EP0709999A2

Method and apparatus for multi-channel acoustic echo cancellation

Abstract

A variation in the cross-correlation between current received signals of different channels is extracted which corresponds to the cross-correlation between previous received signals, and the extracted variation is used as an adjustment vector to iteratively adjust the estimation of the impulse response of each echo path. Furthermore, by additionally providing a function of actively varying the cross-correlation between the received signals to such an extent as not to produce a jarring noise, it is possible to reconstruct acoustic signals by individual loudspeakers and utilize received signal added with the cross-correlation variation to obtain the adjustment vector for an estimated echo path vector.

EP0709999A2, drawing sheet 1 — Sheet 1 of 234

Term

Term ended

Projected expiry passed 24 October 2015, 10.9 years ago.

22 claims: 6 independent, 16 dependent

1
An N-channel echo cancellation method comprising the steps of:(a) radiating an input signal xn(k) of each channel n, n=l, ..., N, as an acoustic signal by a loudspeaker of said each channel, said input signal being inputted as a sample every discrete time k and N being an integer equal to or greater than 2;(b) picking up, by one microphone, said N acoustic signals propagating thereto over N echo paths from said N loudspeakers to obtain an echo y(k);(c) obtaining a combined input signal vector x(k)=[x1 T(k), x2 T(k), ..., xN T(k)]T by combining input signal vectors xn(k) of said N channels, each composed of an input signal sequence xn(k), xn(k-1), ..., xn(k-Ln+1) counted back to time Ln-1 from said discrete time k in said each channel n;(d) generating an estimated echo ŷ(k) by inputting said combined input signal vector x(k) into a combined estimated echo path which has a combined estimated impulse response vector (k) obtained by combining estimated impulse response vectors n(k), n=l, ..., N , of N estimated echo paths corresponding to said N echo paths, each of said N estimated echo paths having Ln taps;(e) obtaining a residual echo e(k) by subtracting said estimated echo signal ŷ(k) from said echo y(k) obtained in said one microphone;(f) extracting a variation in the cross-correlation between said input signals of said N channels on the basis of said residual echo e(k) and said combined input signal vector x(k);(g) generating an adjustment vector v(k) on the basis of said cross-correlation variation;(h) adjusting said combined estimated impulse response vector (k) of said combined estimated echo path by said adjustment vector v(k) to obtain combined estimated impulse vector (k+1) for use at the next time k+1;and(i) repeating said steps (a) to (h) upon each increment of time k.
4
An N-channel echo cancellation method comprising the steps of:(a) radiating an input signal xn(k) of each channel n, n=1, ..., N, as an acoustic signal by a loudspeaker of said each channel, said input signal being inputted as a sample every discrete time k and N being an integer equal to or greater than 2;(b) picking up, by one microphone, said N acoustic signals propagating thereto over N echo paths from said N loudspeakers to obtain an echo y(k);(c) obtaining a combined input signal vector x(k)=[x1 T(k),x2 T(k), ..., xN T(k)]T by combining input signal vectors xn(k) of said N channels, each composed of an input signal sequence xn(k), xn(k-1), ..., xn(k-Ln-1) counted back to time Ln-1 from said discrete time k in each channel;(d) calculating an approximate estimated echo, by a convolution zT(k)x(k), which is obtained by inputting said combined input signal vector x(k) into a combined approximate estimated echo path which has an approximate estimated vector z(k) obtained by combining approximate estimated impulse response vectors zn(k), n=l, ..., N, of N estimated echo paths corresponding to said N echo paths, each of said N estimated echo paths having Ln taps;(e) calculating an auto-correlation vector rp-1(k) of said combined input signal vector x(k);(f) generating a supplementary estimated echo on the basis of said auto-correlation vector rp-1(k) and a smoothing coefficient vector sp-1(k-1) at time k-1;(g) calculating the sum of said approximate estimated echo and said supplementary estimated echo and outputting it as a combined estimated echo ŷ(k) for said echo y(k);(h) calculating a residual echo e(k) by subtracting said combined estimated echo ŷ(k) from said echo y(k) obtained in said one microphone;(i) detecting, as pre-filter coefficients β₁,...,βp, variations in the cross-correlation between said input signals by using current and pervious ones of said combined input signal vector x(k) through the use of said residual echo e(k) and said combined input signal vector x(k);(j) generating a smoothing coefficient vector sp-1(k) by smoothing said pre-filter coefficients β₁,...,βp;(k) adjusting said approximate estimated impulse response vector z(k) on the basis of said smoothing coefficient vector sp-1(k) and said previous combined input signal vector x(k-p+1) to obtainan approximate estimated impulse response vector z(k+1) for use at the next time k+1;and(l) repeating said steps (a) to (k) upon each increment of time k.
7
An echo cancellation method for a multi-channel teleconferencing system, said method comprising the steps of:(a) adding a variation to the cross-correlation between input signals of respective channels;(b) radiating said input signal of each channel added with said variation, as an acoustic signal, by a loudspeaker of said each channel;(c) picking up, by one microphone, said acoustic signals propagating thereto over respective echo paths from said loudspeakers;(d) subtracting an estimated echo from an echo obtained in said one microphone to obtain a residual echo and calculating an adjustment vector from said residual echo and said input signal of said each channel added with said variation;(e) iteratively adjusting the estimation of an impulse response of said each echo path by said adjustment vector;(f) generating respective estimated echo paths each having said adjusted impulse response;and(g) generating said estimated echo by applying the corresponding ones of said input signals to said respective estimated echo paths.
12
An N-channel echo canceller comprising:N loudspeakers for radiating input signals xn(k), n=l, ..., N, as acoustic signals of respective channels, said input signals being inputted as samples every discrete time k and N being an integer equal to or greater than 2;a microphone for picking up said N acoustic signals propagating thereto over N echo paths from said N loudspeakers to obtain en echo y(k);a vector combining part for obtaining a combined input signal vector x(k)=[x1 T(k), x2 T(k), ..., xN T(k)]T by combining input signal vectors xn(k) of said N channels, each composed of an input signal sequence xn(k), xn(k-1), ..., xn(k-Ln+1) counted Ln-1 back from said discrete time k in each of said N channels;an estimated echo generating part which has a combined estimated impulse response vector (k) obtained by combining estimated impulse response vectors (k), n=l, ..., N of N estimated echo paths corresponding to said N echo paths, each of said N estimated echo paths having Ln taps, and into which said combined input signal vector x(k) is inputted to generate an estimated echo ŷ(k);a subtractor for obtaining a residual echo e(k) by subtracting said estimated echo ŷ(k) from said echo y(k) provided from said microphone;a cross-correlation variation extracting part for extracting a variation in the cross-correlation between said input signals of said N channels on the basis of said residual echo e(k) and said combined input signal vector x(k) and for generating an adjustment vector v(k);and an adjusting part for adjusting said combined estimated impulse response vector (k) by said adjustment vector v(k) to obtain a combined estimated response vector (k+1) for use at the next time k+1.
14
An N-channel echo canceller comprising:N loudspeakers for radiating input signals xn(k), n=1, ..., N, as acoustic signals of respective channels, said input signals being inputted as samples every discrete time k and N being an integer equal to or greater than 2;a microphone for picking up said N acoustic signals propagating thereto over N echo paths from said N loudspeakers to obtain an echo y(k);a vector combining part for obtaining a combined input signal vector x(k)=[x1 T(k),xT(k),...,xN T(k)]T by combining input signal vectors xn(k) of said N channels, each composed of an input signal sequence xn(k), xn(k-1), ..., xn(k-Ln+1) counted Ln-1 back from said discrete time k in each of said N channels;an approximate estimated echo path adjusting part which has set therein a combined approximate impulse response vector z(k) obtained by combining approximate estimated impulse response vectors zn(k), n=l,...,N, of N estimated echo paths corresponding to said N echo paths, each of said N estimated echo paths having Ln taps, and into which said combined input signal vector x(k) is inputted to calculate an approximate estimated echo by a convolution zT(k)x(k);a auto-correlation calculating part for calculating an auto-correlation vector rp-1(k) of said combined input signal vector x(k);an adder for calculating the sum of said approximate estimated echo and a supplementary estimated echo and for outputting it as a combined estimated echo ŷ(k) of said N estimated echo paths corresponding to said N echo paths;a subtractor for obtaining a residual echo e(k) by subtracting said combined estimated echo ŷ(k) from said echo y(k) obtained in said microphone;a cross-correlation variation coefficient calculating part for detecting, as pre-filter coefficients β₁, ...,βp, variations in the cross-correlation between current previous ones of said combined input signal vector x(k) through the use of said residual echo e(k) and said combined input signal vector x(k);a smoothing coefficient vector generating part for generating a smoothing coefficient vector sp-1(k) by smoothing said pre-filter coefficients;an approximate estimated echo path adjusting part for adjusting said approximate estimated impulse response vector z(k) on the basis of said smoothing coefficient vector sp-1(k) and a previous one of said previous combined input signal vector x(k-p+1) and for outputting an approximate estimated impulse response vector z(k+1) for use at the next time k+1;and convoluting part for calculating the inner product of said auto-correlation vector rp-1(k) and said smoothing coefficient vector sp-1(k) to obtain said supplementary estimated echo for use at said time k.
18
An echo canceller for a multi-channel teleconferencing system, comprising:a cross-correlation variation adding part to the cross-correlation between input signals of respective channels;loudspeakers of N channels for radiating as acoustic signals said input signals each added with said cross-correlation variation;a microphone for picking up said acoustic signals of said N channels propagating thereto over respective echo paths from said loudspeakers to obtain an echo;a subtractor for obtaining a residual echo by subtracting an estimated echo from said echo obtained in said microphone;a cross-correlation variation extracting part for obtaining an adjustment vector from said residual echo and said cross-correlation variation;an adjusting part for iteratively adjusting the estimation of an impulse response of each of said echo paths by said adjustment vector;and an estimated echo generating part which has said adjusted impulse responses and into which the corresponding to said input signals are inputted to generate said estimated echo.

Independent claims6

68 paragraphs in 4 sections, as filed

BACKGROUND OF THE INVENTION

The present invention relates to a method and apparatus for multi-channel acoustic echo cancellation which cancel a room echo that causes howling and presents a psycho-acoustic problem in a teleconference system provided with a multi-receive-channel system.

With the widespread proliferation of ISDN, LAN and similar digital networks and the developments of high efficiency speech and image coding techniques in recent years, much attention is now focused on a TV conference system which allows participants to communicate with each other while watching the other's face and a multi-media communication system using personal computers or workstations which also enables data communications or the like. These systems adopt in many cases a hands-free communication system using loudspeakers and microphones to thereby create a more realistic teleconferencing environment. However, the hands-free communication system is attended with the problem of echo and howling generation; the acoustic echo canceling technique is indispensable to avoiding the problem. In fact, acoustic echo cancellers are widely used, but they are mostly for one-channel use and can cancel only an acoustic echo from one (one channel) loudspeaker to one (one channel) microphone. At present, stereo is not uncommon in many TV broadcast programs and music media, for instance, and there is also a growing demand for a multi-channel hands-free communication system. To meet this requirement, it is necessary to implement a multi-channel acoustic echo canceller which permits cancellation of acoustic echoes from a plurality of speakers (a plurality of channels) to a microphone; in recent years, technical problems and solutions thereto have been investigated very actively toward the realization of such a multi-channel echo canceller.

A description will be given, with reference to Fig. 1, of a one-channel echo canceller. In hands-free communication, speech uttered by a person at a remote place is provided as a received signal to a received signal terminal 11 and is radiated from a loudspeaker 12. In Fig. 1, k indicates discrete time and x(k) a received signal sample value. In the following description, x(k) will be referred to simply as a received signal. An echo canceller 14 cancels an echo y(k) which is produced when the received signal x(k) radiated from the loudspeaker 12 is picked up by a microphone 16 after propagating over an echo path 15. The echo y(k) can be modeled by such a convolution as follows:<maths id="math0001" num="(1)"><math display="block"><mrow><mtext>y(k) = </mtext><apply><sum /><lowlimit><mtext>n=0</mtext></lowlimit><uplimit><mtext>L-1</mtext></uplimit><mrow><mtext>h(k,n)x(k-n)</mtext></mrow></apply></mrow></math><img file="EP0709999A2_D0001.tif" /></maths> where h(k,n) is the impulse response indicating the transfer function of the echo path 15 at time k and L is the number of taps, which is a constant preset corresponding to the reverberation time of the echo path 15. At first, received signals x(k) from the current time to L-1 are stored in a received signal storage and vector generating part 17. The L received signals thus stored are outputted as a received signal vector x(k), that is, as<maths id="math0002" num="(2)"><math display="block"><mrow><mtext>x</mtext><msup><mrow><mtext>(k)=[x(k), x(k-1), ..., x(k-L+1)]</mtext></mrow><mrow><mtext>T</mtext></mrow></msup></mrow></math><img file="EP0709999A2_D0002.tif" /></maths> where *T indicates a transposition. In an estimated echo generating part 18 formed by an FIR filter, the inner product of the received signal vector x(k) of Eq. (2) and an estimated echo path vector <img file="EP0709999A2_D0003.tif" /><img file="EP0709999A2_D0004.tif" />(k), which is a filter coefficient vector provided from an echo path estimating part 19, is calculated as follows:<maths id="math0003" num=""><img file="EP0709999A2_D0005.tif" /></maths> As a result, an estimated echo or echo replica ŷ(k) is generated. This inner product calculation is equivalent to such a convolution as Eq. (1). In the echo path estimating part 19, the estimated echo path vector <img file="EP0709999A2_D0006.tif" />(k) is generated which is used in the estimated echo generating part 18. The most common algorithm which is used for the echo path estimation is an NLMS (Normalized Least Mean Square) algorithm. With the NLMS algorithm, the received signal vector x(k) at time k and a residual echo e(k), i.e. the following error, obtained by subtracting the estimated echo signal ŷ(k) from the output y(k) of the microphone 16 by a subtractor 21,<maths id="math0004" num="(4)"><math display="block"><mrow><mtext>e(k) = y(k) - </mtext><mover accent="true"><mrow><mtext>y</mtext></mrow><mo>^</mo></mover><mtext>(k)</mtext></mrow></math><img file="EP0709999A2_D0007.tif" /></maths> are used to calculate an estimated echo path vector <img file="EP0709999A2_D0008.tif" />(k+1) which is used at time k+1, by the following equation:<maths id="math0005" num=""><img file="EP0709999A2_D0009.tif" /></maths> where α is called a step size parameter, which is used to adjust adaptation within the range of 0<α<2. By repeating the above processing upon each increment of time k, the estimated echo path vector <img file="EP0709999A2_D0010.tif" />(k) in the echo path estimating part 19 gradually converges toward an true echo path vector h(k) whose elements are impulse response sequences h(k, n), (where n=0, 1, 2, ..., L-1) of the true echo path 15 at time k, that is, toward the following echo path vector:<maths id="math0006" num="(6)"><math display="block"><mrow><mtext>h</mtext><msup><mrow><mtext>(k)=[h(k,0), h(k,1), ..., h(k,L-1)]</mtext></mrow><mrow><mtext>T</mtext></mrow></msup></mrow></math><img file="EP0709999A2_D0011.tif" /></maths> As the result of this, the residual echo signal e(k) given by Eq. (4) can be reduced.

