United States Patent (19)
`Gilloire et al.
`
`73) Assignee:
`
`(54 ECHO CANCELLING DEVICE WITH
`FREQUENCY SUB-BAND FILTERING
`75) Inventors: André Gilloire, Lannion, France;
`Martin Vetterli, New York, N.Y.
`ETAT Francais représenté par le
`Ministre des postes,
`Télécommunications et de l'Espace
`(Centre National d'Etudes des
`Télécommunications), Issy les
`Moulineaux, France
`21 Appl. No.: 322,947
`(22
`Filed:
`Mar. 14, 1989
`30
`Foreign Application Priority Data
`Mar, 15, 1988 FR France ................................ 8803341
`51
`Int. C. ............................................... HO4B 3/23
`52 U.S. C. ................................... 370/32.1; 379/388;
`379/410
`58 Field of Search ............... 379/388, 390, 406, 409,
`379/410, 411; 370/32, 32.1
`References Cited
`U.S. PATENT DOCUMENTS
`4,564,934 1/1986 Macchi ............................... 370/32.1
`4,644,108 2/1987 Crouse et al. ....................... 379/406
`FOREIGN PATENT DOCUMENTS
`0.122340 6/1987 Japan ................................... 379/410
`0125722 6/1987 Japan ................................... 379/410
`
`56)
`
`
`
`
`
`
`
`BANK OF SNTESS
`FILTERS
`
`11
`45
`
`Patent Number:
`Date of Patent:
`
`4,956,838
`Sep. 11, 1990
`
`0.163424 7/1987 Japan ................................... 379/406
`Primary Examiner-Jin F. Ng
`Assistant Examiner-Magdy Shehara
`Attorney, Agent, or Firm-Fleit, Jacobson, Cohn, Price,
`Holman & Stern
`ABSTRACT
`57
`An echo cancelling device for use between a line re
`ceiving an incoming signal and a line transmitting an
`outgoing signal, for cancelling out echo, comprising a
`plurality of processing channels connected in parallel
`relation and assigned to successive mutually adjacent
`sub-bands of the spectral band of the outgoing signal,
`each channel having: a first analysis band-pass filter
`receiving the echo-affected signal to be transmitted,
`whose output is connected to the additive input of a
`subtractor; a second analysis band-pass filter, identical
`to the first filter, receiving the incoming signal and
`feeding an adaptive filter delivering an estimated echo
`value in the sub-band to the subtractive input of the
`subtractor; and a synthesis filter, symmetrical with the
`analysis filters and whose output feeds the transmission
`line. Each processing channel receives an estimation of
`the aliasing component originating from another sub
`band at least and that component is eliminated by add
`ing it to the signal originating from the filter analyzing
`the respective incoming signal. Estimation may be made
`by synthesis using at least one adaptive cross-filter fed
`by the output of an adjacent channel.
`
`9 Claims, 3 Drawing Sheets
`
`D-2
`
`CROSS-FILTER
`
`AAAM
`
`BANK OF
`ANALYSES
`FILTERS
`
`- 1 -
`
`Sony v. Jawbone
`
`U.S. Patent No. 8,467,543
`
`Sony Ex. 1016
`
`

`

`U.S. Patent Sep. 11, 1990
`
`REMOTE TERMINAL
`
`AA
`-Z-r
`
`g
`
`Sheet 1 of 3
`72
`
`4,956,838
`
`A-E" y
`?t-
`
`1, 1
`
`A e
`
`c4g /
`
`70
`46
`
`(AA/OA 4AR7)
`
`- H3-'
`
`
`
`
`
`FG.2
`(AA/OA AAP7)
`
`at ADAPTIVE FILTERS
`AA/f
`
`FIG.5
`
`frage/arcy
`
`- 2 -
`
`

`

`U.S. Patent
`
`
`
`Sep. 11, 1990
`
`Sheet 2 of 3
`
`4,956,838
`
`
`
`SI SI?HLN?S ŠIO XINW9
`
`
`
`SHGIJ, TIJ
`
`- 3 -
`
`

