Increasng Supported VoIP Fows n WMNs through n-based Aggregaton J. Oech, Y. Hamam, A. Kuren F SATIE TUT Pretora, South Afrca oechr@gma.com T. Owa Meraa Insttute Counc of Scentfc and Industra Research (CSIR) Pretora, South Afrca M. Odhambo Eectrca and Mnng Engneerng Unversty of South Afrca (UNISA) Pretora, South Afrca Abstract As Voce over IP (VoIP) becomes a reaty, servce provders w be abe to offer the servce to remote and over popuated areas that currenty are not or are ony partay reached by avaabe Pubc Swtched Teephone Networ (PSTN). The combnaton of wreess mesh networs (WMNs) wth VoIP s an attractve souton for enterprse nfrastructures; presentng avaabty and reduced cost for both consumers and servce provders. The arge number of cents n WMNs eads to ncreased number of concurrent fows. However, ony a handfu of these fows reaches ther destnaton whe st wthn the quaty of servce (QoS) bound for VoIP. Ths performance degradaton can be attrbuted to protoco overhead, pacet coson and nterferences. Ths paper ntroduces VoIP over WMNs and uses a n based pacet aggregaton scheme to mprove VoIP performance n IEEE 802.11 based WMNs operatng under dstrbuted coordnate functon (DCF). Smuaton resuts show that the proposed aggregaton scheme ncreases the number of supported fow whe aso reducng endto-end deay, tter, and pacet oss of VoIP n WMNs. Keywords-component; SNIR; VoIP; WMNs; QoS. I. INTRODUCTION Voce over Internet Protoco VoIP) refers to the transmsson of voce usng IP technooges over pacet swtched networs. Internet teephony s one of the typca appcatons of VoIP. As compared to the tradtona resource dedcated PSTNs, VoIP provdes for resource sharng. Thus, IP based VoIP appcatons presents a cost effectve means of factatng voce communcaton. The ncreasng popuarty of IEEE 802.11 based networs n homes and offces aso provdes a motvaton to use wreess VoIP. For exampe, wth wreess oca area networs (WANs) t becomes easer for users to access teephone servces anywhere anytme through portabe handsets. Wreess mesh networs (WMNs) provde an attractve souton n areas where networs are not easy to nsta or uneconomca to set up. ac of proper networ structures creates aenated areas caed dead zones where there are mted or no networ coverage. Thus, WMNs technoogy presents a vabe aternatve to create an enterprse-scae or communty-scae wreess bacbone wth mutuser wreess VoIP connectvty. However, a maor chaenge s that as the number of VoIP fows ncrease n a networ so does the number of supported cas drops. Tshwane Unversty of Technoogy (TUT) and French South Afrca Technca Insttute n Eectroncs (F SATIE) The number of supported cent capacty s affected by the networ forwardng performance, shared contenton and sef nterference [1]. For IEEE 802.11 based WMNs, the man chaenge n provdng hgher pacet transfer rato es on management of the medum access contro (MAC) protoco overhead. Ths overhead s attached to every pacet transmtted and therefore consumes sgnfcant porton of networ bandwdth that can be used to carry addtona pacets. Thus, the dsma performance assocated wth channe access protoco and transmsson overhead magnfes for sma pacets such as VoIP. Ths wor proposes a n based pacet aggregaton mechansm that adusts aggregaton pacet sze based on oca n quaty to provde guaranteed QoS for VoIP pacets n WMNs. The remanng part of ths paper s organzed as foows. Secton II dscusses reated wor. In secton III detas of the mpact of protoco overhead on VoIP ca capacty for VoIP over WMNs are presented. The aggregaton agorthms are anaysed n secton IV. Obtaned smuaton resuts are presented and dscussed n secton V. Fnay, secton VI concudes the wor. II. REATED WORK The probem of transmttng sma szed pacets n IEEE 802.11 based networ has exsted for qute sometme. Authors such as Hoe and Tobag [2] found that each Access Pont (AP) can