In general, a teleconferencing system of the type having an N (≧2) channel loudspeaker system and an M (≧1) channel microphose system employs, for echo cancellation, such a configuration as shown in Fig. 2. That is to say, an echo cancellation system 23 is composed of N-channel echo cancellers 22₁, 22₁, ..., 22M for processing N-input-one-output time sequence signals, which are each interposed between all of N channels of the receiving side where received signals xl(k) to xN(k) are provided to input terminals lll to llN and are radiated from loudspeakers 12l to 12N and one channel of the sending side composed of microphones 16l to 16M. In this instance, the echo cancellation system has a total of N×M echo path 15nm (1≦n≦ N, 1≦m≦M). The N-channel echo cancellers 22l, 22₂, ..., 22M, which are each connected between all of the N channels of the receiving side and one channel of the sending side, have such a configuration as shown in Fig 3, which is an extended version of the configuration of the echo canceller 14 depicted in Fig. 1. This is described in detail, for example, in B. Widow and S. D. Stearns, "Adaptive signal processing," Prince-Hall, Inc. pp. 198-200 (1985). Now, consider the N-channel echo canceller 22m connected to an m-th channel (1≦m≦M) of the sending side. The echo signal ym(k) that is picked up the m-th channel microphone 16m is obtained by adding together respective received signals of all channels at the sending side after propagation over respective echo paths 15lm to 15Nm. Hence, it is necessary to devise how to make the echo path estimation by evaluating only one residual echo em(k) in common to all the receiving side channels. At first, for the received signal xn(k) of each channel, the following received signal vectors are generated in the received signal storage and vector generating parts (17₁, 17₁, ..., 17N):<maths id="math0007" num=""><img file="EP0709999A2_D0012.tif" /></maths> where L₁, L₂, ..., LN are the numbers of taps, which are constants preset corresponding to reverberation times of the respective echo paths 151m, 151m, ..., 15Nm. The vectors thus generated are combined in a vector combining part 24 as follows:<maths id="math0008" num="(8)"><math display="block"><mrow><mtext>x</mtext><mtext>(k)=[</mtext><msub><mrow><mtext>x</mtext></mrow><mrow><mtext>1</mtext></mrow></msub><msup><mrow><mtext> </mtext></mrow><mrow><mtext>T</mtext></mrow></msup><mtext>(k), </mtext><msub><mrow><mtext>x</mtext></mrow><mrow><mtext>2</mtext></mrow></msub><msup><mrow><mtext> </mtext></mrow><mrow><mtext>T</mtext></mrow></msup><mtext>(k), ..., </mtext><msub><mrow><mtext>x</mtext></mrow><mrow><mtext>N</mtext></mrow></msub><msup><mrow><mtext> </mtext></mrow><mrow><mtext>T</mtext></mrow></msup><msup><mrow><mtext>(k)]</mtext></mrow><mrow><mtext>T</mtext></mrow></msup></mrow></math><img file="EP0709999A2_D0013.tif" /></maths> Also in the echo path estimating part 19m, estimated echo path vectors <img file="EP0709999A2_D0014.tif" />1m(k), <img file="EP0709999A2_D0015.tif" />2m(k), ..., <img file="EP0709999A2_D0016.tif" />Nm(k), which are used to simulate N echo paths between the respective receiving side channels and the m-th sending side channel, are combined as follows:<maths id="math0009" num=""><img file="EP0709999A2_D0017.tif" /></maths> In the case of using the NLMS algorithm, the updating of the combined estimated echo path vector <img file="EP0709999A2_D0018.tif" />m(k) is done as follows:<maths id="math0010" num=""><img file="EP0709999A2_D0019.tif" /></maths> In the estimated echo generating part 18m, an estimated echo ŷm(k) for the echo ym(k) picked up in the m-th sending channel is generated by the following inner product calculation:<maths id="math0011" num=""><img file="EP0709999A2_D0020.tif" /></maths> By combining vectors in the respective channels into one vector, the flow of basic processing becomes the same as in the one-channel echo canceller of Fig. 1.

Of the defects of the conventional echo cancellation system for application to the teleconferencing system composed of an N-channel speaker system and an M-channel microphone system, the defect that the present invention is to solve will be described in connection with a two-channel stereo teleconferencing system. Referring now to Fig. 4, speaker/microphone systems at points A and B which are connected via a network NW respectively comprise two speakers 12a₁, 12a₂ and 12b₁, 12b₂ and two microphones 16a₁, 16a₂ and 16b₁, 16b₂. Between two receiving channels and each sending channel there are provided echo cancellers 22a₁, 22a₂ and 22b₁, 22b₂.

In the case of applying the conventional echo canceller to the stereo teleconferencing system which sends and receives signals between the points A and B over two channels, there is presented a problem that each time a speaker at the point A moves or changes to another, an echo from the point B by the speech at the point A increases even if the echo paths 15₁₁ and 15₂₁ remain unchanged. The reason for this is that the echo path impulse responses are not correctly estimated in the echo cancellation system at the Point B side.

To explain this problem, attention is paid to the operation of the echo canceller 22b₁ connected to a first sending channel of the echo cancellation system at the point B. Let two-channel received signal vectors be represented as xl(k) and x₂(k). Letting echo path vectors of true echo paths 15₁₁ and 15₂₁ of the respective receiving channels be represented as h₁₁(k) and h₂₁(k), respectively, an echo y₁(k) that is picked up by the microphone 16b1 is given by<maths id="math0012" num="(12)"><math display="block"><mrow><msub><mrow><mtext>y</mtext></mrow><mrow><mtext>1</mtext></mrow></msub><mtext>(k)=</mtext><msub><mrow><mtext>h</mtext></mrow><mrow><mtext>11</mtext></mrow></msub><msup><mrow><mtext> </mtext></mrow><mrow><mtext>T</mtext></mrow></msup><mtext>(k)</mtext><msub><mrow><mtext>x</mtext></mrow><mrow><mtext>1</mtext></mrow></msub><mtext>(k)+</mtext><msub><mrow><mtext>h</mtext></mrow><mrow><mtext>21</mtext></mrow></msub><msup><mrow><mtext> </mtext></mrow><mrow><mtext>T</mtext></mrow></msup><mtext>(k)</mtext><msub><mrow><mtext>x</mtext></mrow><mrow><mtext>2</mtext></mrow></msub><mtext>(k)</mtext></mrow></math><img file="EP0709999A2_D0021.tif" /></maths> On the other hand, an estimated echo ŷ₁(k) that is generated in the echo canceller is expressed by the following equation using estimated echo path vectors <img file="EP0709999A2_D0022.tif" />₁₁(k) and <img file="EP0709999A2_D0023.tif" />₂₁(k) that are generated in the echo canceller:<maths id="math0013" num=""><img file="EP0709999A2_D0024.tif" /></maths> When one speaker speaks or utters at the point A, the received signal vectors x₁(k) and x₂(k)have a very strong cross-correlation. When the received signal vectors x₁(k) and x₂(k) have a constant high cross-correlation, the combined vector {<img file="EP0709999A2_D0025.tif" />₁₁T(k), <img file="EP0709999A2_D0026.tif" />₂₁T(k)} as the solution of the following equation (14) exists infinitely, forming a subspace Hx inherent in the cross-correlation between the received signal vectors x₁(K) and x₂(k).<maths id="math0014" num="(14)"><math display="block"><mrow><msub><mrow><mover accent="true"><mrow><mtext>y</mtext></mrow><mo>^</mo></mover></mrow><mrow><mtext>1</mtext></mrow></msub><msub><mrow><mtext>(k)=y</mtext></mrow><mrow><mtext>1</mtext></mrow></msub><mtext>(k)</mtext></mrow></math><img file="EP0709999A2_D0027.tif" /></maths> On this account, in the case of using an ordinary iterative error minimisation algorithm such as the NLMS algorithm, the combined vector {<img file="EP0709999A2_D0028.tif" />₁₁T(k), <img file="EP0709999A2_D0029.tif" />₂₁T(K)} converges to a point in subspace Hx nearest the initial point; in general, it does not converge to the true value {h₁₁T(k), h₂₁T(k)}.

For simplicity, consider the case where the received signal vectors x₁(k) and x₁(k) are expressed by constant scalar values p₁ and p₂ and the source signal vector s(k) as follows:<maths id="math0015" num="(15)"><math display="block"><mrow><msub><mrow><mtext>x</mtext></mrow><mrow><mtext>1</mtext></mrow></msub><msub><mrow><mtext>(k)=p</mtext></mrow><mrow><mtext>1</mtext></mrow></msub><mtext>s</mtext><mtext>(k), </mtext><msub><mrow><mtext>x</mtext></mrow><mrow><mtext>2</mtext></mrow></msub><msub><mrow><mtext>(k)=p</mtext></mrow><mrow><mtext>2</mtext></mrow></msub><mtext>s</mtext><mtext>(k)</mtext></mrow></math><img file="EP0709999A2_D0030.tif" /></maths> The subspace Hx where [<img file="EP0709999A2_D0031.tif" />₁₁T(k), <img file="EP0709999A2_D0032.tif" />₂₁T(k)] is allowed to exist can be regarded as a straight line A1 in Fig. 5A which satisfies the following equation:<maths id="math0016" num=""><img file="EP0709999A2_D0033.tif" /></maths> When the adaptation starts from the initial value (0, 0), the steady-state solution [<img file="EP0709999A2_D0034.tif" />11pT(k), <img file="EP0709999A2_D0035.tif" />21pT(k)] is obtained as follows:<maths id="math0017" num=""><img file="EP0709999A2_D0036.tif" /></maths> Hence, Eq. (16) is no longer satisfied when the rate between the scalar values p₁ and p₂ varies, with the result that no echo can be canceled and the echo increases accordingly.

As above, in the case of applying the conventional echo canceller to the teleconferencing system composed of an N-channel loudspeaker system and an M-channel microphone system, if received signals of the respective channels have a cross-correlation, the echo path impulse resoponses cannot correctly be estimated--this poses the problem that the echo increases whenever the cross-correlation between the received signals varies.

SUMMARY OF THE INVENTION

It is therefore an object of the present invention to provide a method and apparatus for echo cancellation for a multi-channel teleconferencing system which enable echo paths to be correctly estimated regardless of the cross-correlation between multi-channel received signals, and hence permit of echoes even if the cross-correlation changes.

In one aspect, the present invention is directed to an N-channel echo cancellation method which comprises the steps of: <ul id="ul0001" list-style="none" compact="compact"><li>(a) radiating an input signal xn(k) of each channel n (n=1, ..., N) as an acoustic signal by a loudspeaker of each channel, said input signal being inputted as a sample every discrete time k and N being an integer equal to or greater than 2;</li><li>(b) picking up said N acoustic signals propagating through N echo paths from said N loudspeakers to one microphone to obtain an echo y(k);</li><li>(c) obtaining a combined input signal vector <maths id="math0018" num=""><math display="inline"><mrow><mtext>x</mtext><mtext>(k)=[</mtext><msub><mrow><mtext>x</mtext></mrow><mrow><mtext>1</mtext></mrow></msub><msup><mrow><mtext> </mtext></mrow><mrow><mtext>T</mtext></mrow></msup><mtext>(k), </mtext><msub><mrow><mtext>x</mtext></mrow><mrow><mtext>2</mtext></mrow></msub><msup><mrow><mtext> </mtext></mrow><mrow><mtext>T</mtext></mrow></msup><mtext>(k), ..., </mtext><msub><mrow><mtext>x</mtext></mrow><mrow><mtext>N</mtext></mrow></msub><msup><mrow><mtext> </mtext></mrow><mrow><mtext>T</mtext></mrow></msup><msup><mrow><mtext>(k)]</mtext></mrow><mrow><mtext>T</mtext></mrow></msup></mrow></math><img file="EP0709999A2_D0037.tif" /></maths> by combining input signal vectors xn(k) of said N channels, each composed of input signal sequences xn(k), xn(k-1), ..., xn(k-Ln+1) back to time Ln-1 in each channel n;</li><li>(d) generating an estimated echo ŷ(k) by inputting said combined input signal vector x(k) into an combined estimated echo path which has a combined estimated impulse response vector <img file="EP0709999A2_D0038.tif" />(k) obtained by combining estimated impulse response vectors <img file="EP0709999A2_D0039.tif" />n(k) (n=1, ..., N) of N echo paths estimated for said N echo paths, each of said N estimated echo paths having Ln taps;</li><li>(e) obtaining a residual echo e(k) by subtracting said estimated echo signal y(k) from said echo y(k) picked up by said one microphone;</li><li>(f) extracting a variation in the cross-correlation between said input signals of said respective channels by using said residual echo e(k) and said combined input signal vector x(k);</li><li>(g) generating an adjustment vector v(k) on the basis of said variation in said cross-correlation;</li><li>(h) adjusting said combined estimated impulse response vector <img file="EP0709999A2_D0040.tif" />(k) of said combined estimated echo path by using said adjustment vector v(k) to obtain a combined estimated impulse response vector <img file="EP0709999A2_D0041.tif" />(k+1) at the next time k+1; and</li><li>(i) repeating said steps (a) to (h) upon each increment of time k.</li></ul>

