`Stewart et al.
`
`11•11111111n1111
`
`US005761634A
`[11) Patent Number:
`£451 Date of Patent:
`
`5,761,634
`Jun.2, 1998
`
`[54) METHOD AND APPARATUS FOR GROUP
`ENCODING SIGNALS
`
`[75)
`
`Inventors: Kenneth A. Stewart. Palatine; Michael
`D. Kotzin. Buffalo Grove. both of m.
`
`(73) Assignee: Motorola, Inc .. Schaumburg. fil.
`
`(21) Appl. No.: 6'1:1~72
`
`[22) Filed:
`
`Apr. 4, 1996
`
`[63)
`
`[5 1)
`(52)
`
`[58)
`
`[56]
`
`Rebted U.S. Application Data
`
`Continuation of Sec. No. 198,750, Feb. 17, 1994, abandoned.
`Int. CJ.6
`.......... ........ ...................................... GOIL S/tO
`U.S. CJ . .......................... 704/220; 7041243; 704/211;
`704/229; 704/230
`Field of Seardt ................................ 395/22.21. 2.29.
`395/2.59. 28. 32. 34, 38-39
`
`References Cited
`U.S. PATENT DOCUMENTS
`
`4.455.649
`4,535,472
`4,896,362
`4.949,333
`4,956.871
`5,150,387
`
`6/1984 Esteban et al ...... - ................ 39512.38
`8/1985 Tomcik .................................. 39512.38
`111990 VeJdbuis et al. ··--··········· ..... 395/2.38
`8/1990 Koh et al. .............................. 39512.38
`911990 Swaminathan ......................... 39512.37
`911992 Yoshikawa et al. . .................. 39512.38
`
`5,159,447
`5,204,876
`5,301,255
`5,341 ,457
`5,349,645
`5,367,608
`5,383219
`5,416,797
`5,535.239
`
`1011992 Haslcell et al . .......................... 348/419
`4/1993 Brockert et al. ............................ 37 511
`4/1994 Nagai et al ........................... 39512.39
`811994 Hall, ll et al .......................... 39512.31
`911994 Zhao ...................................... 39512.52
`1111994 Veldhuis et al .
`....................... 395/2.38
`111995 Wheadey. III cl al ..................... 375/I
`511995 Gilhousen et al ..................... 375n05
`7/1996 Padovani el al .
`....................... 375/205
`
`PrirMry Ewmin.er-Allen R. MacDonald
`Assistant Examiner-Vijay B. Chawan
`Attome}1 Agent, or Finn-Richard A. Sonnentag
`
`[57]
`
`ABSTRACT
`
`A code division multiple access (CDMA) communication
`system reduces system self-interlerence and enhances sys(cid:173)
`tem capacity by making rate selection decisions for indi(cid:173)
`vidual speech encoders in concen with other speech encod(cid:173)
`ers. The system utilizes perpetually weighted error metrics
`(4tl) as input into a rate controller (414) which determines
`and provides selected rates ( 402) back to the encoders ( 1&5).
`The system provides optimum voice quality and system
`capacity in that it allows specific encoders to decrease their
`rate. which improves capacity. as necessaiy while allowing
`other encoders to maintain their rates. Th.is prevents needless
`degradation in voice quality at those ti.mes when system
`capacity needs to be temporarily inacased.