ony support a few VoIP fows due to the arge overhead of IEEE 802.11 MAC n processng sma pacets. Studes conducted to understand the capacty of WMNs n [1] show that the throughput of each node decreases at order O(1/n), where n s the number of hops. To mprove performance of n such networs, severa approaches have been proposed for both snge and mut-hop wreess networs. However, ths wor narrows to ony the terature n sync wth the proposed methodoogy. The use of pacet aggregaton to mprove performance of VoIP appcaton on the networ s presented n [3], [4], [5] and [6]. The basc decson for an aggregaton agorthm n WMNs s the pacement of de-aggregaton capabty. Ths choce defnes the apped pacet aggregaton approach. There are two basc approaches to pacet aggregaton: end-to-end aggregaton and hop-by-hop aggregaton. In end-to-end aggregaton, pacet aggregaton taes pace at the ngress nodes whe the egress nodes do the de-aggregaton. The hop-by-hop aggregaton does
aggregaton at every node from source to destnaton. Important parameters for mpementng pacet aggregaton are maxmum aggregaton pacet sze and maxmum aggregaton deay. These parameters can be mpemented as fxed, dynamc or a combnaton of fxed and dynamc at varous stages n the networ. Thus, sutabe mx of these parameters can be made to dversfy basc aggregaton mechansms and acheve maxmum benefts. In [3], the use of concatenaton mechansm to reduce protoco overhead s proposed. It assumes a networ wth homogeneous nodes. Ths assumpton presents an neffcent usage of bandwdth. In [4], IP based adaptve pacet concatenaton agorthm for mut-hop WANs s proposed and smuated. The smuaton resuts revea that more than doube the throughput can be acheved n hghy oaded networs but at the expense of ncreased end-to-end deay. The authors n [5] descrbe IEEE 802.11 overhead and the mportance of pacet aggregaton n Ad Hoc networs. Two aggregaton agorthms are proposed: forced agorthm and adaptve agorthm. The forced agorthm ntroduces addtona deay at every hop from source to destnaton. The agorthm can resut n hgher cumuatve deay whch s not sutabe for rea-tme appcaton. On the other hand, the adaptve agorthm proposed n [5] does not usuay have suffcenty enough pacets to aggregate to provde good bandwdth savngs. The authors n [6] nvestgate the mpacts of aggregatng mutpe sma VoIP streams n wreess networs. The resuts of the experment revea the exstence of reatonshp between number of VoIP cas, output n rate and certan teetraffc metrcs. However, the aggregaton agorthm used a n rate whch s not adustabe to the networ stuaton. Frame aggregaton and optma frame sze adaptaton for IEEE 802.11 WANs are presented n [7] and [8]. In [8], a mode for cacuatng the successfu transmsson probabty of a frame of a certan ength s proposed. The resuts of ths experment show that the eves of networ contenton ony has a mnor nfuence on transmsson and that the proposed aggregaton outperforms fxed frame aggregaton. However, the paper fas to deta out how the frames are deayed. It was deveoped and verfed for snge-hop where ony sef nterference s more promnent. These stuatons do not appy to WMNs. In [7], a method to adapt the frame sze dynamcay to the channe quaty and networ contenton s presented. By ntermarryng end-to-end and hop-by-hop aggregaton agorthms, the proposed accreton agorthm expots the advantages of the two whe aso routng out ther shortcomngs. The accreton agorthm uses forced deay at the ngress to coect pacets of the same fow and natura meda access deay for ntermedate nodes. The paper shows that for hgher offered oad, the optmum frame sze ncreases up to a droppng pont. Thus, t s benefca to reduce the channe rate and pacet sze to mnmze the nterference. III. VOIP OVER WIREESS MESH NETWORK When rong out VoIP servces n IEEE 802.11 based WMNs, the man chaenge s the satsfacton of users