Sei sulla pagina 1di 4
H.263+ PACKET VIDEO OVER WIRELESS IP NETWORKS USING RATE-COMPATIBLE PUNCTURED TURBO (RCPT) CODES WITH JOINT SOURCE-CHANNEL CODING ¥. Pei ECSE Department Rensselaer Polytechnic Institute Troy, New York 12180, USA ABSTRACT Te this paper we consider eae video coding and transmission cover packet switched wireless IP networks using RCPT codes In ‘joint soutce and channel coding OSCO) approach. I parc Tat, the performance ofthis JSCC approach employing RCPT cx Ing schemes fr RTP-H.263+ packet video transmission ove siow fading Rican chancel is studied, Ress nccate that an RCPT- USCC spproach is atractive for realtime video applications and dads to increased system capacity together with more acceptable delivered vdeo qualky over time-varying witless networks. Keywords: Video trnsmision, RTP-H 263s, RCPT codes, {oil soure-channe cing, Rela fding, passive err recovery. 4, INTRODUCTION “The ost prominent characteristics ofthe future wireless necvorks ‘whlch diferentes them from fixed wireline metworks the re {qulrement to share a ited spectral bandwith while coping with the mata contesdon and co-channel imerfrence effects res ing froma large numberof eendomly located and paps mesile ‘secs wi diverse service and taffc requirements all tying 10 ‘communicate trough 2 relatively small number of acess points to a fixed wireline backbone network, Farhermor, the wire less communication links themselves an generally be described as time-varying, frequency-selective facing mulpalh channels. The topology, link performance and quality of service (QoS) delivered to user application fn thls communications environment can be characterized as highly me-varying. 'A second distinguishing Feature of wireless mobile networks that bas sigalBicant impact on applicable video compresion techy niques ithe relative ungeliabiy of transmission Hinks compared to thelr fixed witline counterparts. Most existing video comptes sion techniques have been developed under the assumtion of a tellable ranspor or in some cass, storage channel. While his i approximately the case fr high-quality fined wireline networks. is totally inconsistent withthe conditions expected on mast wie- lese mobile networks, Asa result the need exins for video coving and tansmlssion schemes which not only exhibit efficient com pression peformance but provide relatively robust perfomance In the face a ink ers. "As aesul some form of adaptive hybed combination of active forward error correction (FEC) and purely passive err recovery {PER) would appear usefl here although this needs to be applied Judicial to avoid the bandwidth penlies associated with FEC ‘Overheads. Aa important equlement of ary such Scheme Is that (0-7803-7622-6/02/817.00 ©2002 IEEE. 1-541 J. W. Modestino ECE Department University of Miami Coral Gables, FL 33146, USA ‘tt adequately protet the compressed information bitstream, according othe characteristics ofthe ven channel, without ex ‘essve toting of the source coding rte to accommodate the ‘coding overheads. Since an arbitrarily chasen FEC design can lead to a prohibitive amount of overhead for ie-varying ear ‘condition, ajoit source end chanel coding 'SCC) approach for Image o vdeo tarsmission Is necessary (1). Te this paper we wall invesigate the performance of JSCC. schemes or dig video ansmssion over wireless IP netocks Using RCPT codes (2} as 2 dieetextonsion of cur previous re search on JSC. Considering the superior performance of turbo Ceres, some farther performance improvement may be possible ‘ompazed to the use of rate-compalble punctred comotional ARCPC) codes [3]. Our objective In ths paper sto quantitatively sss he elative performance of SCC schemes using RCPT and TRCPC codes as wel as lnvesigating the sroce of JSCC ap- roeches using RCPT codes over wireless channels modeled as an AWGN chanel as well as more general slow-fading Rician chan nels which incorporate fading/nutpah effects 2, RTP.AL263+ PACKET VIDEO OVER WIRELESS NETWORKS In this paper we describe and investigate a video coding and de- livery system for wireless IP networks which consist of a H.2635 ‘video codec RTP-H 263+ packeties, RCPT channel coding and QPSK modilaton a5 asrated in Fig. “To ransit eam H.253 video over wees IP networks, the H.263+ bitstream must fst be packetzed. The protocol of ‘choice for IP-based real-time packat vdeo applications is the el ‘ime