Optimal Allocation of Packet-Level and Byte-Level FEC in Video Multicasting over Wired and Wireless Networks

Transcription

1 1 Optimal Allocatio of Pacet-Level ad Byte-Level FEC i Video Multicastig over Wired ad Wireless Networs T.-W. Agus Lee S.-H. Gary Cha Qia Zhag We-Wu Zhu Ya-Qi Zhag Departmet of Computer Sciece Microsoft Research, Chia Hog Kog Uiversity of Sciece ad Techology 5F, Beijig Sigma Ceter, No. 49, ZhiChu Road Clear Water Bay, Kowloo Haidia District, Beijig , Hog Kog P.R. Chia Abstract Multicast is a efficiet techique to deliver video cotet over a etwor. I this paper, we cosider such a multicast system to serve both wireless ad wirelie users whe there is error over the wired etwor ad wireless hop. Sice pacets are liely to be dropped i the wired etwors while bit errors are more liely over the wireless hop, a combiatio of both pacet-level ad byte-level FEC is required to recover these errors. Give the estimated error ad badwidth characteristics reported by ed users, the server eeds to optimally allocate the pacet-level ad byte-level FEC to achieve maximum video quality. We study two schemes pertaiig to whether or ot the wireless gateway is able to trascode the video pacets from the wired etworefore forwardig it to the wireless users. We first develop a model to aalyze the system; ad theropose a efficiet algorithm for the FEC computatio. We fially compare the schemes i terms of the optimal parameters used i the FEC, ad the video quality achieved. Keywords Video Multicastig, Optimal FEC Allocatio, error recovery, wireless Iteret I. INTRODUCTION Multicast is a efficiet techique to deliver video to its ed users. I such a system, the video files stored or captured i a server are multicast to its cliets distributed i a etwor. Data is usually multicast i a TCP-friedly maer, by which we mea that the streamig rate is ot more tha the available badwidth i the etwor [1], [2]. The ed cliet may be wired or wireless. 1 I the case of wireless etwor, the base statio is geerally coected to a gateway. I its simplest form, the gateway simply forwards whatever pacets it receives to the wireless cliet without ay re-pacetizatio or fragmetatio. A more sophisticated gateway, o the other had, ca tae ito cosideratio of the wireless characteristics ad repacetizes the pacets it receives from the wired ifrastructure, adds or removes some iformatio i them e.g., error redudacy codes before forwardig them to the ed cliets. This techique is ow as, as data pacets are re-ecoded i the process. The gateway is beeficial sice the pacet error characteristics are differet i the wired ad wireless etwors. I wired etwors such as the Iteret, pacets are lost maily due to cogestio at the routers, while i the wireless hop, pacets are ofte lost due to radom bit error caused by fadig or multipath effect [3]. To recover pacet loss, feedbac recovery or forward error correctio code ca be used [4]. I geeral, feedbac recovery does ot wor very well over large scale etwors with real-time guaratee. FEC such as the Reed-Solomo RS code, o the other had, is more appropriate for real-time commuicatios by itroducig some redudacy [5]. It cosists of arragig the data ad redudacy bits i such a way that eve if a partial fractio of the bits are received, the origial data may still be recovered. A FEC scheme adaptig to the etwor error coditios is very efficiet to maitai video quality. I this paper, we will maily cocered with FEC to achieve maximum video quality. FEC strategies would be differet over the wireless ad wired etwors, due to their differet error characteristics. I the wired etwor, some pacet-level FEC should be used, i which redudat or This wor was supported, iart, by the Hog Kog Telecom Istitute of Iformatio Techology 97/98.EG01, ad Sio Software Research Istitute at the Hog Kog Uiversity of Sciece ad TechologySSRI00/01.EG04. 