who are aready accustomed to hgh quates provded by PSTN. Such a quaty n WMNs s compromsed by the archtecture of the IEEE 802.11 and the transmsson overhead for sma szed VoIP cas. A. WMNs Archtecture The genera archtecture of WMNs s a mx of fxed bachau mesh routers and fxed or mobe mesh cents as shown n Fgure 1. Mesh cents can be W-F enabed VoIP handsets, aptops or any other wreess handhed devces and have connectons across the WMNs to other wreess devces. Communcaton from these mesh cents go through the oca mesh networ to other wred or wreess VoIP phones, out to the Internet wth the hep of gateways, or to PSTN through oca Prvate Bag Exchange (PBX) [7]. Wred or wreess phones that extend networ coverage are caed mesh routers. These routers provde bachau connectvty at the n eve or networ ayer. Typca IEEE 802.11 nodes use two man MAC access protocos; Dstrbuted Coordnaton Functon (DCF) and Pont Coordnaton Functon (PCF). Athough PCF offers adequate support for QoS needs of rea-tme traffc, t s uncommon and s amost never depoyed. In ths wor, a wreess nodes are based on IEEE 802.11b W-F nterfaces and runnng on DCF channe access mechansm. Both wred and wreess nodes are uses IP eve addressng so as to excude the probems resutng from routng at the n eve. However, the wor can be taored for n ayer routng. PSTN PBX Mesh Router Cent Mesh Router Gateway Internet Cent Gateway Fgure 1. Voce over WMNs. Communcaton paths are mantaned among wreess mesh routers. Each mesh router has enough nterfaces to connect to cents and bachau. Cents can connect to fxed wreess cent, nternet or to PSTN through the PBX. B. Overhead n IEEE 802.11 based WMNs VoIP systems use codecs to harmonse nteractons between the dgta and anaogue words. The codec nterface receves anaogue voce, converts t to pacets and reeases them at a defned rate. To date, there are severa vocoders avaabe n the maret such as G.711, G.723, GSM and G.729A each comng wth ts pros and cons. Notaby, G.729A s ncreasngy becomng more popuar. For correctness, ths study uses the behavour of G.729A codec for the generaton of VoIP pacets. However, the genera ssues addressed n ths paper are aso appcabe to other codecs. When usng G.729A [9], a voce payoad of 20 bytes s generated at a rate of 50 pacets every second. Therefore, after
40 bytes IP/UDP/RTP header s added, the mnmum channe capacty needed to support a voce stream n one drecton s 24 Kbps for 11 Mbps channe. Ths capacty s equvaent to about 229 VoIP cas. However, expermenta and anaytca resuts ndcate that there s ow VoIP ca capacty. The decrease n capacty can be attrbuted to the arger aggregate tme spent by networ n sendng headers and acnowedgements, watng for nter-frame separatons, and contendng for the medum. For exampe, 20 bytes VoIP payoad contrbutes 14.5 µs at 11 Mbps but IP/UDP/RTP header, MAC headers and physca headers, traers, nter-frame perods, Bac-off and acnowedgements (ACK) need a tota of 818 µs [7]. The contrbuton of the VoIP payoad ncreases the transmsson tme to 832.5 µs. The number of supported cas s cacuated usng the formuae beow. operates at MAC eve. Pacet assemby s usuay done coser to the source of traffc wth the aggregate pacet forwarded to an aggregaton target. Upon arrva at the target, the orgna sma VoIP pacets are recovered from the aggregate pacet. Ths recovery process s nown as fragmentaton or deaggregaton dependng on the ayer n whch aggregaton s done. ( 2. β. α ) 1, (1) where β, s the number of pacets generated by a coder per second and α s the tota transmsson tme for VoIP payoad overheads. Ths yeds ony about 12 VoIP cas supported per hop. The cacuatons above revea that per-frame overhead n the IEEE802.11 standard sgnfcanty mts the capacty of VoIP over WMNs. Apart from hgh protoco overhead, provdng voce servces over WMNs faces other technca chaenges based on