transpoet protocol (RTP). A paylcad format for H.263+ video has been defined for se with RIP (RFC 2428) [4]. According to the RTP-H.262+ payload format speciation, the H.263+ en ‘coded bitstream ie packetized and then transmited as RTP packs Tn our experimen, the group of block (COB) mode was se lec forthe 1263+ coder and packetzation was always per formed st COB boundares, Le, each RTP packet contains one ‘or more complete GOBS. Sine every packet bins with a picture ‘ot GOB stat code, the leading 16 zeros are omitod in 2ecordance wth RFC 2429 3. CHANNEL INDUCED LOSS MODELS By means of FEC, some ofthe channel induced bit errs can be cowecid. Depending on the FEC code parameters andthe channel IEEE ICIP 2002 DEE He cre Fig Illustration of vdeo coding and delivery system over wire- les neowo, conditions, there wil be residual bit eros. Generally, over exist ing wired TP networks. UDP is conigsed to dscard any packet wih even «single evar detected in the end packet. including the Reade, although UDP itself might not plement this erro ‘detecting functionality. a the wieess video telephony system deseribed by Cherian and Hanzo [such packets ae also ds- ‘carded without further processing. However, we have shown in 6] ‘the advantage of a transparent transport layer for vdeo transis ‘Sion over nals channels Ta his paper, we wll gain asume the wansprt aye is vans parent to the application lave, Le. packet wth errs sn m= Diy diseaeded in he cansport ayer Inswead, tho aplication layer sould beable o access the receved data although Such dat may Inve one or more it errors Sacha model corresponds to a ane or layer scheme allowing bi errors in the pyoad, Th chan. Induced impaiement 10 Ue video quality Is then ia the form of ‘sida bit ears nthe vide stream. [tthe esponsbsy of he ‘pplication layer to deal with the possibe bit ero. Specially hee we make use of the H.253+ cding scheme where, based on ‘syntax vilations, cata eror paterrs may be detected by the ‘ideo decoder and use ofthe carespondingerored data can be ‘voided by employing PER techniques. In this work, we make use ofthe eror-detecting and recovery Scheme suggesed In Test Meda! 8 (7. The major objeciveof this PER scheme is o detect the severe err patems and prevent the use of such rors which may substantially degrade he veo qual- ly. The remaining undeecied ror patra in the payload which _ate not detected by the H.263+ deccer will result nthe wseofn- ‘Correctly decoded tage data which can cause qualty degradation ofthe reconstructed vdeo 4. PERFORMANCE OF RCPT CODES OVER RICIAN FADING CHANNEL. In this work, a set of RCPT codes ae goneratd using the algo sithm described in [2] employing two memory Mf = 4 consent encoders with generator (1,23/35)aca and puncturing period P's 8. Simlations were conducted to obtain the BER pore ‘mance ofthe resulting RCPT cades over the AWGN and Rican fading chanel with speculr-to diffuse energy ratio ¢* = 7 4B. In, ‘ozder to achieve siatstal reliability, for each SNR, simlacons employed up to 10" tansmited information Bt or wil 1000, Fig. 2, Simulated BER Peformance of RCPT Codes over the Ri clan Fading chanel with ¢? = 7 4B when used with QPSK, crrored blocks were detected. whichever occured fist, For Ri- ‘an fading chanel, we asume perfect channel inerieaving, Le, ‘hanael errors ae randomly dispersed ‘We demonscate the BER performance for selected RCPT codes generated by a Re = 1/3 mesber code with memory RE Bock length of 1024 information bis, aed a corresponding pune ‘utng period P = 8 when used with QPSK modulation in Fi, 2 for 3 Rician fading chanel with ¢? = 7'¢B. As shown In Fig these area et of codes with increasing eroreoectng capacities at coding rte deceases. In parca, we compare the BER performance of selected RCPT codes with correspond- ng RCPC codes when used with QPSK medulaton In Fig. 3. For BER = 10~, acoding galnof about 25 Bin B/N; exists for the rate /3 ub code wth memory M = 4 compre orate 1/3 ‘omolutional code with memory Af = 8 which are assumed tobe ‘of similar complexity: forthe corresponding rate 72 RCPT code, ‘heres 2 gan of about 15 Bin Big/M; compared tothe rte U2 [RCPC cote wth similar complexity. 