Cooperatve Multcast Schedulng Scheme for IPTV Servce over IEEE 802.16 Networks Fen Hou 1, Ln X. Ca 1, James She 1, Pn-Han Ho 1, Xuemn (Sherman Shen 1, and Junshan Zhang 2 Unversty of Waterloo, Waterloo, Ontaro, Canada 1 {fhou, lca, ames, pnhan, xshen}@bbcr.uwaterloo.ca Arzona State Unversty, Tempe, AZ, USA 2 unshan.zhang@asu.edu Abstract Explotng the broadcast nature of wreless communcatons, multcast transmsson s an effcent way to mprove the network throughput by transmttng the same contents to multple recevers smultaneously. It has been consdered as a key technology for supportng emergng servces n next-generaton IEEE 802.16 based wreless metropoltan area networks (WMANs, such as Internet Protocol TV (IPTV and moble TV. Therefore, t s crtcal to devse effcent multcast schedulng schemes to support these multmeda servces. In ths paper, we propose a novel multcast schedulng scheme, usng downlnk cooperatve transmsson for achevng hgh throughput not only for all multcast groups but also for each group member. Extensve smulatons are conducted to demonstrate the effectveness and effcency of the proposed scheme. I. INTRODUCTION Wth the ncreasng demand on multmeda applcatons, multmeda multcast servces have been attractng great attenton from both academa and ndustry. Multmeda Broadcast Multcast Servce (MBMS has been standardzed by the thrd generaton partnershp proect (3GPP [1]. Internet Protocol TeleVson (IPTV, as one of the most mportant broadband multmeda applcatons, s expected to contrbute mmense market value to the servce provders n next generaton wreless networks [2]. Meanwhle, IEEE 802.16 networks have been consdered as a promsng alternatve to cable and dgtal subscrber lne (DSL for provdng hgh-qualty broadband wreless multmeda servces, due to ts capablty of provdng QoS satsfacton [3]. Therefore, how to fulfll effcent multcast transmsson s a crtcal ssue for supportng IPTV servces over IEEE 802.16 networks. Multcast schedulng plays a crtcal role n achevng effcent multcast transmssons. In general, the users amng to receve the same copy of data are logcally grouped as a multcast group, and each user s called a group member of the multcast group. A smple multcast schedulng scheme could be developed by takng a default transmsson rate for each multcast group and schedulng each group n the roundrobn fashon. The current CDMA 2000 1xEV-DO networks take such an approach wth a fxed rate of 204.8 kbps [4]. Ths scheme s neffcent snce t always uses a fxed rate for data transmsson and does not consder the dverse channel condtons of multple members n a multcast group. Another general approach n selectng the transmsson rate of a multcast group s that all the group members can support ths rate. Note that each multcast group may have multple group members possbly subect to dfferent channel condtons at a tme nstant. Thus, the overall throughput bottleneck would be easly formed by the group member wth the worst channel condton f the base staton (BS ntends to satsfy all group members n a multcast group, whch leads to the underutlzaton for the group members wth good channel condtons. Ths approach s too conservatve and especally neffcent when the group members wth bad channel condtons only account for a small part of the members of a multcast group. To properly select a transmsson rate at the BS, the ntergroup proportonal farness s proposed n [5], n whch the BS schedules the multcast groups based on the proportonal farness polcy and the transmsson rate s selected n such a way that the summaton of logt g k for all multcast groups s maxmzed, where logt g k s the group throughput for multcast group k. Rate optmzaton for threshold based multcast polces s studed n [6]. Aforementoned studes are focused on allevatng the negatve mpact caused by the dverse channel condtons between the BS and multple group members of a multcast group. They do not explot any potental advantages provded by the channel dversty n multcast scenaro. On the other hand, cooperatve communcaton s a promsng technology that can greatly mproves the system performance by explorng the broadcastng nature of wreless channels and cooperaton among multple users. Cooperatve communcaton used for uncast transmssons has been extensvely studed [7], [8], [9]. However, lttle work apples cooperatve communcaton technque for multcast transmssons. In ths paper, we propose a novel cooperatve multcast schedulng scheme that explots the spatal dversty gan across multple users n a network by usng a two-phase cooperatve transmsson. By takng the advantage of the channel dversty of multple group members and cooperatve communcaton, the proposed scheme can sgnfcantly mprove the transmsson rate and mantan the relablty of the downlnk transmsson for SSs wth bad 978-1-4244-2075-9/08/$25.00 2008 IEEE
