T phone networks, brought about by the gradual
`
`he rapid increase i n digital connectivity of tele-
`
`removal of most analog links, suggests a new look at
`enhancing the quality of audio transmitted over the
`telephone network. Pulse-code modulation (PCM) with
`64 kbit/s p-law or A-law (G.711) arose in response to
`the need for multiple analog-digital-analog conversions of
`standard 300-3,400-Hz audio signals. Such signals are
`considered here to be narrowband audio. Modern speech
`coding techniques permit reduction of the transmitted bit
`rate, while preserving audio quality, as in C C I T T
`(International Telegraph and Telephone Consultative
`Committee) Recommendation G.721) where the cus-
`tomary 300-3,400-Hz-wide telephone signal is encoded
`at 32 kbit/s [ l ] . Alternatively, one can provide improved
`a u d i o quality a n d maintain the transmission rate at 63
`kbitls. Such improvements are most important for audio
`o r audio-visual conferencing applications where one
`would like to approach the quality of face-to-face
`communication. C C I T T Study G r o u p XVIII recognized
`the need for a new international coding standard on
`high-quality audio to allow interconnection of diverse
`switching, transmission, a n d terminal equipment and,
`thus, organiLed a n Expert G r o u p in 1983 to recommend
`a n appropriate coding technique. H u m m e l [2] provides
`a good introduction to the working methods of the
`C C I T T . T h e coding method described in this paper
`constitutes the group's recommendation, which was
`approved by the C C I T T through a n accelerated pro-
`cedure in 1986. T h e algorithm represents the results of a
`joint effort of contributors from around the world,* a n d
`is best described i n a series of papers presented a t
`Glohecom '86 [3]. T h i s paper is meant to be a tutorial
`discussion, responsibility for which lies completely with
`the a u t h o r . Bit-level particulars of
`the algorithm,
`a l t h o u g h important for correct implementation of the
`standard, are not discussed i n detail. For more complete
`information, the reader should refer to the forthcoming
`C C I T T document.
`
`Introduction
`
`G,722, A New CCITT
`Cod i ng Sta nda rd for
`Digital Transmission
`of Wideband
`Audio Signals
`Paul Mermelstein
`
`Requirements
`
`T h e main objective for the new standard is to allow
`speech transmission at 64 kbit/s with quality as high as
`poissible and significantly better than that provided by 8-
`bits/sample, 8-kHz sampled PCM coding. If the signal is
`sampled a t 16 kHz, or twice the PCM rate, the spectrum
`of the signal to be encoded can be extended to about 7
`kHz (3-dB point), a n d this results in a major improve-
`
`*Acomplete list of the participants in the Expert Groupcan
`be found in Appendix 1 t o the Report CCITT, COM XVIII-R
`17-E, April 1986. Participating organizations included B N K ,
`Canada; CNET, France; FTZ, Federal Republic of Germany;
`CSELT and SIP, Italy; N T T , Fujitsu and KDD, Japan; P ' I I ' ,
`Switzerland; B T , United Kingdom; Bellcore and Comsat,
`lrnited States. Technical contributions of all participant5 are
`recognized and acknowledged without specific credit on
`individual items.
`
`January 1988-Vol. 26, No. 1
`IEEE Cornrnunlcations Magazine
`
`8
`
`0163-6804/88/0001-0008$01.00 0 1988 IEEE
`
`IPR2016-00704
`SAINT LAWRENCE COMMUNICATIONS LLC
`IPR2017-01075
`Exhibit 2001
`Saint Lawrence Communications
`Exhibit 2001
`
`

`

`conlerence bridges is best carried out with a uniformly quan-
`tized representation of the digital signal. To allow for multi-
`ple bridges in one connection, provision is made for a small
`number (up to three) of digital encoding/decoding sequences.
`Furthermore, both narrowband and wideband audio signals
`may arrive at audio bridges and bridge output should also
`be available in narrowband or wideband form.
`
`I
`
`Narrowband
`
`Wideband I
`
`9
`
`January 1988-Vol. 26, No. 1
`IEEE Communications Magazine
`
`

