THE INCREASED demand for more bandwidth and bandwidth

Transcription

1 JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 25, NO. 1, JANUARY Dynamc Wavelength and Bandwdth Allocaton n Hybrd TDM/WDM EPON Networks Ahmad R. Dhan, Chad M. Ass, Member, IEEE, Martn Maer, Member, IEEE, and Abdallah Sham, Member, IEEE Abstract We dscuss a wavelength-dvson-multplexed-based passve-optcal-network (PON) archtecture that allows for ncremental upgrade from sngle-channel tme-dvson multpleaccess PONs n order to provde hgher bandwdth n the access network. Varous dynamc-wavelength and bandwdth-allocaton algorthms (DWBAs) for wave-dvson multplexed PON are presented; they explot both nterchannel and ntrachannel statstcal multplexng n order to acheve better performance, especally when the load on varous channels s not symmetrc. Three varants of the DWBA are presented, and ther performance s compared. Whle the frst varant ncurs larger dle tmes (and, hence, poor performance), the other two algorthms acheve better but dfferent performance wth crtcal dssmlartes. Our analyss also focuses on the far assgnment of excessve bandwdth n the upstream drecton to hghly loaded optcal network unts. We compare the performance of DWBA to another algorthm that reles on statc-channel allocaton. Furthermore, a study s presented wheren the number of wavelengths ncreases, and a comparson wth nterleaved pollng wth adaptve cycle tme s shown. We use extensve smulatons throughout ths paper. Index Terms Dynamc-bandwdth allocaton (DBA), ethernet passve optcal network (EPON), smulaton and modelng, wavelength-dvson-multplexed (WDM)-passve optcal network (PON). I. INTRODUCTION THE INCREASED demand for more bandwdth and bandwdth servces [1] n the access network has been growng rapdly, and there have been great efforts to develop economcal subscrber networks based on optcal technology [2] [6]. Currently, the predomnant broadband-access solutons developed are the dgtal-subscrber-lne and cable-modem-based networks. Both of these technologes have lmtatons, because they are based on nfrastructure that was orgnally bult for carryng voce and analog TV sgnals, and ther retroftted versons to carry data are not optmal [2]. Passve optcal networks (PONs) and Ethernet PONs (EPONs) [4], [8] are vewed by many as an attractve and promsng soluton for the broadband-access-network bottleneck. EPON s a pont-tomultpont access network wth no actve element n the sgnal s Manuscrpt receved January 23, 2006; revsed September 12, A. R. Dhan and C. M. Ass are wth the Faculty of Engneerng and Computer Scence, Concorda Insttute for Informaton Systems Engneerng, Concorda Unversty, Montreal, QC H3G 1M8, Canada (e-mal: a_dhan@ cse.concorda.ca; ass@cse.concorda.ca). M. Maer s wth INRS Énerge, Matéraux et Télécommuncatons, Montreal, QC H5A 1K6, Canada (e-mal: Martn.Maer@emt.nrs.ca). A. Sham s wth the Department of Electrcal and Computer Engneerng, Unversty of Western Ontaro, London, ON N6A 5B9, Canada. Color versons of one or more of the fgures n ths paper are avalable onlne at Dgtal Object Identfer /JLT path from source to destnaton; the only nteror elements used n ths archtecture are passve components such as optcal spltters and optcal fbers. EPON has been standardzed by the IEEE 802.3ah workng group, and t comprses one optcallne termnal (OLT) and multple optcal-network unts (ONUs). Currently, EPON systems deploy two wavelengths: typcally 1310 nm for the upstream transmsson and 1550 nm for the downstream transmsson. In the downstream, Ethernet frames are broadcast by the OLT and are selectvely receved by each ONU. Alternatvely, n the upstream, multple ONUs share the same transmsson channel to s data and control packets to the OLT. Snce ONUs are unable to detect collson occurrng at the OLT, and due to the dffculty to mplement a carrersense multple access wth collson detecton, t s necessary to desgn a mechansm that arbtrates the access of ONUs to the shared medum. Ths s acheved by desgnng medumaccess-control (MAC) protocols to prevent collson between Ethernet frames of dfferent ONUs transmttng smultaneously. Current MAC supports tme-dvson multplexng (TDM), where each ONU s allocated a fxed or dynamc tme slot (transmsson wndow). Transmssons from dfferent ONUs to the OLT are arbtrated through the use of the multpont-control protocol (MPCP). Gven the steadly ncreasng number of users and emergng bandwdth ntensve applcatons, current sngle-channel TDM EPONs are lkely to be upgraded n order to satsfy the growng traffc demands n the future. One approach for upgradng EPON systems s to ncrease the current lne rate from 1 to 10 Gb/s [2]. However, ths mples that all EPON nodes need to be upgraded by nstallng new hgher speed transcevers, resultng n a rather costly upgrade. Another approach s to deploy multple wavelengths n the upstream/downstream drectons, resultng n a wavelength-dvson-multplexed (WDM)-based topology. WDM provdes a cautous upgrade, wheren wavelengths can be added as needed. Furthermore, only EPON nodes wth hgher traffc may be WDM upgraded by ether deployng fxed-tuned and/or tunable transcevers [7]. In ths paper, we ntroduce an archtecture for ncremental mgraton from TDM-PON to TDM/WDM-PON. Dynamcbandwdth-allocaton (DBA) algorthms ntally desgned for EPON requre approprate modfcatons to handle the multplechannel archtecture and to explot the nterwavelength statstcal multplexng. We present the new bandwdth-allocaton schemes for the hybrd WDM/TDM PON, and we show ther dfferences. These schemes enable dfferent ONUs to effcently share (both n tme and wavelength domans) the access-network bandwdth. In Secton II, we do an overvew /$ IEEE

