WORKSHOP REPORT Bo co hi tho The views expressed here are those of the participants of this workshop and do not

necessarily reflect the views of any organization public or private

Quan im th hin y l nhng ngi tham gia hi tho ny v khng nht thit phn nh quan im ca bt k t chc cng cng hay t nhn.

Executive Summary Tm tt As described in the recent report of the Presidents Information Technology Advisory Committee (PITAC), improvements in information technology are transforming the way we work, learn,communicate, access information, practice health care, buy and sell goods, conduct research, understand our environment, build things, and run government agencies. Behind many of these developments and the developments needed for future information technologies are past and future advances in communications
Nh c m t trong bo co gn y ca Ch tch ban t vn cng ngh thng tin (PITAC), s pht trin trong cng ngh thng tin ang thay i cch chng ti lm vic, hc tp, giao tip, truy cp thng tin, thc hnh chm sc sc khe, mua v bn hng ho, tin hnh nghin cu, hiu r mi trng ca chng ta, xy dng mi th, v vn hnh cc c quan chnh ph. ng sau rt nhiu nhng pht trin v nhng pht trin cn thit cho cng ngh thng tin trong tng lai l nhng tin b trong qu kh v tng lai ca truyn thng

For the past 50 years, the field of communications has developed under the assumption that the two basic components of the communications problem can be performed independently with no performance degradations relative to joint design. Briefly, this idea, known as the separation principle, states that source coding or efficient information representation and channel coding or protection of information from corruption by noise can and should be performed independently. By dividing a single difficult problem into two simpler problems, the separation principle has led to 50 years of enormous advances in communications theory and technology. Examples of technologies made possible by communications advances include CD ROMs, fast modems, wireless communication systems, and the Internet
Trong vng 50 nm qua, lnh vc truyn thng pht trin theo gi nh rng hai thnhphn c bn ca vn truyn thng c th c thc hin c lp khng c s gim sthiu sut so vi thit k chung.Tm li, tng ny, c gi l "nguyn tc tch," ni r hn rng "m ha ngun" - i din thng tin hiu qu v "knh m ha" - bo v thng tin t sai lch bi nhiu - c th v nn c thc hin c lp.Bng cch chia mt vn kh khn vo hai vn n gin hn, nguyn tc tch dn n 50 nm tin b to ln trong L thuyt thong tin v truyn thng.V d v cc cng ngh c th c thc hinbi nhng tin b thng tin lin lc bao gm CD ROM,fast modems, h thng wiles v internet

Ironically, many of the modern information technology systems made possible by advances achieved through the use of the separation principle actually violate the conditions upon which the optimality of that principle relies. For such systems, performance improvements may be achieved by moving from separate design and operation of source and channel codes to joint source-channel code design and operation
Tr tru thay, rt nhiu cc h thng cng ngh thng tin hin i c th c thc hin bi nhng tin b t c thng qua vic s dng cc nguyn tc tch bit thc s vi phm cc iu kin ti u ca nguyn tc .i vi h thng nh vy, ci thin hiu sut c th t c bng cch di chuyn t thit k ring bit v hot ng ca ngun v m ha knh ti thit k" joint source-channel code " v vn hnh

The following report describes discussions and findings of the NSF-Sponsored Workshop on Joint Source-Channel Coding. The workshop was held in San Diego, California on October 12 and 13, 1999. Twenty-two experts, coming from a variety of different disciplines and points of view, participated in the workshop. The goals of the workshop were to: provide a contextual basis for understanding the array of applications in which source-channel coding can play a central role;assess the potential impact of source-channel coding technology in those applications; identify and characterize the major open research problems associated with the design, development, and deployment of source-channel coding technology; and develop a conceptual framework for future cross-cutting research activities in this inherently interdisciplinary area.
Cc bo co sau y m t cc cuc tho lun v kt qu ca Hi tho c ti tr trncc knh m ngun phn NSF.Hi tho c t chc ti San Diego, California vo ngy12 thng 10 v 13 nm 1999.Hai mi hai chuyn gia, n t nhiu lnh vc khc nhau v quan im, tham gia hi tho.Cc mc tiu ca hi tho: cung cp mt c s theo ngcnh cho s hiu bit cc mng ca cc ng dng, trong ngun knh m ha c thng mt vai tr trung tm, nh gi tc ng tim tng ca cng ngh m ha ngun knhtrong cc ng dng, xc nh v m t c im m lnnghin cu cc vn lin quann thit k, pht trin, v trin khai cng ngh m ha ngun knh v pht trin mt khung khi nim cho cc hot ng nghin cu xuyn sut trong tng lai trong lnh vc ny vn lin ngnh.

The committees results include a careful description of system properties that lead to the violation of the separation principle. These properties include existence of critical resource constraints (such as power constraints for hand-held wireless devices), networks (such as the Internet) in which users must share channel resources, applications (such as multimedia communications) characterized by source or channel heterogeneity, and systems (such as mobile communications devices) with unknown or time-varying source and channel properties. The committees assessment is that these properties pervade an enormous array of information systems, including many key network technologies. Wireless communications systems and the Internet are two prime examples considered in this report. Given the pervasive nature of these properties, the committee

recommends focused research in this area in order to: (1) advance the theory needed for characterizing the per-formance and fundamental limitations of joint source-channel coding systems, (2) develop methods for designing joint sourcechannel coding systems, and (3) apply the theory and design techniques to address practical problems. The committee also puts forward recommendations for encouraging productive research interaction between industry and academia in this area of enormous interest and importance to both.
Cc kt qu ca y ban bao gm mt m t cn thn ca cc thuc tnh h thng dn nvic vi phm nguyn tc tch.Cc tnh cht ny bao gm s tn ti ca hn ch ngun lcquan trng (chng hn nh hn ch nng lng cho thit b cm tay khng dy), mng(chng hn nh Internet) m ngi dng phi chia s cc ngun ti nguyn knh, cc ng dng (chng hn nh truyn thng a phng tin) c c trng bi ngun hoc khng ng nht knh,v h thng (chng hn nh cc thit b lin lc di ng) vi m ngunkhng r hoc bin thin theo thi gian v cc c tnh knh.y ban nh gi l cc thuc tnh ny trn ngp mt mng rt ln ca cc h thng thng tin, bao gm cc cng ngh mng quan trng.H thng thng tin lin lc khng dy v Internet l hai v d in hnhc xem xt trong bo co ny.Do tnh cht ph bin ca cc thuc tnh ny, y ban ngh nghin cu tp trung trong lnh vc ny : (1) ci tin nhng l thuyt cn thit m t hiu nng v c bn ca doanh ngun h thng knh m ha , (2)pht trin cc phng php thit k ni ngun v m ha knh, v (3) p dng cc l thuyt v k thut thit k gii quyt vn thc t.y ban cng a ra khuynngh khuyn khch s tng tc nghin cu sn xut gia cng nghip v hc thut trong lnh vc ny,quan tm rt ln v tm quan trng cho c hai.

II Introduction and Motivation Advances in information technology are transforming the way we live our lives. The impact is pervasive, affecting everything from communications to health care, commerce, and education. Since much of information technology is, at its core, communications technology, developments in communications represent key enabling advances for the information technology revolution currently underway
Nhng tin b trong cng ngh thng tin ang thay i cch chng ta sng .Tc ng lan ta, nh hng n tt c mi th t vic trao i thng tin chm sc sc khe, thng mi v gio dc.K t khi cng ngh thng tin pht trin, ti ct li ca n, cng ngh truyn thng, pht trin ca truyn thng i din cho nhng tin b quan trng thun li cho cuc cch mng cng ngh thng tin hin ang tin hnh

The field of communications describes the theory, methodologies, and technologies for transmitting and storing information. The information considered includes sources such as text, speech, images, audio, video, and scientific data. Traditionally, the notion of a communications system divides into two separate functions or subsystems. The first, called a source code or compression algorithm, aims to provide an efficient representation of the information; efficient data representations speed information transmission and minimize information storage-space requirements. The second, called a channel code or error control

code, works towards correcting or concealing errors arising from flaws in the storage or transmission system; channel codes protect information from corruption due to effects like electronic noise in a circuit, damage to a storage device (such as a scratch on a compact disk), loss of information packets in a network, and failure of a communications link
Cc lnh vc truyn thng m t l thuyt, phng php, v cng ngh truyn ti v lu tr thng tin. Cc thng tin c coi l bao gm cc ngun nh li ni, vn bn, hnh nh, m thanh, video, v d liu khoa hc. Theo truyn thng, khi nim v mt h thng truyn thng chia thnh hai chc nng ring bit hoc h thng con. Vic u tin, c gi lmt "m ngun" hay "thut ton nn," nhm mc ch cung cp mt i din hiu qu ca thng tin, d liu hiu qu i din tc truyn ti thng tin v gim thiu yu cu khng gian lu tr thng tin. Th hai, c gi l mt "knh m" hoc "M kim sot li," cng trnh hng ti sa cha hoc che giu li pht sinh t cc l hng trong h thng lu tr hoc truyn ti, m s knh bo v thng tin t nhiu do hiu ng nh n in t trong mt mch, thit hi cho mt thit b lu tr(chng hn nh mt vt try xc trn mt a nh gn ) , mt mt ca cc gi thng tintrong mt mng , v li ca mt lin kt truyn thng