channel condtons, thus resultng n sgnfcant throughput enhancements for both multcast group and ndvdual group member. Dfferent from prevous work on cooperatve uncast communcatons, the proposed cooperatve multcast schedulng focuses on network performance study based on dfferent network scenaros. Frst, the partner(s or cooperator(s n uncast cooperatve transmsson are usually fxed, e.g., preplaced relay statons, for protocol desgn and mplementaton smplcty. In multcast scenaro wth all users n a multcast group requestng the same data, bascally any user wth good channel condtons can forward the receved data to the remanng users n the same group; and thus the cooperatve transmtters are varable. Second, most prevous studes n uncast cooperatve transmsson focus on the performance study n the physcal layer (PHY, n terms of outage probablty, bt error rate (BER, and optmal power allocaton, etc. In addton, dfferent from uncast transmssons, multcast servces are nherently unrelable (due to no acknowledgement, and we need to carefully determne crtcal parameters for mult-user cooperaton to assure hgh throughput for all users. The remander of the paper s organzed as follows. The cooperatve multcast schedulng scheme s proposed n Secton II. The performance of the proposed scheme s nvestgated n Secton III. Smulaton results are presented n Secton IV to demonstrate the effcency of the proposed scheme, followed by concludng remarks n Secton V. II. THE COOPERATIVE MULTICAST SCHEDULING (CMS SCHEME We consder an IEEE 802.16 network composed of one BS and multple subscrber statons (SSs whch are classfed nto several multcast groups accordng to ther subscrbed servces. For smplcty, we refer to multcast groups and the SSs consstng of a multcast group as MGroups and group members, respectvely. The Raylegh flat fadng channel s assumed throughout. Fg. 1 llustrates the meda access control (MAC structure and the prncple of the proposed cooperatve multcast schedulng. At the MAC layer, the tme doman s dvded nto MAC frames of equal duraton. Each MAC frame conssts of a downlnk sub-frame (DL subframe followed by an uplnk sub-frame (UL subframe. Each DL subframe s composed of a frame header and several downlnk bursts. An MGroup s selected at each frame. For nstance, MGroup s selected at the frame n and the BS assgns downlnk burst (denoted as TS to t. In our proposed scheme, the tme nterval of TS s dvded nto two phases, depcted n Fg. 1(a. In Phase I, BS multcasts data, at a hgh data rate of R1, to the group members of MGroup, amng to guarantee that a certan percentage of SSs n MGroup can successfully decode the data, depcted n Fg. 1(b. After Phase I of tme duraton T 1, some SSs n MGroup can successfully receve the transmtted data (denoted them as SSs wth good channel condtons whle others may only be able to decode part of the data (denoted them as SSs wth bad channel condtons. In Phase Frame n-1 Frame n Frame n+1 Frame n+1 DL subframe UL subframe Frame TS 1 TS Header Phase-1 (R 1 Phase-2 (R 2 BS (b BS does the transmsson. SSs wth Good channel condtons. Fg. 1. (a BS (c SS does the transmsson. SSs wth Bad channel condtons. The proposed multcast schedulng scheme II of tme duraton T 2, all SSs wth good channel condtons cooperatvely transmt the receved data smultaneously at the rate of R 2 and satsfes R 1 T 1 = R 2 T 2 to assure all members n the MGroup can receve the same data, as shown n Fg. 1(c. It s worth notng that R 1 and R 2 are much hgher than the default conservatve rate n the worst case scenaro and the two-phase hgh rate transmsson can outperform one phase low rate transmsson [10]. By takng the advantage of the spatal dversty gan of wreless channels n Phase II, the cooperatve multcast schedulng scheme can sgnfcantly mprove acheved throughput of each MGroup. How to select an MGroup durng each frame and then how to select the rate of R 1 and R 2 wll be elaborated further as follows. A. Multcast Group Selecton Dfferent MGroups have dfferent sets of group members dstrbuted at dfferent locatons. Generally, dfferent group members have dfferent long-term channel condtons whch depend on ther geographcal envronments and ther dstance from the BS. On the other hand, due to fast fadng, dfferent group members may experence dfferent nstantaneous channel condtons at each frame, even f they have smlar longterm channel condtons. In the multcast schedulng scheme, to explot the mult-group channel dversty gan, the selecton of an MGroup should consder the channel condtons on the group bass, rather than a sngle group member bass. If an MGroup s selected based on the best channel condton among all members n an MGroup, gnorng the channel condtons of the remanng group members, the acheved group throughput may not be hgh f most of the other group members are experencng bad channel condtons. If a scheme selects an MGroup based on the sum of channel condtons of all group members, t may lead to serous farness problem because MGroups whch are close to the BS and have good channel condtons are more lkely to be scheduled and thus domnate the bandwdth consumpton. Consderng the farness performance whle explotng the mult-group channel dversty, we proposed a crteron of selectng an MGroup based on the