2 278 JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 25, NO. 1, JANUARY 2007 of related lterature. Secton III presents the proposed archtecture, and n Secton IV, we present our bandwdth-allocaton schemes. Secton V presents a comparson between the proposed schemes. In Secton VI, we study numercally the performance of these schemes, and fnally, Secton VII concludes ths paper. II. RELATED WORK Varous early work has consdered the deployment of WDM technology n the access network, and some WDM-PON archtectures have been proposed, namely the composte PON, the local-access router network, the remote ntegraton of termnal network, the multstage AWG-based WDM-PON, and, more recently, the WDM-Super PON. See [2] for a lterature overvew of these technologes. One straghtforward approach to buld a hgh performance WDM-PON s to employ a separate wavelength channel from the OLT to each ONU for each of the upstream and downstream drectons [9]. Ths approach effectvely creates a pont-to-pont lnk between the central offce (CO) and each ONU; ths archtecture results, however, n a poor resource utlzaton and hgh deployment cost. The authors of [5] proposed a new hybrd archtecture (referred to as SUCCESS), whch provdes a practcal mgraton from current TDM PONs to future WDM access networks whle mantanng backward compatblty for users on exstng TDM PONs. The SUCCESS archtecture s based on a collector rng and several dstrbuton stars connectng the CO and the users. The authors proposed a partcular WDM-PON MAC protocol for ths archtecture but, however, dd not present any WDM-DBA algorthms. Furthermore, the archtecture does not allow for any nterchannel statstcal multplexng to better harness the avalable bandwdth on dfferent PONs. More recently, the authors of [6] proposed a SUCCESS-DWA PON that employs dynamcwavelength allocaton (DWA) to further provde bandwdth sharng across multple physcal PONs and, hence, acheve both cost-effectve and hgh-performance archtecture. The authors presented an upstream and downstream system upgrade; tunable lasers, arrayed-wavegude gratng, and coarse/fne flterng are combned to create a flexble access n the downstream. Alternatvely, several dstrbuted and centralzed-access schemes are proposed for the upstream upgrade. The authors of [3] have smlarly proposed a new WDM-PON, n whch each upstreamwavelength channel can be shared among multple ONUs by means of TDM. Here, the ONUs can use ther wavelengthselecton-free (.e., wthout wavelength tunng) transmtters to operate on any wavelength. No WDM-DBAs algorthms were dscussed, however. Wth respect to bandwdth and resource management, access control, and qualty of servce (QoS), some work only recently started to appear and remans very lmted. The authors of [7] have presented extensons to the MPCP protocol for WDM- PON, where wavelength channels, n addton to tme wndows, can be assgned; they presented both onlne and offlne schedulng. The authors of [10] proposed WDM IPACT-ST scheme, based on the nterleaved pollng wth adaptve cycle tme (IPACT) [12]. Here, IPACT protocol was adopted and appled on a multchannel WDM-PON, where the ONUs are equpped wth fxed transcevers. Furthermore, they appled strct prorty schedulng to support QoS n WDM-PON. A byte-sze clock (BSC) protocol [11] wth QoS support that allocates wavelengths on a user-bass rather than ONU-bass s proposed. The approach s scalable n bandwdth assgnment and acheves reducton n packet delays; however, n BSC, all nodes need to be synchronzed and, as a result, the TDM frame does not comply wth IEEE 802.3ah. III. WDM-PON ARCHITECTURE The protocols and algorthms for WDM-PON are currently at ther ntal stage of study, and whle varous types of archtectures have been proposed, no specfc one s domnant yet [2]. We assume two dfferent archtectures [14] for ths paper: The frst (scheme A 1 ) assumes a fxed groupng of ONUs [6, scheme A]. Here, the ONUs are dvded nto multple subsets, each allocated a fxed wavelength channel for upstream transmsson. Hence, every ONU mantans a fxed transcever, whereas the OLT mantans a bank of fxed transcevers. Wthn each subset, the transmsson of dfferent ONUs s arbtrated by the OLT through ether a fxed or dynamc tme-dvson slot-assgnment scheme. Clearly, ths archtecture lmts the shareablty of dfferent wavelengths among ONUs, snce a sngle wavelength s statcally allocated to each subset of ONUs, and hence, no nterchannel statstcal multplexng s possble. Ths archtecture can be vewed as a straghtforward upgrade from conventonal TDM-PON and provdes a baselne for comparson wth the proposed WDM-DBAs. The second archtecture (scheme A 2 ) s more flexble and allows for smultaneous tme-sharng and wavelength sharng [14]. Ths archtecture s smlar to scheme C presented n [6]. For upstream transmsson, every ONU can be equpped wth one or more fxed transmtters, allowng for an ncremental upgrade depng on the traffc demand at the ONU. In ths case, the ONU nforms, durng the regstraton process, the OLT of the wavelength(s) t can support for approprate resource allocaton and management. The OLT, upon recevng bandwdth requests, allocates transmsson wndows for the varous ONUs, takng nto account the wavelengths they support. Alternatvely, the ONU could optonally mantan a fast tunable laser to allow for more flexblty. To develop our dynamc-wavelength and bandwdth-allocaton (DWBA) algorthms, we assume n ths paper the latter approach and we assume a tunng speed n the range of mcroseconds. Ths archtecture enables the ONU to tune ts upstream transmsson from one wavelength to another at dfferent tmes depng on the DWBA algorthm deployed at the OLT. Here, the WDM-PON resources act as a pool, and all ONUs share these resources; resource sharng s arbtrated by the OLT usng DWBA. Ths scheme makes the mplementaton of the DWBA more challengng and requres an upgrade n the MAC. In our archtecture, we upgrade the MAC to support both tme and wavelength assgnment, where each ONU wll be allocated both a transmsson wndow and a wavelength. The OLT may have a bank of fxed transcevers to be able to smultaneously receve data from the varous ONUs on dfferent wavelengths and transmt data and control messages to the ONUs.

