throbber
aJ
`
`,-I N0. 3.wuIctt Iwt
`
`-
`
`.55: JOURNAL or‘aouo-s'rA'r: ctnottrts. VOL 26. No. 3. MARCH 1991
`
`L11‘
`
`Ill-li.T..S.}tn El-Icltodc,
`: 3.5:. d-pen in «tee.
`and the II‘. deuee In
`INS {mm the Untveu
`TllI_NO1.lIB|']|IId.L
`= Wotltint II the Uni-
`tl levels of-terpoo:ihl+
`”flbI'I1'I.fl|'lEmf
`nt"J lo the Lnbcmon
`UlIl'|I¢‘lIiI'r III Titan.
`1 temtett and ma.
`
`-
`
`A’Chip Set for Lossless Image Compression
`Irnran Ali Sltalt. Olu Axtwunti-Assam. and tam Johnson
`
`3il t
`E.53
`II-51$!
`olnpilltflllulptllllllluulll-lIIIol'
`7.tMpttatIm[d.ThtnptrtteutnnuttettueItr--tteullInl|I-
`rllIfl.thndlInI‘IIetIIlhIIIuttHbeblp|.|.|dIItll.luuue
`put-IttfiloIwu&%.
`
`'
`
`l. IN'|'RODUc‘l'l0t-t
`the
`S APPLICATIONS of Gill!!!
`ilflllfl Dfllllifetate.
`ooetormfinandnutmlnlmthetmapedhube
`a:moInoomtnotIptultlem.LnuleniIon.pcompruslnnted't-
`niquu pmvldo I penenl-pupou-. nalutinn fun of mangan-
`rnhte. The oompmnd Image is stored cu tnnentltted In
`typicallylmtltunlulltltesputeotlestthsnhelfthetimeol
`the cuisine]. Louhu oootpteaion can be Icliieved by I
`tnntfotm ptneea hallowed by entropy ending. We hm
`dniznnd mo application-spI:¢i.fi¢ itttcgntlad clmtit: (HE'S)
`to perform the .5‘ tnttslotttt [1]. I21: I hierattitlcal transfor-
`tnntion.
`and I dun Itotrtpreuot/dttcamprenor
`for
`!.:n:pol—Zlv etttruw audio: [51 [6]. The chips my be used
`indetaontlently or togethat for itattne compression. One of
`the applications of such I chip Ill ll in the medical recs
`environt-nettt [7]. The l.l.|Dl'llIlllII and chip ucltilecluroi will
`be delctibed. Sumo oi‘ the important perfionmttec parame-
`ter: will Ibo b-I tlllcttued.
`
`II. S-Ttthttsrotttt Pttocusoa
`1'lt¢p|int[pnlob,locll\tdod'lhntnnsfotmpmonsIl|ln
`uchieveupecttnlditttllnttlonolnsivensourcetipnnltttclt
`lllll molt ol its etttttu it concentrated In I subset of the
`eoeflleionu in the ttlntlottn dotmin. If the signal touroe
`ettltt'ht'u it high dance of cot-teiatiott between suooenlve
`sunples. then it follows that lot I given trtnsflorut the tignal
`enemy il cottoenmlod in I ttlnuw Ipectnl none. In pne-
`tlee. the lnttttomtn norntllly applied to image eodl.n.| Ind:
`off pnckitti ldecorrelntionl cllficlet-tey with itnplententatlon
`complexity. The 3 lrulfiolm il mttp-ul.|Iion.ttl.ly Illnple. Ild
`Illwidea I reasonable Induction olteduudancv In tho intttu.
`The tlecorteletlon lndtteetl by the S trutnlat-tn was themati-
`cllly analyzed by Lu: [I]. In [3]. thin um experimentally done
`by computing the math-ottlet euttopy
`N
`5‘ ' :.P'r1°l:t(.Pt)
`
`hlnnultxipl mollvnd Aunn 11'. 19%: rflhd November l2. 1990.
`I. A. Shall I'll wen FIIHII Llbolaloriu. North Arturitau Htilb
`Corporation. Btindlfl Mltut. NT ID610. HI ll now llllll Phlllpc lu-
`snteh Ltlzot-Itotiee. Sill M Elodltaven. The Netlterihntls.
`O.
`.I\H\VllflIlr.I\llllIl and I. Johtuon In with Philip! Llbotuoriet.
`North At-nerltan Philip: Cotpontlon. fltlareiifl Manor. NY IDSIO.
`IEEE 142: Ntttubnt HNIMJ.
`
`the trutttorm In I set of 15
`ttetme end that sppl;-tat;
`computed mttosnph (Cltl imtpee N calm to the numbet
`of tttgittzattmt limit. and p, the tztub-bllltr of occurrence It
`Vat observed that the uterus ellllflll‘ 0‘ “ll ifllvfilfl Will‘
`tltduoed {tom 8.! to 4.6 It. The S tnn.tform_‘u not Pflffi-‘I 01'
`ettutt no the Knrhuoen-boevo tnttafnntt. but it am: I mood
`Ind:-off between perfinrntenoa and simplicity of 'imnlemflI-
`mien. The trmlornt also provides at man: for hietm‘-hill
`inane teptesetttatlott. Ind has been med tat some lmue
`etthnntnmenu [S1
`
`A. S~'.l'iwt.tfon-It Algorldltil
`The S Innsfot-tn ltthefladantndtnhsfottn rut 2x2-plxzl
`intanehloolt|.ToapplytJte.SttI.n:Iorttt.nnimnmlflIusth.we
`a sieeof znxzm platelsor ho apitrnprluuly pnddfll.
`Ila. lt.c.ttttd duethepixelslntltelxzltlocltusltown:
`.a
`.5
`.e
`.4‘
`then the ta-turd u-utttonu ooefnutenu at aim bvtlh
`E-¢+b+c+d
`Ag,-I-b+.:-6
`Ay-o+lt-c-ti
`(1)
`A,-at-5-e+d
`where E is the nuooooffieiettt and the A coefllcients repre-
`tent hot-lxtantnl. vertical. and dieponal spatial. frequency nom-
`ponentt. oi’ the image.
`-
`11:: S‘ tnaufiomt ptuvltlee a flIll.1'|I
`lo hientelticully de-
`euntpoee:Ia._txe|tttue.‘l'oeehievetltis.I.lte zxzlstoeluot
`the lntqe are tnnetttt-nted_nnent-din; to (ll. reettltltt; In [our
`Iuhilnllol. one lot emit coelliclent. Eich of the Iulthmpes
`hat half the Ipntlll retolndon of In original. ‘nth transfor-
`tlllliotl tIt.i.flI most ct! the energy of the 1x2 Diul block tall!
`the E cooElit:l:111:tMll the tttstt-thntimt lot the A intent: is
`narrow and reqatl.ru liner bits lot coding. pcnnifliug better
`ontopteulon. To ieltieve ltientehlcnl decomposition of an
`ltnue. the lower spaniel-teootntion 2 bone at each my I:
`further tnntlctnted while the three A in-tapes ue entropy‘-
`ooded fat stony: ot trettttttluion. This docotttpottilion step
`can be tnccenlvely applied until : lull: louse of much lower
`resolution is achieved. This result: in I tapretentatiem coo-
`tlttlngonbuia intue. foilowetlbylayertofcoded A images
`of ihcteutnc spatial resolution (tee Fil. 1). To tuconurucl
`higher resolution Imus. the E it-nnse at a given later is
`teomnltitted with in corresponding aft according to the
`Inverse tn.t-Mort-tt equations.
`In our implementation of the S ttantlorrn. the forward
`ll‘I.I1Sfnl'Il‘I
`is computed by a slightly modified set ofe¢l1Ifl-
`
`UJIS-WZGIJ,-’°1/Dfilm-D'4‘J?50l. W O l99l IEEE
`
`Page 122 of 437
`
`Page 122 of 437
`
`