the nature of VoIP traffcs and behavour of WMNs. VoIP has strct QoS requrements and ths gets threatened n WMNs as chances for pacet oss s more profound n channes wth more nterference. Channe nterferences ncrease wth ncrease n number of fows, a characterstc common n WMNs. Because pacet aggregaton reduces pacet overhead, t s mperatve to note that t can be used to mprove the performance of VoIP over WMNs. IV. PACKET AGGREGATION AGORITHMS Aggregaton agorthms enta the process of assembng and forwardng of pacets wth smar destnaton caed aggregaton target and eventua recovery of the orgna pacets at the target as shown n Fgure 2. Fgure 3. Aggregaton of two pacets Pacet aggregaton can be adopted to boost the throughput of IEEE 802.11 based WMNs. Fgure 3, shows the transfer of two pacets wth and wthout aggregaton. It s found that VoIP pacet taes 832.5 µs and 1665 µs transfer tmes for one and two pacets respectvey. When the two pacets are aggregated, t taes ony 84 µs to transfer them together, whch s about 50% tme savng. Thus, ony a sma number of VoIP pacets can be supported n WMNs snce a good porton of the bandwdth s taen by the protoco overheads. Fgure 2. Pacet Aggregaton The process of assembng mutpe sma pacets nto a snge pacet s caed pacet aggregaton when t operates at IP-eve and frame aggregaton or frame concatenaton when t To ustrate the beneft of pacet aggregaton, assume that pacets of the same sze ρ bytes are transmtted at a channe rate of are transmtted at a channe rate of λ Mbps. The beneft of aggregatng κ pacets durng transmsson can be determned by cacuatng the dfference between transmsson wth aggregaton and wthout aggregaton. The saved tme τ (seconds), can then be expressed as foows. τ = τ.( κ 1) 8. γ, (2) 0 λ
where denotes the sze of aggregaton header and τ 0 s the channe tme. Snce and may be assumed constant for IEEE 802.11b based WMNs, by nspectng equaton (2), t can be noted that the aggregaton beneft, τ, ncreases wth ncrease n the number of pacets. Athough ths mpes that the arger the aggregaton sze the better, the mpementaton prompts for further consderatons on end-to-end deay, deay varance and pacet oss parameters whch are cruca for quaty VoIP. When aggregatng, an extra overhead of 20ms s usuay added to the frst pacet. Ths maes t ogca to use aggregaton n ghty oaded networs. However, under a heavy oaded networ, whch usuay happens n WMNs, the sma pacets experence heavy contenton. The ncreased contenton causes voce pacets to drop or be retransmtted resutng nto ncreased networ traffc. In such networs, pacets have to be queued whe watng for meda access. A. The Fxed Pacet Aggregaton Agorthm Ths s aso caed forced-deay aggregaton agorthm. The agorthm mars arrvng pacets wth a tmestamp. The mared pacets are then deayed for a pre-defned tme caed maxmum deay perod (δ). After the expry of δ pacets destned to same next hop are aggregated. The sze of the aggregated pacet s however mted by the maxmum transmsson unt (MTU), whch s 2300 bytes for IEEE 802.11 standard [10]. The rght choce of δ s mportant. Hgher deays yed a hgher aggregaton rate, but aso a hgher end-toend deay. In ths wor, MTU and δ has been fxed at 1500 bytes and 10 msecond respectvey. Pacet aggregaton s done by frst coectng a pacets havng same next hop. Ths s mpemented at the outbound queue n the MAC ayer. Nodes capabe of aggregaton mantan vrtua queues; each for one out-ns. These queues temporary eep pacets as they wat to be aggregated. When a node s de, t checs each n s queue n a round-robn manner f t s ready for aggregaton. The decson s nfuenced by two parameters: maxmum queue sze ϕ, and deay tme χ. If a n has a queue sze greater than ϕ or a head-of-ne pacet tmestamp ndcates t s χ od, then the pacets n the queue are aggregated. Durng ths tme, VoIP pacets are paced together unt the sze of the new