In another aspect, the present invention is directed to an N-channel echo cancellation method which comprises the steps of: <ul id="ul0002" list-style="none" compact="compact"><li>(a) radiating an input signal xn(k) of each channel n (n=1, ..., N) as an acoustic signal by a loudspeaker of each channel, said input signal being inputted as a sample every discrete time k and N being an integer equal to or greater than 2;</li><li>(b) picking up said N acoustic signals propagating through N echo paths from N loudspeakers to one microphone to obtain an echo y(k);</li><li>(c) obtaining a combined input signal vector <maths id="math0019" num=""><math display="inline"><mrow><mtext>x</mtext><mtext>(k)=[</mtext><msub><mrow><mtext>x</mtext></mrow><mrow><mtext>1</mtext></mrow></msub><msup><mrow><mtext> </mtext></mrow><mrow><mtext>T</mtext></mrow></msup><mtext>(k), </mtext><msub><mrow><mtext>x</mtext></mrow><mrow><mtext>2</mtext></mrow></msub><msup><mrow><mtext> </mtext></mrow><mrow><mtext>T</mtext></mrow></msup><mtext>(k), ..., </mtext><msub><mrow><mtext>x</mtext></mrow><mrow><mtext>N</mtext></mrow></msub><msup><mrow><mtext> </mtext></mrow><mrow><mtext>T</mtext></mrow></msup><msup><mrow><mtext>(k)]</mtext></mrow><mrow><mtext>T</mtext></mrow></msup></mrow></math><img file="EP0709999A2_D0042.tif" /></maths> by combining input signal vectors xn(k) of said N channels, each composed of input signal sequences xn(k), xn(k-1), ..., xn(k-Ln+1) back to time Ln-1 in each channel;</li><li>(d) calculating an estimated approximate echo by a convolution zT(k)x(k) and inputting said combined input signal vector x(k) into a combined estimated approximate echo path which has a combined estimated approximate impulse response vector z(k) obtained by combining estimated approximate impulse response vectors zn(k) (n=1, ..., N) of N estimated echo paths for said N echo paths, each of said N estimated echo paths having Ln taps;</li><li>(e) calculating an auto-correlation vector rp-1(k) of said combined input signal vector x(k);</li><li>(f) generating a supplementary estimated echo from said auto-correlation vector rp-1(k) and a smoothing coefficient vector sp-1(k-1) at time k-1;</li><li>(g) calculating the sum of said estimated approximate echo and said supplementary estimated approximate echo and outputting it as an estimated echo y(k) for said echo;</li><li>(h) calculating a residual echo e(k) by subtracting said estimated echo y(k) from said echo y(k) picked up by said one microphone;</li><li>(i) detecting, as pre-filter coefficients β₁, ..., βp, variations in the cross-correlation between said input signals by using current and previous ones of said residual echo e(k) and said combined input signal vector x(k);</li><li>(j) generating a smoothing coefficient vector sp-1(k) by smoothing said pre-filter coefficients;</li><li>(k) adjusting said estimated approximate impulse response z(k) on the basis of said smoothing coefficient vector sp-1(k) and said previous combined input signal vector x(k-p+1) to obtain z(k+1); and</li><li>(l) repeating said steps (a) to (k) upon each increment of time k.</li></ul>

In still another aspect, the present invention is directed to an echo cancellation method for a multi-channel teleconferencing system, which comprises the steps of: <ul id="ul0003" list-style="none" compact="compact"><li>(a) adding a variation to the cross-correlation between input signals of respective channels;</li><li>(b) radiating, as an acoustic signal, said input signal of each channel added with said variation, by a loudspeaker for each channel;</li><li>(c) picking up said acoustic signal propagating through each echo path from said loudspeaker to a microphone;</li><li>(d) subtracting an estimated echo from said microphone and calculating an adjustment vector on the basis of the resulting residual echo and said input signal of said each channel added with said variation;</li><li>(e) iteratively adjusting the estimation of impulse response of said each echo path by using said adjustment vector;</li><li>(f) generating estimated echo paths each having said adjusted impulse response; and</li><li>(g) generating said estimated echo by applying a corresponding one of said input signals to each of said estimated echo paths.</li></ul>

The echo cancellers according to the present invention have configurations for performing the steps of the first-, second- and third-mentioned methods, respectively.

BRIEF DESCRIPTION OF THE DRAWINGS

<ul id="ul0004" list-style="none" compact="compact"><li>Fig. 1 is a block diagram showing a one-channel echo canceller;</li><li>Fig. 2 is a block diagram showing a multi-channel echo canceller system;</li><li>Fig. 3 is a block diagram showing a conventional multi-channel echo canceller;</li><li>Fig. 4 is a block diagram showing a stereo teleconferencing system;</li><li>Fig. 5A is a graph showing an echo path estimating operation when received signals have a cross-correlation;</li><li>Fig. 5B is a graph showing an echo path estimating operation when the cross-correlation varies;</li><li>Fig. 6 is a block diagram illustrating the construction of a multi-channel echo canceller embodying the first-mentioned method of the present invention;</li><li>Fig. 7 is a block diagram illustrating the construction of a cross-correlation variation extracting part 31 in Fig. 6;</li><li>Fig. 8 is a block diagram illustrating an example of the construction of a cross-correlation variation coefficient calculating part 31A in Fig. 7;</li><li>Fig. 9 is a block diagram illustrating an embodiment in which a cross-correlation variation emphasis vector generating part 31B, an adjusting part 32 and an estimated echo generating part 18m in Fig. 7 are formed by a simplified estimated echo generating part 180m using approximate echo path vector;</li><li>Fig. 10 is a block diagram illustrating the construction of the simplified estimated echo generating part 180m in Fig. 9;</li><li>Fig. 11 is a block diagram illustrating the configuration of a multi-channel echo canceller embodying the third-mentioned method of the present invention;</li><li>Fig. 12 is a block diagram illustrating an example of a digital filter for implementing the third-mentioned method of the present invention;</li><li>Fig. 13 is a graph showing the convergence of the estimated echo path vector according to the first- or second-mentioned method of the present invention;</li><li>Fig. 14 is a graph showing convergence characteristic of the estimated echo path vector according to the first- or second-mentioned method of the present invention, with the order p of a projection algorithm used as a parameter;</li><li>Fig. 15 is a block diagram illustrating a teleconferencing system which provides a sound image localization function in four-point communications;</li><li>Fig. 16 is a graph showingthe comparison between a cross-correlation extracting method and a conventional method in terms of the magnitude of an error vector between a true echo path and an estimated one in the case of fixed cross-correlation stereo signals being received; and</li><li>Fig. 17 is a graph comparing error vectors by a correlation variation extracting method embodying the third-mentioned method of the present invention and a conventional extracting method.</li></ul>

DESCRIPTION OF THE PREFERRED EMBODIMENTS

The first- or second-mentioned method of the present invention will be described concretely. A description will be given first of the operation of the echo path estimating part of the multi-channel echo canceller in the case where the cross-correlation between multi-channel received signals varies.

For brevity, consider that as a variation in the cross-correlation between the two-channel stereo received signals x₁(k) and x₂(k) expressed by Eq. (15), the scalar values p₁ and p₂ in Eq. (15) change to q₁ and q₂,where the ratio of q₁:q₂ is different from that of p₁:p₂. The estimated echo path vector {<img file="EP0709999A2_D0043.tif" />₁₁T(k), <img file="EP0709999A2_D0044.tif" />₂₁T(k)} converges first to {<img file="EP0709999A2_D0045.tif" />11pT(k), <img file="EP0709999A2_D0046.tif" />21pT(k)} of Eqs. (17) and (18). Next, when the scalar values p₁ and p₂ change to q₁ and q₂, the estimated echo path vector converges to a point {<img file="EP0709999A2_D0047.tif" />11qT(k), <img file="EP0709999A2_D0048.tif" />21qT(k)} of the shortest distance from an initial value {<img file="EP0709999A2_D0049.tif" />11pT(k), <img file="EP0709999A2_D0050.tif" />21pT(k)}. This operation is interpreted geometrically as shown in Fig. 5B. Here, the steady-state solution {<img file="EP0709999A2_D0051.tif" />11qT, <img file="EP0709999A2_D0052.tif" />21qT} is the intersection of a straight line<maths id="math0020" num=""><img file="EP0709999A2_D0053.tif" /></maths> and its perpendicular line which includes {h11pT,h21pT}. Accordingly, it is obvious that the norm of filter coefficient error vectors ep and eq generally bear a relationship ∥ep∥>∥eq∥. This means that upon each variation in the cross-correlation between the received signals, the norm of the filter coefficient error vector becomes smaller. That is to say, the variation in the cross-correlation between the multi-channel received signals causes an increase in echo when the echo path estimation is incorrect in the echo canceller, but it can be considered as effective information that permits estimation of the true echo path in the echo path estimating part after the variation in the cross-correlation repeats an infinite number of times. In the past, since such a fact has not been noticed, the cross-correlation between multi-channel received signals has been treated as fixed; therefore, it could have been not supposed that the norm of the error vector becomes small.

As described previously, the present invention extracts and utilizes information on the variation in the cross-correlation between received signals of respective channels. In the present invention, the multi-channel received signal vectors are combined into a single vector. On this account, the variation in the cross-correlation between the received signals of the respective channels appears in the auto-correlation which is the correlation between the current combined received signal vector and the past combined received signal vector. Hence, to remove an unvaried component of the cross-correlation of the received signal of each channel to extract therefrom a varied component of the cross-correlation, it is possible to use a scheme that reduces the auto-correlation of the combined received signal vector, that is, removes its auto-correlation.

This can be done by a decorrelation scheme using a projection algorithm, RLS algorithm or linear prediction, which are applied to the conventional one-channel echo canceller. Of these schemes, the scheme using linear prediction is to linearly predict a speech signal and remove its auto-correlation by using the linear predictive coefficients. The RLS algorithm can be regarded as a specific form of the projection algorithm. Thus, the projection algorithm will be described below as an example of the most practical decorrelation scheme.

The projection algorithm has been proposed as a scheme for improving the convergence speed of a speech or similar signal having a high auto-correlation. The projection algorithm is described in detail in Ozeki and Umeda, "An Adaptive Filtering Algorithm Using an Orthogonal Projection to an Affine Subspace and Its Properties," Journal of Institute of Electronics, Information and Communication Engineers of Japan (A), J67-App. 126-132, (1984-2).

In general, a p-th order projection algorithm for multi-channel inputs adjusts or corrects the estimated echo path vector <img file="EP0709999A2_D0054.tif" />(k) in such a manner as to obtain correct outputs y(k), y(k-1), ..., y(k-p+1) with respect to p previous input signal vectors x(k), x(k-1), ..., x(k-p+1). That is to say, an estimated echo path vector h(k+1) is calculated which satisfies the following conditions:<maths id="math0021" num=""><img file="EP0709999A2_D0055.tif" /></maths> where<maths id="math0022" num=""><math display="block"><mrow><mtext>x</mtext><mtext>(k)=[</mtext><msub><mrow><mtext>x</mtext></mrow><mrow><mtext>1</mtext></mrow></msub><msup><mrow><mtext> </mtext></mrow><mrow><mtext>T</mtext></mrow></msup><mtext>(k), </mtext><msub><mrow><mtext>x</mtext></mrow><mrow><mtext>2</mtext></mrow></msub><msup><mrow><mtext> </mtext></mrow><mrow><mtext>T</mtext></mrow></msup><mtext>(k), ..., </mtext><msub><mrow><mtext>x</mtext></mrow><mrow><mtext>N</mtext></mrow></msub><msup><mrow><mtext> </mtext></mrow><mrow><mtext>T</mtext></mrow></msup><msup><mrow><mtext>(k)]</mtext></mrow><mrow><mtext>T</mtext></mrow></msup></mrow></math><img file="EP0709999A2_D0056.tif" /></maths> and<maths id="math0023" num=""><img file="EP0709999A2_D0057.tif" /></maths> When the number p of the equation is smaller than the number of unknowns (the number of taps), <maths id="math0024" num=""><math display="inline"><mrow><msub><mrow><mtext>L=L</mtext></mrow><mrow><mtext>1</mtext></mrow></msub><msub><mrow><mtext>+L</mtext></mrow><mrow><mtext>2</mtext></mrow></msub><msub><mrow><mtext>+ ... L</mtext></mrow><mrow><mtext>N</mtext></mrow></msub></mrow></math><img file="EP0709999A2_D0058.tif" /></maths>, the solution <img file="EP0709999A2_D0059.tif" />(k+1) of the simultaneous equations of Eq. (19) is indefinite. Then, the estimated echo path vector is adjusted or corrected in such a manner as to minimize the magnitude of adjustment ∥ <img file="EP0709999A2_D0060.tif" /> (k+1)- <img file="EP0709999A2_D0061.tif" /> (k)∥. The p-th order projection algorithm is expressed by the following equation:<maths id="math0025" num=""><img file="EP0709999A2_D0062.tif" /></maths> where<maths id="math0026" num=""><math display="block"><mrow><mtable><mtr><mtd><mrow><mtable><mtr><mtd><mrow><mtable><mlabeledtr><mtext>(22)</mtext><mtd><mrow><mtext>X</mtext><mtext>(k)=[</mtext><mtext>x</mtext><mtext>(k), x(k-1), ..., </mtext><mtext>x</mtext><mtext>(k-p+1)]</mtext></mrow></mtd></mlabeledtr></mtable></mrow></mtd></mtr><mtr><mtd><mrow><mtable><mlabeledtr><mtext>(23)</mtext><mtd><mrow><msub><mrow><mtext>e</mtext></mrow><mrow><mtext>p</mtext></mrow></msub><msup><mrow><mtext>(k)={e(k), (1-α)e(k-1), ..., (1-α)</mtext></mrow><mrow><mtext>p-1</mtext></mrow></msup><msup><mrow><mtext>e(k-p+1)]</mtext></mrow><mrow><mtext>T</mtext></mrow></msup></mrow></mtd></mlabeledtr></mtable></mrow></mtd></mtr><mtr><mtd><mrow><mtable><mlabeledtr><mtext>(24)</mtext><mtd><mrow><msub><mrow><mtext>e</mtext></mrow><mrow><mtext>p</mtext></mrow></msub><mtext>(k)=y(k)-</mtext><mover accent="true"><mrow><mtext>y</mtext></mrow><mo>^</mo></mover><mtext>(k) </mtext></mrow></mtd></mlabeledtr></mtable></mrow></mtd></mtr></mtable></mrow></mtd></mtr></mtable></mrow></math><img file="EP0709999A2_D0063.tif" /></maths><maths id="math0027" num=""><img file="EP0709999A2_D0064.tif" /></maths><maths id="math0028" num="(26)"><math display="block"><mrow><msub><mrow><mtext>g</mtext></mrow><mrow><mtext>p</mtext></mrow></msub><msub><mrow><mtext>(k)=[β</mtext></mrow><mrow><mtext>1</mtext></mrow></msub><msub><mrow><mtext>, β</mtext></mrow><mrow><mtext>2</mtext></mrow></msub><msub><mrow><mtext>, ..., β</mtext></mrow><mrow><mtext>p</mtext></mrow></msub><msup><mrow><mtext>]</mtext></mrow><mrow><mtext>T</mtext></mrow></msup></mrow></math><img file="EP0709999A2_D0065.tif" /></maths><dl id="dl0001" compact="compact"><dt>+:</dt><dd>generalized inverse matrix</dd><dt>-1:</dt><dd>inverse matrix.</dd></dl> In the above, gp(k) is a solution of the following simultaneous linear equations with p unknowns<maths id="math0029" num="(27)"><math display="block"><mrow><mtext>[</mtext><msup><mrow><mtext>X</mtext></mrow><mrow><mtext>T</mtext></mrow></msup><mtext>(k)</mtext><mtext>X</mtext><mtext>(k)]</mtext><msub><mrow><mtext>g</mtext></mrow><mrow><mtext>p</mtext></mrow></msub><mtext>(k)=</mtext><msub><mrow><mtext>e</mtext></mrow><mrow><mtext>p</mtext></mrow></msub><mtext>(k)</mtext></mrow></math><img file="EP0709999A2_D0066.tif" /></maths> and it indicates coefficients of the variation in the cross-correlation between the received signal vectors x₁(k) to xN(k)by changes in time k. The second terms having the coefficient α on the right side of Eq. (21) is the adjustment vector, with which the estimated echo path vector is iteratively updated. For the projection algorithm, there have been proposed several fast calculation methods intended to reduce the computational complexity; such methods are described in detail in Japanese Pat. Appln Nos. 102419/94 and 184742/93. The case where the input/output at the minus time is made zero and p is set at infinity corresponds to the RLS algorithm.

