Optimal Resource Allocation for. Communication Networks

Transcription

1 Optma Resource Aocaton for deo Streamng over Dstrbuted Communcaton Networks ng Guan Ryerson Mutmeda Researc aboratory & Centre for Interactve Mutmeda Informaton Mnng Department of Eectrca and Computer Engneerng Ryerson Unversty t Toronto Canada guan@ee.ryerson.ca ttp:// /7/9

2 Acknowedgment Te presenter woud ke to tank Dr. Yfeng He for s persstent effort n makng ts researc a true success Te presenter aso woud ke to tank Dr. Ivan ee for s contnuous contrbutons to ts work Ts researc s supported by Te Canada Researc Car (CRC) Program Canada Foundaton for Innovatons (CFI) Te Ontaro Innovaton Trust (OIT) and Ryerson Unversty /7/9

3 Major Pubcatons Y. He I. ee and. Guan "Dstrbuted trougput optmzaton n PP od systems to appear n IEEE Transactons on Mutmeda. Y. He I. ee and. Guan Dstrbuted agortms for network fetme mamzaton n wreess vsua sensor networks IEEE Transactons on Crcuts and Systems for deo Tecnoogy (subject to mnor revson). Y. He I. ee and. Guan Optmzed vdeo mutcast n wreess ad oc network usng network codng to appear n IEEE Transactons on Crcuts and Systems for deo Tecnoogy. Y. He G. Sen Y. Xong and. Guan Optma prefetcng sceme n PP od appcatons wt guded seeks accepted by IEEE Transactons on Mutmeda Y. He and. Guan Optma resource aocaton n dstrbuted vsua communcatons n Integent Mutmeda Communcaton: Tecnques and Appcatons C.W. Cen Z. and S. an eds Sprnger-erag to be pubsed n 9. /7/9 3

4 Outne Motvaton and contrbutons Prncpes of conve optmzaton Resource aocaton n dstrbuted vdeo communcaton systems Trougput mamzaton aton n PP od appcatons Network fetme mamzaton n wreess vsua sensor networks Optmzaton for vdeo streamng over wreess ad oc networks Concusons /7/9 4

5 Motvaton Many mutmeda eda appcatons nvove rea- tme vdeo transmssons over dstrbuted networks. Some eampes: PP od appcatons deo streamng over wreess ad oc networks Wreess vsua sensor networks Dstrbuted agortms to optmze resource aocatons n dstrbuted networks Eac node as oca knowedge No centrazed controer Scaabty /7/9 5

6 Caenges PP od appcatons mted bandwdt Unreabe and dynamc peers Random seeks deo streamng over wreess ad oc networks Hg transmsson error rate Source rate aocaton Optmzed routng sceme Wreess vsua sensor networks deo compresson consumng a arge amount of power Trade-off between network fe tme and vdeo quaty /7/9 6

7 Contrbutons Formuate resource aocaton o probems n dstrbuted networks based on conve optmzaton and Sove tem usng dstrbuted agortms Trougput mamzaton n PP od appcatons Optmzed vdeo streamng over wreess ad oc networks Network fetme mamzaton n wreess vsua sensor networks /7/9 7

8 Outne Motvaton and contrbutons Prncpes of conve optmzaton Resource aocaton n dstrbuted vdeo communcaton systems Trougput mamzaton aton n PP od appcatons Network fetme mamzaton n wreess vsua sensor networks Optmzaton for vdeo streamng over wreess ad oc networks Concusons /7/9 8

9 Conve optmzaton probem Convety y s often vewed as te watersed between easy and ard optmzaton probems. Conve optmzaton: te prma probem [Boyd4] mnmze : w Subject to : w... m q... p n were R s optmzaton varabe w q Souton: s affne functon. w are conve functons Dstrbuted agortm: agrange duaty propertes construct a dua probem dua decomposton sove dua probem wt subgradent metod obtan prma optmzaton varabe from dua varabes /7/9 9

10 Conve functon and affne functon Conve functon: q Wat s affne functon? Defnton: et q and q y R f we ave q y Ten q q s affne functon. near functon s aways affne. [Boyd4] /7/9

11 Dua souton Construct a dua probem [Boyd4] Te agrangan: v λ p m q v w w varabes. are dua wer v λ Te dua functon: p m q v w w g mn mn v λ v λ Te dua functon: q v w w g mn mn v λ v λ X X Te dua functon s aways concave Te mnmum of severa near functons s aways concave Te dua probem: Subject to : mamze: λ v λ g Te dua functon s aways concave. D bj t f t : Subject to λ Dua objectve functon v λ X mn w q v w w q v w w g p m p m /7/9 Dua objectve vaue s no arger tan prma objectve vaue.