`
`27 Claims, S Drawing Sheets
`
`JOJ
`
`---1
`
`STP fIL TER
`A(z)
`
`307
`
`PERPETUAL
`WEIGHTING
`HLTER
`W(z)
`
`LTP GAIN
`G
`305
`
`JfJ
`
`306
`
`CODEBOOK
`
`PERPETUALLY WEIGHTED
`ENCODING ERROR METRIC
`JOB
`
`J02
`
`312
`CODEBOOK
`INDEX I
`
`64kbps_.__1_04_ ...,_ ___
`PCM
`
`JOO
`
`AUTOCORRELATION
`ESTIMATOR
`
`AUTOCORRELATION
`FUNCTION
`309
`
`SCHUR
`RECURSION
`
`JOI
`ENCODED STP
`STP
`Jll
`----1 COEHICIEtH .........,_. PARAMETERS TO ERROR
`JIO ..._ ___ _
`QUANTIZER
`CORR£CTION BLOCK 106
`STP
`COEffICIENTS
`
`IPR2016-01710
`UNIFIED EX1029
`
`
`
`F IG.f
`
`L-----------------------------------------------
`
`I
`
`MODULATION
`
`8PSK
`
`CORRECTION
`
`ERROR
`
`ENCODER
`SPEECH
`
`•
`•
`•
`
`•
`•
`•
`
`•
`•
`•
`
`MODULATION
`
`BPSK
`
`CORRECTION
`
`ERROR
`
`ENCODER
`SPEECH
`
`107
`
`106
`
`105
`
`MODULATION
`BPSK
`
`CORRECTION
`
`ERROR
`
`ENCODER
`SPEECH
`
`104
`PC
`64k~s
`
`FRAMES
`SPEECH
`ENCODED
`
`100
`
`(1.544Ub/s)
`T1 LINKS
`
`101
`
`~----------------------~------------------------,
`
`1~3
`
`ATION
`MOBILE
`
`ST
`
`102~
`
`STATION
`BASE
`
`
`
`FRAMES
`SPEECH
`ENCODED
`
`
`!/
`
`I
`_______________________________________ J
`I
`I
`
`FIC.2
`
`ENCODER
`SPEECH
`CELP
`I
`
`206-..,
`
`SELECTED
`
`....__
`
`CODER
`
`DE
`
`;-204
`
`I
`
`-v
`"'
`/
`v
`r<20J
`
`~
`
`.....__
`
`THRESHOLDS
`
`SET
`
`r202
`
`I
`
`AVERAGING
`
`FILTER
`
`201,
`
`R(O)
`
`AUTOCORRELATION
`
`ESTIMATOR
`
`I
`I
`I
`I
`
`I
`I
`I
`I
`I
`205~ I
`RATE
`I
`I
`I
`I
`I
`I
`I
`i
`I
`I
`I
`I
`I
`I
`
`----------------------,
`SPEECH ENCODER
`
`COMPARATORS
`
`/105
`
`r----------------J_
`
`200
`
`/
`
`I
`
`\1
`
`PCM
`64kbps
`
`104
`
`
`
`QUANTIZER
`
`STP
`
`SCHUR
`
`ENCODING ERROR METRIC
`PERPETUALLY WEIGHTED
`
`JOB
`
`W£IGHTING
`PERPETUAL
`
`W(z)
`FILTER
`
`307
`
`+
`
`FILTER
`----1 L TP IIR
`
`LTP GAIN
`
`305
`G
`
`LTP LAG
`
`304
`L
`
`302
`
`301
`
`AUTOCORRELATION
`
`FUNCTION
`
`309
`
`AUTOCORRELATION
`
`ESTIMATOR
`
`PCM
`64kbps.~_t_o4______ STP FILTER
`
`A(z}
`
`JOO
`
`JOJ
`
`F IG.3
`
`ENCODED STP
`
`311
`
`312
`
`INDEX I
`CODEBOOK
`
`CODEBOOK
`
`306
`
`J1J
`
`CORRECTION BLOCK 106
`RECURSION -.--~ COEFFICIENT .......,_. PARAMETERS TO ERROR
`
`COEFFICIENTS
`
`STP
`310
`
`
`
`U.S. Patent
`
`Jun. 2, 1998
`
`Sheet 4 of 5
`
`5,761,634
`
`""
`104
`64kb
`ps
`PCM
`104 ""'\
`
`PERPETUALLY
`WEIGHTED
`ERROR METRIC
`BY RATE
`
`401-.,.
