Scalable Multichannel Coding with HRTF Enhancement for DVD and Virtual Sound Systems *

Transcription

1 PAPES Sclble Multicnnel Coding wit HF Enncement for DD nd irtul Sound Systems * M O J HAKSFOD Centre for Audio eserc nd Engineering, University of Essex, UK CO4 3SQ A sclble nd reverse comptible multicnnel metod of sptil udio using trnsurl coding designed for multiple-loudspeker feeds is described wit focus on ttining optimum er signls A Fourier trnsform metod for computing HF mtrices is employed, including te genertion of subset of bnd-limited reproduction cnnels Applictions considered embrce multicnnel udio, DD, virtul relity, nd telepresence 0INODUCION e purpose of tis pper is to investigte ow trnsurl processing cn ennce conventionl multicnnel udio bot by embedding perceptully relevnt informtion nd by improving imge stbility using dditionl loudspekers integrted wit supplementry digitl processing nd coding e key objective is to cieve sclbility in sptil performnce wile retining full comptibility wit conventionl multicnnel formts is enbles te system in its most bsic form wit unprocessed loudspeker feeds to be used in conventionl multicnnel instlltion However, by pproprite signl processing dditionl loudspeker feeds cn be derived, togeter wit te option of exploiting buried dt to extrct more signls in order to improve sptil resolution e system is terefore ierrcicl in terms of number of loudspekers, cnnels, nd ultimtely sptil resolution, wile in its simplest incrntion it remins fully comptible wit te system configurtions used wit multicnnel DD-A nd SACD reply equipment e multicnnel cpbilities of DD 1 tecnology [1], [2] were designed to ennce stereo 2 sound reproduction by offering surround imge nd improved envelopment cpbilities Normlly multicnnel udio encoded onto DD ssumes te IU stndrd of five-loudspeker configurtion driven by five discrete wide-bnd loudspeker feeds However, limittion of tis system is te lck of metodology to syntesize virtul imges cpble of tree-dimensionl udio (tt is, perception of direction, distnce, nd eigt togeter wit coustic envelopment) *Presented t te 108t Convention of te Audio Engineering Society, Pris, Frnce, 2000 Februry 19 22; revised 2001 October 10 nd 2002 July 15 rter tn just sound effects often (ltoug not exclusively) ssocited wit surround sound in ome teter context e IU five-cnnel loudspeker configurtion cn lso be poor t side imge locliztion, ltoug tis deficiency is closely llied to sensitivity to room coustics Neverteless, DD formts still offer only six discrete cnnels, wic if mpped directly into loudspeker feeds remin deficient in terms of imge precision, especilly if eigt nd dept informtion is to be encoded e tecniques described in tis pper support sclble sptil udio tt cn remin comptible wit conventionl multicnnel systems It is sown tt in tis clss of system, under necoic conditions, signl processing cn be used to mtc teoreticlly te er signls to eiter rel or n equivlent sptilly syntesized sound source Also, in order to improve imge robustness, directionl sound-field encoding is retined s exploited in conventionl surround sound to mtc te imge syntesized troug trnsurl processing It my be rgued tt s te number of cnnels is incresed, tere is convergence towrd wvefront syntesis [3], were by defult optimum er-signl reconstruction is cieved However, te proposed system is positioned well into te middle territory 3 nd is fr removed from te rry sizes required for brodbnd wvefront syntesis Consequently, from te perspective of wvefront syntesis te trnsition frequency bove wic sptil lising occurs is locted t reltively low frequency, implying tt for te proposed system te core concepts of wvefront syntesis do not pply 1 Includes bot DD-A nd SACD formts 2 Of Greek origin, mening solid, stereo is pplicble universlly to multicnnel udio 3 A rnge of 5 to 32 loudspekers is suggested 894 J Audio Eng Soc, ol 50, No 11, 2002 November

2 PAPES over muc of te udio bnd It is empsized tt n n-cnnel system does not necessrily imply n loudspekers Indeed, s is well known, it is possible for two-loudspeker system to reproduce virtul-sound sources [4], wile using more tn n loudspekers cn elp crete more robust nd stble illusion Also, te mture tecnology of Ambisonics [5] [7] is sclble nd cn ccommodte bot dditionl loudspekers nd informtion cnnels However, ere te encoding is ierrcicl in terms of sptil spericl rmonics, ltoug no ttempt is mde to reconstruct te er signl directly t te listener Consequently te pproc tken in tis pper differs in number of fundmentl spects from tt of Ambisonics, especilly since tere is no ttempt to trnsform sound field directly into spericl rmonic set us it remins for future work to estblis te reltive merits of tese pproces ltoug, becuse similr loudspeker rrys re used, tere is no fundmentl compromise sould te system be used eiter for Ambisonics or for conventionl surround sound encoded udio e metod of sptil udio described in tis pper uses conventionl loudspeker rry to surround te listener nd to reproduce directionl sound field In ddition, er signls re simultneously syntesized using ed-relted trnsfer functions (HFs) mtced to te source imge, were it is ssumed in ll cses tt loudspeker trnsfer functions ve been equlized or oterwise tken into ccount A number of exmples illustrte te computtionl metods, wic include pirwise trnsurl imge syntesis 4 reported in erlier work, were some preliminry experimentl results were lso discussed [8] [10] to estblis te efficcy of te metod is tecnique is especilly well mtced to multicnnel multiloudspeker instlltions, were trnsurl coding cn be pplied during encoding nd recording wile processing witin te decoder locted witin te reproduction system cn ccommodte bot dditionl loudspekers nd loudspeker positions tt differ from tose ssumed t te encoder Consequently for n n-cnnel system it is strigtforwrd to employ only n loudspekers, ltoug dditionl loudspeker feeds cn be derived, wile still retining correct er signls, eiter by using mtrix tecniques or deterministiclly witin te DD-A formt using dditionl embedded code However, it is empsized tt in te simplest configurtion, using only direct loudspeker feeds nd provided te loudspekers re correctly locted, tere re no dditionl decoding requirements nd te system remins fully comptible wit ll existing recordings Alterntive tecnologies suc s Ambisonics [11] ve used fewer loudspekers togeter wit sopisticted mtrix encoding Also, tere s been substntil reserc into perceptully bsed processing to reconstruct treedimensionl environment using only two cnnels nd two loudspekers More recently DOLBY E 5 s been introduced s mens of syntesizing center rer cnnel using nonliner Prologic 6 processing pplied to te rer two cnnels of five-cnnel system However, tis tecnology is imed principlly t surround sound s conceived for cinem nd ome teter, wit bis towrd sound effects nd mbience cretion Neverteless tere SCALABLE MULICHANNEL CODING IH HF ENHANCEMEN exists grey re between cinem pplictions, music reproduction, gming pplictions, nd te syntesis of virtul coustics, especilly s t teir core te sme multicnnel crriers cn link ll systems It is terefore not unresonble to nticipte some degree of convergence s similr teoreticl models pply Also, wit conventionl multicnnel tecnology it is often te listening environment nd te metods used to crft te udio signls tt impose te gretest performnce limittions Multicnnel stereo on DD llows for improved metods of sptil encoding tt cn trnscend te common studio prctice of using just pirwise mplitude pnning wit blending to mono It is conjectured tt by including perceptully motivted processing, tree-dimensionl soundscpes cn be rendered rter tn just periperl surround sound Complex HF dt by defult encpsulte ll relevnt sptil informtion [tt is, interurl mplitude difference (IAD), interurl time difference (ID), nd directionl spectrl weigting] nd form generlized pproc However, to reduce signl colortion, metod of HF equliztion is proposed wit n empsis on crcterizing te interer difference signl computed in te lterl plne e extension to eigt informtion in te equlized HFs is lso discussed briefly In Section 521 specil cse is presented for nrrow subtended ngle, two-cnnel stereo were it is sown tt er signls derived from rel coustic source locted on te rc of te loudspekers cn be closely syntesized using mono source wit mplitude-only pnning Criticlly in tis exmple, te HFs re defined by te ctul loctions of te loudspekers, nd so re mtced utomticlly to n individul listener s HF crcteristics is is n importnt spect of te proposl, wic is directly extendble to te metod of pirwise ssocition were mismtc sensitivity between listener HFs nd trget HFs is reduced is pproc lso encpsultes succinctly te principles of two-cnnel mplitude-only pnning stereo wile exposing inerent errors s te ngle between te loudspekers is incresed o summrize, four core elements constitute te proposed sclble nd reverse comptible sptil udio system: A vector component of te sound field is produced s loudspeker rry surrounds te listener following conventionl multicnnel udio prctice Pirwise trnsurl tecniques re used to code directionl informtion nd to crete er signls mtced to te required source signl Mtrix processing cn increse te number of loudspekers used in te rry wile simultneously preserving te er signls resulting from trnsurl processing nd nonoptimum loudspeker plcements Embedded digitl code 7 [12] is used to crete dditionl cnnels for ennced resolution wile remining com- 4 Subject to Britis elecommunictions ptent ppliction 5 Dolby Lbortories, cnnel extension tecnology to AC-3 perceptul coding 6 egistered trdemrk of Dolby Lbortories 7 Applicble only to te DD-A formt J Audio Eng Soc, ol 50, No 11, 2002 November 895