1 I this paper, we distiguish a user from a cliet i that a user is the perso requestig a video, while a cliet is the statio/machie the user uses. parity pacets are added so that pacet loss ca be recovered [3], [4], o the other had, byte-level FEC should be used over the wireless hop, i which redudat or parity bytes withi a pacet ca be added to recover bit error [3]. We cosider that the cliets periodically feedbac to the source its estimated ed-to-ed badwidth, pacet loss ad bit error rate betwee itself ad the source How to estimate the ed-to-ed badwidth ad pacet drop rates accurately is beyod the scope of this paper. Iterested readers are referred to [6] ad refereces therei. Give these feedbacs, it is therefore of particular iterest to address the followig issue: uder the multicast eviromet ad give the heterogeeous error ad badwidth characteristics of its ed users, how should the pacet-level ad byte-level FEC be allocated for a sigle video stream i order to maximize the overall video quality, i both cases of simple ad gateways? We primarily cocer the optimizatio for a sigle stream i this paper, sice there are already may rich issues to be cosidered pertaiig to FEC allocatio ad trasodig gateways. Our approach ad discussio here readily exted to ad would be useful for the layered multicast case, i which the video is multicast via multiple layers, ad each layer has differet FEC capability. Traditioally video quality is measured by distortio give by PSNR [7]. It has bee widely observed that such PSNR is proportioal to the video goodput defied by useful data bits per secod received by the ed cliets after FEC, give that the residual pacet error rate is below a certai low value 3% [8]. Therefore, maximize video quality i PSNR is equivalet to maximize the overall goodput of the system, subjected to a certai low loss costrait. Our cotributios i this paper are 3-fold: i We have studied video multicast to wired ad wireless users; ii We have ivestigated optimal allocatio of pacet-level ad byte-level FEC for video delivery; ad iii We have developed a model ad preseted a aalysis of gateway at the wireless hop. Our results show that the scheme with gateways outperforms the scheme with o gateways by a small margi, for wide rages of pacet drop rate. We briefly preset some previous wors as follows. I [1], [2], Zhag et al. discussed rate adaptatio approaches for video delivery. It cosists of badwidth estimatio ad adaptig the trasmissio rate of the videos. Multiple video streams are carefully cotrolled by a quadratic rate-distortio fuctio i order to uicast it i a TCPfriedly maer. Our wor differs from theses by cosiderig video streamig i a multicast eviromet. Also, we cosider the applicatio of FEC o the video stream ad the optimal partitio of the available badwidth for video data ad parity bits. Recoverig dropped pacet by retrasmissio ad delayig the playout time at the cliet have bee discussed i [9], [10]. However, these wors are related to wired eviromet ad have ot cosidered the wireless media ad the optimal allocatio of FEC codes. Other recovery schemes pertaiig to limited retrasmissio ad FEC the socalled hybrid ARQ- FEC scheme, have bee discussed i [11], [12]. Research has also bee doe sedig delayed versio of parity across differet multicast groups, ad cliets ca subscribe to differet groups, accordig to their loss profiles ad desired level of protectio [13]. However, all of them

2 p pacets b bytes bytes pacets - p pacets p pacets pacets - p pacets Geerate Parity Video Data Pacet-level Parity Byte-level Parity Fig. 1. Pacet-level ad byte-level FEC scheme for o- gateway. Trascodig at Gateway after recovery from pacet loss b bytes bytes... Video Data Pacet-level Parity Byte-level Parity are discussed uder the cotext of a sigle medium oly, ad have ot discussed mixed media wired ad wireless etwors ad how pacet-level ad byte-level FEC ca be optimally combied i such case. Furthermore, there has ot bee wor examiig the desig of gateways for FEC. This paper is orgaized as follows. We first preset the schemes depedig o whether the gateway trascodes the pacets or ot, ad the goodput aalysis i Sect. II. I Sect. III, we preset some illustrative examples, ad demostrate the effectiveess of the schemes. We coclude i Sect. IV. II. SCHEME DESCRIPTION AND ANALYSIS Sice every cliet has to receive the video with good quality, the badwidth allocated to the video stream icludig FEC ecodig should be equal to the miimum ed-to-ed