12 Dua souton Duaty gap: * * d w - g g w * * s te mamadua objectve vaue g(uv) s te mnma prma objectve vaue w( ) w Strong duaty: under Sater's condton duaty gap gpd tat s : w Sater s condton: Tere ests a tat satsfes: w... m and q... p In oter word tere est a strcty feasbe pont * g * * g * Weak duaty: d > /7/9

13 Optmaty Objectve vaue of prma probem Objectve vaue of dua probem Duaty gap= Obtan mnma prma objectve and mama dua objectve at te same tme Optma prma varabes Optma dua varabes /7/9 3

14 Subgradent A subgradent of functon f at pont s any vector g tat satsfes te nequaty: f(z)>=f()+g T (z-) for a z f s a conve functon [Boyd4] /7/9 4

15 Subgradent Subgradent metod to update dua varabes [Bertsekas3] k k k ma k k k k v v f... p were f k... m k k k k f k * k k w s te subgradent of λ v... m * k k λ v... p k q -g at v Non-summabe dmnsng Step sze for convergence: k k k m k k respectvey * Fnd optma prma varabes from dua varabes: λ v arg nf λ v /7/9 5

16 Eampe: conve optmzaton mnmze : Sbj Subject to : w [ 5]. w objectve vaue Prma Optma prma varabe=.56 Dua objectve vaue4 Optma dua varabe= Prma varabe Prma Objectve aue Dua varabe Dua Objectve aue /7/9 6

17 Reguarzaton term makes dua functon dfferentabe mnmze : Subject to :-. agrange dua functon: mnmze : Subject to :- g( ) - functon vaue Non-dfferentabe.5.5. agrange dua functon: g( ) - agrange dua.5 Dfferentabe = =.5 =. = /7/9 7

18 Dua decomposton Wy sove te prma probem by frst sovng te dua probem?. Dua probem s aways conve can be effcenty soved.. Dua probem can be decomposed easy eadng to dstrbuted agortm Dua D decomposton: ( to obtan dstrbuted b t d agortm) Prma probem: mnmze : Subject to : f f c Dua probem: f f c f nf f mamze : nf nf Subject to : λ c Dua varabes update: k k k ma c /7/9 8

19 Prma decomposton Prma probem: mnmze : Subject to : f f c ma mze α : f * f * Master probem mnmze : Subject to : f mnmze : Subject to : f c Subprobem Subprobem /7/9 9

20 Dstrbuted souton Probems for vdeo streamng over dstrbuted networks Formuaton Orgna optmzaton probem (prma probem) Dua probem Update dua varabes Subprobem Subprobem N Compute prma varabe at node Compute prma varabe N at node N /7/9

21 Outne Motvaton and contrbutons Prncpes of conve optmzaton Resource aocaton n dstrbuted vdeo communcaton systems Trougput mamzaton aton n PP od appcatons Network fetme mamzaton n wreess vsua sensor networks Optmzaton for vdeo streamng over wreess ad oc networks Concusons /7/9

22 Outne Motvaton and contrbutons Prncpes of conve optmzaton Resource aocaton n dstrbuted vdeo communcaton systems Trougput mamzaton aton n PP od appcatons Network fetme mamzaton n wreess vsua sensor networks Optmzaton for vdeo streamng over wreess ad oc networks Concusons /7/9

23 Cent/server od vs. PP od Cent/server od Server bandwdt botteneck PP appcatons PP od PP ve T Snge-ayer coded PP od Eac user receves te same quaty Server Peer Cent/server Source node Scaabe PP od ower bandwdt ower quaty Hg bandwdt ger quaty PP Peer /7/9 3