`
`104
`
`/ 105
`SPEECH
`ENCODER
`
`1
`-
`
`2
`
`SPEECH
`ENCODER
`•
`•
`•
`SPEECH N
`ENCODER
`
`I
`
`" I
`
`ENCODED
`SPEECH
`FRAMES
`
`v40 2 SELECTED RATES
`
`404-....,
`
`RATE
`CONTROLLER
`
`FIC.4
`
`CANDIDATE RATE
`1/4
`1/2
`1/8
`250
`150
`525
`400
`310
`140
`300
`385
`125
`•
`•
`•
`•
`•
`•
`280
`300
`105
`
`1/1
`60
`40
`35
`•
`•
`5
`
`1
`2
`3
`•
`•
`N
`
`FIC.5
`
`SPEECH
`ENCODER
`NUMBER
`
`
`
`U.S. Patent
`
`Jun. 2, 1998
`
`Sheet 5 of 5
`
`5,761,634
`
`Tl LINK
`103
`
`60 1
`
`Tl
`INTERFACE
`
`602
`
`TO ERROR
`CORRECTION
`CIRCUITRY 106
`
`ENCODER
`DIGIT AL SIGNAL
`PROCESSOR #1
`
`ENCODER
`DIGIT AL SIGNAL
`PROCESSOR f 2
`
`ENCODER
`• • • DIGIT AL SIGNAL
`PROCESSOR #M
`
`604
`
`RATE INFORMATION
`AND RATE SELECTION
`COMMANDS
`
`603
`
`SUPERVISING
`PROCESSOR
`
`FIC.6
`
`401
`404
`r--- - -------L----------------,
`402
`700
`703
`706
`
`I
`I
`I
`I
`I
`I
`t MEANS FOR
`I
`ACCEPTING
`I
`I
`I
`I
`I
`I
`L---------------------------~
`
`MEANS FOR
`DETERMINING
`
`MEANS FOR
`ADJUSTING
`
`FIC.7
`
`
`
`5.761.634
`
`1
`METHOD AND APPARATUS FOR GROUP
`ENCODING SIGNALS
`
`This is a continuation of application Ser. No. 08/198.750.
`filed Feb. 17. 1994 and now abandoned.
`
`FIELD OF THE INVENTION
`
`The present invention relates to communication systems
`utilizing code division multiple access (CDMA) techniques
`and more specifically to variable rate speech encoding
`methods for reduction of system self-interference and
`enhancement of system capacity in point-multipoint
`multiple-access links with colocated digital speech encoding
`in such communication systems.
`
`BACKGROUND OF THE INVENTION
`In recent years a variety of techniques have been used to
`provide multi-user mobile communications within a limited
`available radio-frequency spectrum. These methods have
`included frequency division multiple access (FDMA). time
`division multiple access (TDMA). and code division mul(cid:173)
`tiple access (CDMA) or. more usually. hybrids of these
`methods. All of these methods have been employed within
`the past decade in the design of commercial cellular tele(cid:173)
`communications systems: witness the use of FDMA in the
`North American AMPS system. FD!fDMAin the European
`Groupe Speciale Mobile (GSM) standard. and-more
`recently-the adoption of a direct sequence FD/CDMA
`approach by the United States Telecommunications Industry
`Association as embodied in its IS-95 standard. In the IS-95
`standard. subscribers share one of several wideband radio
`channels in the cellular band. Several proposals for so-called
`personal communications systems (PCS) are also being
`designed on similar FD/CDMA principles.
`Almost all recent cellular and PCS systems have used
`digital speech coding and forward channel error correction
`as the physical layer for voice communication. More inter(cid:173)
`esting in this context. is the use of voice activity detection
`(V AD) to recognize the presence or absence of speech on the
`part of the either calling party. In the absence of speech. the
`speech encoder may instruct the modulator or transmitter to
`which it is linked to reduce its output power to zero. or
`transmit occasional packets of information describing only
`the background noise at the either user's location. Reducing
`the radio transmitter's duty cycle in this fashion provides the
`twin benefits of a reduction in power consumption (which
`increases battery life in the case of the mobile unit) and a
`reduction in interference between users sharing the same RF
`spectrum. Depending on the circumstances of the 50
`conversation. a reduction in transmitted power of between
`40% and 65% can be achieved. The amount of power
`reduction is ultimately limited by the extent to which the
`degraded voice quality which accompanies significant V AD
`techniques is considered acceptable.
