JPS6172420A - Multi-path echo erasure system - Google Patents

Abstract

PURPOSE:To attain multi-channel echo erasure by providing a private echo path characteristic parameter storage circuit at each echo path so as to obtain each echo characteristic parameter. CONSTITUTION:The correcting amount is calculated by using a correction amount calculation circuit 32 and a reception power calculation circuit 34 from a reception signal and a residual signal in order to obtain each echo characteristic to an echo signal of each channel. Then each echo characteristic parameter storage circuit 31 uses each correcting amount to obtain a sequential echo characteristic parameter at each channel and store its content. Further, a convolution integration circuit 33 applies convolution integration between an output signal of the storage circuit 31 and the output signal of a reception signal storage circuit 29 to obtain respectively an approximate echo signal. The approximate echo signal is used to erase the echo signal from a speaker 25 to a microphone 20. Thus, echo erasure of multi-channel is attained.

Description

DETAILED DESCRIPTION OF THE INVENTION [Industrial Field of Application] The present invention is used in stereo echo cancellation devices, multi-site audio conference echo cancellation devices, and the like. The present invention relates to a multi-channel echo cancellation method that aims to remove all the echo signals in a system in which a plurality of received signals pass through a plurality of echo paths and turn into a transmitted signal as the sum of the echo signals. It is something.

[Conventional technology]

There is a conventional method shown in FIG. 7 as a method for canceling echo signals on only one path in normal monaural communication.

Examples of this type of system include 1988 Institute of Electronics and Communication Engineers General Conference No. 2034 ``A Study of Echo Cancellation Methods for Teleconference Rooms'' and IEICE Electroacoustic Research Group EA83-11 ``Echo Control System in Teleconferences''. ” and so on.

FIG. 7 shows a configuration in which there is a monaural communication system using microphones and speakers between ground points A and B, and echo signals from only the negative path of the monaural communication system are eliminated. An audio signal input from a microphone 1 on the ground A is sent to a transmission line 3 via an amplifier circuit 2 suitable for the transmission line.

At ground B, the received signal from ground A is amplified by an amplifier circuit 5 and a speaker 4 is driven. The audio signal from the speaker 4 is received by the microphone 7 via the reverberation path 6 of the room. In this way, the audio signal input from the microphone 1 on the ground A is transmitted through the microphone 1 -> amplifier circuit 2 - transmission line 3 - amplifier circuit 5 - speaker 4 - echo path 6 - microphone 7 - amplifier circuit as shown in Fig. 7. 8-transmission line 9-amplification circuit 10-speaker 11-reverberation path 12-microphone 1 loop. Therefore, even if no audio signal is input from the microphone 1 on the ground A, when the gain of the loop is 1 or more, a so-called howling phenomenon occurs.

In order to prevent this, an echo cancellation circuit 13 is provided which creates an approximate echo signal equivalent to the echo signal via the amplifier circuit 5 - speaker 4 - echo path 6 - microphone 7 - amplifier circuit 8 and subtracts the echo signal. Similarly, an echo canceling circuit 14 is provided which creates an approximate anti-reverse signal equivalent to the echo signal via the amplifier circuit 10 - speaker 11 - echo path 12 - microphone 1 - amplifier circuit 2, and subtracts the echo signal (details). (see the above-mentioned document).

In general, as shown in FIG. 2, the monaural communication echo cancellation circuit 14 includes a reception signal storage circuit 15 that stores a reception signal X (tl) received from a remote location via a transmission path 9, and a reception signal storage circuit 15 that stores a reception signal A subtraction circuit 16 for calculating the difference signal E (tl) between the echo signal Y(t) input to the microphone 1 via the microphone 12 and the approximate echo signal ?(L), and the residual signal E
(t) and the received signal
−−−−−−−.

M-1) Reverberation characteristic parameters to be corrected and memorized・[@
It consists of a circuit 18 and a convolution integrating circuit 19 that generates an approximate echo signal (t) from the received signal It is expressed by the following formula, and is generally used as a learning identification method algorithm (J, Nagumo and Noaa
: Learning Method for S
system Identification” [EEE Trans, AC-12, 3P282.
June, 1976). however,
t represents time.