3 HAKSFOD ptible wit te bsic system lredy ennced by pirwise trnsurl processing 1HF NOAION A set of HFs is unique to n individul nd describes continuum of coustic trnsfer functions linking every point in spce to te listener s ers HFs depend on te reltive position of te source to te listener nd re influenced by distnce, reflection, nd diffrction round te ed, pinn, nd torso In tis pper HFs were derived from mesurements tken t B Lbortories (BL) utilizing n rtificil ed nd smll micropones mounted t te entrnce of ec er cnl Mesurements of edrelted impulse responses (HIs) were performed in n necoic cmber t 10 intervls using mximumlengt-sequence (MLS) excittion, nd te corresponding HFs were computed using time window nd te Fourier trnsform o define te nomenclture used for vrious HF subfunctions, consider te rrngement sown in Fig 1, were te listener s ers re lbeled A (left) nd B (rigt) wen viewed from bove In sound reproduction system ll sound sources nd loudspekers ve ssocited pirs of HFs, uniquely linking tem to te listener, weres in tis pper tese trnsfer functions re clled te HF coordintes for ec object given In Fig 1 te single sound source s te HF coordintes { x, }, wile te tree loudspekers 1, 2, nd n, wit rbitrry positions, ve te coordintes { (1), b (1)}, { (2), b (2)}, nd { (n), b (n)} In specifying te loudspeker HF coordintes, left rigt designtion cn be included wen te loudspeker rry is known to be symmetricl bout te centerline Consequently l (r) denotes te HF between te left-nd loudspeker r nd te left-nd er A However, for rrys ving only tree symmetriclly positioned loudspekers (left, center, nd rigt) simpler nottion is used in Section 3, nmely, { l, Fig 1 Definition of HF nottion for n loudspekers positioned rbitrrily round te listener PAPES lb }, { c, cb }, nd { r, rb } It sould be observed tt in typicl HF clcultions, suc s te evlution of te positionl trnsfer functions G nd G L used in trnsurl signl processing (see Section 31), er-cnl equliztion need not be incorported, provided te sme set of HFs is used for bot loudspeker nd imge loctions en te er-cnl trnsfer functions cncel, ssuming tey re not directionlly encoded For exmple, in Section 4 Eqs (23) nd (23b) describe typicl trnsurl processing to derive te positionl trnsfer functions G nd G L,were ny trnsfer function components common to ll HFs cncel 2 HF EQUALIZAION e HFs used in trnsurl processing revel frequency response vritions tt my contribute tonl colortion wen sound is reproduced In tis section strtegy for equliztion is studied tt reduces te overll spectrl vrition, yet retins te key ttributes deemed essentil for locliztion A simplified form of HF is lso defined, wic cn prove useful in multiloudspeker systems 21 Metods of Equliztion en sound is reproduced over conventionl multiloudspeker rry, were for exmple signl is sptilized using pirwise mplitude pnning, equliztion s function of direction is not normlly employed In suc system sound is perceived generlly s uncolored, even toug te ers, ed, nd torso impose direction-specific spectrl weigting However, ltoug HFs used in trnsurl processing tke ccount of bot te source loction nd te loudspekers, reducing frequency response vritions cn meliorte tonl vrince, wic my become ccentuted s te pntom imge moves wy from te loudspeker loctions nd wen te listener turns wy from te optimum forwrd orienttion Also, systems re rrely optimlly ligned nd exibit sensitivity to smll ed motions, bot of wic mp into frequency response errors in te reconstructed er signls Consequently te im is to introduce minimum spectrl modifictions commensurte wit cieving sptiliztion It is proposed tt for imge locliztion witin te lterl plne te reltionsip between te complex interurl difference signl nd te signl components common to bot ers is te criticl fctor is conjecture is bsed on te premise tt for lterl imges, spectrl components common to bot ers relte closely to te source spectrum weres te interurl spectrum is strongly influenced by source direction Consequently it is rgued tt modifiction to te common spectrum cuses principlly tonl colortion, weres te reltionsip between common spectrum nd interurl spectrum is more criticl to locliztion, even toug spectrl cues embedded in te source cn induce n illusion of eigt is pproc my be extendble to include eigt locliztion, ltoug it is recognized tt dditionl spectrl weigting of te monurl component cn be required following, for exmple, te boosted-bnd experiments performed by Bluert [13] 896 J Audio Eng Soc, ol 50, No 11, 2002 November