`The possibility of power reduction is particularly impor(cid:173)
`tant for CDMA systems. In such systems, user capacity is
`inversely proportional to the amount of system self(cid:173)
`interference. In the TIA IS-95 FD/CDMA standard. the
`approach is slightly broadened by the use of a variable rate 60
`speech encoder in place of simple on-off or discontinuous
`transmission methods. In the IS-95 standard, the encoded
`speech is separated into 20 ms intervals which the speech
`encoder may elect to encode at a effective bit rate of 8000
`bps. 4000 bps. 2000 bps, or 800 bps. Both the base-station
`to mobile station (forward) and mobile station to base(cid:173)
`station (reverse) IS-95 links exploit variable rate encoding.
`
`5
`
`2
`In the case of the forward link. mean transmit power is
`reduced by scaling down the output power as the encoded
`rate decreases. Channel symbol repetition allows symbol
`combining at the mobile receiver and hence maintenance of
`the energy per symbol to noise power spectral density ratio
`which determines link performance. It should be noted that
`mean transmit power-and hence system self(cid:173)
`interference-is reduced by a factor of four during 800 bps
`transmission. By averaging over the aggregate voice activity
`10 for typical two-way conversations. it has been estimated that
`when using the standard speech encoding and voice activity
`detection algorithm defined in TIA standard IS-96 the mean
`transmit power will drop to around 41 % of its nominal
`value. This has a significant effect on system forward link
`15 capacity.
`In current implementations, however. of the IS-95 air
`interface standard and its associated IS-96 speech encoder
`standard, each forward voice link is encoded in isolation.
`That is. speech encoders make individual determinations of
`20 the minimum encoded rate required to maintain acceptable
`voice quality without regard to the other voice channels
`sharing the same RF spectrum. This requires that the rate(cid:173)
`determination algorithm in each speech encoder should
`always minimize its encoded rate. even when the encoded
`25 rates of the other speech encoders sharing the same spectrum
`does not require this. For example. if all speech encoders
`sharing the same channel at a base-station should simulta(cid:173)
`neously seek to transmit at a low rate. the reduction in the
`total output power at the base-station means that each speech
`30 encoder could relax to the next higher rate at no risk to
`system capacity. Since minimizing the mean transmitted rate
`for a variable rate speech encoder requires that the voice
`quality be compromised. isolated speech encoding gives up
`voice quality needlessly. Also. as the CDMA system
`35 approaches capacity, and constraints are placed on the
`transmitted rate of each speech encoder in order to tempo(cid:173)
`rarily boost capacity. such constraints must be applied
`blindly. with all voice links subject to the same constraint
`irrespective of the effect on voice quality. This is wasteful.
`40 since it is known that voice quality depends on many
`different factors and so opportunities will exist for reducing
`the rate of specific encoders with the least overall effect on
`the voice quality experienced within the sector/cell.
`Thus a need exists for a method and apparatus for global
`45 speech encoding on the forward link of an FD/CDMA
`system by making rate selection decisions for individual
`speech encoders in concert with all other speech encoders
`feeding the same sector/cell and RF channel.
`BRIEF DESCRIPTION OF THE DRAWINGS
`FIG. 1 generally depicts. in block diagram form. a prior
`art CDMA base-station transmitter.
`FIG. 2 generally depicts. in block diagram form. the prior
`art rate determination apparatus specified in speech encod-
`55 ing standard TIA IS-96.
`FIG. 3 generally depicts, in block diagram form. a code(cid:173)
`book excited linear predictive speech encoder of the type
`utilized in the preferred embodiment and described in detail
`in speech encoding standard TIA IS-96.
`FIG. 4 generally depicts, in block diagram form. the use
`of a supervising processor or rate controller to group speech
`encode in accordance with the invention.
`FIG. S generally depicts a rate/quality table utilized by the
`rate controller as part of a number of algorithms for opti-
`65 mizing the overall speech quality of a sector/cell subject to
`a constraint on transmitted power from the sector/cell in
`accordance with the invention.
`
`
`
`5.761.634
`
`3
`FIG. 6 generally depicts. in block diagram form. an
`apparatus for implementing the CDMA group encoding
`method in accordance with the invention.
`FIG. 7 generally depicts. in block diagram form. a rate
`controller which may beneficially implement group encod(cid:173)
`ing in accordance with the invention.