E (U = Y ft) -? (t)
・−・・・・・・・・・(1) Δold (t)
=α・E (t)・X(t-i) /Old(t)=Old(t
−1) Ten Δ old (tl −−−−１・・・−・−
(3) Here α indicates a correction coefficient.

In this way, approximate echo signal? (t) is the received signal X (t
An approximate echo signal ♀ is created from l and is approximated from the echo signal Y (t).
(The echo signal can be canceled by subtracting tl.

Now, when an external noise signal N1n (for example, indoor noise, external voice, etc.) that has no correlation with X (t) is inserted into the echo path 12, steady-state echo cancellation is used to evaluate the transmission characteristics of echo cancellation! (ERLE) is ERLE (d
B) =101og ((2-α) /α]+110
1o ((power M of echo signal Y)/(external noise signal N
(Electric energy of in)] −−−]1−−−・・・−(5)
5. P2O3-P312, May 1960).

If the external noise signal Nin is present in the echo path and its amount is large, as is clear from the second term on the right side of equation (5), the echo cancellation l (IERLE) decreases, which is required from the viewpoint of speech quality. It becomes impossible to obtain the desired amount of echo cancellation.

Regarding the correction coefficient α, it converges to a steady value fastest when α = 1, and by making it smaller than l, the echo cancellation f
It is known that fi (ERLE) can be further increased. That is, by changing the correction coefficient α, the values of the convergence time and the amount of echo cancellation change. This is also stated in Noda's paper mentioned above.

On the other hand, in audio conference communication, stereo communication has come to be used because it is desired to have a sense of realism similar to that of a general conference room. In this case, two systems of speakers and microphones are required. Multi-point audio conference calls, which connect three or more conference rooms into one, have also become available. In this case, it is desirable to install as many speakers as the number of speakers to identify the ground. However, with these devices, 2
This results in multiple echo signals coming from more than one speaker, and a technique for simultaneously canceling the multiple echo signals is required.

In this way, when there is echo signals coming from multiple speakers, if we simply apply the conventional monaural echo cancellation circuit shown in Figure 8 and look at only one end of the ground A, we can see that the model is as shown in Figure 3. That is, the audio transmission signal Sin input to the microphone 20 is sent to the transmission transmission line 22 via the amplifier circuit 21. Receive transmission line 23 on the first channel. The signal R11n received from
are connected to the speaker 25 . via the amplifier circuit 24 . to drive. Also in the second channel, the signal R21n received from the receiving transmission line 23□ is sent to the speaker 25□ via the amplifier circuit 24□.
to drive. All channels below are configured similarly,
In the Nth channel, the receiving transmission line 23. The signal RNin received by the amplifier circuit 24N drives the speaker 25°. Therefore, the microphone 20 is connected to the first channel speaker 25. More echo path 26. The echo signal T
l, and a reverberation path 26 from the second channel speaker 25□.
□, and further the N-th channel speaker 25. More echo path 26. A total of N types of echo signals up to the echo signal TN are manually generated. Therefore, a total of N types of echo signals (T I
, T 2 , --------, TN ) 27. ．． 27□,・----1・,27,
are required respectively. However, the echo cancellation circuit 272.
27□, -・-・, 27. has the same configuration as the echo cancellation circuit shown in FIG.

Here, considering the case of two channels (N=2), first, the first channel received signal R11n is converted to X 1 (t
), the second channel received signal R2in is X 2 (t),
Let the signal Sin input to the microphone 20 be Y(t), the output of the echo canceling circuit 271 be E1(tl), the output power of the echo canceling circuit 27□ be E2(tl, and the transmitted signal S0 be E(1)). Let the echo characteristic estimation parameter in the echo cancellation circuit 271 be Hli (tl (where, i = 0.1.----, Ml-1: Ml is the impulse response length), and the echo characteristic estimation parameter in the echo cancellation circuit 27 □ The reverberation characteristic estimation parameters in H2j(t) (where j=
0.1. --------, M2-1: M2 is impulse response length). Approximate signal of echo signal Tl? 1(11), and the approximate signal of the echo signal T2 is ?2(1).