5. RESULTS AND DISCUSSION We present some selected resus using RCPT codes anda ISCC approach fora representative QCIF video-conferencing sequence, ‘Sse a75 fps. These els were abiained using H 263+ coder ‘in conjunction with RCPT channel codes togeter with QPSK ‘modulation. The suis for both AWGN and Rican fading chan- nel are demonstrated, The maximim symbol transmission ae fs Sec tobe r2 = G& Ksps, such thatthe overall bate afer chan- el coding) is constzalned as Ree = 128 Kops duc to the use of ‘QPSK modulation. The packet ize anged rom 1 to 8 COB per RIP packet was optimized for each value of E's/¥s Figure 4 demonstraze ests forthe RTP-IL263+ packet video sransmision over an AWGN channel sing fixe source and chan- ne coding schemes. We provide results for this sytem using PER together wth ro coding with Re = 1/3 and memory M = 4 ‘compared to an uncoded sytem using oly PER. For comparison, the results forthe same rate convoitional codes, with memory Af and 8, ace alo iustated, Tt shouldbe obvious tha the absence of channel impalement, such as with large E/N, ‘he addtional channel coding overheads force 2 decease Inthe spallble source coding biuate? and this resuls in a correspond Teal wea Kbps. ng ll amie badge ot Rang = 128 1-502 Fig. 3. Simulated BER Performance of Selected RCPT and RCPC Codes over the Riian Fading chanel wih (= 7 dB when used with QPSK, Jing decrease inthe video qualty. This canbe seen if we compare results for the coded and uncoded systems for large E/N the difernce ts about 7 4B in PSNR. However, shouldbe nated that passive ero ecavery by self snot effective as #5/N de- creates below a thneshold value, approximately B/N; = 13 dB In this case. Asa result the reconstructed video quality degrades ddramatlaly. On the cther Rand, the coded cases can malnsin the video quality t acceptable levels for much smaller values of ‘B/N; compared tothe uncoded sytem which can be atrbuted {0 ls exor-comecting capably. This sa good indication ofthe rnecestity of emplaying FEC coding in witeless networks. Fur thermore, the system employing tubo coding ouperforms thase ‘sing comolusional codes with similar complexity. This i also lusrated In Fig. 4 if we compare the pesfrmance betwee tl rate 1/8 tubo code with Mf = 4 andthe rate 1/3 convolutional code with M = 8: a performance gan of about 2 4B in Fs/N ‘is obtained. Note thatthe Block size ofthe employed tubo coding. heres 1024 bts which s reasonably short resulting in negligible ardional delay for video transmission where a reasonably large rurmberof bits are generated and transated instanly. This sug {est that turbo codes are a reasonble cholce even for ral-ime ideo applications resling in improved power efficiency. Next we considera more realise wireless channel model, the Rician fading channel. Figae 5 demonstrates resus for RTP- 1.263 packet video transmission over a Rican fading channel with ¢? 7 oB. using fixed source and channel coding schemes, ‘Again, we provie resuls for a system using PER together with turbo coding with Re = 1/3 and Mf = 4 compared wo the une coded sytem using only PER. For comparison, the els fo the samme rate convlutoral codes, wit Mf = 4 and 8, ae also lus twaed. Its clear that to provide 2 fixed quality of reconstructed video, say 30 dB in PSNR, an B/N; in excess of 30 aB is re- {quired for an uncoded system. For inerferencelimited wireless Tinks, such lege power level will cause excessive interference to ‘other users, resulting in reduced system capecity. To improve the system capacity, FEC must be employed a clearly demonstrated ‘in Fig. 5. For example, forthe same quality level, 30 dB in PSNR, the selected turbo coded system requires only 2 4B in B/Ns, a substantial coving gun of over 28d in #5 /V. This achlved, however, a the expense ofa video quality degradation of about 7 4B n PSNR for large Bs/N;. Nevertheless, the significa pefor- ‘mance improvement at low B/N; clearly suggests the advantage ig. 4. Performance of RTP-H262+ packet video transmission ‘veran AWGN channel using fixed turbo and convolutional codes, ‘over a Rican fading camel with ¢ ‘convoludonl codes. 