24 Reated work PP od appcatons Estng PP arctectures Buffer-forwardng arctectures: Tree-based: PoD [Do4] Mes-based [6] Storage-forwardng arctecture: Mes [Yu7] Hybrd-forwardng arctecture [[YHe-ICASSP8YHe-TMM8a] Estng optmzatons n PP Mnmum-Deay for constant-bt-rate (CBR) PP meda sesson [Wu5] Dstrbuted aucton agortm for rate aocaton [6] Mamzaton of trougput n scaabe PP od systems takng nto account packet oss due to ecessve deay at eac nk [YHe- ICME7a] Prefetcng n PP [YHe-TMM8b] /7/9 4

25 Scaabe od: prortzed codng sceme Prortzed codng sceme Orgnay proposed n [Cou3] usefu for vdeo broadcast ayered codng + prortzed packetzaton + network codng (at source and ntermedate nodes) Advantages: Scaabe Resent to packet oss Dupcate-free A arger trougput at a recever eads to a ger quaty /7/9 5

26 Buffer-forwardng overay for PP od Estng approaces to construct buffer-forwardng overay Tree-based Mes-based Our work s on te modue of trougput mamzaton Peer Modue of buffer-forwardng Modue of trougput Aocate rate at overay constructon mamzaton eac outgong nk /7/9 6

27 Grap mode p A network can be modeed as a drected grap G=(N) N s te set of nodes t t f k s te set of nks Matr to represent te node-nk reatonsp A: te reatonsp between te node and ts connected nks A + A Node A + : te reatonsp between te node and ts outgong nks A t t b t t A A - : te reatonsp between te node and ts ncomng nks A= A + A A A network /7/9 7 A= A + - A

28 Grap mode Network grap A PP s modeed by a drected grap G = (N ) Matr A: a f nk s an outgong nk from node f nk s an ncomng nk nto node oterwse. Matr A + : a f nk s an outgong nk from node oterwse. Matr A - : a f nk s an ncomng nk nto node oterwse. /7/9 8

29 Trougput mamzaton n bufferforwardng systems Probem P formuaton: mamze Aggregate aa trougput p p N g mnmze subject to N a a a m sr I O c m m. Aggregate trougput N Source rate constrant N Downoad bandwdt constrant t N Upoad bandwdt constrant nk-forwardng constrant Converted Orgna to: optmzaton probem: Strcty near conve Programmng optmzaton (P) probem probem dstrbuted agortm to sove t /7/9 9

30 Trougput mamzaton n buffer- forwardng systems (cont.) Smuaton resuts:.5 Centrazed P Dstrbuted optmzaton Proportona aocaton Equa aocaton Acevabe trougput n buffer-forwardng arctecture Average trougput [M Mbps] Any better arctecture to mprove te acevabe trougput? t? Number of peers Comparson of average trougput t /7/9 3

31 Impact of reguarzaton factor 4.4 Number of ter ratons for conve ergence trougput [Mbp ps] Average t Reguaton factor Reguaton factor Compety Sub-optmaty Trougput mamzaton n buffer-forwardng PP od /7/9 3

32 Impact of step sze to convergence speed Non-summabe dmnsng step sze sequence: k k nk rate.6 =. = Iteraton No. Iteraton of a nk rate n a -peer PP od buffer-forwardng system /7/9 3

33 Dynamcs andng n PP od u X= 3 4 X3=3 X4= u Convergence speed durng transton X=5 Steady state (fow conservaton): =+3 Dua varabe: u=.4 u=.5 u3=.3 Peer 3 eft peer 4 joned Update dua varabes: u=.6 u=.3 u4=. Update prma varabes: =4 =3 4= Reac new steady state: =+4 on vaue Dua funct No. of teraton Dynamc -peer and 4-parent/peer scenaro peers eft and new peers joned n te prevous tme sot /7/9 33

34 Communcaton overead for trougput mamzaton probem n buffer-forwardng systems 3 4 nk nk nk 3 Te dua varabes at node : Te dua varabes at nk : u v Te prma varabe (nk rate) at nk : Eac node s responsbe for updatng te prma and dua varabes at ts node and at ts outgong nks Node computes te nk rate at nk 3: 3 p 3 u v 3 Communcaton overead: node needs to request u from node request λ from node 3 and λ from node 4 Te update of dua varabes requres te nk rates of connected nks no communcaton overead u : ncomng nk rates v : outgong nk rates λ : ncomng nk rates /7/9 34

