January 1985 LIDS-P-1428 A PERSPECTIVE ON MULTIACCESS CHANNELS* by R. G. Gallager** ABSTRACT The Information Theoretic Approach and the Collision Resolution Approach to Multiaccess Channels are reviewed in terms of the Underlying Communication Problems that both are modelling. We give some perspective on the strengths and weakness of these approaches and argue for the need of a more combined approach focused on coding and decoding techniques. *This research was conducted at the M.I.T. Laboratory for Information and Decision Systems with partial support provided by NSF under Grant NSFECS-8310698 and by DARPA under Contract ONR/N00014-84-K-0357. This paper is an expanded version of the Shannon Lecture at the International Symposium on Information Theory at St. Jovite, Quebec, in September, 1983. **Room No. 35-206, Laboratory for Information and Decision Systems, Massachusetts Institute of Technology, Cambridge, MA 02139. F'RSETIVE A 1 a.. i.i"1 Fo!r br -di r-l fsome t he I.ast rtua i. solat:i. .s th er e h a. e multiaccess t.e o,. from and to vr - -en oni f<iodjelti. del -. 4'r'!.:: C)- RO.. T JCT . it JN o-.f reseearch f di terent addres j.yive somi.e persnpective rk i.: d 7" of :i.s i ,..i orruF:':i L?- tnie a: sEfi r !et i both . ' Eac ::: r(eIcei' e ,ve .,- . and 'i-ther¢ 'ani'sm: r.: ra.:.:n . n etwri LCr I te. ephone line. ft netio4.:rks ~~-" re'', -r.r ng-nai r = is l t hese re i. to r i r -dies bj t:o prov'ide needed tie in, orm.1 i tionrr ~ tharr. t- r ee theora lI tip rec:ei v-ds he three hese ' r'-e There rr' i i,, r, d si .5 -e:-idbac-:: ]. ~ also a common " - in channel - - f a.i vr ~ most i::, ,: I:o-u..ur t.? . t of -' c rtra . is node - tcie r ir-.Or forfpo the h u. l i nk -' networ rere Woneik where adectuIate .r'.i ,nr . the ino rm-a ti'o. on d= e s . -a h 1. tr-a'ffic to th,:e ce ntra is et.eHen fedbrack m,:-s,! i 'L:v" i bfy an a arreas. .e b "t.-, n ri-iisro i at rad io is am.oun:t th ev.rii? l]L. it' 7 :ar a whi-nere ~ t i err-s s.a:l i. t.e'r-., i or and f or- tih e ... 1: mdel repe a.i-er, st.:udvir'n tci ct.ali shal tuation t r ansmi iom o. y.'--. 7This 'ieo.f ite tsi.n o c.ntras . ruitiaces nt'.i fr' . a t.i. oin 'l-hereir is -some Lu.ilv s atel i We f.i h time. .e Lr-noceedin i n.oi-e arnd b'ii mu't=ual .iterterencei y t,r ,-T an., d e.a i r'l(:]i 01.. each on vihat areas co mmuns cat:I. i r 1tra-: e .in d !,"a"fR1 r- f i t 'ter. s'. lt , elast three s.pectr-'"rm. i ':r.' er." snd t-r'-ans-r'!e. each L,:. i,7,ere areas as coll.ision- resolut ion The i-. at. been channiels ti-hers and T?-e unLifi-:,.'iftion. bethwteean and spread CH:ANN H ELS i-MULTfIA=CCESS Oi on a multi dro for' rep.ects. aljlows all nodes to hear ~ - - - a.L other nodes nodes, a andri C.ommon ex..amoies are a cab e 'L.v.iy connecrted The beoinninl communi.z-i iot mutltaess network: Camle n 1 rn trais.. niri te tr transmi;m. rstter other ai.l was in tr-ansmitting encre w ould pre venit er inI:terfF recei'ved. tihe cyclic acknCowt. ecig ement a.gain latLer; avi:f-id th ie wai'te-d an over-laLp.ing ar'e ( RC ) would .izeci,, - a sion if r-eso:l u'Lio.n use, Se.:rt:i.c:," r c'Lt'-r t..h i si o i tter Ia:na si5 ' pr-ocess: rece:i.ved chosein to ti:ransi.r-itters 4r'''ovid s andi 'iai_ io- var in'i-;erest-?. e t-he pI'"Co.] as1 a wa a message su aR .e mf s. s -F:" use:i iri't"er'erenc.._ ono i Erled co'_lllision piroblems !.%o-.n l.-.cLi: iathi 'mios:- ignores the of 0 - ir dt- ro uct:. on to: ti"eszde' 'ic ih aoene ra I I. v he abse-nce an . nein a-st i o'n e r-e, r o.nlv that s numer A ;-ai e.ir i ii r e.. f.: the r' t. his ap.proach as..sunming ay 't'.pic o-j etene-rai r. ol!i si on chall -!'- n± messa..e. but:) A h a. many iany . -: r'e. t: ar'r":i.,vals a criener rally try has beeni-mpriove:i the E ,.'-.F ':i' ....i::]''i-l plo."'fi l,:L em.s ,:an.di:nm'.. i::)'..-rs.tv in arnd has pro-vided m ::-r'i: resea.-rch CoI bnth -cno ch heck: , w ould be pseudorandomly crr'l.i. another a nal yze 3r'F, wi. ticiespr"ead ti noa t s.ame tinme. te : evf::t e.::j in!..ervati., and the transmit ter time woJuld certair'.tv nf' networki. I¥f anotiher; e +rom beina co rrr ectly 1 r edundarcy ch rck: the later _or s1mpny be tran.-mit-tecd the r O,'ver t:he-e ars. thi.s basic svtraec.'iey genera a messare in t-h e me s cr w ouid be sent approach to .ni.' Alioha. th ibra.mso -in.9',:]:,i it. wOU.li arrived at a transmitter, 1;-rgn- reol:ution 'The idea here was that whenever 1] , oacki-e'i- radio network.:: the collisioin o connectino many ani-i en betf.':'ene th r, e noi s.e.e- oar e .unc1er:lyino communicat.ionr ltr.ansmi-ssiori and ilncorrectyi is corr-ectly receiv.,. ed -1.3(]': i" 5, e.r w.i s e. Thi-e mu it i ac; i normation mu! .. rbecan ia"cess in manyil an d ways fE r-,bl. ems.i give n are ii nterf'rei-r-..!ce io.' t:he random f.'-:ire proreeding.r i i Io t-fh ihe e ri e. .i.. .::-i s sse alrtay I] (] J. !"1"t poin·t h rq n t, i channel.si, ~_i' i'1 t,'.- or - t rie,-::'..:.'; rl:-:il.::j'L'j- "i:-he r .rec i ra . l n e 1., i i -l7 n e r-e, er-vi-r i i . l;_ir thiat i't i i is rwl-e5r-'l Fi rr ' j...-£- ns-r i '1::r' i:=.~ iic: {: wr !li.~r c(:i)':ier th~e 1-ci.i t un r t i.a' i t t rar r c.cl. it' :i., cr.a )o i , .- u ir" :: 1.:m 1.: ,n ore+: c t. .(sd ur-s:tsy messace arr--i val s ·rfor mul a ti sources '-ess cf-hannels, aind can and shou :c:ld have Z: r nf v-i L7 wtL -' nave . f or point i c ana n l -fi.or -.'-ce - F.orl detail to : it.-. fif adti h- ns. i"-forml i' there -iommun ti c a -s " e nioth iingc The arri be deal t ic' m Tt - fi s i then IIo out h3er' to do wi th c-ordini with c:h : theZf shar n th i , iv'! 'i-A c)oi n-it vals: ,I. The -tE! so p".'o cti si].i i~aci: i,.--.k ciat-:,a minitr a s Fo:)r mi. ti a;ce' cha Cnnels ::.n '''.r to 1 nr furt.hermore I~ ;n a The p robi.:,M ari. iw o u .d - pr-actl-icai ter'ss.a; h!- mu i. t:i. as, i ' a mi.i i::i.,r .a.: r'i,:::d o'~n tv a ±few are bu~svy :,. ,is . tro be t rr-ai:F.-itted., ai. ,. r- o r :i.i-'liired* ner ofr:- :l.t'a . chnnle' r!.: the~'ofm'j~de' i desi ar rivals ag es Forfntsend i rc .- a pr opri.a"ei¥ usa].ly IunrFeasorn-.-1 . sa rF are -ter c and to undersitand i .. i 1 u re r- me.ssa..es ai--e -ne m dr noise channel mportan is resear-ch c. of thi_ the 4t-.ii i aitsnd com mulnicat i on L1v .9.rti infiri i -a: 5ies if ancih wv -t.hi-, a -j::. r c,f; i_ gfsni' 5. i- accsI mutl aIrrvals it . a'.cri rc'-ar'i- ienr.eral j. i5 been uisp a se iarate a*rea in !:hi .s a..erstr."t u t, thei- fmo :)deli .,e:u,: i,-lhas alo, siummari.s and descr ipions [4,576, to theorern developi-ed by This w-or-: op lened Ih as ilent i approach 197:3 with a c:dingl in :[2] arid Liao [:3J Aihi 1.rswede theoretic have to dco throui t C .l heor-ems ri -th the soutr-ce codinga. e -i.4- without. i--sven; s'o:...r-ce coGd. r ,r thi erl. o-v er ti-he ar-bi trar ih -ieorems th e i.-:lt3at. i...!"_er no.Iot sa].I.v jc :r-a iim:i. ch an;nre ci eat1er- CIa n._m.-ine tii :::'i i to-, be smoo:',c'thed out) tha -ofismenoh,:'_~w Sourlc""c2e prec.i J. -:onf_~ . Lu.'ai:, ij'.i:-' a.t-rvii r m-i: ·u:cs-i1u ] l i:hen sme ' ,. t'.he a.:':'E:i.v' eri ini ceret i.s C. aDpear smotihed ca.i i (iCe. v s the po' all er hibilit' t e t- -e iim::it iCs ini-ervai o.t are oirv Ler ective calse- achievabie. i time on nt; i i b b ot I i `r'articaiiC 1 he ef p uudes ad dinI ccss decocdirlgc simult ianeously i thi' . ,-. 0 a neiw .y r~ar'n"smi s 'e thre 't-ime foor t'han trh e t iCmoosing nOl qu red ca-iy fpi a For" r h-se' e of - ar I:tim i-nter'vai no= I.i the lera bie ongo delays . on ] i n' tL ih i s t arse a., it the presence red seq.e-nc i s af: n ew several since otinerwise orQ.-wave-:tlr'., :)eir.'i:' wo£u._ ti.-a.na1: d i titeJ i op t3 to t de - v ei i: i ri s-. to coding theoernem) This understanidinii o-f each or that' toL devel op seem"s:. -od3i ng. (as ouposed envirionment . inr [i: on si._u-c:e- , o. i created -frJ:s rnoi'se ani-d whati.ever' was ~J L:ers. ar':' LK i ..number iiio i.Crve",d when' the m'essa ej i :r The i'teo. r-ecei of I ii"te transmit er ei ien i. ; the coi used a r in -sum.e eri t tr.fu,'ct:i.ion wnich- is a-i e.to ano .iessa :-e unrici-erstand i. noC, of: mu! iltiai is is that V i.e'i ulidl be sin03].e i. thie (nes.sareS.s. L rine atppr oach r esl,. .. ew c Ce er-abie delav.5, 'ci- to 'i i~t-r e cd:in-i a chi- evablee a nd b* ecaus- Co- arr-ivat- in .J:orLc no i_ mat-er: ar e -t howe er- s.__- '....c d7ore - or er" i.me -inter-vals usLed po1. i' i i what oin -:c:r t-ne recqi"- t-ca or-oces ies 'aiL arr--J wll ar .f rom winat t;-: . c ~ss t 1_ c.- ti e t the arrival-s c,'.:-F-,;n-?O tnil r t-ieo'e1-s m "l..' iy i bi_:rst¥ mo re Fromaf if messaues the pr-ob shouid being in s: a infvolve trarnsmitted em simply reduces to e counlict tnie Li resolution absene cf mo"re ode'ta ii and co-da3ing, svstems we tacklin in section ., multiacess coding in .kn-'own the discussion is restricted to the si1mpl est-. interest with for this is context to before many sourci-ls and m t ters. spectrum approrachi not bee discuss'ed .Ji.il d1 scussed ir munliti.access c:oi i. si '-, or der Ymore, t i :Imes in m:d ..i l ti,itd than military il ook se seen iv tC.. t2 ac: :' o can seiquen'Ices: others-. multi rIie broad band naoise -the tCO i:'he rece:iver' af ter de,::duation The sJigrnai f r e , Le nc y broader jamim.er t o !:t .the s signal s-i nal wi 1.i to b.andi jammer does wi l and the be r-educed by i:' -. o:'.rl multi so::uies, ]. ike the invent-i anai S-pread ;-Q environment. AssuL.minortha br.ieflyi Posf-ibilities ginal I y d c-Y' vel ped over a much necessarv, is out-side -the c o ori no't.ij know tnie fr-i.duiatiln:seIuencei essentialy the t-ypes oCf lie, catio n a bt theorny appr!acnes. ,.:cdiin amming i in- thi:s papere that i78]j tultf-access channels to illustri:ate a.nci i. t ted ia. to any detail comimunic-at;ion aoai nst be tl-irntm P. in is a mode o:.fcomnuni prote'ct sav in here resc..onr u pe '.ctru a ion thieory i -ttleis what The rationale in . tihe mr'obiem ot real i nf'-ord matio i we discuss In both sections, The spread i;lf:i for reli .able transmis mul.1 ti ac ,e-.7 discu.ss with only two sources, understan-d tr a n added coding. con-fiictrs. recrt ion 2. in abLout with access 'i-rar'- mit c ommunnication using at diifferent once and each will If ;irin, we compare usinig this usinq . ook like type p frequency snspread spect rum, broad m,::dul atino. band noise to the o-f system bands. it several to frequency aoupears at f irst thiat spread siectrum is not' a very transmit- tooaetfer goo(d idiea. using spread spectrum: cronsicierable.. a.nd the resultingg in terms: ot capaci ty. typically ma;ny femwemr 3undamental the same reiso-iution appr,' to eachi in mesc;ag other problem handled Since many ach). seq,ences can be chosen noi. se tirmes tihat are aimo-st spread -Thi.s soiLt ioni st:il. n-eeds; at oncei: t:.-wev .'er. r:, m.:def; approah th ian i ._ (i co .L :i. s nfot the decoding is miajr aI ra to . send, of if.t a,- es on r,:' i ,-ut - needed are Oi o-'. ri- there or-1 - or thoonal r (t-lhi modulatiion andI. liike to the busy since one it s transmit-es i 1us namely tihilat n et ii'ook an automatic c:for mfiulti access . la is a transmitters more t-ihan too many - it by the collision compllex 'iti" is paper tiat but enltirely satis-factory very to FDM is ystems tihe channel cis:L.'n i-onresol uti.'~on wien anc the spect rum provides solution to the problem of allocating user, clearly interior the frequencies to thie busy probliemi all.iocating is p with thian noise becormes with FDM:i however, transmitters p p transmitters tihe seilf system is The problem there are matny more than When endi rates a bet-ier- set cLR,mmun icat.on r y a l on e. The 2. in-If orimati on Tin'eoretic Tihe codi nu tiheor ems of holw set much assumed send. thiat in -ormation reliably nave a the sources ouestion ,'"if ' T'hre ci ass Fig Each Ii . svmbooi x channeis unnit of .i from an tr:asmi'itter prob.3iii and s i 'i wzWi. ' 4J 4' 'X a= i ilit':.' as gnment. s3i FP The ,h-annel n i7i- , ;i. i each (o nris i is is an output sen ds i the e c -o:i e e"n'ds a if.j niat 4 w'. wi. same way nX x ii 11 I We assume j. t a nin the j.,,'twt.o t'iransmi'tters and resoecti veiy vvi ngver' units,':her' thei :i's one ie probabiliLy of recei y N successivye -(y'i . gio'evn Xw, F:'y xw:. ;:y Ti F.v,.:: w bL .'i Y and oi ilfnputs :-.'. sense n.i a alpabet :.determ:.n -i. v;Y for e ac:h r in i lust:rated transmi't( l 'y me moryless e vl s i in I' e syvtem*s transmi 'ata tted ransmitt er second Th er e eeivinn .ii-i ri to dbe O-i alpohabet- W. pro-i.bi ' X and tnie toacitly data ir io- m ulkaccess rst is it area do not-- addr-ess to be con-sidered let e one pointr Itor reservoir this in the questi on nts. empty in ti-imes ti i From an alohabet '.!'::bol .w' I' o never arises tne 'ii random arrival f-ro3m orpoiset o¥ results the delay that theory treat - crmmuricatedC oi nt, Thu-s the theor-etical be cause . 'fi can be to another poii nts. f- the data of Ao.oroacri or" the .j So.l time be oovi(3 as -for- singl.e indicated nAs sour ces which bei no that in .--. input the are encoded tha t he al t this p--abets ca n be ar'e all r]enrn.=..r time the. 'for . di scri.-e. zii-a 17 cl-ianrnel s there .figure, ar-e two independently independent in-to the two channel T'-i . i.niuts. W.Con.ider- bl ockP ais inu'nd I{!-_t . } code _ {w i w +or L.... } channel codino ib4th a i ven block for t ransmitte r s transmitter 2: each paramet+ers as an (ii'l.HL) code .4ord is The rates code. N r.ing word s i and L code a seouence o. N (_-ode wi i'h tihese For convenience wi areieto jnouts. lenth of -tie two sources are de i ned as R (i n )l /N, Each N4 uni.its o.f R time. 1ser i unfi-forinv i o'Li- I- in =- -,-_r o !ini.mized .'i. ieah,_-i y as i< iite :, ; cas (n t fworof s ouif ' ou.'sce L'';n.