`
`
`
`IEEE JOURNAL OF SOI.lD-STATE CIRCIJIIS. VOL Je. ND. 3. Mo-IRCKID1:
`a
`
`
`
`intensity vlltlee. The block Tribllf pmvides :'.sl:tI|,|.|un
`pixel
`between inlerilll and external bum. The PCB bineict in-tplg.
`merit the data transfer ntntneol between the pipeline regu.
`ION.
`The pixel port consist: of two 1!-is ninl buses for lhg
`ruler linen {mill and [ad] with aeeerltpanyina pipeline
`control Jlmais. Thus.
`this pol‘! behave: II ileouttectee In
`two mflrlilll pipeline source: (forward transform) or in my
`branching pipeline oulptltl (Inverse Innlfetml. l.tn¢£l:a.|i1r, tin
`mefflcieui port bchlwl in I similar flnhinn. HnI'e1ler._ Ite-
`clulethecleltinaliumufthl Earl-dthe Av eneificiennua
`netalwayIlheutne(A'amtntheeetier),wernalhritnv;
`three pipeline anureu (units) and uneciated etmtrule. ‘nu
`[E,A,]hun la 1-i build: to I-eenrnmndarn the lnrurvlluelef
`the A, coefficil. Sintiiarly. the [ilg,Ag] bu I115 13 ‘Wide.
`1‘L1Ite multiplexing elf the 1:lpel.i.n:
`i.np1tt/output buses and
`the inherent
`two-pltaae eoenpumion u! the S u-nrrsfunn
`enable the device In mlintain I needy me or npentlnrt at
`the sweet duel:
`trequ
`in either forward or in-vet-re
`mode. The PCB’! on the
`put: Ind ili.I.I'p‘I.lI’e oltlie device
`ensureiltnttheeeuelctll-nehrunizatinrtnnhednulrnmet
`lntltelI!oIo§icaJpi|)elineI:rtIncheelI|tlu'Itlteed.11Ie
`mictnptueennr PM Provide: a 4-2: bidireetinnai dl.l'a.bu|
`and animal Jinnah.
`lninrvtard n'lude,ihepixelporli.IIhe inputlolliedevlco
`vrhiletheeuefllcienlporti:tlteouIpuz,I.ndlhelowuhaIfof'
`the dale path h inactive. The efl'ect.l\le internal pipeline
`letental is fuur clock crelee. Only‘ the Ital! leedblck. pill: is
`aetivesndliul-edeileI';'othereydetoinnpiernenIal'Irn~ataue
`butterfly cntnpntatian. It takes two clock cycle: to aequite all
`iortrp'uteisinlhenmri:.urd:iruiwoc9:iutonrttputthe
`lent eeeflieienu. Tim the toll! input-to-manna: leteltelr ll six
`cyc|eI.[uilrt¢rIemot.ie.aiIeiententanttherimpnthnre
`operative.
`l.ll.|ZflIKILI:lIl,| an additional ad: to the pipeline
`latency. [iIIhllflIl.lIi.¢l.llpI.Ili’cilhedeVleel|.!Il.l ennffinient
`pnunlvlllietheeutnntilthepinnlpnrthgainnloeioet
`cyclelarerequitedlnfullyloednfl fnt.Ireqe!fl¢ieI1n,wIIich
`are then tranelerted oliet lite lt:lrl.|er feedbackpllll. £32, to
`the input-sun swirclt it Reduet‘. The nntnnt item the
`proneaattritpatseti through the inweritallntlhetlnta path.
`Again two t.-yeien are required no output the reeamtrueted
`viult
`'
`‘ho I-it ptngratnrnaitle registers define the noel-nine
`meduutdpannenreanunerlenflnutheeflmipu
`rnndee nfq'.lerII.lnlrt dlthe tlcvlnc: forward tniufiorm, inverse
`trana!nm1,lorwIrdbIP0ll.lndlJrvenebypa-.'l'hehrpIan
`mode: are pruvitied uurtnltle direct tnnafnroipiatelt through
`the tllip.
`OneoflIInIeatlchI.ilIIeattheaeleetetltnntppn~rtd'lu-
`noa1'u:n.'i"hnmnIrnlai|n.nlat:inck. naeynndzeueuahlethn
`device to be placed into the principal. diagnnnlic. or reset
`state. Upon newer-up nr 1 tenet condition. the device in
`nutnnurieully planed into the lorwatd truutnt-ea mode of
`operation.
`_ M min; one prneeaanr. fleedbaelt dull UIJ-ice. and little
`ntttllipietitag of the pipeline I/O, we have been able In
`achieve an eiiieiertt V15! implementation. yet one that natu-
`rnily fit: the raster i'o1-nut of scanned images. The chip is
`bund on 3 standard eeil design. in It L5-tell! CMDS double-
`rnetai process.
`it will be available packaged It I 84-Din
`!‘LCC component trperalitlt al 5 V rI:quirin| no more than
`_6iJi] mw at
`Ii] MHz. At this clock lrequenttt. the chip can
`sustain a processills rate 0120 mi1iiort,pjxe.is per scenad.
`
`'
`
`I:-[e+.t+c+d]/4
`5,,-[(n+.-.)—(a +d)],'z
`Av-ii-=+bJ—(c+d}]/2
`(2)
`A,-a—e—e+d.
`Given the tnnltoml eneifleiettn. the Oflllllll pilei: can be
`obtained by the invene trnnnfnt-nt realized by
`
`c -{(2-11+ r[s2[) + am
`5 -[[2-11+ lr[;2]) -I2]/'2
`e -[(z- e'l+ .l='[.t2]) + lay:
`a -[(2-41+ F[d2.]) — dz]/2
`when the lltlerrnediale variables :1. til, :2. and oil are
`
`(31
`
`:1-[(1-lZ+ r|n..]) + .i.,],r2
`d1- [(2-:+rlayl)—eu]/2
`:2 - [(2-A. + 1-'iAai)+ AD]/2
`(‘ll
`‘fl"l(3'5lIr""l3oD'5al/3
`nndtltefuncxhe Fl-lretunutlte inn: siutriileantbitnliu
`Il:ttn1ent.1'hlIfnlnIcllhI6eflIIull.rIn.lfeIrn eqttltlonwu
`ehetentnenrurntnnttnnnynuntlerangenttttnplnelntnany
`Eirnage rernaitutiteume t-eprdiunofthenurniterot
`hien:ehiu1lleeurnpotitionI(otreeeutrlretiens)In image Lt
`rubj¢ette.'l'hereitneinnei_ah:tifiernthr'ubeenmn ti-tn
`ieaat-t1sI1il5eut-iaitia£nrntatitniaeI.rrledintItentIte:ennffi-
`eienta.3ytne.an.tnfthell1-|i'uraetinnintheiuvetuetrnns-
`fnrntellratilhleaueteeellledtolheirfltllteeeiulinnplier
`lflluvenmputntbmoivinthettiteninelnintheottainai
`liflllelledllililedioe bitlfpfliflve ln'Ie|erIl, tIteEcDefli-
`eienturii.|reIrtIinebiu(pnd|iveLutegnrl,theA anday
`eeeffieieeuwilleaeltheb-I‘-1 biufitianed i.nteger:.an¢ithe
`a,eneuIninrtla+2hiu(:i|udinmurl.
`
`B. 5-nI‘ll.Ilf¢'fll CM: Auilltmlwn
`Fig. 2 shall the Itnhileetnle of the .|'-trutafotrn pipeline
`prneeuot[SPlPE)ei1ip. The device has three pom for
`Pl1°lI.'mcEflt:ieI1u. and the tnierwrroeeaaer-interfntx. Data
`Irlnlfer ihmulh the pill! and nnetlicinrnt porn is nnntrnllen
`W I self-rerulntu; eyrteitrnnntu pmtneol {PCB}. The proces-
`sor
`nl ene array of four aritltn-letir: unit: that imple-
`ment nddttien, subtraction. tnultiplltatien. and division by 2.
`"No feedback tine bltaenifili anti‘!-'52) amine this aI1:!litee-
`lure optimum for the cmnputltlorl of both the liurwani as
`well at the irtvene Iruufinrut. The short fnedbaek nut. FBI.
`enables 100% utilization of the ptoeeunr. In inverse mode.
`the block Mu: &.
`l..in:titer_e|uureI correctly reconstructed '
`
`Page 123 of 437
`
`Page 123 of 437
`
`