3 DHAINI et al.: DYNAMIC WAVELENGTH AND BANDWIDTH ALLOCATION IN HYBRID TDM/WDM EPON NETWORKS 279 IV. WDM/TDM DBA To develop our DWBAs, we assume MPCP extensons for WDM-PON, as proposed n [7]. The MPCP GATE message proposed n the standard s modfed by addng an addtonal feld (1 B) ndcatng the channel number assgned by the OLT to the ONU. Thus, the OLT wll provde each ONU wth ts approprate transmsson start tme T start, transmsson length T length, and correspondng wavelength-channel dentfer. A. Statc Wavelength Dynamc Tme (SWDT) Ths scheme reles on the smple archtecture A 1 ;theolt allocates wavelengths statcally among ONUs, and the upstream bandwdth s assgned dynamcally depng on the request of each ONU. Here, ONUs are dvded nto as many classes as the number of wavelengths, and each class wll share a predetermned wavelength. Snce the number of ONUs on each wavelength s dentfed, SWDT runs on each channel separately (the OLT wats untl all reports from one subset s receved and then runs the allocaton algorthm). Ths scheme s easy to mplement; however, t may under utlze the network bandwdth, snce t does not explot the nterchannel statstcal multplexng. Namely, when the load on one partcular channel s lght and hgh on another, the OLT cannot use the avalable bandwdth on that lghtly loaded channel and reassgn t to hghly loaded ONUs on another wavelength, whch therefore, could result n better performance. Note that ths scheme although does not allow for dynamc channel allocaton, t however allows for DBA [13] on one partcular wavelength. Hence, SWDT s used as a bass for our comparatve study. In order to motvate the need for dynamc channel allocaton, we frst consder a rather uncommon case, wheren the ONUs (hghly loaded and lghtly loaded) 1 are not symmetrcally dstrbuted on the channels. Ths scheme s referred to as SWDT-WC (WC for worst case). We also consder a more common case, where the ONUs are evenly dstrbuted on the channels (SWDT-BC, where BC means best case). Now, although ths case s lkely more common, t essentally s smlar to multple EPON networks, where each EPON runs a sngle DBA for optmal performance. It s worth notng that SWDT falls short n the case where the load on varous channels s not symmetrc or not evenly dstrbuted. B. Dynamc Wavelength Dynamc Tme (DWDT) Unlke the prevous approach where the channel s predetermned and fxed for every ONU and the OLT arbtrates only the transmsson of ONUs, the second approach reles on the second archtecture A 2 and enables the dynamc allocaton of bandwdth for dfferent ONUs n both wavelength and tme domans. Here, the OLT mantans a varable for every channel that desgnates the tme Tfree k for wavelength k when the next transmsson s possble on that partcular channel. For every 1 A hghly (lghtly) loaded ONU s one that requests bandwdth more (less) than the mnmum bandwdth guaranteed [B MIN, (2)] by the OLT n each cycle. REPORT message receved from any ONU, the OLT allocates a channel wth the least Tfree k to ths ONU; furthermore, t also determnes the length (e.g., n bytes) of the transmsson wndow allocated to ths ONU on the assgned channel. We refer to ths procedure as DWBA, and we present three varants namely DWBA-1, DWBA-2, and DWBA-3. In these varants, the mnmum bandwdth guaranteed B MIN defned n [8] s depent on the weght assgned to each ONU, based on the servce-level agreement (SLA) between the servce provder and users. We consder a PON access network wth N ONUs. The transmsson speed of the PON s R N (n Megabts per second). Let T cycle be the grantng cycle, whch s the tme durng whch all ONUs can transmt data or/and s REPORTs to the OLT. Let T g be the guard tme that separates the transmsson wndow for ONU n and ONU n+1 and w be the weght assgned to each ONU based on ts SLA such that N =1 w =1. Therefore, the mnmum bandwdth guaranteed per cycle the OLT can allocate for an ONU() s computed as follows: B MIN = (T cycle N T g ) R N K w 8 where K s the total number of wavelengths. In case of no SLA classfcaton per ONU, w = w =1/N,, and N =1 w =1; then B MIN (1) = B MIN = (T cycle N T g ) R N K. (2) 8 N 1) DWBA-1: The OLT wats untl all the REPORTs are receved from all ONUs (on all channels). Upon that, the OLT runs a bandwdth-allocaton algorthm to determne the bandwdth and channel for every ONU. Here, f Breq B MIN, where Breq s the requested bandwdth by ONU, and B MIN s the mnmum bandwdth guaranteed [4], [8], then Bassgn = Breq, and a GATE message s sent to ONU. Alternatvely, f Breq >B MIN, then the OLT computes the excessve bandwdth resultng from the lghtly loaded ONUs and assgns to ONU a bandwdth Bassgn, depng on the excess bandwdthallocaton type, and ss a GATE message accordngly. There are two ways to assgn transmsson wndows usng the excess bandwdth, namely controlled excess (CE) and uncontrolled excess (UE). In UE scheme, the OLT collects from the receved REPORTs all the excessve bandwdth avalable for the next cycle and assgns ths total excess unformly to all hghly loaded ONUs, regardless of ther requested bandwdth. The total excess bandwdth s Then B total excess = N (B MIN Breq) B req B MIN. (3) =1 B excess = B total excess/m (4) where M denotes the number of overloaded ONUs. The advantage of ths uncontrolled scheme s that hghly loaded ONUs are assgned enough bandwdth to satsfy ther hgh demands