`DEfAILED DESCRIPTION OF A PREFERRED
`EMBODIMENT
`A code division multiple access (CDMA) communication
`system reduces system self-interference and enhances sys(cid:173)
`tem capacity by making rate selection decisions for indi(cid:173)
`vidual speech encoders in concert with other speech encod(cid:173)
`ers. The system utilizes perpetually weighted error metrics
`(401) as input into a rate controller (404) which determines
`and provides selected rates ( 402) back to the encoders ( 105 ).
`The system provides optimum voice quality and system
`capacity in that it allows specific encoders to decrease their
`rate. which improves capacity. as necessary while allowing
`other encoders to maintain their rates. This prevents needless
`degradation in voice quality at those times when system
`capacity needs to be temporarily increased.
`The preferred embodiment of the invention is described as
`it relates to a CDMA digital cellular communications system
`based on the Telecommunications Industry Association stan(cid:173)
`dards IS-95 and IS-96. It will be appreciated by one skilled
`in the art that the invention may be applied to any CDMA
`point-to-multipoint link (generally the forward link of a
`digital cellular system) in which self-interference reduction
`by variable rate speech encoding is to be applied However.
`the technique discussed may be beneficially utilized in any
`communication system. and in fact is not restricted to
`communication systems. For example. the technique may be
`utilized where speech encoding occurs for storage in a
`memory means having limited memory space. In essence,
`the technique is applicable to any application where encod(cid:173)
`ing (be it speech. video, data. etc.) is utilized and constraints
`related to the encoding (be it power level. encoding quality.
`system capacity. memory space. etc.) are present.
`The method and apparatus group encodes signals by
`accepting rate determination information from at least two
`encoders and determining the rate of at least one encoder
`based on the rate determination information for the at least
`two encoders. In the preferred embodiment. rate determina(cid:173)
`tion information is quality information relating reconstruc(cid:173)
`tion quality as a function of encoding rate (on a 20 ms
`segment-by-20 ms segment basis). The quality information
`includes. but is not limited to. perceptual weighting error
`metrics generated by the analysis-by-synthesis speech
`encoders. signal-to-noise (SIN) ratio. segmented SIN. cep(cid:173)
`stral distance. an LPC distance measurement and a BARK
`spectral distance measurement. all of which are well known
`in the art.
`The determination of the rate of the at least one encoder
`is based on a threshold criterion (which is typically
`predetermined). In the preferred embodiment. the threshold
`criterion may include. but is not limited to, the total output
`power of the sector/cell to which the encoders are assigned,
`the total output power of an adjacent sector/cell. the current
`power level of transmission by a serving base-station, the
`current data rate of the at least two encoders. the memory
`available in a memory means. the processing power avail(cid:173)
`able in a processing means and the bandwidth available in a
`predetermined spectrum. Also in the preferred embodiment.
`the encoders are variable rate analysis-by-synthesis encod(cid:173)
`ers. These encoders may encode signals including. but not
`limited to. speech signals. video signals and data signals.
`
`4
`FIG. 1 shows the high-level architecture of the forward
`link of a CDMA base-station (102) designed for the pre(cid:173)
`ferred embodiment of the TIA IS-95 digital cellular radio
`standard. The base-station (102) of FIG. 1 performs, inter
`5 alia. variable rate speech encoding. forward error correction.
`forward link power control. multiple access spreading. and
`modulation and transmission. In FIG. 1. several standard
`µ-law encoded. multiplexed. 64 kbps pulse code modulated
`(PCM) Tl links (101) from the public switched telephone
`10 network (PSTN) (100) are brought to a demultiplexer (103).
`Each 64 kbps voice link (104) is then passed through a
`digital speech encoder (105). In a conventional
`implementation. the speech encoding function is performed
`by a number of general purpose digital signal processors
`15 (DSP's) such as the Motorola DSP56156 processor. ROM
`coded DSP' s. or application specific integrated circuits
`(ASICs). Several such processors are generally grouped
`onto a single printed circuit board (although this is not
`necessary for the invention) which is then capable of pro-
`20 cessing a full Tl trunk of multiplexed voice channels. After
`speech encoding. error correction (106) is applied in the
`form of convolutional and cyclic codes. followed by BPSK
`baseband modulation (107). Walsh cover and short pseudo(cid:173)
`noise (PN) sequence spreading (108). low-pass filtering
`25 (109). transmit power level adjustment (110) and power
`amplification (111). and finally transmission to the mobile
`station (113) (for simplicity. frequency shifting to RF is not
`shown).
