JPH0748679B2 - Eco-Cancer - Google Patents

Description

The present invention relates to an echo canceller for canceling an echo signal (hereinafter referred to as an echo) in a long-distance telephone communication line or the like, and more particularly to an echo canceller when a disturbance signal is superposed on the echo. The present invention relates to an echo canceller including a superposed call (hereinafter referred to as double talk) detection circuit that detects a signal.

[Conventional technology]

Conventionally, echo control devices, voice switches, echo cancellers, and the like have been used to control echoes in long-distance lines, loudspeakers, two-way repeaters, and the like. Of these three echo control means, the former two have a problem in that troubles such as disconnection of the talk for turning the line on and off and click noise occur. The echo canceller is developed to compensate for such a defect. The echo canceller synthesizes an approximate copy (hereinafter referred to as a replica) of the echo based on the characteristics of the echo path obtained by measuring the echo path, and cancels the actual echo with this replica. It does not have any configuration that causes interference such as click noise.

However, the echo canceller reduces the measurement accuracy of the echo path when a disturbance signal is superimposed on the echo and is input to the transmission input terminal, and the amount of cancellation also depends on the level ratio between the echo at the transmission input terminal and other signals. Basically, it has a problem that the value drops to the value determined by. Therefore, in order to prevent the reduction of the cancellation amount in the superposed state, a double talk detection circuit has been conventionally used.

A conventional echo canceller equipped with a double talk detection circuit will be described with reference to FIG.

In the figure, the reception signal from the 4-wire side is the reception input terminal 10.
Added to. This signal is sent from the reception output terminal 20 to the echo path 300 that is simulated. Then, the echo component echoes to the transmission input terminal 30 and tries to return from the transmission output terminal 40 to the 4-wire side. Here, the adaptive digital filter (ADF) 100 creates a predicted value of the echo component from the received input signal component, and the subtractor 110 tries to eliminate the echo component. In addition, adaptive control is performed by using the unerased signal component so that the unerased signal approaches zero. on the other hand,
When the pseudo noise external noise (speech of the near-end speaker) 400 is added to the transmission input terminal 30, the adaptive control diverges in the wrong direction, so the double-talk detector 200 detects this, Operates to stop adaptive control. The double-talk detector shown in FIG. 2 is generally used, and compares the reception power of the echo canceller with the transmission power and, for example, judges that the transmission power exceeds the reception power as double-talk.

[Problems to be solved by the invention]

According to the above conventional technique, when double talk occurs,
If the extraneous noise (ie, near-end speaker's voice) level is low compared to the far-end speaker's voice, then the double-talk detector obviously does not work, and thus false adaptive behavior continues, and thus echo cancellation. There was a problem of reduced ability.

In order to solve such a problem, the residual echo level is estimated, the estimated residual echo level is compared with the actual measured residual echo level, and if the latter exceeds the former, it is determined that double talk is present. A method of doing so is proposed in the following document.

"A Method of Superposed Call Detection in Adaptive Echo Canceller and Experimental Results" 1984 IEICE General Conference. N
o2343.

An object of the present invention is to solve the above-mentioned problems in double-talk detection by an approach different from that of the above document, and to obtain a high-performance double-talk detector, and further, a high-performance echo canceller.

[Means for solving problems]

In order to achieve the above object, the present invention identifies an echo path characteristic by an input / output signal of an echo path, and includes an adaptive digital filter that synthesizes an approximate echo replica signal, so that the echo replica is subtracted from an actual echo component. In the constructed echo canceller, the transmission input signal y (n) (sample value expression, n is a time number)
And the approximated echo replica signal (n)
y (n), (Where E {•} represents the expected value calculation), and when ρy (n) increased with time, it was judged that there was no double talk, and ρy (n) decreased with time. In this case, it is determined that it is a double talk, and the adaptive operation of the adaptive digital filter is immediately stopped.