pacet becomes arger than MTU or the queue becomes empty. If no queue satsfes the condtons, the node stays de. Ths reeases the wreess channe to be used by other nodes. The two parameters, ϕ and χ, are reated by equaton ϕ = β. χ, (3) where β, s the average nput rate of n. When s gven, the prmary probem s to determne how to choose χ for each wreess n. The pacet aggregaton rate of n s defned as ψ 1 χ. (4) Here, the optma vaue of equaton (4) mnmzes pacet deay n WMNs. However, the optma vaue for ψ s constraned by fow conservaton (FC), Capacty mt (C) and MTU sze propertes. The FC property emphaszes that the ncomng data rate of a n s equa to the outgong data rate. Ths data rate s aso the aggregaton rate. The capacty constrant ensures that the utzed capacty s no more than the capacty that the channe can offer. As for the MTU sze, the aggregated pacet sze shoud not exceed MTU. B. The Proposed Pacet Aggregaton Agorthm VoIP ca capacty s determned by the pacet that meets VoIP QoS constrant. By reducng pacet oss occasoned by bt errors whe transmttng aggregated pacet, the VoIP ca capacty can be mproved. Ths agorthm ams at dynamcay readustng maxmum aggregaton pacet sze to maxmze the number of fows accommodated n WMNs. Snce aggregaton ams at achevng hgher capacty by combnng smaer pacets, n the proposed agorthm, the pacet rate formuaton narrows down to determnng the maxmum pacet sze that woud optmze equaton (4). For a gven channe quaty, contenton eve and traffc necton rate, dfferent pacet szes produce dfferent pacet oss ratos. To mnmze ths oss, t s desrabe to determne the optma frame sze. Pacet oss n WMNs s dependent on the bt error (BE), queue overfows, and cosons. Pacet oss due to coson and queue overfows can be reduced by ncreasng pacet szes. However, arger pacets ncrease pacet oss due to BE. The BE occurs when a receved sgna cannot be decoded propery. The extent of BE caed bt error rate (BER) s dependent on the moduaton scheme, sgna-to-nose and Interference rato (SNIR) of the receved sgna, the codng scheme and data rate [11]. Here, apart from SNIR, other factors are usuay constant n IEEE 802.11b based networs. The BER s therefore ony dependent on SNIR. Accordng to [12], SNIR can be defned as Ps SNIR = 10 og 10 Pn, (5) where P s, s the strength of the sgna and P n s the strength of nose produced by therma nose and nterference. Therefore, by defnng the foowng varabes: a reatonshp ( 1 (, )) ( ( )) ( ( )) 8. D, 1, and 1, = α β R D = α β R D = α β R, between frame error rate (FER) and BER may be expressed as foows. FER = 1 D. D. D, (6) where, α s the BER, β s the SNIR vaue, R s the transmsson rate of preambe, R s the transmsson rate of physca ayer contro protoco (PCP) header, R s the transmsson rate of MAC frame, s the ength of the
preambe bts, s the ength of PCP header n bts and s the ength of MAC frame n bytes. If the engths of the preambe and header, and transmsson rates are consdered to be constant, the FER s a functon of SNIR and the frame ength. For any networ, as the SNIR goes to nfnty the average error rate goes to zero. Ths means that the networ becomes more accommodatve to arger pacets as the SNIR gets hgher. Fgure 4 ustrates the reatonshp between pacet sze and SNIR assumng IEEE 802.11 standard overheads The ns-2 smuator does not come wth an aready deveoped VoIP traffc agent. In ths paper, a bdrectona VoIP conversaton wth sence suppresson s modeed as an on-off Marov process. The traffc fow s assgned a ta spurt of 35% and sent perods of 65% as typca wth G.729A vocoder. VoIP s transmtted over UDP/RTP/IP protocos to form a tota pacet sze of 60 bytes. Fgure 5 ustrates the networ topoogy used n the smuaton. It comprses of mesh cents that are ether wred or wreess, access ponts (AP) that provdes access to the Internet and wreess mesh routers to extend the coverage of