In this case, the operational formula for a two (N=2) channel echo cancellation circuit, which is a simple application of the echo cancellation circuit for monaural communication, is expressed as follows.

E 1 (tl= Y(t)-tl (tl
・−・−−−1−−−(61ΔH1i(tl=α・
E 1 (t)・X 1 (t-1) /H1i(tl
= )I li (t-1) +ΔH1i (tl
−−−−−−−(8) (However, i = 0
, 1 , --------, M 1 1 )E 2
(t) = E 1 (tl-?2 (t)
−−−−−・−・−00)ΔH2i(tl=α・E
2 (tl・X 2 (t-i) /H2i(t)-H
2i (t-1) +ΔH2i (tl −-
---03 (however, i=o, 1.・・・・・−, Mz
-t)E(t)=E1(t)
(2) As mentioned above, when an external signal is inserted into the echo path, it cannot be erased if there is no correlation between that signal and the received signal. Therefore, in equation (6), the echo canceling circuit 271 cannot remove the echo signal T2 from the second channel received signal, which is uncorrelated with A similar action is performed on the cancellation circuit 27. Therefore, as is clear from 2(5), the echo signal T2 from the second channel received signal is
Accordingly, the anti-go two circuit 27. The amount of echo cancellation during adaptive control decreases, making it impossible to obtain the required amount of echo cancellation. In this way, echo signals from received signals of other channels cause echo cancellation circuit 27 located at the first stage. In this case, the deterioration of the anti-field 1 cancellation characteristic as described above occurs.

This also applies when the number of channels increases to 3 or more, and in equation (6), X 1 mixed into Y (t)
(Since the number of echo signals from other channels uncorrelated with tl increases, the characteristics of the echo cancellation circuit located at the first stage will further deteriorate.

In general, the adaptive operation of the echo canceller is performed only when there is no voice transmission signal from the speaker and there is a received signal. That is, as shown in FIG. 9, when the speaker's voice transmission signal is input to the microphone 20 (double talk state), the echo cancellation circuit 27. ．． The adaptive operation of 27□, . . . -227, is stopped, and the state immediately before the adaptive operation is stopped is maintained. Therefore, it becomes even more important to obtain the required amount of echo cancellation faster during adaptive operation.

[Problem that the invention seeks to solve]

From the above, the monaural communication echo cancellation circuit 27, .
A multi-channel echo canceler simply combining 27□, -..., 21N cannot cancel the sum of echo signals circulating from multiple channels like a plurality of speakers.

The present invention solves this problem, and aims to provide an echo cancellation method that cancels the sum of echo signals generated in multiple paths.

[Means for solving problems]

The first aspect of the present invention is a communication system in which two or more received signals generate echo signals passing through two or more echo paths, and the sum of these echo signals is input to the transmission signal, and an approximation is generated to the transmission signal. In an echo cancellation method for subtracting an echo signal, means for storing each of two or more received signals received from one or more remote locations via a transmission line, and inputting each correction amount for each of the received signals, means for respectively modifying and storing each echo characteristic parameter of each echo signal that goes around from each of the received signals to the transmitted signal; means for generating respective approximate echo signals; residual calculating means for outputting a residual signal obtained by subtracting the sum of the respective approximate echo signals from the transmitting signal; □1 The apparatus is characterized by comprising a correction amount calculating means for calculating each of the correction amounts for each of the received signals using a correction coefficient.

The second invention of the present invention includes, in addition to the configuration of the first invention,
means for generating a training signal in advance before starting communication; switching means for sequentially transmitting the training signal to the echo path as each of the received signals; and while transmitting the training signal to the echo path as the received signal; , means for blocking transmission of the transmission signal to the transmission path; and after calculating the respective echo signal characteristic parameters based on the training signal, connecting the transmission path and the received signal by the switching means, and opening the blocking means. The present invention is characterized by comprising a control means for sending a transmission signal to a transmission path.

. [Function] Has a dedicated echo characteristic parameter storage circuit for each echo path, uses a common residual signal, and uses a correction coefficient and correction amount calculation circuit controlled for each channel to calculate each echo characteristic parameter. This realizes multi-channel echo cancellation.

