JPH10303786A - Echo canceller - Google Patents

Abstract

PROBLEM TO BE SOLVED: To reduce the estimation error of an impulse response, when a double talk occurs and to prevent the deterioration of echo cancel quantity by using a delay impulse response dalayed, compared to the time sequence of a reception input signal. SOLUTION: A register 104 calculates the correction quantity of an estimation impulse response by setting a shift register 100 storing the reception input signal Xn at every input time sequence, a shift register 101 storing a transmission output signal at every input time sequence, a resister 102 calculating the power of the register 100 and a time constant μ 103 to be inputs. An adder 107 adds the output (b) of the register 104 and the output of the register 105 storing the impulse response, and it therefore corrects the delayed time sequence of the impulses response. A computing element 108 convolutes the impulse response obtained by correcting delays and the register 100 and generates a pseudo-echo signal. An adder 30 subtracts the pseudo-echo signal from a transmission input signal dn, canceling an echo and setting it to be the transmission output signal.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an echo canceller, and more particularly to an echo canceller for canceling an echo generated when a received input signal wraps around a transmission output signal.

[0002]

2. Description of the Related Art A conventional echo canceller of this type is disclosed in, for example, Japanese Patent Application Laid-Open No. 62-24726 and Japanese Patent Application Laid-Open No. 7-28.
As disclosed in Japanese Patent No. 3859, in a configuration for canceling an echo generated when a reception input signal wraps around a transmission output signal, a so-called double talk state when a reception input signal and a transmission output signal are present at the same time can be realized at high speed. The detection of the impulse response necessary for estimating the pseudo echo component is stopped at the time of the double talk to prevent the divergence of the echo canceller.

FIG. 2 is a block diagram showing an example of a conventional echo canceller represented by Japanese Patent Application Laid-Open No. 62-24726. This echo canceller includes a shift register 9 for inputting a received input signal, a product-sum operation unit 10, a subtractor 11, power registers 14 and 17, and save registers 15 and 1.
8, 21, power calculator 16, power comparator 19, estimated impulse register 20, square sum calculator 22, correction amount calculator 24, step gain register 24, adder 2
5 is comprised.

In FIG. 2, a reception input signal input to a reception input terminal is input to an input shift register 9,
The sample signal sequence is stored. Input shift register 9
Is input to the subtractor 11 via the product-sum calculator 10, and the subtractor 11 outputs the echo signal input from the transmission input terminal and the superimposed transmission signal of the transmission input from the product-sum calculator 10. Power register 1 for subtracting the output signal from the output signal and storing the output via the output signal power calculator 13.
4 and its output is stored. The evacuation register 15 is connected to the power register 14.

Further, another output from the transmission input terminal is input to a power register 17 for storing the output via a power calculator 16 for the output signal, and the output is stored. The evacuation register 18 is connected to the power register 17.

Next, the outputs of the power registers 14 and 17 are input to a power comparator 19, and the output signal S of the power signal comparator 19 is output to the power registers 14 and 17 and the respective save registers 15 and Estimated impulse register 2 for storing between 18 and the estimated impulse response
It controls data transfer between 0 and its save register 21.

Another output of the shift register 9 is input to a sum-of-squares calculator 22, where the sum of squares of the data of the shift register 9 is calculated and input to a correction calculator 23 for correcting and calculating the estimated impulse response. Is done. The correction calculator 23 outputs another output of the input shift register 9, an output of the square sum calculator 22, another output of the subtractor 11, and a correction coefficient α of the estimated impulse response from the step gain register 24. Then, the correction amount of the estimated impulse response is calculated, input to the estimated impulse register 20 via the adder 25 for correcting the estimated impulse, and the outputs of the estimated impulse register 20 and the evacuation register 21 are multiplied. The sum operation unit 10 performs an operation together with the output of the input shift register 9.

In FIG. 2, the determination of double talk is, for example, f (j) = 10 log (Epj / Spj)> − 15 (A). However, when starting, the estimated impulse register 2
0, save register 21, power registers 14, 17
And its save registers 15 and 18 have been cleared,
f (0) = 0.

When the double talk detection function f (j) is calculated by the power comparator 19 at the time of starting, the above-mentioned conditions are satisfied, and the data of the estimated impulse register 20 and the data of the power registers 14 and 17 are stored in the save register 21 respectively. , 1
Go to 5, 18 and perform 128 estimations.

However, during the 128 estimation operations, when the power calculator 19 satisfies f (j) = 10 log (EPM / SPM) +3 (B), the estimation impulse register is immediately executed. 2
0, the data of the power registers 14 and 17 are returned to the respective save registers 21, 15 and 18, and the estimation calculation is performed again.