[Action]

As is well known, the adaptive digital filter calculates the replica (n) of the actual input signal y (n), and the error signal e (n) = y (n)-(n) (2) On the other hand, the adaptive operation is performed so that E {(n) ² } approaches zero infinitely. In fact, if y (n) does not include extraneous noise, then E {e (n)
The operation of ² } → 0 proceeds smoothly. This means that the waveform of the replica (n) is as close as possible to the waveform of the signal y (n), and therefore the cross-correlation function ψy (n) = E of y (n) and (n) is obtained. {Y (n) ・
(N)} should be increasing. However,
This measure is unsuitable for a signal whose level fluctuates over a wide dynamic range such as voice, and the cross-correlation coefficient ρy (n) such as the equation (1) normalized by its power is suitable. Cross-correlation coefficient ρy (n)
However, if double talk occurs, the signal y (n) contains a component that is uncorrelated with (n), so that ρy (n) decreases with good sensitivity. Like The echo canceller of the present invention is configured so as to catch this moment and stop the adaptive control of the adaptive digital filter. That is, the cross-correlation coefficient ρy (n) has a property of being lowered from the value 1 with high sensitivity when an external signal is generated, and its detection is instantaneously performed. Therefore, the erroneous correction work of the adaptive digital filter is quickly stopped, and as a result, the deterioration of the echo cancellation amount can be suppressed to a minimum.

〔Example〕

An embodiment of the present invention will be described below with reference to FIGS. 1, 2, and 4.

FIG. 1 is an overall block diagram of an echo canceller according to an embodiment of the present invention. In FIG. 1, 10 is a received signal input terminal, 20 is a received signal output terminal, 30 is a transmitted signal input terminal, and 40 is a transmitted signal output terminal. Further, 100 is an adaptive digital filter means, 110 is a subtractor, and 200 is a double talk detector which is a feature of the present invention.

Now, the speech signal x (n) from the far end is applied to the reception input terminal 10, and the signal x (n) is applied to the adaptive digital filter 100 for synthesizing the echo replica, and at the same time from the reception output terminal 20 to the near end speech. It is sent to the person. The echo component y (n) of this signal x (n) returns to the transmission input terminal 30 through the echo path (not shown in FIG. 1). On the other hand, the adaptive digital filter 100 creates an echo replica (n) from the signal x (n), and the subtracter 110 makes e (n) = y.
Echo cancellation of (n)-(n) is performed. The cancellation residual echo component is transmitted from the transmission output terminal 40 to the far-end speaker, and at the same time, it becomes a driving signal for adaptive control of the adaptive digital filter. The content of adaptive control is a well-known learning identification method, and is the algorithm shown below.

e (n) = y (n)-(n) (4) Where N is the number of taps of the adaptive digital filter, hn
(Μ) is the μ-th tap coefficient value at time n, α
Is a constant value called a correction coefficient.

The double-talk detector 200 takes in the two signals y (n) and (n), and always performs the calculation of the equation (1). When a decrease in the correlation coefficient is detected, it is determined that double talk is present, and the correction coefficient α is controlled. That is, when double talk is not present, for example, α = 1 is set, and when it is determined that double talk is present, α = 0 is set and the coefficient correction of the above equation (5) is stopped.

FIG. 2 is a detailed configuration diagram of the double-talk detector 200. In FIG. 2, 201,202,203 are multipliers, 211,212,213
Is a low-pass digital filter with the same characteristics, 204 and 205 are multipliers, 206 is a divider, and 207 is a multiplier. Reference numeral 221 denotes an averaging operation means, 222 a thinning-out switch means, and 223 means for selecting and holding a larger value.
The point part where 1 and 222 and 223 are put together is a means for estimating by averaging the correlation coefficients. Further, 230 is a comparing means, and 240 is a means for controlling the correction coefficient α according to the result of the comparing means 230.
Also 250 is a means to monitor signal power.

Multipliers 201, 202, 203 are respectively y ² (n), y (n)
(N), ² (n) are created and these multipliers 201, 202, 2
The output of 03 is the filter 211,212,21 provided in the subsequent stage, respectively.
By 3, the time average operation outputs the expected value. That is, the filters 211, 212, 213 perform the expected value operation E {.} Of the equation (1). Then, the multipliers 204 and 205 and the divider 206 obtain the squared operation result of the equation (1). In this embodiment, in order to avoid the square root calculation, the output of the divider 206 is set as the correlation coefficient ρy (n) again.

Correlation coefficient ρy at time n calculated in this way
(N) is input to the means 220 for averaging and estimating the correlation coefficient, and the estimated correlation coefficient y is calculated. That is, the averaging means 221 averages over a relatively long time window,
The switching means 222 decimates, the result in the previous time window is compared with the result in the current time window, and the larger value is selected and held by the means 223 to obtain y. The average value means 221 and the thinning-out switch 222 are specifically constituted by a circuit such as an accumulator. In this way, the correlation coefficient ρy is approaching the value 1 every moment.
The average estimated value y in the time window of (n) is accumulated in the holding means 223. On the other hand, the instantaneous value ρy (n) is amplified by β = 1 + ε (ε is a small positive number) times in the multiplier 207 and compared with the above-mentioned yy by the comparison means 230.