`
`‘me
`
`..--;.-~ .
`
`._
`
`.. ,_
`
`i
`
`.:im-..-'_
`
`_ ,t1.M-“rd: CHIP SISITOR LOSSLBS IMIAGE. COMPRESSION
`
`plpolnn
`sltllral
`
`Ft‘. 2.
`
`stenptilled 5 tneelertn pipelim pt-oeoaor (SPIFE) etchitecutte.
`
`III. Dams Cutttt-tustsovt Dacqueltusan
`The data compressor/doeotltpnesor (DCD) uses 1 version
`of ma l.etnpel-Ztv (L-ZJ eompmatnn -tsnrmun is]. £91. The
`L—.z ulgotithm repisces input string: with code words, creat-
`ing a eornpretted output. A node table, containing the dot:
`string‘ that each node represents. is ootI.It:I'ttt‘.ted bleed on the
`data enqountered during compression. During decompres-
`sion. the some code lnbie 1| retreated from-tlte sequence of
`code words. In the table. strings are tepnetented by the uncle
`tor I shorter string plus an additional data word. During
`encoding. a string of a single element will alums be found.
`The next longer string represented by lltlt siting plus the
`next dltl. word will he Iettltiscd for. This mucus continues
`recursively until the ioruer string is not found in the tibia. In
`this case. the longer string is entered into the settle and the
`code for the shorter string in ttlnmtitted.
`While encoding. in truurniniort is made after every fnilttrc
`and is Iccumpttnied by the creation of it new code-tnhie
`entry. This knowledge in tiled bit the decoder. Each time the
`decoder re£eiv¢_s n node. it nukes I ontruponding entry no
`its tutti: by appending the tint eltnncter of the currently
`received code to the pmriotst code and assigning 1 new code
`word to it in the oode ttthis. At the some time, it outputs the
`string corresponding to the current woe. Ilium opet-sting
`jun it stop behind the encoder.
`Ziv and Lempel [SI hue derived upper bounds for the
`oompression ratio attainable with full a pried knowledge of
`the data by fitted code-.tnh1e schetttes. They have then pm-
`mood to show that the efllcienctr of their node with no
`it priuri knowledge of the data ttpptoachet thou: bounds.
`
`A. Solutlrttfltt L-Z5¢¢rt':Pt...RflWh|'llmrti'
`When ineneodittg. n ttewdetaword is received, I prefix is
`nppended to it, and tlteoode tebie isseuehetl for leads
`word reptucnthtg tltil otrltbhtndott. In the DCD. W WNW‘
`theoodewordsi:.etoi6h'.I.hdheneehevatoIenIt:.hthrott;h
`64K toestions. Hashing [mi-[I31 was etuvtwed to perform
`the high-speed senrch and retrievnl.
`
`B..h"e.dt-irtg
`I-Iuhins it I technique
`designed to speed the nation] at‘
`ctmunnt.-tstedttnthniney K[1Di.TiterInneotx in-mlllltf
`extmncty tum; honoe Kannothedlree-ttytnodusn index
`ouddreutottndthednta.Onccotnputetatunetionf(KJ.
`tlseh|.ehfuJ!¢Ki0fl.1fltI¢h1It|IeIt!t:Iti0ItofKnndthe
`sssocisteddatninthetnhie.‘I'heh|thttIectionPruvidest
`napping {tons this tune range of K to I antler range. Such
`antappingisnptttn:'qtte;tntstefltIno!Ieentntcinbe
`mapped onto the tune ioentlon. This it called 1 oolliiion.
`The difference between ituitin: eehentei line in the III? 1110?
`ruohre oollisionl.
`The schemes mute on moment for pertotmanee. The two
`main categories of _{.'0IIil§0I'I tnloiution Ire “I:-||aIlII|J‘Il" Ind
`"open addressing." in cheinim. I link is med It each tebie
`entry to point to it chained list ot edditionoi entries in an
`overflow rnesrtn-111' that had been tnttpped elitist to the same
`location. in open nddretsing. the additions! entries are placed
`within the tune table by some predefined ntethod. In chain-
`ing. the mapping is presented. While in OPGII ldiimlifli. “"9
`mapping it lost.
`
`Page 124 of 437
`
`Page 124 of 437
`
`

