Bottom-Up View: Link Layer Impact. What to adapt, and to what? Power Control Adaptation. Adaptive Techniques for Wireless Ad-Hoc Networks

Transcription

1 EE360: Lecture 9 Outlne Resource Allocton n Ad Hoc ets Announcements Pper summres due next Wednesdy Overvew of resource llocton n d-hoc networs Cross-lyer dptton Dstrbuted power control Jont scheduln nd power control for wreless d hoc networs (Hleh Tbrz) Adptton nd nterference (wdebnd CDMA) Adptton v me theory (Mns Deb) Adptve Technques for Wreless Ad-Hoc etwors etwor s dynmc (lns chne, nodes move round) Adptve technques cn dust to nd explot vrtons Adptvty cn te plce t ll levels of the protocol stc etve nterctons between lyer dptton cn occur Wht to dpt, nd to wht? QoS Adpts to pplcton needs, networ/ln condtons, enery/power constrnts, Routn Adpts to topoloy chnes, ln chnes, user demnds, coneston, Trnsmsson scheme (power, rte, codn, ) Adpts to chnnel, nterference, pplcton requrements, throuhput/dely constrnts, Adptn requres nformton exchne cross lyers nd should hppen on dfferent tme scles Bottom-Up Vew: Ln Impct Connectvty determnes everythn (MAC, routn, etc.) Ln SIR nd the trnsmt/receve strtey determne connectvty Cn chne connectvty v ln dptton Ln lyer technques (MUD, SIC, smrt ntenns) cn mprove MAC nd overll cpcty by reducn nterference Ln lyer technques enble new throuhput/dely trdeoffs Herrchcl codn removes the effect of burstness on throuhput Power control cn be used to meet dely constrnts Power Control Adptton G G P Ech node enertes ndependent dt. Source-destnton prs re chosen t rndom. GP R G P Topoloy s dynmc (ln n G s tme-vryn) Dfferent ln SIRs bsed on chnnel ns G Power control used to mntn tret R vlue P Power Control for Fxed Chnnels Semnl wor by Foschn/Mlnc [1993] Assume ech node hs n SIR constrnt Wrte the set of constrnts n mtrx form F GP R G P 0, G G, Scled Interferer Gn I FP u 0, P 0 γ1η1 γ η u,, G11 G T Scled ose 1

2 Optmlty nd Stblty Then f r F <1 then unque soluton to * -1 P I F u P * s the lobl optml soluton Itertve power control lorthms PP * Centrlze d : P( 1) FP() u Dstrbuted : γ P ( 1) P () R ( ) Wht f the Chnnel s Rndom? Cn defne performnce bsed on dstrbuton of R : Avere SIR Oute Probblty Avere BER The stndrd F-M lorthm overshoots on vere E[lo R ] lo γ ER γ How to defne optmlty f networ s tme-vryn? Cn Consder A ew SIR Constrnt GP R γ η G P EGP γ η GP 0 I F P u F 0, γeg EG 0, Ornl constrnt Multply out nd te expecttons Mtrx form γ η1 u, E[G11] 1 γη, E[G Sme form s SIR constrnt n F-M for fxed chnnels T ] ew Crteron for Optmlty If r F <1 then exsts lobl optml soluton * -1 P I F u For the SIR constrnt E GP γ Eη GP Cn fnd P* n dstrbuted mnner usn stochstc pproxmton (Robbns-Monro) Robbns-Monro lorthm P( 1) P()- Where e s nose term ε Step sze: 0 Under pproprte condtons on n1 P() ε F F() P() u u() * P() P ε n1 2 Admsson Control Wht hppens when new user powers up? More nterference dded to the system The optml power vector wll move System my become nfesble Admsson control obectves Protect current user s wth protecton mrn Reect the new user f the system s unstble Mntn dstrbuted nture of the lorthm Trcn problem, not n equlbrum problem 2

