‘The Picturephone® System: Digital Encoding of the Video Signal By J.B. MILLARD and Il. 1. MAUNSEIT. (Ofacuaioteecve Bepoer & 170) We dacrite in this paper the thry nd srplemantation of @ dif ‘ntal pulse sade modulation eto ha fe omployd in the Picturophone ‘network, Belblshed and original ending locnuze have been combine to allow tranemitsion of x coed managers analog simal ever a 0-312- ‘megabit-persecnd digital channel Pulse code modulation (PCM) ix used for processing of snag wi le for digital truntnisvion, Quanlizing noe and increased ban trith are the ponlice Zor ibe noise immunity inhorent in digital ‘Difeendal ayateme? such us delta solution sad the mere gen teal digerentia, palo code modalution (DPC3E have been invect- felted because of the ablity to sope the spel wn af the noise ad ta lake avenlage of the aample-te-sraple earlation in. Ue niga. Tn ‘cries of papers ia 1952, BAL. Oliver? EK. Kretrmort end C2 W. [arison® eoneidored the use of Hear precieton fr "decorelnion” of video signs. Ts 1986 W. #. Sehteibert performed entropy ensure Inenle on television sigele to determine the prfominuntrehindancien Present. Tn 1968 RE, Grabim® demons: by computer simulation the easily of using -bit TPC fore Ielewison teanemiesion of will black and mite pietarn Mor rently J. BL O'Neal, 31 ‘us analysed deka modulation aed DPCM for the wansmisson of vileo signals. Te noite structure of DECM systems in slope over= loud las been investigated by K. N. Protonotarios"® Subjective evalus tions Have hoa reaorded by B.C. Briard?" Thee sti have le the way for the application of DPCM to Picurephone transision, 23 Pulue Cate Modulation The conventional mated of digital procesing of anclog iona’s ie PCM. The signal, e((), i soiled, a(aP', quantized to dicerece level, 800 rm. GreMMM TreMTW.AAL. JOURKA, SEARMEARE 6 fn), and ener, 2607/0), in Wnary for fr Unwcsion et the digital Tine rae Cae Fi 1) ‘The main source of impairment in the dovoded und Siteredwignal 1s noe introduced ak the qantiner aT) = a(0) — no wo "This quantixing noise kes Ue fort of Twekeroun’d nave of & gene craly nandora nature. In addition, the auaber of bite por sstple ie ested, Whe aekrosnd raise Towel vse and wonton paler spent fn the video ict. 22 Preise Feedba’ Systeme Independent processing of euch samplo is asewzed for PCM ystems For sgual sourees having statiteslprepercis lke thereof white noi, ero atamle-to-saraple eorelation, sie in appropriate. Homers, met ideo signals are highly coreleted aot ex's from eample-to-sample ‘at foun Tine-to-ne and from famnoto-famne a wel Th preiosive quanlisng syste Ue value 0° Une signal ix preted (predator P) at eaoh auzpling instant from ita recorded pate history (Store) and only the diferenee, ofa}, bacrcen tha signal sample and ite prodited vale, ap(, in auanaed, aCe, and lesnerited (ee Fig. 2). ‘The Pictrophons signal han @ 1-MHx bendidth with an ska Tine rie, There ate 271 Tine per fle with a 2:1 interlace, Table T howe te minimum emount of storage tequired (tensmitter and re- csiver) don various feedback aye wsutzing tht Saormation for use in the pradieeor, Pie stored with aa aeeuaey of cht bite por somple ‘The coete and’ complestics of sysiors containing Jorge storce ‘nsighed hesvilyagainet the we of frameio-frameeorelation®+ at the time the devizn of a eadee for the Pictarhono neimarle woe begun "The work of O'Neal on tolevicon sigatls, whoee line-to-line eta- ‘istics ere much tho sume ae thoee of Pitirephone sigaals, demon- ste), a0) for Sees Leto | Lae La Tig 1 Puke toda madoledon (POO apt urea mein 401 worn) a alte ator 1 i i i 1 i i 4 i Lae mle} pay Fi, Posi untae en, strated that only bout 9 2d vin iv sgualdosnise ratio S/N} ould be achieved weing linear pzedition if Hevion‘al and versical oroltions were soprusinaely ancl. Somat es could be tchinved with a 2:1 interhce fora! The les eer was cherefore feontod om enunplo-to-ecm's desursT li procure. Wor sarp- nase ploy eysteng che stare takes the orm of sc owuighved capped. delay Tine (7 be the samp ing interval) with the vreehte, ay to he dcterine! hy lu aise of Ue signal (ao ig. 3." 