In the case of adjusting the estimated echo path vector <img file="EP0709999A2_D0067.tif" />(k) by the conventional NLMS algorithm based on Eq. (10), it is adjusted in the direction of the input signal vector x(k). On the other hand, according to the first method of the present invention, the second term on the right side of the fourth equation of Eq. (21) by the projection algorithm is set as follows:<maths id="math0030" num="(28)"><math display="block"><mrow><mtext>v</mtext><msub><mrow><mtext>(k)=β</mtext></mrow><mrow><mtext>1</mtext></mrow></msub><mtext>x</mtext><msub><mrow><mtext>(k)+β</mtext></mrow><mrow><mtext>2</mtext></mrow></msub><mtext>x</mtext><msub><mrow><mtext>(k-1)+...+β</mtext></mrow><mrow><mtext>p</mtext></mrow></msub><mtext>x</mtext><mtext>(k-p+1)</mtext></mrow></math><img file="EP0709999A2_D0068.tif" /></maths> and the estimated echo path vector <img file="EP0709999A2_D0069.tif" />(k) is adjusted in the direction of the vector v(k), that is, in the direction in which the correlation to all of p-1 previous combined input signal vectors x(k-1), ..., x(k-p+1) has been removed from the combined vector x(k) of the multi-channel input signals. In other words, the coefficients β₁ to βp are determined so that the vector v(k) becomes a signal having removed similar previous combined input signal vectors from the combined input signal vector. Such coefficients β₁ to βp can be obtained by calculating the vector gp(k) which satisfies Eq. (27). While αv(k) is used as the adjustment vector in Eq. (21), v(k) will hereinafter be referred to as an adjustment vector. When the combined received signal vector x(k) is a combined vector of multi-channel signal vectors having a high cross-correlation, the vector v(k) of Eq. (28) becomes a vector having the variation in the cross-correlation emphasized, with the non-varying component removed.

As described previously, the prior art does not recognize that when the cross-correlation between multi-channel received signals varies, the norm of the error vector of the estimated filter coefficients (the estimated echo path vector) becomes small; only an increase in echo is recognized in the prior art. Hence, the model is simplified and it is assumed that a state change which causes the variation in the cross-correlation, such as the shift or change of the speaker, does not occur. Accordingly, there has not been recognized in the prior art an essential difference between the NLMS algorithm and the projection algorithm as to whether the true transfer function of the echo path can be estimated or not when they are applied to the multi-channel echo canceller; hence, the effectiveness of the projection algorithm in the estimation of the true echo path has not been recognized. According to there present invention, attention is directed to the case where the cross-correlation varies, and the variation is extracted for use as the adjustment vector--this has clarified that the estimated transfer function of the estimated echo path can be made to converge to the true transfer function of the echo path at high speed.

In Eq. (27), XpT(k)Xp(k), which is a matrix with p rows and p columns, is a p-th order covariance matrix or auto-correlation matrix and will hereinafter be identified as follows:<maths id="math0031" num="(29)"><math display="block"><mrow><msub><mrow><mtext>R</mtext></mrow><mrow><mtext>p</mtext></mrow></msub><mtext>(k)=</mtext><msub><mrow><mtext>X</mtext></mrow><mrow><mtext>p</mtext></mrow></msub><msup><mrow><mtext> </mtext></mrow><mrow><mtext>T</mtext></mrow></msup><mtext>(k)</mtext><msub><mrow><mtext>X</mtext></mrow><mrow><mtext>p</mtext></mrow></msub><mtext>(k)</mtext></mrow></math><img file="EP0709999A2_D0070.tif" /></maths> The coefficients β₁, β₂, ..., βp in Eq. (28) will hereinafter be referred to as pre-filter coefficients, and a vector using them as elements is expressed by the following equation (30) and will be referred to as pre-filter coefficient vector.<maths id="math0032" num="(30)"><math display="block"><mrow><msub><mrow><mtext>g</mtext></mrow><mrow><mtext>p</mtext></mrow></msub><msub><mrow><mtext>(k)=[β</mtext></mrow><mrow><mtext>1</mtext></mrow></msub><msub><mrow><mtext>, β</mtext></mrow><mrow><mtext>2</mtext></mrow></msub><msub><mrow><mtext>, ..., β</mtext></mrow><mrow><mtext>p</mtext></mrow></msub><msup><mrow><mtext>]</mtext></mrow><mrow><mtext>T</mtext></mrow></msup></mrow></math><img file="EP0709999A2_D0071.tif" /></maths> The pre-filter coefficient vector gp(k)can be calculated by a recursion formula composed of the following two equations:<maths id="math0033" num=""><math display="block"><mrow><mtable><mtr><mtd><mrow><mtable><mtr><mtd><mrow><mtable><mlabeledtr><mtext>(31)</mtext><mtd><mrow><msub><mrow><mtext>g</mtext></mrow><mrow><mtext>p</mtext></mrow></msub><mtext>(k)=(1-α)</mtext><mfenced open="[" close="]"><mrow><mtable><mtr><mtd><mtext>0</mtext></mtd></mtr><mtr><mtd><msub><mrow><mtext>f</mtext></mrow><mrow><mtext>p-1</mtext></mrow></msub><mtext>(k-1)</mtext></mtd></mtr></mtable></mrow></mfenced><mtext> + </mtext><mfrac><mrow><msub><mrow><mtext>a</mtext></mrow><mrow><mtext>p</mtext></mrow></msub><msup><mrow><mtext> </mtext></mrow><mrow><mtext>T</mtext></mrow></msup><mtext>(k)</mtext><mtext>e</mtext><mtext>(k)</mtext></mrow><mrow><mtext>F(k)</mtext></mrow></mfrac><msub><mrow><mtext>a</mtext></mrow><mrow><mtext>p</mtext></mrow></msub><mtext>(k)</mtext></mrow></mtd></mlabeledtr></mtable></mrow></mtd></mtr><mtr><mtd><mrow><mtable><mlabeledtr><mtext>(32)</mtext><mtd><mrow><mfenced open="[" close="]"><mrow><mtable><mtr><mtd><msub><mrow><mtext>f</mtext></mrow><mrow><mtext>p-1</mtext></mrow></msub><mtext>(k-1)</mtext></mtd></mtr><mtr><mtd><mtext>0</mtext></mtd></mtr></mtable></mrow></mfenced><mtext> = </mtext><msub><mrow><mtext>g</mtext></mrow><mrow><mtext>p</mtext></mrow></msub><mtext>(k-1) - </mtext><mfrac><mrow><msub><mrow><mtext>b</mtext></mrow><mrow><mtext>p</mtext></mrow></msub><msup><mrow><mtext> </mtext></mrow><mrow><mtext>T</mtext></mrow></msup><mtext>(k-1)</mtext><mtext>e</mtext><mtext>(k-1)</mtext></mrow><mrow><mtext>B(k-1)</mtext></mrow></mfrac><msub><mrow><mtext>b</mtext></mrow><mrow><mtext>p</mtext></mrow></msub><mtext>(k-1)</mtext></mrow></mtd></mlabeledtr></mtable></mrow></mtd></mtr></mtable></mrow></mtd></mtr></mtable></mrow></math><img file="EP0709999A2_D0072.tif" /></maths> where ap(k) is a p-th order forward linear prediction coefficient vector which satisfies a normal equation <maths id="math0034" num=""><math display="inline"><mrow><msub><mrow><mtext>R</mtext></mrow><mrow><mtext>p</mtext></mrow></msub><mtext>(k)</mtext><msub><mrow><mtext>a</mtext></mrow><mrow><mtext>p</mtext></mrow></msub><msup><mrow><mtext>(k)=[F(k), 0, ..., 0]</mtext></mrow><mrow><mtext>T</mtext></mrow></msup></mrow></math><img file="EP0709999A2_D0073.tif" /></maths> concerning the input signal x(k) and its leading element is l. F(k) is the minimum value of the sum of forward a posteriori prediction-error squares. bp(k-1) is a p-th order backward linear prediction coefficient vector which satisfies a normal equation <maths id="math0035" num=""><math display="inline"><mrow><msub><mrow><mtext>R</mtext></mrow><mrow><mtext>p</mtext></mrow></msub><mtext>(k- 1)</mtext><msub><mrow><mtext>b</mtext></mrow><mrow><mtext>p</mtext></mrow></msub><msup><mrow><mtext>(k-1)=[0, ..., B(k-1)]</mtext></mrow><mrow><mtext>T</mtext></mrow></msup></mrow></math><img file="EP0709999A2_D0074.tif" /></maths>, and its last element is 1. B(k-1) is the minimum value of the sum of backward a posteriori prediction-error squares. It is disclosed in J. M. Cioffi and T. Kailath, "Windowed fast transversal adaptive filter algorithms with normalization," IEEE Trans. Acoust., Speech, Signal Processing, vol. ASSP-33, No. 3, pp. 607-625 that the linear prediction coefficient vectors ap(k), bp(k-1) and the minimum values of the sum of a posteriori prediction-error squares F(k) and B(k-1) can be calculated at high speed with a small amount of calculation through utilization of, for instance, FTF (Fast Transversal Filter) algorithm. fp-1(k) is called a pre-filter conductor coefficient vector and is given by the following equation:<maths id="math0036" num="(33)"><math display="block"><mrow><msub><mrow><mtext>f</mtext></mrow><mrow><mtext>p-1</mtext></mrow></msub><mtext>(k)=</mtext><msub><mrow><mtext>R</mtext></mrow><mrow><mtext>p-1</mtext></mrow></msub><msup><mrow><mtext>(k)</mtext></mrow><mrow><mtext>-1</mtext></mrow></msup><msub><mrow><mtext>e</mtext></mrow><mrow><mtext>p-1</mtext></mrow></msub><mtext>(k)</mtext></mrow></math><img file="EP0709999A2_D0075.tif" /></maths> By substituting elements of the pre-filter coefficient vector gp, obtained from Eqs. (31) and (32), into Eq. (28), the adjustment vector v(k) can be calculated with a small amount of calculation.