`

`Audio
`
`audio part
`
`Audio
`signal
`
`Transmit
`audiopart
`
`14 bas
`16 kHz
`
`I
`
`I
`
`I
`
`I
`
`Auxiliary data
`channel input,
`0, 8 or 16 kbit/s
`
`64 kbit/s
`
`I
`
`I
`I
`
`1
`DMUX 4-
`
`Data extraction
`device
`(Determines mode)
`
`input
`
`Auxiliary data
`
`Receive 1"
`
`' H
`Higher sub-band
`ADPCM decoder 16 kbi"sb
`
`-
`
`quadrature
`mirror
`
`t
`
`Lower sub-band
`
`ILr
`
`ADPCM decoder 48 kbit/sb
`(3 variants)
`
`t
`
`allocating the bits across the different bands, the error
`variance in the reconstructed signal can he shaped n i t h
`frequency. With the audio signal subdivided into two
`I-kHz-wide hands, a high signal-to-noise ratio in the
`lower band hecomes perceptually more important than
`in the higher band. An advantage ot ;I design that uses
`tlvo equally wide subhands is that each compotient is
`suhsampletl to 8 kHz and the total transmission rate may
`he reduced in 8-kbit's steps by reducing the number of
`bits assigned to samples of one or the other hand. While
`the hit rate may also he reduced by reducing the sampling
`rate, those processes generally arc more complex to
`implement. These considerations led to design and
`e\xluation of tivo alternative subband ADPCM systems,
`o n e using 5 and 3 hits'sample for the lolv- and high-band
`components, respectively, the other, 6 and 2 bits/sample.
`T h e G.721 ADPCM design employs a n adapti1.r
`predictor with two poles and six zei-os. A fixed predictor
`design w i s also tested for the \videhand coder, h i i t i t led
`to a generally lower speech quality. A time-wrying
`adaptive allocation of bits to the t i v o su1)hands according
`to the short-time signal characteristics w a s also tried. For
`\,oiced sounds carrying significant low-frequency energy,
`one can assign additional bits to the low h n t l ; for
`fricative sounds, the atlditional hit m a y be assigned more
`advantageously to the high band. IIowever, for the two-
`band, 4-kHz-wide subband dcsign, thr. advantage of a n
`adaptive bit assignment is only apparent at the .4 low-
`2 high hits samplr. assignment a n d is found too small to
`\va rra n t the add i t i on a 1 r o m p 1 exi t y .
`Overall block diagrams for thr \videhand encoder and
`decoder ;ire shoivn in Fig. 2. These blocks are discussed in
`greater detail in the following sections.
`
`Subband Filtering
`
`T h e nominal 3-dB hand of the codcc w a s chosen as
`50-7,000 Hz. Two sets of identical quadrature mirror
`
`January 1988-Vol. 26, No. 1
`IEEE Communications Magazine
`
`10
`
`filters ( Q M F ) are used to di\.itle the ~videhantl sigii;rl
`sampled ;it a 16-kHz rate into t'ivo 8 - k H 7 s m i p l ~ d
`components to tic transmitted, ;I low hand and ;I high
`hand, and reconstruct the ivitlehand signal from its
`rrceived low- and h i g h - h m d componmts. QMF liltet-s
`are finite-impulse response, impose ;I fixed tlcla)~ \vi thollr
`phase distort ion, ;I nd c t i s lire t h a t :i 1 i x i n g prod I I c t s
`resulting from subsampling the input signal at the
`transmitter are canceled at
`the recei\w. H o \ v e \ e r ,
`quantization noise components introdric-r.d in coding the
`l o w - and high-band signals ma)- not be elirnitiatctl
`complctely by thr. receiver QMF fi1tr.r. Bccausc the 1c1~c.l o f
`the high-band component of the signal may be :is m u c h
`a s 40 dR 1o1vr.r than the low-band componenr, aliasing
`noise introduced into the high-hand frcqiivnc ies d u e t o
`coding the low-hand signal might h e i n a d e q u a t c l ~ ~
`masked hy the high-band signal componcwt. ' I o achicw
`
`I
`
`i
`
`0
`
`0.1
`
`0.2
`0.3
`Normalized Frequency (Hz)
`
`0.4
`
`0.5
`
`