35 A Hybrd-forwardng Arctecture for PP od Server deo- buffer-forwardng forwardng overay 75 Ide peers deo- buffer-forwardng overay Hybrd-forwardng arctecture Buffer-forwardng nk Storage-forwardng nk Sare bot buffer and storage to mprove trougput Peers contrbute ter stored segments to oter peers Impement servce dfferentaton among vdeos Stored segments are stabe robust to peer dynamcs /7/9 35

36 Buffer-forwardng and storage-forwardng g overay constructon Buffer Source node Source node Stored segment Requestng segment Peer 4 5 Payback progress Buffer-forwardng overay Stored segment 4 3 Requestng segment 5 Stored segment Requestng segment Storage-forwardng overay /7/9 36

37 Trougput mamzaton n ybrd- P b f t g p y forwardng arctecture Probem formuaton: Aggregate wegted trougput mamze p a N Aggregate trougput N p a mnmze Source rate constrant Downoad bandwdt constrant subject to N I a N s a r Upoad bandwdt constrant Buffer-forwardng constrant c N O d B m m Buffer forwardng constrant F S N m m m Storage-forwardng constrant Orgna optmzaton probem: near Programmng (P) probem Converted to: Strcty conve optmzaton probem dstrbuted agortm to sove t. /7/9 37 near Programmng (P) probem Strcty conve optmzaton probem dstrbuted agortm to sove t

38 Trougput mamzaton n ybrd- forwardng arctecture (cont.) Smuaton resuts: Hybrd-forwardng centrazed Hybrd-forwardng dstrbuted Buffer-forwardng dstrbuted Storage-forwardng dstrbuted Avera age trougpu ut [Mbps] Number of peers Comparson of average trougput /7/9 38

39 Oter observaton Number of Peers Smuaton resuts (Posson dstrbuton): Buffer forwardng Hybrd forwardng /7/9 39

40 Oter observaton Scaabty Smuaton resuts: Comparson of te cost ntroduced by te proposed dstrbuted agortm n buffer-forwardng forwardng PP od systems wt dfferent network szes: (a) te number of te overay nks (b) te average number of teratons per nk and (c) te average communcaton overead per node /7/9 4

41 Outne Motvaton and contrbutons Prncpes of conve optmzaton Resource aocaton n dstrbuted vdeo communcaton systems Trougput mamzaton aton n PP od appcatons Network fetme mamzaton n wreess vsua sensor networks Optmzaton for vdeo streamng over wreess ad oc networks Concusons /7/9 4

42 Wreess vsua sensor networks Appcatons of WSN deo sensor deo surveance envronmenta trackng Te proposed dstrbuted agortm: Mamze te network fetme by jonty optmzng ng source rates encodng powers and routng sceme Tota power consumpton at a Snk node Encodng power + transmsson power + recepton power Network fetme: defned as mnmum node fetme A wreess vsua sensor network /7/9 4

43 Reated work Network fetme mamzaton for Wreess vsua sensor networks (WSNs) Conventona wreess sensor networks Coect data (e.g. temperature) neggbe power consumpton on sgna processng at sensor node Estng dstrbuted optmzatons for conventona wreess sensor network Tradeoff between te source rate aocaton and te network fetme [Nama6] Dstrbuted agortm to mamze fetme [Madan6] Tese metods cannot be apped drecty to WSNs snce tey omt te processng power consumpton at te sensor nodes Mamzaton of network fetme for WSN by jonty optmzng source rates encodng powers and routng sceme. [YHe-ICME7b YHe- TCST8a] /7/9 43

44 Markov mode and encodng gpower Canne error mode Two-state Markov mode bt b q Average bt error probabty: p q q packet oss rate (PR): p p. p ( ) Power consumpton mode Encodng power consumpton p Power-rate-dstorton mode [He6] b G q q q q d s e / 3 s. P s Under te same encodng power ncreasng rate reducng dstorton Under te same rate ncreasng encodng power reducng dstorton /7/9 44