While in the above the multi-channel projection algorithm has been described in connection with the case where the adjustment vector v(k) is derived from the pre-filter coefficient vector gp(k) and is used to adjust the estimated transfer function (the estimated echo path vector) <img file="EP0709999A2_D0076.tif" />(k) by Eq. (21), it is possible to obtain the estimated echo ŷm(k) from the pre-filter coefficient vector gp(k) at high speed and with less computational complexity instead of calculating the adjustment vector v(k), as described below. At first, the auto-correlation vector rp-1(k) of the input signal vector x(k) of each channel is defined by the following equation:<maths id="math0037" num="(34)"><math display="block"><mrow><msub><mrow><mtext>r</mtext></mrow><mrow><mtext>p-1</mtext></mrow></msub><mtext>(k)=[</mtext><msup><mrow><mtext>x</mtext></mrow><mrow><mtext>T</mtext></mrow></msup><mtext>(k)</mtext><mtext>x</mtext><mtext>(k-1),</mtext><msup><mrow><mtext>x</mtext></mrow><mrow><mtext>T</mtext></mrow></msup><mtext>(k)</mtext><mtext>x</mtext><mtext>(k-2),...,</mtext><msup><mrow><mtext>x</mtext></mrow><mrow><mtext>T</mtext></mrow></msup><mtext>(k)</mtext><mtext>x</mtext><msup><mrow><mtext>(k-p+1)]</mtext></mrow><mrow><mtext>T</mtext></mrow></msup></mrow></math><img file="EP0709999A2_D0077.tif" /></maths> The pre-filter coefficients βi, which are the elements of the pre-filter coefficient vector, are smoothed by the following equation to define smoothing coefficients si(k):<maths id="math0038" num="(35)"><math display="block"><mrow><msub><mrow><mtext>s</mtext></mrow><mrow><mtext>i</mtext></mrow></msub><msub><mrow><mtext>(k) = s</mtext></mrow><mrow><mtext>i-1</mtext></mrow></msub><msub><mrow><mtext>(k-1)+αβ</mtext></mrow><mrow><mtext>i</mtext></mrow></msub><mtext>(k) for 2≦i≦p</mtext><mspace linebreak="newline" /><msub><mrow><mtext> = αβ</mtext></mrow><mrow><mtext>i</mtext></mrow></msub><mtext>(k) for i=1 </mtext></mrow></math><img file="EP0709999A2_D0078.tif" /></maths> An approximate transfer function z(k) which approximates the estimated transfer function <img file="EP0709999A2_D0079.tif" />(k) is defined by the following equation:<maths id="math0039" num="(36)"><math display="block"><mrow><mtext>z</mtext><mtext>(k+1)=</mtext><mtext>z</mtext><msub><mrow><mtext>(k)+{β</mtext></mrow><mrow><mtext>p</mtext></mrow></msub><msub><mrow><mtext>(k)+β</mtext></mrow><mrow><mtext>p-1</mtext></mrow></msub><msub><mrow><mtext>(k-1)+...+β</mtext></mrow><mrow><mtext>2</mtext></mrow></msub><msub><mrow><mtext>(k-p+2) +β</mtext></mrow><mrow><mtext>1</mtext></mrow></msub><mtext>(k-1)}</mtext><mtext>x</mtext><mtext>(k-p+1)</mtext><mspace linebreak="newline" /><mtext> =</mtext><mtext>z</mtext><msub><mrow><mtext>(k)+s</mtext></mrow><mrow><mtext>p</mtext></mrow></msub><mtext>(k)</mtext><mtext>x</mtext><mtext>(k-p+1) </mtext></mrow></math><img file="EP0709999A2_D0080.tif" /></maths> Furthermore, defining by the following equation a smoothing coefficient vector sp-1(k-1) which uses the smoothing coefficients si(k) (i=1, ..., p-1) of Eq. (35) as its elements:<maths id="math0040" num="(37)"><math display="block"><mrow><msub><mrow><mtext>s</mtext></mrow><mrow><mtext>p-1</mtext></mrow></msub><msub><mrow><mtext>(k-1)=[s</mtext></mrow><mrow><mtext>1</mtext></mrow></msub><msub><mrow><mtext>(k-1), s</mtext></mrow><mrow><mtext>2</mtext></mrow></msub><msub><mrow><mtext>(k-1), ..., s</mtext></mrow><mrow><mtext>p-1</mtext></mrow></msub><msup><mrow><mtext>(k-1)]</mtext></mrow><mrow><mtext>T</mtext></mrow></msup></mrow></math><img file="EP0709999A2_D0081.tif" /></maths> the estimated echo ŷm(k) can be calculated by the following equation from the inner product rp-1T(k)sp-1(k-1) and the convolution zT(k)x(k):<maths id="math0041" num="(38)"><math display="block"><mrow><msub><mrow><mover accent="true"><mrow><mtext>y</mtext></mrow><mo>^</mo></mover></mrow><mrow><mtext>m</mtext></mrow></msub><mtext>(k)=</mtext><msup><mrow><mtext>z</mtext></mrow><mrow><mtext>T</mtext></mrow></msup><mtext>(k)</mtext><mtext>x</mtext><mtext>(k)+</mtext><msub><mrow><mtext>r</mtext></mrow><mrow><mtext>p-1</mtext></mrow></msub><msup><mrow><mtext> </mtext></mrow><mrow><mtext>T</mtext></mrow></msup><mtext>(k)</mtext><msub><mrow><mtext>s</mtext></mrow><mrow><mtext>p-1</mtext></mrow></msub><mtext>(k-1)</mtext></mrow></math><img file="EP0709999A2_D0082.tif" /></maths> A description will be given of a concrete procedure of the above-mentioned multi-channel fast projection algorithm for the m-th channel echo ym(k). <ul id="ul0005" list-style="none" compact="compact"><li>Step S1: As initialization procesisng, the auto-correlation vector of Eq. (36) at time k=0 is set by the following equation:<maths id="math0042" num="(39)"><math display="block"><mrow><msub><mrow><mtext>r</mtext></mrow><mrow><mtext>p-1</mtext></mrow></msub><mtext>(0)=[</mtext><msup><mrow><mtext>x</mtext></mrow><mrow><mtext>T</mtext></mrow></msup><mtext>(0)</mtext><mtext>x</mtext><mtext>(-1),</mtext><msup><mrow><mtext>x</mtext></mrow><mrow><mtext>T</mtext></mrow></msup><mtext>(0)</mtext><mtext>x</mtext><mtext>(-2),...,</mtext><msup><mrow><mtext>x</mtext></mrow><mrow><mtext>T</mtext></mrow></msup><mtext>(0)</mtext><mtext>x</mtext><msup><mrow><mtext>(-p+1)]</mtext></mrow><mrow><mtext>T</mtext></mrow></msup></mrow></math><img file="EP0709999A2_D0083.tif" /></maths> And the residual echo ep(k), the pre-filter conductor coefficient vector fp-1(k), the smoothing coefficient vector sp-1(k) and the minimum values of the sum of a posterirori prediction-error squares F(k) and B(k) at time k=0 are respectively set as follows:<maths id="math0043" num=""><math display="block"><mrow><mtable><mtr><mtd><mrow><mtable><mtr><mtd><mrow><msub><mrow><mtext>e</mtext></mrow><mrow><mtext>p</mtext></mrow></msub><mtext>(0)=0</mtext></mrow></mtd></mtr><mtr><mtd><mrow><msub><mrow><mtext>f</mtext></mrow><mrow><mtext>p-1</mtext></mrow></msub><mtext>(0)=</mtext><msub><mrow><mtext>s</mtext></mrow><mrow><mtext>p-1</mtext></mrow></msub><mtext>(0)=0</mtext></mrow></mtd></mtr><mtr><mtd><mrow><mtext>F(0)=B(0)=δ</mtext></mrow></mtd></mtr></mtable></mrow></mtd></mtr></mtable></mrow></math><img file="EP0709999A2_D0084.tif" /></maths> where δ is a small positive constant.</li><li>Step S2: Input signal vectors of N channels are calculated as follows:<maths id="math0044" num=""><img file="EP0709999A2_D0085.tif" /></maths> The correlation vector at time k-1 is used to calculate the correlation vector at time k by the following equation:<maths id="math0045" num=""><img file="EP0709999A2_D0086.tif" /></maths></li><li>Step S3: The estimated echo ŷm(k) is calculated by the following equation:<maths id="math0046" num="(42)"><math display="block"><mrow><msub><mrow><mover accent="true"><mrow><mtext>y</mtext></mrow><mo>^</mo></mover></mrow><mrow><mtext>m</mtext></mrow></msub><mtext>(k)=</mtext><msub><mrow><mtext>z</mtext></mrow><mrow><mtext>m</mtext></mrow></msub><msup><mrow><mtext> </mtext></mrow><mrow><mtext>T</mtext></mrow></msup><mtext>(k)</mtext><mtext>x</mtext><mtext>(k)+ </mtext><msub><mrow><mtext>r</mtext></mrow><mrow><mtext>p-1</mtext></mrow></msub><msup><mrow><mtext> </mtext></mrow><mrow><mtext>T</mtext></mrow></msup><mtext>(k)</mtext><msub><mrow><mtext>s</mtext></mrow><mrow><mtext>p-1</mtext></mrow></msub><mtext>(k-1)</mtext></mrow></math><img file="EP0709999A2_D0087.tif" /></maths> Step S4: The residual echo em(k) is calculated by the following equation:<maths id="math0047" num="(43)"><math display="block"><mrow><msub><mrow><mtext>e</mtext></mrow><mrow><mtext>m</mtext></mrow></msub><msub><mrow><mtext>(k)=y</mtext></mrow><mrow><mtext>m</mtext></mrow></msub><mtext>(k)-</mtext><msub><mrow><mover accent="true"><mrow><mtext>y</mtext></mrow><mo>^</mo></mover></mrow><mrow><mtext>m</mtext></mrow></msub><mtext>(k)</mtext></mrow></math><img file="EP0709999A2_D0088.tif" /></maths></li><li>Step S5:<maths id="math0048" num="(44)"><math display="block"><mrow><msub><mrow><mtext>e</mtext></mrow><mrow><mtext>p</mtext></mrow></msub><msub><mrow><mtext>(k)=[e</mtext></mrow><mrow><mtext>m</mtext></mrow></msub><msub><mrow><mtext>(k),(1-α)e</mtext></mrow><mrow><mtext>m</mtext></mrow></msub><msup><mrow><mtext>(k-1),...,(1-α)</mtext></mrow><mrow><mtext>p-1</mtext></mrow></msup><msub><mrow><mtext>e</mtext></mrow><mrow><mtext>m</mtext></mrow></msub><msup><mrow><mtext>(k-p+1)]</mtext></mrow><mrow><mtext>T</mtext></mrow></msup></mrow></math><img file="EP0709999A2_D0089.tif" /></maths> is calculated by the following iterative equation:<maths id="math0049" num="(45)"><math display="block"><mrow><mfenced open="[" close="]"><mrow><mtable><mtr><mtd><msub><mrow><mtext>e</mtext></mrow><mrow><mtext>p</mtext></mrow></msub><mtext>(k)</mtext></mtd></mtr><mtr><mtd><mtext>*</mtext></mtd></mtr></mtable></mrow></mfenced><mtext> = </mtext><mfenced open="[" close="]"><mrow><mtable><mtr><mtd><msub><mrow><mtext>e</mtext></mrow><mrow><mtext>m</mtext></mrow></msub><mtext>(k)</mtext></mtd></mtr><mtr><mtd><mtext>(1-α)</mtext><msub><mrow><mtext>e</mtext></mrow><mrow><mtext>p</mtext></mrow></msub><mtext>(k-1)</mtext></mtd></mtr></mtable></mrow></mfenced></mrow></math><img file="EP0709999A2_D0090.tif" /></maths> where the value * is not required.</li><li>Step S6: By using the afore-mentioned fast calculation methods by J. M. Cioffi and T. Kailath for the linear prediction coefficients and the minimum values of the sum of forward and backward a posteriori prediction-error squares, a p-th order forward linear prediction coefficient vector ap(k), its minimum value of the sum of forward a posteriori prediction-error squares F(k), a p-th order backward linear prediction coefficient vector bp(k) and its minimum value of the sum of backward a posteriori prediction-error squares B(k) are calculated which satisfy the following equations:<maths id="math0050" num=""><math display="block"><mrow><mtable><mtr><mtd><mrow><mtable><mtr><mtd><mrow><mtable><mlabeledtr><mtext>(46)</mtext><mtd><mrow><msup><mrow><mtext>X</mtext></mrow><mrow><mtext>T</mtext></mrow></msup><mtext>(k)</mtext><mtext>X</mtext><mtext>(k)</mtext><msub><mrow><mtext>a</mtext></mrow><mrow><mtext>p</mtext></mrow></msub><msup><mrow><mtext>(k)=[F(k), 0, ..., 0]</mtext></mrow><mrow><mtext>T</mtext></mrow></msup></mrow></mtd></mlabeledtr></mtable></mrow></mtd></mtr><mtr><mtd><mrow><mtable><mlabeledtr><mtext>(47)</mtext><mtd><mrow><msup><mrow><mtext>X</mtext></mrow><mrow><mtext>T</mtext></mrow></msup><mtext>(k-1)</mtext><mtext>X</mtext><mtext>(k-1)</mtext><mtext>b</mtext><msup><mrow><mtext>p(k-1)={0, ..., B(k-1)]</mtext></mrow><mrow><mtext>T</mtext></mrow></msup><mtext></mtext></mrow></mtd></mlabeledtr></mtable></mrow></mtd></mtr></mtable></mrow></mtd></mtr></mtable></mrow></math><img file="EP0709999A2_D0091.tif" /></maths> where the leading element of the vector ap(k) and the last element of the vector bp(k) are both 1s.</li><li>Step S7: The pre-filter coefficient vector gp(k) is calculated by the following equation:<maths id="math0051" num="(48)"><math display="block"><mrow><msub><mrow><mtext>g</mtext></mrow><mrow><mtext>p</mtext></mrow></msub><mtext>(k)=(1-α)</mtext><mfenced open="[" close="]"><mrow><mtable><mtr><mtd><mtext>0</mtext></mtd></mtr><mtr><mtd><msub><mrow><mtext>f</mtext></mrow><mrow><mtext>p-1</mtext></mrow></msub><mtext>(k-1)</mtext></mtd></mtr></mtable></mrow></mfenced><mtext> + </mtext><mfrac><mrow><mtext>a</mtext><msub><mrow><mtext></mtext></mrow><mrow><msup><mrow><mtext>p</mtext></mrow><mrow><mtext>T</mtext></mrow></msup></mrow></msub><msub><mrow><mtext>(k)e</mtext></mrow><mrow><mtext>p</mtext></mrow></msub><mtext>(k)</mtext></mrow><mrow><mtext>F(k)</mtext></mrow></mfrac><msub><mrow><mtext>a</mtext></mrow><mrow><mtext>p</mtext></mrow></msub><mtext>(k)</mtext></mrow></math><img file="EP0709999A2_D0092.tif" /></maths></li><li>Step S8: The pre-filter conductor coefficient vector fp-1(k) is calculated by the following equation:<maths id="math0052" num="(49)"><math display="block"><mrow><mfenced open="[" close="]"><mrow><mtable><mtr><mtd><msub><mrow><mtext>f</mtext></mrow><mrow><mtext>p-1</mtext></mrow></msub><mtext>(k-1)</mtext></mtd></mtr><mtr><mtd><mtext>0</mtext></mtd></mtr></mtable></mrow></mfenced><mtext> = </mtext><msub><mrow><mtext>g</mtext></mrow><mrow><mtext>p</mtext></mrow></msub><mtext>(k) - </mtext><mfrac><mrow><mtext>b</mtext><msub><mrow><mtext></mtext></mrow><mrow><msup><mrow><mtext>p</mtext></mrow><mrow><mtext>T</mtext></mrow></msup></mrow></msub><mtext>(k)</mtext><msub><mrow><mtext>e</mtext></mrow><mrow><mtext>p</mtext></mrow></msub><mtext>(k)</mtext></mrow><mrow><mtext>B(k-1)</mtext></mrow></mfrac><msub><mrow><mtext>b</mtext></mrow><mrow><mtext>p</mtext></mrow></msub><mtext>(k)</mtext></mrow></math><img file="EP0709999A2_D0093.tif" /></maths></li><li>Step S9: The smoothing coefficient vectors sp-1(k) and sp(k) are calculated by the following equation:<maths id="math0053" num="(50)"><math display="block"><mrow><mfenced open="[" close="]"><mrow><mtable><mtr><mtd><msub><mrow><mtext>s</mtext></mrow><mrow><mtext>p-1</mtext></mrow></msub><mtext>(k)</mtext></mtd></mtr><mtr><mtd><msub><mrow><mtext>s</mtext></mrow><mrow><mtext>p</mtext></mrow></msub><mtext>(k)</mtext></mtd></mtr></mtable></mrow></mfenced><mtext> = </mtext><mfenced open="[" close="]"><mrow><mtable><mtr><mtd><mtext>0</mtext></mtd></mtr><mtr><mtd><msub><mrow><mtext>s</mtext></mrow><mrow><mtext>p-1</mtext></mrow></msub><mtext>(k-1)</mtext></mtd></mtr></mtable></mrow></mfenced><mtext> + α</mtext><msub><mrow><mtext>g</mtext></mrow><mrow><mtext>p</mtext></mrow></msub><mtext>(k)</mtext></mrow></math><img file="EP0709999A2_D0094.tif" /></maths></li><li>Step S10: The approximate value zm(k) of the combined estimated echo path vector<maths id="math0054" num=""><img file="EP0709999A2_D0095.tif" /></maths> is updated by the following equation:<maths id="math0055" num="(51)"><math display="block"><mrow><msub><mrow><mtext>z</mtext></mrow><mrow><mtext>m</mtext></mrow></msub><mtext>(k+1)=</mtext><msub><mrow><mtext>z</mtext></mrow><mrow><mtext>m</mtext></mrow></msub><msub><mrow><mtext>(k)+s</mtext></mrow><mrow><mtext>p</mtext></mrow></msub><mtext>(k)</mtext><mtext>x</mtext><mtext>(k-p+1)</mtext></mrow></math><img file="EP0709999A2_D0096.tif" /></maths></li><li>Step S11: Upon each increment of time k, Steps S2 through S10 are repeated.</li></ul>

Fig. 6 illustrates in block form an embodiment of the first method of the present invention applied to the echo canceller 22m in Fig. 3 which is used in the Fig. 2 teleconferencing system provided with the N-channel loudspeaker system and the M-channel microphone system. The parts corresponding to those in Fig. 3 are identified by the same reference numerals. According to the present invention, a cross-correlation variation extracting part 31 is provided in each of the echo cancellers 22₁ to 22M in Fig. 2. The received signal x(k) and the residual echo em(k) in each channel are inputted into the cross-correlation variation extracting part 31, wherein the adjustment vector v(k) is generated.