{ , *i-P assi -~~~~~,n~~~~~~~~~~.P1----·--~---~--- 'ee n:men'ts 1 ----- h nro . eF-:' ech '~ou: t p ut-'L to result e ma,,it a r" [23 ffhlswede asiognmentiS on the o-f rate ac, ~ ,..' . satisfy · -- pairs ~·----··---·--- · IF-. 3._. i:. at t.i e i e 4: c r:de decoder -ieui---:-w "o.,. tn ese can]..r],:i ' i".3 E .i Lei:t X and W input rate 3. i (x) i-ihe and alp-habtet'c r-egion R as the co nve'x (R 1 ,,.i '.,which .for each otI the . .... e;:-rr.or - or a choosiC-no o- h- and L au a f :ti'erw.w.-r-e m1:-in F'i outputC r ini ia- ur n.ko abou'i 7 messa.ae dlcoder- e or' o channel decodino L t hat rt<n--m. De-firne thle achievable the . i; ni..lu i damelnt-al be probabili'ty of o"? ' h ima:n rfun. due m io and ma. 'imum ii tkel ii'jhood ;' oener-ates anr- decoded Wii_-o are knownIti oure n -ce .. tn,:ei respectively. hull by a minunliormlv Thie transmit ters is decodfino fcorrect- lq . t i he, 1orem c:odi in i(w! . , into a mapomed The pr_.bab'.i.i c hoe n wi be rios't: "he t the i. inteoer 1 to L. _~r-C0Hilbl~lit%!, "s, r' , W i:(y x rom ie OCCU-S. occ'-:-ur an and the corres o on and d'Jec:':odin-,-o e,:.'-r'-or " distriiiute ithe decroder and'l is y enters 2 source 2 independently ri-iVy rr xwmand L)N /N.. 1 oenerates source dist-ributed .from 1 t-o HI and intecer (In = some choice inequalities: of i-'i. "r W *~. lF- h.>.. I- k iy F W Y W * .. '-W, iras yFP io i _< ' | rsf ;1 ~ iZsr3 1'"iI~ i'}r ioi:· { i~ '-' '. -, J::'': ':¢' il ' jr u bound ' co":: n ci:i i:i i .i. exp i.,i l anr d In possible an . · . 'i, . ... byL' 2.:) suzcn q R 2" codes e ffe ect Ri P( ( i w ,, (y · ' i i ' -2 ftor I: 1-' gc ) ex'p ac t l (2.F4 i" ' i - { 'C iiv i (t Jh *' w are r!d:F alsoe ... C,: i i" r'd"el'n' i"iO.L'n o.. ' L( -i v. > each alli >':- , E N (N.,I',L) >' 'No, code L -the interior - exp ) 6 i~t witih suc h NiRi+6), in .4 :--w .. Y K f I£ ar-nd x anrdW t he thi-eor-em s ays th hat for- source rates i'n. S ' n s-: . i:]h . f'" r a' iven; e bre'.:k point. o-f there fexists witlh iM i i. F'or- t:hai~.i: Is ws. . (y d i_ it' Xjare Nlisince r --a Get"ieFa' equaiiiv i¥tt iier'e exists an iiR) (Fi i P t y w w)F~~~y~xw~~ln . w . ,Aih5sw'ede, Li aoi: iH, t. ' nJ es! ,.'::end en t i or ve i u a i" K.it,... ; (Nfi,M,L) L 5.i ex:Lsi. 6> ,;;: ' w;·' O dd' is:i iihee.' th b - :LX:' v ..Y'JIn i' iheL:iT.,'::)m n·ler",e s't fL i' L-i: ini"( q i.3si'.--" e--a Gr y ' :iX;[.. 7 i' g) 'i"he ;. iX'"e3i . ,.i' in '-lto fr _ei ~ w r fn(w xV4 " i "'x_- L wAW;Y (':~ e;. p FP tha-t ' (Ri2 "R N(Rz-6), For% FP'- i eacn N(R,,+&).n thte achievable ' L E for al r-el iable communrication of s is <- -2.4is reoion and Sleoian and Wolf a thi.rd sou::,rce tr ans:i. t -eThiey r- 5, -lshowed tc1iat outside of impossible [9J latier that used in .-his ran-dom coding cdCinig insi.ui.i- intoi throuih thie arr1iument Let Q i (x) and Codinci is a very both arument which with N and behavior. si mpie extensior. ofgives a gr-eat access channels, we now deal oft oo assianments and consider respectively codes wnhere each coCe ia Theorem be probability Q., w. W alphabets selected accordi- for or tihr e cwo sour-ce case. Mul tiacc ess 2,i. by considering eXrponen+rialiy input channelis and muitiple codinrLo 1for result a r-andom cod inog argument single *tor- this tChat most codes nave this a sense- random reoion, achievable jointly Pe can be made r-o decrease s7howed also. i nrce eneralized be encoded could .als the x , to te i m cro aob itv orn the X and an ensemble o-i(N:MiL) i1i, isiwordind eoendrn'tI as . si( nment- N and eiach c1ice a c [:oir di r] n] D.. wor-cd wq? i _ , ID._ (.w eh t1 d code in likelihood vaslue F' e i L is inde ,pen w =.-w ..... W1, i e the entl y seiected t c( (w_ 1 For each I S o+ in ense the ensemble. decoding. F'E ) , tfor this deod the decoder and we want ensemble. i,,_ uses maximum to upper- bound the expected LDeine an error event to bIe of type the de co ded pair - 1 if (mn - sat.is-fy . and i 12 1), and fi! = m. mi is of type be the probability. .*- Consider encoder, (Hi-) i pairs, 3 if hm that we can directly Xi,, 0 , p, <:[ (H-i) (L-i) ]F' T' and the · 1-he statement of words are chosen (xfit .W ) fiii with (x rW) as input alphabet theorem 5.6.1 of X".'',W [_l1-], 1l (: x. and P1, E's. rates in theorem 5.6. (2.2) , of independently, but m, i2, or independent of rmecarding this 2 F• 1 . (2.-'1 2. 2.7 si.mpli.fies, t o [i-0] assumes that all code the proof only uses pairwise independence between the transmitted word word Z i For each such pair Thuss ' for ('.y) ffr r m OT f. dl,.' fi -- tni tion oDf 1 error choices is statistically w input channel that for all t he product 'ei, a type i (L-1i a type 3 error. yield y Let ri = m (mk)) enters the Note that when -_single 2 if tyvp + apply the coding theorem, which asserts, Uis n r of source pair e 'i wq o ver the en semble of condes input to is '12 f 1. m and choices for ,T and the code word pair combined event An error . PeiPe+ , first. there are M-i L-1) (niTi), Xmq P,; and the original over the ensemble obviousl.Y Pee = event: (im,i ) (Xm,w) and each other e i+F' N ...I C..+ L-(v Y er dconi Ne;-'t n F' this en-teringi- w,) transmitted viev;- the cihannel as a transition ma1k :e F .y |x m.w m--. im inc this 'e evern t u p p etr r..t bo li; meust by the at thi seconrd input. e input decoder t;: occure IW t 1 error T:akingl the expected = We 1 code with a .iven w., with input: si nge that Wie can x Iand w:i one m.- 1 error on ~~~---------·------------~ ~ ~ ~ ~ ~- ~u-.n ~ any j y ! i Q(xi C(- PF [.' 1. : o-f fer xw over- w the--·-·---·~--·---·-~ - p.~ saent Z vf on a mi o i th 4hieor:-'.n m (2. _11) - or carm W, beinq, .,e ihave, , Y tL o+ c hannel P m. e r-r r tiv.f ising n, (M-I) P input uous)- if rrondit:ional pr foD a iij. value for wh,.ene,-er- a ty'e i±nput channel P'Type of channelr leas.t at - inpuFlei't chanre'.e o1t ....'i s si.n1gie - mess-age a particular a choice a itypep :.-Ci, f nd e d -'.,tf:v n-sl.: W ) .F ¢ m ani d and (,o-rbe am.big P '--'y yX !] P(.yx nw .)i . i hood an-, err-o-r :prob-.Dbi j. ii.t a. singi probabil ities, A ma--i mum3. lii kel ,1i on probability the sec-nd encoder, partiicullar . the probabilit y that ol ' - first condition X (14A U 0 ] is- '-?h e aga.irn !::_ 1 2 and thien using t h r - 11) - h it :ol.-n ri'-of-duct Q again. 1 and .t-, L.1 R F' Applying same arigument-i-to type the . 1+ w :, . 2 J -- ' r! .- ±p) for all errlors, p, 0 p S f _., PF p RIp...F.IP L- x e p C e. ' 2.. 1,) [.'...' 2 (w) Puttting '"'-wF (2.. 2.) (i on N. e.por'nenti al d en T-hieorem (SiepiaJn-Wc.!f): 2.2 ..n :nd....; d "']onseln h .i h accor-dino .-- F' to . f.p a in F' - i-Fi ei nb 1 e 'sense; (22. for- sa:.tis-f the i. - ('):i. the (N MLA indep.,n ib.t iver-i te J'r-ba A- e.'-pected i t vy er'r('3r i-4l ;' . f *,enenti ie :' -t eto emphasize an en'semble of *la-re Ti(:.. rhen 1.-. form we have: Confsider and i zen. _'.w) :' 3) 1 and i tv over- thine pr'.,ibabi 1 i' dence ; ' (Y - all.I I' 1 F' L ., 1!1 _ R f ln R= R._ t L R-= R.,2 R, zf·-1 01 -i~~i~ . 2 iEj t.P ( -,+ = -in in'.e - of ~(]~~~~~~~ *sfm. Lv Pc.~ii·- · 7- n p it and ore Lve a nd Q6 5.c I 4 et, ., i. i mna- decr eaia -ii oi-U 1: ince''"-'.::iegieb' n N. iFi ~he . e t r i t linctfi th-e s . f!.. iol-ervi o- 10 w) . . -- Ex.1 ipp - i th e, L s "F:~P~·_1~~~~~~ t rei -1i i mum c). i-Q. Of in h n P * [ .(v . O expres;i for- p r-ocs) I s+ w ',*v C'.f E i+.R - - n,;.ric Inn I( pG.) 1P for tpo~: -p,. ve c . ovRr Ps Thehr-em -'Li I0 2en2 r 1 ipL Ip ; ee theorems theni asserts decr-eses epoeritia. to 1..y - ith 5. Ob thi.: it There are two questions we want to explore in this section. First. priobabil ityt anid second, practicality of- coding as upper bounded s-econd source messaig aliso clearly a F-:- is to P', errors, W ur is ieo et rem 52 problem. a. the b est- c rs d ': "idle' is we shall X t. e pair :? 1(l-) ,Q2(w, or i is errors a--re aided error the bound f-or inputs with for a probability the bound on PF ii that tie given ensemble o s ee t hrougCiCh tihe a set of n always repr ese! .tat iv in ci odes. 'the t ia- e. amples, ri t he'l e'lsem i :- s, or, Channel K} and if 0 is and W "collision i ()j . K} ,:c. input for- a - the . i mil m( ar l . F-irst consider the two soon if ' C:' and wJ n .1i thn, 'ight are not {:,i,. = input, Finaily rIr it quite "':"IThe CIrIioii:i Let concl usi on about input random coding bound sinqle single code of rate R R.i+ 2 using combnei i.(x) ii type Similarly, Pe . in probability so tnhat when the genie t.he probability the decoder error the To e' plorer the error i in-formed approximates the conventional t gi ve of the terms PF This genie aided P. cause of closely (2. 12) "genie3' lower bound thie predominant probability e inter-pret in o-f bound on error multiacceSS channels. we first wic'uld resuilt if a that this what ind ication does i for question 0o. tightness.= Pi- is how tight the rest i¥ ie n w input': t":: hhe._ ou. tputt i w the ou tpt This is W We regard C) as an is a special SSioWn i in Siig. it is easy to see that, conditional inputs are statistically independent; y (U y the achievable rate region. then symbo.:l fr.]; fr c K- For any given on the output thus I(X;YW) = y, iX:Y) rec-ta n l-Ie (2.5 is tJe nexnt e,2 .n-Hd }LV,(wsr of union i r he ri,= or]i abi union;' rates exn!: onent all i n . !murli. t:l e. 2 it nroC, C rone: ·9.3..r)ii s - is -reoion. p s d I (XW Y) max;imization -CQ was fir-st is, a bu i.i 1 t.! ai. r- i the each- relij iabe T h ui's-i. : eore i! n uc in unc. : ti c.ornve j D n1. - ... 'n hifhi r :,w. Q't 'f ..'e recion. either over ext.i-erma can i i-)71 oa a.s-- a. . n-trcocate non rtregeir. n i tunion i r maxnimization Either way, multiple isolated int.erior- o-tr rates pa i r- and Q-, can be viewed or a concave that d: i. visi,- rcanrve a of e- m 2i, e t ::vr- G-. is nTi u. inter C. C. co nav. ve e as a a" r e s ul i i.n f thie the thatthe s tihe shown by the set "ratepa.ir , fa f or sor me Q1) over t ii me absh t hor- 'x pr-o mbabit i orf in i ven r aci a. t t . s. t h e '_.w pai.rr's i nf chioc-s non-convex..' (the non- pa.ir tihe ia set st-e s, rr id s,, Threiorem 2.2 assures I (X W Y )i t 2C.4 4- region the . i vei fg tha all channels eeilioni .- o r 0 r and ifn over li.ne be'i'_-e unin . effi . CA$)(,ir'' (h-2 n.ax i nmiz i nr concave 1 be c...ar x need nr rat eas a r-e a chi evable o7n a straioitt \:, : -,-'-'.'. ! r.iFc'. i: L'5... n b'i, is 1-1nv01 eIn rate i... !". e 6 1 rates all for .(2:- aE'i.t sh.ul It fth' F t -his-un4ion ror r-t ote n:. t,,(::c:.: r ) , U(A., o- ratiher 'sur!orhisino rir-st f-'u c ...ion : (:)t oi. ii:.i.'.,r.j oF -theoremi2 i f :.ns ac,: _e'.:: ,,x *:..R ' h ul. -i ) codes. fo no bet-een se e:ir. co: ':un1 .!' i ,:: cai:-'i 2 and i multi-access or of'.e tih eri i:C i (2:3)-2 g -') 'the set rec:. on lL!. i s i ::! n n v deer- t:tm c:;onx fi;r 0 rates F. F should be constant satisfsviJ . t J O of choice .i. - orms so - ant for certain achitevable ao-¥ nst m'The con vex, [1I.] ), some C!t rK r' p- a.iiit y convex c ,,)-(25) ( f ind h trhe setftesr-ao nt ¥of -f.ri satifiE.d. t mmecrvV t t ha. from . the o.- rates satisfvino and the set Tin.us- as a nona non-convex e;xist and there3 s o-- the no anairo u t Arim oto-lah also rmight be surprising colli sion the .. !C;i u..using" th-e o'ther r'' eison is ra.-ctcal ./ For aI:I hfa nne i.n we t.Crh-i : we"can C i. ieva .' n p r. L nc C in it... ,ind this - "C I... in R-.'-,': i:;." iin tihe ~ ...... t':) ') 1- .. E. n -.::can ' :L = a 2.r 'The f .@-p I b be -ax imi :c .d th rinouhC .. in.rtormati3.on. isorthe ti di cttr:. thre in - Djor user bu'tI. i3'. (tx. w) 1orm: 2 .ver pr.-.cm. .ict 'b creates an exponent principle u. nion har.e i oi i n (a e'--'tlra aiin . np - ut ir one to be i.dl.e also error- probabii t-i i between idle). is s olut Es 1,'_~scs thJ. .J.; e.r-, . pen 1]o..? leS. r 8: rnot le.ittipn r..:l 1 arge KI) ,ti'-- nr- by i.e. . =0 t.he other wihether r- at fr-- reqgionr by imiultipie.i-ing Qi._- i, while chiannel the co.ns -i der- ::.l..i. os.'-J.i 7 n ( = rl:.i n- -rmatio.'tn . and t h e emi nent].'-.p - aid i . t. hat the choice connve:s i :N;ext = i that the achievable not achieved is channei alg orithm th 13] rate reoion, can be used to -find the achiev-abl.e It 12] reqion m.{ !]~- . c+ .'.-. ::[i.-f-9i a.4, ( ':...... of 'I'The error decay same kinds of +or each oroblem e-- acln:i;eve s;.-:. *-l - *-. Af the : by time and for R a f .... ibie are cfo' i. n,::exp-o'. net e r' sem J1 -I i diownin +' i --r n:'1H.. ' . 'u;". .:U : a "-ine r-eR-.t' bi' ') Rx defined ori tR ..- o dr. lfthe ". ; p robabi -as i . b I oc: V k This wou , r i- and r-2, d ftound to. achiieve ex.ponentia.l ~1-.1-___ 1---__~___ ------- /I + (-X)E' approachi in as. tfrm i-the .i F'e for all radm th e codi. i g rin.q t y eadead': r replaced a by g, a s-, (2.25) enilarges the sufficient: rate .s tL-,he o i approach is r-Fando'ml a c h odi c. e's o.f where this decays .::s nin( ci3 fO be.rig : Si f-ef d: i: t,-ar rn this i-! dep er :i.' i, above,; to F it = pair-s in R-. R -st R. ,R, + or which. Ru' shr ! a the tim-e and r{-ate is_ i,, . -- - ap' r.ah r2. .5) No examples have been ."~" ^~~-~-~~- time. a in for codes lenqgth ±rac-tion Ankof t -iesrom - 'i on us in ,:: iS andri ER F block' ( .l-the other- of on; . :w.e.-i oh';:j i tde d:-tvea. , w,.i et E. equal ,1 . to severa.l -,ereti ].ett er are two rate ~c or mo-re can be achiev-ed, ) ,, . i.i..'xc s e .,( culd i 1oni .:ti there '' to we wan.i 2. 2 code a ,a wt a-s a if : ,a-i a p isi r C re-:i.:. on R i."re t .s between sharing 'ir_ E:, -istin :e pair '. an exponent of ne- nivex ' iulli c-.,and if sRu -=ch th.at '.