The separation principle, which divides a single communications system into separate source coding and channel coding subsystems that are then designed and run independently, is both extremely powerful and extremely limiting to the advancement of communications technology. The power of the separation approach is that it divides a single complex problem into two simpler problems. This divide and conquer strategy has led to a wide variety of advances in both the theory and the practical design of source coding and channel coding systems. Theoretical advances include greater understanding of optimal source coding and channel coding methodologies and their performance. Practical advances include good, fast algorithms for source coding and channel coding. For example, modern compression systems typically halve the file size or transmission time needed for text and reduce by a factor of 40 or more those required for good image reproduction. Similarly, advances in channel coding have led to channel coding methods achieving reliable information communications and storage; in fact, the performance of these techniques is nearing the theoretical limits for simple common channels (e.g., single-user Gaussian noise on linear channels). These advances have enabled a wide array of information technologies, including CD ROMs, fast modems, wireless communications, the Internet, and digital libraries
Cc nguyn tc tch, phn chia mt h thng truyn thng thng duy nht vo cc m hangun ring bit v cc knh m ha h thng con ny sau c thit k v chy c lp, c hai cc k mnh m v cc k hn ch s tin b ca cng ngh truyn thng.Sc mnh ca phng php tch l n phn chia mt vn phc tp thnh hai vn n gin hn.Phn chia v chin lc chinh phc cc ny dn n mt lot cc tin btrong l thuyt v thit k thc t ca m ha ngun v cc knh m ha h thng.L thuyt nng cao bao gm s hiu bit su sc hn ca cc m ha ngun ti u v phng php m ha

knh v hiu sut ca chng. Tin b thc hnh tt, cc thut ton nhanh chng cho m ha ngun v m ha knh. V d, h thng nn hin i thng gim mt na kch thc tp tin hoc thi gian truyn cn thit cho vn bn v gim bi mt nhn t ca 40 hoc nhiu hn nhng ngi cn thit ti to hnh nh tt. Tng t nh vy, nhng tin b trong m ha knh dn n phng php m ha knh t c cc truyn thng thng tin ng tin cy v lu tr, trn thc t, vic thc hin cc k thut ny l gn gii hn l thuyt cho cc knh n gin ph bin (v d, ngi s dng duy nht Gaussian trn cc knh tuyn tnh). Nhng tin b ny cho php mt mng rng ca cng ngh thng tin, bao gm c CD ROM, fast modem, truyn thng khng dy, Internet, v cc th vin k thut s

A communication system in which the source code and channel code are designed or operated in a dependent fashion is called a joint source-channel code. Unlike separation-based techniques, joint source-channel code design techniques rely on the joint or cooperative optimization of communication system components. Further, the joint source-channel coding approach allows for strategies where the choice of source coding parameters varies over time or across users in a manner that in some way depends on the channel or network characteristics. Likewise, joint source-channel coding techniques allow for systems in which the choice of channel code, modulation, or network parameters varies with the source characteristics. In short, joint source-channel coding applies to any system in which the conditions of the separation theorem are violated and thus dependence of the source and channel code arises.
Mt h thng thng tin lin lc, trong cc m ngun v m knh c thit k hoc hotng trong mt cch ph thuc c gi l mt "doanh ngun knh m".Khng ging nhtch da trn k thut, doanh m ngun knh k thut thit k da trn vic ti u hadoanh, hp tc x ca cc thnh phn h thng thng tin lin lc.Hn na, cc knh tipcn m ngun chung cho php cc chin lc ni m s la chn m ha cc thng sngun thay i theo thi gian hoc qua ngi s dng trong mt cch m mt cch no ph thuc vo cc knh hoc cc c tnh mng.Tng t nh vy, cc k thut m hangun knh chung cho php cc h thng m trong s la chn ca m knh, iu ch,hoc cc thng s mng khc nhau vi c im ngun.Trong ngn hn, doanh ngun knh m ha p dng i vi bt k h thng trong cc iu kin ca nh l tch vi phm v do ph thuc vo ngun v m knh pht sinh.

Given the above observations about the conditions under which the assumptions for separationfail, we next list four examples of conditions under which we expect significant gains from jointsource-channel coding. First, gains are expected for systems with critical resource constraints. Examples of potential resource constraints include rate or bandwidth for data transmission, complexityfor low-cost systems, power for hand-held devices, and delay for real-time systems. Joint source-channel coding yields performance improvements in such scenarios by allowing for optimal allocation of a single users resources between the source code, channel code, and modulation scheme of his system. The goal here is to choose the allocation that yields the best end-to-end system performance. For example, in a system in which power is a critically limited resource, the power required to compress a signal should be no larger than the

power savings achieved by reducing the source description length. Likewise in realtime systems, where delay limitations represent a critical constraint, the delay introduced by the source code must be justified by a greater reduction in the time required to transmit the data using the more efficient data representation. By encouraging careful tradeoffs between the resources consumed by the source code and those consumed by the channel code and modulation schemes, critical resource constraints create fertile environments for the application of joint source-channel coding techniques.
Vi nhng quan st trn v cc iu kin theo cc gi nh cho ly thn khng, chng titip theo danh sch bn v d v cc iu kin theo chng ti hy vng li nhun ng k t jointsource knh m ha. Th nht, li nhun d kin cho cc h thng c hn ch ngun lc quan trng.V d v cc hn ch ngun lc tim nng bao gm hoc tc bng thng truyn d liu phc tp cho h thng chi ph thp, nng lng cho cc thitb cm tay, v s chm tr cho cc h thng thi gian thc. Lin ngun knh m ha ci tin hiu sut sn lng trong cc tnh hung nh vy bng cch cho php phn b ti ucc ngun ti nguyn ca mt ngi dng duy nht gia cc m ngun, m knh, v iu ch s h thng ca mnh. Mc tiu y l chn phn b sn lng h thng end-to-endhiu sut tt nht. V d, trong mt h thng trong quyn lc l mt ngun ti nguynquan hn ch, sc mnh cn thit nn tn hiu khng ln hn cc khon tit kim int c bng cch gim di m t ngun. Tng t nh vy trong h thng thi gian thc ", trong hn ch s chm tr i din cho mt hn ch quan trng, s chm trgii thiu ca m ngun phi c iu chnh bi mc gim nhiu hn trong thi gian cn thit truyn ti d liu bng cch s dng biu din d liu hiu qu hn. By khuyn khch s cn bng gia cc ngun ti nguyn tiu th bi cc m ngun v nhngngi tiu th bi cc m knh v n iu ch cn thn, hn ch ngun lc quan trngto ra mi trng mu m cho vic p dng cc k thut m ha ngun knh chung.

A second example of a scenario where the potential performance benefits of joint sourcechannel coding are high is multiuser systems with shared channels. Wired data networks, where packets of information travel through a sequence of shared links on their way from their point of origin to their point of destination, are one example of this type of shared-resource environment. Wireless communications systems, where users attempting to send information through the system compete for the limited system bandwidth, are a second example. In systems in which users share a common channel, the signals transmitted by different users can interfere with each other. Thus one users source contributes to the noise in another users channel. The competition for network resources and the dependence of the channel on the sources feeding information into the channel make shared-channel environments prime candidates for application of joint sourcechannel coding solutions. In particular, while joint design of source and channel codes across users in a shared-channel environment is typically not possible, some form of cooperation can limit