In this embodiment, the auto-correlation of the combined received signal vector x(k) is removed using the projection algorithm, by which the variation in the cross-correlation between received signals to generate the adjustment vector v(k). That is to say, the cross-correlation variation extracting part 31 stores information about the combined received signal vector x(k) and the residual echo em(k) defined by Eq. (8), then solves the following equation (52) corresponding to Eq. (27) to obtain gp(k) and generates, by Eq. (28), a decorrelated vector, i.e. the adjustment vector v(k).<maths id="math0056" num="(52)"><math display="block"><mrow><mtext>[</mtext><msup><mrow><mtext>X</mtext></mrow><mrow><mtext>T</mtext></mrow></msup><mtext>(k)</mtext><mtext>X</mtext><mtext>(k)]</mtext><msub><mrow><mtext>g</mtext></mrow><mrow><mtext>p</mtext></mrow></msub><mtext>(k)=</mtext><msub><mrow><mtext>e</mtext></mrow><mrow><mtext>m</mtext></mrow></msub><mtext>(k)</mtext></mrow></math><img file="EP0709999A2_D0097.tif" /></maths> An adjusting part 32 updates the current combined estimated echo path vector <img file="EP0709999A2_D0098.tif" />m(k) by calculating <img file="EP0709999A2_D0099.tif" />m(k+1) on the basis of Eqs. (21) and (28) using the adjustment vector v(k) as follows:<maths id="math0057" num=""><img file="EP0709999A2_D0100.tif" /></maths> The combined estimated echo path vector <img file="EP0709999A2_D0101.tif" />m(k) thus updated is fed to the estimated echo generating part 18m.

In this way, the variation in the cross-correlation between multi-channel received signal vectors is extracted in the cross-correlation variation extracting part 31 and new multi-channel signal vectors are generated in which the variation in the cross-correlation between the respective received signal vectors are emphasized. Since such a multi-channel signal vector is fed to the adjusting part 32, it is possible to speed up the estimation of the true echo paths.

Fig. 7 is a block diagram showing a modified form of the cross-correlation variation extracting part, which is formed by a cross-correlation variation coefficient calculating part 31A and a cross-correlation variation emphasized vector generating part 31B and in which the former solves Eq. (27) to obtain the pre-filter coefficient vector gp expressed by Eq. (30) and the latter calculates the adjustment vector v(k) by Eq. (28). In this case, the cross-correlation variation coefficient calculating part 31A can reduce the amount of calculation of the pre-filter coefficient vector gp(k) through utilization of linear prediction. That is to say, as shown in Fig. 8, the cross-correlation variation coefficient calculating part 31A is formed by a linear prediction part 31A1, a pre-filter conductor coefficient vector adjusting part 31A2, a pre-filter coefficient vector adjusting part 31A3 and an error signal vector generating part 31A4. The linear prediction part 31A1 calculates the forward linear prediction coefficient vector ap(k) of the combined input signal vector x(k), the minimum value of the sum of forward a posteriori prediction-error squares F(k), the backward linear prediction coefficient vector bp(k) of the combined input signal vector x(k) and the minimum value of the sum of backward a posteriori prediction-error squares B(k).

On the other hand, the error signal vector generating part 31A4 holds the p latest error signals (residual echoes) em(k), em(k-1), ..., em(k-p+1) iteratively generates the error signal vector ep(k) of Eq. (44) by Eq. (45). The pre-filter conductor coefficient vector adjusting part 31A2 calculates the pre-filter conductor coefficient vector fp-1(k-1) by Eq. (32) and the filter coefficient vector adjusting part 31A3 the pre-filter coefficient vector gp(k) by Eq. (31). The pre-filter coefficient vector gp(k) thus obtained is provided to the cross-correlation variation emphasized vector generating part 31B.

Fig. 9 illustrates a configuration which reduces the amount of calculation in the Fig. 6 embodiment by generating the estimated echo ŷm(k), instead of calculating the combined estimated echo path vector <img file="EP0709999A2_D0102.tif" />m(k), by using the approximate vector zm(k) of the vector <img file="EP0709999A2_D0103.tif" />m(k) on the basis of the fast projection algorithm described previously in Steps S1 through S11. In this instance, the echo path estimating part 19m and the estimated echo generating part 18m in fig. 6 are substituted with the cross-correlation variation coefficient calculating part 31A and the simplified estimated echo generating part 180m; the former is identical with that shown in Fig. 7 or 8 and the latter corresponds to the set of the cross-correlation variation emphasized vector generating part 31B, the adjusting part 32 and the estimated echo generating part 18m and, as depicted in Fig. 10, it comprises a correlation calculating part 181, a smoothing coefficient vector adjusting part 182, an approximated echo path adjusting part 183, inner product calculating parts 184 and 185 and an adder 186.

The correlation calculating part 181 calculates the combined input signal vector x(k) and the auto-correlation vector rp-1(k) of x(k) by Eqs. (40) and (41). The smoothing coefficient vector adjusting part 182 calculates by Eq. (50) smoothing coefficient vector sp-1(k-1) and sp(k) from the pre-filter coefficient vector gp(k). The approximated echo path adjusting part 183 calculates by Eq. (51) the approximated echo path vector zm(k) from sp(k) and x(k). The inner product calculating part 184 calculates the inner product rp-1T(k)sp-1(k-1) of the smoothing coefficient vector sp-1(k-1) and the correlation vector rp-1(k). The inner product calculating part 185 calculates the inner product zmT(k)x(k) of the input signal vector x(k) and the approximated echo path vector zm(k). These inner products are added together by the adder 186 to obtain the estimated echo signal ŷm(k).

Next, the third method of the present invention will be described concretely. According to this method, a function of actively changing the cross-correlation between received signals of respective channels are adopted, and the received signals added with the variation in the correlation between them are radiated by loudspeakers and used to derive the adjusted vector of the estimated echo.

The variation in the correlation between received signals of respective channels can be added thereto by (a) filter processing, (b) signal multiplication processing, (c) signal addition processing and (d) pitch shift processing. These methods will hereinafter be described one after another.

(a) Method by Filter Processing

The received signals x₁(k), x₂(k), ..., xN(k) of respective channels are inputted to time-variant filters with different time-variant characteristics, wherein they are convoluted (indicated by *) with impulse responses q₁(k), q₂(k), ..., qN(k) of the filters for conversion into signals <maths id="math0058" num=""><math display="inline"><mrow><mover accent="true"><mrow><mtext>x</mtext></mrow><mo>¯</mo></mover></mrow></math><img file="EP0709999A2_D0104.tif" /></maths>₁(k), <maths id="math0059" num=""><math display="inline"><mrow><mover accent="true"><mrow><mtext>x</mtext></mrow><mo>¯</mo></mover></mrow></math><img file="EP0709999A2_D0105.tif" /></maths>₂(k), ... ,<maths id="math0060" num=""><math display="inline"><mrow><mover accent="true"><mrow><mtext>x</mtext></mrow><mo>¯</mo></mover></mrow></math><img file="EP0709999A2_D0106.tif" /></maths>N(k) which are expressed as follows:<maths id="math0061" num=""><img file="EP0709999A2_D0107.tif" /></maths> By this, the variation in the cross-correlation between the received signal of respective channels can be added thereto.

(b) Method by Signal Multiplication

The received signals x₁(k), x₂(k), ..., xN(k) of respective channels are multiplied by different functions u₁(k), u₂(k), ..., uN(k) for conversion into signals <maths id="math0062" num=""><math display="inline"><mrow><mover accent="true"><mrow><mtext>x</mtext></mrow><mo>¯</mo></mover></mrow></math><img file="EP0709999A2_D0108.tif" /></maths>₁(k), <maths id="math0063" num=""><math display="inline"><mrow><mover accent="true"><mrow><mtext>x</mtext></mrow><mo>¯</mo></mover></mrow></math><img file="EP0709999A2_D0109.tif" /></maths>₂(k), ... <maths id="math0064" num=""><math display="inline"><mrow><mover accent="true"><mrow><mtext>x</mtext></mrow><mo>¯</mo></mover></mrow></math><img file="EP0709999A2_D0110.tif" /></maths>N(k) which are expressed as follows:<maths id="math0065" num=""><img file="EP0709999A2_D0111.tif" /></maths> By this, the variation in the cross-correlation between the received signals can be added thereto.

The received signals x₁(k), x₂(k), ..., xN(k) of respective channels are added to different functions n₁(k), n₂(k), nN(k), respectively, for conversion into signals <maths id="math0066" num=""><math display="inline"><mrow><mover accent="true"><mrow><mtext>x</mtext></mrow><mo>¯</mo></mover></mrow></math><img file="EP0709999A2_D0112.tif" /></maths>₁(k), <maths id="math0067" num=""><math display="inline"><mrow><mover accent="true"><mrow><mtext>x</mtext></mrow><mo>¯</mo></mover></mrow></math><img file="EP0709999A2_D0113.tif" /></maths>₂(k), ..., <maths id="math0068" num=""><math display="inline"><mrow><mover accent="true"><mrow><mtext>x</mtext></mrow><mo>¯</mo></mover></mrow></math><img file="EP0709999A2_D0114.tif" /></maths>N(k) which are expressed as follows:<maths id="math0069" num=""><img file="EP0709999A2_D0115.tif" /></maths> By this, the variation in the cross-correlation between the received signals of respective channels can be added thereto.

(d) Method by Pitch Shift Processing

The frequency characteristics of the received signals x₁(k), x₂(k), ..., xN(k) of respective channels are each subjected to different time-variant frequency axis expansion or compression, that is, by pitch shift processing, by which the signals are converted into <maths id="math0070" num=""><math display="inline"><mrow><mover accent="true"><mrow><mtext>x</mtext></mrow><mo>¯</mo></mover></mrow></math><img file="EP0709999A2_D0116.tif" /></maths>₁(k), <maths id="math0071" num=""><math display="inline"><mrow><mover accent="true"><mrow><mtext>x</mtext></mrow><mo>¯</mo></mover></mrow></math><img file="EP0709999A2_D0117.tif" /></maths>₂(k), ..., <maths id="math0072" num=""><math display="inline"><mrow><mover accent="true"><mrow><mtext>x</mtext></mrow><mo>¯</mo></mover></mrow></math><img file="EP0709999A2_D0118.tif" /></maths>N(k) signals. By this, the variation in the cross-correlation between the received signals of the respective channels can be added thereto.

Fig. 11 illustrates an embodiment of the third method of the present invention. The parts corresponding to those in Fig. 2 are identified by the same reference numerals. In this embodiment, a cross-correlation variation adding part 24 is provided at the received signal input side. In the cross-correlation variation adding part 24, the received signals x₁(k), x₂(k), ..., xN(k) of the respective channels are converted into <maths id="math0073" num=""><math display="inline"><mrow><mover accent="true"><mrow><mtext>x</mtext></mrow><mo>¯</mo></mover></mrow></math><img file="EP0709999A2_D0119.tif" /></maths>₁(k), <maths id="math0074" num=""><math display="inline"><mrow><mover accent="true"><mrow><mtext>x</mtext></mrow><mo>¯</mo></mover></mrow></math><img file="EP0709999A2_D0120.tif" /></maths>₂(k), ..., <maths id="math0075" num=""><math display="inline"><mrow><mover accent="true"><mrow><mtext>x</mtext></mrow><mo>¯</mo></mover></mrow></math><img file="EP0709999A2_D0121.tif" /></maths>N(k) by being actively added with the variation in their cross-correlation. These signals <maths id="math0076" num=""><math display="inline"><mrow><mover accent="true"><mrow><mtext>x</mtext></mrow><mo>¯</mo></mover></mrow></math><img file="EP0709999A2_D0122.tif" /></maths>₁(k), <maths id="math0077" num=""><math display="inline"><mrow><mover accent="true"><mrow><mtext>x</mtext></mrow><mo>¯</mo></mover></mrow></math><img file="EP0709999A2_D0123.tif" /></maths>₂(k), ..., <maths id="math0078" num=""><math display="inline"><mrow><mover accent="true"><mrow><mtext>x</mtext></mrow><mo>¯</mo></mover></mrow></math><img file="EP0709999A2_D0124.tif" /></maths>N(k) thus converted are radiated from the speakers 12₁, 12₂, ..., 12N, while at the same time they are inputted into the echo canceller system 23.

In the implementation of the cross-correlation variation adding part 24, care should be taken not to impair the psycho-acoustic quality of the converted signals <maths id="math0079" num=""><math display="inline"><mrow><mover accent="true"><mrow><mtext>x</mtext></mrow><mo>¯</mo></mover></mrow></math><img file="EP0709999A2_D0125.tif" /></maths>₁(k), <maths id="math0080" num=""><math display="inline"><mrow><mover accent="true"><mrow><mtext>x</mtext></mrow><mo>¯</mo></mover></mrow></math><img file="EP0709999A2_D0126.tif" /></maths>₂(k), ..., <maths id="math0081" num=""><math display="inline"><mrow><mover accent="true"><mrow><mtext>x</mtext></mrow><mo>¯</mo></mover></mrow></math><img file="EP0709999A2_D0127.tif" /></maths>N(k) when they are reconstructed as acoustic signals. Next, a description will be given of concrete examples of the above-mentioned methods in the implementation of the cross-correlation variation adding part 24.

(a) Concrete Example of Implementation by Filter Processing

A description will be given of the time-variant filters for converting the received signals x₁(k), <maths id="math0082" num=""><math display="inline"><mrow><mover accent="true"><mrow><mtext>x</mtext></mrow><mo>¯</mo></mover></mrow></math><img file="EP0709999A2_D0128.tif" /></maths>₂(k), ...,<maths id="math0083" num=""><math display="inline"><mrow><mover accent="true"><mrow><mtext>x</mtext></mrow><mo>¯</mo></mover></mrow></math><img file="EP0709999A2_D0129.tif" /></maths>N(k) of the respective channels into those <maths id="math0084" num=""><math display="inline"><mrow><mover accent="true"><mrow><mtext>x</mtext></mrow><mo>¯</mo></mover></mrow></math><img file="EP0709999A2_D0130.tif" /></maths>₁(k), x₂(k), ..., xN(k) which satisfy Eq. (54), through utilization of the impulse response q₁(k), q₂(k), ..., qN(k) of the filters having different time-variant characteristics. It is said that amplitude information is mainly important for the psycho-acoustic perception and that phase information is not so important. Hence, it is considered that the psycho-acoustic influence of filter processing would be lessened by implementing the time-variant filter so that the amplitude characteristic remain flat with respect to the frequency of each input signal.