〔Example〕

FIG. 1 is a block diagram of an embodiment of the present invention. This circuit is characterized by the configuration of the multi-channel echo cancellation circuit 28.

First, looking at the first channel, the code 29. 30. is a received signal storage circuit; is a correction coefficient control circuit, code 31
．． 32. is a reverberation characteristic parameter storage circuit for the reverberation signal TI; Reference numeral 331 indicates a correction amount calculation circuit for estimating the echo signal TI, and a superimposition integrating circuit 331 obtains an approximate echo signal of the echo signal T1. Next, looking at the second channel,
Reference numeral 29□ represents a received signal storage circuit, and reference numeral 30. 31□ is a correction coefficient control circuit, 31□ is an echo characteristic parameter storage circuit for the echo signal T2, 32□ is a correction amount calculation circuit for estimating the echo signal T2, and 33□ is an approximate echo signal of the echo signal T2. This is the convolution integrator circuit obtained. All the channels below are constructed in the same way, and for the Nth channel as well, reference numeral 298 is a received signal storage circuit, reference numeral 30 is a correction coefficient control circuit, reference numeral 31. 32.4 is a correction amount calculation circuit for estimating the echo signal TN, and 33N is the echo characteristic parameter storage circuit for the echo signal TN.
This is a superimposed integration circuit that obtains N approximate echo signals.

Further, reference numeral 34 denotes a power calculation circuit for all received signals;
Reference numeral 35 is a circuit that calculates the sum of all approximate echo signals in each channel, and reference numeral 36 is a circuit that calculates the sum of approximate echo signals in each channel.
This is a subtraction circuit that subtracts from

Next, to explain the operation of this circuit, first, the received signal of the k-th (k=1.2.-.-, N) channel is
k (t), and the echo signal Tk corresponding to the channel
approximating the signal? Let k (tl be the first channel reverberation characteristic parameter Hk i (t), and let its impulse response length be Mk. Subscript (i-=0.1゜----
-., Mk-1) represents an element of the echo characteristic parameter. The input signal Sin of the microphone 20 is assumed to be Y (t), and the transmitted signal S0 is assumed to be E ([).

At this time, the operational formula for the N-channel echo cancellation circuit when applied is ΔHki(t)=α・E(tl・Xk(L−i)
/HkHtl = Hki (t-1) +ΔHk
i (t) −−−−−−−・− surface, where k
=1.2.・−・・−・−1N1i = 0. 1
, -------, becomes Mk-1.

In the present invention, in order to overcome the drawbacks of a multi-channel echo canceler that is simply a combination of monaural communication echo cancelers, 209 is used in place of 2 (6) and 2 aφ, and subtraction circuits 35 and 36 are used. I am looking for the residual signal. By using this residual signal in common and controlling the echo cancellation circuit for each channel, the received signal from the other channel that is uncorrelated with the mixed X 1 (t) in 2 (6) is Echo signal T 2 , -------・-1,, I
This makes it possible to remove TN at the same time. As a result, the amount of echo cancellation necessary for improving speech quality is ensured.

In the circuit shown in Figure 1, first, the echo signals T1 and T2 of each channel are
．．・・・In order to obtain each echo characteristic parameter for TN, the correction amount calculation circuit 32 and the reception power calculation circuit 3 are used from the received signal and the residual signal as shown in 2αG.
4 to calculate the correction amount. Next, as shown on the second side, each reverberation characteristic parameter storage circuit 31 sequentially obtains the reverberation characteristic parameter Hki(t) for each channel from the respective correction amounts and stores its contents. Further 2 OB
Accordingly, the superposition integration circuit 33 performs superposition integration of the output signal of the echo characteristic parameter storage circuit 31 and the output signal of the received signal storage circuit 29 for each channel, and obtains an approximate echo signal '?
k (tl can be obtained respectively, where '
9 k (t) is used to cancel the echo signal Tk from the channel speaker 25 to the microphone 20.