If the result of the estimation operation 128 times is f (j)> 10 log (EPM / SPM) (C), it is determined that the estimated impulse has not converged, and the estimated impulse register 20 Register 14,
17 data is moved to each save register.

If the condition (C) is not satisfied, it is determined that the estimated impulse has converged, and the estimated impulse register 20 and the power registers 14 and 17 are determined.
Are transferred to the save registers, and the conditional expression (A) is determined.

As a result of the determination of conditional expression (A), conditional expression (A)
Is satisfied, the estimation calculation is performed again 128 times. If the conditional expression (A) is not satisfied, the processing shifts to the single talk processing.

In the single talk process, an estimation operation is performed for each sample, and the calculated values of the power registers 14 and 17 are respectively transferred to the save registers.

In the case of transition to the double talk process,
When the conditional expression (A) is satisfied, the data in the estimated impulse register 20 and the power registers 14 and 17 are transferred to the save registers, respectively, and the estimation operation is started 128 times.

That is, as the double talk processing, the update of the estimated impulse is not stopped, but the data of the estimated impulse register 20 and the power registers 14 and 17 are transferred to the evacuation register, so that the impulse response at the time of starting is improved. Is converged, and the estimation calculation can be stopped at the time of double talk.

FIG. 3 is a block diagram showing an example of an echo canceller represented by JP-A-7-283859. The echo canceller includes a delay unit 3 that delays a reception signal RIN for a predetermined time to generate a delay reception signal RRIN, a delay reception signal RRIN and a transmission-side input signal SIN.
, A double talk detector 5 for detecting a double talk state based on the correlation result from the correlator 4, an impulse response estimator 6 for estimating an impulse response of the echo path 13, and an impulse response A pseudo-echo generation unit 7 that convolves the impulse response IM estimated by the estimation unit 6 with the reception signal RIN to generate a pseudo-echo PE, and attenuates the echo by subtracting the pseudo-echo PE from the transmission-side input signal SIN; Transmission signal (residual echo) SOU
It is composed of an echo attenuator 8 for generating T.

In FIG. 3, when the double talk state is detected by the double talk detecting means for detecting the double talk state based on the correlation between the delayed received signal and the transmission side input signal, the estimation of the impulse response by the impulse response detecting means is stopped. Therefore, even when there is an echo propagation time delay due to the echo path, the situation in which a non-double talk state is erroneously detected as a double talk state can be significantly reduced, the double talk state can be accurately determined, and the echo in the non-double talk state can be reduced. Can be surely erased.

[0019]

The first problem with the conventional echo canceller shown in FIG. 2 is that the estimated impulse response stored in the save register does not sufficiently converge during double talk. Therefore, the amount of echo cancellation deteriorates. The reason is that the impulse response when it is determined that the transition from the single talk to the double talk has been made is stored in the save register after the double talk is determined, and thus is affected by the estimation error of the impulse response due to the double talk. Because.

The second problem is that a circuit configuration and a control method for holding an estimated impulse response of single talk are complicated. The reason is that, since the impulse response at the time of single talk is stored in the save register, it is necessary to exchange data between the impulse response register and the save register at any time. The conventional echo canceller shown in FIG. 3 also has a problem that the influence of the impulse response estimation error due to double talk cannot be completely eliminated.

An object of the present invention is to provide an echo canceller capable of minimizing an estimation error of an impulse response at the time of occurrence of double talk and preventing deterioration of an echo cancellation amount.

Another object of the present invention is to provide an echo canceller capable of correcting an estimation error of an impulse response at the time of double talk into a sufficiently converged impulse response at the time of single talk and preventing an error in the estimation operation. That is.

[0023]

According to the present invention, there is provided an echo canceller for canceling an echo generated when a reception input signal wraps around a transmission output signal, wherein the delay impulse is delayed from the time series of the reception input signal. An echo canceller characterized by including an echo estimating means for estimating a pseudo echo component using a response and an echo canceling means for eliminating an echo component in the transmission output signal by the pseudo echo component is obtained.

[0024] The present invention further includes a double talk detecting means for detecting a double talk state based on the reception input signal and the transmission output signal, wherein the detecting operation of the echo estimating means is stopped when the double talk state is detected. Features.

Then, the echo estimating means uses the time series of the erasure output of the echo canceling means, the time series of the received input signal, and a predetermined constant to obtain m sambles (m: m) of the time series of the received input signal. It is characterized by comprising generating means for generating a delayed impulse response delayed by an integer (two or more integers), and means for generating the pseudo echo component according to the delayed impulse response and the received input signal.