`

`:I stop-hand rejection of 60 d B , we e m p l o y ;I 24-tap filtei-
`design, introducing a total signal delay of only 3 tns
`(sec Fig. 3 ) . T h e resulting signal distortion is helow 1 d B
`o \ w the 100-6,"M-Hz band.
`T h e numerical precision with which the partial s u m s
`in the QMF filters ;ire accumulated has a n important
`hearing on the ac~ciiracy of the loiv- antl high-band signal
`components grnerated. T h e overall goal for the \videband
`signal representa tion (after analog-to-digital conversion
`a t the input and Iiefore digital-to-analog convrrsion at
`the o u t p u t ) is a precision of 14 tiits. To this end, the
`internal coding computations are performed with 16
`hits. For the suhhand signals to have 16 significant hits,
`the partial sum computations were found t o require 21
`pi-ecision of 24 bits. Since the widehand sign ;I 1 I S ' acciirate
`to 14 hits, the s u m and difference signals to lvhich the
`QMF filter coefficients are applied arc only precise to 13
`bits. To prex'ent the intt-otluction of noise due t o
`differently specified analysis and synthesis filters, the
`QMF filter coefficients are also represented with 13 bits.
`
`ADPCM Coders
`
`T w o A D P C M coders are required, one for the low-
`hand signal and one for the high-hand signal. Thecoders
`employ identical adaptation strategies to modify the
`
`qiiantirers and predictors based on the previously
`ohser\wl characteristics of the input signal. T h e low- antl
`high-hand coders are very similar, except for small
`differences due t o the need to vary the number of hits
`output h y the loiv-hand coder and the fact that the high-
`hand quantizer output is always 2 hits/sample.
`'I'hc adaptive predictor design is bonowed directly
`frotn that investigated in detail in developing the G.721
`standard. T h e two-pole, six-zero design combines good
`prediction gain for speech with relatively simple stability
`control. Robust adaptation is assured by leaky integrators
`allowing the effects of transmission errors to dissipate
`rapidly [8]. Transmission errors may introduce d i f -
`ferences hetlveen the predictor memories at the trans-
`mitter and rvceivcr. Adapting the predictors and q u a n -
`tizers using the residual signal alone and not the
`reconstrurted signal eiisures that the predictors a t the
`transmi ttei- and receiver recover tracking rapidly f o i - all
`signals [9]. T h e adaptive qiiantirer design is also
`horroived directly from the G.721 standard since the 1 0 ~ ' -
`hand signal cotnponen t resembles the narrowhand
`speech signal in most of i t s properties. G.721 employs ;I
`dual-mode quantizer, a locked or s l o w ~ l y adapting mode
`for voicehand data signals, and a n unlocked or rapidly
`adapting mode for speech signals. Since the G.722
`standard was not required to encode voice-hand data
`
`lndkallm -L
`
`M o d s
`
`Indlcalbn
`
`Y
`
`U
`I
`I
`I
`
`P
`I
`
`1 . r
`
`D
`
`m
`U
`I
`t
`I
`P
`I
`
`I . I
`
`I
`
`I In -
`
`.
`
`Fig. 4. Detailed Block Diagram of the Subband-ADPCM
`Encoder and Decoder.
`
`DECODER
`
`January 1988-Vol. 26, No. 1
`IEEE Communications Magazine
`
`