`
` ‘
`
`IEEE JOUII-NA.L OI’ SOLID-STATE. C‘lllCU|'l‘S. W31. 25. ND. 3». MARLTH I39]
`
`1-to
`
`
`
`I13. 3. Averqe nrendaerntnemry prebaa vatrau merttnry required {in
`sirasbtutl.
`
`lit the “open a.dilr'eI:t':r|" due of bathing achetrter. uni-
`latm hashing has been rbcrtnt
`to be optimal
`[12], while
`easleeeert haaiI.i.n|
`is IIIDVIII. Ofllllllnl H1] in the "ciIa£n!lt|"
`other at huhlnr schemes. We aiaalyzed the pet-forrnence of
`nine rtirterent taaitlsinn remhtdon achemes: 1lnear(LtN. [10]).
`nlliforrn lu.u2. I101. I121}. nueieaend {Co.C-12. I101. [11l3- Ind
`separate chaining (Sr:1,Sc2.5cJ.Sel,
`I10],
`I111.
`I131). We
`dirt not
`t.-enaider
`the enrnn-teniy used performance
`rrtes.sure——nttmher of memory pr-nhea versua load-teeter
`(traction of the table filled)--as an important men?-IN. since
`the new load factor is diflerem aeheurea represents difleru
`em asteurttrettnen:t:rrrettulredand.rtenee.'tanetaa'uod
`measure of relative performance. The perfarmlnee measure
`eetrsedarutbeaveruemtntberofnternnryptnlterlnr
`teecttttingaa-ucr:eaaenti_aEallu:eaaaftrtn:rio1-tnrtrrrnt
`Rltmofvunplvrvi-i,'flreurrtenmeefthe anely-rirluhmtm in
`Etc. 3. From the eerfet-nanne and tntplernentatiert point or
`view. separate chaining I (Sell irll meet attractive. In chain-
`ing. the eddteae of the uwrtlw nnwtrery implicitly connlna
`the code word Ileoctlledwilhlhlreyandwedo rtuttleed to
`:atore it. thus saving rrsetrrrrsy.
`
`c. Memory smttgaretetureexryznttdrtpatirtiag
`3|!!!-‘G lht IIIIPPIIII waa presented by ehelning. the eddreea
`of the keyjit-ea information about the her and only a arealler
`lIanofitneeattrhelered.Fore:entple.lftrtehuh
`l’u.n:tionlaadivia.|nIt.thenthetpaoeienteanIreua:dealhe
`addreseend the remainder ahnreal aa the ebbrevluted key.
`
`D. Has‘: Function:
`
`The D=I'fOll'I'lIl|C= 0'’ tn! huhlu scheme depend: on the
`hub function. A need hath function spread: all the input
`key: umiarmlr screen the nttnnery range. Mrtitiottttt eon-
`urainta were plead due to the desired use of abbrnlatett
`Wt’! Ind maple Etardwara. A bull function haaed on Ilium’
`|'|'II|f'v| Tmllliniieatinn Ina developed. which satisfies all the
`above constraints. The hath lunetiort end abbreviated key
`generator car:it ennsirt cl an array of s.ttct.ustve-on gates
`WMEPI mmhiflfl bite of the input key to form the hash
`hanction and abbreviated key,
`
`'
`
`
`
`I-Dl.‘ll1'E88
`
`lflk .
`Fla. 4. Merle: etunbar of eulriet in earth Iddreat (eheaen Iran ma-
`I
`
`It is convenient to re1Ireaenr the generator: as matrices
`(Hui) oi ONE’: Ind a:ru3'l. Ind the igrjrul hey. ltaah func-
`tion. and ahlrreviltetl heat as vectors (hi?) of ante‘: and
`zEl.D'l. With this repreeenratirm. the get-teralare for the huh
`ftrttclion and ahbretriated key may he expressed '11
`(5)
`5 - HI?
`(6)
`i’ - AF
`where binary multiplication '5 performed by Art1>irt|'orl bits.
`and binary addition is perforated by e.xt:r.t:srvs min; of bill-
`II the hash function eeneretnr nnttltt H ‘ta ra.ndon1.|y created,
`the neetrltin; heal: llrnctiert lrea pod prepertiea. The diffi-
`ntlnr is in treating an aranehhed abbreviated key generator.
`We emnblne (5) and (6) into one equetlen and obtain
`
`(7)
`
`|:l-lf.‘l*-
`as last; as the etlnblned matrix
`A|"|
`il lnver-trhle. III! cflrllllnlliolr oi hllll filntfliolt it and Ibhl'B-
`viated key a uniquely represenra the orlslnal key it at
`requlred,sint:eictrtaybeetatrpuhedfrnnri:andeeaahnwn
`
`below:
`
`,1
`
`can
`lt|-
`~-lat
`Tide allow! us to create a eorrrlalned peetrdanndnm matrix.
`restricted to invertible rt1at1'lr:e, which will result in n gener-
`ator having good sratietlui rttwertim fer the hurt function
`and valid abbreviated ken.
`Once the enmbined matritt la determined. it me}? be seps-
`reled into the indltrlduel rraetrlm H and A to generate the
`hash ftrnclion and Ihbrwiated key. respectively.
`Eartemive software dmulationa urine mpreaenrattve data
`were pert‘-nrmetr to select the binary rrnrritt. Fig. 4 than the
`eveme number of entries at each memory‘ location for the
`chosen rnsu-ix. It shows the ttnifortn mending at the irtllul
`key: srtrusa the memory range. Simulations eunflnn out
`theoretical
`results of less than L2 probes per key. A: a
`contrast. Fig. 5 show: a different matrix used as the hash
`function. It is ell!’ to see that mernury utilization is not ea
`unifnnn. resulting In many "clusters" of over and under--
`
` .__
`
`Page 125 of437
`
`Page 125 of 437
`
`