*-" .. .... ...1'R th-at t-hat Note decay qq .racrtion ! Ie (i- ~.t.rns man ·"i:it regions R us-ina the first ,R'- seco nd t h i ;;- for any rate'i. *i1 sens rei3 rniow .5nd then, t.h na this m ax lza. ti. on a a givein exponnential JJpa l F _.q wi i thie feasibie b ind.'iin'4 .' , inrertormi ist howeverl pairs in th7e i in terior oi' R. Another whici'h approach successiv.e Co .1i.1jsioi' ¥:::,r ui--bl1ocks ,- f -'ur ?.: --e e:ui l.a :J ni proibability. probabl y each ) t1o t(: have either' . :p r Ri. Finally, chosienf from sui.b-blck of length x t he blo i: and f)orm the ii b wi,,: x (-)i . i s),: or to a noc sciheme -for :!.sy'.l'.croni f ale rather or i.-: :::' s w.it I "the ab."ove mu .. .i t spher'e somewil-at c hia rie n s t.lhere .Si s acc,"e ss ch anni codingi pac'-kin1 -the tw(o enrc:ders bo.::)und. b1(tnd loose since for s it for [1' 4 ]) The -for the m arginal random in t mul..i iaccess does not accountl [15] h a-ln s ame e but s.rle low.er r r ,r t butt e...- it al s b .o A un. t:) i i ne n i has developed channels, for tohe mo7 re c ,rmpl i ' _.:. ica s pher e pack-ing that .1 v el ds t i e f- i n' channell, or s iln AruLt-yunyan ve was notiP to collision t h i.s s pro babilities. are f{ar the type 3 :rrors. vchannei resulting e'. pone i . t' iimp le a nd thinan hthne t-P·hatI thcrw-,ss wher-e disus-_sio:]n t -.i: a-n. r.:o codin t he same ti.tlZ. 'i:r-S _e why errlr .ta'i. .st-.str why as th-ie randim o.: n i.~r,'"{'r e:.'.:poneern:ts:. achievable to...,. : .r. achannels cei.arl..iy o-f n ulti ancL.i:::es s :i.. iu is i.pen nde .: Tfhie- p.-'"p3ipsfe 1arcges'. Massey's coeii. sion i ze.i ,.'::.::.t:in~ no:-po:neln'it Tcet.ters b with iseless e. at.mpie Ieste, the .arr.angement this with 21K inp ut su_ int o zei '-a ry w are independent ly val ent tihe we (i iw,i .w or in di.vided is Wj.-ikth 2`'K input"s:, and n a noiseiless wJ channel was r F- :.C- hS.~ With eui '.L,- iFn i.r;. { 1 .K). .2 4 is ensembles Within .i si each sub--bl~ck:s each, ) has .. -; it her thne form t''W2.. . .. equal r3s coding derpendent.n ly sfupp . o.=. ex.a mpI e le-tt-.- "o?'2.',:.' .- ,n,I :.:: * '( ,;:, to consider random letft].er-s are stati.stic-al ch anr'eJ.? choose is a type is the separation o-f -2142.1 Addit:ive We now turn imp-fortancen n L ct t whlere Su ippoe I. rea:L White anothei r of exaFmple the random te1 N)oise Channel W, codino art.1 and the output num!nbers-- y Gausrsian somewhat exrtonents Y apih y is (AWGN;' a zero mean Gaus:sian random variable + x and The . indepiendent otf Constr-Lrained to have mean w, channel at values consider. the I inplDut input and w square Fj-+ we see that' wit iih the channel S1 most is inot ar variance and S, as a cascade cia.nnei I(XW:Y) of -26 inpu't_are each .xiand w +ollo;wed nor:Lse.. ess channel. adding the s:i. n-le si1_ tihe (2. is Gaussian each by given '..' + .we respec:veley. practica.-i ftr-out mfore 4:- are abt:j-s = .l-where z reatrf- .ingl1 e by a of a input at most the capacitv of ned to ener.y cnt -strai ~~:1i~~~~. ~ I· is -i a so_ o-mat ion eaI v t c' ee bt~e · that and I--X v :LL I W i.n.. t..e + · ;i the is ab.sene a o.f w remtIuS als .. Thus . ''. ~22 -; iT + I(sWaY:X) -. ihese Finec:ual ties l are s satisfied elo for tu229) 2 al.l independen ths di.,stribiu tions on .vxand w and are all satis-fied and Lw arte independent' zero mean Gau.ls.ian ihus respec :i vely. the i r't S - R ri:i S w d.s.tribution is regi on , v conve . alread.y is it iis the achievable rate o!n R. Th'ii. s:i r 2t LT 2 _ and R 2 12 *_i rnce thliis Si i-f- 1 - + 5.7c and + variances -_Fr .1)-(2. which relion for some independent- ) satisfied with with equLality l:.L reiorin R ='" --S 2 tr:i anqul ar a code are limited i:.the axis is "iDMl o;ar s'.cal.l Aq -TD)Mi s is moist clear-l.y and our model) limitation is almrost 1 arge A , - oar r ce..i"p, ' ts:= " ,: A, (Rt;.o) region a code (C,R)i th-e while it..e boupndair.ry al most -for then irst quite as od :.3: -The inferi)r. t he case R 1 wastes hal-f transmitter transmitting. 2 if f i. g2t reason -Cor hal.f t t its i6}Tus whiel -reas this power, stays within rate.s betweenl- can Alternating the available Losing ht line -i TD le codi ng. bes-ti -= R. R u-e s one the ach ievab rstr-ai Si.e as- th e arla--Fsr small rectangu .wt. then by the o, case where Note thiat or Earious value s o:r S + i-the S = to the recii. f-in bounded :i fig,, 25 in A = -For x and ·fo:r::. 1,arge seen S/ the tethat ote anid a:l fmst biet,- weer-- ratios r ni- s to is s:etched be between since (by power the available power Fraction oinlv a small loses whereas. .for smnal.l goino to a continuous model here ancd cn-.ing achie vli rg the a A. larae model frequency power. available;= fraction time the same simpli available of the lost. r-ather than division of- t reqCluencyi between the discrete the resulting 2.6 shows rthe tr-ansmitterS: s large A suggests time thiLus mUi.itiplex-infi assuming tihe same power- i¥or- each transmitt-:er splrit for This as TDM, but beingi able city Figure is capacity -io: use all raLte region and the optiimal as a function of tne r-ates. Niext Usino consider the above Gaussian cal culate Ea. : I-a I ini e: the random coding from P) ".Ther-esult C: S... Q)-p P: i we can easily sums with S + i+PJi It channels. i Si/ we can m., , ic i ',at t.t imize io. r l In.I 2 _g; A. these is over' p to oCet the parametr i 2 replacing i2.19 (2. 17)- Letti-i-ilnq + for for- x and w. distribution F'+ whert-, exponent = li 2 i ~ + 1 i+Ai 1 +h 2.s t e h34) For rates lower than those where Pi = 1, -2.17- . r=i El .1 } . in 1 n - - -} for iAsi i (2. 2) and the minimum of thjree do minan t. Ei ons subrgi A-is for source 1 cr i2 For a small single thing resi.ul ts t-i having an in c- nsi . of rates fig. ria r R di vi des (Ri) Er for type dominated s (R1R ri for each p1roboabili tyb£ iy i into i . i of+ type :- 1 and 2 errors.. by errors in bothn. addi jtiv e Ga_ ssian noise con-t'1raint; very clo se is to NS i en.-sembile code w.or-ds. di st for the first is set s u-si: -ii to aaii.n r-ai t ini. erprei.-tat..ion n of on type by tlhe pr-obabi ity with the additive the second set a N -ri'· exp[-NE lri i phere c apac-ri-t V. --;h-ell h 11 upperbounded .ee as the reasonable a nci worr.d ribuion to F' toi of code words noise but without we we,2, have PF' i--1,~, ~~~~~~~ei t From the- rq:eni.Fet we see that- t-he are -chosen lon the re' iti clo.en i't seit chain nel not qui'e code word ihis ly is choC.i i ng The bes.t di-tr-ibuti On that--- su.-.ff iieC ' oding channel . -a.~sian i d5i.stribiul, lnn yieldls ti he same ex'ponenr dom E region 2.7 where that 2i i e:,-ponenti Te n di s.tr-i.but i on conditiona a. for- i for 1 ,2,.. than w'i-h the f der Gaussian i5 the bound is input a-mu il tl r,-i' error in = more rapidly bol iunid f -or rates and 2 errors, i do f-or error exponenLts. energy of i ver as shown F'or each set + -. the random coding (Ri ) -ec:,:..~':':i. on 7.. 4,:i]) pa-.cr: .A [ n; (2. 2"), from a sihe 1 wit. h a iau.ssia <1 bu. nort i--odina ens-emblie beis-Lt R. ~ R. ' >3 I1-----Z the rate-s decrease. the error error's. decreases so that t ++ ri~r Lroi | Ei:. R ) of code words. Thus. ], where from s-ection i t'4 -if a [i1i:}] Et t (. i is , constant a = +, and Ei 1 1 ' L i + '+ 'i ~ ')J i V r ~~~-1 'ri .-- R+ t 17, a t h e " Ks raties - -t satis:-'yincg sinFgle inpu.t c ha.nrel1 R(.i : t-`2 1+-.,-- . 1 F'o--r n ' T. .i by: ( W rai E. '.sr given , 1 l; ,.r~r - i i/. is (R i ) Iin I 1 .3. i -'- I i. --. , * 1Ai i .... :+ , (2-.54) 4 the sphere pack:ing bfound. fior gives. f . lower bot-undi the -2. :pE L Pei combined code words x wei thei-n ,cons which (on corresponds e multipi in thqe cii;ode words x,I (iX) = ' (2.'i. t.i' (tii ttppel; - for .. : - ' n act if -tohave enerv-y TN') , we would Q then be 1ito be orth3ogonal ai shel Choose each x and each w usinci i-lhe to itime the freqit..ecncry Oiusly. -.usin.r on Fi:., a bound ano" w(. previ N. simple si'nc c hannel continuous discussed i.' We n('::!W cieveloip ciersit a N(Si+S. the two codes the code words iof division less + w are not constrained. x+w to have energy cinstrained raining is the situaition increasing x to have enerQy N.S1 and w con'-traininl after i=1 2 ] approahes 0 with ' errorsq -. type F-or + o (N ri(Ri i 1or ai iaN" cod es, where 19- constreaint-. on independently the density t-. (w) using the w1here a::p 1 NS 6<i 1< ,.i otherwi se. wh.-ere iEPis an:tarrbi trarv; poisi number, constan.t to 1. t (,Ax)and to make CiX(x) into kYw) upper bou'nding '-=~ + ~ xl intearate (2.19) by b replacing and Substituting is a normalizino (2.442) sums with integrals. far and ( xx w,-e o.f inJ. r expi that (2. ) Af{ter [10]. P3 '1 6 + 1 i i U inn some tedious - intepratior', _ ,. +rr .- Fo r t he ca-ri ad . · · Fs . .1 vi. ntly v i mp,3rtan t 1ui eX i r i ver' car-n wih er'-e 'e L , r. i i for all andAi A. ·-- ·-- e - A1 b y sy .for is inrntrt y, 4 .:rV any by aN" 'i ation '- i ,:al n valid, 1 and and for F':, r:-. optimiz also X- it v.er o't"i-miz:ed 'I a solu ti on sUsinq simply bou.nd b_ 1, ca.s +F proportio na.l. to 6,:so we The e;.:oonent anld suc h -- a-y '1n.1 S 4: (2 is r,-, and are equal, ~ap~------~-~--~~---- ar 7.. -- . +p irsE: term in sEuitable a. elLqu section -i .- ci yVen i,-hoi ce o,- r, (t3'- in 11 [he :.fL-, (as + + 44) ' : E * ri we g et, !J. in+A..-'-~+- i+F') :.*.,.:p, 1 1 1 a product -form into bre-aks iex p I 165(r- r (F:,r~ ........ r, . , .- can be i an d but not optim al, aA f+or A A ?+ -_7 21- E the e xponnt- Optimi zin 1 E n. ( + F. .. - [ we Find 1 Li + that R In | -4 J: E t2-1| I (-'(1 I + _+ p ) + i 2 +:t 4 I i 1--+ .. :1 For R-:.less tha.n 2 1 the n I .- I A- (2 5 A+ : ptrI, P-f .. [. 2 I -A) ++ -:Z lower 4i or 71. I 2.t + /21 '-' bound in (2.49), 22 E(Ri-) 2_1n = + _i n( i - derived exponent without a constraint shell c hannel ,i with a shell way between and the exponent with constraint would resul-t to noise ratio the previously for a single A input Gaussian that (i. e. given by s(2 ., )- 41) ( = w:e 1,2L 3 samr-e .as i-he in d.ifferent .R FR..') r. f: (2.6 ,oCdes. ne were l-ne We hw i. n For I.:no o .. vanisnhes.~ :;)ptimum w.hiter ie is i'.2 :i) the symmetric wiheneTver ti-e ex?)nont cet.. is, ip es the co;nmin'ei an tisi..omias .s givevn b' Ri r-at-e .- pair t rae i nr rate-. r u' l ate aal R:o .here R.. R.?ta the:::tn e,?n ai t r -For optimr the poJi nnt w:ir critical the recgin R.2 into 5ons look i ar e -s.om he f or l? br-eak s t hv, ab-ve : .po..i i2n:]:7-2, thh e cri tical nal and l (Ri Eri 1-re t ndeed the ex-'ponent R2 Rif crbti:cal . si hnv that (or f Ri n l case where =' R1 enough abnh. e R:,- v-anis hwei.eever ,..R; and -,.. ishe.:: fonr smal bound. a-l thouQh n umeriaii (or R. t'henS eacih 2.77 -i exponents tind tihat the achievable regioin again e ir;ns I subr o-f the three we takle the minimum When i 5 A+42 + that -sig nal -_R.F'i4 -t + in(i 1] + lies roughly half exponent Th-iss 3 - - t.he thle i?:S phenomen?.lron n.t occturs to noise ratiio A-i is bel ow about ; Coding 3. Techniques channels for multiaccess is that technolcw, gy which a of of coding a transmitter-s to the simpler of- problem of always have where both sources fig. 2 .1 rding. simultaneously use the channel. ourselves we restrict the two input channel subset is needed is large set a for applicable has been little for multiaccess=. what introduction? but variable small however, Here, ithie in As pointed out techniques coding theorems very is quite advanced. to general done with respect of development While the theoretical sormethinCg- to send. .we observe First evaluated linear 6.2 in The arqument codes.. [1C:. each -1 any desired achievincg in13Lt p bounds for for- t(!-he s ng fr'-om.n sinqle Il withi C1 . ie :Lnt np r.nenei t rn di.gitf- or be thu,o-r modelling to the hearts- of en .1'ar-edex i s e Thus t herTin input channels, dij.Cussing decoding, is to m codes have the in order. a is brief The discrete als-.: t:iacci'.e'ss poanent as ,occur i em n;"~ pr-b or convor lutionaa. : .ci 4i.her bli ock as with sing le Bee:or-e channel section iilzation convor.iutional input channrleis :he s-rlne type ofi gen er at ing cr:tood codes ].em.t prob as in e tter-. been generalized ; uti of of to ensembles input alphabet.? trequency relative bounds codes can be generat ed tfhes-e binary thie channel ling -id.arldom cod clhanrnelis the same is linear sub-bc iflJ.rnr into to one many :i. t- :s and S well equally for this binary n general, transmi tter mapped apply last section thie probability that the error T'he with decoding. discussion of time channels dear information theorists implicitly assume that the channe-il iL element. Furthermore? u sualii ' assume -v is channels it ch rInnel a mul' tiacces increiase o the icord-s F. ra- can be viewed a der .r-ei3s .-. -F.er4back: v i 1mposls:i. I e ie t thi one C:-.'i" At h r. tters transmiti and anqroxxi :cr 1 y ImTar-i n.:r pepr i .ai: c..:p: ph.-ase chal nel t ie f'fc t -" t and baud l evel however, , the continuousi ? time narr-:owiwband shelel cr'onstrai-.7nt)r ie of be notfl . thainr inln t. .a it carr- ier o phase .. pase t'it fron syn:,'aronami.s reas o,-ab -- Ic,=i wvi em ine ri-l' 't-hI ., Ienera i ttl.e of. a For- ti-he spec< :if Gaussian on a receiver m b etJe en -fie-e is Cituat i on is arnen a ctino (al.ong with i ndeoFefdeln, nt a t.ynchoni..mn to qenera..te rain re'-e'ived reason.abl.e :n mcI ni*. i cai-i. i.: : fc ap i:':.ear-s - cofn nave 'nti r E. '7ti exceeds the additional -y-'mb oi svnchrin i- ,.v-chr-cnIssm r :.a! n r 3'.a! . e..?e da !' be fsai.. id ,ab..u e s a c an to S.I and S-., While thi s rOo abhl be sligShtly.vmore ate h '"h e r e *.ie t n a n!T nn f rom antenna ex-ect shioul.ic- C: iarasmi-tter iv is and that not e dback!ia. the receiver tin cC.rl'Jination. Typical liv frrom energy at r ming 3i certait i.