APs. Ths arrangement of nodes repcates the current snge rado networs where the cosest gateway s usuay no more than two hops. The networ assumes that there s ony one AP n the networ. A wreess nodes are based on IEEE 802.11b wth DCF channe access mechansm and RTS/CTS are dsabed snce they reduce networ performance for sma pacets. Nodes n the networ are confgured for herarchca routng. Fgure 4. Correct Pacet ength for a gven SNIR [13] AP VoIP Cents Wth these arguments, an optma pacet determnaton scheme can be deveoped as a functon of SNIR. The scheme shoud ncorporate the sender and the recever handshae. The recevng node measures the SNIR of the comng pacets, cacuates the maxmum toerabe pacet sze based on the current SNIR and transmts the cacuated vaue to the sender. The current SNIR vaue ( S ) for each n s cacuated and stored n the routng tabe. The formua used s S = 1 S + α( S + m S ) (7) where S defnes SNIR vaue before recevng the current pacet, S m s the SNIR of the ncomng pacet and α s the smoothng factor defned by the equaton 0 < α < 1. Snce statc WMNs are stabe, the vaue of α s adequate. In ths wor, α s chosen as 0.1. VoIP Cent Wred Router Fgure 5. Smuaton topoogy Mesh Router Smuaton resuts are reported n Fgures 6, 7, 8 and 9. The potted vaues are obtaned by varyng concomtant fows per smuaton that asts for 150 seconds, then the average end-toend deay, tter and pacet oss for the current smuaton are cacuated and potted aganst the nected fows. For each performance metrc, a maxmum vaue s seen beyond whch performance begn to degrade rapdy. These vaues correspond to the threshod for supported concomtant fows. V. PERFORMANCE EVAUATION In ths secton, the performance of the DA s evauated n terms of end-to-end deay, tter and pacet oss rate of VoIP pacets under dfferent number of concomtant fows. The resuts are compared wth those obtaned wthout aggregaton and under fxed aggregaton scheme. The ns-2 smuaton envronment s used. The varaton of the number of parae fows by use of nected fows mode dfferent degrees of networ contenton and nterference. Ths ads n understandng the performance of the proposed agorthm over rea mesh networ depoyments. Fgure 6. End-to-end deay for VoIP n WMNs
Fgure 6 ustrates the end-to-end deay characterstcs for three scenaros. oong at the fgure, t can be seen that for ow traffc, aggregaton agorthms have hgher traffc end-toend deay compared to no aggregaton. However, as the number of nected fows ncreases, more pacets get aggregated and thus reducng the average pacet deay. The proposed agorthm presents superor performances wth a brn experenced from 105 fows compared to 45 and 30 for fxed and no aggregaton. In Fgure 7, the reatonshp between pacet end to end tter and nected fows s presented. From the fgure, t can be seen that the use of pacet aggregaton reduces deay varaton. By sendng arger bocs of pacets, aggregaton agorthms reduce chances of havng unnecessary onger queues that causes tter n the networ. The proposed aggregaton experences a brn after 105 fows whe fxed aggregaton and no aggregaton have ther tter rsng from 30 and 25 fows respectvey However, for fows ess than 20, no aggregaton scenaro has superor tter and end-to-end deay vaues compared to aggregaton technques as shown n Fgures 6 and 7. Ths s because, for ower traffc some pacets are deayed due to the δ deay parameter and queung. As a resut pacets requre dfferent tme to be transferred. If δ s sma, most pacets w be sent wthout aggregaton thereby demystfyng the use of aggregaton. Fgure 8. VoIP pacet oss rate n WMNs Fgure 9 shows the number of supported fows for each scenaro when the number of concomtant fows s vared. Fxed aggregaton schemes support the east number of fows and above 30 fows t supports amost nu. The proposed aggregaton however shows remarabe performance wth neary 90% support for