45 Grap mode Network grap A WSN s modeed by a drected grap G = (N ) Matr A: a f nk s an outgong nk from node f nk s an ncomng nk nto node oterwse. Matr A + : a f nk s an outgong nk from node oterwse. Matr A - : a f nk s an ncomng nk nto node oterwse. Eac sesson foows te aw of fow conservaton: a N /7/9 45

46 Acevabe mamum network fetme Wtout oss Probem formuaton: B T T Network fetme subject to : mn mn mamze N a y a c y c a P T T r s s net Fow conservaton Network fetme 3 /.. D e y P s s Aggregate fow rate deo quaty requrement. P s s Orgna optmzaton probem /7/9 46

47 Acevabe mamum network fetme (cont.) Probem converson: mnmze subject to: q og P s s a y N 3 / D / P / a s s s r c y c a y qb N P s. Fow conservaton Aggregate fow rate deo quaty requrement Power constrant Frst converson: cange te varabe: q=/t net /7/9 47

48 Probem converson n wreess vsua sensor networks (cont.) P q q P: constrants. subject to : mnmze q q 3 q q 3 q 3 q (q q q ) P3: bj t t N mnmze N q (q =q =q 3 ) P4: subject to : mnmze N a q N s / / subject to : /3 P D y N a / / og subject to : /3 s P D y N a s / / og /3 N qb y a c y c a P s P D r s s s q a N B q y a c y c a P N r s s /7/9 48. P s s. P s s

49 Acevabe mamum network fetme (cont.) Probem converson: mnmze s q N Fow conservaton deo quaty requrement / / og subject to: 3 / s P D N a s N q a N q B a c c a P r s s Power constrant Auary varabes are equa. P s N q q a s N Second converson: ntroduce auary varabes: q /7/9 49 y q

50 Acevabe mamum network fetme (cont.) Smuaton resuts Acevabe mamum netwo ork fetme [s] 6 Source-and-Routng Optmzed (Proposed).7 Source-Optmzed Sceme (SOS) Routng-Optmzed Sceme (ROS) Powe wer consumpton n [W] ABC Encodng gp power Transmsson and recepton power Encodng dstorton requrement Network fetme comparson Sensor node No. Power consumpton at eac node /7/9 5

51 Network fetme for arge-deay appcatons arge-deay appcatons: Mamum netw twork fetme [s] 6 Average PR= Average PR=.7 Eampe: vsua data Average PR=.34 Average PR=.64 coecton Retransmssons to recover 9 te corrupted packets 8 Retransmsson consumes 7 etra power reduce 6 network fetme Encodng dstorton requrement Network fetme wt retransmssons /7/9 5

52 Network fetme for sma-deay appcatons Sma-deay appcatons: Eampe: rea-tme traffc montorng FEC to recover te corrupted packets Introduce etra encodng and decodng power consumpton reduce network fetme [s] Mamum network fetme Average PR= Average PR=.7 Average PR=.34 Average PR= Encodng dstorton requrement Network fetme wt FEC /7/9 5

53 Outne Motvaton and contrbutons Prncpes of conve optmzaton Resource aocaton n dstrbuted vdeo communcaton systems Trougput mamzaton aton n PP od appcatons Network fetme mamzaton n wreess vsua sensor networks Optmzaton for vdeo streamng over wreess ad oc networks Concusons /7/9 53

54 Optmzed vdeo uncastng over wreess ad oc networks deo uncastng From a source to a recever Eampe: A group of vstors n a park person A receves vdeo streamng from person B va reays. Te proposed optmzed vdeo uncastng sceme Prortzed codng + network codng Mnmze te dstorton by jonty optmzng te source rate aocaton and te routng sceme Reay node Source Recever deo uncastng /7/9 54

55 Reated work deo streamng over wreess ad oc networks Dstrbuted optmzaton for data communcatons over wreess ad oc networks [Xao4 Cen6] Te optmzatons for data communcatons cannot be apped drecty to rea-tme streamng appcatons Estng optmzatons for vdeo streamng over wreess ad oc networks Mamze te epected vdeo quaty usng genetc agortm [Mao4] (centrazed tus g compety) Optmzaton of resource competton among mutpe uncast vdeo sessons [Zu6] (dstrbuted and ow compety at eac node) Optma resource aocaton for bot vdeo uncast streamng [and vdeo mutcast streamng [YHe-ISCAS6 YHe-PCM7 YHe-TCST8b] over wreess ad oc networks. /7/9 55