-. is thiat n t s+ Firgy two transmit the - i-I] *are limited as a phased array essentiial l energy means- trhat the AWGii. -'Fnor that achieva'ble by a s.rine ond transmitters the signal buT T.his Sl+SZ. n eir,- Et ri individual respectively, knowwn [t] ' t"is f or mu lti acces- F or e'xample Ri+R-b 2 beyo nd tih'-e achievable sx. urcee 1well, is bu t of ithe-se o.f i npuzt channel s r-eas, e na quesi.Tionable. it perormance ideal iior s ngie .1 usua.ly persectiy i.s o-ten physical channels are p.rt in model srp aration ti--e time phase and sampiin.j carrier t a4ii urre thai .'a ees ...een h."i i(:: case mnu ch simplerD. en--semble code wordsC, ~of¥ r::r c= can ! at a. t' AWGN Usi nrq a (wiith cor tlhe discrete without a time rcode I- * wrords o-f lat1.:.ection can the the b i,:ck ol- 13 i'-'i o with i. ternate _: re. r ['uLks rom for wl.l,h-hane st-ati bio ck i i..1'r_ bi r. ..... fn ,rl r3 sF-' y '- een 'i;-he "! i .i_. . n]v.y I.-Ify . ; .,- a tni -,, d .. -F . c:me i (1 mu L:. ,i-'i acess b 7 rtt']i o(': L r-. ,r-f.a:.7. i.u .:5 'y..(.: ..[.. .... : chilangea t:,'hel i.:Frlm.,i4 musrt f at t. hr but b ut tJ, hi et.-e r Tn m.d onlyj oFt,:mh. n I vy is i -,f, - ireooa !o h r - le of: bi oc:: :s .:1-F 4t-L,o! tr annsmiter-s t-he 4: -- m ef-nds the reliat ive- to b e.syn:c.o f-it,.m..(=;.t_-._] ti a'et .-- ca !n -i i a;Jeqte'ie -luatd t qFied d- ;ast. 'i.e tn .. ti ;lt l e frr r transmi r- w r.t'l 'i. . £eedback .e b;-C.c: p are r- "tCrtt .: in - ai b ei t n os urnceritainty e sim. roan: r u'_~.' e hull.. is ,m wee .'-"Ln ' as w. z h i'9 thrant t i. pe- - ct t -fi: d r a; d -i r:i r .-r:, a1 hnr sy r:hr-'o i n For co.MlpLete iti: has.: beenn shiown The essenti::iaLl of fig, idea here is . that ri-at f t.:-- Im me. (i l-:o utC e ute.a :s.'.. i b ';- i of C zat i on e, the tunion region tha'i i .t ..... 3 t, ^i b.... c,'_ain]: const'ira.ini: so i ar-ge hand, i o n f o.r r- s.oo h....- b.:ecomes i n gl iigi +easible re:jion HI S' si a on 'lthe other n chr,oni zJat t-k_ e r_,r :. J/ 'i '''ir iim . inz a '_.I o.n b '. ar":: is .. ................................... t Ess.F'enl]t:i..al I1y c:(]lovex t . ].em of '-L r e.:i: i oin: P R i:.' is timingr, for each i-s r- a r-e c :-a .s e - t-e s- .. "is * .. cianl .i...-.. c:sL'e nd e' m tim.e not ti::ne 's-':.n c-hr '":.r i , sym bc:, sh. a-. , Tihe decrider .%] C.? v .=. .octn,-hrr.:ni..-..-ii. -i r cin i and saa.am,'1, end eff''ec f3i or non-- e Tne h pro rof i phace i me n c.de b,.ut will fi,-r t t ;_.r f .:i.q-t. r], s.'. -ti.u-e.si i ample ph ee. c ihoseL n wJav -F irm i r and o:f pro ress Gau s.ian p hasr-- and o ui- o vt-r'er carrier ig n : r-e). and 'i j--} rr***t 1.s j sU.r rmma4r\' we ',:t,n oj ned se s; set ly i. :aside Esome from ic whici7 ri.-r phaser car; o¥ ree· thie di s--crt:r. en semne.:. in a gfien - co de~wor3Yd~:!a cha rrtien ti i a aro -r1 w band s at.ionary l.eter .nnitn compc)nents. cons.idered as time samples over be th-tat .theii. ti m:i no unc.ertai nty [55] [20] 2 4 rather i-n that thani he its time shlar-ing rrann ot the total in be used absence o- betw.!.een timing relaitive the transmi titers. Having car-.i oned tihe reader inherent in channer-l., a discrete tLime miemorviess return w,, .e now. t to modael his prblem-s a.bout the model in mui- tiacce- ss modei asaid see wi-iiat can be 4tio about codingI First., cani n codes using .i I3i ties ensrnembl1e o.F codes -eet o deco3der tr aranslition for a Fniy w i" sin le and F (v lw ror i th e pr vary with tus s !'quential deco'dirng chiannel n. a bu..i.t n.umner ol Viiterbi int er:i. or- o-f thle achievable pairs, one of Wihicih, (R'1 reogion to i:' iv3en can decodie us .in satisfies , fepend ie- decodi0[ ;i (23 any ')5.- s'uf.:'" with this in (RiRi)2 ' or of a .i -'for on tecChniques as a convex, combination R..) !'nood': w reliably decodi ng catn deal g _lhe somewhat uncolnventio.na itr i :n p ro babii 'fi As can be seen f:rom ,Q..-, can be represented input channel dec; odinq is Xnthe trans andci cain-f decorde with ' precisely a .. by Over- the Tinhv s =. ir om w to -' rL channei input is th is W en=.-mnhle w) encoder a singf.e sc3i-ind i n that: rs t i'-lputi sin-gle Ti-iis inou.' ts j. Over ithe .s 'P(y 1i rabi. decoder- for anotier w, Q!,(w) i (ix) and assumino- a single code word proabi iti R| .that < i x a decoder can deco de the w code word -. tthe x igionoringoi transi assignment -For some iY) source decod:tn.g sa Jsi (R 1R.) Firs-i supose prioblems th at a pproach tihe decodiing problFem to -several single redue * (L. si.mFple general fair-iy a thiere is nhe v en two rate "r]ias F.i I and the other tor each of F-'i nal :.v z-rate Ion'-e'.in is ich.i sat : is. ies two thi-e .interi:oLr o.f a't mst 4 c des. n resp vvo ].... Q and :i i ti:. :for joist:i; error that inrll-)- decodinig of decoding cecoding.. tri is iNote. by ,) can be decodedi of the (._ however, o-f which s i i b:et"weer' atih error 1js -f:or two e-poneilreln.' ts tocgether decoding (aL ;:ac'f'-cr' muc:1 smaller simpler that p4oint satisf ' the two code wor-ds potentially crsle U.! -1 t-yfr an arbirry rando'csding m of e one of by' t im e two are often pairs, leq - uali3 i ent:irel-y at measure the achievabi twou rate Thu . s ot her- not -' -:his appr oacih ex.porne"lts as a crude joint on. o *' -' thPe is a.th.at .i:irst in is (iR.. 1' the in terior whicih-lyn u se rats :-.2 T'ini.s approachn p::lr'.:babt: in tiher h two of :X.21 such codes. R can be reL. iably sl'arina decood . e[.'s.'te above and comnonat;';i some , X I (i Y swith flr : t-hese rate pairs can be decc.ded point a convex the o her ecL a>ny i[ tiA;Y) satisfies Procedure described between : F:!. iR I (X by time sinarina an which R ... : (odes the two step of .. ti-iW f'or- ei-rror-(]r{ tihan If we u'lse simpl icitv, than the a'bove exponfents those can we see single be m.sie.ading cofmet'imes as a the random coding example. channrt- el i is rnt- larae, to dercoding guide for exponent to pcrovide much i nsight sources. a small but. urknown subset decode the channel r-egl.ican is (/2 I!n Ai / (1+A wlnc.;ine A..:is sma'l. c ase -r C''f a as a P se a f::ilde. m 'du uen.,:e or ) .'1. i the advantaCe numbeEr a , perhaps simillarl.y individual Aal.yti case o f arbi . i22 , transmitters is seen ratio- to Ior tihat ' the ac:hievabie j I (X i (X:YjW) = (1/2) lnr[i+ 1] foro ot iecodin .i joint general izing imm. di atel w.i'Ln an unknownsu ..c56 is ov oer a giien- d(j i n g can where period fl u it i a ccess -bIe discr-e r trary approacn . to o- bset t t':- e viL Led time i- ways in ito chioose code such a way that the use is sifmpi same erIyless rr a o-f v an p ru s: i t ii f .A s. e. with more than two transmitters investigate oFr SpreEad spec'trum with pn se-eruences ca t5Cscr:ompiement p msn I 2 it 4is calily a to so:)urces s ers. raptroach the exponent err`or o ai to the use oS tra,:nsmi4 tt 2-.5; to noise close thtis case. special i is it i.nridependently is almost the same as .ae For" an channel, transmitter nal that joi nt decodini -:, of Fr-om fic. si are trvial.. is leiss gieneral, each which co : I-. t r-ai i i-i: o n t iie e ,nl I.at 1large i but small the scurces transmi ting ise viewed a as noise. In approach ha Ls lhfe of Forl binary not, gen: er-a ize almost r-ectan;i:ul arr. i:di vid ua l decodin hhi -ruom with rate i does simpier reqar dirng the other the fAiWGi certainly coode wora-ds approach into the qtestiornn of several A second. a noiseless whereas coding and decoding The oti-her o bji-ection to this It complexity. n wa V m .nltiacc 'ss one can word sets .for t'_he they are mutual ll non 1 !-r er i rrmorre precise i decod ed wi'.t smali 1 code worrld o!e po' -sii. :ii j. . possib iiities chann el, word mIore A sets property in pr-obability). ty, We have t1asv; and asvincnron'ousuolii s time mieoylesv it- the non-intr'i"--ferencen srt fymboil that ma, t-c i d av.ar- c-f in this ao-..-.ther i 1 i [14] ari-e known c hian nel will .I 1 al:::i: i:. a r ::.ir;i::.:i!;ned: poss.ib.e number oC:f 'tates-,, s i. (..,- s, o-.fal gebraic Vi.ter-bi thi-i s objecti-,'e; at discreite gener'al i-not coins:i. der a stat:es The pr-obiem With morle is tie I be am n:)ot geerral r:r convolutional codes very- promi-si ng decoder is t1ha--.t tlhe .i- ich iod ier of-: both enc vhichi- and hess appear igebraic must st -ructure decod ino of joint true consider charact eristic- approar: does not buot o:"f sta'tes p roducet is as a shoui.'--: .ea ds to a i ndi vi.dual. oi-I tni-e than t wo trans-.mi'i tters the probiLem - Finally 4 approac-h qso in teThnique. tnum.iT-:rs ' evenl' to -ense .the channel iit possibi0 t:i. :decod for' the f-or to decoding, c .o ode w o rd-s any e...¢amples .j::i.'i: be done -for the ca; ch-annels. e s mnl.Ai t access m betiween s yncrhron is i 'Ii clchni q ues here since analgebrai oerciii-i that co e to cinoose the also achieves no, approaches As a thc-Iird approach decofdin,.'' of problem is codina scheme on channel si inowev er, ' seen ot-er AWGN lack or o¥ .ithin a as we have seen for the as to maintain the presence AWG3N c'.hanlneil Time sharing depending on the channel, r elated difficult tter-. biu t be preferabie. su.ch a ,.ay :in thie transmin:i i:prese:.t err-r migh. they cran be indijviduai.v so that sequential to mui'i access lack of block decoding appears joi nt decodi n synchronization is and it to be a general has been not a serious shown C23Vr impediment to Un-torun irts operatio. il to meake sequentia expl i n t+he d ific ity sequence oar sie rcihes nt f"l1metriic: backwi.ward st1ocastically tkhe actuai Fiti.L owin: t.he d ecoder now, ini thr,~ee ways in secti of the d ti-, recent. Y srme i:,or- erfa"_ure: E: : ,,- y = _ c:ael cul.:ete de'::-deer thie d~~eA as is. decoder can nIn oI-'f thle te anId madei ypoes of tiree errors to use metric inowltedoqe is this the Arikan (eJe)3n} , cdirrQii n g c--insiders ih a L= With proba1bilii:y . wIiher eas with probabili.t.y i-x...w) input, with i i iuLt [24] Ii ,.Jher-e I3 -- u.i::i).,') i.' ? i 3) = e qu i p: r : b,a. b e a when problem with sequentia e rX ra u;-'e (.3 cnhanneis beino error discovered by Ari kan .ce.s binary ihaInei independent ot the encoded downward o 1the approprl i ilundair-iental bveen mu .L__:La- i to a fcoder. Another has-. up the val.ue o. and drifts to ti= r.-resondin jdepends on the type uiirnown on T is a sequence by e)x'tending thi is'that U! ir.F-oirtu-nately 2. or i a nne ,h.w tr o-fi-theiack.. ets The problemn track lear- upward when the decoder sequence encoded not cha iaccess thle encoded deoending drifts is sequen tiall decoding proceeds forward aind either pc:.int thet ecall that h-ypothesizes search pro-edure gi.:ven r -for a m ul work dec.ding imeq i. ti i nahis at ely . f:_~eah -:' i.n (ee ef'fect tof'f Cnalascomp I cu ~er co:)mputeaiti 1:'d-7 ti- t wo sures Ln.iiera ng correl.ated we : ransmi we have w reriion R cean 4-o r.'l for a I y1 jo:nt _;W-.. i---', a 41 rCj ol F._ ' i E . -' 1~_:5 · ~, I=fi-iLn 1 ) ( | 4 I ' E ' R- 4 ) -(:3.5) we notfie that j [n > rihus . foi r- R ' 1nr - = R. ,. i is 35) ; E all the active 0< (3.6) r constraint, 1. :: and evean the mcetir-ic problei-ims discussed above.(. , 3.5) th1-e achievable rate However, U s Ing s eq uen-i withinout ary se D-parat e (.3 *. .. ma k: th e Ihi e s it-ati:ie, co lmputati icna -t io- cnvoi ai J':L n riesi: cnahnnel iut i ona e :i.n 't.h '?.'.'- i- hIe potential which tt:or encoded are as f' se. we r-ate ,Q- a likely digits and decoders hI i gi-e i h'e r- r. partijul. L :ofcodes ws n as two binary r ens.:emble o-f cortvol utlri pair;s tl e twor I aiel ati on i we can ac cui encoiders a'c :ii ev' or decoding. es eve th e binary as the pairs o'f /- '-] 4 Inpurt np t,a'i - -rnary q to the anld using separa:te the two di qits, m i s-. si 1-gi 1 -- bei ng inputs) , the expected (or higQ Iher rat-ies i n I ( it . ere n- t d.: Over inpi:ut number sequences) transmitted sequence we;e can used. cofdes 'for the quaternary sequences and t4h a t - d .i i:' , i c i a igor h i search ,llal 'h t:or limits transmit ters -1owever,.r by regardingq the inpui-ts er'asure c hannel.i ~~quate!