4 280 JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 25, NO. 1, JANUARY 2007 Fg. 1. DWBA-2 protocol. (assumng the excess s enough); however, f some ONUs are only slghtly hghly loaded, they are beng assgned an unfar share of the excess bandwdth that could ultmately be not utlzed. Hence, the assgnment of the excess bandwdth must be controlled (.e., CE) by the OLT n order to guarantee a far bandwdth allocaton for all hghly loaded ONUs. A more controlled scheme may work as follows: B assgn Breq, = Breq, B MIN +Bexcess, f Breq B MIN f B MIN <Breq B MIN +Bexcess f B MIN <B MIN +B excess <B req where the assgnment of the excess bandwdth s controlled n the followng way. Let χ = {ONU } =0,...,M 1 be the set of hghly loaded ONUs. Then, B excess s computed as follows: B excess { B total = Breq B MIN, excess/(m ), (5) f B MIN + ( Bexcess/(M total ) ) <Breq otherwse (6a) where the total excess B total excess s updated as follows every tme B excess s assgned B excess total = B excess total B excess. (6b) However, the CE scheme allocates the excessve bandwdth n a round-robn fashon. Thus, some hghly loaded ONUs mght not have the chance to receve any share of ths bandwdth due to the fact that Btotal excess wll be 0 before vstng all ONUs, or n a very common case, these last ONUs mght get a less share than the frst ones. For that reason, we propose a far-excess (FE)-allocaton scheme that assgns portons to hghly loaded ONUs accordng to ther bandwdth demand. Let Breq excess, = Breq B MIN be the excess bandwdth requested from a hghly loaded ONU and Bexcess req = N =0 Bexcess, req be the total excess requested bandwdth from all ONUs; then B porton, excess = Bexcess, req B excess total Bexcess req (7a) where Bexcess porton, s the computed porton of excess bandwdth for each hghly loaded ONU. Hence, to prevent the waste of bandwdth, Bexcess s computed as follows: B excess = mn ( B excess, req,bexcess porton, ). (7b) As a result, FE wll ensure far excess bandwdth allocaton among all hghly loaded ONUs. Note that unlke CE, FE ensures a far bandwdth allocaton but mght not satsfy any hghly loaded ONU; on the other hand, CE makes sure to satsfy the demand of a hghly loaded ONU, f enough excess bandwdth s avalable, but not all ONUs, n case all the total excess bandwdth s fully exploted. Now, for the wavelength selecton crtera, as mentoned before, the OLT mantans, for every wavelength k, the tme t becomes avalable for next transmsson Tfree k,k =1,...,K, where K s the total number of wavelengths n the WDM PON. The channel wth smallest s selected for next transmsson. 2) DWBA-2 (see Fg. 1) : Here, upon recevng a REPORT from ONU, the OLT checks whether Breq B MIN ;nths case, the OLT assgns on the fly a GATE to that ONU wth Tfree k

5 DHAINI et al.: DYNAMIC WAVELENGTH AND BANDWIDTH ALLOCATION IN HYBRID TDM/WDM EPON NETWORKS 281 bandwdth Bassgn = B req. Otherwse, the OLT wats untl all the REPORTs from the other ONUs are receved and, then, assgns a bandwdth of Bassgn computed usng UE, CE, or FE. The dfference here s that ONUs that are lghtly loaded can be scheduled mmedately on the partcular channel wthout watng for the rest of the ONUs to s REPORTs. Ths early allocaton wll result n mproved delay performance. However, such a scheme may ncrease the complexty of the desgn and mplementaton of the DWBA due to the fact that the OLT wll have to keep track of each REPORT message receved from each ONU (e.g., sometmes one ONU can s two or more REPORTs before the OLT receves all the other REPORTs because of the grant-on-the-fly manner). Hence, the OLT wll have to store excess nformaton that holds the status of each ONU (hghly or lghtly loaded) to be able to assgn the approprate transmsson wndow. 3) DWBA-3: Here, the OLT wll always assgn on the fly, a GATE to the ONU regardless of ts requested bandwdth. However, the sze of the transmsson wndow s depent on the requested bandwdth. Upon recevng a REPORT from ONU, the OLT checks f Breq B MIN. In ths case, as n DWBA-2, the OLT wll assgn, on the fly, a GATE wth Bassgn = B req; otherwse, t wll assgn, on the fly, a GATE wth Bassgn = B MIN. Subsequently, the OLT wats untl t receves all the REPORTs from all ONUs and collects the nformaton about the excess bandwdth from each channel as well as the number of hghly loaded ONUs M. Each hghly loaded ONU s allocated ts share of the excess bandwdth n ether an uncontrolled manner [Bexcess as n (4)], n a controlled manner as n (6a), or n a far manner as n (7b). Note, here, the REPORT message s always transmtted once by the ONU n the frst assgned transmsson wndow (.e., not n the excess wndow) regardless of whether an excess bandwdth s assgned or not. Ths s because 1) the allocaton of the excess wndow cannot be guaranteed for a partcular ONU and 2) snce the OLT ss a GATE upon the recept of a REPORT (.e., on the fly), the ONU should not s a second REPORT (.e., n the excess wndow) n the same cycle. That s because the OLT may already have done the schedulng of other ONUs over the same channel, and ths second REPORT cannot vod the frst receved one. Ths scheme s consdered complex as well, snce the OLT wll have to use ts excess table and, at the same tme, wll have to keep track of the two to-be-sent (f applcable) GATE messages to each ONU. V. C OMPARISON AND ANALYSIS As mentoned earler, SWDT reles on a statc and predetermned allocaton of wavelengths to the ONUs and the OLT performs DBA on each channel. In the case where the load on dfferent wavelengths s not symmetrc (.e., some wavelength channels are more loaded than others), SWDT cannot explot the bandwdth avalable on one channel (.e., the lghtly loaded) to allocate t for ONUs resdng on another more congested channel. Ths results n under utlzng the avalable resources and, hence, n ncreased delays on the congested channel, and t s manly attrbuted to the lack of nterchannel statstcal multplexng. Note that when the load s evenly dstrbuted among the PON wavelengths, the problem reduces to performng effcent bandwdth allocaton on each ndvdual channel. Alternatvely, DWDT enables ONUs to share the network resources both n tme and wavelength domans. Frst, DWBA- 1 s straghtforward; the OLT allocates GRANT messages only after recevng all the REPORTs from all the ONUs (N) n the network. Evdently, ths smple algorthm has defcences; namely, consder a two-channel PON network, where t 1 and t 2 are the tmes where each of the channels s avalable (assume t 1 t 2 ) for transmssons. We can compute the perod durng whch channel 1 s not beng utlzed: T dle 1 =(t 2 t 1 )+ζ (8) where ζ = τ + T transmsson + T