normalzed relatve channel condton, whch s gven as, = arg maxx (1 G X = γ /γ (2 G where X represents the normalzed relatve channel condton of MGroup, G represents the set of all group members n MGroup, G s the total number of group members n MGroup, γ and γ denote the average channel condton and the nstantaneous channel condton between the -th group member n MGroup, SS, and the BS, respectvely. Based on (1, BS selects the MGroup, whch has the maxmum value of the normalzed relatve channel condton, to be served n each DL sub-frame. In summary, by consderng the dfferent channel condtons of multple MGroups, the proposed scheme explots the mult-group channel dversty for achevng a hgh network throughput. Meanwhle, by averagng out the long-term channel condtons and normalzng the total number of MGroup members, the proposed scheme can obtan good farness performance. B. Cooperatve Transmsson After an MGroup s selected for servce, the next step s to effcently multcast data to all group members n the selected MGroup. If the transmsson rate s determned based on good channel condtons, the group members wth bad channel condtons cannot successfully decode the receved data. On the contrary, f the transmsson rate s determned based on the bad channel condtons, the wreless resources would suffer from underutlzaton because the group members n good channel condtons use a conservatve low rate for data transmssons. Ths dlemma s manly caused by the dverse channel condtons of group members n the same MGroup. To explot the dversty gan of the wreless channels, a twophase cooperatve transmsson scheme s used to effcently multcast data to all group members. Durng the Phase I, the BS multcasts data to all group members at a hgh transmsson rate determned by the channel condtons of a certan porton of the group members. That s, the rate s chosen to guarantee the relable transmssons of the group members n good channel condtons. Therefore, the transmsson rate s much hgher than the conservatve rate consderng the worst channel condton n the MGroup. However, due to the hgh transmsson rate n Phase I, the remanng group members that are n relatvely poor channel condtons may not be able to successfully decode all the transmtted data n Phase I and have to use Phase II cooperatve communcatons for achevng the relable transmsson. Denote S g the set of group members that can successfully receve the data n Phase I and S b the remanng members n the MGroup. In Phase II, all members n S g cooperatvely transmt the same data to those members n S b. In ths way, group members located n dfferent locatons form a vrtual MIMO system, n whch group members n S g are transmtters and those n S b are recevers. Although the channel condton between the BS and group members n S b are relatvely poor durng ths frame, t s very lkely that there are some group members n S g close to the S b members and have good channel condtons. By explotng the spatal dversty gan of wreless channels n Phase II, the transmsson rate for relable multcast transmsson can be sgnfcantly mproved. Note that n Phase II, the downlnk transmsson to any one of the members n S b s a vrtual multple-nputsngle-output (MISO system because multple members n S g are transmttng data to one recever. The sgnal power receved at a group member n S b s the sum of sgnal powers from all cooperatve transmtters. Thus, a group member n S b s able to successfully receve the data n Phase II even at a very hgh data rate. In the proposed scheme, the group members n bad channel condtons stll have a hgh probablty to successfully receve the data when both the frst and the second phases use hgh transmsson rates. By usng approprate parameters for two-phase cooperatve transmssons,.e., the transmsson rates n both Phases I and II, the proposed scheme can acheve hgh throughput for each group member, consequently a hgh group throughput and network throughput. Furthermore, the propose scheme can satsfy the relable transmsson for the members wth bad channel condtons as well. The selecton of transmsson rate n Phases I and II (.e., R 1 and R 2 s crtcal to the system performance. For easy