260) = ¥alesin = OM tae = 9TH a whet the a, f= 3,3, +++ Letty the oquations By = Rig Re bo OP @ sad Bs the correlation funtion, Ry = Bislin — THs — ATH c) Mourorements have shown that, for samples sien on the eae Ling, lelevision signale have an sutgcomelation funstien which is very erly Lapinevsas ii cay saul eerie! i ator i i, Rikcush pin sie oa Se vere wil exc pe af he ete Reptle IG a ea of te wine bw Sng ‘Tanun 1 Fuaue Brongn Reqoamp rox ‘Vanrore Pasaicrivt Semis Pome Crating Police 1084000 bie iil Cclaae Prtieae eee Rie} emp (mal bh 6 {Wor thie type of souree, the optimar eolation to minnie the mean ouare tor between input end ontput (orediesed signs) i one how All the a, = 0 except a. "Thix meare tha only the previous sample is elective in decrselating the signi, The peeetve feedbacs coder ‘hen takes the form shown in Hig. 4. A mesure ofthe validity of thu {scomptions and the efectivencs: of the decorrelation proc is tho ‘Nout of correlation temeved from the vio sigaal and Ube amount remaining in Ue error signal. The adjacent esmple earelation eo- ficient aK lo oa Be Ptalotoause sete gating malas ae «td dy rare. sens 46 sy pty, en en wl ak pry suena =) 2 OT ° ‘as ect from spss oud tobe 08 forthe it ne “Rom” need te ton 2. Te act spe Sarl to Bless — 21) Bien) o fos the wane picture iv 0.407 9 exielnla rom Inbormtary dala for ‘ue three bit DPOB ceder using the quantizer of Fig. 5. This shows ‘that considerable redundancy Le Uoen recoved Iu the cna but hae nome sttoplosta-eemple earrclation genio 2 Quantoing Neiwe ‘Tw deionstate that the noige introduced ia preditive fede syslems can be determined by anelysing the ener st the quantizer ike, courder the one-cap syst of Fig. 4. The quantizing vor QU is given by QF = en) — sia ° whee 400) = vin) — Hoh, o 6G) = ye) — en 10) wot) = Mo = 112). ay Substituting equstions (0), (10) and (11) into (8) gvee QF = ain) — Hw + YT a) whic isthe aytema err in transmitting the signal inthe eheones of ‘ny chennel errae. ‘The syetem nike enn therefore be determined 464 tun nous averme troxsiou. Jonna, PERRUARY HL Aircaty from the cbaraoesstic ofthe ero elnal and a knowledge of the quantizer. The manner in waieh the quantsing soise manifests ‘Held ae piane impaaunts will he discussed in later sections 24 Difirential Pulte Code Mnadetion CConmler Ue single tap predictive feetbuek wie shown inHg, 4 ‘The feedback igual, 2a ie evn by i) 16) = Eat ~ ‘haloes fvioe <1 intraduce into the delay section thea equation (03) is modifiel to ire ton) = Zonta — an as Ps hislony ie weighted giving greater importance vo the mere recont samples, In practioe, the feoubnok notvrk in the Piturephone ele is replaced ty an analog network which consila of an intogratar, F swith lnk fe Fig 8 Fe ~ a, a ana ig, |irelit DPCM wane bene, orn mxcooise $85 stom es, oft ine} a7] ig, ODPOM opto with Sota inter ‘where the jt lenustion (14)) snd w [equation (15)] ave related in the folloing man pep (al) any ‘A eystem cnganined in this fashion ie DPCM system A cites ‘ential system very measly tranemia the differences beten successive Samples zether thar the samples Uheuselves, ‘Tho reeiver aceumulatee ‘he reosied difenenee ond recontuets an npproximntion le the ‘orignal waveform, ‘The preecoe of a quantines provides the potential {or an seeumlition of eystomarie qusntising errr. The feedback loop around the quantizer prevents :hie weeur ulation "The soedback network used in tie Pisteephons coer Ias a 2we time cowstent, Wor abject eee the time constane ras made 96 luge as possible without anduls raugnfying the efele of transient errors on tho viewod picture, This is dveused farther in Section 28 128 Subjective Considerations ond Nowe Shavbwy In DPCM, the subjective effec of = video ahverver can be taken into neeonnt and the spesirum of the nase shiped aaronlingy. Ax Grahann pointe out? the “concep of pregictive quentning at 2 tly peteeplua coding teehniqua" is ona “in which the prmery objet is ot necessarily aczarate prediction wl} ot wl:noe ll othe time, which Isquenty ie impossible, be: only acoorete pension ix thateeigne] regions where che obwerver pereaplion high.” T i known that the pe is more tolerant of ros onaind sl blagk-whiteintrfecastasn in {at regione of tho pienire. Interfaces are charsaeviaed by lage valuet ‘of the dierene signal and st regen hy smal ves Tei therefore possible to use eompsnding on the diferenee signe to loeate the noise in rogion of the siuure where chat note i Lent abjcetionsble Companding * accorslaked through te use of a nonlinear qm tising ehneeeleviatc provicing a fine strutare and whieh S/N for Tow level illus signals (nt reviors} end anarar abrsue with ite lower 8/10 Zar sarge eflerenor nigra Ulan), The ronnear 498 oy nuts aewrme CUBICAL JOURNAL, PEERUAIY