nected fows. Fgure 9. Support Fows versus Inected Fows Fgure 7. Average deay varaton for VoIP pacets Apart from end-to-end deay and tter, pacet oss rate s aso a cruca parameter n evauatng networ performance. Pacet oss ncudes both pacets that do not reach the destnaton at a or reaches wth unacceptaby onger deay. Athough aggregaton technques uses the meda we by transmttng arger bocs of pacet thereby reducng contenton and overhead, the ager pacets have hgher chances of beng dropped due to frame errors condtons. As ustrated n Fgure 8, fxed aggregaton that uses an nvarabe aggregaton pacet sze experences arger pacet oss compared to other technques. The use of no aggregaton experences hgher pacet oss as a resut of tter buffer beng overwhemed by arge number of pacets. The better performance reazed by the proposed agorthm s attrbuted to the abty of the agorthm vary pacet sze n response to n characterstcs. The fxed aggregaton agorthm may create pacets that are too arge to be accommodated n a channe eadng to a drop to pacet oss. However, even beow the capacty threshod t happens that some fows have bad quaty. Ideay a fows beow threshod are to be supported and ths dvergence can ony be attrbuted to the dfference n confdence eves between fows. VI. CONCUSION Ths paper has shown that VoIP performs poory n WMNs. It further proposed a n based aggregaton agorthm that adusts aggregaton pacet sze based on oca n characterstcs. The proposed agorthm has been smuated
and ts performances compared wth no aggregaton and fxed aggregaton approaches. The smuaton resuts show that the proposed aggregaton scheme yeds superor VoIP QoS performance compared to other approaches by ncreasng the number of supported fows whe aso reducng end-to-end deay, tter and pacet oss rato of VoIP pacets. Thus, the resuts have proven that by consderng n quaty parameters to adust aggregaton pacet sze, pacet VoIP performance n WMNs can be enhanced. REFERENCES [1] J. Jun and M.. Schtu. The nomna capacty of wreess mesh networs. IEEE Wreess Communcatons, Oct 2003. [2] D. P. Hoe and F. A. Tobag, "Capacty of an IEEE 802.11b wreess AN supportng VoIP," In Proceedngs of IEEE Int. Conference on Communcatons (ICC), 2004. [3] Y. Xao, "Concatenaton and Pggybac Mechansms for the IEEE 802.11 MAC," n IEEE WCNC, 2004. [4] Raghavendra, R., et a., IPAC-An IP-based Adaptve Pacet Concatenaton for Muthop Wreess Networs. n Proceedngs of Asomar Conference on Systems, Sgnas and Computng. 2006. [5] Jan, M. Gruteser, M. Neufed, and D. Grunwad, Benefts of pacet aggregaton n ad-hoc wreess networ, Dept. Comput. Sc., Unv. Coorado, Bouder, CO, Tech. Rep. CU-CS-960-03, 2003. [6] R. Komoafe, O. Gardner, "Aggregaton of VoIP Streams n a 3G mobe networ: A Teetraffc Perspectve," n European Persona Mobe Communcatons Conference (EPMCC), 2003. [7] Ganguy, S., et a., Performance Optmzatons for Depoyng VoIP Servces n Mesh Networs. Appeared n IEEE Journa on Seected Areas n Communcatons, Vo. 24, no. 11, Nov. 2006. p. 2147-2158. [8] n, Y. and W.S. Wong, V. Frame Aggregaton and Optma Frame Sze Adaptaton for IEEE 802.11n WANs. n Proceedngs of IEEE Goba Teecommuncatons Conference. 2006. San Francsco, CA. [9] ITUT Rec. G.729A (11/96): Reduced Compexty 8bt/s CS-ACEP Speech Codec. [10] J., C. Bae, D. S. J. De Couto, H. I. ee, and R. Morrs, Capacty of ad hoc wreess networs," n Proceedngs of the 7th ACM Internatona Conference on Mobe Computng and Networng, Rome, Itay, Juy 2001, pp. 61-69. [11] S. Mangod, S. Cho, and N. Esseng, An Error Mode for Rado Transmssons of Wreess ANs at 5GHz. n Proceedngs of 10th Aachen Symposum on Sgna Theory. 2001. [12] Xuchao, W., Smuate 802.11b channe wthn ns2, Natona Unversty of Sngapore: Sngapore, 2004. [13] W. u, Y.Fang, Courtesy Pggybacng: Supportng Dfferentated servces n Muthop Mobe Ad Hoc Networs, WINET, Unversty of Forda, 2004