`

`signals, o n l y a single rapidly adapting mode had to be
`implemented. T h e d y n a m i c range of the lo\\-band signal
`quantizer was set to he the same as for G.721, namely 54
`dB. A higher dynamic range is a l l o ~ v e d for the higIi-l)atid
`quantizer, 66 dB, mostly to accommodate music signals.
`Robust adaptation is emplo)-ed also for the quantiret,
`scale factor to combat the effects of transmission crrot~s.
`Embedded quantirers alloiv for possible stripping of
`the less significant hits from the quantized signal d u r i n g
`transmission b y not making use of those bits in the
`quantizer adaptation process [ I O ] . T h e l o w h a n d q u a n -
`tizer anticipates that the one o r two least significant hits
`m a y he stripped from the transmitted code ~ ’ o r d and
`adapts the quantizer and predictor using only the four
`most significant bits. T h i s results in 4- and 5-hit
`q u a n t i ~ e r s that are slightly suboptimal in quantiration
`noise-to-signal ratio compared to the c ~ u ~ i n t i r e r s that
`may be designed without this cons tr;t i t i t . T h e cmhedding
`property requires that the 4- m d 5-hit quantization
`hoiindaries coincide with a subset of those employed for
`
`t h c 6-hi t q 1i;t ri t i rer. Ma t i y t r;t 11s mission systems rcq ui t-e 21
`minimal number of zero-one alternatives to maintain
`s) tic.hroniz~ttioti. ?To prevrnt the all-zero code from
`appeiiritig even in the 4-hit data representation, only 15
`quantirer levels are used in that mode; this also restricts
`t h y higher niodcs to 30 m d 60 levels. Esperinicntal
`e\xluations ha\.c shoivti only a fraction of a decibel is lost
`in quantizing speech signals with the embedded q u a n -
`tirer comp:ired to a n unconstrained quantirer design. In
`tcrms of subjective performance, the embedded design
`was not found to be significantly different from the
`noneinbedded design, e t m aftei- four transcodings.
`A significant systems a d \ m t a g e resulting from the
`embedded coder design is that the en( oder i n a y opcrate
`without regard to the momentary data transmission
`rcquirements. T h e speech c o d i n g a n d data multiplexing
`operations are separated logically a n d possihly even
`physically. ‘Thus, data may he introduced ;it a point
`downstream in the transmission path r e m o \ d froin the
`encoding terminal. T h e receiver o r decoding terminal
`
`I
`
`d . ouantiued
`4- difference
`
`Predictcn
`mrnpulations
`
`January 1988-Vol. 26, No. 1
`IEEE Communications Magazine
`
`12
`
`

`

`the amount of data
`must, of course, be aware of
`introduced in place of speech information so that it may
`interpret the respective bits appropriately.
`A more detailed diagram of the subband-ADPCM
`encoder and decoder is given in Fig. 4. As illustrated,
`identical encoding by the transmitter and receiver in
`each of the three modes of low-band quantization is ensured
`by stripping the two least significant bits in the feed-
`back paths of the predictor and quantizer. The decoder
`interprets
`the received data in accordance with the
`current mode indication. T h e 6-, 5-, or 4-bit words are
`converted using 60-, 30-, or 15-level inverse quantizer
`tables to arrive at the appropriate signal estimate.
`T h e two-pole, six-zero adaptive predictor data-flow
`structure is illustrated in greater detail in Fig. 5.
`Alternative implementations are, of course, available;
`the figure shows a flowchart for perhaps the most simple
`configuration.
`
`Subjective Performance
`
`Before selecting the final design, the Expert G r o u p
`carried o u t a series of subjective experiments with
`different speech signals (diverse languages), music,