mplementaton, n the proposed scheme, R 1 and R 2 are determned based on the long-term channel condtons of all group members n MGroup and the coverage rato, C, whch s defned as the percentage of group members that can support R 1. For nstance, C = 50% means that the BS transmts at arateofr 1 such that on average half of the group members of the MGroup can receve the data successfully, and R 2 s set n such a way that the remanng group members n MGroup can successfully receve the data n Phase II. Note that R 1 and R 2 are decded based on the long-term channel condtons of group members, nstead of the nstantaneous channel condtons, so that BS does not need to reconfgure the transmsson rates frequently. III. PERFORMANCE ANALYSIS In ths Secton, an analytcal model s developed to nvestgate the performance of the proposed schedulng scheme n terms of steady-state servce probablty of each MGroup and acheved throughput of each group member. The notatons used n the rest of the paper are lsted n TABLE I. A. Steady-State Servce Probablty for MGroup Steady-state servce probablty s defned as the probablty that an MGroup s selected to be served at an arbtrary frame when the system s stable. Based on (1, the MGroup wth the best average channel condton n a frame s selected to be served. We defne a random varable Y = γ /γ n, then X can be expressed as X =. For a Raylegh fadng Y G channel, γ follows exponental dstrbuton wth mean γ. Therefore, Y also follows the exponental dstrbuton [11], and ts probablty dstrbuton functon s gven by
M G G g G b n R1 R2 X SS γ γ ς C TABLE I TABLE OF NOTATIONS The total number of MGroups The set of all members n MGroup The set of members that can successfully receve data n Phase I The set of members that fals to receve data n Phase I The total number of group members n the MGroup G The transmsson rate of the BS n Phase I for MGroup The transmsson rate n Phase II for MGroup The normalzed average channel condton of MGroup The -th group member n MGroup The average SNR of the channel between the SS to the BS The nstantaneous SNR for the channel between SS to the BS The receved SNR for the SS n Phase II Coverage rato used to decde the transmsson rate n Phase I φ(y = y =n e ny, (3 where n s the total number of group members n MGroup. Accordng to the relatonshp between X and Y, X follows gamma dstrbuton, whch s gven by X Gammar(n, 1 n Thus, the steady state servce probablty for MGroup s obtaned as π = Pr[X = max(x 1,X 2,..., X M ] M = f (X = x( H (X = x dx x=0 =1, where the functon f (. denotes the probablty dstrbuton functon (p.d.f. of X, and H (. denotes the cumulatve dstrbuton functon (CDF of X. B. Throughput For the Raylegh flat fadng channel, gven a receved sgnalto-nose rato of γ, the achevable data rate wth a neglgble error probablty s log 2 (1 + γ for unt bandwdth [12]. Therefore, gven that the transmsson rates n Phases I and II for MGroup are R1 and R2, respectvely, the probablty that an arbtrary group member n MGroup, SS, can successfully receve the data n Phase I s gven by (4 Pr[γ 2R 1 1] = e (2 R 1 1/γ (5 Even f SS cannot successfully receve the data n Phase I, t s stll possble to receve data n Phase II, and whether or not to successfully receve the data depends on ts receved sgnal-to-nose rato (SNR n Phase II. Wth cooperatve communcaton, the receved SNR for SS s dependent of the number of cooperatve transmtters (.e., the number of members n G g and the receved power from each cooperatve transmtter. Let ς be the receved SNR of SS n Phase II, TABLE II SIMULATION PARAMETERS Transmsson power of BS s 41.8 dbm Transmsson power of SS s 30 dbm DL/UL sub-frame duraton 0.5 ms/0.5 ms OFDM symbol duraton 23.8µs Bandwdth 10 MHz Nose fgure 7 db Pass loss exponent 4.375 Close-n Reference dstance 100 m Frequency band 3.5GHz Number of MGroups 10 Coverage rato C 50% the probablty that SS can successfully receve the data n Phase II s gven by Pr[ς 2R 2 1] = 1 G(ς =2R 2 1 (6 where G(. functon s the CDF of ς. The throughput acheved by the group member SS s gven by Th = π [R1 Pr[γ > 2R 1 1] +R2 Pr[γ 2R 1 1] Pr[ς 2R 2 1]]. (7 The network throughput s gven by M n Th = Th. (8 =1 =1 IV. SIMULATION RESULTS In ths secton, smulatons are conducted to demonstrate the effcency and effectveness of the proposed scheme. We smulate an IEEE 802.16 WrelessMAN-OFDM network composed of one BS and 50 SSs. SSs are randomly dstrbuted n the coverage area of the BS, whch s a crcle wth a radus of 8 km. The group members n each MGroup s randomly selected from these 50 SSs. Other smulaton parameters are lsted n TABLE II. We repeat the smulaton 50 tmes wth dfferent random seeds and calculate the average value. To verfy the effcency of the proposed scheme, we compare