DWBA. Here, τ and T transmsson are the RTT and the transmsson tme of a GATE message from the OLT to the ONU. T DWBA s the computaton tme. Clearly, the upper bound of T1 dle corresponds to the maxmum of (t 2 t 1 ). Ths maxmum, n turn, corresponds to the case where the last ONU n ths cycle starts ts transmsson (on channel 2) at tme t start t 1. Accordngly, the upper bound of (t 2 t 1 ) s Tassgn N (the tme wndow n seconds assgned to the last ONU). Therefore T dle 1 T N assgn + ζ. (9) Moreover, f t 2 t 1 s large (e.g., when the majorty of ONUs on channel 1 are lghtly loaded and hghly loaded on channel 2), then the perod of tme where channel 1 s dle s much larger than that of (8). Overall, ths dle tme experenced by a channel results n poor bandwdth utlzaton and, thus, ncreased overall packet delays. DWBA-2 and DWBA-3 solve ths effcency problem by sng GATE messages on the fly to all ONUs requestng bandwdth less than the mnmum guaranteed (B MIN ).Ths on the fly bandwdth assgnment mtgates the effects of the channel dle tme experenced by DWBA-1 and results n a better throughput and delay performance. However, these two schemes exhbt dfferent behavors. Namely, under DWBA-2, the OLT defers ONUs wth B req >B MIN untl all REPORTs are receved, and then, t performs ts assgnment. DWBA-3 rather assgns on the fly all ONUs ncludng those wth B req >B MIN. In other words, the bandwdth allocated to a hghly loaded ONU n DWBA-2 s a sngle entty, whereas DWBA-3 segregates the excess from the mnmum bandwdth. Ths results n two transmsson wndows beng allocated at dfferent tmes to a hghly loaded ONU n the same cycle. The mmedate mplcaton of ths segregaton s that a large packet may not ft n the frst granted wndow and gets deferred to the second granted wndow (or perhaps to a subsequent cycle), blockng other packets and wastng a fracton of the allocated bandwdth. Ths mplcaton ncreases the packet delays by holdng unnecessarly these packets. On the other hand, n DWBA-2, every ONU s allocated only one transmsson wndow n the same cycle; ths wndow (combnng both B MIN and B excess ) s large enough and may thus mtgate the mpact of the

6 282 JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 25, NO. 1, JANUARY 2007 prevous problem. We demonstrate ths through an example. Frst, let BMIN Total be the total number of bytes of transmtted data n the wndow allocated for B MIN ;letbexcess Total be the total number of bytes of transmtted data n the wndow assgned for the excess bandwdth B excess.letbmin+excess Total be the total number of bytes of transmtted data n the wndow assgned for B MIN combned wth B excess. Let x be the remanng bandwdth from B MIN, y from B excess (when DWBA-3 s used), and z from B MIN+excess (when DWBA-2 s used). In DWBA-3, snce the allocated bandwdth n one cycle s splt nto two wndows (B MIN and B excess ), one packet P of large sze p may not ft n x and, hence, gets deferred to the next transmsson wndow. Ths wll effectvely prevent other packets from beng transmtted and result n neffcent use of the allocated bandwdth. A total bandwdth of (x + y) s, hence, not beng utlzed by the ONU under DWBA-3. On the other hand, n DWBA-2 combnng B MIN and B excess n one transmsson wndow, enables the transmsson of P and, hence, releases or unblocks the rest of the buffered packets. Ths ultmately allows the transmsson of larger number of packets; therefore, reduced packet delays and ncreased bandwdth effcency (wasted allocated bandwdth s only z, z x + y). Another mplcaton of allocatng two wndows n the same cycle for a hgh loaded ONU stems from the fact that the ONU reports ts buffer occupancy n the frst wndow of cycle n 1, and subsequently, after some tme, some already reported packets for the next cycle wll be transmtted durng the excess wndow. The OLT wll allocate bandwdth for cycle n based on the reported traffc from the prevous cycle. Ths wll result n grantng bandwdth more than needed snce the buffer occupancy has decreased. Let t 1,n 1 be the tme where the REPORT message s transmtted by the ONU n cycle n 1 and Q(t 1,n 1 ) be the buffer occupancy (n bytes) at tme t 1,n 1. Smlarly, let t 2,n 1 be the tme where the same ONU fnshes sng traffc n the excess wndow of cycle n 1 and Q(t 2,n 1 ) be the buffer occupancy (n bytes) at tme t 2,n 1. Here, the requested bandwdth for cycle n s Breq n = Q(t 1,n 1 ). Now, we can wrte ( ( B req n = Breq n Q t 1,n 1 ) ( Q t 2,n 1 )) (10) where B req n s the amed value of Breq n before the OLT performs DBA n cycle n, and Q(t 1,n 1 ) Q(t 2,n 1 ) s the sze of the excess wndow of cycle n 1, whch s known by the OLT. Hence, the OLT wll allocate more bandwdth than the ONU requres at the tme t 2,n 1. Ths wll result n ncreasng the cycle tme and, hence, neffcent use of the allocated bandwdth. To overcome ths defcency, we propose a modfed verson of DWBA-3 (DWBA-3a) that elmnates the effect of the outdated nformaton at the OLTs. Here, the OLT keeps track of the allocated excess bandwdth Bexcess,n 1 n cycle n 1 and, then, extracts ths excess out of the allocated bandwdth n cycle n. Consequently, the allocated bandwdth, n cycle n, B,n alloc s computed as follows: B,n alloc = Bn req B,n 1 excess. (11) Fg. 2. Average packet delay comparson (K =2). VI. PERFORMANCE EVALUATION In ths secton, we study the performance of the dfferent wavelength and bandwdth-allocaton algorthms we presented. We developed a WDM-PON event drven smulator n C++. The followng are some parameters used n our smulaton: number of ONUs N =64; number of wavelengths K =(2, 4, 6, 8); maxmum cycle tme =2ms; speed of each wavelength s 1 Gb/s; guard tme =1 µs; dstance between OLT and ONU s 20 km; dstance between ONU and -users s 5 km; and bufferng queue sze s 1 MB. Out of the 64 ONUs, 32 ONUS are lghtly loaded, whereas the remanng ONUs are hghly loaded. A lghtly loaded ONU generates traffc at a rate of 10 Mb/s(load =0.1). We consder bursty traffc; to model ts bursty nature, we generated self-smlar traffc based