badwidth. Thus, the oly cocer here is how the error cotrol should be applied to serve both wireless ad wired cliets so that its overall quality is optimized. As oted before, the quality is measured by the aggregate goodput i the system, or equivaletly, average goodput of the cliet. I this sectio, we first describe pacet-level ad byte-level FEC schemes i Sect. II-A. I Sect. II-B, we aalyze ad optimize video quality i terms of system goodput for o- ad gateway, give cliet pacet loss ad bit error rate. A. FEC Scheme Descriptios We propose mixed pacet-level ad byte-level FEC to protect the video stream. We study the scheme with ad without a gateway. A.1 No- gateway With o- gateway, both pacet-level ad byte-level FEC ecodigs have to be doe at the video server, ad error correctio are oly doe at the ed cliets. At the server, the compressed stream is first ecoded with byte-level FEC followed by pacet-level FEC. The decodig part is the reverse of the ecodig process. Note that with this system, the byte-level FEC does ot really help those wired cliets where pacet drops occur i improvig their error resiliece capability. We show i Fig. 1 how to geerate the two levels of FEC based o RS code. For the byte-level FEC, the ecoder processes i symbols, where each symbol cosists of m = 8 bits. Give a pacet of size b bytes, 1 bytes of source data is paced with b parity bytes, where = b, b 2,.... This is the so-called RS b, code, which is able to correct up to t b symbol errors i a pacet, where t b = b /2. The pacet size b is limited by 2 m 1 symbols; therefore, for m = 8, b 255. With every p of these byte-ecoded video pacets, pacet-level FEC is the applied to geerate p parity pacets for a bloc of pacets, where p =, 1,..., 1. This is geerated as follows. The ith byte of each of the p video pacets 1 i b is tae out Fig. 2. Pacet-level ad byte-level FEC scheme for gateway. to geerate b parity bytes. The geerated parity bytes are the redistributed as the ith byte of each of the p parity pacets. Sice all the pacets are sequeced, up to t p = p pacet losses i a bloc ca be corrected. Clearly, as a bloc of pacets has to be ready before pacet-level FEC is doe, the delay of the system icreases with. Therefore, i reality user delay requiremet determies the that ca be used. The server computes the optimal allocatio betwee the video data rate, the pacet-level FEC rate i.e., the umber of pacet-level FEC parity bits per secod ad the byte-level FEC rate i.e., the umber of byte-level FEC parity bits per secod give the feedbacs from the ed cliets. Let G be the multicast group size. The feedbacs for cliet g 1 g G are i terms of the estimated ed-to-ed available badwidth ˆB g ad the pacet drop rate ˆP l,g ˆP l,g may be estimated by the missig sequece umbers of the pacets, 2 ad, for wireless cliets, the bit-error-rate of the wireless hop ê b,g ê b,g may be estimated by usig a two-state Marov process as give i [14]. Give the feedbac iformatio, the server has to first decide the pacet-level ad byte-level FEC rates for the video stream, with its trasmissio rate icludig all the redudat bits is equal to the least ed-to-ed badwidth i the multicast group i.e., R 0 = mi g ˆBg. Let the pacet-level FEC rate be R p ad byte-level FEC rate be R b. Give, p ad b,, R p ad R b are clearly give by p p R p = R 0, 1 ad p b R b = R 0. 2 b The video source rate R s, defied as the data rate excludig all the FEC, is the give by R s = R 0 R p R b = R 0 / b p /. The omeclature used i this paper is listed i Table I. A.2 Trascodig Gateway A gateway trascodes video pacets from pacet-level FEC to byte-level FEC before forwardig the pacets to the wireless cliets. The gateway first recovers ay dropped pacet by the pacetlevel FEC, ad theads the video pacets with byte-level FEC parity. Note that the wired cliets eed to perform pacet-level FEC operatios oly, ad, i cotrast with the o- gateway, bytelevel FEC ecodig is doe at the gateway rather tha the server. We cosider a simple gateway which does ot do ay 2 We defie a pacet as dropped if the pacet is i error durig its trasmissio over the etwor. A dropped pacet is permaetly lost if it caot be recovered after pacet-level FEC.