interference between users and thereby lead to better overall system performance. System protocols establish a framework in which these types of cooperation take place. For example, in packet based networks, system protocols describe methods for breaking a single users information into packets and routing those information packets, along with the packets from other users, through the network. In some sense, the protocol becomes the channel through which the data description passes. Here joint sourcechannel coding techniques lend insight into how to achieve better system performance. In particular, we can engineer the source to match the channel by modifying source rates to send more information when network congestion is low and less when network congestion is high. Likewise, we can engineer the channel or the network protocols to match the source statistics. For example, in packet networks we can use the bits in the headers of data packets to instruct the routers on which packets have low priority and can be dropped if necessary to avoid overflow. This approach, which is analogous to existing methods for using bits in the data packet body to describe to the source and channel decoder how to decode a particular transmitted message, allows for control of the channel based on the importance of the information in the source. These joint source-channel coding approaches for engineering sources that yield better channels and channels that behave in ways better suited to the source are a natural match for shared-channel environments
Mt v d th hai ca mt kch bn m nhng li ch v hiu sut m ha sourcechannel phn cao l h thng a ngi dng vi cc knh chia s. C dy mng d liu, ni m cc gi d liu thng tin du lch thng qua mt chui cc lin kt chia s trn con ng ca mnh t quan im ca h v quan im ca h n, l mt trong nhng v d ca loi hnh ny chia s ti nguyn mi trng. H thng truyn thng khng dy, ni ngi dng c gng gi thng tin thng qua h thng cnh tranh cho cc bng thng h thng gii hn, l mt v d th hai. Trong cc h thng trong ngi s dng chia s mt knh chung, cc tn hiu c truyn bi ngi dng khc nhau c th can thip vi nhau. Nh vy, "ngun" ca mt ngi s dng gp phn vo ting n ca ngi s dng khc cnh tranh "knh." i vi ti nguyn mng v s ph thuc ca knh trn ngun nui dng thng tin vo cc knh lm cho mi trng chia s knh chnh ng c vin cho cc ng dng gii php m ha sourcechannel doanh. c bit, trong khi thit k chung ca ngun v m knh trn ngi s dng trong mt mi trng chia s knh thng khng c th, mt s hnh thc hp tc c th hn ch s can thip gia ngi s dng v do dn n hiu sut h thng tng th tt hn. H thng giao thc thit lp mt khun kh trong cc loi hp tc din ra. V d, trong gi tin da trn mng, cc giao thc h thng m t cc phng php ph v thng tin ca mt ngi dng duy nht thnh cc gi v nh tuyn cc gi thng tin, cng vi cc gi tin t nhng ngi dng khc, thng qua mng. Trong mt ngha no , giao thc tr thnh "knh" thng qua m t d liu i qua. Di y doanh ngun knh k thut m ha cho vay ci nhn su sc vo lm th no t c hiu sut h thng tt hn. c bit, chng ta c th k s m ngun ph hp vi cc knh bng cch thay i t l ngun gi thng tin khi tc nghn

mng l thp v t khi tc nghn mng l cao. Tng t nh vy, chng ta c th k s knh hay giao thc mng - ph hp vi s liu thng k ngun.V d, trong cc mng gi, chng ti c th s dng cc bit trong cc tiu ca gi d liu hng dn cc b nh tuyn m trn cc gi tin c mc u tin thp v c th c gim nu cn thit trnh trn. Cch tip cn ny, l tng t nh phng php hin c s dng cc bit trong c th gi d liu m t ngun v b gii m knh lm th no gii m mt thng ip truyn c th, cho php kim sot cc knh da trn tm quan trng ca thng tin trong ngun. Nhng phng php tip cn ngun knh chung m ha cho cc ngun k thut nng sut tt hn cc knh v cc knh hot ng trong nhng cch tt hn ph hp vi ngun l mt trn u t nhin cho mi trng chia s knh

A third basic scenario in which significant gains are expected from the application of joint source-channel coding methodologies is in multiuser systems with heterogeneous sources, channels, topologies, or users. Source heterogeneity arises from the fact that different data types traversing a single communications system have different sensitivities to loss, corruption, and delay. Channel heterogeneity arises from the fact that different channels within a single network may have different noise characteristics, available rates, and, in packet networks, delay and jitter. User heterogeneity refers to the differences between the needs and qualityof-service requirements of different users, the different computational capabilities of their systems, and the different bandwidth access available to them. Heterogeneity of topology refers to the fact that the same network can be simultaneously used as a point-to-point communication system (with or without feedback), a broadcast system, a multiple access system, and so on. A variety of joint source-channel coding techniques may be employed in networks exhibiting these types of heterogeneity. For systems with heterogeneous sources, unequal error protection can be used to protect more error- or channels or users, layered or multire solution source coding allows transmission at multiple rates (for multiple channels) or makes information simultaneously accessible with different reproduction qualities to multiple users with varying needs and system capabilities; the result is a democratization of information access, where information can be made available to all users despite the potentially very different bandwidths and computational capabilities of their access devices. Finally, network topology can affect the choice of source and channel codes. For example, separation is known to fail for some network topologies even when the sources and channel are known and the system delay and complexity are unconstrained. Further, feedback in a network system topology can affect source and channel coding choices. For example, using feedback to request repeated transmission of missing or corrupted data packets can obviate traditional channel coding; prioritization of packets by relative importance of the included source content can then lead to prioritized repeat requests for scenarios where delay constraints are tight. The matching of source coding parameters to heterogeneous channels and networks and channel

coding parameters, modulation techniques, and network protocols to source characteristics in these heterogeneous environments are inherently joint sourcechannel coding approaches
Mt kch bn c bn th ba, trong mc tng ng k t vic p dng cc phng php m ha ngun knh chung trong cc h thng a ngi dng vi ngun khng ng nht, cc knh, cu trc lin kt, hoc ngi s dng. Ngun khng ng nht pht sinh t thc t l cc kiu d liu khc nhau vt qua mt h thng thng tin lin lc duy nht c nhy cm khc nhau tht thot, tham nhng, s chm tr v. Knh khng ng nht pht sinh t thc t l cc knh khc nhau trong mt mng duy nht c th c c im ting n khc nhau, t l c sn, v, trong cc mng gi tin chm tr v jitter. Ngi s dng khng ng nht cp n s khc bit gia nhu cu v yu cu cht lng dch v ca ngi s dng khc nhau, kh nng tnh ton khc nhau ca h thng ca h, v bng thng truy cp khc nhau c sn cho h. Tnh khng ng nht ca cu trc lin kt cp n thc t l cng mt mng c th c ng thi s dng nh mt h thng thng tin lin lc im-im (c hoc khng c phn hi), mt h thng pht sng, mt h thng truy cp nhiu, v do . Mt lot cc k thut m ha phn ngun knh c th c s dng trong mng li trin lm cc loi khng ng nht. i vi h thng vi cc ngun khng ng nht, bt bnh ng bo v li c th c s dng bo v li hoc nhiu knh hoc ngi s dng, lp hoc multiresolution ngun m ha cho php truyn nhiu mc thu (i vi nhiu knh khc nhau) hoc lm cho thng tin ng thi tip cn vi cht lng sinh sn khc nhau cho nhiu ngi dng vi nhu cu v kh nng h thng khc nhau, kt qu l mt "dn ch" ca truy cp thng tin, ni thng tin c th c thc hin cho tt c ngi dng mc d bng thng c kh nng rt khc nhau v kh nng tnh ton ca cc thit b truy cp ca h. Cui cng, mng topo c th nh hng n s la chn m ngun v knh. V d, tch c bit n tht bi cho mt s mng cu trc lin kt ngay c khi cc ngun v knh c bit n v s chm tr v phc tp h thng khng b gii hn. Hn na, thng tin phn hi trong mt cu trc lin kt h thng mng c th nh hng n ngun v knh m ha s la chn. V d, s dng thng tin phn hi yu cu truyn lp i lp li ca d liu b mt hoc b hng cc gi tin c th phong nga m ha knh truyn thng, u tin cc gi tin vo tm quan trng tng i ca ni dung ngun bao gm sau c th dn n yu cu lp li u tin cho cc kch bn hn ch s chm tr cht ch. Ph hp vi m ha cc thng s ngun knh khng ng nht v mng li v knh m ha cc thng s, k thut iu ch, v cc giao thc mng vi c im ngun trong cc mi trng khng ng nht l phng php tip cn ngun vn doanh knh m ha

The fourth and final category of problems in which joint source-channel coding is expected to achieve significant gains is applications characterized by unknown or time-varying sources, channels, or networks. The issues arising here are the same as those considered for systems with heterogeneous sources, channels, networks, or users, and the techniques that arise to combat these problems are likewise similar. The difference here is that the potential variation occurs across time for a single source, channel, or network rather than varying across space or system use. The goal for time-varying systems is to achieve graceful performance fluctuations as the system evolves and changes with time
Cc th loi th t v cui cng ca cc vn trong m ha ngun doanh knh d kin s t c mc tng ng k l cc ng dng c trng bi cc ngun khng rhoc bin

thin theo thi gian, cc knh, hoc mng.Cc vn pht sinh y ging nhnhng ngi xem xt h thng vi ngun khng ng nht, cc knh, mng, hoc ngi s dng, v cc k thut pht sinh chng li nhng vn ny l tng t nh vytng t.S khc bit y l s bin i tim nng xy ra qua thi gian cho mt knh duy nht, ngun, hoc mng ch khng phi l khc nhau qua khng gian hay h thng sdng.Mc tiu cho cc h thng thay i theo thi gian l t c hiu sut bin ngduyn dng nh h thng pht trin v thay i theo thi gian