3 Fxed Step Sze Alorthm Propertes Hve non-sttonry equlbr So cnnot llow 0 Step sze: n1 n1 A fxed step sze lorthm wll not convere to the optml power llocton ~ P() P where E P* P ~ O() Ths error s cost of trcn movn tret 2 Exmple:..d. Fdn Chnnel Suppose the networ conssts of 3 nodes Ech ln n the networ s n ndependent exponentl rndom vrble 1 E[G] γ 5 η 1 ote tht r F =.33 so we should expect ths networ to be frly stble Power Control + Power control mpcts multple lyers of the protocol stc Power control ffects nterference/sir, whch other users rect to Useful to combne power control wth other dptve protocols Adptve routn nd/or scheduln (Hleh) Adptve modulton nd codn Adptve retrnsmssons End-to-end QoS Trffc Genertor Recever Multuser Adptton Dt Buffer Source Coder Chnnel Coder Chnnel Cross- Adptton Modultor (Power) Chnnel nterference s responsve to the crosslyer dptton of ech user 3

4 Power Multuser Problem Formulton Optmze cross-lyer dptton n multuser settn Users nterct throuh nterference Cretes Chcen nd E control problem Wnt n optml nd stble equlbrum stte nd dptton for the system of users The ey s to fnd trctble stochstc process to descrbe the nterference Lner Mult-User Recever Assume ech of K mobles s ssned -lenth rndom spredn sequence 1 S V 1,, V Interference term T 2 ( c S ) ( t) z ( t) SIRK (, t) T 2 T 2 ( c c ) ( c S ) ( t) z ( t) The recever c tes dfferent vlues for dfferent structures (MMSE, de-correltor, etc.) Interference Models Jontly model the stte spce of every moble n the system Problem: Stte spce rows exponentlly Assume unresponsve nterference Avods the Chcen nd E control ssue Problem: Unresponsve nterference models provde msledn results Approxmtons use men-feld pproch Model rete behvor s n vere Cn prove ths s optml n some cses CDMA Wdebnd Lmt K= number of users, =spredn n K Let K, nd the system lod Prevous reserch hs proved converence of the SIR n the wdebnd lmt [Tse nd Hnly 1999,2001] Cn pply wdebnd pproxmton to the stochstc process descrbn CDMA system nd the correspondn optml control problem K Hz Optmzton n the Wdebnd Lmt Wnt to fnd optml mult-user cross-lyer dptton for ven performnce metrc, subect to QoS constrnts Approxmte the networ dynmcs wth wdebnd lmt Optmze the control n the wdebnd lmt Chec converence nd unqueness to ensure the soluton s ood pproxmton to fnte bndwdth system Specl cse of usn men feld theorems Defne mtrx Equlbrum n the Wdebnd Lmt For ny K,, the system stte vector K (t) frcton of users n ech stte P K ( t), s the snle user trnston s the In the wdebnd lmt we hve determnstc nonlner dynmcs for the system stte ( t) lm K, ( t) Furthermore P, K nd ( t 1) ( t) P ( t), hs unque fxed pont 4

5 Avere Power Wdebnd Optml Control Problem subect to: ( ) P T mn ( ) r( ), ( ) ( ), ( ) 1, f ( ) Very smlr to the snle user optmzton The non-lner constrnt cn ntroduce snfcnt theoretcl nd computtonl complctons The non-lner prorm s not convex Cn show tht t cn be solved by sequence of lner prorms Exmple: Power Adptton Wth Dedlne Constrned Trffc Assume dedlne senstve dt (100ms) 50 m/h Mcrocell (sme chnnel s before) Mnmze vere trnsmsson power subect to dedlne constrnt Assume we hve mtched flter recever Wht hppens s system lod ncreses? Let number of users per Hz vry between 0 nd 1 Power vs. System Lod vs. Dedlne Constrnt Crosslyer Desn n Ad-Hoc Wreless etwors Infesble Reon Applcton etwor Access Ln Hrdwre Users/Hz Substntl ns n throuhput, effcency, nd end-to-end performnce from cross-lyer desn Hrdwre Ln Access etwor Applcton Crosslyer Desn Dely Constrnts Rte Requrements Enery Constrnts Moblty Optmze nd dpt cross desn lyers Provde robustness to uncertnty Crosslyer Adptton Applcton Desn optmzton crteron Dt prortzton Adptve QoS etwor Adptve routn MAC Access control MUD/nterference cncellton/smrt ntenns Ln Adptve rte, codn, power, frmn, etc. Adptve retrnsmsson/herrchcl codn Ln, MAC, nd networ hve the most obvous syneres, but the pplcton lyer dcttes the optmzton crteron 5