`

`U.S. Patent Sep. 11, 1990
`FG.4,
`
`Sheet 3 of 3
`al
`
`4,956,838
`70
`
`BANK OF ANALYSIS FILTERS
`
`
`
`2 . 1 FILTER
`
`
`
`CROSS
`FILTER
`BANK OF
`ANALYSIS
`FILTERS
`
`4.
`f
`
`
`
`
`
`BANK OF
`SYNTHESIS
`FILTERS
`
`
`
`
`
`
`
`
`
`/0EXEEER-70
`
`
`
`
`
`
`
`/21 affe,
`
`y
`
`
`
`- 4 -
`
`

`

`1.
`
`ECHO CANCELLING DEVICE WITH FREQUENCY
`SUB-BAND FILTERNG
`
`5
`
`O
`
`5
`
`BACKGROUND OF THE INVENTION
`1. Technical Field
`The present invention relates to echo cancelling de
`vices for use in signal transmission installations for full
`duplex transmission. It finds a particularly important
`application in telephony for solving problems raised by
`teleconference installations and so-called "free hand”
`telephone sets or at least having amplified listening.
`These problems are formed by danger of Larsen effect
`and the existence of an echo of acoustic origin.
`The origin of such phenomena appears in FIG. 1
`which shows, with continuous lines, the general dia
`gram of a teleconference terminal. The signals x enter
`ing through a reception line LR and coming from a
`remote terminal 8 are amplified by an amplifier 10 and
`then broadcast in a listening hall by a loud-speaker 12.
`20
`In the listening hall, the sound waves coming from the
`speaker whose speech is to be transmitted and from the
`loud-speaker 12 through acoustic coupling (shown
`schematically by a channel 13) are picked up by one or
`more microphones 14. Microphone 14 is connected to
`25
`an amplifier 16 and the output signal y thereof is trans
`mitted over the send line LE to the remote terminal. A
`listener-speaker placed at the remote terminal 8 will
`consequently hear not only the speech which is in
`tended for him, but also an echo of his own speech with
`30
`a delay proportional to the length of the lines LE and
`LR, perturbated by the transfer function of the acoustic
`channel 13. This echo is all the more troublesome the
`higher its level and the longer the delay. In satellite
`transmission, the delay may reach 600 ms and consider
`35
`ably disturb intelligibility. The Larsen effect occurs
`when the coupling formed by channel 13 is sufficiently
`tight for the gain in the loop formed of the two termi
`mals and the lines exceeds 1.
`2. Prior Art
`Different devices have already been used for dealing
`with the acoustic echo and the Larsen effect, particu
`larly automatic amplification gain controllers or echo
`cancelling devices in each terminal or "free hands' set.
`The gain varying devices operate by introducing
`45
`attenuation before the loud-speaker or after the micro
`phone, depending on the detected communication di
`rection. These devices have the drawback of producing
`a subjective impression of speech cut-off when the de
`gree of inserted attenuation is high. Now, a high attenu
`50
`ation is necessary when the communication installations
`in which the delay is large and in which the echo must
`be greatly attenuated so as to remain tolerable.
`The invention relates to echo cancelling devices for
`directly dealing with the echo by compensating for it by
`55
`means of an equivalent signal and of opposite direction.
`For that, an echo cancelling device comprises an adapt
`ive filtering device connected to the incoming signal
`input line for delivering an estimation of the echo of the
`signal received at the subtractive input of the subtractor
`which receives, at its additive input, the useful signal
`affected by the echo.
`The principle of such a device is shown with broken
`lines in FIG. 1. The echo canceller comprises an adapt
`ive filtering device FA which receives the incoming
`65
`signal x and whose output feeds the subtractive input of