Examples of existing technologies and applications in which the above-described characteristics appear are bountiful. For example, second generation (G2) voiceoriented digital cellular phones combine critical resource constraints, a sharedchannel environment, and extreme channel variations over time and space. Critical resource constraints include power constraints for handheld devices and delay constraints for real-time voice communications. The very noisy shared channel environment arises in part from extreme competition for bandwidth, a problem that increases as cellular systems increase in popularity and use. Further, wireless channels experience extreme variations over time and user location due to changes in such characteristics as distance from base stations, interference from other users, shadowing effects, and fading characteristics. The combined effect of all of these characteristics makes digital cellular telephony a prime candidate for the application of joint source-channel coding techniques.
Cc th loi th t v cui cng ca cc vn trong m ha ngun doanh knh d kin s t c mc tng ng k l cc ng dng c trng bi cc ngun khng rhoc bin thin theo thi gian, cc knh, hoc mng.Cc vn pht sinh y ging nhnhng ngi xem xt h thng vi ngun khng ng nht, cc knh, mng, hoc ngi s dng, v cc k thut pht sinh chng li nhng vn ny l tng t nh vytng t.S khc bit y l s bin i tim nng xy ra qua thi gian cho mt knh duy nht, ngun, hoc mng ch khng phi l khc nhau qua khng gian hay h thng sdng.Mc tiu cho cc h thng thay i theo thi gian l t c hiu sut bin ngduyn dng nh h thng pht trin v thay i theo thi gian

Another existing system exhibiting many of the above characteristics is the Internet. Due to its shared-channel environment and heterogeneous components, the Internet is characterized by extreme channel variations over space and time. Variations over space describe the variations between the channels observed by different sender-receiver pairs in the network. These variations can be extreme, with bandwidths ranging from 10s of kbps (kilobits per second) to 10s of Mbps (megabits per second), packet loss probabilities varying from fractions of a percent to 25% or more, and delay and jitter varying from milliseconds to seconds. In addition to these variations over space, are variations over time. A single links bandwidth, packet loss probability, delay, and

jitter can vary enormously over time as competing processes begin and end. In distance learning, for example, downloading a web page during a real-time video transmission can cause severe performance degradations. User mobility adds further heterogeneity to the system, bringing with it all of the time and space variations described above for cellular telephony systems. Finally, network topology further complicates the issue since the Internet is simultaneously used for pointto-point, broadcast, multiple access, and feedback communications applications
Mt h thng hin c trng by rt nhiu cc c im trn l Internet.Do mi trng chia s cc knh ca n v cc thnh phn khng ng nht, Internet c c trng bi s thay i knh cc v khng gian v thi gian.Bin th trong khng gian m t cc bin thgia cc knh quan st bi ngi gi, nhn cc cp khc nhau trong mng.S bin i ny c th c khc nghit, vi bng thng khc nhau, t 10 kbps (kilobits mi giy) n10 Mbps (megabits mi giy), xc sut mt gi tin khc nhau t vi phn ca mt phn trm n 25% hoc nhiu hn, v s chm tr v jitter khc nhau t phn nghn giyn giy.Ngoi ra cc bin th trn khng gian, s thay i theo thi gian. Mt lin kt duynht ca bng thng, xc sut mt gi tin, chm tr, v jitter c th rt khc nhau theo thi gian khi cc qu trnh cnh tranh bt u v kt thc.Trong o to t xa, v d, ti v mt trang web trong mt truyn video thi gian thc c th gy ra s gim st hiu sut nghim trng. Ngi s dng di ng b sung thm tnh khng ng nht hn na vo h thng, mang theo tt c cc thi gian v s bin ikhng gian m t trn cho cc h thng in thoi di ng. Cui cng, cu trc lin ktmng li phc tp thm vn k t khi Internet ng thi s dng cho pointto im,pht sng, a truy cp, v cc ng dng thng tin phn hi thng tin lin lc

Other examples of current systems and applications characterized by the above properties of critical resource constraints, shared channels, system heterogeneities, and time variance are numerous. The tight delay constraints of video telephony, video conferencing, and video-on-demand applications create a resourceconstrained environment. In video broadcast applications, differences between the systems and links of different users create a system characterized by user heterogeneity. The desire for ubiquitous access, fast browsing, and easy downloading in a large data bases such as a digital library creates another heterogeneous user and channel environment. By failing to meet the separation theorems conditions for optimality, each of these applications becomes a candidate for the application of joint source-channel coding techniques
Cc v d khc ca cc h thng hin hnh v cc ng dng c trng bi cc tnh cht hn ch ngun lc quan trng trn, cc knh chia s, heterogeneities h thng, v ng thi gian rt nhiu.Hn ch s chm tr cht ch ca in thoi video, hi ngh truyn hnh, vcc ng dng video theo yu cu to ra mt mi trng ti nguyn hn ch.Trong cc ng dng chng trnh truyn hnh, s khc bit gia cc h thng v lin kt ca cc ngi dng

khc nhau to ra mt h thng c trng bi tnh khng ng nht ca ngi sdng.Mong mun truy cp mi ni, duyt web nhanh v ti v d dng trong mt c sd liu ln chng hn nh mt th vin k thut s to ra mt ngi s dng khng ng nht v mi trng knh. Bng khng p ng cc iu kin ca nh l tch cho ti u, cc ng dng ny tr thnh mt ng c vin cho vic p dng k thut m ha knh ngun doanh

With advances in information technology, the properties that cause the separation theorem to fail are becoming more common. For example, G2.5 and G3 digital cellular systems will add simultaneous data transmission to their voice services, allow variable data rates (with up to 2 Mbps for stationary conditions and up to 144 kbps for high mobility), run the Internet protocol on top of the data link, and offer multimedia (image, video, and audio) services on top of the Internet protocol. As a result, these future systems will be characterized by even greater source, channel, and user heterogeneity and more time variance than G2 systems. Similarly, next generation Internet will add even greater channel and source heterogeneities to an already diverse environment. New Internet channel heterogeneities will arise from both the increase, by several orders of magnitude, of allowed transmission capacity for high-end users and the increase in use by low-data-rate mobile users. Increased use of multimedia applications over the Internet will likewise lead to greater source heterogeneities. Other future applications in which joint source-channel coding may play a significant role include last mile technologies, distributed sensor arrays, network transport of digital video and images, packet-switched video conferencing, multimedia delivery on digital subscriber loops, and digital audio delivery systems
Vi nhng tin b trong cng ngh thng tin, cc ti sn gy ra cc nh l tch bit khng tr nn ph bin hn.V d, G2.5 v G3 h thng k thut s di ng s thm ng thi truyn d liu vi cc dch v ting ni ca h, cho php tc d liu bin (ln ti 2Mbps i vi iu kin vn phng phm v ln n 144 kbps cho di ng cao), chy giao thc Internet trnlin kt d liu, v cung cp a phng tin (hnh nh, video, v m thanh) dch v trn u trang ca giao thc Internet.Kt qu l, cc h thng ny trongtng lai s c c trng bi ngun thm ch cn ln hn, knh, v tnh khng ng nht ca ngi s dng v phng sai nhiu thi gian hn hn so vi cc h thng G2.Tng t nh vy, Internet th h k tip s thm knh ln hn v heterogeneities ngun n mt mi trng c a dng. Heterogeneities knh Internet mi s pht sinh t c hai tng, mt s n t hng cacng , dung lng truyn dn cho php ngi dng cao cp v tng s dng bingi s dng in thoi di ng tc d liu thp. Tng cng s dng cc ng dnga phng tin trn Internet tng t nh vy s dn n heterogeneities ngun ln hn.Cc ng dng khc trong tng lai, trong doanh ngun-knh m ha c th ng mtvai tr quan trng bao gm cng ngh dm cui cng, mng cm bin phn phi, mng li giao thng vn ti ca video k thut s v hnh nh, hi ngh truyn

hnh chuyn mch gi, giao hng thu bao k thut s a phng tin trn cc vng lp, v cc h thng phn phi m thanh k thut s

There are many other examples of the successful use of very simple joint sourcechannel coding techniques in existing systems. For example, error concealment is used in video telephony and conferencing to hide the effects of uncorrected channel errors. Video-on-demand systems use prioritized automatic repeat requests so that higher priority data such as audio may be reliably received at the expense of losing lower priority data such as video when high loss conditions are experienced. Finally, video broadcast applications very commonly use layered coding in order to simultaneously provide lower quality data reproductions to users with lower available datarates and higher quality data reproductions to users with higher available data-rates using a single embedded data description
C rt nhiu v d khc ca vic s dng thnh cng k thut rt n gin lin ngun knhm ha trong h thng hin c.V d, che giu li c s dng trong in thoi video v hi ngh truyn n nhng nh hng ca li knh cha c sa cha.Video theo yucu h thng s dng cc yu cu lp li t ng u tin cc d liu u tin cao hnnh m thanh c th c mt cch ng tin cy nhn c ti cc chi ph ca vic mt d liu u tin thp hn nh video, khi iu kin mt mt cao c kinh nghim.Cui cng,cc ng dng video pht sng rt thng c s dng nhiu lp m ha ng thi cungcp d liu cht lng thp hn bn sao cho ngi dng vi datarates c sn thp hn vcht lng cao hn d liu bn sao cho ngi dng d liu c sn cao hn t l s dngd liu m t mt nhng