`differing transmission conditions, and various hardware
`coding devices. Starting with four different algorithms,
`as implemented in hardware, the list was pruned to two,
`a n d , finally, one compromise design. T h e discussion o n
`performance that follows pertains only to the final
`design as embodied in the recommendation.
`T h e subjective measure of audio quality adopted is the
`mean opinion score (MOS) on a five-point scale:
`excellent, good, fair, poor, bad. Since in the anticipated
`applications of audio conferencing and. hands-free
`telephony listening over speakers a n d not handsets is
`most likely, audio was provided over loudspeakers a t a
`level of 70-dB S P L (sound pressure level) at the listener.
`Seven different language tapes were processed by the
`same codec hardware. Listening experiments were
`
`t - - BER =
`
`-- BER-IO"
`0 L
`56
`64
`48
`Transmission Rate (kbit/s)
`
`3 -
`
`P
`
`2 -
`
`G.711 PCM -
`
`15
`
`20
`
`25 30 35 40 45 Ow [dB1
`
`I
`
`Fzg. 7. Average Subjective Quality Ratings for
`Multiplicative Noise Reference Conditions.
`
`conducted in seven laboratories; the results quoted are the
`average results obtained.
`Audio quality is best a t 64 kbitls, drops slightly when
`the transmission rate is reduced to 56 kbit/s, and more
`significantly with a further reduction to 48 khitls (Fig.
`6). However, even at 48 kbit/s, the audio quality is
`significantly better than narrowband PCM. Received
`audio quality is only slightly affected by transmission
`bit error rates up to lom4, but a significant quality drop is
`noted when the error rate reaches
`T h e subjective MOS may show differences in speech
`quality due to speaker and listener effects as well as the
`language used. T o allow different experimental condi-
`tions to be compared more precisely, a set of reference
`conditions of speech mixed with multiplicative white
`noise was evaluated by each g r o u p of listeners evaluating
`the coded speech. In each case, Q w indicates the signal-
`to-noise ratio in decibels. Figure 7 gives MOS scores as a
`function of Q w . T h e hest mean MOS score of 3.3,
`obtained under error-free 64-khitIs transmission, cor-
`responds to a reference condition of Q w = 45 dB. Note
`that the direct uncoded speech is assigned the same MOS
`rating, implying n o measurable quality degradation due
`to one stage of wideband coding. In contrast, narrowband
`PCM (G.712) results i n a Q w of 32 dB. T h u s , the overall
`subjective gain for G.722 coded speech relative to G.712
`coded speech is equivalent to some 13 dB of noise
`reduction. Wideband signals coded at 128 kbitls, 8
`hitslsample at a sampling rate of 16 kHL, are found to
`correspond i n quality to a Q w of 38 dB. T h i s finding
`suggests that roughly 6 dB of quality improvement
`results from expanding the signal bandwidth to the
`range of 50-7,000 Hz, and 7 dB of additional improve-
`ment is obtained by more precise encoding of that signal.
`As the available transmission rate is reduced, the
`quality becomes slightly degraded. At 56 kbitls, we
`observe a Q w of 43 dB, a very minimal degradation. At 48
`kbitls, Q w is 38 dB, which is still significantly better
`than the quality of G.712. Whenever data transmission is
`intermittent, conference participants may not even be
`aware of the audio-quality variations due to dynamic
`mode switching.
`
`Mode Initialization and Mode Switching
`
`Fig. 6. Average Subjective Quality Ratings of t h e
`Final Algorithm.
`
`Although the procedures for mode initialization a n d
`mode switching will be incorporated into a separate
`
`13
`
`January 1988-Vol. 26, No. I
`IEEE Communications Magazine
`
`