56 deo dstorton mode deo dstorton mode Dstorton vs. receved trougput Modeed as: d D R Dstorto on Akyo QCIF sequence d= /(R+.3) Fttng dstorton mode Epermenta data Conve functon Data fttng tecnque to fnd Parameters: D D : Parameter reated to encodng dstorton : Parameters for transmsson dstorton t Trougput [Kbps] /7/9 56

57 Optmzed vdeo uncastng over wreess ad oc networks (cont.) Probem formuaton: deo dstorton Fow conservaton nk capacty constrant 45 Orgna optmzaton probem Converted to: strcty conve optmzaton probem deveop dstrbuted agortm to sove t PSNR [db] Smuaton resuts: Comparson of frame PSNR 5 Proposed routng Congeston-mnmzed routng Doube-dsjont-pat routng Frame No. /7/9 57

58 Optmzed vdeo mutcastng over wreess ad oc networks deo mutcastng: Streamng from a source node to H recevers smutaneousy Wt network codng a mutcast fow = H conceptua uncast sessons [Aswede] Te proposed optmzed mutcastng sceme Mnmze te dstorton by jonty optmzng te source rate aocaton te routng sceme and te power Source node Source node Source node Recever Recever Recever 5 Conceptua uncast (a) sesson Conceptua uncast (b) sesson Mutcast (c) fow Recever /7/9 58

59 Optmzed vdeo mutcastng over wreess ad oc networks Packet oss rate (PR) n wreess ad oc networks oss due to transmsson error (p ) oss due to congeston (p ) PR at nk : p p p C R were Pr obdeay T ep T ep T p p k SIRtGk P G P k /7/9 59

60 Optmzed vdeo mutcastng over wreess ad oc networks (cont.) Probem formuaton: Tota Aggregate dstorton g trougput Fow conservaton Mutcast fow rate y Capacty constrant nk capacty under CDMA Transmt power constrant Orgna Network codng optmzaton emnates probem dupcate packets cange te objectve functon to mamze te aggregate trougput /7/9 6

61 Optmzed vdeo mutcastng over wreess ad oc networks (cont.) Dstrbuted souton usng erarcca dua decompostons Frst-ayer dua decomposton Optmzaton varabes (syp) Network fow varabes (sy) Second-ayer dua decomposton Power varabe (P) Second-ayer decomposton usng game teory Compute nk rate and transmt power for eac outgong nk Node Compute nk rate and transmt power for eac outgong nk Node N /7/9 6

62 Optmzed vdeo mutcastng over wreess ad oc networks (cont.) Smuaton resuts: Unform-power Optmzed Snge-tree Doube-tree Optmzed.3.3 Trougput [M Mbps].5..5 Trougput [M Mbps] Recever ID 3 4 Recever ID Compare to unform-power sceme Compare to tree-based scemes /7/9 6

63 Optmzed vdeo mutcastng over wreess ad oc networks amda 3.5 Power [W] Iteraton No Iteraton No Conceptua source rate [Mbps] Mutcast nk rat te [Mbps] Iteraton No Iteraton No. Optmzaton resuts for a 5-node wreess ad oc network A mutcast fow for a source to 8 recevers /7/9 63

