ( k t a )( m m=1 m=1 + P ( k) k ( ) k t a ( τ ( k) m m )τ ( k) m (1 τ ( k) ) (k m) ml ) (1 τ ( k) ) (k m) (m + N nc k )L )] (1 P ( k) col )/T slot, (23) (24) where T slot can be calculated from the followng equatons: ( Nc ) T slot = P d σ + P s (1 P e ) T s + P c T c, (25) N nc1 N nc2 P d =... k 1 =0 k 2 =0 =1 N ncn c P r (k1) 1 P (k 2) r 2...P (k nc r ) nc k n c =0 (1 τ ( k) 1 )ñ1 (1 τ ( k) 2 )ñ2...(1 τ n ( k) c { 1+kj for k wth ñ j = j N ncj k j for k j = N ncj )ñc (26) N c P c =1 P d P s. (27) =1 Applyng the above equatons, we evaluate the throughput of the general scenaro where P se of each cluster s chosen unformly from the nterval [0, 1]. Each tme the throughput of each cluster and the aggregate throughput are calculated. Then P se values are set to Aggregate throughput [Mb/s] Number of STAs new random numbers. The results are shown n Fgure 7. Although the throughput s degraded compared to the case where all STAs are event-synchronous, t s stll much hgher than that of the CSMA/CA system. C. Impact of clusterngs on CB-CSMA/CA non-saturaton throughput In unsaturated condtons STAs may or may not have a packet to transmt, dependng on the traffc arrval rate. In order to calculate the transmsson probablty under unsaturated condton, we can apply the Markov chan model as proposed n [21]. In ths model, compared to the saturaton model, a new state s ntroduced, whch ndcates the probablty that there s at least one packet to be transmtted n the buffer. Ths probablty s denoted by q. As t s shown n [21], n ths case the transmsson probablty can be obtaned from: τ = CB CSMA/CA Reference Fgure 7 Throughput of the general case where STAs n each cluster C suffer from event-synchronsaton errors wth randomly chosen probablty P se. 2(1 2p)q q[(1 2p)(W +1) + pw(1 (2p) m )]+2(1 q)(1 p)(1 2p), (28) Assume l as the average packet arrval rate at each STA. For a traffc model wth Posson packet arrval process and for a small buffer sze the probablty q can be calculated from [22]: q =1 e λt slot, (29) where T slot can be obtaned from (7). In non-saturaton scenaros, groupng dfferent STAs nto a cluster can mpact the network throughput. Ths happens snce STAs n a cluster may or may not have a packet to send. Here we dstngush between two types of clusterngs: non-adaptve clusterng where clusters are formed regardless of the presence of queued packets

13 Page 13 of 14 and adaptve clusterng where STAs whch have packets to transmt are allocated to the same clusters. Adaptve clusterng can for nstance be acheved by usng the pollng nformaton n the contenton-free (CF) perod. The AP can adaptvely change clusters based on the nformaton obtaned durng the contenton-free perod. Accordng to IEEE e [23], STAs whch do not have any queued packet to send, reply the CF-Poll by a Null frame. As a result, the hybrd-coordnator (here the AP) can form clusters by takng only STAs wth queued packets nto account. As each CF perod s followed by a contenton perod, the clusters can be formed durng CF perod and can reman the same durng the contenton perod. 1) CB-CSMA/CA wth non-adaptve clusterng Frst we consder a scenaro where clusters are defned ndependently of the presence of packets n ther queues. Let us assume a worst case scenaro where a cluster begns to compete for the channel as soon as all of ts members have packets ready to send. Although n practce members of a cluster should be able to transmt even f other members reman dle, ths scenaro shows an extremely neffcent way of clusterng under unsaturated condton. Accordngly, for the CB-CSMA/CA throughput analyss we have to replace q n (28) by q N nc. Non-saturaton throughput of CB-CSMA/CA protocol wth the consdered non-adaptve clusterng s depcted n Fgure 8. For the purpose of comparsons, the throughput of the MPR as well as that of the IEEE n DCF basc access mechansm are also shown. At q = 1 each STA has a non-empty buffer wth probablty one and the system s saturated. It can be observed that the CB-CSMA/CA acheves hgher Aggregate throughput [Mb/s] CB CSMA/CA, non adaptve clusterng CB CSMA/CA, adaptve clusterng 5 MPR CSMA/CA Probablty of havng a non empty buffer at each STA (q) Fgure 8 Unsaturaton throughput of the adaptve CB-CSMA/ CA, the MPR and the CSMA/CA vs. probablty of havng nonempty buffers, for N nc = N a = 2 antennas at the AP and n =24. throughput above a certan threshold. For very low values of q, CB-CSMA/CA throughput s below that of the other systems. Ths happens snce clusters are defned ndependently of the packet arrval rate and each cluster transmts only f all of ts members have packets to transmt. The results n Fgure 8 are calculated for n = 24. CSMA/CA acheves slghtly hgher throughput for very low value of q as compared to MPR. Whle MPR s performed usng RTS/CTS handshake, the CSMA/CA s based on the basc access mechansm. Hence, n the regon where the collson probablty s small CSMA/ CA benefts from smaller overheads. 2) CB-CSMA/CA wth adaptve clusterng In ths part we assume an adaptve clusterng method. Accordng to ths method, for each transmsson attempt, we select STAs whch have packets n ther bufferandputthemtogethernclusters.inthsway, assumng n STAs and N nc =2,wecanformacluster wth two members as soon as there are at least two STAs whch have packets n ther buffer. Ths happens wth the probablty 1 (1 q) n ( n 1 )q(1 q)n 1. Throughput results for ths scenaro wth n =24are depcted n Fgure 8. It can be observed that the CB- CSMA/CA protocol wth adaptve clusterng outperforms both the MPR and CSMA/CA for most values of q. V. Concluson In ths artcle we proposed a novel cluster-based CSMA/ CA scheme whch supports multuser streams and reduces the collson probablty n a network. The CB-CSMA/CA protocol showed a promsng throughput mprovement compared to a reference system based on IEEE The analyss of event-synchronsaton errors shows that the CB-CSMA/CA s relatvely robust to the event-synchronsaton error and n the worst case t performs smlar to the standard CSMA/CA. The CB-CSMA/CA outperforms both MPR and CSMA/CA n saturaton as well as unsaturaton mode wth medum and hgh probablty of non-empty buffers. However, to beneft from CB-CSMA/CA for low packet arrvalrates,weshouldapplyanadaptveclusterng. The adaptve clusterng takes the presence of packets n the STAs nto account and t can be performed by usng the nformaton obtaned n the pollng phase. Endnotes a The BSS s defned as the basc buldng block of an IEEE network [2]. It s assumed that the statons wthn a BSS are n the communcaton range of each other. b We denote all errors whch cause dfferent backoff wndow values at nodes wthn a same cluster, as eventsynchronsaton error.