`

`recommendation, they are discussed here to indicate holv
`the G.722 coding procedures may he applied in practical
`communication systems.
`To avoid the need for multiple audio terminals o n
`one's desk-a high-quality widehand terminal to com-
`municate with other wideband terminals and a normal
`telephone to reach parties equipped only with narrow-
`hand telephones-most wideband terminals will incor-
`porate a narrowband PCM communication mode. Calls
`can then be set up with the terminal in the narrowband
`mode (Mode 0). As soon as the called party answers, a n
`exchange of flags takes plare. T h e calling terminal
`transmits one of
`two flags denoting its capabilities.
`Terminals may he of Type 1 , which has only ii 64-khit s
`t r;i n sm i ssi o n cii pa hi 1 i t y , narrow ha 11 d or wideha n d, or
`Type 2, which implements a t least 64- a n c l 56-khit s
`~videhantl motles (Modes 1 and 2 ) and possihly even 48
`khit ' s (Mode 3 ) . T h e flag sent identifies the terminal
`type; the calling terminal awaits a similar response flag.
`D u m b terminals having only ;I narro\vband capability
`cannot respond to the received flag, which leads the
`c;illing terminal t o conclude, after a suitable time-out,
`that i t must remain in the narrowband mode. Intelligent
`term i na 1 s ;ic knoivlcdge t lie received flag by t ra 11 smi t t i rig
`the appropriate response flag and adopt ;I widehand
`inotle (Mode 1 ) . On receipt of
`that flag, the calling
`terminal assumes the same wideband mode. If
`the
`exchange of flags takes place on the least significant bit
`or hit 8 o f every word, it introduces only ;I slight a m o u n t
`of noise into the audio path. T h u s , narrowband
`communication is availa tile even during the flag ex-
`change. T h i s plan allows widehand terminals to he
`introduced gradually into the telephone network, retain-
`i n g connectivity and gradurllly enhancing quality with
`increased penetration of ividehand terminals.
`Many North American network connections optionally
`modify the least significant bit of PCM words to transmit
`signaling information. To allow the use of widehand
`terminals in such situations, the initial widehand mode
`could he Mode 2. It has heen suggested that both hits 7
`and 8 he employed here for reliable flag transmission.
`Corruption of the flagon bit 8 would then indicate to the
`terminals that their transmission channel is limited to 56
`khi t/s.
`
`Data-Speech Multiplexing
`
`In many conference situations, i t is desirable to
`transmit conference-related data, such as speaker-iden-
`tification information 01- document facsimile, on the
`established connection without interrupting the speech
`communication path. T h e fact that high-quality audio
`transmission can he maintained down to 48 kbit 's with
`hut minor degradation alloivs u p to 16 kbit s of data
`transmission.
`Type 2 terminals may switch from Mode 0 (64-khit s
`speech, no data) to Mode 1 (56-khit s speech, 8-khit! s
`ser\,icr channel) b y exchanging flags. It is envisaged that
`this service channel will provide terminal-to-termiii~il
`signaling and incorporate a frame alignment signal
`(FAS) o f 8 hits '80octt.t frame, a bit-rate~illocation signal
`(BAS) o f 8 bits, 80 octet frame, and u p to 6.4 khit s of
`data. ITnder the control of the BAS, the senice chanricl
`may tx expanded in increments of 8 khit s b y stealing
`additional bits from the speech chaiinrl.
`If more than 16 khit s are to be devoted to d:rta
`transmission, the benefits of ividehand audio heconic.
`marginal. Additional modes of audio transmission, e.g..
`narrou.hand audio within 32 khit s (possibly C.721) 01-
`even 16 khit) s , can be readily defined and \vould allow
`the capacity of the service channel to increase to 52 and 48
`k hi t s , I-espec t i ve 1 y . Sprerh t ran sin i ssion , ;I 1 though s t i 1 I
`quite intelligible, would I>r of lesser qualit\..
`
`Communication between Narrowband and
`Wideband Terminals
`
`To a1 low ;I utlio conferences het w e m participants,
`so in e ha 1.i 11 g w i de ha nd term i ria 1 s , o t hers na rrotvba n d
`terminals, con\wsion of narroivlxirid signals to widehand
`representation and widehand signals t o narrowhand
`represen t;i t ion is required. Narroivha nd term i na 1 s c;i n no t
`reproduce the high-band components of the widehand
`signal; this permits their derivation from only the low-
`txind component by filtering and PCM coding. However,
`generating a wideband signal using a QMF synthesis
`filter with only
`low-band a n d n o high-band
`input results in audible high-frequency distortion due to
`the uncanceled aliasing product. It is preferahle, there-
`
`samples U at 8 kHz
`
`. O
`
`Insert
`alternate
`zero level
`
`filter
`(high
`
`- 1 0
`
`Higher
`sub-band
`(8 kHz
`sampling
`
`alternate
`
`Delete
`alternate
`
`sub-band
`(8 kHr
`sampling
`rate)
`
`1 -
`
`filter
`(high
`
`(G.7113
`
`uniform
`
`January 1988-Vol. 26, No. 1
`IEEE Communications Magazine
`
`