The properties given above as examples of appropriate scenarios for the use of joint source channel coding and the joint source-channel coding techniques briefly described there barely scratch the surface of the problem and its solution. A more detailed description of a single system gives a better picture of the system characteristics of which joint source-channel coding may eventually take advantage
Cc ti sn a ra trn l v d v cc kch bn thch hp cho vic s dng m ha knhlin ngun v cc knh k thut chung ngun m ha m t ngn gn c hu nh khngxc b mt ca vn v gii php ca n.Mt m t chi tit hn v mt h thng duy nht cho mt hnh nh tt hn cc c tnh h thng trong cc knh m ha ngun doanhcui cng c th tn dng li th

Example: The Internet As described earlier, the Internet is an example of a network in which there are tremendous channel variations over time and space. Not noted earlier, however, is the fact that the channel variation is also correlated in time and space. The correlation in time arises from channel memory. Much of the packet loss experienced in packet-based channels arises from buffer overflow. Since buffers,once full, take time to drain, packet losses in a network often occur in

bursts. The correlation in space arises from the shared-channel environment. For example, two neighboring paths through a network may share one or more nodes or links. Overflow in a shared node generally results in lost packets for both channels. Correlations may also exist for two paths not sharing any nodes since both may be affected by high transmission rates on a third path that crosses both of them
V d: Internet Nh c m t trc , Internet l mt v d ca mt mng, trong c nhng bin thknh to ln v thi gian v khng gian. Khng ghi nhn trc , tuy nhin, l mt thc trng s bin i knh l c lin quan trong thi gian v khng gian. Cc mi tng quantrong thi gian pht sinh t b nh knh. Phn ln s mt mt gi tin c kinh nghim trongcc knh da trn gi tin xut pht t li trn b m. K t khi b m, mt ln y ,mt thi gian ro nc, tn tht gi tin trong mt mng thng xy ra trong v n. Ccmi tng quan trong khng gian pht sinh t mi trng chia s knh. V d, hai con ng ln cn thng qua mt mng li c th chia s mt hoc nhiu hn cc nt hoc lin kt. Trn trong mt nt c chia s ni chung kt qu trong cc gi d liu b mt cho c hai knh. Tng quan cng c th tn ti hai con ng khng chia s bt k nt v c hai u c th b nh hng bi tc truyn cao trn mt con ng th ba i qua c haingi trong s h

These characteristics create many challenges and opportunities for joint sourcechannel coding in networked environments. Even in the simplest case, with only a single sender and a single receiver, the channel between the sender and receiver is not known a priori, is in fact time varying, and the memory of the channel is usually large compared to the delay constraint for the application. If there is a single sender broadcasting (actually, multicasting) to multiple receivers, then the sender must deal with coding for a heterogeneous mixture of unknown but correlated channels. This is the problem of joint source-channel coding for the broadcast channel. One way to deal with this problem is with feedback from the receivers to the sender. Each receiver can report, for example, its average transmission rate, loss rate, delay, and jitter. Such receiver reports are highly correlated because the network paths between adjacent receivers have many links in common. Although it is known that feedback from a single receiver to a single sender does not increase the capacity of a memoryless channel, the broadcast feedback problem involves multiple receivers and a channel with memory. In this environment, there is good reason to take advantage of feedback, which in a highly connected network is readily available (although over unreliable channels). Feedback from the multiple receivers to the single sender can be considered a problem in joint source-channel coding for the so-called multiple access channel. It is known that source coding for correlated sources plus channel coding for the

multiple access channel is inferior, in general, to joint source-channel coding. In other words, in multiuser communication networks, it is known that the separation theorem does not hold, even when the sources and channels are known and the delay is unconstrained
Nhng c im ny to ra nhiu thch thc v c hi cho doanh ngun knh m ha trong mi trng mng. Ngay c trong trng hp n gin, vi ch ngi gi v ngi nhn mt, knh gia ngi gi v ngi nhn khng bit mt u tin, trong thi gianthc t khc nhau, v b nh ca knh thng l ln so vi hn ch s chm tr cho ng dng. Nu c mt ngi gi pht sng duy nht (trn thc t, multicasting) nhiu ngi nhn, sau ngi gi phi i ph vi m ha cho mt hn hp khng ng nht ca cc knhkhng r nhng tng quan. y l vn ca doanh ngun knh m ha cho cc knhpht sng. Mt cch i ph vi vn ny l thng tin phn hi t ngi nhn cho ngi gi. Mi ngi nhn c th bo co, v d, tc truyn ti trung bnh, t l mt, chm tr v jitter. Bo co nhn nh vy l lin quan cht ch bi v mng li ng ligia thu lin k c nhiu lin kt chung. Mc d n c bit n thng tin phn hi tmt my thu duy nht mt ngi gi n khng lm tng nng lc ca mt knh memoryless, vn qung b thng tin phn hi lin quan n nhiu ngi nhn v l mt knh vi b nh. Trong mi trng ny, l l do tt tn dng li th thng tin phn hi, trong mt mng li kt ni c sn (mc d trn cc knh khng ng tin cy). Thng tin phn hi t nhiu ngi nhn cho ngi gi c th c coi l mt vn trong ccknh m ha ngun doanh cho cc knh truy cp nhiu ci gi l. c bit, ngun m ha cho cc ngun tng quan cng vi knh m ha cho cc knh truy cp nhiu l km,ni chung, cc knh m ha ngun chung. Ni cch khc, mng li truyn thng a ngi dng, n c bit rng cc nh l tch bit khng t chc, ngay c khi cc ngunv cc knh c bit n v s tr hon l khng b gii hn

The transport of multimedia data across networks like the Internet poses even more problems. Existing reliable transport protocols (TCP/IP) are matched to data applications requiring guaranteedrobust performance. To guarantee robust performance, TCP/IP requests retransmission of all lost or corrupted packets, thereby achieving the desired protection from noise at the expense of an imposed delay. Unfortunately, the resulting delay makes these protocols a poor choice for real-time data traffic, which is characterized by high delay sensitivity and variable tolerances to noise. In the networking field, two approaches to this problem have emerged: integrated services and differentiated services. The integrated services approach stresses guaranteed qualities of service for different service classes, while the differentiated services approach stresses different, though not guaranteed, qualities of service for different service classes. Guaranteed quality of service is more difficult to implement and consumes more network resources, but provides stable channels for traditional separation-based source coding algorithms such as MPEG video. On the other hand, differentiated services, in conjunction with joint source-channel coding, could prove to be significantly more resourceefficient than integrated services. As an example, a simple differentiated

services protocol that calls for lower priority packets to be dropped during congestion before higher priority packets would obviate the need for explicit channel coding and unequal error protection. This could result in a lower transmission rate, higher quality, lower delay, and lower network resource usage compared to a separation-based solution. In general, information placed in packet headers (such as priority tags, protocol parameters, or even programmatic scripts destined for active routers) can be used by joint source-channel coders to change the way different packets are handled by the network, potentially as a function of buffer delay, packet loss, changing transmission bandwidth, and so forth, either measured by the routers directly, or received as feedback from downstream routers. The Internet provides a rich variety of research problems in joint source-channel coding. The solutions to these problems would likely have a major impact on networks of the future
Vic vn chuyn cc d liu a phng tin qua mng nh Internet t ra vn thm ch nhiu hn na. Hin ti cc giao thc giao thng vn ti ng tin cy (TCP / IP) l ph hp vi cc ng dng d liu i hi hiu sut guaranteedrobust. m bo hiu sut mnh m, TCP / IP truyn li yu cu ca tt c cc gi d liu b mt hoc b hng, do t c bo v mong mun t ting n ti cc chi ph ca mt s chm tr p dng.Tht khng may, s chm tr kt qu lm cho cc giao thc ny mt s la chn cho ngi ngho thi gian thc d liu lu lng truy cp, c c trng bi s chm tr nhy cm cao v dung sai bin vi ting n. Trong lnh vc mng, hai cch tip cn vn ny ni ln: cc dch v tch hp v dch v khc nhau. Cc dch v tch hp phng php tip cn nhn mnh cc phm cht ca dch v bo lnh cho cc lp hc dch v khc nhau, trong khi cch tip cn cc dch v khc nhau nhn mnh khc nhau, mc d khng m bo, cht lng ca dch v cho cc lp hc dch v khc nhau. m bo cht lng dch v l kh khn hn thc hin v tiu th ti nguyn mng nhiu hn, nhng cung cp cc knh n nh cho truyn thng da trn thut ton m ngun tch m ha, chng hn nh video MPEG. Mt khc, dch v khc nhau, kt hp vi cc knh m ha ngun chung, c th chng minh c nhiu hn ng k ti nguyn hiu qu hn so vi cc dch v tch hp. V d, mt s khc bit n gin dch v giao thc m cc cuc gi cho gi u tin thp hn b rt trong khi tc nghn trc gi u tin cao hn s phong nga s cn thit m ha knh r rng v bo v li khng ng u. iu ny c th dn n mt tc truyn ti thp hn, cht lng cao hn, s chm tr thp hn, v s dng ti nguyn mng thp hn so vi mt gii php da trn tch. Ni chung, thng tin c t trong tiu gi tin (chng hn nh th u tin, giao thc thng s, hoc cc kch bn thm ch c chng trnh dnh cho cc b nh tuyn hot ng) c th c s dng bi cc lp trnh vin doanh ngun knh thay i cch cc gi d liu khc nhau c x l bi h thng mng, c kh nng nh mt chc nng ca b m gi tin, chm tr mt mt, thay i bng thng truyn ti, v vv, hoc o bng cc b nh tuyn trc tip, hoc nhn thng tin phn hi t cc b nh tuyn h lu. Internet cung cp mt phong ph a dng ca cc vn nghin cu trong cc knh m ha ngun doanh. Cc gii php cho nhng vn ny c th s c mt tc ng ln n mng li ca tng lai