`A block diagram of the TIA IS-96 standard processing
`30 performed by the DSP or other device used to implement the
`speech encoder (105) is shown in FIG. 2. As shown. speech
`encoder (105) can be broken down into two main elements:
`rate determination and encoding. Consider first the rate
`determination function. In the IS-96 standard, each speech
`35 encoder (105) divides its associated PCM signal stream into
`contiguous 20 ms frames consisting of 160 samples of the
`source speech waveform. The power level of each frame
`(which is the zeroth lag R(O) of the autocorrelation function
`estimate of the frame produced by the autocorrelation esti-
`40 mator (200)) is fed to a bank of comparators (203) which
`establish which of three monotonic-increasing threshold
`levels the frame power exceeds. These levels are generated
`by 2nd order interpolation of a non-linear average of the
`power level of the speech signal formed by block (201).
`45 Note that all these processing steps are completely defined
`in TIA standard IS-96. If the current frame energy is less
`than the lowest of the three thresholds. the frame is declared
`an :,3 rate frame; if the frame energy lies between the lowest
`and middle of the thresholds. the frame is declared a 14 rate
`50 frame; if it is between the middle and highest threshold. the
`frame is the declared a 'h rate frame; and finally. if the frame
`energy exceeds the highest threshold level. the frame is
`declared a full rate frame. This final step is performed by
`comparators (203) and decoder (204) to produce the selected
`55 rate (205).
`The selected rate (205) is then input to the codebook
`excited linear predictive (CELP) speech encoding function
`(206) which forms a parametric description of the speech
`frame using the specified number of bits for that rate. In the
`60 preferred embodiment. the number of bits used to express
`the encoded parameters of an 1/s rate frame is 16 (ignoring
`additional bits used for error correction/detection); for a 1A
`rate frame. 40 bits; for a 'h rate frame. 72 bits; and for a full
`rate frame. 160 bits. While CEIP is depicted and discussed
`65 in the preferred embodiment, other encoding techniques
`such as. inter alia. waveform coding, linear predictive cod(cid:173)
`ing (LPC). sub-band coding (SBC). code excited linear
`
`
`
`5.761.634
`
`25
`
`5
`prediction (CELP). stochastically excited linear prediction
`(SELP). vector sum excited linear prediction (VSELP).
`improved multiband excitation (IMBE). and adaptive dif(cid:173)
`ferential pulse code modulation (ADPCM) coding algo(cid:173)
`rithms may likewise be beneficially employed.
`For the purpose of clarity. it is necessary to describe in
`more detail the CELP speech encoding procedure. A high(cid:173)
`level block diagram of the signal processing used in the
`CELP speech encoder of the preferred embodiment appears
`in FIG. 3. As shown in FIG. 3. an estimate (309) of the
`autocorrelation function of consecutive 20 ms frames of the
`64 kbps speech signal (104) is first obtained (this is usually
`done in common with block (200) of the rate determination
`procedure). Next. solution of the so-called Normal Equa(cid:173)
`tions using. for example. the Schur Recursion (301). pro(cid:173)
`vides the short term linear predictive (STP) filter coefficients
`(310). Often, the STP filter (303) is a lattice filter. and the
`STP coefficients are lattice filter reflection coefficients. After
`quantization (302) by line spectral pairing or some other
`robust quantization method. the STP coefficients are used to
`filter the speech signal. The resulting signal is next passed to
`the long term prediction (LTP) filter (313) and (in the case
`of a CELP linear predictive coder) the codebook search
`procedure. The LTP filter is generally a first order recursive
`filter whose feedback delay and gain are variable-they
`appear in FIG. 3 as LTP lag L (304) and LTP gain G (305).
`Encoding then proceeds by simultaneously adjusting the
`LTP lag and gain and the codebook index 1 (312) so that the
`square error at the output of the LTP filter is minimized. L.