`
`
`
`
`
`utilized locations. This huh function requires rnote than
`[our probe: in resolving each key.
`Since all
`the algebra involved in the above binary upm.
`tit:-its is modulo two. the lurtlurrte Ir sharpie. consisting only
`of zxctmtvz-on plea.
`Additionnl memory savings are obuinod when the main
`memory marlin: only it link field. This field points to the
`overflow rnerrtonr containing the bar. the inrpllttit oode word.
`and further link fields. Refer to FEI. 6. K it hultvad to the
`link table lmlin rnernorr.'I by tlte hash Ettnction. If there it no
`linlt at tlttt location. .tltert we have I hilure end a link I:
`created to the next mileblu erttr location in the overflow
`memory where K is inserted. Ortstlttt other lurid. it there il a
`little at Ilm'locrtion..we follow it and loolt for K in the
`chained list of the overflow memory.
`A diradtrantaie or the reduced main memory sclterne la
`lhlll a rninitnurn of two meteor!’ recesses is required to find
`each K. This problem can be oltercome by plp£Hl1|I1.| the two
`ICEEIIEI.
`.
`
`
`
`“tnMICROHDEBSDIJKHGDRY IITLIFACE
`
`HP-Pnlfl‘
`K.!1'_CI'lD.hTA_.'llEfl)i PORT
`TAO-|l!HJ|.IJ'D Full‘
`
`at 1.
`stwuttcrocottnrxaturlm.
`
`Simulation: [3] have lbulm that the L-Z II[Ol'ilIl.Ill‘fl3l1l-
`prerre|aniuta¢ehyatIcnorot2to3.witltnvennert:'lnl-
`lengtluofbettreen 3and4.1'hotnfure,trhileoe|rchinsllora
`pn-.fl.t-dun eornbinnion. ruuceu will occur lhree timer or
`often as hilute. with tho knowledge. we can anticiplte It
`prefix oefouthereuehllrooolvodlndwecnnbomnet-1
`about 7595 of the time. I! we anticipated Incorrectly. I'=_
`ditteglrd the incorrect
`link Inrl use the out anticipated _
`prefbr. T311: bIt:!lttrIc.HD| method llloul efficient pipeflnlnl
`oftlre mainlliuklmd overilm-‘(test Induode wotdlnremonr
`Iwesset to improve weed.
`
`F. Cirtprtrtznitrrmn and Fatwa
`'l1teDCDel-tipeoruiruoftwe-16-hFIFO‘s.reodee
`(Including enoodef. decoder, LIFO luoefiry for deoodlnll
`node-word unentlon block. and I II-llb-hvel mntrollotl. I
`reprelter. and a rniu-otrroeeuor interface section (Fig. 7}.
`The bidirectional FIFO’! In In smooth the htmnt flu fllf-I
`inherent
`in the compression and
`I 9'03"“-
`Durirt; coding. the decoder is idle. Ind during tiewdlfltt 3313
`encoder is idle. ‘Rte bidirectional rt.-meter converts n_rl:l:le
`width codes to I rm output width during eotnpteutoo or
`vice Venn in deeodiru. The mieruotoeeasor interiaee pro-
`vides control and dlunottie: of the DCD.
`The cadet: is shown in FL: 3. The code-Ward generation
`block (cw bloclt) is common to both Gfll73dlJ'I[ and dfl70l'.lllI.I.
`The eodec int: a hierarchical control structure. Each at the
`block: has various low-level state machines to mike local
`
`E rllrarithrrt Impfenrutlarrbn Mtfl .HI.ltHl'l.[
`When the idea of pioelirlins main and overflow nrentory
`were: is encoded to the L-2 algorithm, 3 pownllll prob-
`lem arises. The rlwirt memory cannot be Ioceascd without the
`prefix and the prefix will not be Inlhblu without lCDelsil'I|
`and resolving the search in the overflow ntemory. However.
`inherent pmpetties of the algorithm do allow us to pipetine
`ll.
`
`Page 126 of 437
`
`Page 126 of 437
`
`