6 Why crosslyer desn? The techncl chllenes of future moble networs cnnot be met wth lyered desn pproch. QoS cnnot be provded unless t s supported cross ll lyers of the networ. The pplcton must dpt to the underlyn chnnel nd networ chrcterstcs. The networ nd ln must dpt to the pplcton requrements Interctons cross networ lyers must be understood nd exploted. Source Adptve Routn Routn estblshes the mechnsm by whch pcet trverses the networ As the networ chnes, the routes should be updted to reflect networ dynmcs Updtn the route cn entl snfcnt overhed. Destnton Route dessemnton Route computed t centrlzed node Most effcent route computton. Cn t dpt to fst topoloy chnes. BW requred to collect nd dessemnte nformton Dstrbuted route computton odes send connectvty nformton to locl nodes. odes determne routes bsed on ths locl nformton. Adpts loclly but not loblly. odes exchne locl routn tbles ode determnes next hop bsed on some metrc. Dels well wth connectvty dynmcs. Routn loops common. Relblty Pcet cnowledements needed My be lost on reverse ln Should netve ACKs be used. Combned ARQ nd codn Retrnsmssons cuse dely Codn my reduce dt rte Blnce my be dptve Hop-by-hop cnowledements Explct cnowledements Echo cnowledements Trnsmtter lstens for forwrded pcet More lely to experence collsons thn short cnowledement. Hop-by-hop or end-to-end or both. MIMO n Ad-Hoc etwors How to use Feedbc n Wreless etwors Antenns cn be used for multplexn, dversty, or nterference cncellton Cncel M-1 nterferers wth M ntenns Wht metrc should be optmzed? Cross- Desn Output feedbc CSI Acnowledements etwor/trffc nformton Somethn else osy/compressed 6

7 Dversty-Multplexn-Dely Trdeoffs for MIMO Multhop etwors wth ARQ Multhop ARQ Protocols ARQ H1 ARQ H2 Multplexn Error Prone Bemformn Low P e Fxed ARQ: fxed wndow sze Mxmum llowed ARQ round for th hop stsfes Adptve ARQ: dptve wndow sze Fxed Bloc Lenth (FBL) (bloc-bsed feedbc, esy synchronzton) L 1 L L MIMO used to ncrese dt rte or robustness Multhop relys used for covere extenson ARQ protocol: Cn be vewed s 1 bt feedbc, or tme dversty, Retrnsmsson cuses dely (cn desn ARQ to control dely) Dversty multplexn (dely) trdeoff - DMT/DMDT Trdeoff between robustness, throuhput, nd dely Bloc 1 ARQ round 1 Bloc 1 ARQ round 2 Bloc 1 ARQ round 3 Bloc 2 ARQ round 1 Vrble Bloc Lenth (VBL) (rel tme feedbc) Bloc 1 ARQ round 1 Bloc 1 ARQ round 2 Recever hs enouh Informton to decode Bloc 1 round 3 Recever hs enouh Informton to decode Bloc 2 ARQ round 1 Bloc 2 ARQ round 2 Bloc 2 ARQ round 2 Asymptotc DMDT Optmlty Dely/Throuhput/Robustness cross Multple Theorem: VBL ARQ cheves optml DMDT n MIMO multhop rely networs n lon-term nd short-term sttc chnnels. Proved by cut-set bound A B An ntutve explnton by stoppn tmes: VBL ARQ hs the smller oute reons mon multhop ARQ protocols Accumlted Informton (FBL) re Short-Term Sttc Chnnel 12 t 1 t Chnnel Use 39 Multple routes throuh the networ cn be used for multplexn or reduced dely/loss Applcton cn use snle-descrpton or multple descrpton codes Cn optmze optml opertn pont for these trdeoffs to mnmze dstorton Cross-lyer protocol desn for rel-tme med Vdeo stremn performnce Loss-reslent source codn nd pcetzton Coneston-dstorton optmzed scheduln Rte-dstorton premble Applcton lyer Trnsport lyer s 5 db Trffc flows Coneston-dstorton optmzed routn etwor lyer Ln stte nformton Jont wth T. Yoo, E. Setton, X. Zhu, nd B. Grod Cpcty ssnment for multple servce clsses Adptve ln lyer technques Ln cpctes MAC lyer Ln lyer fold ncrese 1000 (lorthmc scle) 7