`G. and I are then quantized (often using simple biased linear
`quantizers methods). and passed along with the STP coef(cid:173)
`ficients to the error correction block. The performance of
`this analysis-by-synthesis procedure can be improved by
`weighting the error metric which is to be minimized by the
`human auditory frequency response. This is done with a
`perceptual weighting filter (307) which modifies the error
`metric (308) to emphasize those frequency components to
`which the human ear is most sensitive. One skilled in the art
`will appreciate that the perceptually weighted error metric is
`made available by almost all sophisticated analysis-by- 40
`synthesis speech encoders. The present invention. as men(cid:173)
`tioned above. is therefore not limited exclusively to CELP
`speech encoders.
`With this background. group speech encoding in accor(cid:173)
`dance with the invention may now be described. It is clear 45
`from FIG. 1 and FIG. 2 that, in the prior art, the encoded rate
`of each forward link speech encoder is determined in
`isolation. That is. the encoded rate of each 64 kbps voice link
`is determined exclusively by signal processing that speech
`signal. Since the amount of self-interference (and hence the 50
`capacity) in the forward link of a CDMA system depends on
`the mean encoded rate of each encoder. it is also clear that
`in order to operate at the maximum possible capacity. the
`rate determination algorithm of each speech encoder must be
`designed to always seek the minimum possible rate, since 55
`each encoder operates in isolation and has no knowledge of
`the total power (and hence system self-interference) being
`emitted at the base-station antenna (112). Since speech
`quality must be sacrificed to achieve low mean encoded
`rates. this implies that overall system speech quality is 60
`unnecessarily sacrificed when the system is not at its maxi(cid:173)
`mum capacity--or equivalently. is not transmitting its maxi(cid:173)
`mum allotted power. Put another way. isolated speech
`encoding allows the total instantaneous output power at he
`base-station to have a large variance.
`Since. in many CDMA power control algorithms, a strict
`limit is placed on total emitted power from a cell or sector.
`
`6
`the rate used to encode individual links must be kept
`unnecessarily low. In addition. it is known that the percep(cid:173)
`tual quality of a digitally encoded voice link is dependent not
`only on the speech encoder being used. but also on factors
`5 such as the gender. accent. loudness of the speaker. and
`environmental factors such as type/levels of acoustic back(cid:173)
`ground noise. Thus. by encoding each link in isolation. no
`recognition is made of situations where one link may be
`reduced in rate with a smaller loss in perceived overall voice
`10 quality than an equivalent reduction in rate on another link.
`and hence another speaker. Further. the current art embodied
`in TIA standard IS-96 makes no use of the perceptually(cid:173)
`weighted encoding error in performing rate determination.
`The method shown in FIG. 4 can be used to overcome
`15 these deficiencies. In FIG. 4. each speech encoder (105)
`evaluates. for each 20 ms frame. the perceptually weighted
`error metric (401) produced by encoding the speech frame at
`each of the four candidate rates (more than four rates may be
`possible in alternate embodiments). This information is then
`20 passed back to a supervising rate controller (404). Rate
`controller ( 404) then forms a rate/quality table similar to that
`of FIG. 5. which depicts the perceptually-weighted error
`produced by encoding at each of the candidate rates for each
`of the N speech encoders reporting to the rate controller.
`A simple approach to optimizing the overall voice quality
`of the cell or sector starts by assuming that all N voice
`channels have equal transmit power. All of the encoders
`(105) are placed in the lowest candidate rate and the total
`transmit power P is calculated by rate controller (404). In
`30 this case. Pis simply equal to the sum of the rate values for
`all N encoders. where the rate value for 1;8 rate is ''8. for V..
`rate is 14. and so on. Rate controller (404) then finds the
`largest entry in the rate/quality table corresponding to the
`current candidate rate for any of the N encoders. This is
`35 equivalent to identifying the encoder with the worst voice
`quality (i.e. the largest perceptually weighted error) for the
`current set of selected rates. The rate for that encoder is
`increased to the next highest rate. and Pis recalculated. This
`process continues until P exceeds some total power thresh(cid:173)
`old T at which time the procedure terminates. An improved
`approach would be to apply the procedure to rate/quality
`table entries which have been weighted by the transmit gain
`associated with each encoder. This would be extracted from
`power level block (110). It will be appreciated by one of
`ordinary skill in the art that the overall effect of this
`procedure is to reduce power by sacrificing the rate of those
`encoders which will suffer the least reduction in quality by
`operating at a lower rate.