A filer that serves this purpose is an IIR filter which has an all-pass transfer function. The transfer function G(z) of this filter can be expressed as follows:<maths id="math0085" num="(57)"><math display="block"><mrow><msup><mrow><mtext>G(z)=A{z</mtext></mrow><mrow><mtext>-K</mtext></mrow></msup><msub><mrow><mtext>-c</mtext></mrow><mrow><mtext>1</mtext></mrow></msub><msup><mrow><mtext>z</mtext></mrow><mrow><mtext>-(K-1)</mtext></mrow></msup><msup><mrow><mtext>+...+(-1)</mtext></mrow><mrow><mtext>K-1</mtext></mrow></msup><msub><mrow><mtext>c</mtext></mrow><mrow><mtext>K-1</mtext></mrow></msub><msup><mrow><mtext>z</mtext></mrow><mrow><mtext>-1</mtext></mrow></msup><msup><mrow><mtext>+(-1)</mtext></mrow><mrow><mtext>K</mtext></mrow></msup><msub><mrow><mtext>c</mtext></mrow><mrow><mtext>K</mtext></mrow></msub><msub><mrow><mtext>} ÷{1-c</mtext></mrow><mrow><mtext>1</mtext></mrow></msub><msup><mrow><mtext>z</mtext></mrow><mrow><mtext>-1</mtext></mrow></msup><msup><mrow><mtext>+...+(-1)</mtext></mrow><mrow><mtext>K-1</mtext></mrow></msup><msub><mrow><mtext>c</mtext></mrow><mrow><mtext>K-1</mtext></mrow></msub><msup><mrow><mtext>z</mtext></mrow><mrow><mtext>-1(K-1)</mtext></mrow></msup><msup><mrow><mtext>+(-1)</mtext></mrow><mrow><mtext>K</mtext></mrow></msup><msub><mrow><mtext>c</mtext></mrow><mrow><mtext>K</mtext></mrow></msub><msup><mrow><mtext>z</mtext></mrow><mrow><mtext>-K</mtext></mrow></msup><mtext>}</mtext></mrow></math><img file="EP0709999A2_D0131.tif" /></maths> where K is the order. Let the sampling interval and the angular frequency be represented by T and ω, respectively. No matter what values the filter coefficients c₁, c₂, .., cM are set to take, the denominator and the numerator cancel with each other and the amplitude characteristic equation | G(ejωT)| takes a fixed gain A independent of the angular frequency ω. On the other hand, the phase characteristic differs with the filter coefficients c₁, c₂, ..., cM. When the coefficients c₁, c₂, ..., cM are varied with time, a time-variant phase characteristic is provided, but the influence on the amplitude characteristic is very slight.

Now, a description will be given of a secondary all-pass filter which is simple-structured and easy to implement. The configuration of the secondary all-pass filter is described in detail in, for example, Alan V. Oppenheim and Ronald W. Schafer, "DISCRETE-TIME SIGNAL PROCESSING," PRENTICE HALL. For brevity, setting the gain A=1 in Eq. (57), the secondary all-pass transfer function G(z) is given as follows:<maths id="math0086" num="(58)"><math display="block"><mrow><msup><mrow><mtext>G(z)=(z</mtext></mrow><mrow><mtext>-2</mtext></mrow></msup><msub><mrow><mtext>-c</mtext></mrow><mrow><mtext>1</mtext></mrow></msub><msup><mrow><mtext>z</mtext></mrow><mrow><mtext>-1</mtext></mrow></msup><msub><mrow><mtext>+c</mtext></mrow><mrow><mtext>2</mtext></mrow></msub><msub><mrow><mtext>)/(1-c</mtext></mrow><mrow><mtext>1</mtext></mrow></msub><msup><mrow><mtext>z</mtext></mrow><mrow><mtext>-1</mtext></mrow></msup><msub><mrow><mtext>+c</mtext></mrow><mrow><mtext>2</mtext></mrow></msub><msup><mrow><mtext>z</mtext></mrow><mrow><mtext>-2</mtext></mrow></msup><mtext>)</mtext></mrow></math><img file="EP0709999A2_D0132.tif" /></maths> When configured as a lattice model, the filter having this transfer function can be formed by six adders, two multipliers and two unit delays as shown in Fig. 12. Replacing the coefficients in Eq. (58) by<maths id="math0087" num=""><math display="block"><mrow><mtable><mtr><mtd><mrow><mtable><mtr><mtd><mrow><mtable><mlabeledtr><mtext>(59)</mtext><mtd><mrow><msub><mrow><mtext>c</mtext></mrow><mrow><mtext>1</mtext></mrow></msub><msub><mrow><mtext> = -γ</mtext></mrow><mrow><mtext>1</mtext></mrow></msub><msub><mrow><mtext>(1+γ</mtext></mrow><mrow><mtext>2</mtext></mrow></msub><mtext>)</mtext></mrow></mtd></mlabeledtr></mtable></mrow></mtd></mtr><mtr><mtd><mrow><mtable><mlabeledtr><mtext>(60)</mtext><mtd><mrow><msub><mrow><mtext>c</mtext></mrow><mrow><mtext>2</mtext></mrow></msub><msub><mrow><mtext> = γ</mtext></mrow><mrow><mtext>2</mtext></mrow></msub><mtext></mtext></mrow></mtd></mlabeledtr></mtable></mrow></mtd></mtr></mtable></mrow></mtd></mtr></mtable></mrow></math><img file="EP0709999A2_D0133.tif" /></maths> the transfer function of Eq. (58) is expressed by the following equation (61) and the response signal to the input signal X(z) is <maths id="math0088" num=""><math display="inline"><mrow><mtext>Y(z)=X(z)G(z)</mtext></mrow></math><img file="EP0709999A2_D0134.tif" /></maths>.<maths id="math0089" num="(61)"><math display="block"><mrow><msup><mrow><mtext>G(z)={z</mtext></mrow><mrow><mtext>-2</mtext></mrow></msup><msub><mrow><mtext>+γ</mtext></mrow><mrow><mtext>1</mtext></mrow></msub><msub><mrow><mtext>(l+γ</mtext></mrow><mrow><mtext>2</mtext></mrow></msub><msup><mrow><mtext>)z</mtext></mrow><mrow><mtext>-1</mtext></mrow></msup><msub><mrow><mtext>+γ</mtext></mrow><mrow><mtext>2</mtext></mrow></msub><msub><mrow><mtext>}/{1+γ</mtext></mrow><mrow><mtext>1</mtext></mrow></msub><msub><mrow><mtext>(1+γ</mtext></mrow><mrow><mtext>2</mtext></mrow></msub><msup><mrow><mtext>)z</mtext></mrow><mrow><mtext>-1</mtext></mrow></msup><msub><mrow><mtext>+γ</mtext></mrow><mrow><mtext>2</mtext></mrow></msub><msup><mrow><mtext>z</mtext></mrow><mrow><mtext>-2</mtext></mrow></msup><mtext>}</mtext></mrow></math><img file="EP0709999A2_D0135.tif" /></maths> An advantage of the lattice-type filter configuration in the present invention is that γ₁ and γ₂ can be handled as parameters of independent meanings. When the group delay characteristic of the transfer function G(z) of Eq. (61) has a peak, the peak frequency depends virtually only on γ₁ and the steepness of the peak depends virtually only on γ₂. That is to say, there is no need of varying all of the filter coefficients with time; for instance, only by causing γ₁ corresponding to the peak frequency of the group delay characteristic to undergo a different variation for each channel, effective time-variant characteristics can be obtained and the filter configuration can be simplified. By making these time-variant characteristics differ with the channels, it is possible to produce the effect of varying the cross-correlation between the original multi-channel received signals.

(b) Concrete Example of Implementation by Signal Multiplication Processing

A description will be given of functions u₁(k), u₂(k), ..., uN(k) by which the received signals x₁(k), x₂(k), ..., xN(k) of the respective channels multiplied for conversion into signals <maths id="math0090" num=""><math display="inline"><mrow><mover accent="true"><mrow><mtext>x</mtext></mrow><mo>¯</mo></mover></mrow></math><img file="EP0709999A2_D0136.tif" /></maths>₁(k), <maths id="math0091" num=""><math display="inline"><mrow><mover accent="true"><mrow><mtext>x</mtext></mrow><mo>¯</mo></mover></mrow></math><img file="EP0709999A2_D0137.tif" /></maths>₂(k), ..., <maths id="math0092" num=""><math display="inline"><mrow><mover accent="true"><mrow><mtext>x</mtext></mrow><mo>¯</mo></mover></mrow></math><img file="EP0709999A2_D0138.tif" /></maths>N(k) which satisfy Eq. (55).

If the functions u₁(k), u₂(k), ..., uN(k) take both positive and negative values in Eq. (55), it is considered that the quality of the received signals would seriously be degraded regardless of the magnitude of the variations in the functions. Hence, it is desirable that the functions u₁(k), u₂(k), ..., uN(k) be always take positive or negative values and that their variations be also limited. Now, consider the case where the functions u₁(k), u₂(k), ..., uN(k) are in such forms as mentioned below. Let the functions u₁(k), u₂(k), ..., uN(k) be expressed by the following equation (62) using different functions ω₁(k), ω₂(k),..., ωN(k) each having the maximum amplitude normalized to 1, scalar values k₁, k₂, ..., kN (0<∥k₁∥, ∥k₂∥, ..., ∥kN∥<1) and gain coefficients A₁, A₂, ..., AN.<maths id="math0093" num=""><img file="EP0709999A2_D0139.tif" /></maths> Since at given time k the following equation holds, the signs of the values u₁(k), u₂(k),.., uN(k) remain unchanged.<maths id="math0094" num=""><math display="block"><mrow><msub><mrow><mtext>-1<k</mtext></mrow><mrow><mtext>1</mtext></mrow></msub><msub><mrow><mtext>ω</mtext></mrow><mrow><mtext>1</mtext></mrow></msub><msub><mrow><mtext>(k), k</mtext></mrow><mrow><mtext>2</mtext></mrow></msub><msub><mrow><mtext>ω</mtext></mrow><mrow><mtext>2</mtext></mrow></msub><msub><mrow><mtext>(k),..., k</mtext></mrow><mrow><mtext>N</mtext></mrow></msub><msub><mrow><mtext>ω</mtext></mrow><mrow><mtext>N</mtext></mrow></msub><mtext>(k)<1</mtext></mrow></math><img file="EP0709999A2_D0140.tif" /></maths> From the viewpoint of the psycho-acoustic balance, it is desirable that the values of the gain coefficients A₁, A₂, ..., AN be equal or take values of the same sign; in this instance, the functions u₁(k), u₂(k), ..., uN(k) all take values of the same sign. Furthermore, the amount of distortion of each processed signal can be adjusted by the scalar values k₁, k₂, ..., kN.

Moreover, from the viewpoint of a frequency shift, by setting the functions u₁(k), u₂(k), ..., uN(k) as follows:<maths id="math0095" num=""><img file="EP0709999A2_D0141.tif" /></maths> and by shifting the frequency characteristics of the received signals x₁(k), x₂(k), ..., xN(k) through the use of functions ω₁(k), ω₂(k), ..., ωN(k) which have different time-variant characteristics and vary within certain limited ranges, the variation in the cross-correlation can be added to the original multi-channel received signals.

A description will be given of functions n₁(k), n₂(k), ..., uN(k) to which the received signals x₁(k), x₂(k), ..., xN(k) of the respective channels are added for conversion into signals <maths id="math0096" num=""><math display="inline"><mrow><mover accent="true"><mrow><mtext>x</mtext></mrow><mo>¯</mo></mover></mrow></math><img file="EP0709999A2_D0142.tif" /></maths>₁(k), <maths id="math0097" num=""><math display="inline"><mrow><mover accent="true"><mrow><mtext>x</mtext></mrow><mo>¯</mo></mover></mrow></math><img file="EP0709999A2_D0143.tif" /></maths>₂(k), ..., <maths id="math0098" num=""><math display="inline"><mrow><mover accent="true"><mrow><mtext>x</mtext></mrow><mo>¯</mo></mover></mrow></math><img file="EP0709999A2_D0144.tif" /></maths>N(k) which satisfy Eq. (56).

From the viewpoint of the psycho-acoustic influence, it is desirable that the functions n₁(k), n₂(k), ..., nN(k) be moderately smaller in amplitude value than the received signals x₁(k), x₂(k), ..., xN(k) of the respective channels. In view of this, different functions nr1(k), nr2(k), ..., nrN(k), normalized so that the maximum amplitude becomes a rated value Ar of the amplitude of the received signals x₁(k), x₂(k), ..., xN(k) of the respective channels, are multiplied by scalar values λ₁,λ₂, ..., λN, by which the functions n₁(k), n₂(k), ..., nN(k) are expressed as follows:<maths id="math0099" num=""><img file="EP0709999A2_D0145.tif" /></maths> Based on the scalar values λ₁, λ₂, ..., λN, it is possible to determine the degree to which the magnitude of the signal to be added to each of the received signals x₁(k), x₂(k), ..., xN(k) is made smaller than the rated value Ar.

(d) Concrete Example e of Implementation by Pitch Shift Processing

A description will be given of an embodiment in which the received signals x₁(k), x₂(k), ..., xN(k) of the respective channels are converted into signals <maths id="math0100" num=""><math display="inline"><mrow><mover accent="true"><mrow><mtext>x</mtext></mrow><mo>¯</mo></mover></mrow></math><img file="EP0709999A2_D0146.tif" /></maths>₁(k), <maths id="math0101" num=""><math display="inline"><mrow><mover accent="true"><mrow><mtext>x</mtext></mrow><mo>¯</mo></mover></mrow></math><img file="EP0709999A2_D0147.tif" /></maths>₂(k), ..., <maths id="math0102" num=""><math display="inline"><mrow><mover accent="true"><mrow><mtext>x</mtext></mrow><mo>¯</mo></mover></mrow></math><img file="EP0709999A2_D0148.tif" /></maths>N(k) by subjecting the frequency characteristic of each received signal to different time-variant frequency axis expansion/compression processing or pitch shift processing.

The pitch shift processing involves time expansion/compression as well as frequency axis expansion/compression and can be processed in the time domain. When the time axis is compressed, interpolation processing is needed to make the duration of the processed signal equal to the duration of the original signal, whereas when the time axis is expanded, the duration of the processed signal becomes longer than the duration of the original signal; hence decimation processing is needed. Such interpolation/decimation processing can be implemented by detecting a silent duration and extending or removing it.