·c.v sequential al decoderis e corr nd .F "ti:3 with joint th!' S'3Lr ei : ci igr -ri a .so3 o-f at (or- pair) the (o3 of length N is in exponentially increasing -in[ is t(li-:') ;4]J Thle conclusion not really a same exampile, discussed fundamenft.al by iMassey much F:rit3oi.wiedge e:· i i~% about paramet-er coding is cess rate reach in excess o-it is. that Romp oft communicatlion. channel , has been 26]. ary before any chesive and decoding s 'Ti approaches to decoding, previousL ithe must photonr the and Humblert 25] more researc-i one that the contex.t o-i in -ummrarizinng that N for- ainy combirned for multiacces.s we see body channels_ of wi l3.l 4, , COLLIS-3ION RE7SLHUTiUrNi iThie coilision co;i'municajt·- i_4ior, the channrel i'(t vsites -E. approach resolution as menti'oned among a ias in large set- the weakness ormmutniCtcatji assump iions. t(hat of users on ClOCL:-e alloicatin at different oT es-sertiaiv ly the class limit mt].tiaresfs section i, of the problem. aspeits tc of transmitirin t-he gnoring We start svsternms by a we will set o-Tibe cronsi er ing liotte a.) starts transmission small ti me of clotEks, for slot) lonlig pac-ket- s is ends before Not'-e t'-hat packets. th:is assumption d o short. car-rie-r ng , sensir sytems can be undersr'lZ.ood r "i which the nexti: secon arind ,-vat dsci af ter tihis sitrnlmpl'y eacih icult ,d,a thir d, precludes both t' e t r-ster make pack1et.: of too dif usuall y not betweenr spac to be transmission. f-eedback from the receiver, timing aEnd of time and zation QLuard of in send F.possi .f:i'liy Suchi synchironih a smail aimi.,.:(nt s~t!i:able at an integer Sci n first, gCivi en, (a (pacl:ket) hhers are synchronized so tnhat the reception tratnimit i i l..teerr each message that f¥,its into one time unit transimitted l1 We assume S ystem: or ons Ltss la_ter as-ic madei is undd e r- s t' oo d:. b) Collision i oCn':i.:. '-l'*"ansm:. tn*i iter co 1 i. si on or oriCiin? sends a or . p Ferf ect _ R eception a s send s a r a:ke c an irthe receive.r of packet the assumption i- n a cets eino .. llme a slot, that ·from t ihe problem; it is i n. orfrat i If received fhW a': i-hen i:her'e sl ot., the transmitted packets. in i on about itfh : mo-r'e j.s a t'he contents jlus.t one tra nsmit'ter with no error-s. removes the noise and communication it_ allows collision than resolution This is aspects to be studied in th e simplest context strateoies and c:) Inf inite but also severely limits the class oi- tradeoffs of Set that can be considered. Tr-ansmi tters: Assume that each arrivino parket arrives at a transmitter that has never previously This precludes received a. packet. transmitters queueing iLthat, a finite set of regarding each co-ncept-iual packet transmitter. on a .inite set is sionir'1ficant ronl u sef,l 1 or at one transmii tter; Coll i si on sfet of tran.smittters i n this The maaximum al. g or inm an a vod d dt i mpi"t-ovemenl-its achievable d) iAr-rivals: IAs'sunme PoF3i.3sson Poisson at independent e) (, provioes an overall arrival i trans mitti-er wl-hen queueino 1, c Immediate Feedback: the p1ri marn.R pac:t-c. rarely iy quteue with up a finite o-f an algorith-im under- the of ithi-fe g'oodnei measure un d amental less proceses at .waiti nq at approx i iated by that . wi t h an s a q u al]iitativie r-ate X. the difference t ihe pert e rforman c:e thrcugthput that that al gori thims ar e reg tcin i e case bound a worst paickets are multiple sh ould be l wel :i.ri-if: i. i-te se't assuimptiio (:',-f tie rc resolution '1ere a separate colliding multiple packets. two or more ratesr. to this case, a physical assumption when low inut set. in of trans*mitters and, second4 · ane tr-ansmit ter. inin:t nt the transmitters send as corresponding simultaneous, that first, shoiws, T-his an transmitters can use the same algorithm, arrival wo(uld sometimes send is point of view. but note gqiven any alagorithm determining when pa ,:kets? individual This and precludes the use of TDi'i. unreasonable assumption from a practical at tinhrt occurs at each hrip ut transmitter. nekw. pac.ket ar-rivals This is reasonable the individual are given nodes. Assume that by the end of each s .- each slot, learns transmitter (c more than one packet slot. This is immediate feedbi-ack algorithms c of is benefit e: dj .scrusCseCi in wnhat It f+or n Some alternative feedback; compliicat-es of i, i'l the transmitters idle also'- of resolution assumption f-The between an is;teni col.ision however inigh:i (or gets with delayed feedback: e in transmitter that and a collision. slot i-iat implf- thi feedbacki s forms s of idle , w. fich feedback is will not be foiJ.lows. SLOT.ED ALOHA The as'.mrit.ionsr of form stimples-1t aab]v' is desct~-ribedi arrivyes at ne:.t slot.:, pac-::et involved remains nbacklogged sLuch backlogged ofd: the In the untij. packet with some fi:.ed probability of other packets. A.l oha Arloha is it iS is. saiRi 0. C to t is in tran'E--t.yi't-ted a slot be backloogged independent be a packet in each 't ted wil. iy I and Eacih subsequent of past in each I[y transmitted. suc-essful transmitted p whenever occurs Sli , [i2which t-hat pack-e a coiliion 1g i ' in (Rober-ts L2''1].o (AIbramson slot'ted collisiosi"on strat resolution ':tra.nsmitt-ers. Whenever in isi on pur-e Ar]. oha subsequentlVy. one col Sl otted Alioha is a var iati on 0 th e but the receiver are alwavs alway-s desirabl th-e th not alwavs reasonable tranmisMtte rs 4:, 1 in delay in The assumption to deal be modified can distinguish themselves) each often unrealistic. implies that feedback whlichi is little with that or 1 packet. were transmitted collision) of packets elsewhere. can usua.ll the intr(oduction analysis for the only information the e<ist-ence about C) packets, whether slot slots and Note that if p were 1, backlogged packets would continue colliding and no more packets would ever be -4. 4successu-n tril ansmitted. -effect. iveiv resolved irlfinite of a known set staie at number slotted Aloha pa.cl-:ket is n'"e,,. i p.roackt ne,/ pack::et and k+i ttate, trans 1 paci k:et artrives be the state = anyv in at t+l. is l rly =i if: same way, s probabilities fk. k,+ are given by i L-p ) [ 1-kp( -p) El .4e A e (i2 e k =.'+ ot i1 eitler o rs ino or :Cone transmitted. the cases is the back1 oogcgged orne > i in the i :ftt ttfthatLLs ti'ransmission successfu.lly but does the number and Sim nila pack:ts to t Thus 1 es -f{t-'i as a Let k be the r,{t+!? r in {t Ct+l in and no successful :: p P like time t from t-l to t. in [It- ,t) arriives ar'rives and ition d;etermined of backlogged the slot from t-l (if transmi.ii tted Analy7'zing be can be used 'The state at in new packet arrivals no nev slots can be analyzed using the number o.F new parck-et arrivals if of cannot Such strateiles as the state. su,-,'-~='=~uf l tran.smissions = -1 the looading. time t time t of and can be made to behave includes packets that collided n-0ot include some number transmitters M'arkov chain, at each integer because the collision the transmitter. can be seen that homogeneous that transmitters. waiting of TDMA under heavy it of bv transmitters by tile identity with set also Note (l-p) _ that. t;e he ee i e -, 3 i 1A (e Ž2 (4' . -4.i5- at the In understandin(g how this chain behaves. we look first drift, D (i-e. at as the expected defined k. the e.pected -[(i--p . .. at + on k conditional the state between on the state t conditional - difference i value of at t+1 and that t). kp(i-p) ] e The first term departure rate or throutghOpt. p will positive for all sufficiently larce i-::. This .:. CD. k is .x be the arrival rate term is and the second >: ( f: rn any X N ote that the system becomes sufficierntly bac.kloqedd drif-ts in the direction of a.i Js..,") gives a noi th e when but i simple unsti-able legarnt ,ge .t ly suffic ient this otf slotted oha .l approach especial :lly [28 ,hain is type of thi,-at on almost i-.ap lan . are proo-f instability resolution collision rl modii-i:i ed since collisions occur i.e. 'i non--ergod ic) IDesfpite the use'uftt backlo it more and more backlogged; becoming be surpri si ng the and, any means that if thi-s shoul.d , ( tc:avoid or 5Using a.-. small c-aLas'trcph.ic value r'cover oi: b3ehavior l;iC .from the heavil'y' the syste -m .is i-f ibac-loqqged p helps pos'tponre tihe on.set of iabove, and for' simall p, be a can stili it ii(4 2. state. the ca-i be weil an p 1r ,! i mza'lt-i d: b y h: 3 '.X+p *I.. -4..-Xpk3) -~. Fig. 4.1 - :: .. illustrates this equation. > 0 'for all k For .X .:: e-, For X > e - there is a range of we see that rD. k: for which D and a '0- the size as p dcreases. riather would oi retrials ot this range Unfortunatlely not: X is the aririval p is stable recgion. chiange p as the backloqg solution The engrineerinjg k::changes. this solution is5 -tunc: it snoi.w.ed ims in f: the Is i , .. nl' y a c.o':!:d wn^hen p p is functions twou.ICiij:e to ei For small we see Dk that a prox pn- i nmat ion :is sm all .D. su fic:Lenti .-. Aloha is unstable in Thus. y lari k must each,,islit be p must a cl.ass si m1 for be o a1 in ormation. feedbaci anyrt e and f.or ao l /e, k no matter this (4. 2 P mu st be s .::> smaller Dk is i: T 'r e si a' X .. 1 /e if ut the app r-o all. I when k positive s t -o is i mat ionis 1lar..e k: is ti . -for all chiosen: so tha't how p is case even delays, w...henever i - posi tive the next subsection we s-mhow that and presumably ,: on tlhe The problem with 'f29] have analyzed ca, n be chosen t For = :., s yis-.5tem sta5ble mi(r i:niz' I In we or an appropriate value of ,utda-ted at t he almos-t obvious-- and either s p and the p '' vir'- :- !he-. r e-ultino3 -'riom -r:- i 'hiis unknow. feedbiack: which-, t;h-,a--.t su..ichi ima:ei t t'hati k is I--iajek:: and VanLoon 'ion packets i n Dk, which occurs at pk+.;X(-l-p) ,from the et-.tim a t ed air.? - p makes it p is iarij..e and thus delay is small, hiand, estim;aa;ted is Ideally. larqe i.:,this miamintains a t-hrouSqhputf otheri- between delay ied reo ion, large delay for coli ded small. 1 p ca u ses adj:st p to-iminim:ize which v.:e rate lar-e very und-eirable; very easv to ernter the hunstabl.e heavily backlogi w er ea and pac1---:e. a collided in as X decreases L p means i. ar-ge and smari decrease, This tr:adeoff increases slot-ted known. much higher -hroughputs, are possible when newly arriving -4,7pac.lets.1 a-.re sometimes 'iioh Slotted Aloh !no-- requ.iring all many phi-v-ical muitia accs-s chiannesi- a collision uish slotti-ed that- this :~,how.n that- slo-.. tuted thanrl 1/e( . he:never mo rdel oe .ed tihe ou.!r ass1,mpti in t' Iha t [1:31 ot eac h h ' 'li. :m-.di.t n :tateit a dispersi.ve from idle chil-ack,: is ,to or collision, suf+icient and for tihe more sopiisticaLted much miore dif ic_.ult tro erstimate Cruz [4.'i]. hnowever, -for throu. hpu-s ican be modelled rmat ioughion slot passed tihr capaci.ty: and as the a inas: .1=-ess idle,p discrete the ca-se ahbovie '-an wayv. was the precur--rsor slt'o-ted pac ket: svstem. requires EacLh newly up on arrival !v an expolen -ial y coll:ision it i s Fior usually feerdback for eedbark g:.)ver- i. api. c i. pac ::e ts afi:terr transmission i 4we h a ve a-- sumed a cvr-lic redcundanrv of positive i n thi--j F-Pure Alr.oha a) use of is Aloha ccan be stabilized c ,annel , and It i. n more inantage of nauJSh.from a n idle dii type of-ieedback. c i t'F ::.nsi or memorivis-s--:2 I, -. that . tf kind of rith this the bac*kloog as.ul.me -fic. ut Unfortunatr''iely has the ad v parJticiularly Al.oha but not su.fficient 't.. .ate-..3tem-. ob throuih a success.ful it can be seen · i . - ' -.. f h h-owever with high reliabil.ity. strai;-ht-orward d:istingo:§. d are resolved infnormat-ion fee-iback c hannel-., i.. t is i.no t the colli -ions up and ra, wavs istiica-ted iheld c. :ili de, star-ting anywiere at nre thie than i ed del.ay. in time t interval a slotted will collide (t- lt+l). we continue nf as ' sumi , and backl oaaaed . i.s hignher here transmission that arr I ved p ac ket av-oiids unit. for tran'mI si:,:i... o n. i m and Alo ha andf: slotte ait'horgh onr e di stribut startinO 0 z c , ions i t rar ans i mi't t ac-et-::ts ar-e tiran:-mi tte'd The prol:-a:i.li syst-em; a ty of packet with other packets The upper bound on throu:h,.- h:oui: becoomes arise as t,-r pure Aloha. die-,erent In - is however, ALGO6RITHMS slotted Aioha, is upper bounded by lie were a:lwa/-s transmitted and arrival. thi-e tr-ansmi iton f cofnrsi::-:er a slot" arri-vals , were netl ru unt..il c,'ot-h~er :ri. t. er!'at,- p i en-5 . 'L si.:: t'- iy *5t? t then, 1 prbi. S i- n:iio W':i. w-..'. f. e, t ran . f i:]r-oa io' ti.i. i-../ 1/2, t e d aind in sn th e until 'The resolution yields an effective period., wh t":hen a coll ii ding pa.cket:. in i this a.i . ot to- tv1./2yi the arn slot. ca'e, enn :tre W:i.th ur-s c another colis io.n or an i, Jle each pa.k ti n n. v et tnth w. the two pack:ets ar, e..pec(teed y transmit tihe two pac.-ke:ts .to s ucctessfull1 which no idea .ole., the -f;ol1c.w.p.n in fo1.] oni wi f.nr-t:h ' . Y t ransmitted . were H;..A ain, lot. ive If - t(he newfi. rans ision 1/2. tbili'iy ]1](:i s intit. nsl-wiwnith oprob-..iit su a j their af.ter a singile cn1 J. iide ts each be for w.would w :itH. wL.:.. l.b Ci sC:- the ne>xt slot rcollision fi.,oultd be tran*smi tted .ac'e-t vei the ofallog ti:he ; :!.1!.es' /2, new should sometiimes be postponed, tlw oia,. , packe whif'h in nmiL:. tism:n'(: pr - b ab:-. iiy of collided To get an new ar-i.val reaso~..na.-b.i.z.estlrategy re'. ].n retransmissionn. . the strategy iupondepended back loggled packets that p for parafn-eter of regardl-ss the imposed by the restrictrion..n that This bf-und was arrivals we saw that n probability used to adjust the retransmissio packlets. to handle packets of . our discussiion o+ throuigiput issues advanta.]e of practical Ai major ability its iencths [-2;0,-1] st-abi i. ity annd the same k:ind .s o.f systemy the slotted FSPLITTING 4..2 (2e) throutghput of number is of, Slit.S required to be easily seen to'., 2/:3 during the collision.r This concept of involved in pac-,kets non-transmittirg idrea ! of achieve thiro.-ih'r we call i these :or a:il owing ti r'ans.i-t_ of la rer .involve MIi t et khai I 'v i [I]4] i et weenri-.w. tw-: cc::llis.ior'i C:::),t 3,. iLi, f·ir, on .i: m. We r::,::r o.i -. i n: i1o t".:"'.,am i. '.'. t all lmode s v g.' : .+ r-ir h .ri_: i.- :o i i ' e t is is 1slccessrful or' idie, in t1hie folilo wing slot. tihe collision thietree al of t on-].r'i- I'ticrn general: ii.-! s1lot, of t'hin : u bb s e ' :sw i t -,j ef-:: oif . t ie . r .n ?.. i-c i S h . l : c. f u ri_!i-le r T'he firs't this the subsets is whenever .1 :a.-i i ti e 4.2). pi and it _. a b-ra *) 'i sseu,:e b:.e :e t he branch. next the netax nvl] ( ee ee f. sub s e t s moo.de= -gro irto ! in'--i o f. i tr s and .ltl.-ernates , .i.. c.or-r-es.-pondi n tr- iithe second In thee sY-stem .. ....... c:c. i. is:li s onj occr i1"l the oiri itih 'ms and Tsybak::ov r pac :et's all thlh·e j-- the rul.es in a corllision'tLo in .a. i_ as tr,o These .wait unt-:Ji. '-Liews.. tL o s .::sa i.l .er .I. tr-anrlmittetd ithms. new arr:i.valis b:i.i ar'y hEtt r ?. I ode o ne th.at assumotions a) tral'si-.. ' i ttrs haches . rQow:i.G g fr.'rm:;inal the or i subi~:e'{:s al.gorithms ali 'i land s an .n the ved ac mis mode, i ie i.