`A more complex approach would be as follows. Assume
`that. as above, the goal (i.e .. the predetermined criterion) of
`the rate-reduction scheme during periods of high traffic
`loading is to maintain the overall transmitted power to be
`less than some threshold T. where T is set according to the
`current load conditions. Define a global measure Q of speech
`quality for the sector/cell served by the base-station to be the
`sum of the perceptual errors for the current set of selected
`rates for the N voice channels. Each encoder is initialized to
`encode at the maximum rate. Q is then evaluated and the
`corresponding transmitted power calculated using either the
`equal power assumption or the weighted transmit power
`method described above.
`A simplification of this method would occur where the
`rate controller (404) was not available. but where each DSP
`was encoding several voice links by time-sharing its avail-
`65 able computational resources. In that case, the rate selection
`procedure would be applied over the number of voice
`channels for which the DSP was performing encoding. FIG.
`
`
`
`5.761.634
`
`10
`
`7
`6 generally depicts an apparatus which may be used to
`implement this scenario. In FIG. 6. a single DSP(603). such
`as the Motorola DSP56156. communicates via a time(cid:173)
`division multiplexed serial bus or a conventional parallel
`address/data bus. Rate determination information and rate 5
`selections are passed between the controlling DSP (603) and
`the DSP's (602) used for speech encoding via bus (604).
`Alternatively. the controlling DSP (603) may be eliminated
`and one of the encoder DSP's (602) promoted to fulfill the
`global rate controller function and speech encoding for one
`or more voice channels.
`FIG. 7 generally depicts. in block diagram form. a rate
`controller (404) which may beneficially implement group
`encoding in accordance with the invention. Rate controller
`(404) comprises means for accepting (700) rate determina(cid:173)
`tion information (401) from a plurality of encoders (105). In 15
`the preferred embodiment. rate determination information is
`quality information which includes a perceptually weighted
`error metric. Means for accepting (700) has its output
`entering means for determining (703) which determines
`encoding requirements based on predetermined criterion. 20
`The predetermined criterion include those stated above as
`threshold criterion. The output of means for determining
`(703) is input into means for adjusting (706) which adjusts
`the encoding rate for any encoder out of the plurality of
`encoders based on the rate determination information and 25
`the predetermined criterion. In a scenario where the prede(cid:173)
`termined criterion is total transmit power or available
`memory space. means for adjusting (706) will typically
`increase the encoding rate for the encoder having the worst
`quality (based on a determination/calculation of either total 30
`transmit power or available memory space and a threshold)
`as described above. However, certain predetermined
`criterion. such as system capacity. may require means for
`adjusting (706) to decrease the encoding rate for a particular
`encoder.
`While the invention has been particularly shown and
`described with reference to a particular embodiment. it will
`be understood by those skilled in the art that various changes
`in form and details may be made therein without departing
`from the spirit and scope of the invention.
`What we claim is:
`1. A method of encoding signals in a wireless communi(cid:173)
`cation system. the method comprising the steps of:
`accepting quality information from at least two encoders;
`determining encoding requirements of at least one 45
`encoder based on the quality information from the at
`least two encoders;
`adjusting the encoding rate of at least one encoder based
`on the determined encoding requirements; and
`outputting the encoded signal from the at least one so
`encoder, wherein the encoded signal is encoded using
`the adjusted encoding rate of at least one encoder.
`2. The method of claim 1 wherein the at least two
`encoders further comprise analysis-by-synthesis encoders.
`3. The method of claim 1 wherein the quality information 55
`further comprises perceptual weighting error metrics gener(cid:173)
`ated by the analysis-by-synthesis speech encoders. signal(cid:173)
`to-noise (SIN) ratio, segmented SIN, cepstral distance. a
`linear predictive coding (LPC) distance measurement and a
`spectral distance measurement.
`4. The method of claim 1. wherein the step of determining
`encoding requirements of at least one encoder based on the
`quality information from the at least two encoders further
`comprises the step of determining the encoding require(cid:173)
`ments of the at least first encoder based on the qual