on a Pareto dstrbuton wth a hurst parameter H =0.8, and packet szes are unformly dstrbuted between 64 and 1518 B. Fg. 2 presents the network average delay for the varous allocaton algorthms presented earler when K =2, all under the UE scheme. The traffc load of a hgh loaded ONU s vared between 0.1 and 1 (.e., 10 and 100 Mb/s). The results frst show clearly that when the traffc load s not symmetrc and evenly dstrbuted on the varous channels, SWDT shows the worst performance (.e., SWDT-WC), especally at medum and hgher loads. Ths s due to the lack of nterchannel statstcal multplexng, wheren the OLT could allocate avalable resources on one channel and ONUs on another channel, snce the channels n SWDT are already preallocated and the allocaton s fxed. Alternatvely, when the load s more evenly dstrbuted (.e., lghtly and hghly loaded ONUs are unformly dstrbuted on the channels), SWDT (SWDT-BC here) exhbts a good performance, n comparson wth the other schemes, especally DWBA-1. Ths shows that under such crcumstance, there s no need for dynamc channel allocaton, and hence, only DBA on each channel s requred. However, as the load becomes less symmetrc, dynamc channel allocaton becomes essental (as shown n the other extreme, SWDT-WC) n order to explot all the avalable network resources. It s worth notng that SWDT-WC and SWDT-BC at very lght load (e.g., 0.1) perform alke snce both channels have smlar lght loads. Note also

7 DHAINI et al.: DYNAMIC WAVELENGTH AND BANDWIDTH ALLOCATION IN HYBRID TDM/WDM EPON NETWORKS 283 Fg. 3. Delay measurements wth CE and FE (K =2). Fg. 4. Delay measurements wth K =4, 6, and 8. that SWDT performs smlar to DWBA-1 at lght load, whle DWBA-2 and DWBA-3 slghtly outperform the other schemes because they both allocate bandwdth on the fly for lghtly loaded ONUs (.e., almost all ONUs are at lght load), whereas the other schemes wat untl all REPORT messages are receved (e.g., SWDT wats untl REPORT from ONUs on each channel are receved and DWBA-1 wats untl REPORTs from all ONUs are receved). As the load gets hgher, DWBA-1 performs better than SWDT-WC because the former allows for sharng the resources on all channels and slghtly underperforms SWDT- BC. Moreover, as the load gets hgher, the DWBA-2 and DWBA-3 schemes outperform both SWDT algorthms; frst, they outperform SWDT-WC, snce the latter lacks the resource sharng property due to the statc-channel allocaton. They also outperform SWDT-BC, although the load s symmetrc under SWDT-BC, for two reasons: 1) DWBA-2 and DWBA-3 allocate lghtly loaded ONUs on the fly, and 2) although the load s evenly dstrbuted among the channels, we have seen n some nstances that a channel could have some free resources that could be mmedately exploted by DWBA to allocate for ONUs on another channel. Ths s manly due to the bursty nature of traffc used n our smulatons. Fg. 2 also shows that at a load of 0.3, the average packet delay under DWBA-2 (DWBA-3) s ms (15.66 ms) better than that under DWBA-1. Furthermore, our smulaton results show a better performance of DWBA-2 over DWBA-3. The man reason s due to the fact that n DWBA-3, the OLT allocates the excess bandwdth to a hgh loaded ONU n a separate wndow n the same cycle. Ths results n under-utlzng the allocated bandwdth. The average packet delay of DWBA-2 s slghtly better than that of DWBA-3; for example, at load =0.4, a dfference of 12.8 ms s shown. Note, when the assgnment of the excess s controlled by the OLT (as dscussed n Secton IV), better results are obtaned n terms of overall average packet delays; however, the relatve dfference between the dfferent algorthms s the same. Fg. 3 shows the delay performance of DWBA-2 and DWBA-3 usng both CE and FE allocaton schemes and presents a comparson wth IPACT-ST [10] when K =2. Clearly, DWBA-2 and DWBA-3 exhbt better per- formance than IPACT-ST. For example, at load =0.4, there s 16-ms mprovement of DWBA-2 (CE) over IPACT-ST. Here, although IPACT-ST assgns bandwdth for every ONU on the fly as soon as t receves ts REPORT, t does not beneft from the excess bandwdth, whch may be avalable n the network to assgn t for hghly loaded ONUs (as n DWBA schemes). Hence, the avalable excess bandwdth n each transmsson cycle s effcently utlzed under DWBA, and as a result, the overall network performance s amelorated. Moreover, FE mproves the performance of DWBA-3 over CE, whle t shows a smlar performance for DWBA-2 usng the CE allocaton. Ths s due to the fact that FE farly allocates the excess bandwdth among hghly loaded ONUs, n contrast to CE, that concentrates on satsfyng hghly loaded ONUs untl the avalable bandwdth s fully consumed. Next, we study and compare the performance of DWBA-2 and IPACT-ST when the number of wavelengths K =4, 6, and 8 (Fg. 4). It s mportant to frst note that ncreasng the number of wavelengths whle usng the same number of ONUs (64 s the lmt) 2 yelds fewer ONUs on the same wavelength channel, and accordngly, the mnmum bandwdth guaranteed [B MIN, (2)] ncreases. Ths has clear mplcatons on the performance of DWBA. Recall that DWBA assgns on the fly, lke IPACT-ST, those ONUs requestng bandwdth less than B MIN and, unlke IPACT-ST, defers the allocaton for ONUs requestng more. Now, as B MIN becomes larger, when an ONU becomes hghly loaded, t would be stll requestng ether less than B MIN or slghtly more than B MIN (the most loaded ONU ss at 100-Mb/s rate). If the ONU requests less than B MIN, t wll then be allocated bandwdth on the fly, just lke IPACT-ST does. Therefore, IPACT-ST and DWBA-2 show smlar performance, as s clearly shown n Fg. 4 for loads varyng between 0.1 and 0.7. However, as the load ncreases further, nterestngly, IPACT-ST outperforms DWBA-2. Ths s manly due to those ONUs requestng more than B MIN, as t s the only dfference wth IPACT-ST. When B MIN s large, as we have mentoned earler, a hghly loaded ONU may only request a bandwdth 2 Ths s determned by the physcal lmtatons of the optcal spltter.