`
`‘tau
`
`‘
`
`IEEE roustmu. or SOLID-STA1'fiCllt{.‘l.ll'l'S. vol. at. no. r. t-siutctr mt
`
`decisions. The high-level state rnschine.hi1evel.sni.. monitors
`the blocks end controls all oonsmurtiestlons with the outside
`world under interrupt control. Breaking the control into I
`hicrsrcifl enlblcd sir:npli.Mn¢ the state machines and provid-
`ing a urtifonn view of the processing (encoding or decoding)
`to the outside world. This spproscis. however. made synchro-
`nization of various events more cosnpllcatcd. which was
`resolved after extensive simulations of the hardwtsre.
`Use of the DC!) chip requires trln estemsl memories to
`store tables. One memory is
`IJSKXIG ll or RAM plus
`lzsxxl is of resettsble memory. The second memory is
`64Kx.'i: bnfRs\M.TIseDCl3ltssress:tanddcich:inputs.e
`iiidiwfiiififlll ‘I311 I‘-‘DH. I bidirectional code port. a micropro-
`cessor interface. and I direct mentor? sccess port for diag-
`nostics. ‘1'he.dI1I eon consists or s 154: bidirectional port
`and three pipeline control signals. The pipeline control sis-
`nsls are handshake siutsls tn ucilitele synchrotron: transfers
`to and from the
`t)uringcodlng.theDt‘.‘Dwillaeceptdstastsword traits-
`fer rate of appmnimetely 1.5 MR2.-During decoding. _lltc rats
`lsh.iIller.Il:out8MTr‘Iz.'l'l:eaet1ss.lntuIr|ierttrl'I:late bluuscd
`bytheDCDiI9IIllIIlltsns.hle.'l'hecodeporrennsistsola
`16-h htdlsectinnsl post and three pipeline control signals.
`The insurer rare at the code port is llmieed by the transfer
`rate at
`the dsta- port and the actual compression rstlo
`achieved. The rnssietuer number elcede hits used by the
`DCD is props-atttnssltle.
`The chip has sixteen 84: control registers visible in the
`ptostsrnmer. All the transfers are under synchronous micro-
`processor protocol cnntrnL The registers control the direc-
`lionioompress Ot&al1't||tBnJ,tllIlII.1lI'IiiIc1'oft‘htl sndoode
`hits. and the first sliocatsrl code word (sliows for keeping a
`range of reserved code words). Till? Ibo provide for insert-
`ing and sernoviiu code words in the code streets. The
`cstensaleodie-hookrtnsneriesranheseeessedlythe DMA
`cl’ 1 start sddrul ill the control segkten [useful for disp-
`nostlcs). The output code words csp be packed into any
`nstnriser oi hits (He) in the repsclier. wltlclt it pmsramnsed
`using control registers. 'l1re DCD can be programmed to flu
`an interrupt under vsrinsts conditions (for example. on re-
`ceiving e renewed codeword creeeousttering an error). 11::
`chip can also be operated in I shulevstep mode for diagnos-
`tics. The DC!) on be programmed for automatic reset when
`sheusdebooklsfuil.orcnsrcnrtlin1retvithoutttssil.inssny
`new table entries. The DCD is s srsrtdssd cell design. imple-
`mented in Sipeetlce’ 1.5-um CMOS lecltnolo-gy.
`Tito DCD lID0l'lI'|‘llfibl8'fll‘IIl.cxiIllll]ullI'lpfeIIiDfl soft-
`ware, rwrrpuu Ind lznorrtp,
`linund on UNIX and VMS
`systems. The speed and oorttpatililllty of the DCD makes it
`I-KIM ‘W wflilmession in dist/tspe storage and satciiitef
`micron-eve/IAN eermnunloseinn Imerns.
`
`IV. TI-is Cosmmstnu Ssrteu
`Fig. 9 shows s diagram or an insane compression/decom-
`pression system (CD8! employing the chip set described
`above. The configuration shown assumes that
`the CD5 is
`part of a letter inure processing system with a common
`i.n'ls3e system but for control end imlpe transfer. However.
`simpler ones are possible. The figure shows only the princi-
`Pll em Pltils or the CD5. Apart from the srms and DCD
`chips iincludlnl external memories {or the code tables). the
`CD8 conIains1tiree_fF_lFO's. a local in-tau hutfer (LIB).-an
`notional quantiser..and e controller. ‘No date interfaces
`
`accomplish Lise transfer of pixels and coefficients. We have
`designed such s srltem around the VME. bus as s two-bmrd
`set. i.e.. SPIPEXLIB board and DCD/qunntizcr board.
`The quantizer is optional and is provided for spplleations
`pemtilting lousy ootnprtssion. All FIFO’: are needed in
`either mode of the CDS. Two or the FIFO‘: operate as s
`multiplexer or demultipleser oi their inputs: the third oper-
`stes in srrsigltt input/output mode. The LIB is lsrgo enough
`to hold the fires E huge of the largest imsgle size the system
`is delittned fer. e.e.. tor a 2lDc2.K-plsei image. the LIE holds
`1 million piseis. The LIB substantially reduces the traffic on
`the image Iyslclll hits for hierarchical compression or decom-
`pressicrt or large images with men)‘ decomposition levels.
`In compression, phtels from the original image stream into
`the CDS end up dcmultlplened by the pixel FIFO into two
`tester stresnss. i.e.. the [o.h]- -- end [e.di-'- lines or the
`image. subject to pipeline syncttrnnizslinn principles. Then
`the SPIPE ttlrrsforrsrs the image. ‘Ilse E coefficiertts are sent
`to the LIB where they
`internally or-cruised as two raster
`streams—tht.tt the LIB be
`use sstsro Fll-‘0‘s' Meanwhile.
`the A‘: are huflered up in the coefficient E-‘l'FD. The enchi-
`clent FIFO snolripicmcs the A's that pass Iiunush the quan-
`wear and the DCD. The output of the DCD is buffered up in
`the code FIFO and is
`that
`transmitted over the image
`system bus to its final destination. Tltls sequence cnntirtues
`until the entire origins] in-tau hgs been processed, l.e., level
`I of the hicrllclty |t.I.s been completed In this phase,
`the
`isnspc system bus ishuslest, having in support simultaneously
`piselandeodetnnsters. Itthellnihimseeistnbe further
`decomposed, Le..forleveI2.the Llllpreiridestltetworsster
`streams [st.b]---
`sud [r:.n']---,
`thus tile pixel FIFO is
`insctive.1'lteenntpressbnio1lowstltessmesequenccss
`iscinre,httt oostthe L[Bi.shnt.hssourwttnd aslnkforf.
`sulsinssgandtll-eonls'1rui‘l'leontireirns¢eI'J'I|£I=IlvIIIi-I
`that or code tssnslers. Further tilerssehicsi decomposition
`steps may be aclsieued until the nrisilest lmsps_is in the LIE».
`At this point, we lsnlre the option of direcfl! lwttirlng the
`smallest
`losses from the LIE thml-Isis the pisel FIFO or
`passing it to the DCD for coinpreerion.
`lntelfi
`In decompression. we first
`load up the smallest
`either directly into the LIB tlsrnush lite pixel FIFO or after
`passing it through the DOD for decentpressiert. To recon-
`struct the nest higher resolution image, the coded n’s stream
`Into the DC!) and are decotnpreseed.'dernuiIipIes:ed by the
`ooeffltzient FIFO. combined with the sinsllest image from the
`LIB. end inverse-tr-stsrionncd hy the SPIPE. In this phase,
`the Llflls holltssottroa and ssinkoltimdsel. and rheonbt
`t1-stfle on the image system hus is that or coded A's. Once sll
`the A's for this layer are processed, the higher resolution
`image is resident in the LIB. At title point. the itttspe may be
`directly accessed throulh the pixel FIFO or dwndirig of a
`higher resolution image may he initiated. The reconstruction
`process proceeds as before until the full-rrcsolulion image it
`achieved. Al the fins! level. the pixels are directly pulsed
`throuph the pixel FIFO to the system bus so that in this
`phase the image system bus Irsmc is made up of both ended
`A's and reeeststruetml pixels.
`The above system has heen designed around the industry-
`stsndard VME bus. For this implementation. the LIB ac-
`eomrnodatcs I million planets, the pixel FIFO consists of two
`[K-word bullets. the coefficient FIFO consists at one 2K
`and 4K word buffers. and.
`the node FIFO of it
`lK-Word
`hufler; Several Dl-iA_ channels are provided to speed)-ID
`transfers in the lnnsse system bus. Although uttty one D-CD
`
`Page 127 of 437
`
`Page 127 of 437
`
`