64 References [Bertsekas3] D. P. Bertsekas A. Nedc and A. E. Ozdagar Conve Anayss and Optmzaton Atena Scentfc 3. [Boyd4] S. Boyd and. andenberge Conve Optmzaton Cambrdge Unversty Press 4. [Do4] T. T. Do K. A. Hua and M. A. Tantaou PoD: Provdng Faut Toerant deo-on-demand Streamng n Peer-to-Peer Envronment n Proc. of IEEE ICC vo. 5 no. pp. 9-3 Jan. 4. [6] Z. and A. Maant A Progressve Fow Aucton Approac for ow-cost On-Demand PP Meda Streamng n Proc. of ACM QSne Aug. 6. [Yu7] W. P. Yu X. Jn and S. H. Can Mes: Dstrbuted segment storage for peer-to-peer nteractve vdeo streamng IEEE Journa on Seected Areas n Communcatons vo. 5 no. 9 pp Dec. 7. [YHe-ICASSP8] Y. He I. ee and. Guan Dstrbuted trougput mamzaton n ybrd-forwardng PP od appcatons n Proc. of IEEE Internatona Conference on Acoustcs Speec and Sgna Processng (ICASSP) pp as egas USA Apr 8. [YHe-TMM8a] Y. He I. ee and. Guan "Dstrbuted trougput optmzaton n PP od systems IEEE Transactons on Mutmeda vo. no. 3 pp Apr 9.. [YHe-TMM8b] Y. He G. Sen Y. Xong and. Guan Optma prefetcng sceme n PP od appcatons wt guded seeks IEEE Transactons on Mutmeda vo. no. pp Jan. 9. [Wu5] C. Wu and B. Optma Peer Seecton for Mnmum-Deay Peer-to-Peer Streamng wt Rateess Codes n Proc. of ACM MM pp Nov. 5. [YHe-ICME7a] Y. He I. ee and. Guan Dstrbuted rate aocaton n pp streamng n Proc. of IEEE Internatona Conference on Mutmeda & Epo (ICME) Speca Sesson on PP Mutmeda Content Access and Dstrbuton pp Bejng Cna Juy 7. [Xao4]. Xao M. Joansson and S. Boyd Smutaneous routng and resource aocaton va dua decomposton IEEETransactons on Communcatons vo. 5 no. 7 pp Ju. 4. [Cen6]. Cen S. H. ow M. Cang and J. C. Doye Cross-ayer congeston contro routng and scedung desgn n ad oc wreess networks n Proc. of IEEE INFOCOM pp. -3 Apr. 6. [Mao4] S. Mao X. Ceng Y. T. Hou and H. D. Sera Mutpe descrpton vdeo mutcast n wreess ad oc networks n Proc. of IEEE BROADNETS pp Oct. 4. /7/9 64

65 References (cont.) [Zu6] X. Zu J. P. Sng and B. Grod Jont routng and rate aocaton for mutpe vdeo streams n ad oc wreess networks Journa of Zejang Unversty Scence A vo. 7 no. 5 pp May 6. [YHe-ISCAS6] Y. He I. ee and. Guan Optmzed mut-pat t routng usng dua decomposton for wreess vdeo streamng n Proc. of IEEE Internatona Symposum on Crcuts and Systems (ISCAS) pp New Oreans USA May 7. [YHe-PCM7] Y. He I. ee and. Guan deo mutcast over wreess ad oc networks usng dstrbuted optmzaton n Proc. of Pacfc-Rm Conference on Mutmeda (PCM) pp Hongkong Dec. 7. [Nama6] H. Nama M. Cang and N. Mandayam Utty-fetme trade-off n sef-reguatng wreess sensor networks: A cross-ayer desgn approac n Proc. of IEEE ICC vo. 8 pp Jun. 6. [Madan6] R. Madan S. a Dstrbuted agortms for mamum fetme routng n wreess sensor networks IEEE Transactons on Wreess Communcatons vo. 5 no. 8 pp Aug. 6. [YHe-ICME7b] Y. He I. ee and. Guan Network fetme mamzaton n wreess vsua sensor networks usng a dstrbuted agortm n Proc. of IEEE Internatona Conference on Mutmeda & Epo (ICME) pp Bejng Cna Juy 7. [He6] Z. He and D. Wu Resource aocaton and performance anayss of wreess vdeo sensors IEEE Transactons on Crcuts and Systems for deo Tecnoogy vo. 6 no. 5 pp May 6. [Aswede] R. Aswede N. Ca S.-Y. R. and R. W. Yeung Network nformaton fow IEEE Transactons on Informaton Teory vo. 46 pp. 4-6 Ju.. [YHe-TCST8a]Y. He I. ee and. Guan Dstrbuted agortms for network fetme mamzaton n wreess vsua sensor networks IEEE Transactons on Crcuts and Systems for deo Tecnoogy vo.9 no. 5 pp May 9. [YHe-TCST8b] Y. He I. ee and. Guan Optmzed vdeo mutcast n wreess ad oc network usng network codng IEEE Transactons on Crcuts and Systems for deo Tecnoogy vo. 9 no. 6 pp June 9.. /7/9 65

66 Tank You! /7/9 66