`

`Powered by TCPDF (www.tcpdf.org)
`
`References
`
`[ I ] \V. R. Daumrr, P. hlermelstein, X. Maitre, and I .
`T o k i m r v a , “O\,rrvie\v o f the ADPCM coding algorithm,”
`Cortf. Record Globrcom ’81, Atlatita, GA, pp. 2 3 . l . l -
`23.1.4, 1984.
`[2] E. Hiirnmel, “The CCITT,” ZEEE Coinmun. .&Ing., vol.
`23, 110. I , pp. 8-11, 1985.
`[3] IEEE: (;loha1 ~ I ’ e l e c o m m i ~ n i c a t i o ~ ~ s
`C:onfrretltc--(;lol)e-
`corn ‘86, Houston, ~1.X. Papers 17.1 thtough 17.5, 1986.
`[-11 T. Nishitani, I. Kui-oda, M. Satoh, T. Katoh, :iricl Y . Aoki,
`stantlard 3 2 k b s ADPCM I.SI c o c k , ” ZEEE
`“A C:C:I?‘T
`Tmn.\. , 4 ~ oust, Speech Sig-. Pro(.., \ol. ASSP-35, pp.
`219-225, 1987.
`[ 31 R. E:. (:rochiere and J. Flan;rgan, “Curt ctit pu-spcc t i v r in
`digital sperch,” IEI.:E C o m m i o t . itlag,, v o l . 21, no. I ,
` pp.
`32- 10, 1S83.
`[6] G . \Villiatiis ant1 H. Suytlcrhoud, ”Suhj(~( tive pctfot -
`
`tiiaii(r r\~aI~~:itioii of ihc 32 kl) s ADP(:hl iilgorithm,” it]
`t . Corif.-C;lobecom
`‘81?
`
`[ 71 R. E. <:roc-liicrc, S . .A.
`\Vchber, ;incl J. L. Flanagan,
`“Digital coding 01 ~precli in sub-hancls,” B P I / S y \ l . Tec I t .
`J., pp. 1069- 1085, 1976.
`[ 8 ] N . S. J a y a n t and P. N o l l . Zligitcil Codiit,q of li’rcuefoi-rrz.\.
`F,tiglc\vood CIiJfs, N J : Prentice-Hall, 11. 306, 1984.
`[9] D. M i l l a r a n c l P. hleriiielstcin, “ P r e v e n t i o n of prrdictoi
`mistrac king in ADPChl todrrs,” Pro(, I E E E Z r i t . C o r t f .
`Corn ni i t ti. -ZCC ‘81, A m s tcrtla in, pp. 1 508- 1 5 1 2 , 1 98-1.
`[ I O ] D. J. Gootiinan. “Eml~cddetl DPCXI f o t - I’ariahlr Bit Kate
`~1.1-ansinissioti,” ZEEE Trcitt.\. Conzinuii.. \,ol. COM-28,
`pp. 10 10-10-l(i. 1980.
`
`fore, to generate a pseudo-high-hand signal from the
`narrowband i n p u t arid use i t later to cancel the aliasing
`products of the loiv-hand signal.
`T~vo alternative procedures suggest themselt,es for the
`nari-owhand~ wideband conversion. T h e first and more
`straightforward is to upsample the uniform PCM
`representation t o 16 kHz, loivpass the result t o eliminate
`the 4-8-kHz aliasing component, arid thcn \videhand
`encode the result ; i s if i t ~ v c r e a normal widehand signal.
`A second 111-ocedure is simpler and avoids the need for a
`new lo\v-p;iss filter design. As shown in Fig. 8, i t
`generates 21 lower suhhand signal b y a series of tlvo Q M F
`high-piss opera tioiis on the aliased narrowhand signal.
`I t also generates a n artificial high-hand signal ljy ;I series
`of high- and low-pass operatiom on the same aliased
`signal. \l.’hen the t l v o suhtxirid cornpotieIits are p;issed
`
`through Q M F synthrsis in :I ~ ~ i t l c h a ~ i d terminal, ;I
`naiiowhand signal is heard Lvith no additional noise.
`<:on ferencc bridges combine several input signals and
`ma) ti-ansmi t different o u t p u t signals, depending on
`whether the Ixirticulai port is consitlered acti1.e (currently
`speaking) o r silent. FOI witlehancl audio bridges, i t
`appeai-s preferable to c~omhirie the l o i v - and high-hand
`(oniponents of the input signals f r o m the several ports
`separately, ;is this avoids thc accumulation o f delays due
`to Q M F analysis and synthesis :it conference bridges. T o
`;ichic\,r hest quality when mixing narrowband and
`~vidchand inputs, n;irrowlxind inputs should first be
`c.on\.erted to wideband form. T h e all-widehand bridge
`i n a y then ern ploy signa 1 coin hina t ion 1 ogic ana logo u s to
`that found in nai-roivhand bridges, hut implement i t
`sepal-atrly for the 1 0 ~ ’ - and high-hand signal components.
`
`Concluding Remarks
`
`T h e new widehand coding stantlard represents a n
`iniportant advance in two aspects: first, i t improves the
`quality of audio o n the telephone netlvork; second, i t
`provides for a n audio-associated data channel to carry
`conversation- or conference-related data. Its deployment
`is currently limited t o locations accessihle hy .i6- or 64-
`khit, s digital loops. However, since there is much
`current interriational interest in digital networks such a s
`I S 1) IV , the pen‘ t r:i t i on of end - t o-e nd dig i t ;i 1 con nec t i 1.i t y
`will proha1)ly increase rapidly, and the cost of digital
`transmission will decrease simul tancously. T h e adoption
`of the new standard is timely and should not only prevent
`the pro1 ifem t i on of incorn pat iblc cotli rig techn iq u c s , hut
`also help i 11 m i
` king cost -cf fec t i1.e premium-quali t y
`audio term i na 1 s ra pi dl y :i\.:i i la hle .
`
`15
`
`January 1988-Vol. 26, NO. I
`IEEE Communications Magazine
`
`