- ed Hayes [33], ns im nrmi iiC:3r-inmai u. re .iso 1]2] ti 'ir he r-'oo't:- a,': a. reoft:ed s':.1::,: c t t:)r --- nr i rIt fitd e . ].E d e .i. i nvol .- and a is wait f-or splittino used thiese. allionritms :tm | * .::i d reso.1tution io nI7 .'i.i i.t 'tr-ai.niisl-.i agor al.gorit inms were a..:'C:]. i si on :3. oc-rs 1 : When proi -i:)a modne ngi set resolu.tion sp littin co-l. i si on is In i....... transmitti mor'e tihan two packets) :and by-, Capetanakis (Af- the seht- of 1/e while u.-inc pack::ets not t.e splitting other packeits colljision the rule-s in The f:i.rst sp:itt i developed into a tfhan algorithms waitiing a t. er sion while makin:ii s h mi_riht (i'thi, i a. variety aig orita-ins dji.ffer set a col seet -centra e) probabilistically o'f these tr-arnsmis.:.oion transmitted thie tranrsmissior'; o-f -4. a subset results branches 1of a the tree transmission is in of a transmit subset onel mode is again resolving eacerly be awaiting will re-sounding collision packets oT the preceding number i oe pac.::ets-. br-anqcihes en'sue. rrisino t. irou',:,u'i set - (. incre.ased even is to period, is in is that number oF mode; randomiily with thie lenqgth period s o that of onen the expectei 'The rom the root and two node.c algiori thm or- all attainable input has a maximum ra tes :less t simple improve The first split slot an with tree a.griorithms ment into two subsets of following was first Note what the algorithm does when is the at resolution packet inicreasess this stable that many Thus the a new collision on the order non-leaf a collision empty. 1. and each waiti.ng (0.46due to a [35]. worse is by eliminating the normal sih,_owed : an d by lMavi-xssey involved the first is kp branches ri si ng from each mode and a ollow until the expected 7.4~ maxi:... T4The um througq-hput suggested What resol.tioni has a .petaaIak:is [1.:2]3 later many new arrivals the return to normal per suseti correspond:i. n... triee exhausted, algorithm spends many this h-ie n umber-r collision subset used to subsets have been if entered ..oins one of ,k:: k s:-ubset=a. is t-ihe entered. res olution is- imme diatel. y t so1 a subset becomes on the order of in a collisioi-n period the next Whenever (i.e. the successu-i When all will 1 algorithm can be improved if packet) , will e collisions sucessive enld idle or collision, then typically a sp.it and two new flrom the old branch. should be apparent that Itr is the subset is the nevxt subset. the normal slots collision, are grown empty or coan tains 10- the collision the which is -4. 11then is a idle and the ne-xt in the first collision., collision, involving all the packets Massey's improvement was to avoid this predrictable collison by resplitting the second subset of a set whienever the first cllision The ne;xt improvement and somewhat and later, with7 Miklhailov entiirely. since is the tree First. al nuimber orithm. ofi packets. 'ubt- hen, sets, pac,' kets i t.-us, in as far tati.-,iisti,-al aid t hz 'Lt 1i, t le l.y we view it of collision nnc get the two subsets is inroved i-, e ntit The easiest waiting If all packets it the packets by meanrs of that arrived of e li th e ti ng of distributed. new arrivals, than waCting a wai ti ng ne w how to do this sec ond r subset is and subset is issue of divided, as a then we to consider. way to do the bookhkeeping is this t ime the f i rs-t s.u Lbset more than two subset.s subset the number PoiF-. -on ra.ther t.; 'the b-ookkepirna;h Sepa rate have in to ;0_some tinme int erval v be bu,-:d o 'ite two r-esul concerned, but note 'that i-f we eli mi na te never FPoi sson di stri on these colli.3si ons, .hortLy,i every modrif ication ot for such a the first we simpl y tr eated s er We will as another occurs in as the algiorithim i identi-ali later, than the tree contains a subsets cond-.it:i onal algorithm would ar ri''ls. to analyze occurs -second structure the ewid of a collision resolution period, if' a col lision the tree algorithm precisely easier the kI-niewly found tihen anti'lther i At analysis? by Tsvybak:ov elimirnating describe this however, empty.. was du.e to Gallager [36], in throughput involved considerably al-g or i thfin. is a more complete this We shall it eac.ih of [37]; suo.bset the arrival in a given concerning times of su5bsets and the paczkets. time interval are i.s - p:i:i.t two into suhinterval iir-t a in alw..ays sent time t packets that vi,eC4 sizer t"he pack ac: e't:s a:-one tr3X~qS d..!ari 'i ndt . .... .. atFr: a .i1r1i me l ' ime I'i--A:t , a ortI- i].f:.1,a'. . r.iti. an tr.7-nsM'i ' the wai ni- is te hi n o: t o do sin', best .s. o t '-ino p'ro" o .ab th al.l o,:.s a hi i; r it a- i(t,-E:...ii2, anrd i;c: t wee '- cn c -ha or ir si si5.... e is .in a orc)the tributaed ept-... s1lot i t i qLuee 5.a .s tf t ihat ssoi :.sh t e. :.-c.i on .t ne':: i.n th e i n:har'th. no:e int ri.nO s*oii slo'i , f:ut terms : ts attemp one ibut t-,fhe the queu .. . the to t he qu ee ued I a cn'ci s!-uc ce in based dL the succ:eedi the and help-ffu :... , ..'F. t he given below determines it(r1::. .fteedba!ck arrived ,e jt als possibl v se-e-n in throughput he for the sl.ot Ett+l) the !. i nac V,.:l.=-. i.ti o eni. e e r ron tt at t . i IMit n]hus th-e a ].go r.thm 'SS le wi th an i dle slo i i bib o Th'xe algorithm 1) .... l p ac(:ketI hat....- :t i.:erva-l t]h.e i..iii i:art+nron]'i.: r an-,. W.'-i r .........-.. einT ts at;r"t t. li k e -ito a 1. 1ccat e We v,outiI 4,3) tjr.- ;r ar' sm:i. t- ter ,ma.'..' i m i .z [ s 'Tt at transmitter by= eiac-h time .-(to ee,]bari.:: the arri-,a.ai -ffig queCe c-.ee that L. .e cali nhe time T(t) are transmitted. (t) are determined -_lf ert Sup.pose the packe-ts that all ttt+i-) the slot ilistor y on tihe s (not necessari:iy some time T(t) before bet.4.;een r.(t) and T(t)+F(t) the interval ordrF are transmitted all has successIuilv the algorithm arrived In inte er), (iLFC S precisely. express tihe algorithm integer piacik:et s approach, serve t the in algorithm. a FCFS splitting We noni fir come ihis With t.ir Si-. the tubset- and those ir-'-t as the are regar-ded int'r--ival in and the pack::ets suibintervals 'eq,,a.l then the a coili sion resuL tsr and the second subsI.et. second as this a siot in tir-a-nslmitted p(t)t, of r T(t) ' p(t-1) [t-i.,t). annd T(t Th.e - ) - state Q(t) cr-n fideration. , _ ' Lt-,t I has been 'Ti' r otherwi.se resolutio. .:imum s tablie intterv-a L thata -iaor is w.a i foar itin, itf i -fetadback 1R i'f3: t: 1n --. as, o a i 'i,': ';:' 'i . : c, i- sb ib i. :_-.r, i- r vl.a i. ., e:5.: i sten·ce · -i 1 2 .. aof =ubi nt e.rvi a th-en - (t :2 these tw] t s ":orrgeft " iabout part of the waiting z. 'I6. 4* . t ; - a 4t IL: - t4 t . +- -. -'Tr n[t -a1 -- , thei it at this queue. f Eq. nto n Lt t o 'is,ia ( t-1 ) +2 .(tl- there f: (".44;, two e J 4 -h.. -e il.- i "r e r mmeth er h t w.as a previous the algorithm regar.rding that pointed !'-'5 .. -vi ( t- I·ii If:x point, As 1, s. subinterval. t-1( ,.5.... f-!t) lILi1, nin =r''v'. t ( t-i .. - rn tim=-s a rrival _i. f ,and ni.(t-= F; ..- - on - iJin ~~ ~ ~ ~ ~ rIti-' o- oc ati i:-.-t3n thV- ; · 4..~~~ .- the ai i n ter-,ai L ad.t-, r- -!.:i-pt(':--'5= min....!i... , --,t T(t) iT( 3-3= T~t-)+p~t-); -I .... thhat to achieve at N teth 2t on ien - f.D:eedbac5: = . to ra1.nsissmioLr n ( i,- = 5' .t, .'+-) =r t s It t - him: t eedb f= ,.:'-: " and a-ter a cllisi out It turns = t ! _e t - i.. to be used completed iy the intervals ,-or s. o, iunder- a p.arametera".t tlh--ht determines has interval t t i Ini- ,:mi a i., .lanl f l:i msited by t-T -(t) rt-e still F'F -1 S i thin crtrrentliv . one i n--to 2 - or tir 1-ui. oput .i-., is to 2 i set d5ivideid period n i s alLca ti on c- -ithe size Ta 1ofh sub-' . i .-e.--e -In _-. thi-e nLumbrl, Qr;(.- out before, subinterv..l the number .as oSt -4' 14packet'=i. PF'oiss:n is sub'sequent 8a. o-f the queue head t i- sn nd rr or! a4. i.s r-vI: .. i.i..-(4_~ .-/!<.:7i nr- ithn ;a!o4 w hen - i d lee .: J.s col i l owed The FiCFS homo I a. e.n , o.:i-s 'mt t.,"r tie ' w,,: re a : :i. :- t o,,r !e - !.. c-i, ::10.:1 ]. otn ::i -. of t_1 ent t.he f ecorndo l . i1. s i o r "i 4 a -3 ef..ct ....-. i ts- I s , i i ti p ic- i , ,t i -:! r J.s3i. C:.. -i: ori·· _is li: fo cha:i n ar't time 't s tate at ~i 51l time re ti is ' r i ': 'i..r-"s_.. p: :F t,-I T'i-t I I ii .* 4 d-'. so 2 Ivte for- some i t byj desc(ribed i where o'. . -c: ii o n r 'i. r i. = r :, C ' satet ea i r: : 3.L..:Y. to:IL s so iS i ; ie-1i -A' . nn. n,1 rier I,.ti i ios r. is =,; i i" .t n :i.. . ..,.... the e T'he st.te !(t) and Fi i ure 4 4 shows the posQsib1e a-_ `! ... . t:i: '5'-t,'lh r _ i. and by Qf(t) j evea nij, de '. s-.Co i- a - j .3 - s s pe'"r ee 'i 1 eaSchl is tree "IC nor'r e.pondino casee ar:.e 'P i., r r-o r'! f3 r' 'l41t ~4lt- , nr-t;"i i ... i-.f.. ta] : s'u interv ....r' ,-' i] o 4, -i',-.J .'j-.: e- o e'n t:therwi':e !,. o-,. ;_d 3 I t o ,, , 1 e¢:rno 's i o:i. n, r eso n.::i].f.,::or' '. it -. .-.. the '..t r'-.ee.,.= w...._,here .ibu' der t-he~m, period (.:or~nei ote ;-0-f t-Tbeinin, t.i.-) r',end en-'t::. · id e-n.r.-ir := nr-, per hlps 'i-t1 , hn:-t t. .,... 'ni: s!e O i i u . a'-s a i t ? i -'t|.,- _4r o_j. _ - r±l - r (4.6) second seubin'terval. r t , i i I-.. -t Triov -. onf thEli .iran.... n (o i.die I t... _ ,n .i. ino r- can be anal -id 'Ci T_ ,... n firs: i imprtovemen: rl the no t-'-i ---. :=-. l !....s o. the of i .ind move ch a iv r,I,, 'i'I', I: v' ence, li i '-: ion 1 kon!' l i mp l. Eq. inter-viX. on ta iiowed by an e: .... a n'-, .mis by spi.ittring ir Ma kCi .. r .or. i..to:.r i-= !' all1oca.tes nJ. 1anor:_~i '.-ii tt in .=.:~..,. .s.*t':-'1.! ~ r) ._ i te .... a.nd il hisoryt r-els .-t t no coi. ision. alnd c d with per-i tO Ma.ee. .rrs'-·es nnd·~. .t.'i n--5p 1.i ~-0~.r r ; 'e ,^ c- iJi Tion i the ends o! ± coil ii i oa i_Ci L .Fi. -teedbrc the X.U (t - 1 ). .. l_ce.-i: fra.n; s on c onditional 'ho ('4 .- .l1"' -orr'ee~nrd' pericd or r- ,=-- paraa w-ith E:-: the ubint rval 1 tha.t in the. we d:eno't'e such transition of ,ne arme -the pi(t) that eas -4. t15di':imned ' by' 4 4) 4: t. i 3o.ii'?ii l1 eads 1eads -a 1r-r to to complete Ini prbabiiis ,at. t* te, ;eacn j-'-ize h i ai. r Pii sie the chain c- The is ;X..i rconditional on the sum o0f the number a tIi being o _i. isrt ihe ths gC two or nmoreL. number is pait - i of ; j.n with para9. mfi-i r iin packei-s The transitiion >-actiy_one acces u issuin s .nterva the hv e t4 wo o-f ._btinterva iiont'iiains ir-st transmission p r ob a.n C- From to calculate P'oisson random variable,1 tie two sulbinterv;-als i-f the is we e sLbinterval the or iasres i eai. c_ oc: cusl an Fo remains subintier.'ilvas ii i that transitir From SaIwt1hereas - rEds S..i S.-. ...+t* i :i ...- H:i 4,7). .idie t'Si (i pa*ic: ,_e ' Te thien L 1. e e . f'.- 'ini sta.te i.i-C 't':~..(.SrDirtb.ntei"V..;i].! , -sub :. Lt,.1val is tPhe sum be i n(. .:''ic 1.t.- y se fi'd one ex;actly )' i: . i0 son: i ,: tin "is is of sire pc-I meas t- ter so -ra. nsmi tt pacne-s nterval nu mbeer the to S. . o: '.eton . :e i cond i tona. .arae .i: :First. tihat t riP~iji' umber is then achI, bc'z h oni- -oLntaini:nr ot on be i n:oI !oi F.'oi 'sson cD ndil i ional t a transition orne packe't- abou.:t i TheF m,'z-re and the p aC:e sui ID :i. nt'ieiva Th-e probi..bility: of :eachi- two 4e are 1. paci::ets i e ofrf in mol-re the probab:i..i'i.iy tie -4. 1,-- I -e of the probbiii',tyI Fin.aly from S!-:1 to Si t ra nsition dir ect I , is P. + "'he numbaer Te oT:- slots. 'Thhe i1 queue ,:i] '.:i.s:i. on t--T (it) peri i resolution'l 1' c.,eriAOd i ess 'ie·iT(); change i redidue-d b-y Ibut. is the5 ;=~ (.V.) in? =C5remen'. CaIl. i... in ev it l-rI t:'..']r" e.:-h .- ke .ee s now .. o . . ai in tis i - re perifjd oJf thile alqori thm period .:::* ,,4S71, a = .: ihat i fhe dr .ft i i t is la t: h. . periods st'artingr with a queue the o¥f Thi i J. alg 5ror :i. riod P rl ' s 4 ae mum i.s thif is a. b i5 d ef inle colt 1 i.si on : -y and sion reso:luti on queue lengths i.i max i mu.m noaI. ( length t:-T (ti i V be i length it string o f the nex3t coll 'TThe sequence .. A --. i.n 0. ! the aF: mrst M,:-i X( a n d mo e3p-e.:se. i Inr- a consec,:uti vee with t-r(t) r that i-le ah up 'to the beC1ginni i nof resol V in ettinc9 .:o i: iie s"io-s is i.n the que-?ue b .asi p ion vwhich c and o-. SLi fLni r a X for .. in t over a in- "F( t colis .i. a s max-imum i i on'L' ! n remenit inc to the number over peri ord is. -:oore the ir-st r-etu-n co ll.siIibon occurs - .......... " runnlingi a i the cJhane i cal a c: is- hth-ere *od equal ien 'ih ne:-ive resolution a 'i- i-f beO has an n ':_eue i - Si ncE-e * 5es en-er' ngit'h 4 e1 l.'J.ision reso-lu-tio.ion a in I t Isn'ii of uinb 'Lt .impLi i at the -4, beriinning witih of each colli-sion ai-n abs-rbing barrier queue lercgtit and (the pt(] F'.-r r I.ef3ngth a moment-l gcfenerating cier-ayving f:-r period i- i n1. as: an 487 X * in a busy period si s in is is also di.it. st'r -i bUt ioT ft (;p.. ._e 'i.t pier'iaps -th l.h v .' o se. · n " "i. *i' 4 U.:.: p p"'~ er.. r. an-csp i 1i i delav i.Un t.f r The FCFS splittino ot- the possibility and of on iprio i i , di tsir' .. n::rl .f.-3 i bb u- t stable, /2' s:lo,ts an optimal more than sl.t.. f the n umntuber ot y dec:av'i n [here-tore '. v'v. i e - a f unct ion - o7-f t' I1 ' i,t dT .W i. periods-tin in S'i..-'='-"i ; -i..v b ivals l arr nusv -·s! per then o. l -i f thus has ,- it t.hat. the , F.:i variai::ie= r ' iior7n there-Fore pC tne r n6 int i '.r T:ijf' ii somewhat way a-fter collisions. two packets ~::ct - suT s. 1/f..s a ind abiout be improved ai*e oac oo '1n- I:ui l ..r . t. J.i i mu 1. at --- J on a algorithim can in reso !. .t distr-i:i as'iin.'berva1. t..- -. j i as '-i. :is aboiu::ti:- 5 inr'tervals are split inuLmbeir iF' i nal'i. .' 