III Research Goals and Challenges

Three pressing research goals for joint source and channel coding are (1) develop appropriate theory for characterizing the performance and fundamental limitations of joint source-channel coding systems, (2) determine methods for designing joint source-channel coding systems, and (3) apply the theory and design techniques to address practical problems. A variety of more specific research objectives addressing these problems is listed below. These objectives represent many of the significant research goals in joint source-channel coding. The order in which these items are listed reflects no prioritization over the described projects. Much of this research is expected to be multidisciplinary in nature, combining expertise from such areas as signal communications and information theory, networking and queuing theory, as well as input from areas including economics and psychology
Ba mc tiu nghin cu nhn cho ngun doanh v m ha knh l (1) pht trin l thuytph hp vi c im hiu sut v cc hn ch c bn ca cc h thng m ha ngunknh doanh, (2) xc nh phng php thit k cc h thng m ha ngun knh doanh, v (3)p dng cc l thuyt v k thut thit k gii quyt vn thc t. Mt lot ccmc tiu nghin cu c th hn gii quyt nhng vn ny c lit k di y. Nhng mc tiu i din cho nhiu nghin cu mc tiu quan trng trong cc knh m ha ngun chung. Trnh t, trong cc mt hng ny c lit k phn nh khng c u tin i vi cc d n c m t. Phnln ca nghin cu ny d kin s c tnh a ngnh trong t nhin, kt hp chuyn mn tcc lnh vc nh truyn thng tn hiu v l thuyt thng tin, mng v l thuyt xp hng, cngnh l u vo t cc khu vc bao gm c kinh t v tm l hc

Performance and Fundamental Limits Many of the major challenges in developing appropriate theory for joint sourcechannel coding are related to understanding the tradeoffs between the quality, delay, complexity, channel usage, and power inherent in code design. Quality refers to the end-to-end distortion between the original data and its reproduction. Delay refers to the difference between the times at which data enters the encoder and exits the decoder. Complexity refers to the requirements of the encoding and decoding algorithms, including such things as number of arithmetic operations per data sample and memory use, which may be heavily influenced by implementation considerations. (Complexity is also a primary factor in determining how much power a particular joint source-channel coding implementation consumes.) Channel usage refers to issues such as the number of channel uses per sample, channel bandwidth, and channel signal-to-noise ratio (which includes both channel noise and transmit power); both implementation and transmit power use become important factors in energy-limited applications (e.g., battery-powered devices). It is also important to limit transmit power when one users transmitted power can significantly impact the performance of others. While development of a comprehensive theoretical framework for joint source-channel coding is a suitable

long-term goal, it is evident from prior work that completely characterizing the relevant tradeoffs between quality, delay, complexity, channel usage, and power is extremely difficult, and that achieving progress in this area is very challenging
Hiu sut v cc gii hn c bn Nhiu ngi trong s nhng thch thc ln trong l thuyt thch hp pht trin cho knhm ha ngun doanh lin quan n s hiu bit cn bng gia cht lng, chm tr, phc tp, s dng knh,v sc mnh vn c trong thit k m. Cht lng cp n s bin dng end-toendgia cc d liu gc v sinh sn ca n. S chm tr lin quan n s khc bit gia cc ln m d liu nhp vo cc b m ha v ra khi cc b gii m. Phc tp lin quann cc yu cu ca cc thut ton m ha v gii m, bao gm c nhng th nh s lng php tnh s hc trn d liu mu v s dng b nh, m c th b nh hng nng n bi nhng cn nhc thc hin. ( phc tp cng l mt yu t chnh trong vic xc nh mt ngun doanh c th thc hin m ha knh tiu th bao nhiu nng lng.) S dng knh cp n cc vn nh s lng cc knh s dng cho mi mu, bng thng knh v knh t l tn hiu-to-noise (trong bao gm c hai knh ting n v truyn ti in nng), c hai thc hin v truyn ti in nng s dng tr thnh yu t quan trng trong vic ng dng nng lng hn ch (v d, cc thit b chy bng pin). N cng rt quan trng hn ch truyn ti in nng khi cng sut pht ca mt ngi s dng c th tc ng ng k hiu sut ca ngi khc. Trong khi pht trin mt khun kh lthuyt ton din m ha ngun knh lin l mt mc tiu di hn ph hp, n l iu hin nhin t cng vic trc khi hon ton c trng cho s cn bng gia cht lngc lin quan, chm tr, phc tp, s dng knh, v sc mnh v cng kh khn, v nhng tin b t c trong lnh vc ny l rt kh khn

Potential projects in this area include: The development of a theoretical framework for joint source-channel coding, suitable for both point-to-point communications and network communications, that includes the relevant tradeoffs between quality, delay, complexity, channel usage, and power. Such a framework is needed for both lossless and lossy data requirements and for situations where synchronization is an issue. The development of performance measures appropriate for joint source-channel code designand evaluation, possibly including such things as performance bounds or different quality metrics for lossless and lossy coding applications
D n tim nng trong lnh vc ny bao gm: S pht trin ca mt khung l thuyt cho cc knh m ha ngun doanh, thch hp cho c hai im-im truyn thng v truyn thng mng, bao gm s cn bng c lin quangia cht lng, chm tr, phc tp, s dng knh in,. Mt khun kh nh vy l cn thit cho c hai lossless v lossy yu cu d liu, v cho tnh hung m ng b ha l mt vn . S pht trin cc bin php thc hin thch hp cho doanh knh nh gi m ngundesignand, c th bao gm c nhng th chng hn nh gii hn hiu sut hoc s liucht lng khc nhau cho lossless v cc ng dng m ha lossy.

Models Modern communication systems are complex. Analytical research is typically based on the study of relatively simple models (abstracted from physical or

idealized systems). There is great need for the development of tractable source and channel models for use in joint source-channel coding research. The models developed should be rich enough to capture the features and constraints of the problems at hand yet simple enough to provide the insight necessary for algorithm development. Such models could be developed progressively from consideration of simple point-to-point communication systems to more complex multiuser networks and should be driven by corresponding developments in theory
H thng truyn thng hin i rt phc tp. Nghin cu phn tch thng c da trnnghin cu cc m hnh tng i n gin (tm tt t h thng vt l hay l tng ha).C l rt cn thit cho s pht trin ca ngun d lam v cc m hnh knh s dngtrong nghin cu chung m ha ngun knh.Cc m hnh pht trin nn c phong ph, nm bt cc tnh nng v hn ch trong nhng vn bn tay n gin, cung cp ci nhn su sc cn thit cho s pht trin thut ton. Cc m hnh nh vy c thc pht trin dn dn t xem xt cc h thng thng tin lin lc im-im n gin vi cc mng a ngi dng phc tp hn v cn c thc y bi s pht trin tng ngtrong l thuyt

Models needed include: Models for speech, video, images, audio, and text that can be used to devise better source codes; Models for the output of source encoders for use in the design of channel codes (e.g., for unequal error protection); Models for wireless channels that capture the effects of mobility and multiple access; Models for Internet traffic that can be used when devising end-to-end protocols
M hnh cn thit bao gm: M hnh bi pht biu, video, hnh nh, m thanh, vn bn c th c s dng a racc m ngun tt hn; M hnh cho u ra ca b m ha m ngun s dng trong vic thit k m s knh (v d, bo v li bt bnh ng); M hnh cho cc knh khng dy nm bt c tc ng ca di ng v truy cp nhiu; M hnh cho lu lng truy cp Internet c th c s dng khi t ra cc giao thcend-toend

Design Systematic design methodologies are required to broaden the applicability of joint source-channel coding techniques. Potential projects include the development of design methodologies that addressing the following problems:
Thit k Phng php thit k h thng cn thit m rng kh nng p dng ca knh ngundoanh k thut m ha. D n tim nng bao gm s pht trin ca phng php thit k gii quyt cc vn sau y:

Channel-optimized source coding, including error concealment and recovery methods; Source-optimized channel coding, including unequal error protection methods; Synchronization; Feedback in source-channel coding; Coding for time-varying sources or channels; Multiple description coding; Robustness to model variations; Flexibility to different data types.
Knh ti u ha m ngun, bao gm che giu li v phng php phc hi; Ngun-ti u ha knh m ha, bao gm cc phng php bo v li khng ng u; ng b; Thng tin phn hi trong knh m ha ngun; M ha cho cc ngun hoc cc knh thay i theo thi gian; Nhiu m t m ha; Mnh m bin i m hnh; Tnh linh hot cc d liu khc nhau.