the acheved throughput of the proposed scheme (denoted as CMS wth that of the scheme specfed n 3GPP (denoted as Conserve, where the transmsson rate of BS selects the sendng rate n a conservatve way such that all group members of the selected MGroup can support ths rate. Fg. 2 shows the steady-state servce probablty of each MGroup for the proposed scheme. It s observed that each MGroup obtans almost the same servce probablty, whch llustrates that the proposed scheme can acheves good farness performance n terms of channel access. In addton, the analytcal results match the smulaton results very well, whch verfes the accuracy of the proposed analytcal model. The acheved throughput of each group member for an MGroup s shown n Fg. 3. It s observed that CMS outperforms Conserve n terms of the throughput for all group members by explotng the mult-group channel dversty. Some
0.18 CMS (Sm CMS (Ana 1 0.9 Steady-state servce probablty 0.16 0.14 0.12 0.1 0.08 0.06 0.04 Normalzed network throughput 0.8 0.7 0.6 0.5 0.4 CMS Conserve 0.02 0.3 0 1 2 3 4 5 6 7 8 9 10 Index of MGroups 0.1 0.3 0.4 0.5 0.6 0.7 0.8 0.9 C Fg. 2. Steady-state servce probablty of each MGroup Fg. 4. Network throughput for dfferent parameter C Normalzed throughput 0.4 0.35 0.3 5 Conserv CMS 2 4 6 8 10 12 14 16 18 20 Index of group members Fg. 3. Normalzed throughput of each group member fluctuaton s observed wth CMS, whch depends on channel condtons and locaton dstrbuton of group members. Some solated and faraway SSs acheve relatvely small throughput enhancement. wth Conserve, the acheved throughput of all group members are dentcal. Snce Conserve selects the transmsson rate based on the worst case channel condton of group members, all group member acheves the same, but low throughput even f some of them are n good channel condtons. The mpact of the coverage rato C on the network throughput s shown n Fg. 4. It s observed that the maxmum network throughput s acheve wth C =0.55. In addton, a medum value of parameter C s better than a relatvely large or relatve small value. V. CONCLUSIONS We have proposed a cooperatve multcast schedulng scheme for achevng hgh throughput n IEEE 802.16 networks. By usng two-phase cooperatve transmssons to explot the spatal dversty gans n the multcast scenaro, the proposed schedulng scheme can sgnfcantly mprove the throughput not only for all MGroups, but also for each group member. In addton, the proposed MGroup selecton can provde good farness performance n terms of channel access. Extensve smulatons have been conducted to valdate the effcency of the proposed scheme and the accuracy of the analytcal model. The proposed scheme can also be extended to other wreless networks n general. REFERENCES [1] 3GPP TR 25.992 V7.0.0, Multmeda Braodcast/Multcast Servce (MBMS; UTRAN/GERAN requrement, ETSI, 2007, http : //www.3gpp.org/ftp/specs/archve/25 seres/25.992/. [2] J. She, F. Hou, P.-H. Ho, and L.-L. Xe, IPTV over WMAX: Key Success Factors, Challenges, and Solutons, IEEE Communcatons Magazne, vol. 45, ssue 8, pp. 87-93, Aug. 2007. [3] IEEE Standard 802.16-2004, IEEE Standard for Local and Metropoltan Area Networks-Par 16:Ar Interface for Fxed Broadband Wreless Access Systems, 2004. [4] P. Agashe, R. Rezafar, and P. Bender, CDMA2000 Hgh Rate Broadcast Packet Data Ar Interface Desgn, IEEE Communcatons Magazne, pp. 83-89, Feb. 2004. [5] H. Won, H. Ca, D.Y. Eun, K. Guo, et. al. Multcast Schedulng n Cellular Data Networks, Proc. INFOCOM, pp. 1172-1180, 6-12, May 2007. [6] W. Ge, J. Zhang, and X. Shen, A Cross-Layer Desgn Approach to Multcast n Wreless Networks, IEEE Trans. on Wreless Communcatons, vol. 6, no. 3, pp. 1063-1071, March 2007. [7] A. Coso, S. Savazz, U. Spagnoln, and C. Ibars, Vrtual MIMO Channels n Cooperatve Mult-hop Wreless Sensor Networks, Proc. Informaton Scences and Systems, pp. 75-80, Mar. 2006. [8] M. Danat, X. Shen, and K. Nak, Cooperatve Far Schedulng for the Downlnk of CDMA Cellular Networks, IEEE Trans. on Vehcular Techno., vol. 56, no. 4, pp. 1749-1760, July 2007. [9] V. Mahnthan, L. Ca, J.W. Mark, and X. Shen, Maxmzng Cooperatve Dversty Energy Gan for Wreless Networks, IEEE Trans. on Wreless Commun., vol.6, no. 7, pp. 2530-2539, July 2007. [10] H. Ocha, Varable-Rate Two-Phase Collaboratve Communcaton Protocols for Wreless Networks, IEEE Trans. on Informaton Theory, vol. 52, no. 9, pp. 4299-4313, Sep. 2006. [11] P.J. Pahl and R. Damrath, Mathematcal Foundatons of Computatonal Engneerng: A Handbook, New York, Sprnger, 2001. [12] M.-S. Aloun and A.J. Goldsmth, Capacty of Raylegh Fadng Channels Under Dfferent Adaptve Transmsson and Dversty-Combnng Technques, IEEE Trans. on Vehcular Techno.,vol. 48, no. 4, pp. 1165-1181, July 1999.