4 ' .3".'; . i :4 - . .:nmcs iJ- '. faa If3 i-y deli v-H ,48'i. v':s 1and].]. ;-then icie,-3. at,:,coi ¢'er~ '" ihm i.s feJe ::edt decayv ing!7 easy tne.T: that r randomf the oz_ iii si,:n a]. so hias an 1- exnn i-, ction initial (4. 10) icollision resolution a't most hNp.). a busy i perio wn.iere the From Wald's identity, e.f' prane.'ia.i.y It Tihe distribution lutnctiion and function, walk are identical ].y and (4,9) and ver a ol] l.ows t-ia th f-he numheir 'fi of bi-us'y thie busy period. _**:,:, i This . means. ithat i ncrement exf''ponernt-iali. v forms a random period increment Observe f-rom i as i queue has anl n at the end of .iortihe first ex'epi?3'Ut is less Lhan : i/ resolutio increfments are indeiendent dis'tribuhted p(t.) 17- a collision: s:l.,:ii if ti-he Beecau'se ei-qual. -4 subi nterva.tS].. 't. T-syba k are v aniid su bi. 'i-r'"'.v'a Recei'zl s V ie-.-denska I37]7'i i ncrea.e- t_ hE in.tere-.a C ons - i.der" E:.'ie t "1-ie mai.-imulm k:fnown iS i to (j.-58'7 Pi.Fpencer' ' au(nd res-ult sho·nwst that if to' i f'l;v ne n-h eg-. iat:i. !bi.-:: ..o *R-. r si: :,'iat-ie- i a':E l o:.ln 1. / si:i)i.p i:' .i' r ·f eedb:as,:'- tiherei ar'sd of 44.4 .itn :i s smai. also of i_ each Tie -thne tightest- bound [F46], since- he interest incre.sed co.1 iisi on. of u ppe r' bounds ievd using particu-_lar is :ew-nterva].--l i any 4S61. o- feedback l it . from the prin ciple acl: 45. b-v tade rm olnfinding cain be t he n,:::i. s' spil l' J. 'i7. i as If a.z co C . i nuer i t to then results paper by Tsybak:ov [,481 li'- anvy the e -_hrogh ihput. of .lgnroithms are +or '. 1 it imes in i lot' . t_ h en .is ] t-hie ii e snar be s v va it '.ter .a :' r ke Iistj, alg ri ir Tihe known [35 4'7, 48]. reviews. manv variations a variety of other th m andsa l .er : 'frs', re1ei a row. on'd rc"-: in al so- ieir th . Si s oni i rand thi- b C pt erneral sti 1 not weel 1 understood., butv is also .grit. f r.s an idle . - l 3ii 'r-vyinn. the entire feedh..-ck: in sc.F ii Ljt t l. ri:_b p . em e ni-/ number lit a.g -iti.r -ns Tree feedbacr-k, number- of partial reso1 ,uion is tr .ns.'in i 'i-ters ln t secon jd su~,i i:: i 4.3. .f11,:-'F:evF r:"J ' c C) w.:i.'] departs ve .aspect of I-Cf- nois;y it ttioh,e htihat been .,-hi(:.e(d" si.. -:ept .y b tfede -aha.s tihis a been f C0 pe i nvol ved in be it a4*nd a ;4z:~7' C due tko Hikhailov' and Tsvbaio::o)v i mproveS.m it ent ,ni Q' o-i s subl::jec:t is the amount cire as seyv ugh(-iput [42,1 - lC, 7 .i. . iaion o'o-ore trving ha s [42] r'!.tff! b(.'r o'F psaac ket.: up coL i i ei hput to A il tihougch s.inc:e e ffort t-hr .u (i i) -asj ors t onie ti Cir 440 chnoosing--the ooti.m inui mum thr.]ua ma..i aid i Humbl.et rMcseiy that E414i aind Fdi ,in:ern ai ay.-'- re-- resol. v.vi!c ti mal. s-.howi er i fpr._'vemert of tiheo.reti:-ica I - o quite Mikhai.1ov a.r ii.j v no i8.. Th onr a r eview re collision s, i1ssumptions.. -4 alno.-:ithms T'-he so3litting transmitters to =:+n.'- interesting to i n1fvesi-: l a e Flajoret i [-49] - have develpe atr--act-i: ve bothr feec-i7ac.: FCF'S spl ittin Very which but j ma:. ni t ai ns t he also .-_": muti. i.i -1 f a...i v ::r or ace- ulI simp li ity can be modified also has t-hi.s modi-fy t: nta;.C:vr ile ofn¥ sysi-te.ms tnu.mber e firi; . tran'sitio"('n tbcti-. s : 5_.ause ,"!'"F1_t3 u.% Lr ,[5 l' of. i-ErA 5R ]. i nos r a nf bete ]i- :.'-4 ileis;ad.'e.. . oe ven i r7 'i-i'-r': in thnttsisc cabp i. of .iliiy users, it is orithims so as to t-r'ansmitt'P+'ers sr) rp.i. JtijT the l las-t come fir'st number [i..5:-I that - 487. resolution 1:2a ln-A1I el ;'1' o fh is a ins sensino t: aprach- -OaC::1h i n as!7. ':.... int 1. 1r 'Tihei ap test 0'san d limited te and and has sfhown intoa a s'plit'rii rl Sp.c:i..ic 'i._?.'d: .'ii(- S. . .i. no thes-e only capabi.lity n [=55] a finite from collision arl'-i.-,; J. r'-ate Inle.es ds i wii.h is r i=. with a ma-.liMrunM stable sensi imb. et samie mai murm thirouo i iacc-ess terest Hui so it Mtathys send to all trimes., ,al and robuistness y, recentl algcorithin serve al*orithm pac:i an aioritrithm td at .whichsen-_in in this limite d for its Ci: ierrors. For has a 4 that o+ Iha feedback ai,: oriithms transmitter after ) discused so far- require the channel required throUoihu,+1 19- to and TDM.i this s .; o f kt-. t!e as are i s teraJ i.so o jn tr -ir e- : ii thm r i .t i i-,A4 Cnill:7'5. l :L e y S i ! antd a; ;C '.- . t.,. ,', , ,SENF:, s. i HC':,,: rinot the other i 7.... a- c .spei! rCrSJ.,,r.Fi... i n .... "-i1'5tic f4, i ir~~ t t. .tr caEn i ,I jn Not · r'. so '.on,it I en d does not the.. i: b... .i- nece:r . t .1 i ? icldn] i....... .: i.mo. vt he n effectivelyan dn.htiiir .. t ~t. , .n ... H.. arc t v i r'aci:iJ 'e-d '..n.u et ........ cu.rrent>. ermino use o -A the infinite:? numriber - of en i -~ b tsense oy., , carr .er- sense_, C a-ir e, but . i. Ci i .I n !. o O±' e send:. -I-ter Carri i c n- t the mi dd' - i i s=o u-r ur-c:s t to.. s "t i.. b.i. .. : dJ: : h ri t-sd ' th . th, n : ... , mo ,:i wen be1t :i. ..e-' ets rre i i -'.ii r t .... J I :i. in ns:cictenat J. s,":._ch ts to sIibe onst. ofc, cot].'J. ].-s.ri r. 'v 't tmri a.- i u ,en-d, at. .i.i .CJ (::i.. tt .aC e · idle t. number Tihe ?i me in a) ndJ. rI [5. ] the.r' t L' i b ,h., r hej,-enstriC.t Cti Tob b a i tm7f ro Ui ir;. _ si n to :end ,-k--.I-t. a i _ transmi'~ tt.r ie_.s -this Ci'the io r-educe KI. c=i n ro ci ti:.t.1?.. t-aGrIt other s, t=o ,, ' t ; ,rs t f mi..Crr.>. h i.::t.,on -d ;,, "nI m:!. a'!-. cot w.......ni . ·..... t.r-rs.mittcr- pr-f j a*i-i- Pesi no.._. We e, t.-erm:i ne 1i.n.t,. et.: thb'i n"d b . .r-.-nr-Y.s tci.. ct i Jsm .. :, it y..- . .. c hi ot,i souirces.. We btet:in first loEtted Al-ha for 'this new situation and1 the.n mod-' I ..... T . .aJ.l r. t h m i . modifyv the 1he ma.i or-dJ. nRr n: Jdu4-t:in ri r ve= r,.. './ a C ;_r;'", ine r.e 1. t is pacl. et wi t-th lot Fiet rth e. hs.].'I-h~ :is that. in ... . , ~: ... ,, i ec t t, d ID 1-1 Ii.. oh w:i. t. i- .n s-e id thef.3 .o of 1 e ci. nt T ti. h- s5ci clt si ,r i p- obb. ib.t tth ult u -sed q:i. . en ab f ore i.a. 'ty :- '. -:. a ex r -ec -.eRd r ji d.i. :. ,-l1- f a , . i vl., ti tt: hh.3e i:t C i js~h tr led. ::r-omi , two 1_ . t rn;. wiB. 'I d simn r h y i. S.I i, c,:.i s.r--.t : be F:rIne a 'i i:srm ; -i. s:i ... .i- .t ni- - ens i.t r'..s tr-n trh :37cirn i. . i tnc st.c t_ i on te t.r:-n;--:-i;. 'F. ,ei ., t new 1:;i -he mode. ittwl-; - i, t. fort EFe:--' 0 hti· ,Cit t 4- i[lie:.. ih : n .:- t '-e i : i.: ,.i. j, '41i..rt i .. i t,,..r1 j:i.: h n ....t 9r-i-citCi *t r 'ncn:i- tte- :i. .li. i cme ;- j ben.7::'L t;i',i. ,f. s'jbseu':enti i.i-td , - :- "p.n ot tne jir was cl r-i- t i" i.i.;, C ...ercj.stent . b..:'. ' 4hi- . Ib :n i a -- i; i.r -r 7 rn:e .. tD·C i f..' e -nt di these i·r.. ;:;;r :- 'i nj et.; :.1 no.re e tIn h -.robi i.*i. t ri . . r o;.arr i ;' p :ot 'ter qi. ID- i i Daciket ? r and bes .. t~chni "l"i-. In one,, et J.. on var. .,ons. .ar:I. i *'d C j.. l r.tr r-,- i...- i r . t att.empt!snc dur.. sc, s: t. ra n'=-m.F i1ion mt :i.die ann inF 'sti -['.:'_l nttn-persi ct.rnit- CSt-ri J..:_ s_.].o;i ae+ter ;ach d'3.n,-i uu. .i a n . oged paci.::et Iroabi ti . tv arriv-:i.n(i. c,acl-:etr slc,:t; ,,h' - hve nrihw lots tr.....mi ~. bac :l as a re.a.rded tranin7 mission :in ference clthtr di 'T' he cs u a sit.-tted Al oh.a C,'.IM'l4 and idle in th t Aii.oih '.tted ,een Tf erence bet '.r di. i o 1 I or,, t._,it nun ei.n n l i n,; .. i.s iiri,, -i - nCc ote .h c oS3 c h - ere f. I iiir- F thieni' . 1.5! h! htlt. rr - . of e ,,,-, r' n ji, the vet~ c In *,,r. h . . j1.....p .,'' s'L .si. :, '.i.,'t. :. 'e .,,.'3 ht.....,, yi F,i5.-. ... nd wit-ith , i o.: L'::';Cd el ends i7eri - b c ;.- 1 'is ,, v:2 t.o t t It a we l. ..uct Lhe correspondina arrivals transi't-lonThe model~~: wias~3-e':l id].;-~ si'~ ............. ~ _ b,:- s-. i mrq -f- p, 4-- fthe 4- ~-o!1owing i4 i t t.his · ~Ie d._ n nL~~mb~.'.::.~O± departures __ .... p,?r 0~~~~~~~~~~~~~~~~~~~~~~ (]i ol r rn..ton ui-I -- "i~mi].-,n, small s l P p~~~~~~~~~~~~~ J., =,:, - k F·~·:0!'" ';"tl........ h d nt a 4 ,7; art' cou·.l~rd be cha-noe u ,_o e'limin!~at e F..o b.- .'..DiI i .i. .)'v i: _ as .l~rene isE '""'"' :_ed i p 01- -.-. ir.: 4~~~~~( 'hhi t~~~~~~~~~~~~~~~~~~~~~~~~~~~ ise mlnim~~iTC_; ...... -- C----fValiSO P it ~ as:- v~'i t,-h dr (1 ~ rLFv sEort t ed1 .t hi 'L e ptima .t..t .,~~~~~~ vat ~ f~ ~p ~ ._then sati..s.es sn con n: .r 2...... I. ..... hv~~~·... rr~~or~~ I n C:; i.I.: he t: Peedbs~~~~c: .: o o~ne :i. Ien m __:_"__.._-'... 1. C)v ' C.:~ssi: eC ,_F m ~ or F' .... s3 sminai].i:rif',/a].'_ '~-',=ifL:~~::; e - ,4 th ba. time a i-ckJ.n a:J, · ;R I.--:C 3I~ a power ~ ~ ~ ~ ~ ~ ~ ~~~~(~. tI~ v ex:pand · in -J. (4. I s e+r:: in I J. e id l-t .- ! vi-F is,/ari ..-.. .;.n.m.,el -.,.h (:-ci~.4nich . -r i.s .... p( +,'')3. ~ c~i.n I p .:ar Di::....;. ,_,:k .....,, r....1.u~p hpeF. t h at t h . b~ rars .......c me' t.c'o :..~ l.=~r ll~~ .t w (.:--,L"=.7 the t:pe t i~s m a ] "~ r. , (:) i' i.i:~ ~re:.-(imt) i.,erq4e r --. (*.-r,:-:, 4-{l~ ...omi cll::-- ec- too I '.". it ~!ueteop.I h o..,.iai a=-. :,·I.u01 ..'; i; L ±oreh p ·. ...i .tI hhi (; : n h, s e ' -i C'T·he d thti-.fo..,r si ' ' ' the i =-. syste :.m n e{:~~~~~~~~~~~~~~.,, vei: a't:.i.j~ c), B\;'· .e>'panrdirng (4 1) - is .-. stabL--e or at. n a po~.~er" series in trfhai- +orP small1 ,:,; the .syst. em i~s . .I h.a.n !.i::..0''~'.:n.I. stablite +ori al! 1 ... ,l'o:. "/he optima! valJ...tm o+~ m then sati.s-,~.1. e t....Awer , A· :les-s p ''-' -{'lind thanl !c:. lit 4 ,. 23 :i. n.1 tJ initerest-i ti.e T the - whertire at.. s ... tir- :d:L -- cci4 _. ps !t..L:thoh..l i. :i.r Ti'·.. . -ti shoIi i..-. ;-e :sp. ]. i -_ h 3 .' .i ] n eR i hDtavde nit Ser an imn .: 3.- -in - he ' f .:, the G -ic03 tl . o a h" ti sbirterval.s. d_.J . p so, n4t- .'.-r-.. rr.; ej.c -fl ivi " .. nii I s orh ..y:_ n......,.., .r.,s. nrte ,:.; 4t-7i tri,,,, h th- Si'ce ,s.i d.: I ti . - n.d..w ! not i"ons 3 i..i1t t: c,t. ',. I I~,fl. O sh~o].d .n b should co 'i h e rrr ;r '-4 ev -2 s a..m od :i..! 1 n' i ,14- s .- t Je _nc ision i.t preasei m:i. .s ;-+:, if ' t'i (4 ; 4 -' t :) mI._.ch s: ... 1h~e -..be h I"jr .t . .c.l... withf ' e qL. .! -.... ...-7-hl .. I.'] . t?,':J e -:L. Sd h= :a.q.:-'!.t:.thuqh prameer t i .. e 7-fiCfl - ti. ide trsLFr= LfioFl, es2 ii c. a :h i.s a -, .spi f oh_ . t t . .n c,_ l.-' s- or it- jFCF ps.-etr... t. r, l inter-v rttl t4 4 ... pJr i. . L . my l ote sI . r-: .deIfi -u r thsn :. t : ri c-. i.. si, ; t:ra.en '.L o.i. equ,:+i itt timert n '..: t h i t ,..:r-s c:- c-:i.n ' i r.. tv c~-_,- frrj.:i.j h ,..uh. incrE- . E l - C, ... 4. ons.i I Pfd or (J'S-.'. je-j... trn-mi . pf_C.i iidc:Le d , t!.fi ,,fl't . t. il . tv .t't - sei.,j. r r.. ._ n _ .m h-.,dE er _ the.4 Ld prohbai:: qI Co 4r-ft-i .x .. .t ar i-. j _.E i- LtJ.l r , i -. th,.n a , pcrai-rrD . p, . .1 * il. te:yi n.Fp3. n ne r'"-.-i . r1 J. 7J. .. i.n..r _4 pr- the hisa;oci d-nt. t .., c .. o]i ' s:i-.e i. h -. 4- w - rJ s- J :'. ,I chn(:lct,:and trc ste .r onAbr be {d.- r . I.'"'.i. i,.- t', ',t.i,' L ta s w hou.. ld e ch s er-i is an a.ys:i.s s:i mpler We first S.LL-:.. 0., i isi i.o, E- ;a rnl1 or i de co in 3 ei,.i - , r co t t' -f lot p I. oc.l i rn. .-r,e j.' re:o. sion i :I. - f ..........here 'letion - . tr 1r . 1' o00 t ti:'i. im(.= ]. 'i..;":. 4;.~ S: al V~ 1j -i . ei-i.!i. s:.,,-4 -= v..s e . fr.. e.o, ..... I ec i: 7 l t.de c '\/,, ..... ].n j4 ' l_ -t cr .,.fiii I ~. i 9m-7.. t:\..e now... We now)i / ic: f. *t , +a_.t. . ..s..'.::. s ....... e't t., ' ... .- (:.....·. . si . hnad the :":ih for ' ro:+ .. . i .... riqht . " +......(.~,2 t d.iia ini o ti + m aximize e ?..d I.~e i. t . ., "[Ond -': ,: i ' -h .aa'n *T'- .1!, m4 ce... o. an e .'. r*. i.- o e.... J ";'~ :i. | I::,:i J. J y:i.ei.;diian j atr~a" o-f t-he.... end -,.. -].,':: r oa b r c ,rb a , . IJ. t. .i, i th iri em pt-y I i t :i. a. idct,'...{,din. f . _j roJ- , 4- h a. num: er . e-.dpected ;- -es aerete iJr t. . .1.t. Lfis the 4-.T 4- e thi iT t..i on ot thus d- t de :i. t' (..i as on' i i ntervl r re }-., iii a nd pen.: 4r 14~r.:..~d peri od Ll, e-, probe:3b:i end c i.' .. .e]C. a. :.... neoJ.:iia3le be coi=.!-i t....on that in tur-n thari before; ,. o.; a~e fmore :Lfc,.ui i re:-... i n This mieans ti.- -Findr =.'S =.i. ll . two n pl::et= t.ha 4W:i.th sm side ) . I + ,o -F ( 4,T1' o v4 1 (, .,*, o.ve :,....y mp ,. p"m orj t 5.s i U a.ymptc,tic * a ter ril 1is.on En i.n terva. cm iut-.es. . si .eme :.=.. ri. cG'..e: u..=.ed > i- .itj i;.. 5 te i .. . .* - n. h mu thr-.hp j t - 3' i. :-1 t- ,n, --i,. .o.i. ;or .f:irst 1. .T I· Iff . m . |n.... same maF'. . t he c n inor C.-;M :: 1"L t'h +-i--': t , .. o t; as '¢ F" I::^:C;;si h .1.loted , e. F- o' : ion n i n terv!i j. wi th v' ,u ; 1-he e co :. !d t.h t ... o ..,h ee bset :,: -. n , t .. e LrLetd e : i- i. sn,,3 et ' s. 4 ' .r " t e OCCUrled. the diL= ,. .,_ ha- in e-san rt1eF-f,' hfsi ...' .J i t onTue. i. ri a> *' 3t 5, "'f ) .. ! 1rt::i... e i FCF-4r:t3 .-- we q 5 e--pre ssiOnr ': :.=:' !roLa ': -- ... '.n the i.imi . - ii .... . -,:_h ] :.*, ... o '-i . * t-i - T.i.C -h t .:R iont . r" in ioth o .:i. t :i.ih7 n m o Aloh-a. ti- i ./. :. j :i.q o r This is t h not- ttit and wih t surpri -in equal: col I isi ons ,f,, at n o. sut.tant:i.al sam5fe van m :.-: 4.. The idle or (. ttroh ,;..:.t e>·ieted e:- time i..: wn:i t l i ci-. n Ft'e ra· a.n o' ..t dtra·t.ion cI, on du..;e to a req. whereas ,: 5 t '. forr iThus a-..: I T. i. nj .^ io. ve in a tranl on i d.e i Li 3 -- orni. an t.hn deIs: . i dle ei.isaeand (a , is ationr, ., is-. the. the same same as as 4t.he the aove above rserv reservation e c- t .m win.