In the embodiment circuit shown in FIG. 1, the correction amount calculation circuit 32 is linearly coupled through the circuit 34, and each reverberation characteristic parameter is calculated using a learning identification method. Therefore, formula αG is used in place of formula (7) and formula αυ. This provides convergence stability. The correction coefficient α is 0 < cX < 1.0 as in the learning identification method.
' -----------(It becomes 11.

FIG. 2 shows an example of the results of a simulation using this example circuit. This is an example of a two (N=2) channel echo cancellation circuit. For comparison, the figure also shows the results of a simulation conducted under the same conditions using the conventional circuit shown in FIG. However, the correction coefficients are both 0.5. As is clear from the figure, the conventional circuit shown in FIG. 9 can only obtain an insufficient amount of echo cancellation of about 10 dB, whereas the circuit according to the embodiment of the present invention has a sufficient amount of approximately 35 dB for speech quality. The amount of echo cancellation obtained is obtained. The effectiveness of the present invention can be confirmed through this simulation.

FIG. 3 is a block diagram of another embodiment of the present invention. The fourth mentioned above
In the multi-channel echo cancellation device having the circuit configuration shown in the figure, if there is a difference in power between the received signals Rkin of each channel, R
The ratio of the maximum eigenvalue to the minimum eigenvalue of the correlation matrix which is an input signal obtained by linearly combining kin becomes 1.0 or more. In this way, when the ratio of eigenvalues becomes 1.0 or more, the convergence speed slows down and it takes a long time to obtain the required amount of echo cancellation (
Regarding this theory, see Widrow, B., et at
, :^dapttvenoise cancelling
: pr, 1nciples and applic
cations'+Proc, IEEE, 63.
12. pp, 1692. See Dec 1975).

The circuit configuration shown in FIG. 3 is a method for improving the deterioration of the convergence speed due to the power difference.

In the present invention, by normalizing the received signal X k (t) for each channel using the power of each received signal of each channel, the above-mentioned eigenvalue ratio is removed and the convergence speed is improved. Along with this normalization, a constraint condition is attached to the correction coefficient depending on the impulse response length of the echo characteristic parameter.

The circuit configuration shown in FIG. 3 differs from the circuit configuration shown in FIG. The difference lies in the operation of the correction coefficient control circuit 30 that generates the coefficients.

The operational formula for the response of the N-channel echo cancellation circuit 37 shown in FIG. 3 is E (tl = Y (11-Σ'9 k ftl
−・−(20) ΔHki(t) = α to −E(t
l・Xk(t-i)/Hki(tl = Hki (t
−1) +ΔHki[tl −・−・−(22)
However, k = 1, 2, --------, N,
i = 0, 1, -----, M k4. The operating equation at this time of adaptation differs from the operating equation of the circuit in FIG. 1 in equation (21). This means that the operations of the correction amount calculation circuit 38 and the power calculation circuit 39 are different. That is, the received signal X k (t) is normalized for each channel by the multiplier in (21). , and the convergence coefficient α is α1 :α2
Knee...-:αN=M1;M2:・---
−−・−・・：MN −・
--(25) and is set in proportion to the impulse response length of the reverberation characteristic parameter of each channel. As a result, there appears to be no difference in power between the received signals RkinO of each channel. Therefore, the ratio of the eigenvalues of the correlation matrix is removed and the convergence speed is improved. Note that α is controlled by a correction coefficient control circuit 30k, and is set so that the sum of both is 1.0 or less in order to guarantee stability of convergence.

0kuΣαk<1.0 ・−・−・−
・(26) - to military 1 - These setting methods are derived from the setting conditions of the correction coefficient associated with normalization according to the magnitude of the human signal (Fujii, Haragi, Miyayo, ``Frequency sampling filter “Transmission path adaptation, etc.”, Transactions of the Institute of Electronics and Communication Engineers B, J 65-8, 9.
p, 1172. (September 1982). Note that instead of using the power calculation circuit 34, the magnitude of the signal of each channel may be determined using an amplitude value or the like.

FIG. 4 shows an example of the results of a computer simulation using the example circuit of FIG. 3, which is 2 (N=2
) is an example of a channel echo cancellation circuit. In the figure, the ratio of the average received signal power of the first channel and the second channel is 1.
0: Set to 1.