Applications Joint source-channel coding theory and design techniques should be applied to address practical problems, including the design of joint source-channel codes for multiuser or network systems. One challenge to the accomplishment of this goal is the fact that network protocols attempt to isolate link and applications layers a separation-style approach. In contrast, joint source-channel coding inherently involves interaction between the link and applications layers. A challenge of joint source-channel coding is to find ways for the application layer to crack through the network stack all the way to the link layer. Another challenge is to use joint source-channel coding research to inform protocol design. Specific projects include
Phn ngun knh m ha l thuyt v k thut thit k nn c p dng gii quyt vn thc t, bao gm c vic thit k cc knh m ngun chung cho cc h thng a ngi dng hoc mng. Mt thch thc hon thnh mc tiu ny l mt thc t rng cc giao thc mng c gng c lp cc lp lin kt v ng dng - mt cch tip cn theo phong cch tch bit. Ngc li, m ha knh chung ngun vn lin quan n s tng tc gia cc lin kt v cc ng dng lp.Mt thch thc ca knh m ha ngun doanh l tm cch cho lp ng dng crack thng qua mng li cc ngn xp tt c cc cch cclp lin kt. Mt thch thc khc l s dng chung ngun knh nghin cu m ha thng bo cho thit k giao thc. D n c th bao gm

Evaluation of the effects of network protocols, including the need for networks to handle a variety of data sources;

Evaluation of the effects of packetization and routing; Re-examination of the typical network separation of channel coding into the physical layer and source coding into the application layer; Application-specific needs for asymmetry in encoding and decoding complexity; Power allocation in shared-resource environments

IV Roles of Academia and Industry

Little is currently understood theoretically about joint source-channel coding, and advances will require contributions from both industry and academia. It is therefore of interest to determine how best to encourage productive interaction between industry and academia in the context of joint source-channel coding. Several characteristics distinguish this topic, and may have bearing on interactions between industry and academia. Foremost, the problem area is of great immediate interest to both academicians and members of industry. In addition, joint sourcechannel coding requires systems-level design and optimization, which industry is perhaps more inclined towards than academia. Further, it is an area in which applications are expected to strongly motivate and inform the theory, and viceversa. Finally, it is an area where fast-paced development outstrips theoretical understanding
t hiu v mt l thuyt v cc knh m ha ngun doanh, v tin b s yu cu ng gp t c hai ngnh cng nghip v cc hc vin. Do , quan tm xc nh cch tt nht khuyn khch s tng tc sn xut gia cng nghip v cc hc vin trong bi cnh ccknh m ha ngun doanh. Mt s c im phn bit ch ny, v c th mang vtng tc gia cc ngnh cng nghip v cc hc vin. Trc tin, cc vn khu vc lrt ln ngay lp tc quan tm n c hai vin s hn lm v cc thnh vin ca ngnh cng nghip. Ngoi ra,ngun doanh-knh m ha i hi h thng thit k v ti u ha, ngnh cng nghip c l nhiu hn nghing v pha hn so vi cc hc vin. Hn na, n l mt khu vc ng dngd kin s thc y mnh m v thng bo l thuyt, v ngc li. Cui cng, n l mt khuvc ni nhp pht trin nhanh chng vt xa s hiu bit l thuyt

The emphasis and role of research in this area is likely to be different for industry and academia. The pace of development in the wireless and Internet industries, as well as the competitive climate in these areas, has largely necessitated the allocation of research resources on the part of industry to shorter-term goals. However, there are important, fundamental, longer-term research topics that need to be pursued if the technology is to advance to its potential. As academics take up more of the fundamental research agenda, it is important that they work closely with industry for a number of reasons. First, close ties between industry and academia allow for fundamental research based on practical insights into where the

most important research problems lie. Second, these ties allow for design choices informed by an understanding of their theoretical implications
Nhn mnh vai tr ca nghin cu trong lnh vc ny l c kh nng th khc nhau cho ngnh cng nghip v cc hc vin. Tc pht trin cc ngnh cng nghip khng dy vInternet, cng nh kh hu cnh tranh trong cc lnh vc ny, phn ln i hi phi phn bcc ngun lc nghin cu trn mt phn ca ngnh cng nghip mc tiu ngn hn. Tuy nhin, c quan trng, c bn, di hn ti nghin cu cn phi c theo ui cng ngh ny l thc y tim nng ca n. Nh cc hc gi ca chng trnh ngh snghin cu c bn, iu quan trng rng h lm vic cht ch vi cc ngnh cng nghipcho mt s l do. u tin, mi quan h cht ch gia cng nghip v cc hc vin chophp nghin cu c bn da trn nhng hiu bit thc t vo cc vn nghin cu quan trng nht nm. Th hai, cc mi quan h ny cho php la chn thit k thng bo mt s hiu bit v ngha l thuyt ca h

Areas of interaction The time from algorithm development to implementation is rapidly decreasing due to advances in processor speeds and memory as well as feature size reduction in VLSI. It is therefore becoming possible to devise sophisticated algorithms for endto-end optimization, whether at the level of a single wireless link, or of an end-toend Internet protocol. As discussed earlier, the development of good source and channel models is critical to such end-to-end optimizations. Given the wealth of real-world experience available in industry and the traditional academic strengths in abstract reasoning, model development is a prime area for industry-academic collaboration. Another broad area for industry-academic cooperation is in the formulation of design criteria. Since the objective of joint source-channel coding is to perform end-to-end optimization, an understanding of the constraints facing the engineers who will ultimately build the system is important even in the earliest phases of algorithm development. For example, the need for low power consumption in small hand-held devices implies certain limitations on the complexity of transmitter and receiver algorithms in such devices. On the other hand, advances in VLSI may temper such limitations in the long term. Such rapid changes in technology make close interaction between academia and industry a must in formulating meaningful problems. Another example pointed out at the workshop was that, from an overall systems perspective, it is often preferable to provide a flexible framework that allows for system-wide optimization, rather than over-optimizing for a specific source or channel model. Multimedia transmission over heterogeneous networks is an example of a broad class of applications that could benefit from such a flexible framework. Industry-academic interactions allows such tradeoffs between flexibility and optimality to be made in an informed fashion. Finally, while the development of fundamental theory and algorithms is, of necessity, based on

somewhat idealized models, simulation and testing in a realistic environment is an area in which industry has substantially more resources than are typically available in university laboratories. Such performance evaluation is therefore a major area of potential collaboration between industry and academia, especially because it is the first step to potential technology transfer. An example of this is the implementation and evaluation of joint source-channel codes in a wireless mobile environment, which would not be possible to recreate at a large scale in a university environment
Cc khu vc ca s tng tc Thi gian t pht trin thut ton thc hin nhanh chng gim do nhng tin b trong tc x l v b nh cng nh tnh nng gim kch thc trong VLSI. Do , c th a ra nhng thut ton phc tp ti u ha end-to-end, cho d cp ca mt lin kt khng dy duy nht, hoc ca mt giao thc Internet end-to-end. Nh tho lun trc , s pht trin ca m ngun v cc m hnh knh l rt quan trng ti u ha nh end-to-end. Vi s giu c ca kinh nghim thc t trong ngnh cng nghip v cc im mnh truyn thng hc tp l lun tru tng, m hnh pht trin l mt khu vc chnh cho ngnh cng nghip-hc hp tc. Mt khu vc rng ln cho ngnh cng nghip-hc hp tc trong vic xy dng cc tiu chun thit k. K t khi mc tiu ca knh m ha ngun doanh l thc hin end-to-end ti u ha, hiu bit v cc kh khn phi i mt vi cc k s ngi cui cng s xy dng h thng quan trng l ngay c trong nhng giai on u tin ca pht trin thut ton. V d, nhu cu tiu th in nng thp trong cc thit b cm tay nh ng nhng hn ch nht nh vo s phc tp ca thut ton truyn v nhn trong cc thit b nh vy. Mt khc, tin b trong VLSI c th bnh tnh hn ch nh vy trong di hn. Nhng thay i nhanh chng trong cng ngh tng tc cht ch gia hc vin v ngnh cng nghip phi xy dng cc vn c ngha. Mt v d khc ch ra ti hi tho rng, t mt quan im v h thng tng th, n thng thch hp hn cung cp mt khun kh linh hot cho php ti u ha h thng rng, ch khng phi l ti u ha cho mt ngun c th hoc m hnh knh. a phng tin truyn ti trn cc mng khng ng nht l mt v d ca mt lp rng cc ng dng c th hng li t mt khun kh linh hot nh vy. Cng nghip-hc tp tng tc cho php cn bng gia tnh linh hot v ti u c thc hin trong mt thi trang thng bo. Cui cng, trong khi s pht trin ca l thuyt c bn v cc thut ton l cn thit, da trn cc m hnh phn no l tng ha, m phng v th nghim trong mt mi trng thc t l mt khu vc trong cng nghip c ngun ti nguyn ng k hn so vi thng c sn trong phng th nghim trng i hc. nh gi hiu sut nh vy l do mt khu vc chnh ca s hp tc tim nng gia ngnh cng nghip v cc hc vin, c bit l bi v n l bc u tin chuyn giao cng ngh tim nng. Mt v d ca vic ny l vic thc hin v nh gi ca cc knh m ngun chung trong mt mi trng in thoi di ng khng dy, trong s khng th ti to quy m ln trong mi trng i hc