8 284 JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 25, NO. 1, JANUARY 2007 that s slghtly hgher than B MIN. In such a case, the ONU wll be deferred untl all REPORT messages arrve n order to receve only a small excess bandwdth (ths s not because there s no excess bandwdth avalable, but because the ONU needs only lttle excess). In such a case, there s no clear payoff for a hghly loaded ONU to wat untl all REPORT messages to arrve n order to only receve a small excess bandwdth. IPACT- ST, unlke DWBA-2, allocates bandwdth to all ONUs on the fly. Ths dfference leads IPACT-ST to outperform DWBA-2 at very hgh loads. Now, as the number of wavelengths ncreases further (e.g., 6 and 8), each wavelength wll support fewer ONUs, and the mnmum bandwdth guaranteed gets further larger. Under such a crcumstance, regardless of the load at each ONU (wth an excepton f the load s allowed to go beyond 100 Mb/s), all or most ONUs wll be granted on the fly, smlar to IPACT-ST, and all channels wll be less loaded. For example, when the number of wavelengths s sx (.e., 6 Gb/s of bandwdth avalable), the total hghest load generated n the network s Mb/s Mb/s = Gb/s, whch s much smaller than the total bandwdth avalable. Ths clearly shows that all wavelengths are not fully utlzed, and hence, smaller delays (Fg. 4) are experenced under DWBA-2 and IAPCT-ST. Furthermore, the performance of both DWBA-2 and IPACT-ST become smlar; as K gets larger, the average packet delays reman almost fxed at varous loads. The average delay equals to almost 0.4 ms, whch accounts for the cumulatve guard tmes, the round-trp propagaton delays, as well as the packet-queung delays. Although the average packet delay s small and smlar at varous wavelengths, we note that the maxmum packet delays have exhbted some dfferences. That s, as the number of wavelengths ncreases, the maxmum packet delay slghtly decreases [e.g., a dfference of 2 ms between DWBA-2 (K =6) and DWBA-2 (K =8)]. Fnally, we should note that ncreasng the number of wavelengths beyond the overall hghest load the ONUs can s to the network may not be wse from a desgn and operaton perspectve. We now study the dssmlartes between DWBA-2 and DWBA-3 schemes. Recall that the man dfference between DWBA-2 and DWBA-3 s the effcent use of the total allocated bandwdth for hgh loaded ONUs. Clearly, f the allocated bandwdth s not fully utlzed (as n DWBA-3), the buffer occupancy of the hgh loaded ONU wll ncrease substantally, and therefore, more bandwdth wll be requested for the subsequent cycle(s). Snce n our expermental setup, about half of the ONUs are hghly loaded, more hgh bandwdth requests wll arrve at the OLT and each ONU wll be rewarded excess from whatever s avalable. Ths neffcency of utlzng the allocated bandwdth may occur more often, and thus, may accumulate throughout the duraton of the burst and after. Alternatvely, under DWBA-2, the behavor s conversed. Namely, the allocated bandwdth s used more effcently, and hence, fewer ONUs wll be requestng addtonal bandwdth. To valdate our reasonng, we measure the probablty densty functon (pdf) of the number of ONUs wth B req B mn for both algorthms and under the two allocaton schemes of excess bandwdth (UE and CE) and when K =2. Clearly, as Fg. 5 shows, more ONUs wll be requestng bandwdth more than B mn under DWBA-3 (both under UE and CE). However, under DWBA-2, fewer ONUs Fg. 5. PDF of number of ONUs (wth B req >B MIN )(K =2). are always requestng more than B mn. Ths clearly ndcates that 1) the neffcent use of allocated bandwdth and 2) the msguded allocaton of bandwdth n DWBA-3 both result n ncreased queung of ONUs traffc (hgher occupancy), and hence, more ONUs are requestng bandwdth larger than the mnmum guaranteed from the OLT. To further compare the performance of the two algorthms wth respect to ther effcent use of the allocated bandwdth, we measure the number of bytes wasted n each cycle for a hghly loaded ONU under CE scheme: B wasted = B alloc B sent, where B alloc and B sent are the amount of bandwdth allocated to the ONU and effectvely used by the ONU, respectvely. Fg. 6 shows the results of ths experment, where we plot B wasted collected n each cycle throughout the smulaton at a partcular hgh loaded ONU. Clearly, the fgure shows that, under DWBA-2, there exsts no cycle where B wasted 1518 B (maxmum packet sze). By concluson, DWBA-2 wll defer at most one packet from one transmsson cycle to the followng one; however, there s an excessve waste of allocated bandwdth under DWBA-3. Ths s largely due to the over-allocaton of unnecessary bandwdth by the OLT. Overall, ths allocaton results n ncreased average cycle tmes, neffcent bandwdth utlzaton, and therefore, ncreased overall packet delays. We mtgate ths problem by proposng a modfed verson of DWBA-3, namely DWBA-3a (see Secton V). We have seen (results are skpped) that DWBA- 3a sgnfcantly decreases the waste of bandwdth found n DWBA-3 and shows a smlar behavor to the one observed usng DWBA-2 (.e., B wasted 1518 B). Fnally, to compare the farness of both CE and FE allocaton schemes, we run our smulaton at load =0.5, and we measure the performance of two partcular hghly loaded ONUs. We choose the frst ONU as the frst hghly loaded ONU (namely ONU 1 ) among the 64, snce t s expected to be always satsfed when applyng the CE scheme (see Secton IV), whle the second ONU (namely ONU 2 ) s chosen to be the last one among the 64. As expected and observed, the average packet delay for ONU 1 wth FE s equal to s, and for ONU 2, t s equal to s. Whle wth CE, t s s for ONU 1 and s for ONU 2. Ths shows the advantage of FE that farly allocates the excess bandwdth and, thus, provde