However, the impulse response lengths of the echo characteristic parameters are set equal, and the ratio of Ml and M2 is 1:1. Therefore, α1=α2=0.5 is set so that α1 and α2 have the same ratio. For comparison, the figure also shows the results of a simulation conducted under the same conditions using the circuit shown in Figure 1.

As is clear from Fig. 4, the received signal X of each channel
It can be seen that the convergence speed is improved by normalizing k (t) using each power. This confirms the effectiveness of the present invention.

FIG. 5 shows a configuration diagram of an embodiment of the stereo echo canceling device of the present invention. In FIG. 5, a plurality of audio signals manually inputted to the microphone 40 are edited into stereo by a mixing device 41, and are sent to the left channel transmission transmission line 421 and the right channel as a left audio transmission signal Loi and a right audio transmission signal Rot. It is sent out to the transmission transmission path 42r. Left channel reception transmission line 437! The signal Lin received from the right channel receiving transmission line 43r is supplied to the left speaker 45β via the amplifier circuit 44N, and the signal Rin received from the right channel reception transmission line 43r is supplied to the right speaker 45r via the amplifier circuit 44r.

The left channel transmission signal Loi has two types of echo paths 4611 from the left speaker 451 and the right speaker 45r! , 46
Similarly, the echo signals LN and Lr via βr are transmitted to the right channel transmission signal Roi via two types of echo paths 46rf and 46r.
Reverberation signals R1 and Rr are input via r. Therefore, an echo canceling circuit 47 is required to cancel a total of four types of echo signals corresponding to the respective echo paths.

In the stereo echo cancellation circuit 47, 481 is a left received signal storage circuit, 48r is a right received signal storage circuit, 491 is a reverberation characteristic parameter storage circuit for the echo signal Lff, 49r is a reverberation characteristic parameter storage circuit for the echo signal Rr, and 50g is an echo. 50r is a reverberation characteristic parameter storage circuit for the signal Lr; 50r is a reverberation characteristic parameter storage circuit for the reverberation signal Rff; 51 is a correction amount calculation circuit for estimating the reverberation signal L1; 52
A is a correction amount calculation circuit for estimating the echo signal Lr;
52r is a correction amount calculation circuit for estimating the echo signal R1, 51r is a correction amount calculation circuit for estimating the echo signal Rr, 532 is a superposition integration circuit for obtaining an approximate echo signal of the echo signal L1, and 53r is Rr. 51 is a superimposition integration circuit that obtains an approximate echo signal of Lr, 54r is a superimposition integration circuit that obtains an approximate echo signal of Lr, and 54r is R7! A convolution integrator circuit that obtains an approximate echo signal of
55 is a subtraction circuit that subtracts the echo signal 11 and the approximate echo signal Lr from the transmission signal Loi, and 55r is the transmission signal R.
A subtraction circuit subtracts approximate echo signals of the echo signals RN and Rr from oi, and 56.57 is a modification coefficient control circuit that generates a modification coefficient.

Here, since the circuit operation is the same for the right channel and the left channel, we will focus on the right channel below and
Explain the operation. First, the right received signal Rin is
(t), the left received signal Lin is written as X l (t), and the right transmitted signal R0 is written as Er (tl).Circuit 49
Let the reverberation characteristic estimation parameter of r be Hrri (tl (where i=o, 1°·----, M rr-1), and the reverberation characteristic estimation parameter of the circuit 50r Hrlj(t)
(However, j = 0, 1, -----------.

Mrl-1). Approximate signal of echo signal Rr? rr
(t) and the approximate signal of the echo signal R1? r II (
Let it be tl. At this time, the operation formula for the right channel of the stereo echo cancellation circuit during adaptation is according to equations (20) to (23).
r II, (t) − (27)ΔHrr 1(t)=
α2 ・Er(tiXr(ti)/ΔHrl 1ft
)=α4 ・Er(tiX2(L-i)/Hr r 1
(tl=Hr r i (t-1) +ΔHr r
i (t)−・−−−−−−−−(30”) Hr l 1(t)=Hr l i (L−1) +
Δ) i r l i (tl・−・−−−−−−(3
1) It is expressed as. Note that the convergence coefficients α2 and α4 are CX
2: α4 = N r r : N r 1 −
(29), and the sum of both is set to be 1.0 or less in order to guarantee stability of convergence. By using this direction, echo cancellation is sufficient for improving call quality■
It is clear from the simulation results shown in FIG. 8 that the following can be obtained.