78 sing Bixed turbo and ‘of FEC coding inthe design of a mulusee witless communi ation system where eficint low power operation isthe key 10 Improved sysiem expacy. Furthermore, a luted In Fig, 8, {ato coding sill outperforms the convolutional codes with imi- Jac complexity by over I dB in E5/N7. Since the Block length of ‘usbo codes used la ths paper shld 10 1024 informatio bis, thls shouldbe considered a reasonably shorc-length code. For high bt trate video encoding and wansmissio systems, such s possible on wireless LANs, canbe expected that larger coding gai is pos: stole using wrbo codes wi larger block length whe maining accepable delay ‘The above results also suggest th uso of JSCC approach to aalaplvely allocating resources between source and channel cod Ing according tothe network condtons. In Fig. 8, we ihstrate ‘be results fr the JSCC scheme togetec witha set of fixed cing schemes fora clan fading channel with ¢? = 7 dB assuming de- ‘coding with perfect CSI. Iris elerly demonstrated tht with fixed FEC, Substantial performance improvement is obtained for small B/N with a performance penalty fo large E/N due to chan- nel cong overheads. However, the JSCC approsch eliminates ‘such peralty through the joint adjustment of source and channel ‘coding rate, such tha sccepabe end-to-end qual ls malnained for reasonably small values of Fs/N a5 wel as improved video ual for large values of Bs/N;. Punhermore, JSCC provides Truck more graceful endtoend quality degradation compared 1-543 Fig. 6. Pesformance of RIP-H.263+ packet video ransmission over aRiclan fading channel with = 7 dBusingJSCC approach ‘with RCPT codes. Also shown ar performance resuls fora set of fixed channel coding rte schemes. Fig. 7. Performance comparison of RIP-H.263+ packet video transmission over an AWGN channel sing SCC approach. to ote fled FEC schemes which 8 highly desirable for time ‘uring wireless communleaton networks Fall, we compare the performance af he bo and coo Juonal oding schemes wen wed In cone with» JSC approach in Figs Tan foran AWGN channel anda Rican fad ing chanel with? = 7 €B,respecvely! Asan be sen, se of SCC i ether cae can provide more grace pate a aly leradation by keeping the vdeo quay at an actepable lve for 2 uch wie range of B/N. Thi achieved by Joy select dog the channel and source cling rate sed onthe revaling ‘hae! condor, hee repeeted by B/3. As fs tht ‘ral stem performance il be forgiven chanel ‘cndions. The system employing REPTISCC outperforms the conesponing RCFC-ISCC scnme wih snl eomplenty due {othe adonal channel coding gai achive sing abo coe. Foe fied sales of PSNR this snouts o about 2B fn E/N ‘rer AWGN chanel or about 75 forthe Ran fang han nel wih = 7B, or smal values of B/N “re west nly the comet hl ove al opeaionl choles of Rated Re Fig. 8. Performance comparison of RIP-H.268+ packet vido ‘eansmission over a Rician fading channel with ¢? = 7 dB using SSCC approseh, 6. SUMMARY AND CONCLUSION We have described a JSCC strategy employing RCPT codes 10 ‘be used in conjuncdon with packstzed H.263e, vdeo transmit. ted ove wireless IP networks. The performance on both AWGN channels and Rion fdng channels evaluated forthe propose SCC scharme. The results clearly demonsirate that with approp- te RCPEISCC, packet video canbe delivered over a witless [P ‘network with improved end-to-end quality while exhibing a more graceful patter of quality degradation compared to fixed chanel ding Schemes. In particular, he impeoverent a small values of 2y/Ny offers the petenal of significa: Impeaverénts in system capacity 1, REFERENCES [1] M. Bysrom and JW. Modestne, “Combined Source-Channel ‘Coding for Video Transmission over an Aadiuve White Caus san Noise Chantel IEEE Journal on Selected Ares n Com ‘munications. ol 18, 0.8, p, 869-880, June 2000. (2) D.N. Rowich and L. B. Milstein, “On the Performance of Hybrid FEC/ARO Syatems Using Rate Compatible uncured ‘Tudo (RCPT) Codes,” IEEE Trans. Commun, vel 8, n0. 6. p- 848-859, June 2000. (3) J. Hagenauer, “Rate-Compatble Punctured Convolutional ‘Codes (RCPC Codes) aad thie Applicalons” IEEE Tran. ‘Commun, vo.COM-38, no. 4 pp, 388-400, Api 1988 (41 C.Bormann L Cline, G. Delser, . Gardos, C. Macioeo, D. Newell, J. tc G. Sullivan, S. Wenger, C. Zh, “RIP pay- Joe Forma forthe 1988 version of ITU-T Rec. H.263 video ‘H.263e)> RFC 2429, Oc 1998. (5) B.Cherinan, C. H, Wang, and L. flanzo, “Turbo and BCH. Codes Wide Band Burst-by Burst Adaptive H263-Assiscd ‘Wiles Video Telephony.” IEEE Trans. Circus and Stems {for Video Tachnology, v0.10, 8, Dec, 200, (61 ¥. Pet and ]W, Modesting, “Robust Packer Video Transmis- sion Over Heterogeneous Wire to-Wieless IP Networks Us- Ing ALF Together with Edge Proxies” in Pro. of EW2002 Feb. 2002. IM) Iivet Corp. “Video Codec Test Model TMNB” tpstandard lctelcom/ideosteh283plus! deaf dc, June 1987 1544

Potrebbero piacerti anche