4 PAPES 211 Lterl-Plne HF Equliztion e proposed metod of lterl-plne HF equliztion first trnsforms ec HF pir into sum nd difference (M-S) coordintes nd ten performs equliztion on te corresponding pir by dividing by te corresponding sum spectrum It is proposed tt ll HFs in set sould be equlized using tis tecnique in order to mintin reltive group dely nd, wit pproprite weigting, reltive level As defined in Section 1, let te HFs for given source loction be x nd, nd let te corresponding complex sum nd difference trnsforms be HSUM x nd HDIFF x us HSUM x x HDIFFx x (2) Four metods of HF equliztion tt mtc tis objective re identified, were { xe, xeb } re te resulting HFs fter equliztion Metod 1: Equliztion by te modulus of te complex sum spectrum, xe xeb x x x nx nx Metod 2: Equliztion by complex sum spectrum, xe xeb x x x nx nx (1) (3) (3b) (4) (4b) Metod 3: Equliztion by te derived minimum-pse spectrum of te complex sum spectrum, xe xeb x exp( conj( ilbert( log( bs( exp( conj( ilbert( log( bs( x x ))))) ))))) nx nx (5) (5b) Here nx re te normliztion coefficients clculted to mintin te reltive levels fter equliztion of ll HF coordintes in te set Ec form of equliztion delivers identicl mgnitude spectr in te HFs, ltoug tere re vritions in te time-domin wveforms resulting from pse response differences o illustrte tese vritions, consider n exmple HF pir corresponding to nominl 30 off-xis imge source Fig 2() sows te mesured HIs, weres Fig 2(b) (d) presents te impulse responses resulting from ec form of equliztion in order tt bot pre- nd postring cn be compred Fig 3 sows te corresponding mplitude spectr before SCALABLE MULICHANNEL CODING IH HF ENHANCEMEN nd fter equliztion, nd Fig 4 illustrtes te sum nd difference spectr, gin before nd fter equliztion In selecting potentil equliztion strtegy, it is necessry condition tt te reltive time difference between HF pirs be mintined Also, te time-domin wveforms sould not ccentute or exibit excessive prering or postring, s tis cn produce unnturl sound colortion Altoug ec equliztion metod meets te principl objective, te tecnique of forging te denomintor from te minimum pse of te sum spectrum yields results wit minimum prering In essence, te minimumpse informtion common to bot er signls is removed, leving minly excess pse components to crry te essentil time-dely informtion Equliztion using te complex sum spectrum (metod 2) lso yields results close to te requirements However, inspection of Fig 2(c) sows tt te rigt-er response, wic in tis cse s te greter dely, exibits prering extending bck in time to te commencement of te left HI However, experience gined wit equliztion s reveled tt certin imge loctions, prticulrly towrd te center rer, cn yield excessive ringing fter equliztion Consequently furter equliztion vrint is proposed is is similr to metod 3, but it differs in te wy te sum spectrum is computed nd is defined s follows Metod 4: Equliztion by te derived minimum-pse sum of te moduli of ec complex spectrum, xe x exp( conj( ilbert( log( bs( ) bs( ))))) xeb exp( conj( ilbert( log( bs( ) bs( ))))) x x nx nx (6) (6b) In te denomintor tis lgoritm fctors out te interurl time difference between left nd rigt signls, wic oterwise mp into rtificil mplitude response vritions in te complex sum spectrum As suc tis procedure could be rgued to be better estimtor of te common spectrum, s umn uditory processing does not sum er signls directly Overll te effect on HFs is minor Fig 5 presents results tt sould be compred directly wit tose in Fig 3 Finlly furter vrint of equliztion is were n verge of ll sum spectr is formed nd te HFs re modified following procedures similr to tose reported in tis section but wit prticulr empsis on te minimumpse nd sum-of-moduli tecniques However, in tis cse, since ll HFs re modified by common equliztion function in wy similr to er-cnl equliztion, wen positionl trnsfer functions re clculted, teir form is uncnged 212 Equliztion wit te Addition of Heigt Cues eserc by Bluert [13] s sown tt by introducing specific frequency-dependent crcteristics into te HFs senstion of eigt is cievble However, te J Audio Eng Soc, ol 50, No 11, 2002 November 897

5 HAKSFOD PAPES () (b) (c) Fig 2 Normlized time-domin left rigt HFs t 30 op left er; bottom rigt er () Mesured Observe reltive time displcement reveling ID nd lck of prering in nturl responses (b) Metod 1 equlized Observe excessive prering tt blurs commencement of te two HIs (c) Metod 2 equlized HIs exibit mirror imges except for initil impulse (d) Metod 3 equlized Prering reduced nd initil ID of HIs mintined 898 J Audio Eng Soc, ol 50, No 11, 2002 November

6 PAPES SCALABLE MULICHANNEL CODING IH HF ENHANCEMEN (d) Fig 2 Continued () (b) Fig 3 Mgnitude HF pir t 30 op left er; bottom rigt er () Mesured (b) Equlized esults re identicl for metods 1, 2, nd 3 J Audio Eng Soc, ol 50, No 11, 2002 November 899

7 HAKSFOD PAPES () (b) Fig 4 Mgnitude HF sum nd difference spectr t 30 Sum dimond line; difference continuous line () Unequlized (b) Equlized, pplicble to metods 1, 2, nd 3 (Note constnt-level sum spectrum following equliztion) Fig 5 Equlized HF pir t 30, pplicble to metod 4 only op left er; bottom rigt er 900 J Audio Eng Soc, ol 50, No 11, 2002 November

8 PAPES question rises s to weter tis modifiction is comptible wit te equliztion strtegies presented in Section 211 For exmple, te following questions need to be considered: Is it sufficient to mesure te HF coordintes only t te required loction bove te lterl plne nd ten pply equliztion, nd will ten sufficient informtion remin buried in te interer difference signl wit unique crcteriztion to discriminte ginst lterl imges wit equivlent interurl time differences? Does te bsolute mplitude response vrition, rter tn just te difference response vrition inerent in te HFs, represent mjor fctor in producing eigt cues? Are tere secondry fctors, suc s ground reflections, wic introduce dditionl cues to id eigt locliztion? Effectively tis would require t lest two interfering sets of HFs to be summed A full investigtion of tese points relting to eigt is beyond te scope of te present study However, if te ground reflection model were responsible, ten te equliztion metods could be pplied individully to te direct source nd to te ground reflection, wit te results combined by tking te pt difference into ccount 22 Simplified HF Models In pplictions were pntom imges re positioned close to te loclity of te loudspekers, it my be sufficient to use simple form of HF is is prticulrly pplicble wit multiloudspeker rrys were vector component lredy forms strong locliztion clue Fig 6 sows source imge t ngle θ defined by x nd wit respect to umn ed of dimeter d meters 221 Simple HF Model 1 e model ignores ed sdowing nd ssumes tt only te interurl time difference is significnt Hence for SCALABLE MULICHANNEL CODING IH HF ENHANCEMEN source t ngle θ from te forwrd position, te respective HF coordintes re pproximtely x d exp * j2πf = sin^θg4 2c (7) d exp * j2πf = sin^θg4 (7b) 2c were te velocity of sound is c m/s nd s is te time dely from te source to te center of te ed In tis model te dvntge of equliztion metod 3 is evident s no equliztion need be pplied 222 Simple HF Model 2 In tis second model bot te front nd te bck wves re considered, were te bck wve results from ed defrction In tis representtion wve incident on er A produces, by ed defrction, secondry signl t er B e ed diffrction trnsfer function from ers A to B, DH A B (r, θ, φ), is function of te direction of te incident wve defined by te spericl coordintes {r, θ, φ} A similr function linking ers B to A is defined, DH B A (r, θ, φ) Hence for te source HF coordintes { x, }, x d exp * j2πf = sin^θg4 2c d DHB - A_ r, θφ, i exp * j2πf = sin^θg4 2c (8) d exp * j2πf = sin^θg4 2c d DHA- B_ r, θφ, i exp * j2πf = sin^θg4 2c (8b) In simple model te diffrction trnsfer functions could be represented s ttenution DH k wit time dely of pproximtely te interurl time dely A B, DH _ r, θφ, i DH _ r, θφ, i A- B B - A DH exp_ j2π f k A- B i (9) 3 MULILOUDSPEAKE AAYS IN O- CHANNEL SEEO Fig 6 Sound source nd simplified ed model used to derive pproximte HFs is section introduces vrints to two-cnnel, twoloudspeker trnsurl processing to demonstrte ow two-cnnel signl formt cn be mpped into n feeds to drive multiloudspeker rry [14] It is ssumed tt more tn two loudspekers re driven simultneously by signls derived from single-point sound source, wile forml metods sow tt te correct er signls cn be retined Besides supporting stnd-lone pplictions, tese trnsformtions re relevnt in te development of multicnnel trnsurl stereo, s described in Section 5 J Audio Eng Soc, ol 50, No 11, 2002 November 901