-.h:in sutccesses have dra.:tioun and .L c, ;o'-s en. -. :" r _ a sF.ots . wou.ltd have a; orth-inm h t w,._f h success ii r .... res..rvation. i. C.SI! , in .t he re-,,rvat:i on-., c. and d Lura.tri n .nd d l z sl ol'3sion i:i rn a . d have the d urat aredic c, or , ouri caI 1. that Ri te n,, o'm). [ 53] Qeneraiie to .. 'r o. r l' to a - var-ie:/--o.-- otvhater It. 7 xtt t-r th'i-e E:.th.net tThedn systre limited d ue t:o M iasse and S=,. i . ue,:: - os: sti and In : a reservation I9+, =lot. ta r... :i .i. t alq orithm nwif.4ith to el.:to allow arbitrary col at usetd here can be u sed s5 ssems is somew cnl isi nson o-f b vHumbtl eL: f54.] is eerva es't ion p.:acket, Ctsa durat 4 ,i. ion It+o:,,. I' t a; t trar; · n the nLumTer .idea, orio-iin aly sl - ts had dur.tin i.. on t he FCFES' < then we are analysi o .idle.. success, had d.:iuriatic,n coil. : . asum.mtions a) ir.f-'cRr........................... tL . i T'his d e radation type -:turther,- d eveIoped ft{or =12: mlticcess ui ati c on d i ti.ons. ori i. j..:.l ie,, increase tie iu_. : e that sub.iinterval s throu. hp....ptt reser; tirj.;-at voccur-' rarely and .:stem.. c t X5 unita i:..,ih .d :i :; ,: th ,, as in have the throu:utto....!tn psc,tem, a REFERENC ES 1) *Abramsont, t'-.."iThe Alo-hna Syvsttem--'inother Alternative for Computer Cionmmun nications", Falt. 1 Joint Comp....ter Conf+.erenc e 37, 197nnAFIFPS Conf. , Vol. ."2 Aihnlswede? .t· ') Liao.. R., "Mlti-twav Cfmmunicatio n lChannels, Pr oc 2nd , 1971; 'T'sahkadsor, Armenian S.S.R. Int. Symrp . Inform. Th. 1 '97; Sc., pp. 2n-5P; Hunriar ia n Acad H. "H"ACodina Cnm muinicrai al-s:o EE., "-Pful'J Unii v i-heorem fo'r Symp Ln-r ii ns:. lTnt piIe Rc -eass Channelsi i 19 . I-772 H.ai tul.piple Access mrtion Thn A-sil omar, PiF. D. Tinesi s Dept. 1k'97' o.f 4) El Gamal, A. and T. M. Cover-. Multiple User Inrormation 'iTheor'y". Proc. of the IEEE. Vol. 68, pp. i466-.!.z8_3q Diec. 1 98- . r5i va n der il ulen, Info:_nrm.a.tion'T'heori---' ·;Jan, 19'77. C1., " 196i-9/6": E i_--iisz ar, I . and .. K hieorems f r [.:-'iscret e . S:)eci;.l /7 Itssue on '. mmu n i. cl t . o n s 6) F',urslet ic- c k eiReport 91 :3.e12.pn. Ac.::-es s Te:.hi C D. and Cha n ne;L l 1 .) f' M, and ti:Leri::)!,,;Lu 3'.t-he Rec.i 31on of 1::";) ' .. J r'. iagter, i IIRi'n . G1 'Ral f: mm t i. a .t i on 1 u ot+ ii tlt-y I. Ci an nels in . 'o o . IT'- Inform atii o n Theor-:y: essin- Systems, Ci iap. .prrctrum, IEEl Coding :i:: ; Acade Transac'rtionsc; on 2. Sat-ie ii * , in . "F'Frequency-lHop Transm-isi.'i1on fr c k e'. tad i Net o r.w.t ks S i t c i. i n. t d !l Tera r isiei t r a i'a i"--i 44, Coordir.nated Sciencee'L:Laboratory. Univ _ i 11 1) ..- 7C 1rner 1 ey IEEE -Trans. Meorl Spread ay Sr T-2'5, ::: wf. .th l Vtl VJo tr f. Wol.f, "A' Codi n, ei ateod F,':rr't" 1 p'7-0 nformatio 1 a n7''hflr-y tn'. John i Wi e y i 9 68 . . Wa. imeler MuiullC tcces.: Chaln .iEEE nl.e" H-. t p. Theorem for tu It i p1I e S vystem -.e I ources Sept 1......... anrF Reliabie n::, e on the Ca ::) T"lr-ans?,i o'n .a yit i rul 4 34. t2)r i moti os S: n &I A"i0onrithm for C omlpu't ti ng thIe Caparc itity of t'lemoryless Channels": IEEE Trans. on D.iscrete Arbitr-ary ,i Jan 19'72, pp. 14-2(0 .IT-18i' No. Infotrmati:ion 1"ibhery, Vol3. B:lahut:, Distortion Vol. R. .T'-l "Computation orf Channel Capacity Function", IEEE Trans. on Information pp. 46(0-4 C70 No, 4, Ju.y 1972, E., and Rate- T'lheor'y. 14) Massey, J. Feedbacki', ,.5) tArutyunyan, E. A., "Lower Bound for the Error ilultiple-Ai ccesss Channels", Problemy Peridachi Vol. i, ,io. 2. ' 2:p.-.-.6, April 1'75. L. and iP. Mathys, "The Collison Channel IEEE 'Trans. IT, this issue. without of Probability In formatsii, 1-. Peterson, R. L. and I. J . Costello, "Error Probability and Free Distrane Bioutnds for T wo-U-ser -Tree Codes on Mfultiplei Access Channels'", IEEE Trans. IT T, ol. n T-26 6. 5pp 8- 67o Nov. i98. i7) Gaarder, N,. T. and JT.. K. Wolf, "The Capacity Region ra- a lllui-iple Access Discre-;ete Memroryless Channel. Can Increase wiiith FeedbLack"::', IEEE Trans. IT Vol. IT-21., pp. 1i0-i2, 1975. Jan. ,1) . zarow, Li H. Aiccess Ci-haniinel pp. 1i9) "The Capacity of the Whire Gaussian Mutilt ipre! Wi'th Feedba-ck<"? IT Trans. on IT VolIT . -3 flyv 62-`;-u29, 19R4. Cover, T.. , R. J. McEliece. and E. C. Posner, "Asynchronous Muiltiple-A ccess Channel Capacity". IT, Vol. iT-27, pp. 409-41, July 1981.r .ui ..NHi P Pnr &. Humblet, -. "Th e Totarl 1y A -ynchro noucs tMultiple Acrs: "i" , J1 Capac ity Region ot tihe Chann' , I:EEE T rans. i-ss'ue. A. R., ,. A. Heller, and A. J. fecininique fort.A-chron.ous Muii iip, e IEEE iTrans. i"om. Te. fch , Vol. CM0f-1-9, Cohen, 19 7:. i 1this IEEE Trans. Viterb-i. " pp "A NewJ Con.di nQ f'mZ mu-.n-m 7iait' '" 849-855f jCt. n r............. a, a "Hi hi Eff Cu.ifntti m r-i ion 'iTr-ouh c .l a 'r- ain.. C-oimmn "Ferha y Vfl M1.L tiple Aics-cr nal i F rocessi n Repe-ater" Apri COi-l-6, pp 222-2:2.2 iciency iEEE L 1 96. o-Ur Rate_ v , and .. L. Shy'dr, T'ne h-or T..t'..f a..an an As-ou n hroand a Synchron ous ML. U ltiC r hannei. r'e T-' he S'a t", 1 EiE ' r -a s. 1 T ? V i ITi -"27 ppi 414- 4 L 9 Ju 1 v Na a-n, 21 A ri ..25j) MasseyF , Detect:i. pp. 26) .. : Priva P te Communication. . "Capaci. Cuttof Ra te, and Codin n Opt.ical Cihannel", IEEE 'Trans. Comm.. 16i 5-1622, LNov. , fmr a i:i-rect Vol COri'-29!: 1983I. Humblet, P., "Error Ex-tponents -for a Direct Detection Optical Channel", Report LIDS-P-1: 57, Laboratory for Information and Decision Systems, M. I.. T. Oct. 198,.; ,27 . t:'pl, 28i ..- ChiaiW .forCoflnd1ition "A S uFi'rient IEEE Trans. iTT. kV . ani 2_ontrol ;:-Ji:'j Vi Fer-gC1uson? SantL D e Jl. J. ': l .fi.'... n*_lJ_; "-' Septn F'releplnoe Hiar a chi ' ; as eiei 1i 8t,"-: ~ '-4- Aug. i :. L -~-::a;,' l-:'..:.t::et .. ; 4. . . '3e ("',* t . 'Lio 37_ FiAi) T 1,1i ,1 o:l 7 i.c 1t: O;- U1-c 1-:: ____ ........................ P 1lf'-F i-.i-c..r.v.a.... i . i._~ 1. f Thfl:t-t! , i i CFi, Oi~lrzi~n : c L or'!n Vol -2. , I 4 1111 ; j ,, B. at Vii I 1,i.44 ardn -App iJ. so Acc s s n !ri1;:ii,, in CRandom .ze 1 (".'L ip cei'- -se I E-F31RWor kihciF RI 1i .E. _ ,ir _i i n L _--/_w_ i P'Iar S. -. N 17 a i , .i on .i: d Raindomf i:. ; on H .3-nCr-h'-n',rE :. F:-:etedbarm w:ith n ci i k - .irI n nV. '- f..i p . Z tl.1S4o7 FaiCk:et Delay in aM i:'eri odachi Inriormatsii ~~~~~~~~~~~.___..~~----~- nel-. Dr''i'- il l i- "Free :c. iRV. ,. A.i l S i:i :o\ "Tyvbakov::v . Vo v` CFhannrel n &ii ql r i - i i. pofir. l) . , , cor : Pac kets n-fi 'r. E f S :-;8 a iti tir..ti"ilr!: ,, Lo.. iemcora-f.-.ndum I'i-'76-:s- i 1i-1 -r t V... H V , .. r 1 -3V~LtC C vryb B,.. ca f- i- e-i', 'i IT' rans, . I L.(D .... Broadcr-.t Fjjir' f 1"n n.et 5 ' ,(D 4rI7j i r i :) a .--. i3ai-'i s i g 4 ,I--; -*nj'-.-. o Comin iEL 'Ei'2Tr-' aEnE Sl 1ic nI L~roadcaE rrl bs':-,s E.E i''I *- iadria,t F'h. . , i.er-1hni1qe r-v ieohinin- 1l Lnab- r , ' in . c' l iie-,n . J A t Q_i i 97J ..nd .- f4 7I7l c i n .i. i f-:)C i-i-: "ri: 1 1 4 2- fi F2 PL.tr t 978. Ac-Icess i t'i,ol T 8 - i-':i ........ :~ i e-'Atcess als 3Y.mp nd Univ. Ls.ai C an-i.i'ne'i I st r i 1422- Muli e I -i':rt rd,ap'ive , 2;.':.....q.. pp. wit Svmofsi. U3onir.dJerar:Lon-i" t-y i............. r'_-. ri.., 1 977. Cc rnel'T IIct. Timeij. o h- A rbi tr r tfi'lQ .Il1eL-- Aioha i Aloha ih ni-tted - C ntroi - nanmc l982. 4th Data CoffmmiE i 975. 1. :t 1"i:-Te iF. Li , Caa ind ,::i :i; G aetana*_ F'-ir-fotoc l a E i Ci; ;D June ' -.t _.lyI_, "t. Urit ln lotte i . 't- frc, 5 2 ' j: tr Inter- ar riv A rb i t r a PErA creF i r ans . Camm, n Voir iii-2 CT 8 ' ' =:i -5 S'i- i "T-iin r-iLoughpuf t q G po1 ACo-2'- rodici r' ov il Non-er IDecentrallied van Loon , .Idt" Ijies' a-Qe Lrn rith _-i a pp Ci a 1 ) r Robert-B. L. Gn. 1 oha rac.ket S*ystem Witn and Wi thou.t Si1 an2d f.-ct rf- H-ob-c Note 8, ARFPA Network in-fo. Ctr , FR i Stanord. Cal i I June 1972. Try N. B. ul tiple , 1hm" P Aorit. pp. 65-79i "Airi P'eridach:i. 198Fi. Upper BFound Frobtlemy Channel 3c t-. N4, pp. '76-84, ACCs.-ss 1i3 Vo1li. F Fren roblemv Oct. Liikhanov, .. t ov- __- - on 1 9803, .:-' *3 Huain i a J -CiiT. and rt e er, 'Di e - Ai - visi s ofi the Modifi iJ d 0in2 487 Contention Riesoi.r ution Alzzorithm". Cornell Univ. iReport, submitted to IEEE Tr ans. on Co.mm. Moseiy: *.J. and P. H-lumbiet, "A Class of Ef ici ent C ont ntion R-esi ..iti "L F,-i Alg: or MfGultiple Accerss ChraTnels i' t appe ar in iE EE rI'rta-., ,in Coimin. also Repo 1r-4 lIS-R--3194. Labo iator ir- iln' r trmai.L n and Dei1 on stI e s i, T. r - March LiDS-TH-T ].982--Reporrt 5 LR I-o-ai Informatir" aarv:or 41. V'edena, s k N. D. anD M. . F'insi-::er '- on - o oi ioT:in. L t of at-and-'T r v Llgrthm:'? it n nt r atiai 'i-Dr .Ti shoi on ilCo'...ti ,inar C ultiuser Co fC 1mm1 u n ic'i t.1n A bstr-ac ts r !Papers? Sf chn , pP 141-14'F, l1983 42. Pippengerr N "'Bontis ' n on the Pertformance of F'rrtircol iult-i plIe Acci -ss Bradc-ast Channel", iEEE Trans TT- . -2'7, MIlarfci ir' 8 tiie -for a Junte 19i8, 4.4) -. L, "On the Capacity i of In-fini e Pr pulat o tcce'r roto~ol:rs' IEEE Trans. VITa, o I'T-25., :,-:--4()'1 ',M; vay l"?F2 (s'...bjmi tted l.ay 'lib8) i-'-,._ lln'' i a N r:pr 7 c:.nd t r e PC LAI. a _4 I ri 'I R ano"4AB., ti Haj.i_ M.].e? IHu. ltipe A ,i'_'S i ,-.:.f -i .q t:f.. 1 i' ii ti -A cc.-ss IT '- "_-pp 4 _2- 4 C:apao rii i _.- SidA 1 47 4 ff .. r' of ft u. i 1.: Lt 1 '. . 'f rr-ma . - ir oadzast Channel" fl,a,May:9o o Rar'idn -p iKT * A n . ':-: at..i E;- -:y y a' l. :i. riIi l"~:·i :"S !hi.y I , 74.) a_ !3 w a '" , *i _. Hu].tipi |.. rii on:: a n . iin Sdf rec:: i. f. ( S¢ -? IL;ti,veiT.- lci r;' ' ar I E Sy~.tj-em", b'ur':.r' T ns 1 n I ' e a db c:, ,. Ua br i ,, I:_' .....- IEEE IVY Tra i-, Ac,-ess Sys.'..e m s :9inSSR ,,, .- =... 3' i, ._epr-_t r I''Z f i fi tiii-T S i_'iif. i I'T, .i. f~ . *...... -f H i- aLth y.R. F', aand P F']a i-et, ". !-ar-y t C -iij ii Ri-. A l c r- it Lhms i- r Randim-A r'-ess Syctems wit h Free r Ch.ianr'lel cc:ss"lsvpSy i:-reseni-ed at IEEE Int p on SJovi'el~e Quebec,: Cana~da 9 Sept.1' i7:]3; Inst of. 'Tfel recrmmuni catins. E TH Zurich. pp. .i _? ol. l. i L.2:i3: ia;/ i:I ':,r; n iuti on B lo,:::ed IT., St. "5:) Hiuc"i ti Chan.el ib i y Con -f -erenc- e, G. Ga1ii.ager , and R RE Many i'iew OCrleans, i and DecFsin tinormation .ti-cc''s of, a Sli. te1 Uyi=_ers" ,iNa tionral. Teiecfmni.unications 198i: ai1ro Laboratorvy -or Der. Sy-tems R eort LIDS-F-i1 .'31 u i:tely iin i'l,ultipte AccessdCo..-. mmuni c an r.r"._ r~f- Coremm VOi7 C.iii-'4-2 , i G roup Testing , 76-'77 '!:. , Ju'ly ... 1 i. - ,_+ .) I'i<einirio k L4 andA iTobaii "Fa,::et Swit'chinr ii-' Radio Car,e 1 f art 1: CISMA Modes a niD l i he i r T hrouchpm:-.. t' -- 'I 1 v COM-23, pp. V5ol on st Ciimm. , IEEE TraIns 1i,'s i-i ar ac'ri :i4i':i_:-1 6 Ij e . 75 _ .55; Aasey t, i L. A1 i <:.o,") 5~4) .i Sec' lWCA . Mmorseiyv. A. and Ci.an'e:' eil": ,.=-~isi,.~r ~ and Cr',_, also in LI140' DS-F'-i , i'. :=,__ Ha i F'rE-oedinis bDe. _ i'i, IT "Codi no : SJ~i. 3. r o Pi -:LC._ i-z .-a - Repor:t ia d; i3 pr i. I i .. Pei rj ji e L'i. : .. o-'f i9'i:h , C,-r2, 19'80 Car f t ir, in an H.'nc-ri" , .i i c _ dSelectd A'.ea~ .. r", i na, ts Tf _ Sjpecti-rutm ina C1.'(hlfrm i93'84,-·u~ic'Lnr,: --------- --· ~fy icien:- A rc sing1 n "E.- iAc,:_e"sir'!Cj (2'.¢ tniie Spr-ead i _U ,; _ .3 Humbl'.et.. F.4 i;ui ti iaccess i'l · ='55)P=' o tyrev,./ si_ - an d .....Eii6...o.i6 . Fn in Mhi o"CiR eol r't t riinmssUC nLcs. ia<1' -- AI U 1 C - '(:L Qf'!, h '... rt.e ir e , Aibuquer-que. : an Nii:1 MA. ,.'ous i. i C ___ d...L A Ch·rannei i July ul-l % a C rn] _ipe-fices-a, i _ i: _ EEE pp. i t |- I '.... p_ it i i e_ Jour"nai onn 4 8J2-4Sq. List of Figure Captions Figure 1.1: Multiaccess Channel Figure 2.2: Multiaccess Channel With Two Transmitters Figure 2.2: Rate Region of Eqs. (2.3)-(2.5) Figure 2.3: Output y = (y ,y ) as a Function of x, w for Collision Channel with K = 2 Figure 2.4: Achievable Rate Region for Collision Channel Figure 2.5: Achievable Rate Region As Function of Signal/Noise A Figure 2.6: Comparison of TDM (as modelled) and FDM Figure 2.7: Regions Ri Where Er i Figure 3.1: Timesharing Construction to Obtain Any Achievable Rate Pair Without Joint Decoding Figure 4.1: Dk as a Function of X,p,k Figure 4.2: Tree Algorithm for Collision Resolution Figure 4.3: Record of a Collision Resolution Period for FCFS Splitting Algorithm Figure 4.4: Markov Chain for Collision Resolution Period Dominates Error Bound; for A=1 T4 -(T3 rn N SOU1RCE rEN E SOURCE CHANNEL E 2 ENCODER p(y/xw) N (Y 47 *ytq DECODER I(W;YIX) I(W;YIX) I(W;Y) RR R R2 I(W;Y) - I(X;Y) I(W;Y) I(X;YIW) I(X;Y) R. R1 NORMAL CASE Rl SPECIAL CASE I(X;Y) =I(X;YIW) 0 0 (0,) 1 (1,0) 2 1 I(X;YiW) 2 (1,0) (2,0) (c,c) (c,c) (2,0) (c,c) (c,c) SPECIAL CASE I(X;Y) = 0 R \ UNION REGION I II- I R1 \ R2 =10 (BITS) (1.29, .44) (.5,.29) A=.4 (.3,. 1)R A=2 1(BITS) R 1 (BITS) ACHIEVABLE BOUNDARY R2 REGION FOR FDM BOUNDARY FOR TDM R1 _0_.51 0.5 -~ (.3,. 5) (R2 (BITS) T) (.5,.3) (.238,.238) E=O 0.238) R1 (BITS) 0.5 I(W,Y/ 'X)|I ®'.% I(W:Y) - I(X,Y) QI I ] I R I(X, Y/W) REGION 1: DECODE X,THEN W GIVEN X TIMESHARING OF POINT R IN CONVEX REGION 2: DECODE X,W HULL BETWEEN R' INDEPENDENTLY REGION 3: DECODE W THEN X GIVEN W REGION 4: USE TIMESHARING BETWEEN 1,3 AND R'" ~ . - x =X+pk SUCCESS SUCCESS NSUBSETZ 121-/SUBSET SUCCESS \/ COLLISION COLLISION SUBSET 1 /SUBSET IDLE \ ~UBSE/UBSET SUBT/ S2 COLLISION COLLISION ORDER OF TRANSMISSION AFTER INITIAL COLLISION 1) SUBSET I 2) SUBSET II (SUBSET OF SUBSET 1) 3) SUBSET 12 (OTHER SUBSET OF SUBSET 1) 4) 5) SUBSET 121 SUBSET 122 6) SUBSET 2 FRONT OF TRANSMISSION TRANSMISSION QUEUE BY TIME t ARRIVAL TIMES OF WAITING PACKETS t ILOTt \ t[t INTERVAL SLOT It,t+1) (tl)_ "X X' SLOT[t t+ TIME ALLOCATION T(t) ) time - ,SECOND, t+, t+2 - XX' L r;SUBSETI CURRENT T(t+l) TIME SLOT [t+l,t+2) SUCCESSFUL BY TIME t+2 _ _ 4(t+2) tt+2 CURRENT T(t+2) TIME SLOT[t+2, t+3) I P2, P2,2 Pi, 0 P1,2 P2,3