`
`SHAH at el:D-H1‘ set} '1-out
`
`mace COMFIIBSSIGH
`
`M3
`
` --e-a---......--..-.- .-.-----.----n-.--..-.-..-_.--
`
`rev. Muueumwameuamummmmu-emmsruzmnmauum
`
`devicciuhmvnL||Fi|.9,itispcIIibhI1ohIvaupUoth:cuLf
`the nppttcatinn require: the A‘: to be sent-ately command.
`Isnppozed la treating than u 1 trlpiet.
`
`V. R.a:uL1‘e
`In [31 extensive sirnuhttnns wen perbrmed int data oom-
`prcssion using the S tmuhm end Ltmpel-zlv coding. ‘me
`nnin result: :1: Iumuuliud heat. The S tnnstmm I¢|'lll'-
`ally tedtteed the cntrmy only about ll effectively us fits!-
`otder DPCM. For III: 15 computed redltunph Wilma: used.
`theuroth-urderennnnwuteduced nnnnevenutrom
`8.8to-t.6b.TIti:ienouurp1ie£n:.u:heflJt:nLnute5
`I1-ansformeteentemehv IIIIPI-¢.1Iflt]ll'aI1dl.11[lV¢»t!'IDDd
`Iepu-ntiun between the frequency bend: (the 2 end the A's).
`ll, howwer. bathe edvmtalenllhmie implemanmion and
`hientchlenl tum: decomposition.
`.
`Thcahovel!imagu.uchtI1thmod|inalwordai::oflU
`b/piuLIvcrecuc|cdbyLhcL—Z|1prithmxflerhein;S-
`II'lnsIuI-Ined. An evenieoulwreuim nth uf2.09 (Iron: 10
`Maine! down to 4.18 b/vi!!!) we: achieved. ‘nuts. for these
`intesemlhecunprestinnperfotmneewuverydue mute
`mductlolt of the tenth-order canopy la‘ the S tnmiutm
`flllfll an man at 5.5 b/pin! to 4.6 hf|gbnL
`
`_
`
`itatitiEamgag5%E§§E:atE.3‘!RE-='Ef:32;;EEEi?!E
`
`9ae“§§“egggT"€§§g§§§:i§Egg?3,-gfigg5%;gig}:*3%?’§%1:???ii§%‘*i%
`
`[4]
`
`I51
`
`I61
`
`I71
`
`VLCGIEMIWN
`The dnocflbedtwflflhiplinrmlhehadaofuhlglt-tpeed
`lossleu image compmllonfdacomvtnsiou system. The S
`truuform. besides damrntlntiu Ihl inane. Mwida I. came-
`nlent method at ltiermtthal Imus deeotnpoeflloa. ‘me date
`commence/deeomprcnor {C is I rent and enldent Incub-
`raentatinn or the L-Z nlprithm.
`Such a system reduxl Itanne remliremenu in Itlghvspced
`image mhivel and dmbue appiicutlotte end reduce: the
`transmission time of dlglnl
`lflllltl our communication
`channels.
`
`tition. and A. Miran for his support.
`
`AcauaowLencIuaH1'
`The authors would like In flunk It Appleptc, I-I. Hlume.
`Ind 1'. Wendie: tar very useful technical dilcutlium, J. Stun
`and G. Hyland for help in mloul Stile: cf VLSI iuwletuen-
`
`I
`
`utlltor of various
`
`
`
`Page128of437
`
`Page 128 of 437
`
`