RL chactoterinie une the Pictarophono eodee is chown in Fx, 5, The Coie of thin parseular ecb of slap slaee Wl bo discussed io a ater rection THotnase of the eubjective myeete of rating picture quality, B/N is nol a rafllene mesure of aye performance. However, itis ue fo compare the theoretical S/N performances of PUM wil DPCM. [Breed on sone simplifying aawizplions" O'Neal! has chow that for ‘he optics one tap predictive ter the 8/1 improvement for DPCM bovr PUI ie elated to the eample-to-sample corelation of the eign, [Ry(T}, ia the following manner /t0 IMPROVEMENT ~ 10. [, by ap y= Rin, 0 For vidoo signals wish only moderato amount of etal this eels in an jmpeovernent of spprosimstely 12 dB which i aquirsent to bro ‘ite nf quantising information per sazple 128 Signal Impairments DPCM systems are subjes: to a unique sat of wlisetive impaire rnvnta, Atzong thew nee overload aise, ga ke, eumfuuting and tiie haynes, all resuling from the qauntining prone (see Section 22) bat diferen: In appearance to th observer, Tntike PCM syrtane whieh exe amplivade ‘iled, DPCM ayetoms uvesioad on slope, If the eample-to-emle d:Terense is geester then the langeet quae step, then the syste is salto e in slope over owl. A rystem exhibiting ale overload cfestvely reduces the hori= rental olution of the sene being procemed, in Iw detail noplone of the pistore, the noise power ix dolermined by te rather ne quialining etructare designed for small difference. {This ir feared te ranlar ooiee region, Figure T sows w wyplea ‘en ert a vinden ed ze a rn Tn the Pituorpheme oodea the small stop is 15 «mY wna the ep 225 nV ona lla eo egal of 2480 (Pi 9). Tho Toulution errr for any beghtues chenge is pls or mings one-hulf of ‘aml atep, The slope handling eapablty of the coded i 50 m¥ per ws. Te sumations on mich Wie Wey i sd ae that) the dton of se ial a be saat Yeh Lape nce acon Ho = ne ping he ain he pl a Sascha ° oma, socom 407 te Fx. 1—ilatied vido wavtfon and DPOB ream shoving an eveond Figure 8 ahowe 4 typionl pitore ("Rosalie") uneaded, (sce Fig, fe), ind exhibiting slope overloud uitorthrse bit encocing (cae Hz. 8b) ‘This ig mest notte ne sinenrmg on Uns cap ring ‘whitetorblak interface) 1 the amit steps are too large the «trove corrslacion between the ‘aventizing noise ad th signe) eat Tea’ ir viii pattern Ue a sbjeerivaly objectionabie. Ta at regis uf the pore (aera aap Tecsampl liferenee), the DPCM feedback lcep eriliaces using alter nately positive ard negative sail step. The integrator lake break nap thie petvem aad if the leek snwe lark i to sally Teale con touring ean occur. This i ustrsted for oro different vale of video level in Pig. 0. Tho dashed component of the sigpal iste ak eonions [Even inthe abeone of leak, contouring patterns can owur in slong rian of the video sgoal A‘slope whienaighly exceeds (fale hort ct} the ability of w particu step sie to follow i wil rquiee pore ‘orzetion, through the uso of the next lager eonaller} sl. T this ‘waveform ie repented through several Hes (esrong vertical eorzela- tion in the picture) and if the difeenoes bacwwum the kdjncent steps in quertion ate oo lage, a Vile eantone pastern wll rele, Figure 410 demonstrates this flo Figure 11 is 2 photograph of an ccalaenpe teoes showing Tene comeusing patra, A nell-designod thoee bit paiva dees nat produce hill eortours, ‘bw piewre reelter from changing the sonst from 27 ys to 3 ws ‘The zequiremete for micimising the peeing inmairmente ste a conic, given a fixed number of etepa In the ouantier, Grenulas noise require eperty fine spcntsing susie. Slope overlond and 408 twm mma vere Te SOUMNAL, PEOREARY @ o Big Su) al Méturptone ima she Puta of it PCa Reephone nt asa Es zy ener Noisonta] resolution requirements spesity a large outer stop Contouring and, a2 wil be tui slow, edge Busynass, can both be race! through the uee of « uniform spacing of the steps. Quantiscr Ales la therefore a mnticr of Gnding Une proper compromise rong the subietive effec of the vovinue impairment, ft Bymcbvnssaton and Lidge Boone ‘Since cho system wi sesiqnel tn eozeentate the naiee in the high troquensinearrespending to she ileface or edges in ae Petar, the resulting degredation botomes « Tiling Totes. Queatising noise a: Slightly vorrerteal eager likr= the fora o: "staizstops." ‘Tie is due to tho diserete nate ofthe warping process fia going from feurs- fovirame in sti, plete Pree ate changes in the responce of he