9 HAKSFOD e outputs of n n-rry of loudspekers combine by cousticl superposition t te entrnce to ec er cnl e principl condition for ccurte sound locliztion is tt tese signls mtc te signls tt would ve been generted by rel sound source, bot in te sttic cse nd in te cse for smll ed rottions Also, by using severl loudspekers plced to surround te listener, sound-field direction cn mke te system more tolernt to ed motion Consequently cnges in er signls wit ed motion mtc more closely tose of rel imge A sttic sound source, wtever its size nd pysicl loction, produces two er signls tt fully define te event, provided te reltive ed position to source is fixed In prctice it is possible to generte te correct er signl from two or more noncoincident loudspekers tt cn tke ny rbitrry position round te ed However, if te position of te ed moves, ten cnge in te er signls results, wic no longer mtc te pntom imge correctly, nd locliztion error is perceived In system of wvefront reconstruction tis distortion is minimized, ltoug te penlty is lrge number of loudspekers nd cnnels However, if limited number n of loudspekers is used (for exmple, n 12), ten ltoug imge position distortion still occurs, te effect is reduced s tere is robust directionl component Also, wen pntom imge coincides wit loudspeker position, te positionl distortion s function of ed position tends to zero, ltoug it is debtble weter tis is desirble sitution s imges from oter loctions re represented differently in terms of teir rdited cones of sound PAPES 31 ree-loudspeker rnsurl Processing o illustrte ow to ccommodte more tn two loudspekers in n rry wile retining te requirements for precise HF formultion, consider tree-loudspeker rry s illustrted in Fig 7 In tis system mono source signl is filtered by te positionl trnsfer functions G nd G L to form L nd,wic in turn form inputs to te mtrix [] By wy of exmple rifield 8 mtrix (fter Gerzon [15]) is selected, wic is defined s S11 12 S S S S (10) S S31 32 S By pplying te coefficients defined in mtrix [], te tree loudspeker signls L, C, nd (left, center, rigt) cn be derived However, to reproduce optimum locliztion, te system requires tt te er signls produced by te tree loudspekers mtc te er signls tt would be produced by te rel source 311 Anlysis e positionl trnsfer function mtrix [G] converts te mono signl to L nd s SL G S S G (11) S L Using mtrix [], te loudspeker feeds L, C, nd re ten derived from [G], S L S11 12 G SC S S L S S G (12) S S S31 32 However, reclling te HFs s defined in Fig 7, were x nd re te HF coordintes of source imge, ten L S S x S l c r SC (13) S Slb cb rbs S were, substituting for L, C, nd, S11 12 Sx Sl c rs SGL S Slb cb S rb (14) SG S31 32 e positionl trnsfer functions G nd G L ten follow by mtrix inversion, SG SG L Z ] S [ S ] \ l lb c cb r rb S S S S _ b ` b 1 S S x (15) Fig 7 rnsurl processing using symmetricl tree-loudspeker rry 8 egistered trdemrk describing two-cnnel to treeloudspeker mpping proposed by Gerzon [15] 902 J Audio Eng Soc, ol 50, No 11, 2002 November

10 PAPES from wic te loudspeker feeds L, C, nd re clculted, Z S L S SC ] S S [ S S ] S \ S S x _ Z b ] S `[ S b ] \ l lb c cb r rb S S S S _ 1 b ` b (16) In prctice L, C, nd re clculted directly using mtrix inversion However, becuse te trnsfer functions cn ve severl tousnd elements, to void lrge-dimension mtrices te solution cn be decomposed s follows Define S l S S c rs S21 22 Slb cb S rb (17) S31 32 giving Sx S11 12 SGL S S21 22SG (18) were, using mtrix inversion, te positionl trnsfer functions re G G L x xr x (19) (19b) wic enble L, C, nd to be clculted Fig 8 sows exmple trnsfer functions linking te system input to te tree loudspeker inputs (L, C, nd ) locted t 45, 0, nd 45, wit source loction t 30 Simultions confirm tt te correct er signls re produced s sown in Fig 9, wile Figs 10 nd 11 present te mgnitudes of te positionl trnsfer functions G L nd G nd teir differentil pse response, respectively 32 ree-loudspeker Mtrix wit Bnd-Limited Center Cnnel is section extends multiloudspeker trnsurl processing by considering cse were te center cnnel is bnd-limited by low-pss filter wit trnsfer function λ( f ) For exmple, λ( f ) could constrin te center cnnel to operte only in te bnd were te er nd brin employ interurl time differences for locliztion Alterntively te center cnnel my be used minly for low-frequency reproduction e inclusion of λ( f ) yields effective HF center cnnel coordintes { c *λ( f ), cb *λ( f )}, were * implies element-by-element vector multipliction Hence, from te equtions derived in Section 31, L, C, nd follow, SCALABLE MULICHANNEL CODING IH HF ENHANCEMEN o illustrte tis system wit bnd-limited center cnnel, Fig 12 sows gin te system input to te loudspeker trnsfer functions for te tree loudspekers locted t 45, 0, nd 45, wit pntom source loction t 30 e low-pss filter λ( f ) in te center cnnel s cutoff frequency of 100 Hz wit n symptotic ttenution slope of 40 db per octve e er signls re formed correctly nd re identicl to tose presented in Fig 9 Fig 8 Loudspeker feed trnsfer functions for rifield mtrix linking input to tree loudspekers locted t 45, 0, nd 45, imge t 30 Fig 9 Input-to-er trnsfer functions for tree-loudspeker system wit rifield mtrix, corresponding to functions sown in Fig 8 SZ S L S S SC ] S S S [ S S S ] S S\ S S x _ Z b ] S `[ S b ] \ l lb c cb * λ _ fi * λ _ fi r rb S S S S S 1 Sλ _ f i S S 1 J Audio Eng Soc, ol 50, No 11, 2002 November _ b ` b 1 (20)