The generating means includes a first storage means for storing the time series of the received input signal for m samples, a second storage means for storing the time series of the erasure output for m samples, A third storage unit for calculating the time-series power of the received input signal and storing the same for m samples; a delay impulse response delayed by m samples by the outputs of the first to third storage units and the constant; Means for generating a corrected impulse response of (m-1) samples to one sample delay for correcting the delayed impulse response.

The operation of the present invention will be described. Using the impulse response delayed from the time series of the received input signal, the time series impulse response of the received input signal is corrected to estimate a pseudo echo, and further, the operation of estimating the impulse response is stopped when double talk is detected. As a result, the estimation error of the impulse response of the echo canceller is corrected, and echo cancellation can be performed using an impulse response that is not affected by the estimation error due to double talk.

[0028]

Embodiments of the present invention will be described below with reference to the drawings.

FIG. 1 is a block diagram of an embodiment of the present invention. Referring to FIG. 1, an echo canceller device 1 includes a shift register 100 (hereinafter, referred to as an X register) for storing a received input signal Xn for each input time series, and a shift register for storing a transmission output signal for each input time series. 101 (hereinafter referred to as an E register) and a register 102 (hereinafter referred to as an X ² register) for calculating the power of the X register 100.

Further, an X register 100 and an E register 1
01, the X ² register and the constant μ103 as inputs, an impulse correction register 104 (hereinafter referred to as ΔH register) for calculating a correction amount of the estimated impulse response, and a register 105 storing the impulse response (hereinafter referred to as H register). And an output a of the ΔH register 104 and an output of the H register 105 to update the H register 104, and the other output b of the ΔH register 104
And an output of the H register 105 to add the adder 107 to correct a time series in which the impulse response is delayed.

Further, the impulse response with the delay corrected and X
The arithmetic unit 108 generates a pseudo echo signal by performing a convolution operation of a register, and the adder 30 which subtracts the pseudo echo signal from the transmission input signal to perform echo cancellation to obtain a transmission output signal.

Next, the operation of the circuit will be described with reference to the drawings. Normally, the adaptive correction algorithm (NLMS (Normalized Least Mean Square) algorithm) of the echo canceller is represented as Yn = Hn.Xn, where the input signal Xn, the input signal dn, the pseudo echo signal Yn, and the impulse response Hn of the echo path are provided. (1) en = dn -Yn (2) ΔHn = μ · en / ΣXn ² (3) Hn + 1 = Hn + ΔHn · Xn (4) Here, n represents time and n ± 1, n ± 2.
Represents a time series. en is a residual signal (transmission output signal) from which the pseudo echo signal has been eliminated, and μ is a constant.

Next, considering the above equation for the time n-1 (when the sample time is delayed), Yn-1 = Hn-1.Yn-1 (5) en-1 = dn-1 -Yn -1 (6) ΔHn = μ · en / ΣXn ² (7) Hn = Hn-1 + ΔHn-1 · Xn-1 (8)

Here, equation (4) can be expressed as follows from equation (8): Hn = Hn-1 + .DELTA.Hn-1.Xn-1 + .DELTA.Hn.Xn (9) Further, when Expression (9) is extended to a case where the time is delayed by m samples, Hn = Hn−m + ΔHn−m · Xn−m + ΔHn−m + 1 · Xn−m + 1 + ·· + ΔHn−1 · Xn -1... (10) That is, the impulse response at time n can be represented using the impulse response Hn-m delayed by m sample times from time n.

That is, the NLMS algorithm is as follows: Yn = Hn · Yn (11) en = dn−Yn (12) ΔHn = μ · en / ΣXn ² (13) Hn−m = Hn -m + .DELTA.Hn-m.Xn-m (14) Hn = Hn-m + .DELTA.Hn-m.Xn-m + .DELTA.Hn-m + 1.Xn-m + 1 + .. +. DELTA.Hn-1.Xn-1 .. ... (15)

Here, the convolution operation unit 10 in FIG.
8 is the equation (11), the X register 100, the H register 105
Correspond to Hn and Xn, respectively. Further, the adder 30 calculates the residual signal e shown in the equation (12) by the equation (12).
The register for storing n for m sample times is the E register 10
1 output.

In the ΔH register 104, the operation of updating the H register in the equations (13), (14) and (15) is performed, and the output a is calculated as ΔHn-m · Xn-m in the equation (14). And the other output b is ΔHn-m in equation (15).
・ Xn-m + ΔHn-m + 1 ・ Xn-m + 1 + ・ ・ + ΔHn-1 ・ X
It corresponds to the term n-1.