Next, the second invention of the present invention will be explained using the circuit shown in FIG. This is the application circuit for Training 1.

In general, it takes time to obtain a sufficient amount of echo cancellation in an echo cancellation circuit, so a training signal (for example, white noise whose characteristic parameters in the used band can be easily determined) is used to determine the echo characteristic parameters in advance. The required method is used. This is achieved by sending out a training signal as a received signal.

In FIG. 6, the echo canceling circuit 58 is the same as the multi-nayanel echo canceling circuit shown in FIG. 4 or 3;
Reference numeral 60 indicates a training signal generation circuit, reference numeral 60 indicates a changeover switch for each channel, reference numeral 61 indicates a selection switch for selecting a channel, and reference numeral 62 indicates a switch for preventing transmission of the training signal to the transmission transmission path 22.

To operate this, a training signal is sent out from the training signal generation circuit 59 before starting communication. First, switch 61 is changed so that the training signal is sent only to the first channel. Next, switch 60. The training signal from the training signal generation circuit 59 is connected to the first channel, and the speaker 2
5. is sounded, the echo cancellation circuit 58 is operated, and characteristic parameters are determined. Thereafter, by sequentially switching the switch 61 in the same manner, the reverberation characteristic parameters of all channels can be obtained.

In this drawing, the echo cancellation circuit is constituted by a digital signal processing circuit, but AD conversion and DA conversion are omitted. In this embodiment, the means for modifying and storing the echo cancellation circuit parameters and the means for obtaining the approximate echo signal described in Claims (11 and (2)) are shown in an acyclic filter configuration. However, the present invention can be implemented with a recursive filter as well.

〔Effect of the invention〕

As explained above, the device of the present invention has a dedicated pressure characteristic parameter storage circuit for each echo path, and a correction coefficient and correction amount calculation circuit controlled for each channel using a common residual signal. The circuit configuration uses the following to modify the reverberation characteristic parameters. This has the effect of making it possible to obtain the required echo cancellation characteristics in multi-channel echo cancellation.

Further, in order to prevent deterioration of the convergence speed due to the power difference between the received signals of each channel, when a means for normalizing each signal by using its power is provided, the convergence speed can be improved by this.

Additionally, a training signal is generated at the start of communication.
However, it has a function of directing the flow to the echo path for each channel, and can quickly obtain the amount of echo cancellation.

With these features, the device of the present invention can quickly obtain the required amount of echo cancellation required from the viewpoint of speech quality in multi-channel echo cancellation.

[Brief explanation of the drawing]