`a subtractor 18 whose additive input is connected to the
`output of the amplifier 16. The coefficients of filter FA
`
`4,956,838
`2
`are adapted automatically responsive to the signal e
`delivered to the line, equal to the difference between the
`echo-affected signal y of amplifier 16 and the estimated
`echo. The algorithm for adaptation of the coefficients of
`filter FA must be such that the filter models the charac
`teristics of the acoustic coupling channel 13, formed by
`the beginning of the impulse response of this channel.
`In the absence of speech from a speaker placed in the
`hall, the outgoing signal delivered to line LE is then
`reduced to an echo residue. When a speaker speaks in
`the hall in front of microphone 14, his speech is trans
`mitted to a line LE without being attenuated or modi
`fied. Often, a device (not shown) is provided for detect
`ing the presence of the speech signal coming from a
`speaker (for example by level detection) and for then
`momentarily blocking adaptation of filter FA so as to
`avoid any disturbance of the filter by the speech of the
`local speaker.
`This solution has the great advantage, over automatic
`gain control of making bidirectional operation possible
`without attenuation of the useful signals, but presently
`existing devices which apply it are not entirely satisfac
`tory, for their construction comes up against two main
`difficulties.
`In the case of telephone installations, the acoustic
`echo cancellers comprise transversal adaptive digital
`filters operating at a sampling frequency from 8 to 16
`kHz. Now, the impulse response of channel 13 is often
`very long and it corresponds to several thousands of
`coefficients at this sampling frequency. The computa
`tions to be carried out at each sampling time (filtering
`by convolution and adaptation of the coefficients of the
`filter) require a huge volume of operations. For carry
`ing them very rapidly, numerous and expensive elec
`tronic circuits are indispensable.
`The characteristics of the acoustic coupling channel
`13 are time-variable, for example when a person moves
`in the listening hall. Theoretical reasons, related to the
`spectrum of the signal received over line LR and to the
`required length of the adaptive filter, limit the ability to
`track these variations (which must be taken immedi
`ately into account to avoid the untimely reappearance
`of the echo) and the initial convergence rate.
`To overcome these problems an echo cancelling de
`vice has been proposed which is inserted between the
`line receiving the incoming signal and the line transmit
`ting the outcoming signal so as to cancel out the echo,
`of the type comprising a plurality of processing chan
`nels assigned to successive adjacent sub-bands of the
`spectral band of the outgoing signal, each channel hav
`Ing:
`a first analysis band-pass filter receiving the echo
`affected signal to be transmitted, whose output is con
`nected to the additive input of a subtractor;
`a second analysis band-pass filter, identical to the first
`filter, receiving the incoming signal and feeding an
`adaptive filter delivering an estimated echo value in the
`sub-band to the subtractive input of the subtractor; and
`a synthesis filter, symmetrical with the analysis filters
`and whose output feeds the transmission line.
`Such a device is described for example in the article
`"Kompensation akustischer Echos in Frequenzteilbän
`den', Walter Kellermann, Frequenz, Vol. 39 (1985) No.
`7-8, pp. 209-215. The general diagram of such a device
`is shown in FIG. 2. The signal x from line LR is frac
`tionated by a bank of M analysis filters BFAR into M
`frequency sub-bands, usually all having the same width.
`
`- 5 -
`
`