The output of the adder 106 is given by the following equation (14).
The H register delayed by m sample times corresponding to the above is updated, and the output of the adder 107 is obtained by the equation (1) that generates Hn corrected by the H register (Hn-m) delayed by m sample times.
This corresponds to 5).

Next, a second embodiment of the present invention will be described in detail with reference to FIG. Referring to FIG. 1, there is provided a double talk detection circuit 2 which receives a transmission input signal and a reception input signal as inputs and detects simultaneous communication of both transmission and reception signals. The double talk detection circuit 2 detects the power level of each of the transmission input signal and the reception input signal, and determines the double talk state when the power level ratio between the reception input signal and the transmission input signal exceeds a certain threshold.

At this time, by stopping the estimation of the impulse response estimating section of the echo canceller 10 and the ΔH register 104, an impulse response with a small estimation error which is not affected by the double talk delayed by m sample times can be obtained.
A pseudo echo can be generated.

When disturbance noise (such as double talk) other than an echo signal is superimposed on the transmission input side, an error has conventionally occurred in the estimation operation of the learning identification method, and an impulse response different from the actual echo path (impulse response is disturbed). As described above, by using the impulse response (Hn-m) delayed from the time series of the received signal, as described above, even if the impulse response (Hn) is disturbed by double talk, the impulse response immediately before the impulse response (Hn-m) is disturbed. Hn-m).

The pseudo echo (Yn) is generated by the convolution operation of the impulse response (Hn-1) and the received input signal (Xn) according to the equation (1), and the impulse response (Hn-1)
From the equation (15), the delayed impulse response (Hn-m
) And its correction term (the second term and below on the right side of equation (15)). That is, the disturbance of the impulse response (Hn) due to double talk is more caused by the disturbance of the correction term than the impulse response (Hn-m).

Therefore, when the double talk state is detected, the correction term of the equation (15) is cleared to thereby make the impulse response (Hn = Hn−) unaffected by the double talk.
m) can be returned.

In FIG. 1, since the impulse response which is not affected by the double talk can be generated without saving the H register in a memory or the like, the configuration circuit and the control circuit can be simplified. It also has the effect.

[0045]

The first effect is that echo cancellation is enabled by using an impulse response delayed from the time series of the received input signal. The reason is that the transmission output signal is stored in the shift register and has a function of correcting the same time-series impulse response as the reception input signal.

The second effect is that the double talk detector
It is possible to perform echo estimation using an impulse response with a small estimation error at the time of double talk. When double talk is detected, by stopping the estimation operation,
This is because a pseudo echo is generated by an impulse response with a small estimation error delayed from the time series of the received input signal.

[Brief description of the drawings]

FIG. 1 is a block diagram of an embodiment of the present invention.

FIG. 2 is a block diagram illustrating an example of a conventional echo canceller.

FIG. 3 is a block diagram showing another example of a conventional echo canceller.

[Explanation of symbols]

Reference Signs List 1 echo estimation unit 2 double talk detector 30, 106, 107 adder 100 X register 101 E register 102 X ² register 103 constant 104 ΔH register 105 H register 108 convolution operation unit

Claims (5)

[Claims] 1. An echo canceller for canceling an echo generated when a received input signal wraps around a transmission output signal, wherein a pseudo echo component is estimated using a delayed impulse response delayed from a time series of the received input signal. An echo canceller comprising: echo estimating means; and echo canceling means for canceling an echo component in the transmission output signal using the pseudo echo component. 2. The apparatus according to claim 1, further comprising a double talk detecting means for detecting a double talk state based on said reception input signal and said transmission output signal, wherein the detecting operation of said echo estimating means is stopped when said double talk state is detected. The echo canceller according to claim 1, wherein: 3. The echo estimating means uses the time series of the erasure output of the echo canceling means, the time series of the received input signal, and a predetermined constant to obtain m samples (m is a time series) of the received input signal. 2. The apparatus according to claim 1, further comprising: generating means for generating a delayed impulse response delayed by an integer (two or more integers); and means for generating the pseudo echo component according to the delayed impulse response and the received input signal. Or the echo canceller according to 2. 4. The first storage means for storing the time series of the reception input signal for m samples, the second storage means for storing the time series of the erasure output for m samples, A third storage unit for calculating the time-series power of the received input signal and storing the same for m samples; a delay impulse response delayed by m samples by the outputs of the first to third storage units and the constant; 4. An echo canceller according to claim 3, further comprising means for generating a corrected impulse response of (m-1) samples to one sample delay for correcting the delayed impulse response. 5. The echo canceller according to claim 4, wherein said corrected impulse response is cleared when said double talk state is detected.