11 HAKSFOD PAPES An ttrction of tis configurtion is tt te center cnnel cn ve limited bndwidt wile offering improvements in bss qulity bot in terms of power ndling nd by improving modl dispersion in te listening room Alterntively, if loss of sptil resolution t low frequency is permitted, ten te center cnnel could function s subwoofer wit n upper response tt extends only into te lower midbnd frequency rnge e left- nd rigt-nd loudspekers would extend to ig frequencies, ltoug wit restricted low-frequency performnce 33 n-loudspeker Arry wit wo-cnnel rnsurl Processing e metod of using more tn two loudspekers cn be generlized to n loudspekers wile retining only two informtion cnnels, were for exmple te loudspekers surround te listener in symmetricl rry e left- nd rigt-nd loudspekers in te rry re fed by one of two informtion signls, nd ec loudspeker s individul weigting defined by coefficient mtrix [] Z _ Z S LS^1 ] S 1 1 b ] SLS^2 S2 ] 2 x ^1 S S S b ] S [ `[ S ] S S b ] Sb ^1 SLS^n S ] b ] \ \ Altoug tis mtrix eqution cnnot be solved in generl s tere re too mny independent vribles, solutions cn be cieved wen te mtrix [] is specified For generl two-cnnel stereoponic reproduction tis system offers little dvntge However, in telepresence nd teleconference environment te coefficient mtrix [] my be trnsmitted for given tlker longside te two informtion cnnels A trnsurl reproduction system cn ten be conceived, were te coefficients re updted dynmiclly to ennce directionl coding is becomes prticulrly ttrctive were tere re number of tlkers, s te Fig 10 Positionl trnsfer functions G L nd G, corresponding functions sown in Fig 8 ^2 ^2 coefficient mtrix could be djusted dynmiclly to ennce locliztion 4 MULICHANNEL PAADIGM EPLOIING PAIISE ANSAUAL SEEO is section reviews sptil udio prdigm tt links multicnnel udio nd trnsurl processing e tecnique ugments te directionl clues inerent in multicnnel stereo reproduction by embedding HF dt suc tt te er signls re mtced more ccurtely to tose produced by te source imge By defult, suc processing includes bot frequency (interurl mplitude response) nd time (interurl time response) informtion nd terefore forms n elegnt metod of virtul imge mnipultion Also, becuse HFs vry wit bot ngulr position nd distnce, sound sources cn be syntesized nd mnipulted in tree-dimensionl spce, togeter wit _ S 1 1 b f ^n S2 2 b S ` f b ^ns b S b b n n n n reflections sptilized using teir HF coordintes, wic furter ennces tis process In multicnnel system tis processing is performed during source coding nd is terefore comptible wit ll DD formts e proposl opertes t six principl levels: Selecting pir of loudspekers wose subtended ngle includes te position of te pntom imge elps reinforce te sound direction nd mtces conventionl mixing prctice for locliztion in multicnnel systems Encoded mplitude differences in signls bove bout 2 khz support locliztion using interurl mplitude differences Encoded time differences in signls below bout 2 khz support locliztion using interurl time differences were n extended bss performnce is desirble e ddition of trnsurl processing bsed on HF dt enbles te construction of er signls tt mtc te originl event nd ids locliztion Closer spcing of loudspekers in multiloudspeker rry reduces sensitivity to te precise form of HF crcteristics, tus mking n verged HF set more pplicble to wide rnge of listeners e effect of moderte ed motion, wic is desirble ttribute for improving locliztion, is supported For reltively smll loudspeker subtended ngles te error in er-signl reconstruction is reduced wen te ed is moved by smll ngle suc tt te vector component reenforces locliztion 904 J Audio Eng Soc, ol 50, No 11, 2002 November 1 (21)

12 PAPES As n exmple, consider circulr rry of n loudspekers, s sown in Fig 13 In tis system pirwise coding (PC) selects te two closest loudspekers suc tt n imge flls witin te subtended ngle t te listening position wo-cnnel HF syntesis is ten used to form te optimum er signls For exmple, if loudspekers r nd r 1 re selected from te n-rry of loudspekers, ten loudspeker feeds LS(r) nd LS(r 1) re computed, SLS S 1 SLS 1 Sb b 1 SGr G S _ r1i 1 S S x (22) Mtrix [G] defines set of positionl trnsfer functions, wic re effectively filters locted between source nd loudspeker feed, were te HF nottion ws defined in Section 1 wit reference to Fig 1 Solving for te positionl trnsfer function mtrix [G], G G r r1 x b 1 1 b b 1 x b b b (23) (23b) SCALABLE MULICHANNEL CODING IH HF ENHANCEMEN If sound source flls between loudspekers r nd r 1, ten r r1 1; oterwise ll remining coefficients in mtrix [] re set to zero Consequently for sound source to circumnvigte te ed, te HF coordintes x nd must cnge dynmiclly, weres s te source moves between loudspeker pirs, te coefficient mtrix is switced to redirect te sound A four-cnnel, four-loudspeker PC sceme is sown in Fig 14 e positionl trnsfer functions {G 1, G 2, G 3, G 4 } re clculted for ec source loction, wic ten filters te source signl to form te loudspeker feeds Becuse trnsurl processing is performed t te encoder, simple reply system is supported Consequently complete comptibility wit conventionl multicnnel udio is retined 5 INCEASED NUMBES OF LOUDSPEAKES An increse in te number of loudspekers cn cieve more even sound distribution, distribute power ndling, nd possibly lower te sensitivity to room coustics is section considers metods by wic te number of loudspekers in n rry cn be incresed In te bsis system, were loudspekers re linked directly to informtion cnnels locted t coordintes comptible wit te encoding HF coordintes, te loudspekers re designted nodl loudspekers An n- is result cn be generlized for n n-rry of loudspekers s Z _ Z S LS^1 ] S 1 1 b ] SLS^2 S2 ] 2 x ^1 S S S b ] S [ `[ S ] S S b ] Sb ^1 SLS^n S ] b ] \ \ ^2 ^2 _ 1 S 1 1 b SG f ^n S2 S 2 b G S ` S f b ^ns b S S b SG 2 b n n n n n 1 (24) Fig 11 Differentil pse response between positionl trnsfer functions G L nd G corresponding to functions sown in Fig 8 Fig 12 rnsfer functions linking input to tree-loudspeker feeds for rifield mtrix, but wit center cnnel bnd-limited to 100 Hz, corresponding to functions sown in Fig 8 J Audio Eng Soc, ol 50, No 11, 2002 November 905

13 HAKSFOD cnnel system terefore s n nodl loudspekers, wic constitute te bsis rry, wit te corresponding drive signls, or primry signls, collectively forming te primry signl set Loudspekers in ddition to te nodl loudspekers re termed secondry loudspekers 51 Compenstion for Inclusion of Secondry Loudspekers e objective is to derive dditionl signls witin te decoder to drive secondry loudspekers locted between te nodl loudspekers However, te er signls must be conserved nd teoreticlly remin identicl to te cse were only te nodl loudspekers re present It is ssumed ere tt ll loudspekers in te rry ve identicl trnsfer functions nd terefore do not ffect te decoder process If tis is not te cse, ten tey require PAPES individul correction A sector of suc n rry is sown in Fig 15 In tis rry nodl loudspekers r nd r 1 ve te respective HF coordintes [ (r), b (r)] nd [ (r 1), b (r 1)] t te listener, weres te single secondry loudspeker p s te HF coordintes [ ( p), b ( p)] e syntesis of te drive signl for secondry loudspeker p employs two weigting functions λ p1 nd λ p2 pplied to te respective primry signls LS r nd LS r1 suc tt LS p, te drive signl to te secondry loudspeker p, is LS p λp1 LSr λp2 LSr 1 (25) However, wen te secondry loudspeker enters te rry, it is necessry to compenste te output from te two djcent nodl loudspekers in order tt te er signls remin uncnged Idelly tis needs to be cieved witout knowledge of te encoding prmeters Oterwise te modified rry cnnot be used universlly in multicnnel udio system A sceme cpble of meeting tis objective is sown in Fig 16, were te compenstion trnsfer functions γ p1 nd γ p2 filter te signl LS p to yield signls tt re dded to LS r nd LS r1 Fig 13 n-loudspeker rry, suitble for trnsurl pirwise stereo Fig 15 Nodl loudspekers wit dditionl secondry loudspeker Fig 14 Four-loudspeker rry wit pirwise trnsurl syntesis Fig 16 Decoder processing to compenste for secondry loudspeker p 906 J Audio Eng Soc, ol 50, No 11, 2002 November