`

`5
`
`15
`
`4,956,838
`3
`4.
`The amplified signal y from the microphone is fraction
`filtering, for the respective sub-band, the aliasing com
`ponent originating from another sub-band for eliminat
`ated into M sub-bands 1, ..., k, ..., M by a bank of
`ing said component by adding it to the signal originating
`analysis filters BFAM identical to bank BFAR. In each
`from the filter analyzing the respective incoming signal.
`sub-band of serial numberk, an adaptive filter FAkis fed
`by the incoming signal xk and its output is subtracted by
`In practice, this result will generally be obtained by
`subtractor Sk from the signal y delivered by the analy
`forming said means as at least one adaptive cross-filter
`fed by the output of at least one adjacent channel. If the
`sis bank BFAM. As in the conventional canceller of
`FIG. 1, the adaptive filter FAkis adjusted so as to mini
`sub-band analysis filters are sufficiently selective, it will
`mize the power of the signal ek at the output of sub
`often be sufficient to provide a channel of order k with
`two cross-filters each using as input the output signals
`tractor Sk.
`10
`Adaptation of the coefficients, shown schematically
`from the adjacent channels of order k-1 and k--1 (or
`on FIG.2 by an oblique arrow, is provided by a specific
`of a single one for channels 1 and M). If, however, such
`circuit using a conventional algorithm which will gen
`selectivity is insufficient, more than two cross-filters
`erally be the gradient algorithm although, in some
`may be assigned to each channel.
`cases, a simpler algorithm may be adapted, for example
`The cross-filters may often be factorized into a fixed
`the sign algorithm, or another adaptation algorithm, for
`part, depending solely on the characteristics of the anal
`example the least squares algorithm.
`ysis and synthesis filter banks, and an adaptive part. The
`The M signals e1, . . . , ek, . . . , eM feed a bank of
`increase in computation volume is small and the overall
`synthesis filters BFS which rebuilds the full band signal
`reduction of this volume with respect to the full band
`sent as outgoing signal to the transmission line LE.
`conventional canceller is maintained in a ratio which
`20
`The device has several advantages compared with a
`may practically be as high as the number of sub-bands.
`conventional echo cancelling device, whose band is not
`The invention will be better understood from the
`following description of embodiments, given by way of
`fractionated:
`The volume of computations to be effected per unit of
`explanation and in no way limitative. The description
`time is considerably reduced since the signals in each
`refers to the accompanying drawings.
`25
`sub-band can be sub-sampled: if Fe is the sampling fre
`BRIEF DESCRIPTION OF THE DRAWINGS
`quency thought necessary for the full band signal, each
`of the M sub-bands of spectral width Fe/2M may theo
`FIG. 1, already described, is a block diagram of a
`retically be sub-sampled at its critical decimation fre
`teleconference installation equipped with a conven
`quency Fe/M. The volume of computations for the
`tional echo canceller device, shown with broken lines;
`30
`same impulse response time is theoretically divided by
`FIG. 2, already described, is a block diagram of a
`M, for the computation load in the analysis and synthe
`frequency sub-band echo canceller device;
`sis filter banks is negligible as compared with the com
`FIG. 3 is the general diagram of a device forming one
`putations to be carried out in the adaptive filters.
`example of implementation of the invention, a single
`channel being shown completely;
`In each sub-band k, the adaptation gain of the algo
`35
`rithm used may be optimized as a function of the power
`FIG. 4 illustrates a modification of the invention, in
`of signal xk in the band, which increases the conver
`which only the cross-filters from the bands directly
`adjacent to a given channel are kept;
`gence rate and the ability to track variations in the
`acoustic channel.
`FIG. 5 shows a possible form of the sub-bands in the
`But these theoretical gains cannot be completely
`case of the modification described in FIG. 4;
`obtained in practice. A complete study of such a device
`FIG. 6 illustrates an embodiment of the invention, in
`A. Gilloire, Experiments with sub-band acoustic echo
`which the cross-filters are factorized as a fixed part and
`an adaptive part.
`cancellers for teleconferencing, Proc. ICASSP-87,
`April 1987, Dallas, pp. 2141-2144) has shown that in
`DETAILED DESCRIPTION OF PREFERRED
`fact it is not possible to adopt the critical decimation
`45
`EMBODIMENTS
`frequency if the sub-bands are directly adjacent. In fact,
`sub-sampling at the critical decimation frequency, re
`The device shown schematically in FIG. 3, in which
`quired for reducing the computation volume as much as
`the elements corresponding to those of FIG. 2 bear the
`possible, cannot be used without causing spectrum alias
`same reference, again comprises a bank BFAM of M
`analysis filters receiving the echo-affected output signal
`ing to appear, at the frontiers between sub-bands, which
`50
`y from amplifier 16, and a bank BFAR for analyzing the
`are not cancelled by the adaptive filters. That leads
`either to using mutually separate sub-bands (without
`incoming signal x arriving over the reception line LR.
`overlap), but with the drawback of introducing gaps in
`The two filter banks are identical. The output yk of
`the spectrum of the signal reconstituted at the output of
`BFAM is applied to the additive input of the subtractor
`Sk which outputs signal ek applied to the corresponding
`the synthesis banks BSF (which adversely affects the
`55
`quality of speech when the number of sub-bands is
`filters of the synthesis filter bank BFS feeding the trans
`high), or to sub-sampling the sub-bands at a frequency
`mission line LE. The signals in each sub-band are deci
`higher than the critical decimation frequency so as to
`mated (sub-sampled) in the ratio M within the analysis
`form guard bands avoiding aliasing, which increases the
`banks BFAM and BFAR.
`computation speed required in the adaptive filters.
`The output of filter FAk is not applied directly to the
`subtractive input of subtractor Sk. It is first of all added
`SUMMARY OF THE INVENTION
`to the output of cross-filters FC, k . . . , FCK-1 k,
`An object of the invention is to provide an echo can
`FCk,k, . . . , FCMk. The device consequently con
`celling device which is improved particularly in that it
`prises, for each channel k, an adder Ak receiving the
`eliminates to a large extent the problem of spectrum
`output of the corresponding filter FAk and that of the
`65
`aliasing. For that purpose, the invention provides a
`associated cross-filters. The coefficients of all cross-fil
`device of the above-defined type, wherein each process
`ters contributing to the formation of the output signal of
`ing channel is provided with means for extracting by
`adder Ak are adjusted by circuits whose input signal is
`
`- 6 -
`
`