9 DHAINI et al.: DYNAMIC WAVELENGTH AND BANDWIDTH ALLOCATION IN HYBRID TDM/WDM EPON NETWORKS 285 Fg. 6. Wasted bandwdth comparson. almost the same performance for all hghly loaded ONUs, whereas CE satsfes one hghly loaded ONU over the other and manly deps on the avalablty of the remanng excess bandwdth. VII. CONCLUSION We proposed a WDM-PON archtecture whch allows for hgh bandwdth utlzaton among multple wavelengths. We presented new bandwdth-allocaton schemes and provded a thorough comparson between them and studed ther advantages and dsadvantages. We showed that statc-wavelength allocaton may penalze ONUs wth hgh load and wll underutlze the PON resources, especally f the load s not symmetrc. We showed that DWA ncreases the network effcency. We presented three ways to effcently allocate excess bandwdth among hghly loaded ONUs, namely controlled, far, and uncontrolled. We showed that by usng CE bandwdth allocaton, we ncrease the bandwdth utlzaton, and as a result, we mprove the overall network performance. We also studed the performance of our schemes and compared wth IPACT-ST as the number of wavelengths n the network ncreases. [7] M. P. McGarry, M. Maer, and M. Resslen, WDM ethernet passve optcal networks (EPONs), IEEE Commun. Mag., vol. 44, no. 2, pp , Feb [8] C. M. Ass, Y. Ye, S. Dxt, and M. A. Al, Dynamc bandwdth allocaton for Qualty-of-Servce over Ethernet PONs, IEEE J. Sel. AreasCommun., vol. 21, no. 9, pp , Nov [9] J.-I.Kan,M.Teshma,K.Akmoto,N.Takacho,H.Suzuk,K.Iwatsuk, and M. Ish, A WDM-based optcal access network for wde-area ggabt access servces, IEEE Commun. Mag., vol. 41, no. 2, pp. S43 S48, Feb [10] K. H. Kwong, D. Harle, and I. Andonovc, Dynamc bandwdth allocaton algorthm for dfferentated servces over WDM EPONs, n Proc., IEEE ICCS, Sngapore, Sep. 2004, pp [11] C. Xao, B. Bng, and G. K. Chang, An effcent reservaton MAC protocol wth preallocaton for hgh-speed WDM passve optcal networks, n Proc., IEEE INFOCOM, Mam, FL, Mar. 2005, pp [12] G. Kramer, B. Mukherjee, and G. Pesavento, IPACT: A dynamc protocol for an ethernet PON (EPON), IEEE Commun. Mag., vol. 40, no. 2, pp , Feb [13] M. P. McGarry, M. Maer, and M. Resslen, Ethernet PONs: A survey of dynamc bandwdth allocaton (DBA) algorthms, IEEE Commun. Mag., vol. 42, no. 8, pp. S8 S15, Aug [14] A. Dhan, Desgn and analyss of next generaton ethernet-based passve optcal access networks, M.S. thess, Concorda Unv., Montreal, QC, Canada, REFERENCES [1] H. Shnohara, Broadband access n Japan: Rapdly growng FTTH market, IEEE Commun. Mag., vol. 43, no. 9, pp , Sep [2] A. Banerjee, Y. Park, F. Clarke, H. Song, S. Yang, G. Kramer, K. Km, and B. Mukherjee, Wavelength-dvson-multplexed passve optcal network (WDM-PON) technologes for broadband access: A revew, OSA J. Opt. Netw., vol. 4, no. 11, pp , Nov [3] D. J. Shn, D. K. Jung, H. S. Shn, J. W. Kwon, S. Hwang, Y. Oh, and C. Shm, Hybrd WDM/TDM-PON wth wavelength- selecton-free transmtters, J. Lghtw. Technol., vol. 23, no. 1, pp , Jan [4] G. Kramer, B. Mukherjee, and G. Pesavento, Ethernet PON (epon): Desgn and analyss of an optcal access network, Photon. Netw. Commun., vol. 3, no. 3, pp , Jul [5] F. An, K. S. Km, D. Guterrez, S. Yam, E. Hu, K. Shrkhande, and L. G. Kazovsky, SUCCESS: A next-generaton hybrd WDM/TDM optcal access network archtecture, J. Lghtw. Technol., vol. 22, no. 11, pp , Nov [6] Y.-L. Hsueh, M. S. Rogge, S. Yamamoto, and L. G. Kazovsky, A hghly flexble and effcent passve optcal network employng dynamc wavelength allocaton, J. Lghtw. Technol., vol. 23, no. 1, pp , Jan Ahmad R. Dhan receved the B.S. degree n computer scence from the Amercan Unversty of Berut, Berut, Lebanon, n 2004 and the M.S. degree n electrcal and computer engneerng wth a best thess award nomnaton from Concorda Unversty, Montreal, QC, Canada, where he s currently workng toward hs Ph.D. degree. He s wth the Faculty of Engneerng and Computer Scence, Concorda Insttute for Informaton Systems Engneerng, Concorda Unversty. Hs current research nterests are n the areas of wred/wreless-access networks, qualty-of-servce, and schedulng algorthms. More specfcally, he s currently workng n the areas of passve optcal networks.

10 286 JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 25, NO. 1, JANUARY 2007 Chad M. Ass (M 03) receved the B.S. degree n engneerng from the Lebanese Unversty, Berut, Lebanon, n 1997 and the Ph.D. degree from the Graduate Center, Cty Unversty of New York, NY, n Aprl He was a Vstng Researcher at Noka Research Center, Boston, MA, from September 2002 to August 2003, workng on qualty-of-servce n optcal-access networks. He joned the Concorda Insttute for Informaton Systems Engneerng, Concorda Unversty, Montreal, QC, Canada, n August 2003, as an Assstant Professor. Hs current research nterests are n the areas of provsonng and restoraton of optcal networks, wreless and ad hoc networks, and securty. Dr. Ass s a recpent of the Mna Rees Dssertaton Award from the Cty Unversty of New York n August 2002 for hs research on wavelengthdvson-multplexng optcal networks. Abdallah Sham (S 01 M 02) receved the B.E. degree n electrcal and computer engneerng from the Lebanese Unversty, Berut, Lebanon, and the Ph.D. degree n electrcal engneerng from the Graduate School and Unversty Center, Cty Unversty of New York, NY, n 1997 and September In September 2002, he joned the Department of Electrcal Engneerng, Lakehead Unversty, Thunder Bay, ON, Canada, as an Assstant Professor. Snce July 2004, he has been wth the Unversty of Western Ontaro, London, ON, where he s currently an Assstant Professor n the Department of Electrcal and Computer Engneerng. Hs current research nterests are n the area of data/optcal networkng, EPON, WIMAX, WLANs, and software tools. Dr. Sham s a recpent of the Irvng Hochberg Dssertaton Fellowshp Award from the Cty Unversty of New York and a GTF Teachng Fellowshp. Martn Maer (M 04) receved the Dpl.-Ing. and Dr.-Ing. degrees (both wth dstnctons) n electrcal engneerng from the Techncal Unversty Berln, Berln, Germany, n 1998 and 2003, respectvely. He s currently an Assocate Professor wth the Insttut Natonal de la Recherche Scentfque, Montreal, QC, Canada. He was a Vstng Researcher at the Unversty of Southern Calforna, Los Angeles, and Arzona State Unversty, Tempe. In summer 2003, he was a Postdoc Fellow at the Masschusetts Insttute of Technology, Cambrdge. Hs research nterests nclude network and node archtectures, routng and swtchng paradgms, protecton, restoraton, multcastng, and the desgn, performance evaluaton, and optmzaton of MAC protocols for optcal wavelength-dvson-multplexng (WDM) networks, automatcally swtched optcal networks (ASONs), and generalzed multprotocol label swtchng (GMPLS), wth partcular focus on metro and access networks. Recently, hs research nterests have concentrated on evolutonary WDM upgrades of optcal metro-rng networks and access networks. He s the author of the book Metropoltan Area WDM Networks An AWG Based Approach (Sprnger, 2003). Dr. Maer s a recpent of the two-year Deutsche Telekom doctoral scholarshp from June 1999 to May He s also a corecpent of the Best Paper Award presented at The Internatonal Socety of Optcal Engneers Photoncs East 2000-Terabt Optcal Networkng Conference. He served on the Techncal Program Commttees of IEEE INFOCOM, IEEE GLOBECOM, and IEEE IPCCC. He s an Edtoral Board Member of the IEEE Communcatons Surveys and Tutorals.