3 TABLE I NOMENCLATURE USED IN THIS PAPER. G : Size of multicast group umber of cliets ˆB g : Estimated ed-to-ed available badwidth for cliet g bits/s ˆP l,g : Estimated pacet drop rate i the wired etwors for cliet g P l : Average pacet drop rate i the wired etwor ê b,g : Estimated bit error rate over the wireless hop for cliet g e b : Average bit error rate over the wireless hop e s,g : Symbol error rate i the wireless hop for cliet g b, : Pacet size of the byte-level FEC bytes ad bloc size of the pacet-level FEC pacets, respectively, p : Data bytes i a byte-level FEC pacet ad umber of pacets i a pacet-level FEC bloc, respectively : Costrait/Requiremet o ed-to-ed pacet loss rate after error correctio ɛ g : Ed-to-ed pacet loss rate after error correctio for cliet g for o- gateway η g : Ed-to-ed pacet loss rate after error correctio for cliet g for gateway Γ g : Goodput for cliet g bits/s Γ : Total goodput = Γ g bits/s pacet fragmetatio or reassembly. We see that a gateway achieves lower badwidth requiremet tha the o- oe or equivaletly higher video quality give a badwidth costrait by tradig off some system complexity. We show the detail of the ecodig process i Fig. 2. The gateway first recovers the p data pacets out of the FEC bloc each of bytes, ad the trascodes the pacets to b bytes by paddig them with some byte-level FEC. Give the trasmissio rate of R 0 bits/s, the pacet-level ad byte-level FEC rates are clearly give by the same expressios as of Eqs. 1 ad 2, respectively. The source rate is, however, give by R s = R 0 p /. B. Quality Optimizatio I this subsectio, we aalyze the system with o- ad gateways ad cosider how the video quality ca be maximized over all the cliets i the system. As metioed before, we cosider miimizig the sum of PSNR over all the cliets. For the error rate of iterest, this is equivalet to maximizig the aggregate goodput Γ bits/s, defied as the useful data bits delivered per secod over all cliets after error correctio. Further let Γ g be the goodput of the gth cliet. Therefore, we study the followig byte-level ad pacet-level FEC allocatioroblem: Give b ad, fid the optimal, p, ad i order to maximize G Γ = Γ g 3 g=1 such that the ed-to-ed pacet loss rate after error correctio is o more tha a certai value say, over all cliets. Here, we cosider the sum of the idividual goodput, i.e., all the cliets i the system have the same priority or importace. If it is ot that case, we eed to assig some weight to each Γ g ad thereof each PSNR. This extesio is straightforward ad would ot be pursued further here. B.1 Optimizatio for o- gateway Let s cosider a particular cliet g ad hece the subscript g i some of our equatios ad obtai its goodput give ˆP l,g ad ê b,g. I the wireless hop, a symbol is cosidered i error if ay of the m bits i the symbol are trasmitted i error. Clearly, give bit error rate ê b,g i the wireless chael, the symbol error rate is e s,g = 1 1 ê b,g m. Sice the RS b, code corrects up to t b symbol errors, the probability that a radom pacet caot be recovered by byte-level FEC is give by α g = b j=t b +1 b j e s,g j 1 e s,g b j. 4 Note that for the wired cliets, α g = 0 as ê b,g = 0 by defiitio. A pacet is dropped if it is dropped i the wired etwors with rate ˆP l,g, or if it is i urecoverable error with probability α g over the wireless hop. Sice the two evets are idepedet, the ed-toed pacet drop rate from the source to the cliet is give by β g = 1 1 ˆP l,g 1 α g. Note that the dropped pacets may be recovered by the pacet-level FEC