14 PAPES 511 Anlysis Consider initilly te cse wen te secondry loudspeker p is bsent from te rry e left- nd rigt-er signls e nd e b re expressed in terms of te HFs corresponding to te two djcent nodl loudspekers r nd r 1, e LS LS 1 r r1 (26) e LS LS 1 (26b) b r b r1 b en te secondry loudspeker p is introduced into te rry, te modified er signls e nd e b become SCALABLE MULICHANNEL CODING IH HF ENHANCEMEN 521 HFs Derived Using Liner Interpoltion Consider pir of nodl loudspekers witin n n-rry PC system were te proximity of loudspekers r nd r 1 is suc tt te imge source HFs cn be pproximted by liner interpoltion, suc tt β _ m i $ m ^ r _ 1 m i 1 x r r r (30) β _ m i $ m ^ r _ 1 m i 1 (30b) r r b r b were m r is te pnning vrible wit rnge of 1 to 0, wic corresponds to n imge pn from loudspeker r to r 1, nd te function β(m r ) is modertor cosen to cieve constnt subjective loudness wit vritions in m r By substitution, te positionl trnsfer functions G r el LSr LSr1 1 $ LSrλp1 LSr1λp2 γ p1 b $ LS λ LS λ γ ^ r 1 $ LS λ LS λ _ pi r p1 r1 p2 p2 r p1 r1 p2 el LS LS 1 $ LS λ LS λ γ r b r1 b r p1 r1 p2 p1 $ LS λ LS λ γ ^ r 1 $ LS λ LS λ _ pi r p1 r1 p2 p2 b r p1 r1 p2 b b (27) (27b) Forcing e e nd e b e b, ten S S b 1 S γ b 1 Sγ p1 p2 S S b _ pi _ pi from wic te correction functions γ p1 nd γ p2 follow, γ γ p1 p2 S _ pi b^ r 1 b_ pi 1 S r r 1 r r 1 ^ b^ b^ ^ S b_ pi ^ r _ pi b S b 1 b 1 (28) (29) (29b) ese equtions revel tt te correction functions γ p1 nd γ p2 depend only on te HFs corresponding to te loudspeker rry Consequently tey re purely function of te reply system, nd its loudspeker lyout tus cn be computed witin te reply decoder s prt of te instlltion procedure However, te weigting functions γ p1 nd γ p2 cn be selected independently, provided te system is stble 52 eltionsip of System Prmeters to Loudspeker HFs Section 51 presented n nlysis of decoder prmeters were precise HF mesurement dt re known for ec loudspeker position However, tere re some interesting observtions nd simplifictions tt cn be mde, wic re considered in tis section nd G r1 ten simplify to G m β _ m i r r r (31) G _ 1 m iβ _ m i (31b) r1 r r Consequently, for n imge source tt lies on rdil rc between te two nodl loudspekers, simple mplitude pnning yields te optimum pnning lgoritm Effectively, HF coding informtion is derived directly from te loudspeker loctions nd terefore is mtced precisely to te listener is ssumes tt intermedite HFs re derived by liner interpoltion It sould lso be noted tt s te imge source moves wy from te rdil rc contining te loudspeker rry, cnges in HFs occur, mking te positionl trnsfer functions complex Neverteless, even wen more exct HF dt re vilble, tere remins strong desensitiztion to te exct form of te HFs wen te positionl trnsfer functions re clculted becuse of te reltively close proximity of loudspekers in n n-rry Consider next secondry loudspeker p tt is dded to te rry, gin ssuming liner interpoltion model It is ssumed tt te secondry loudspeker is locted midwy long te sme rdil rc s te nodl loudspekers nd tt its HFs ( p) nd b ( p) wit respect to te listener cn be determined by liner interpoltion us for te midpoint loction, _ pi (32) _ pi (32b) b b b From tese dt te compenstion gmm functions γ p1 J Audio Eng Soc, ol 50, No 11, 2002 November 907

15 HAKSFOD nd γ p2 follow, γp1 γp2 05 (33) reveling once more simple form for tis specil cse e modified positionl trnsfer functions GM r GM r1 nd GM p for te respective nodl nd secondry loudspekers r, r 1, nd p re clculted s GM GM GM λ G λ G p p1 r p2 r1 `1 γ λ j G γ λ r p1 p1 r p1 p2 r1 G γ λ G G `1 γ λ r1 p2 p1 r r1 p2 p2 j (34) (34b) (34c) Assuming symmetry, let λ p1 λ p2 05 nd consider te following exmples: 1) G r 1 nd G r1 0, yielding GM r 075, GM r1 025, nd GM p 05 2) G r 05 nd G r1 05, yielding GM r 025, GM r1 025, nd GM p 05 For n imge locted coincident wit te secondry loudspeker tere is 6-dB level difference between secondry nd nodl loudspeker input signls Also signl processing is simple nd uses only rel coefficients in te mtrices 522 Compenstion nd Incorportion of Exct HF Dt It is instructive to compre tree oter options for incorporting secondry loudspeker nd imge HF coordintes In ec cse correction functions λ p1 nd λ p2 re clculted to mtc te selected HFs nd te corresponding trnsfer functions for system input-to-loudspeker inputs evluted for loudspekers r, p, nd r 1 locted t 20, 30, nd 40, respectively, wit n imge t 30 e four cses re s follows Cse 1: Imge Secondry loudspeker esults Cse 2: Imge Secondry loudspeker esults Cse 3: Imge Secondry loudspeker esults Mesured HF dt Mesured HF dt See Fig 17() Mesured HF dt HF derived by liner interpoltion See Fig 17(b) HF derived by liner interpoltion Mesured HF dt See Fig 17(c) (b) PAPES () Fig 17 rnsfer functions linking input to tree-loudspeker feeds t 20, 30, nd 40, imge t 20 () Cse 1 (b) Cse 2 (c) Cse 3 (d) Cse J Audio Eng Soc, ol 50, No 11, 2002 November (c)

16 PAPES Cse 4: Imge Secondry loudspeker esults HF derived by liner interpoltion HF derived by liner interpoltion See Fig 17(d) nd discussion in Section 521 Fig 17(c) sows tt wen te HF coordintes for te imge re derived by liner interpoltion, te center cnnel response is constnt wit frequency, even toug te secondry loudspeker HF coordintes re derived by mesurement Also, wen bot sets of coordintes re derived by liner interpoltion, ll tree responses re constnt, s demonstrted in Section 521 However, for cses 1 nd 2 ll responses vry wit frequency, ltoug cse 2 mintins greter ig-frequency content in te center cnnel response, wic is desirble for n imge locted coincident wit te secondry loudspeker 523 Compenstion for Encoder Decoder Loudspeker Displcement Error e encoder ssumes nominl loudspeker rry loction wen determining te positionl trnsfer functions If te nodl loudspekers re plced in te listening spce in equivlent positions, ten no dditionl processing is required t te decoder However, in circumstnces were te loudspeker loctions re displced, positionl compenstion is required It is importnt tt positionl compenstion be independent of source coding nd cn be pplied t te decoder witout knowledge of te encoding lgoritm A pirwise positionl correction sceme is sown in Fig 18, were te compenstion functions GC 11 (r), GC 12 (r), GC 11 (r 1), nd GC 12 (r 2) re used to derive modified loudspeker feeds e form of tis compenstion is not unique, s correction signls cn be pplied to oter loudspekers in te rry vi pproprite filters However, it is suggested tt te loudspeker SCALABLE MULICHANNEL CODING IH HF ENHANCEMEN selected to process te correction signl be te one closest to te loudspeker tt s been displced Hence by wy of exmple, consider te sceme sown in Fig 18 en te only ctive primry fed is LS r, nd equting te er signls for bot optimum nd displced loudspeker loctions, te positionl compenstion filters re s follows: 1 SGC11 Sl l 1 S SGC12 (35) Slb lb 1 Sb Similrly, wen only LS r1 is ctive, ten SGC SGC Sl 1 1 Slb 1 l lb 1 S 1 Sb 1 (36) 53 Stndrdiztion of HF Grid nd Nodl Loudspeker Loctions e tecniques described in te preceding require knowledge of te HF coordintes for ec nodl loudspeker Estblising stndrdized encoding grid were ec grid point is ssigned nominl HF coordintes nd were nodl loudspekers re ssigned nominl grid loctions cn stisfy tis requirement All encoder nd decoder users ten universlly know tis informtion An exmple grid proposl, s sown in Fig 19, is bsed on 60 subtended ngle for nodl loudspekers wit two dditionl () (d) Fig 17 Continued (b) Fig 18 Compenstion process for nodl loudspeker positionl errors () Optimum loudspeker loction (b) Suboptimum loudspeker loction J Audio Eng Soc, ol 50, No 11, 2002 November 909