`

`15
`
`4,956,838
`6
`5
`The simplest solution, usable in telephony, consists in
`formed by the output ek of the subtractor. For the cross
`using two bands in the ranges of respectively from 0 to
`filters FC21,..., FCK, 1,..., FCM1, the coefficients will
`be adjusted in response to the component e1 of the out
`4000 Hz and from 4000 to 8000 Hz. A solution which is
`going signal, etc.
`slightly more complex, again using sub-bands of equal
`The device shown in FIG.3 requires each channel to
`width, consists in splitting the whole band into four
`comprise, in addition to the adaptive filter FA, M-1
`sub-bands each of 2000 Hz or into eight sub-bands each
`cross-filters FC. But, in practice, the aliasing bands at
`of 1000 Hz.
`the low and high limits of a sub-band of orderk often do
`In some cases, however, it will be of advantage to use
`not extend as far as the end sub-bands, of orders 1 and M
`sub-bands of unequal width. But in this case, the correc
`and that makes it possible to reduce the number of
`tion terms must be adapted to the correct sampling
`O
`cross-filters. Such reduction may be all the greater the
`frequency for each sub-band.
`more selective the filters of banks BFAR and BFAM. If
`In each case, the number of cross-filters used for each
`so-called quadrature mirror filters (QMF) or pseudo
`sub-band is selected in dependance of the number of
`QMF filters are used, the aliasing bands corresponding
`sub-bands for which overlap occurs. In the case, for
`to a channel may often be limited to a range which does
`example, of a structure with eight equal sub-bands of
`not extend beyond the two adjacent channels. FIG. 5
`1000 Hz in width, the overflows from a sub-band to
`shows, by way of example, the gain variation as a func
`adjacent sub-bands must not exceed 1000 Hz. This re
`tion of the frequency which can be obtained with QMF
`sult may be readily obtained using efficient filter banks
`filters: the gain becomes practically zero, for the filter
`whose complexity however remains limited, for exam
`of orderk of the bank, well before the central frequency
`20
`ple with 64 or 96 coefficients.
`of the adjacent sub-bands. The QMF and pseudo-QMF
`By way of example, the characteristics will now be
`filters have the property of providing a 3 dB gain at the
`given which can be used for a typical device, for a
`cut-off frequency and an overall response of the filter
`sampling rate of 16 kHz, required for transmission of
`bank which is flat, i.e. without modification of the fre
`speech over a wide band (up to 7 kHz). To provide a
`quency spectrum.
`25
`device operating at 16 kHz and intended to identify the
`In this case, the device may be greatly simplified. As
`first 62 milliseconds of the impulse response of the
`shown in FIG. 4, it is sufficient to provide two cross-fil
`acoustic coupling channel 13, an arrangement as shown
`ters FC-1, and FC 1: for channel k (and even a
`in diagrammatic form in FIG. 6 is used. Each sub-band
`single one, FC21 or FCM-M, for the end channels 1 or
`is sub-sampled at 8000 Hz. Each main filter FA1 or FA2
`M). It can be seen in FIG. 4 that the output of the filter
`has a length of 500 coefficients. Each cross-filter has a
`corresponding to the sub-band of order k of the analysis
`fixed part FO of length 32 (for filter banks constructed
`bank BFAR feeds the input of the main adaptive filter
`from a low-pass prototype of 32 coefficients) and an