see Fig. 1. Sice up to t p = p dropped pacets i the same bloc ca be recovered by pacet-level FEC. By cosiderig the umber of pacet drops i a FEC bloc, the probability that a radom pacet is permaetly lost i.e., the ed-to-ed pacet loss rate after error correctio is give by ɛ g = p =t p +1 The goodput of the cliet g is hece give by p β g 1 β g. 5 b p Γ g = R 0 1 ɛ g. 6 b The allocatioroblem is a two dimesioal search o p ad, which is of complexity OG b ad is ot efficiet. Validated by extesive rus, we foud that pacet-level FEC optimizatio ca be doe idepedetly with that of byte-level FEC without affectig the results much less tha 1%. Therefore, we ca greatly reduce the complexity to OG + b by meas of the followig two-step procedure: 1. Pacet-level FEC optimizatio For all cliets, let P l = max g ˆPl,g. If P l, STOP ad proceed to the ext step The pacet drop rate is so low that p =. Otherwise, for all the cliets with ˆP l,g >, search for the largest p < for miimum overheads such that ɛ g i accordace to Eq. 5 of all these cliets are o more tha. This is the p required. 2. Byte-level FEC optimizatio Give p, fid the largest < b such that ɛ g for all the wireless cliets are o more tha. This is the b required. B.2 Optimizatio for gateway Cosider a cliet g. The probability that a radom pacet is permaetly lost over the wired etwor is clearly give by γ g = p =t p +1 p ˆP l,g 1 ˆP l,g. 7 If it is a wireless cliet, the pacets corrected after pacet-level FEC are trasmitted over the wireless hop. The probability that these pacets caot be recovered due to wireless error has already bee obtaied

4 TABLE II CLIENTS PROFILE USED FOR TRANSCODING AND NON-TRANSCODING GATEWAYS. ˆP l,g % ê b,g 10 4 cliet cliet cliet cliet cliet cliet cliet cliet cliet cliet * bytes o / o 245 Fig. 4. b versus give p for ad o- gateways e b = 10 4, P l = 2%, b =255 bytes, =40 pacets. p * pacets o Fig. 3. p versus εo give b for ad o- gateways e b = 10 4, P l = 2%, b = 255 bytes, = 40 pacets. Γ * bits per secod as α g i Eq. 4 agai, for the wired user, α g = 0. Therefore, the ed-to-ed pacet loss rate after error correctio is give by by the idepedece of error rates i the wired ad wireless etwors ad hece the goodput of the cliet is η g = 1 1 γ g1 α g, 8 Γ g = R 0 p 1 η g. 9 As i the o- case, we agai observe that the pacetlevel FEC ca be doe idepedetly of the byte-level FEC for the case. The optimizatiorocedure is hece the same as that of the o- case, except that ɛ g is replaced by η g i Eq. 8. III. ILLUSTRATIVE NUMERICAL EXAMPLES AND RESULTS I this sectio, we compare the performace of ad o- gateways. We cosider a baselie system of G = 10 cliets, with half of them wireless cliets. We show i Table II ˆP l,g ad ê b,g of each cliet, which are geerated by assumig that they are uiformly distributed with mea P l = 2% ad e b = 10 4, respectively. Note that cliets 1 to 5 are wireless cliets, while the remaiig are wired. The other baselie parameters are = 1%, b = 255, = 40, ad R 0 = 100 bits/s b/s. Optimal FEC allocatio will first be performed give these parameters. The i our sesitivity aalysis, we vary the other parameters, oe at a time. We show i Fig. 3 p versus for the ad o cases. Clearly, both cases have the same optimal p due 800 Fig. 5. Γ versus for ad o- gateways e b = 10 4, P l = 2%, b = 255 bytes, = 40 pacets. to the same optimizatio step i. p icreases with i a stepwise maer due to the costrait o itegral value. Note that p is already very close to, idicatig that little pacet-level FEC is ecessary to achieve a low