`
`IEEEJOUIINAJ. OF 50l.l.D3'TA"l'E C'lRC1.|!'|"5. VOL 25. NOJ, NAIIGI-I LWI.
`
`
`Smear Member (If the Resunzh Siafl ll Inn Johan tuagimuhn 3.3..‘-.'.I-2. dglpua
`
`
`
`
`
`
`
`
`OII Aflfllni-Aalnl ru::‘rved Ih: B5. deans:
`roar.
`hum Ranmtm Pblyiechnie Iwiluta.
`xifl plI1I'u:| bun aha University o.fSc1':nue and
`NY. In 1915 and the M.SJ:'.E. demo hum
`Tochlldoqr. Ghun, in 1913 Ind the M5.
`swam: Lluiwniny. Symun-.. NY. in I9?!-
`dune in aaetrical gngineerins from Lchigh
`Ulviunity. Buthlzhem. PA,
`in 1917. Until
`Hc hu held position In Praise: Lnder ll
`lfl|_ ha 0|: unmilnd in UI: Elcatficll EMF
`P1_|il|.D1' Hwum CATV nnd Ozneul Elu-
`nu:-m Pl1.D.11r0|n-In 1: unigh.
`tnc-HaiIcurr¢uflvaP«:5oc1h-lInII¢1‘II
`Philip: LI-bomorlu. Bliuclilf Mmw, NY.
`He mush: Phyiici it ill: University of Sci-
`when ha il illxivud In VLSI Iitnll amm-
`
`
`cnae Ind Tefllndon. Ghnn. until he join-and
`P|I.||_h=Il.Aboru.ariuiaJ9§O. Hsilnowa
`laa and mlema. Ha lm appllog for and
`
`
`rewind pmnu In the ma ufmlctupuowr
`
`
`.BfiI.r¢lHMI.wf.NV.Hu|1.uhc:nimuhedII1
`Iar:1nI|m,vi:kona'Imhllng,d|umu|pnuiou.I1hid5;iLIlIi;nnl
`~ puiumznlsnimnu.
`moaning.
`V1.51. commuter uchmuun. ml mbruprupusor nntanu rmuuh
` ncth-ides.
`
`Page 129 of 437
`
`Page 129 of 437
`
`

`
`still Picture
`compression
`standard
`Gregcw 14
`\:*.'a!1ac:—
`
`Page 130 of 437
`
`Page 130 of 437
`
`

`
`
`
`
`
`"e.-.._._.rmi-=_sp-is-..—-e--' (
`
`civctnccs over the past decade
`in rnanv aspects oI' digital
`lechnulog_t'—especiall‘v‘dtWices for
`image acquisition, data storage. and
`bitmapped printing and display-

This document is available on Docket Alarm but you must sign up to view it.


Or .

Accessing this document will incur an additional charge of $.

After purchase, you can access this document again without charge.

Accept $ Charge
throbber

Still Working On It

This document is taking longer than usual to download. This can happen if we need to contact the court directly to obtain the document and their servers are running slowly.

Give it another minute or two to complete, and then try the refresh button.

throbber

A few More Minutes ... Still Working

It can take up to 5 minutes for us to download a document if the court servers are running slowly.

Thank you for your continued patience.

This document could not be displayed.

We could not find this document within its docket. Please go back to the docket page and check the link. If that does not work, go back to the docket and refresh it to pull the newest information.

Your account does not support viewing this document.

You need a Paid Account to view this document. Click here to change your account type.

Your account does not support viewing this document.

Set your membership status to view this document.

With a Docket Alarm membership, you'll get a whole lot more, including:

  • Up-to-date information for this case.
  • Email alerts whenever there is an update.
  • Full text search for other cases.
  • Get email alerts whenever a new case matches your search.

Become a Member

One Moment Please

The filing “” is large (MB) and is being downloaded.

Please refresh this page in a few minutes to see if the filing has been downloaded. The filing will also be emailed to you when the download completes.

Your document is on its way!

If you do not receive the document in five minutes, contact support at support@docketalarm.com.

Sealed Document

We are unable to display this document, it may be under a court ordered seal.

If you have proper credentials to access the file, you may proceed directly to the court's system using your government issued username and password.


Access Government Site

We are redirecting you
to a mobile optimized page.





Document Unreadable or Corrupt

Refresh this Document
Go to the Docket

We are unable to display this document.

Refresh this Document
Go to the Docket