17 HAKSFOD secondry loudspekers ree lyers of nodes re suggested t rdii of 15 m, 3 m, nd 6 m It is recognized tt HFs re not unique, being listener specific, but wen multiloudspeker rry is formed using set of HFs tt re sred wit imge syntesis, errors re reduced HF coordintes tt re noncoincident wit te nodl points cn be inferred by interpoltion For exmple, ssume tt n imge is locted t te cylindricl coordintes {r, θ}, were te four nerest nodes re {r 1, θ 1 }, {r 1, θ 2 }, {r 2, θ 1 }, nd {r 2, θ 2 } e interpolted HFs (r, θ) nd b (r, θ) re ten PAPES multicnnel system configurtions However, inevitbly tere is limit to sptil resolution rising from te use of mtrixing only, wic imposes crosstlk between nodl nd secondry loudspeker feeds Some dvntge my be gined by using nonliner decoding wit dynmic prmeteriztion, ltoug for ig-resolution music reproduction liner decoding sould be retined Becuse DD-A is cpble of six cnnels t 24 bit 96 ^ r, θ m 8m _ r, θ i _ 1 m i _ r, θ ib _ 1 m i8m _ r, θ i _ 1 m i _ r, θ ib r θ 1 1 θ 1 2 r θ 2 1 θ 2 2 (37) ^ r, θ m 8m _ r, θ i _ 1 m i _ r, θ ib _ 1 m i8m _ r, θ i _ 1 m i _ r, θ ib (37b) b r θ b 1 1 θ b 1 2 r θ b 2 1 θ b 2 2 were m θ nd m r re te ngulr nd rdil liner interpoltion prmeters defining te imge For imges tt lie eiter witin te inner rdius or beyond te outer rdius, ngulr interpoltion is performed first, followed by n pproprite djustment to te mplitude nd time delys bsed on te rdil distnce from te ed 6 PECEPUALLY BASED CODING EPLOIING EMBEDDED CODE IN PIMAY SIGNALS O ENHANCE SPAIAL ESOLUION e tecniques described in Section 5 cn be extended to system wit ny number of nodl nd secondry loudspekers nd tus cn be mtced to wide vriety of khz, some of te lower bits in te LPCM strem cn be scrificed [12] wile still retining n exemplry dynmic rnge by using stndrd metods of psycocousticlly motivted noise sping nd equliztion [16] e lest significnt bits in te LPCM strems togeter wit rndomiztion function cn ten be used to encode dditionl udio cnnels using perceptul coders suc s AC-3, 9 DS, 10 or MPEG 11 For exmple, 4 bit per smple per LPCM cnnel 9 Proprietry perceptul coding developed by Dolby Lbortories 10 Digitl etre Systems Fig 19 Proposed 18-segment constelltion mp to define stndrd set of HFs 910 J Audio Eng Soc, ol 50, No 11, 2002 November

18 PAPES t 96 khz yields seril bit rte of 384 kbit/s e proposl retins te primry signls in igresolution LPCM nd uses te mtrix metods described in Section 5 to estimte te secondry loudspeker feeds Sptilly relted difference signls re ten clculted from te discrete secondry loudspeker signls vilble t te encoder nd te mtrix-derived signls Also, becuse of close sptil clustering of te dditionl cnnels tere is ig degree of intercnnel correltion wit bot te primry nd te secondry loudspeker signls, fctor tt bodes well for ccurte perceptul coding Close clustering lso implies tt perceptul coding errors re not widely dispersed in spce, yielding n improved msking performnce Given tt DD-A lredy supports six LPCM cnnels, it is suggested tt n extr two encoded signls per primry signl is relistic compromise, yielding totl of 18 cnnels, s proposed in te stndrdized HF constelltion illustrted in Fig 18 In te grnd pln tere would be n perceptul coders in opertion, one per nodl loudspeker feed In suc sceme furter gins re possible by integrting dynmic bit lloction cross ll coders s well s using perceptul model designed specificlly for multicnnel stereo encoding In difficult encoding situtions dynmic sptil blending cn be used to reduce te difference signls prior to perceptul encryption Fig 20 sows te bsic encoder rcitecture were te error signls D 1 nd D 2 re indicted In Fig 21 decoder is sown were identicl estimtes re mde of te secondry loudspeker input signls, but wit te ddition of te difference signls to yield discrete loudspeker feeds Of course, if te embedded perceptully coded difference signls re not used, estimtes cn still be mde for te secondry loudspekers s sown in Fig 16 Alterntively, for bsic sceme n rry of nodl loudspekers only cn be used 11 Perceptul bsed udio coding proposed by te Motion Picture Expert Group 12 egistered trdemrk of Compny nme, New rnsducers plc, UK SCALABLE MULICHANNEL CODING IH HF ENHANCEMEN 7 CONCLUSIONS is pper s presented metod for multicnnel udio tt is fully comptible wit DD (DD-A nd SACD) multicnnel formts, wic ve te cpbility of six ig-resolution signls e key to tis tecnology is te exploittion of sptil coding using HF dt to ennce te positionl representtion of sound sources As suc it forms link between two-cnnel trnsurl tecniques nd conventionl multicnnel udio using mny loudspekers is tecnique s lredy been demonstrted in telepresence nd teleconferencing pplictions [9], [10] to be effective in representing sptil udio However, te metods described ere sow ow prticulr loudspeker rry cn be configured were issues of positionl clibrtion were discussed for loudspekers displced from tose loctions ssumed during coding e metod is sclble nd fully bckwrd comptible In simple system tere is no dditionl processing t te decoder were, for exmple, te outputs of DD plyer re routed directly to n rry of loudspekers However, if dditionl loudspekers re used, s migt be envisged wit tiled wlls of flt-pnel N 12 loudspekers (see, for exmple, Fig 22), ten forml metods exist, enbling te correct er signls t te listening position to be mintined Also for DD-A, metod ws suggested were perceptully coded informtion is embedded witin te LPCM code to enble discrete loudspeker signls to be derived It ws proposed tt n upper limit of 18 cnnels sould be ccommodted, ltoug full comptibility wit systems down to te bsic rry is mintined An interesting observtion for systems using lrge number of closely spced loudspekers is tt imge positioning on te rc of te rry cn use simple liner mplitude pnning pplied between pirs of djcent loud- Fig 20 Hig-level processor rcitecture for encoding discrete secondry loudspeker feeds Fig 21 Hig-level decoder rcitecture to derive discrete secondry loudspeker feeds J Audio Eng Soc, ol 50, No 11, 2002 November 911