8 Approches to Cross- Resource Allocton* Dynmc Prormmn Stte Spce Reducton etwor Utlty Mxmzton Wreless UM Multperod UM etwor Optmzton Dstrbuted Optmzton Dstrbuted Alorthms Gme Theory Mechnsm Desn Stcelber Gmes sh Equlbrum *Much pror wor s for wred/sttc networs etwor Utlty Mxmzton Mxmzes networ utlty functon mx s. t. routn Ar Assumes Stedy stte Relble lns Fxed ln cpctes flow U ( r ) Dynmcs re only n the queues R Fxed ln cpcty U 2 (r 2 ) U n (r n ) U 1 (r 1 ) R R Wreless UM WUM Polces Extends UM to rndom envronments etwor operton s stochstc optmzton lorthm mx st E[ U ( r ( G))] m E[ r( G)] E[ R( S( G), G)] E[ S( G)] S Stolyr, eely, et. l. user vdeo Physcl Physcl Physcl Physcl Physcl Control networ resources Inputs: Rndom networ chnnel nformton G etwor prmeters Other polces Outputs: Control prmeters Optmzed performnce, tht Meet constrnts Chnnel smple drven polces Exmple: UM nd Adptve Modulton Rte-Dely-Relblty Polces Informton rte r() Tx power S() Tx Rte R() Tx code rte Polcy dpts to Chnn chnnel condtons (G) Pcet bclo Hstorcl power use Dt U 3 ( r 3 ) Dt Dt U 1 ( r 1 ) U 2 ( r 2 ) Buffer Physcl Buffer Physcl Polcy Results Bloc codes used 8

9 Gme theory Coordntn user ctons n lre d-hoc networ cn be nfesble Dstrbuted control dffcult to derve nd computtonlly complex Gme theory provdes new prdm Users ct to wn me or rech n equlbrum Users heteroeneous nd non-coopertve Locl competton cn yeld optml outcomes Dynmcs mpct equlbrum nd outcome Adptton v me theory A Fundmentl Lmts of Wreless Systems (DARPA Chllene Prorm) C B D Reserch Ares - Fundmentl performnce lmts nd trdeoffs - ode cooperton nd conton - Adptve technques - n nd Cross-lyer desn - etwor/pplcton nterfce - End-to-end performnce optmzton nd urntees etwor Metrcs etwor Fundmentl Lmts Cpcty Dely Oute Cross-lyer Desn nd End-to-end Performnce Cpcty (C*,D*,R*) Dely Robustness Applcton Metrcs Summry The dynmc nture of d-hoc networs ndcte tht dptton technques re necessry nd powerful Adptton cn trnscend ll lyers of the protocol stc Approches to optmzton nclude dynmc prormmn, utlty mxmzton, nd me theory etwor dynmcs me centrlzed/dstrbuted control chllenn Gme theory provdes smple prdm tht cn yeld ner-optml solutons 9