ed-to-ed pacet loss. The pacet-level FEC is so effective that eve though most of the ˆP l,g s are greater tha 2%, oly a few overhead pacets 2 i this case are eeded to brig to as low as 1%. No overhead is ecessary whe max g ˆPl,g as all cliets has ˆP l,g <. We ext show i Fig. 4 the correspodig b versus for ad o- gateways. Both cases share almost the same b. As compared with p, b is quite isesitive to ; it icreases relatively very slowly. Therefore, as chages, p is a more importat parameter to adjust. Note that for ê b,g = 10 4, a radom pacet without ay byte-level FEC is i error occurs with probability of 1 1 ê b,g m b = Eve with this pacet loss rate, oly a few overhead bytes about 4 6 i our plot is eough to brig this error rate dow to a low level give by. This agai idicates the efficiecy of byte-level FEC. The dips i the figure correspods to the rises i Fig. 3. This is because oce p is icreased, the pacet-level error correctio capability decreases ad hece a lower b ad thereof a stroger byte-level correctio capability is eeded to compesate. As icreases, the b jumps bac up as the system ca tolerate more ed-to-ed pacet loss.

5 p * pacets / o p l Fig. 6. p versus P l for ad o- gateways e b = 10 4, = 1%, b =255 bytes, =40 pacets. * bytes o e b Fig. 7. b versus e b for ad o- gateways P l = 2%, = 1%, b = 255 bytes, = 40 pacets. We show i Fig. 5 the correspodig optimal goodput Γ i.e., with p ad b versus for the ad o- cases. Though the goodput for the case is higher, there is o much differece betwee them oly about 2% differece here. This is expected because, from Eqs. 9 ad 6, the ratio of the o- ad goodputs for cliet g is give by b o-/ b 1 ɛ g /1 η g b / b. From Fig. 4 we have already see that b for the o- gateway is very close to b, ad hece the differece is small. As icreases, Γ i geeral first icreases ad the decreases the decrease is show for the case. This is due to the followig. Γ is affected by two factors: i the ed-to-ed pacet loss rate ɛ g ad η g Γ decreases with them, ad ii p ad b Γ icreases with them. From the figure, we see that whe is small, the effect of p ad b domiate, while whe is higher, the error rate domiate ad Γ decreases. For the cases of iterest i.e., 5%, Γ icreases with. I Fig. 6, we show how p varies with P l. Clearly, p decreases with P l as more error protectio is ecessary. The pacet-level FEC is quite effective, as oly a few parity pacets give by p are able to brig a high P l say, 7 8% to a low value 1%. I Fig. 7, we show b versus e b for ad o- cases. Clearly, there is o much differece betwee these cases. I geeral, b decreases with e b because more parity bytes are eeded. As e b icreases, b remais quite flat at the begiig ad the sharply decreases. This idicates that whe the bit error is high, may more parity bytes are eeded to achieve a certai error rate after FEC. From the figure, we also see that whe e b is greater tha a certai value about 5% i this case, the bit error rate is too high that byte level FEC is o loger effective to brig errors i the wireless hop dow to. Thus, the system ca oly tolerate wireless chael with bit error rate from 10 6 to IV. CONCLUSIONS I this paper, we have studied a video multicast system over wired ad wireless etwors with cliet feedbac. The mai challege is to optimize the overall video quality by meas of layer FEC allocatios for the set of cliets give their heterogeous badwidth ad error characteristics, subject to a certai overall loss rate requiremet. Furthermore, sice there may be a gateway which trascodes