19 HAKSFOD spekers is pproximtion ssumes tt te imge HF coordintes cn be estimted to sufficient ccurcy using liner interpoltion between djcent loudspeker HF coordintes is expedience effectively embeds HF dt mtced exctly to te listener simply becuse of te pysicl loction of te loudspekers However, s te loudspeker spcing increses, tis pproximtion fils, requiring ten te use of more ccurte HF imge coordintes togeter wit trnsurl PC, s described is is prticulrly importnt were n imge is locted wy from te rc of te loudspeker rry nd were reflections re to be rendered to crft more ccurte virtul coustic is work is lso trgeted t new communiction formts for virtul relity, telepresence, nd video conferencing [17], [18], were future reserc sould investigte its ppliction Suc scemes re not constrined by te norml prdigms of multicnnel stereoponic reproduction, nor is comptibility necessrily sougt e pproc is to form n optimum metodology for constructing pntom imges nd to consider coding prdigms pproprite for communiction For exmple, one possible communiction formt ssigns discrete cnnel to ec ptnom imge e cnnel ten conveys te uditory signls togeter wit te sptil coordintes updted t rte comptible wit motion trcking of te sound source At te receiver, processor crries downloded progrm wit knowledge of te positionl dt nd source coustics from wic te required reflections nd reverbertion re computed ese dt would ten be formtted to mtc te selected loudspeker rry Suc sceme s gret flexibility nd cn llow mny mono sources to contribute to te finl soundscpe In conclusion, te tecniques presented describe Fig 22 Multicnnel configurtion using rry of wll-mounted flt-pnel diffuse loudspekers PAPES mens by wic sptil resolution nd imge coding performnce cn trnscend te six-cnnel limittion of te current DD formts, yet witout requiring dditionl storge cpcity Also, by bsing signl processing on perceptul model of ering, it is reveled ow sound imges cn be rendered nd, in prticulr, ow interurl mplitude differences nd interurl time differences cn be ccommodted witout seeking trdeoffs between time nd mplitude clues Essentilly te work s presented sclble nd reverse comptible solution to multicnnel udio tt is prticulrly well mtced to n LPCM formt on DD-A 8 EFEENCES [1] HFN/, Digitl Frontiers, vol 40, pp 58 59, 106 (1995 Feb) [2] M O J Hwksford, Hig-Definition Digitl Audio in 3-Dimensionl Sound eproduction, presented t te 103rd Convention of te Audio Engineering Society, J Audio Eng Soc (Abstrcts), vol 45, p 1016 (1997 Nov), preprint 4560 [3] A J Berkout, D de ries, nd P ogel, Acoustic Control by vefield Syntesis, J Acoust Soc Am, vol 93, pp (1993) [4] D Begult, 3-D Sound for irtul elity nd Multimedi (AP Professionl, 1994) [5] M A Gerzon, Peripony: it-heigt Sound eproduction, J Audio Eng Soc, vol 21, pp 2 10 (1973 Jn/Feb) [6] M A Gerzon, Ambisonics in Multicnnel Brodcsting nd ideo, J Audio Eng Soc, vol 33, pp (1985 Nov) [7] M A Gerzon, Hierrcicl rnsmission Systems for Multispeker Stereo, J Audio Eng Soc, vol 40, pp (1992 Sept) [8] K C K Foo nd M O J Hwksford, HF Sensitivity Anlysis for ree-dimensionl Sptil Audio Using te Pirwise Loudspeker Assocition Prdigm, presented t te 103rd Convention of te Audio Engineering Society, J Audio Eng Soc (Abstrcts), vol 45, p 1018 (1997 Nov), preprint 4572 [9] K C K Foo, M O J Hwksford, nd M P Hollier, ree-dimensionl Sound Locliztion wit Multiple Loudspekers Using Pirwise Assocition Prdigm nd Embedded HFs, presented t te 104t Convention of te Audio Engineering Society, J Audio Eng Soc (Abstrcts), vol 46, p 572 (1998 June), preprint 4745 [10] K C K Foo, M O J Hwksford, nd M P Hollier, Pirwise Loudspeker Prdigms for Multicnnel Audio in Home etre nd irtul elity, presented t te 105t Convention of te Audio Engineering Society, J Audio Eng Soc (Abstrcts), vol 46, p 1035 (1998 Nov), preprint 4796 [11] M Gerzon, Prcticl Peripony: e eproduction of Full-Spere Sound, presented t te 65t Convention of te Audio Engineering Society, J Audio Eng Soc (Abstrcts), vol 28, p 364 (1980 My), preprint 1571 [12] M A Gerzon nd P G Crven, A Hig-te 912 J Audio Eng Soc, ol 50, No 11, 2002 November

20 PAPES Buried Dt Cnnel for Audio CD, J Audio Eng Soc, vol 43, pp 3 22 (1995 Jn/Feb) [13] J Bluert, Sptil Hering, rev ed (MI Press, Cmbridge, MA, 1997) [14] J Buck nd D H Cooper, Generlized rnsurl Stereo nd Applictions, J Audio Eng Soc, vol 44, pp (1996 Sept) [15] M A Gerzon, Optimum eproduction Mtrices for Multispeker Stereo, J Audio Eng Soc, vol 40, pp (1992 July/Aug) [16] J Sturt nd J ilson, Dynmic nge Enncement Using Noise-Sped Diter Applied to SCALABLE MULICHANNEL CODING IH HF ENHANCEMEN Signls wit nd witout Preempsis, presented t te 96t Convention of te Audio Engineering Society, J Audio Eng Soc (Abstrcts), vol 42, p 400 (1994 My), preprint 3871 [17] elepresence eme, B ecnol J, vol 15 (1997 Oct) [18] D M Burrston, M P Hollier, nd M O J Hwksford, Limittions of Dynmiclly Controlling te Listening Position in 3-D Ambisonic Environment, presented t te 102nd Convention of te Audio Engineering Society, J Audio Eng Soc (Abstrcts), vol 45, p 413 (1997 My), preprint 4460 HE AUHO Mlcolm Hwksford received BSc degree wit First Clss Honors in 1968 nd PD degree in 1972, bot from te University of Aston in Birmingm, UK His PD reserc progrm ws sponsored by BBC eserc Scolrsip nd investigted delt modultion nd sigm delt modultion (SDM, now known s bitstrem coding) for color television nd produced digitl time-compression/time-multiplex tecnique for combining luminnce nd crominnce signls, forerunner of te MAC/DMAC video system Dr Hwksford is director of te Centre for Audio eserc nd Engineering nd professor in te Deprtment of Electronic Systems Engineering t Essex University, were is reserc nd tecing interests include udio engineering, electronic circuit design, nd signl processing His reserc encompsses bot nlog nd digitl systems wit strong empsis on udio systems including loudspeker tecnology Since 1982, reserc into digitl crossover networks nd equliztion for loudspekers s resulted in n dvnced digitl nd ctive loudspeker system being designed t Essex University A first in 1986 ws for prototype system to be demonstrted t te Cnon eserc Centre in okyo, work sponsored by reserc contrct from Cnon Muc of tis work s ppered in te JAES,togeter wit substntil number of contributions t AES conventions His reserc s lso encompssed oversmpling nd noise-sping tecniques pplied to nlog-to-digitl nd digitl-to-nlog conversion wit specil empsis on SDM Oter reserc s included te lineriztion of PM encoders, diffuse loudspeker tecnology, nd tree-dimensionl sptil udio nd telepresence including multicnnel sound reproduction Dr Hwksford is recipient of te 1997/1998 AES Publictions Awrd for is pper, Digitl Signl Processing ools for Loudspeker Evlution nd Discrete-ime Crossover He is crtered engineer s well s fellow of te AES, IEE, nd IOA He is currently cir of te AES ecnicl Committee on Hig- esolution Audio nd is founder member of te Acoustic enissnce for Audio (AA) He is lso tecnicl consultnt for N, UK nd LFD Audio, UK J Audio Eng Soc, ol 50, No 11, 2002 November 913