Modes of Interaction There are many fruitful ways in which academia and industry can interact. Perhaps the most obvious is through joint research projects. These can take the form of true collaborations between engineers from both organizations, but more often consist of academia performing research and reporting results to industry. This approach is often a result of the need for industry to remain silent about research approaches that they believe will be important to short term competitiveness. Indeed, even the

knowledge of which problems need attention in the short term can represent a competitive advantage
C rt nhiu cch hiu qu trong hc vin v ngnh cng nghip c th tng tc. C l r rng nht l thng qua cc d n nghin cu chung. y c th mang hnh thc ca hp tc thc s gia cc k s t cc t chc, nhng thng xuyn hn bao gm cc hc vinthc hin nghin cu v bo co kt qu cho ngnh cng nghip. Cch tip cn nythng l kt qu ca s cn thit cho ngnh cng nghip gi im lng v phng php nghin cu m h tin rng s l quan trng n kh nng cnh tranh ngn hn. Tht vy,ngay c nhng kin thc trong vn cn c quan tm trong ngn hn c th i din cho mt li th cnh tranh

Perhaps the most serious impediment to joint research projects between academia and industry is the intellectual property (IP) issue. The usual opening stance of both organizations is sole ownership of all IP, with no rights for the other party without additional licensing fees. Much attention has been focused on large efforts involving centers, with relatively less attention for smaller (e.g., single investigator) projects. Some progress has been made on both fronts, and IP agreements have been reached in many cases. Nevertheless, much work remains to resolve this thorny issue.
C l tr ngi ln nht i vi cc d n nghin cu chung gia cc hc vin v ngnh cng nghip l s hu tr tu (IP) vn . Lp trng m thng thng ca c hai t chcny l duy nht quyn s hu ca tt c cc IP, khng c quyn cho bn kia m khng cnl ph cp giy php b sung. Nhiu s ch tp trung vo nhng n lc ln lin quann trung tm, vi s quan tm tng i t hn cho nh hn (v d, iu tra vin duy nht)cc d n. Mt s tin b c thc hin trn c hai mt trn, v cc tha thun IP t c trong nhiu trng hp. Tuy nhin, vn cn nhiu vic gii quyt vn gai gc.

Other modes of interaction between academia and industry exist and can be quite fruitful. Many of these involve exchange or visits of personnel. Internships for students or faculty in industry can provide perspective, direction, and research ideas to be followed upon return to academia. Likewise, visits by industrial researchers to academic labs can benefit the university lab as well as the industrial visitor. While relatively uncommon in U.S. industry, Japanese and Korean engineers often visit university research labs. While similar to internships (although usually more focused and shorter term), faculty consulting can provide similar benefits to both academia and industry. On the other side of the equation, industrial practitioners frequently teach courses at universities, and can co-adviser for theses and dissertations. One final and highly successful mode of academic/university interaction is through entrepreneurship and start-up companies
Phng thc khc ca s tng tc gia hc vin v ngnh cng nghip tn ti v c th c kh hiu qu. Nhiu ngi trong s ny lin quan n trao i v thm ving canhn vin.Thc tp cho sinh vin hoc ging vin trong ngnh cng nghip c th cungcp quan im, nh hng v nghin cu tng sau khi tr li hc vin. Tng t nh vy, chuyn

thm ca cc nh nghin cu cng nghip phng th nghim hc tp c thmang li li ch cho phng th nghim trng i hc cng l ngi truy cp cng nghip. Trong khi tng i ph bin trong ngnh cng nghip M,cc k s Nht Bn v Hn Quc thng xuyn gh thm cc phng th nghim trng i hc nghin cu. Trong khi tng t nh thc tp (mc d thut ng thng tp trung hn v ngn hn), ging vin t vn c th cung cp cc li ch tng t cho c hai hc vin v ngnh cng nghip. pha bn kia ca phng trnh, cc hc vin cng nghipthng xuyn ging dy cc kha hc ti cc trng i hc, v c th c vn cho cclun n v lun vn. Mt ch cui cng v rt thnh cng ca hc / trng i hctng tc l thng qua tinh thn kinh doanh v bt u cc cng ty

There are a variety of ways in which the NSF fosters cooperation between academia and industry. Perhaps the most direct is via matching funds that leverage industrial contributions. Currently, most of the matching is done via large NSF supported centers (such as the ERC and UICRC). We encourage funding for smaller scale programs (e.g., individual PI) such as the GOALI and CAREER programs. Programs that sponsor industrial implementation of research (e.g., SBIR) help spur innovation and move ideas from academia to industry and should be encouraged as well
C rt nhiu cch khc nhau trong NSF thc y hp tc gia hc vin v ngnh cng nghip. C l trc tip nht l thng qua ph hp vi qu tn dng cc khon ng gpcng nghip. Hin nay, hu ht ph hp vi thc hin thng qua ln NSF h tr cc trung tm (nh ERC v UICRC). Chng ti khuyn khch ti tr cho chng trnh quy m nh hn(v d, c nhn PI) nh GOALI v S NGHIP chng trnh. Cc chng trnh ti tr thc hin cng nghip ca nghin cu (v d, SBIR)gip thc y cc tng i mi v di chuyn t gii hc vin cho ngnh cng nghip v cn c khuyn khch nh

On a more intellectual side, the NSF in cooperation with joint industrial and academic panels can identify areas of mutual interest to both communities and promote them via calls for proposals. Finally, to ameliorate the intellectual property impediments mentioned earlier, the NSF can help establish general guidelines for IPR agreements. This could be done either by publicizing existing standards or conventions, or by convening industrial and academic representatives to devise a general framework, or even model agreements, for the sharing of intellectual property rights
Trn mt kha cnh tr tu, NSF hp tc vi cc tm doanh cng nghip v hc thut c thxc nh cc lnh vc hai bn cng quan tm cho c cng ng v thc y h thng qua cc cuc gi cho cc xut. Cui cng, ci thin cc tr ngi s hu tr tu cp trc , NSF c th gip thit lp cc hng dn chung cho cc tha thun quyn s hu tr tu. iu ny c thc thc hin bng cch cng b cc tiu chun hin c hoc quy c, hoc bng cchtriu tp cc i din cng nghip v hc thut a ra mt khun kh chung, hoc thmch l tha thun m hnh cho vic chia s quyn s hu tr tu

Concluding Remarks

This report describes the findings of the 1999 NSF SponsoredWorkshop on Joint Source-Channel Coding Research. The report and its recommendations address three main goals. First, the report provides a view of the conditions and scenarios in which joint source-channel coding can potentially achieve performance improvements over independent source and channel code designs. These conditions include resource-constraints, multiuser interactions, heterogeneity, and time-variation or uncertainty. Second the report describes key research goals and challenges in the field of joint source-channel coding. This summary includes specific suggestions for research leading to advancement of theory, design, and application of joint source-channel codes. Third, the report comments on the roles of academia and industry in meeting the research challenges discussed. In particular, the report suggests specific research problems for which collaboration could be particularly beneficial and suggests methods for encouraging productive interactions between companies and universities in tackling this problem of great interest to both
Bo co ny m t nhng pht hin ca NSF SponsoredWorkshop nm 1999 v nghin cu chung M Ngun-Channel. Bo co v kin ngh ca a ch ba mc tiu chnh. u tin, bo co cung cp mt ci nhn trong cc iu kin v tnh hung trong cc knhm ha ngun doanh c kh nng c th t c nhng ci tin hiu sut trn m ngunc lp v thit k m knh. Nhng iu kin ny bao gm cc ti nguyn hn ch, tng tc a ngi dng, tnh khng ng nht, v thi gian bin i hoc khng chc chn. Bo co th hai m t mc tiu nghin cuquan trng v thch thc trong lnh vc m ha ca doanh ngun knh. Bn tm tt nybao gm cc gi c th cho nghin cu hng u tin b ca l thuyt, thit k, v p dng cc knh m ngun chung. Th ba, cc kin bo co v vai tr ca cc hc vin vngnh cng nghip trong vic p ng nghin cu nhng thch thc tho lun. c bit,bo co cho thy cc vn nghin cu c th cho s hp tc c th c c bit c li v xut cc phng php khuyn khch tng tc sn xut gia cng ty vtrng i hc trong vic gii quyt vn ny quan tm rt ln cho c hai