`FAk and the inputs of two cross-filters only FC-1
`adaptive part FC" with 160 coefficients, i.e. about one
`and FC: 1. The adder Ak with three inputs receives
`third of the length of the main filters, such length being
`the output of the main filter FAk and the outputs of the
`35
`sufficient in practice.
`cross-filters FC-1 and FC 1: the output of Ak is
`The adaptive filters are all adjusted by means of a
`subtracted from the signal of the sub-band k, delivered.
`stochastic gradient algorithm, the adaptation represent
`by the bank of filters BFAM, by the subtractor Sk.
`ing in this case as many computations as the convolu
`. The adaptive filters FA and the cross-filters FC may
`tion. The volume of computations to be carried out for
`have any one of numerous known constructions. Filters
`each pair of input samples is then:
`may be used employing the algorithm of the gradient
`frequently used in transverse adaptive filters. A filter
`2(2500-260)+232+3*32=2800
`using the least squares search algorithm gives even
`multiplications-summations.
`better results, at the price of greater computing com
`plexity. In some cases again, it is sufficient to use the
`This figure is to be compared with 4000 multiplica
`45
`sign algorithm. For a description of such filters, of their
`tions-summations which a conventional canceller must
`operation and their respective advantages, reference
`carry out for processing a pair of samples.
`may be made to different documents, for example to the
`In a second application which is also typical, in which
`article by O. Macchi et al "Le point sur le filtrage adap
`the device must again operate at the sample frequency
`tatif transverse', 11th colloque GRETSI, Nice, June
`SO
`of 16kHz but must identify the first 126 milliseconds of
`1987, pp. 1G-14G and to U.S. Pat. No. 4,564,934 (Mac
`the impulse response of the acoustic coupling channel, a
`chi).
`device of the kind shown in FIG. 3 is used. This device
`An advantageous solution consists in adopting, for
`comprises M=8 sub-bands. Each sub-band is sub-sam
`the cross-filters, a special structure shown in FIG. 6 in
`pled at 2000 Hz. Each main filter FA has a length of 250
`the particular case of fractionation into two sub-bands.
`coefficients and each cross-filter FC has a fixed part of
`Each of the cross-filters is factorized into a fixed part
`12 coefficients and an adaptive part of 84 coefficients.
`OF and an adaptive part FC". The part FO is equal to
`The analysis and synthesis filter banks are constructed
`the convolution product of the two filters forming each
`from a low-pass prototype of 96 coefficients. The band
`analysis bank decimated with a ratio of 1:2. In this case
`corresponding to the highest frequencies (7000 to 8000
`of two sub-bands, the filters of the banks BFAR and
`Hz) is not processed, for it contains signals of negligible
`BFAM are a low-pass filter for a bank, and a high-pass
`power. It consists of the transition band of the anti-alias
`filter for the other bank.
`ing input filters, not shown in the Figures.
`The same construction of the cross-filters could be
`The volume of computations to be carried out for
`transposed to the embodiments shown in FIG. 3 and
`each block of 8 samples is then:
`FIG. 4.
`Fractionation into sub-bands will be made in each
`2"(7250-1284)+7.12+3*(96+56)=5856
`case taking into account the characteristics of the trans
`multiplications-summations.
`mission installation.
`
`30
`
`55
`
`65
`
`- 7 -
`
`