If W (n) 0 as W (n) = ρy (n) · β−y (6), it is determined that there is no double talk, and the means 204 for controlling the correction coefficient α is ON-controlled. In this case, α = 1. Note that the meaning of β times is to ensure that α = 1 is ensured for small fluctuations of noise or the like, or to ensure α = 1 for slow line fluctuations. Now, when double talk occurs, the instantaneous value ρy (n) starts to decrease.
The result of equation (6) becomes negative with good sensitivity, and as a result, α
It is controlled as = 0. Then, the correlation coefficient value disturbed by the double talk is much lower than 1, so that the average value is lowered, and it is discarded to the holding means 223.

By the way, when the received input signal is not present (when the far-end speaker is not speaking), the echo y is naturally obtained.
(N) and echo replica (n) are zero,
The operation of the equation (1) is mathematically indefinite. In such a case, it suffices to perform the calculation. The monitoring means 250 does this by comparing the two powers σy ² = E {y (n) ² } σ ² = E {(n) ² } with an appropriate threshold and either If it is determined that there is no signal power, operations after division are disabled. Therefore, the previous value of α remains unchanged.

FIG. 4 is a diagram showing measurement data in this example. L _R is the received signal power, L _N is the external noise power, the upper part of the measured value is the residual power before double talk, and the lower part is the residual power after double talk. The hybrid return loss is 6db as the echo loss condition. This 4th
According to the figure, it can be seen that almost no deterioration is recognized. Since this value cannot avoid deterioration of about 10 db in the conventional example as shown in FIG. 3, the effect of the present invention is great.

〔The invention's effect〕

According to the present invention, in an echo canceller including an adaptive digital filter, it is possible to detect the occurrence of double talk quickly and appropriately, and it is possible to immediately stop the coefficient correction work of the adaptive digital filter. Alternatively, there is an effect that the divergence of the coefficient is prevented and the echo cancellation amount can be prevented from decreasing.

[Brief description of drawings]

FIG. 1 is an overall block diagram of an echo canceller according to an embodiment of the present invention, FIG. 2 is a detailed block diagram of a double talk detection circuit shown in FIG. 1, and FIG. 3 is a block diagram of a conventional echo canceller. FIG. 4 is a comparison diagram of erase residual power before and after double talk in the embodiment shown in FIG. 10 ... Receive input terminal, 20 ... Receive output terminal, 30 ... Transmit input terminal, 40 ... Transmit output terminal, 100 ... Adaptive digital filter, 110 ... Subtractor, 200 ... Double talk detector, 201, 202,
203, 204, 207 ... Multiplier, 206 ... Divider, 211, 212, 213 ... Low pass digital filter, 220 ... Correlation coefficient estimating means, 2
30 ... Comparison means, 240 ... Correction coefficient control means, 250 ... Power monitoring means.

Claims (1)

[Claims] 1. An adaptive digital filter means for synthesizing a replica (n) of a signal y (n) (where n is a time point number) containing an echo component generated in a 2-line to 4-line conversion section, and an error e ( n) = y (n)-(n), and an echo canceller configured to adaptively control the adaptive digital filter means by the error e (n). n) and a means for calculating the cross-correlation coefficient ρy (n), and the cross-correlation coefficient ρy (n) calculated from moment to moment is averaged and sampled in the time window M, and the short-term average mutual correlation is calculated. Number y (mM)
In addition, by comparing the short-time average cross-correlation coefficient y (mM) with the maximum short-time average cross-correlation coefficient y as an estimated value obtained and held so far, the larger one is obtained. Means for selectively holding as a new estimated value y, deviation W (n) = ρy (n) · β−y between the cross-correlation coefficient ρy (n) and the estimated value y (where β is a positive constant) ), The adaptive control stopping means for controlling the adaptive control stopping of the adaptive digital filter means according to the positive or negative of the above, and when the cross-correlation coefficient ρy (n) cannot be calculated, the adaptive control stopping means is An echo canceller, which is configured to be disabled and to retain a previous control state.