As described above, the methods of the present invention can be applied to the cancellation of echoes to implement a teleconference system of highly realistic presence which permits transmission of acoustic spatial information from either point by using a multi-channel transmission system and terminal equipment provided with multi-channel microphone and loudspeaker systems between two point.

Now, consider a stereo teleconferencing system which connects two points via a two-channel transmission system and uses two-channel microphone and loudspeaker systems as shown in Fig. 4. For instance, when a plurality of speakers alternately speak in their seats at the point A, the stereo speech signals that are picked up by two microphones bear a cross-correlation therebetween dependent on the positional relationship between the speakers and the microphones. When these speech signals are received at the point B and inputted to the conventional echo canceller, the echo path estimation depends on the cross-correlation and hence is incorrect, with the result that a large echo returns to the point A each time the speaker changes. In contrast to this, when the echo cancellation method of the present invention, a variation in the cross-correlation between the stereo received signals by the change of the speaker is extracted and used to correct an error in the echo path estimation; thus, it is possible to suppress the echo from increasing whenever the speaker changes. Furthermore, in the case where speech uttered by one person is picked up by a plurality of microphones, even if the speaker speaks at one place, the cross-correlation between the speech signals picked up by the microphone delicately changes. The method of the present invention permits effective utilization of such a slight variation in the cross-correlation between signals.

Fig. 13 is a graph showing the results of computer simulations using stereophonically picked up speech uttered by a speaker with his body or head fixed; the prior art method and this invention method are compared in terms of the error vector between the combined estimated echo path vector generated in the echo path estimating part and the combined true echo path vector as indicated by the curves 13A and 13B. This invention method used the secondary projection algorithm for the extraction of the variation in the cross-correlation. From Fig. 13 it will be seen that this invention method captures even a delicate variation in the cross-correlation between signals, permitting the correction of an error in the echo path estimation. It can be said that the present invention effectively detects a delicate variation between channels through utilization of past data.

Fig. 14 is a graph showing the convergence of the estimated echo path vector in the case of stereophonically picked up speech as in the case of Fig. 13, the order p of the projection algorithm being used as a parameter. The number L of taps was set to 500+500 and the step size α was set to 0.5. This graph also indicates that the present invention is effective.

Other than a method according to which at a terminal provided with a stereo receiving system, two-channel stereo signals picked up at the other side are received and reconstructed intact as shown in Fig, 4, there is proposed a method for multi-point teleconference by which the received signal for each point is freely subjected to sound image localization at the receiving side to provide a comfortable receiving environment. The method of the present invention is also applicable to such multi-point teleconferencing terminals. Fig. 15 illustrates the configuration of teleconferencing system for four points interconnected via a network. At each point, the sending or microphone system is one-channel (monaural). Now, the point D will be described. Let it be assumed that at the point D, received signal from the points A, B and C are subjected to sound image localization in a received signal processing part 13d so that the signals are localized at the right, center and left with respect to the point D, respectively, by which new two-channel stereo received signals are generated for stereo reconstruction by loudspeakers 12dl and 12d₂. Also in this case, if the method of the present invention is used, when the speaker changes to a person at a different point, the echo canceller 22d extracts a variation in the cross-correlation between the stereo received signals and corrects the error as in the above-mentioned case; hence, it is possible to suppress the echo from increasing, as compared with the conventional method.

Incidentally, the curves 16A and 16B in Fig. 16 show the convergence of the error vector in the case of using, as received signals, two-channel signals obtained by picking up speech uttered by one speaker with one microphone and applying the speech through two different time-invariant filters on a computer, that is, in the case where the cross-correlation between the received signal is constant. In this instance, even if the first or second method of the present invention is employed to extract a variation in the cross-correlation between the received signals, since the cross-correlation between the received signals does not vary, echo path estimation is made incorrectly, and hence the effect of echo cancellation is not heightened even compared with that by the conventional method.

By the application of the third method of the present invention, the cross-correlation between the received signals always varies and the echo path estimation is made at all times. The configuration therefor is the same as shown in Fig, 11, and the N-channel echo cancellers 22₁, 22₂, ..., 22M are identical in construction with that shown in Fig. 6. The curves 17A and 17B in Fig. 17 indicates the results of computer simulations done on the method of extracting and using the variation in the cross-correlation between received signals in the cases where the same received signals as those used in the case of Fig. 16, that is, stereo signals of a fixed cross-correlation therebetween, were added with and not added with the variation in the cross-correlation, the error vector between the true and estimated echo paths being plotted with the elapse of time. The curve 17A for the case of no variation in the cross-correlation being added is the same as the curve 16B in Fig. 16; in this case, no convergence to the true echo path does not take place. On the other hand, in the case where the cross-correlation variation was added, the convergence to the true echo path proceeded and excellent results (the curve 17B) were obtained as is the case with the curve 13B in Fig. 13. The addition of the variation in the cross-correlation to the stereo (two-channel) signals is performed by the signal multiplication processing indicated by Eq. (55), and u₁(k) and u₂(k) are set as follows:<maths id="math0103" num=""><math display="block"><mrow><mtable><mtr><mtd><mrow><mtable><mtr><mtd><mrow><mtable><mlabeledtr><mtext>(65)</mtext><mtd><mrow><msub><mrow><mtext>u</mtext></mrow><mrow><mtext>1</mtext></mrow></msub><mtext>(k)=1+0.05ω(k)</mtext></mrow></mtd></mlabeledtr></mtable></mrow></mtd></mtr><mtr><mtd><mrow><mtable><mlabeledtr><mtext>(66)</mtext><mtd><mrow><msub><mrow><mtext>u</mtext></mrow><mrow><mtext>2</mtext></mrow></msub><mtext>(k)=1-0.05ω(k) </mtext></mrow></mtd></mlabeledtr></mtable></mrow></mtd></mtr></mtable></mrow></mtd></mtr></mtable></mrow></math><img file="EP0709999A2_D0149.tif" /></maths> where ω(k) is a white noise signal having a maximum amplitude of 1.

That is to say, in the case where the conventional echo cancellation method is applied to a teleconferencing system usually composed of a multi-channel loudspeaker system and one or more microphone channels, when received signals of respective channels have a cross-correlation therebetween, incorrect echo path estimation is made, with the result that the echo increases upon each variation in the cross-correlation between the received signals. The first or second method of the present invention extracts and utilizes the variation in the cross-correlation between received signals to correct an error in the echo path estimation, and hence is effective in settling the above-said problem.

The third method of the present invention has the function of adding a variation to the cross-correlation between received signals, and hence does not completely freeze at a wrong solution in the echo path estimation but continues the echo path estimation in the direction in which to reduce the error; thus, the third method also effective in solving the problem. With the combined use of the first or second and third methods, the first or second method effectively works when the cross-correlation between received signal varies, and when the cross-correlation does not vary, the third method effectively works to make the correction of the estimated echo path to converge in a short time.

It will be apparent that many modifications and variations may be effected without departing from the scope of the novel concepts of the present invention.

Contents4

234 sheets

Sheet 1
Sheet 2
Sheet 3
Sheet 4
Sheet 5
Sheet 6
Sheet 7
Sheet 8
Sheet 9
Sheet 10
Sheet 11
Sheet 12
Sheet 13
Sheet 14
Sheet 15
Sheet 16
Sheet 17
Sheet 18
Sheet 19
Sheet 20
Sheet 21
Sheet 22
Sheet 23
Sheet 24
Sheet 25
Sheet 26
Sheet 27
Sheet 28
Sheet 29
Sheet 30
Sheet 31
Sheet 32
Sheet 33
Sheet 34
Sheet 35
Sheet 36
Sheet 37
Sheet 38
Sheet 39
Sheet 40
Sheet 41
Sheet 42
Sheet 43
Sheet 44
Sheet 45
Sheet 46
Sheet 47
Sheet 48
Sheet 49
Sheet 50
Sheet 51
Sheet 52
Sheet 53
Sheet 54
Sheet 55
Sheet 56
Sheet 57
Sheet 58
Sheet 59
Sheet 60
Sheet 61
Sheet 62
Sheet 63
Sheet 64
Sheet 65
Sheet 66
Sheet 67
Sheet 68
Sheet 69
Sheet 70
Sheet 71
Sheet 72
Sheet 73
Sheet 74
Sheet 75
Sheet 76
Sheet 77
Sheet 78
Sheet 79
Sheet 80
Sheet 81
Sheet 82
Sheet 83
Sheet 84
Sheet 85
Sheet 86
Sheet 87
Sheet 88
Sheet 89
Sheet 90
Sheet 91
Sheet 92
Sheet 93
Sheet 94
Sheet 95
Sheet 96
Sheet 97
Sheet 98
Sheet 99
Sheet 100
Sheet 101
Sheet 102
Sheet 103
Sheet 104
Sheet 105
Sheet 106
Sheet 107
Sheet 108
Sheet 109
Sheet 110
Sheet 111
Sheet 112
Sheet 113
Sheet 114
Sheet 115
Sheet 116
Sheet 117
Sheet 118
Sheet 119
Sheet 120
Sheet 121
Sheet 122
Sheet 123
Sheet 124
Sheet 125
Sheet 126
Sheet 127
Sheet 128
Sheet 129
Sheet 130
Sheet 131
Sheet 132
Sheet 133
Sheet 134
Sheet 135
Sheet 136
Sheet 137
Sheet 138
Sheet 139
Sheet 140
Sheet 141
Sheet 142
Sheet 143
Sheet 144
Sheet 145
Sheet 146
Sheet 147
Sheet 148
Sheet 149
Sheet 150
Sheet 151
Sheet 152
Sheet 153
Sheet 154
Sheet 155
Sheet 156
Sheet 157
Sheet 158
Sheet 159
Sheet 160
Sheet 161
Sheet 162
Sheet 163
Sheet 164
Sheet 165
Sheet 166
Sheet 167
Sheet 168
Sheet 169
Sheet 170
Sheet 171
Sheet 172
Sheet 173
Sheet 174
Sheet 175
Sheet 176
Sheet 177
Sheet 178
Sheet 179
Sheet 180
Sheet 181
Sheet 182
Sheet 183
Sheet 184
Sheet 185
Sheet 186
Sheet 187
Sheet 188
Sheet 189
Sheet 190
Sheet 191
Sheet 192
Sheet 193
Sheet 194
Sheet 195
Sheet 196
Sheet 197
Sheet 198
Sheet 199
Sheet 200
Sheet 201
Sheet 202
Sheet 203
Sheet 204
Sheet 205
Sheet 206
Sheet 207
Sheet 208
Sheet 209
Sheet 210
Sheet 211
Sheet 212
Sheet 213
Sheet 214
Sheet 215
Sheet 216
Sheet 217
Sheet 218
Sheet 219
Sheet 220
Sheet 221
Sheet 222
Sheet 223
Sheet 224
Sheet 225
Sheet 226
Sheet 227
Sheet 228
Sheet 229
Sheet 230
Sheet 231
Sheet 232
Sheet 233
Sheet 234

Every citation, both ways

Citations
Document	Relation	Office	Cited during
US7555116B1	Cited by	United States of America	Applicant
FR2763774A1	Cited by	France	Search report
EP0944228A1	Cited by	European Patent Office (EPO)	Search report
US6556682B1	Cited by	United States of America	Applicant
WO9847276A1	Cited by	World Intellectual Property Organization (WIPO)	International search
US6553122B1	Cited by	United States of America	Applicant
US7925006B2	Cited by	United States of America	Applicant
FR2762467A1	Cited by	France	Search report
US7925008B2	Cited by	United States of America	Applicant
WO9853595A1	Cited by	World Intellectual Property Organization (WIPO)	International search
WO0145375A1	Cited by	World Intellectual Property Organization (WIPO)	International search
US6424720B1	Cited by	United States of America	Applicant
WO03007500A1	Cited by	World Intellectual Property Organization (WIPO)	Applicant
EP0561133A1	Cites	European Patent Office (EPO)	Search report
JP10241994A	Cites	Japan	Applicant
JP18474293A	Cites	Japan	Applicant
US5062102A	Cites	United States of America	Search report
JPH10241994A	Cites	Japan	Applicant

10 priority claims, no other members on record

This recordMember, by publication datePriority claim

Priority claims10

Priority claims
Document	Office	Kind	Date
26273894	Japan	–	1994-10-26
26273894	Japan	A	1994-10-26
26273894	Japan	A	1994-10-26
5000295	Japan	–	1995-03-09
5000295	Japan	A	1995-03-09
5000295	Japan	A	1995-03-09
26273894	–	–	–
5000295	–	–	–
JP19940262738	–	–	–
JP19950050002	–	–	–

21 legal events, as 3 offices reported them to INPADOC

Over the term

Point at a mark for the event

Events

Legal events, newest first
	Event	Code	Office
2016-01-29	Lapsed in a contracting state [announced via postgrant information from national office to epo]LapsedPG25	PG25	EP
2015-11-18	Patent expired after termination of 20 yearsExpiredPE20	PE20	GB
2015-10-24	Expiry of rightR071	R071	DE
2015-01-30	Annual fee paid to national office [announced via postgrant information from national office to epo]GrantedPGFP	PGFP	EP
2015-01-30	Annual fee paid to national office [announced via postgrant information from national office to epo]GrantedPGFP	PGFP	EP
2014-11-28	Annual fee paid to national office [announced via postgrant information from national office to epo]GrantedPGFP	PGFP	EP
2004-06-16	No opposition filedOpposition26N	26N	EP
2004-05-07	Fr: translation filedET	ET	EP
2004-04-30	No opposition filed within time limitOppositionORIGINAL CODE: 0009261PLBE	PLBE	EP
2004-04-30	Information on the status of an ep patent application or granted ep patentGrantedSTATUS: NO OPPOSITION FILED WITHIN TIME LIMITSTAA	STAA	EP
2003-07-31	Corresponds to:REF	REF	EP
2003-06-25	Designated contracting statesAK	AK	EP
2003-06-25	European patent grantedGrantedFG4D	FG4D	GB
2003-05-09	(expected) grantORIGINAL CODE: 0009210GRAA	GRAA	EP
2003-04-23	Despatch of communication of intention to grant a patentORIGINAL CODE: EPIDOS IGRAGRAH	GRAH	EP
2002-12-06	Despatch of communication of intention to grant a patentORIGINAL CODE: EPIDOS IGRAGRAH	GRAH	EP
1997-05-14	Designated contracting statesAK	AK	EP
1997-03-27	Search report despatchedORIGINAL CODE: 0009013PUAL	PUAL	EP
1996-05-01	Request for examination filed17P	17P	EP
1996-05-01	Designated contracting statesAK	AK	EP
1996-03-15	Public reference made under article 153(3) epc to a published international application that has entered the european phaseORIGINAL CODE: 0009012PUAI	PUAI	EP