FIG. 1 is a block diagram of an apparatus according to a first embodiment of the present invention. FIG. 2 is a diagram showing simulation results of this example circuit. FIG. 3 is a block diagram of an apparatus according to another embodiment of the present invention. FIG. 4 is a diagram showing a simulation of this example device. FIG. 5 is a block diagram of an embodiment of the present invention in stereo echo cancellation. FIG. 6 is a configuration diagram when applying a training signal according to an embodiment of the present invention. FIG. 7 is a configuration diagram of a conventional example of echo cancellation for monaural communication consisting of a speaker/microphone system between ground A and B. Figure 8 is a detailed diagram of the monaural echo cancellation circuit. FIG. 9 is a diagram showing an extension of the conventional monaural communication echo cancellation method for multi-channel use. ■...Microphone, 2...Amplification circuit, 3...Transmission line,
4...Speaker, 5...Amplification circuit, 6...Reverberation circuit, 7...Microphone, 8...Amplification circuit, 9...Transmission line, 10...Amplification circuit, 11...・Speaker, 12・
...Echo path, 13.14...Echo cancellation circuit, 15.
... Received signal storage circuit, 16... Subtraction circuit, 17...
- Correction amount calculation circuit, 18... Reverberation characteristic parameter storage circuit, 19... Superposition integration circuit, 20... Microphone, 2
1... Amplification circuit, 22... Transmission transmission line, 23...
Reception transmission path, 24... Amplification circuit, 25... Speaker, 26... Echo path, 27... Monaural echo cancellation circuit, 28... Multi-channel echo cancellation circuit, 29...
- Received signal storage circuit, 30... correction coefficient control circuit, 3
DESCRIPTION OF SYMBOLS 1... Echo characteristic parameter memory circuit, 32... Correction amount calculation circuit, '33... Superposition integration circuit, 34... Power calculation circuit, 35... Addition circuit, 36... Subtraction circuit, 37... Multi-channel echo cancellation circuit, 38-... Correction amount calculation circuit, 39... Power calculation circuit, 40... Microphone, 41... Mixing device, 42... Transmission transmission line, 43... ...Reception transmission line, 44 amplifier circuit, 45...
Speaker, 46... Reverberation path, 47... Stereo H
Erasing circuit, 48... Received signal storage circuit, 49.50.
... Reverberation characteristic parameter storage circuit, 51.52 ... Correction amount calculation circuit, 53.54 ... Superposition integration circuit, 55.
... Subtraction circuit, 56.57 ... Correction coefficient control circuit, 5
8... Multi-channel echo cancellation circuit, 59... Training signal generation circuit, 60... Changeover switch, 61
... Selection switch, 62... Blocking switch.

Claims (6)

[Claims] (1) In a communication system where two or more received signals generate echo signals via two or more echo paths, and the sum of these echo signals is input to the transmission signal, the generated approximate echo signal is subtracted from the transmission signal. In the echo cancellation method, there is provided means for respectively storing two or more received signals received from one or more remote locations via a transmission line, inputting each correction amount for each of the received signals, and inputting each correction amount for each of the received signals, and Means for modifying and storing each echo characteristic parameter of each echo signal that goes around the transmitted signal, and generating each approximate echo signal from each of the echo characteristic parameters and each received signal for each of the received signals. means for outputting a residual signal obtained by subtracting the sum of each of the approximate echo signals from the transmitted signal;
A multi-path echo cancellation system comprising: correction amount calculation means for calculating the correction amounts for each of the received signals using all of the received signals and the correction coefficients. (2) The correction amount calculation means includes correction amount calculation means for calculating each correction amount for each received signal using the residual signal, each received signal, and each correction coefficient. Multi-passage echo cancellation method described. (3) The multi-path echo according to claim (2), wherein the correction amount calculation means includes means for determining the value of each correction coefficient so that the sum of all correction coefficients is 1 or less. Signal cancellation method. (4) The multi-path echo according to claim (3), wherein the correction amount calculation means includes means for setting each correction coefficient in proportion to the impulse response length of the echo characteristic parameter for each received signal. Signal cancellation method. (5) In a communication system where two or more received signals generate echo signals via two or more echo paths, and the sum of these echo signals is input to the transmission signal, the generated approximate echo signal is subtracted from the transmission signal. In the echo cancellation method, there is provided means for respectively storing two or more received signals received from one or more remote locations via a transmission line, inputting each correction amount for each of the received signals, and inputting each correction amount for each of the received signals, and Means for modifying and storing each echo characteristic parameter of each echo signal that goes around the transmitted signal, and generating each approximate echo signal from each of the echo characteristic parameters and each received signal for each of the received signals. means for outputting a residual signal obtained by subtracting the sum of each of the approximate echo signals from the transmitted signal;
correction amount calculation means for calculating the respective correction amounts for each of the received signals using all of the received signals and each correction coefficient; means for generating a training signal in advance before starting communication; switching means for sequentially transmitting each of the received signals to the echo path; means for blocking transmission of the transmission signal to the transmission path while transmitting the training signal to the echo path as the received signal; The apparatus is characterized by comprising a control means for connecting the transmission path and the received signal by the switching means after calculating each of the echo signal characteristic parameters, and for opening the blocking means and sending out the transmission signal to the transmission path. Multi-passage echo cancellation method. (6) The multi-path echo cancellation system according to claim (5), wherein the means for generating training signals includes means for generating the same types of training signals as there are echo paths.