`

`10
`
`15
`
`5
`
`4,956,838
`7
`component by adding it to the signal originating
`This figure is to be compared with the 32000 multi
`plications-summations which a conventional canceller
`from the filter analyzing the incoming echo
`must carry out for processing the same block of eight
`affected signal corresponding to said particular
`samples.
`sub-band.
`The invention is not limited to the particular embodi
`2. The echo cancelling device according to claim 1,
`ments which have been shown or described; these appli
`and further comprising at least one adaptive cross-filter
`cations are not limited to telephony and to acoustic
`connecting adjacent channels and fed by the output of
`echoes. The echo cancelling device may also be used in
`at least one adjacent channel.
`data transmission installations for cancelling out the
`3. The echo cancelling device according to claim 2,
`wherein said processing channels are respectively as
`electric echoes on the line.
`signed to first, intermediate and last mutually adjacent
`We claim:
`sub-bands of the spectral band of the outgoing signal,
`1. An echo cancelling device for use between a line
`receiving an incoming echo-affected signal and a line
`each of said processing channels, except the channels
`transmitting an outgoing signal, for cancelling out echo,
`for the first and last sub-bands, receive, on inputs
`comprising a plurality of processing channels connected
`thereof, output signals from two adjacent sub-bands
`only.
`in parallel relation and assigned to successive mutually
`adjacent sub-bands of the spectral band of the outgoing
`4. The echo cancelling device according to claim 3,
`signal, each of said channels comprising;
`wherein the analysis band pass filters are quadrature
`a subtractor circuit having an additive input and a
`mirror filters or pseudo-quadrature mirror filters.
`subtractive input for delivering an output signal at
`5. The echo cancelling device according to claim 2,
`wherein each of said cross-filters comprises a fixed filter
`an output;
`an adaptive filter having an input and an output con
`part, depending solely on the characteristics of the anal
`ysis and synthesis filters, and an adaptive filter part.
`nected to the subtractive input of said subtractor
`6. The echo cancelling device according to claim 1,
`circuit;
`a first analysis band-pass filter receiving the incoming
`wherein all of said sub-bands are of substantially equal
`25
`frequency width.
`echo-affected signal, the output of said first band
`pass filter being connected to the additive input of
`7. The echo cancelling device according to claim 2,
`and comprising a plurality of adaptive cross-filters
`said subtractor circuit;
`a second analysis band-pass filter, identical to the first
`wherein particular ones of said cross-filters contribute
`band-pass filter, receiving the incoming echo
`to provide an estimation for a particular one of said
`30.
`affected signal and feeding the input of said adapt
`channels and said particular ones of said cross-filters
`being adjusted by an algorithm which processes the
`ive filter for delivering an estimated echo value in
`the sub-band of the corresponding channel to the
`output of the subtractor circuit of the particular channel
`as an input parameter.
`subtractive input of the subtractor; and
`a synthesis filter connected to the output of said sub
`8. The echo cancelling device according to claim 1,
`35
`tractor circuit for rebuilding the full band of said
`wherein all of said filters are constructed for adjustment
`by a gradient algorithm.
`incoming echo-affected signal and whose output
`feeds the line transmitting the output signal,
`9. The echo cancelling device according to claim 1,
`means for extracting through filtering for a particular
`wherein at least some of said filters are constructed for
`being adjusted by the least squares algorithm.
`sub-band, the aliasing component originating from
`40
`another sub-band and for eliminating the aliasing
`:
`
`45
`
`SO
`
`55
`
`65
`
`- 8 -
`
`