from pacet-level FEC to byte-level FEC betwee the wireless cliets ad the wired etwor, we have studied the value of such a gateway. We have aalyzed the system ad proposed a efficiet allocatioolicy. I order to serve all the cliets, the trasmissio rate of the video stream should be equal to the miimum badwidth of the cliets. The issue of the video trasmissio is hece how to allocate pacet-level ad byte-level FEC so as to maximize video quality i terms of goodput. Istead of a 2-dimesioal search, we have preseted a efficiet algorithm for such optimal FEC allocatio. Our results show that the scheme performs oly slightly better i terms of system goodput tha the o- scheme by about 2%. This is maily due to the efficiecy of FEC ecodig which occupies less tha 20% of the data for the pacet-level FEC, ad less tha 10% for byte-level FEC. This small differece may ot justify the complexity of such a gateway. A gateway which trascodes data i some other ways may be more useful. The schemes ca be exteded to apply i deliverig layered video, which each layer is multicast i separate groups with differet loss costrait. Future wor ca be doe i combiig the optimizatio i both FEC ad badwidth allocatio across the layers. REFERENCES [1] Q. Zhag, Y.-Q. Zhag, ad W. Zhu, Resource allocatio for audio ad video streamig over the Iteret, to appeared i special issue o Multimedia over IP i IEEE Tras. o Multimedia, September [2] Q. Zhag, W. Zhu, ad Y.-Q. Zhag, Networ-adaptive rate cotrol ad uequal loss protectio with TCP-friedly protocol for scalable video over Iteret, to appeared i special issue selected from IEEE ICME 00 o Multimedia Commuicatios Joural of VLSI Sigal Processig - System for Sigal, Image ad Video Techology, [3] J. G. Kim ad M. M. Kruz, Badwidth allocatio i wireless etwors with guarateed pacetloss performace, IEEE/ACM Trasactios o Networig, vol. 8, pp , Jue [4] K. Stuhlmuller, N. Farber, M. Li, ad B. Girod, Aalysis of video trasmissio over lossy chaels, IEEE Joural o Selected Areas i Commuicatios, vol. 18, pp , Jue [5] S. Wicer ad V. Bhargava, Reed-Solomo Codes ad Their Applicatios. IEEE Press, [6] J. Padhye, V. Firoiu, D. Towsley, ad J. Kurose, Modelig TCP throughput: A simple model ad its empirical validatio, i Proccedigs of ACM SIGCOMM 98, pp , October [7] A. Ortego ad K. Ramchadra, Rate-distortio methods for image ad video compressio, IEEE Sigal Processig Magazie, vol. 15, pp , November [8] Y. Wag ad Q.-F. Zhu, Error cotrol ad cocealmet for video commuicatio: a review, Proccedigs of the IEEE, vol. 86, pp , May [9] G. Ramamurthy ad D. Raychaudhuri, Performace of pacet video with combied error recovery ad cocealmet, i Proccedigs of INFOCOM 95, pp , April [10] A. Albaese, J. Blomer, J. Edmods, ad M. L. d M. Suda, Priority ecodig trasmissio, IEEE Trasactios o Ifomratio Theory, vol. 42, pp , November [11] R. Puri, K. Ramchadra, ad A. Ortega, Joit source chael codig with hybrid FEC/ARQ for buffer costraied video trasmissio, i Proccedigs of IEEE Secod Worshop o Multimedia Sigal Processig, pp , [12] S. S. Wag, H. Zheg, ad J. Copelad, A QoS ehaced hybrid SR-ARQ for mobile video commuicatios, i Proccedigs of IEEE Iteratioal Coferece o Commuicatios, pp , [13] P. Chou, A. Mohr, A. Wag, ad S. Mehrotra, FEC ad pseudo-arq for receiver-drive layered multicast of audio ad video, i Proccedigs of Data Compressio Coferece, pp , [14] H. S. Wag ad P.-C. Chag, O verifyig the first-order marovia assumptio for a rayleigh fadig chael model, IEEE Trasactios o Vehicular Techology, vol. 45, pp , May 1996.