JPH07123238B2 - Eco-Cancerra control method - Google Patents

Description

Description: TECHNICAL FIELD The present invention relates to a method of controlling an echo canceller for canceling a talker echo in a hybrid circuit that performs 2-wire to 4-wire conversion of a telephone. .

(Prior Art) FIG. 2 is a schematic block diagram of a system using a general digital echo canceller. Reference numeral 1 is an input end of a far end input signal, which is connected to a D / A converter 2 and an echo canceller 7. ing. 3 is a hybrid circuit that performs 2-wire to 4-wire conversion,
The echo path is composed of a power feeder and a two-wire circuit, and is connected to the balance circuit 4, the telephone 5, and the A / D converter 6. The echo canceller 7 is connected to the input end 1 of the far end input signal, the A / D converter 6 and the transmission output end 8. In FIG. 2, x _j is a digital signal of the far end input, and x (t) is its D / A conversion output. Further, y (t) is an echo signal of the D / A conversion output x (t), n (t) is a near-end input signal such as a transmission sound or a room sound, and s (t) is an echo signal y (t). Near end input signal n
It is the sum transmission signal of (t). s _j is the transmission signal s (t)
A / D conversion output, e _j is the residual echo.

To explain the above operation, the far-end input signal x _j is D / A
Although converted into an output x (t), a part of it is leaked to the transmitting side as an echo signal y (t) in the echo path. On the other hand, since the near end input signal n (t) is also superimposed on the transmission side,
Eventually, the output s _j obtained by A / D converting the transmission signal s (t) = y (t) + n (t) is obtained. On the other hand, the echo canceller uses the residual echo e _j and the far-end input signal sequence. The impulse response of the echo path is successively estimated by. And at the next sample time, the estimated impulse response And far end input signal sequence A pseudo echo is generated by the A / D transmission signal s (t).
The echo y (t) is canceled by subtracting from the converted value s _j . As a result, the residual echo l _{j + 1} is transmitted.

However, when the estimation is continued in the double talk state in which the near-end input n (t) is large, the estimated impulse response diverges to a value completely different from the actual impulse response, and conversely the transmission signal s (t). The estimation error is added to the value s _{j obtained} by A / D conversion of. Therefore, in the case of double talk, a function is provided to detect this and stop the estimation of the echo canceller.

Double talk is detected by f (j)> c (j) (1). Here, f (j) is the digital signal x _j of the far end input, the A / D conversion output s _{j of} the transmission signal s (t), and the residual echo e.
a double talk detection function using a number of power and instantaneous values of _j, also c (j) is a threshold function for the double talk detection function f (j), transmitting a residual echo e _j A / D conversion output s _{j of} signal or residual echo e _j
And the far-end input digital signal x _j .

In this way, the double talk can be detected at high speed by the equation (1), and the impulse response can be estimated.
It is possible to control the divergence of the echo canceller due to the increase of the near-end input signal n (t).

(Problems to be solved by the invention) However, in the above method, the echo path changes,
Since the equation (1) is satisfied even when the estimation error increases, there is a problem that the estimation is stopped even though the impulse response estimation calculation should be continued. Furthermore, when the threshold for double-talk detection calculated from the power ratio of the residual echo and the far-end input or the power ratio of the residual echo and the transmission input becomes smaller than the actual cancellation amount, the estimation is stopped by equation (1) and the convergence speed There was a problem that it greatly damages.

Therefore, the present invention provides a method capable of stopping the impulse response estimation calculation during double talk, estimating the impulse response when the echo path changes, and restarting the impulse response estimation due to the setting error of the threshold value function c (j). It is provided.

(Means for Solving the Problems) The present invention relates to an echo canceller control method for removing an echo associated with 2-line to 4-line conversion, including (1) a sampled value sequence of a far-end input and an estimated impulse inside the echo canceller. Pseudo-echo is generated by convolution calculation of the response, and a process of sequentially estimating the estimated impulse response so that the square value of the residual echo obtained by subtracting the pseudo-echo from the echo signal approaches zero, and (2) echo path Pseudo-echo is generated by the estimated impulse response and the sample value of the far-end input, and the echo is canceled by subtracting this pseudo-echo from the echo signal, and the first residual echo is calculated. (3) Far-end input By calculating the residual echo or the power ratio of the echo signal and the residual echo, the cancellation amount of the echo canceller is sequentially calculated and The process of setting the threshold value of double talk detection to be high with the increase of the amount of cancellation, and (4) the cancellation amount calculated in the process of (3) above is set in the process of (3) above. The process of detecting double talk when it becomes smaller than the threshold value, and (5) While detecting double talk, the first residual echo is calculated and output using the estimated impulse response at the time of detection, and it is continued. (6) The power of the second residual echo is a constant multiple of the power of the first residual echo (≤
1) When it is below, it is judged that the estimated impulse response when the estimation is continued is valid, and thereafter, the threshold value for double talk detection is re-established based on the cancellation amount of the estimated impulse response when the estimation is continued. By setting and setting the impulse response used in the calculation of the second residual echo as an initial value to estimate the impulse response and calculate the first residual echo, the impulse response is estimated during double talk. When the echo path is changed, the echo response is followed to make it possible to eliminate the echo, and the threshold for double talk detection set in the process of (3) above is the true cancellation amount. It is possible to avoid a decrease in the convergence speed that occurs when the value is much lower.

(Operation) When double talk is detected by the formula (1), the residual echo is calculated and output using the estimated impulse response immediately before detection, but at the same time, the residual echo is calculated when the impulse response estimation is continued. By comparing the cancellation amounts when the estimation is stopped and when it is not stopped, the one with the larger cancellation amount is validated. Therefore, when the threshold for echo path change and double talk detection is set to a small value, equation (1) And the estimation can be prevented from stopping.

(Embodiment) The present invention calculates and outputs the first residual echo using the estimated impulse response immediately before the detection when double talk is detected, but by utilizing the margin of the number of program steps generated by stopping the estimation. The second residual echo when the estimation of the impulse response is continued is also calculated, and the cancellation amount when the estimation is stopped and when it is continued can be compared. The comparison of the cancellation amounts can be performed as shown in Expression (2) by comparing the powers of the first residual echo and the second residual echo, which are output signals.

E _pj > KE _pj ′ (2) Power of E _pj , E _pj ′: e _j , e _j ′ e _j : First residual echo e _j ′: Second residual echo K: 1 or more constant Echo path change And when the estimation is stopped because the threshold for double-talk detection is made too small, and when the impulse response estimation is continued after several samples,
(2) is satisfied by reducing the residual echo of At that time, the threshold for double talk detection is reset according to the amount of cancellation when the estimation of the impulse response is continued, and the echo is estimated using the estimated impulse response used in the calculation of the second residual echo. Erase and restart the impulse response estimation calculation.

FIG. 1 shows a functional block diagram of one embodiment of the present invention.
Is an input end of the far end input signal, and is connected to the registers 12 and 12 'which store the digital signal sequence of the far end input.
These registers 12 and 12 'are for storing the digital signal sequence of the far end input, and the product-sum operation circuit 13 and
The registers 20 and 21 are connected to the product-sum operation circuit 13 ', the register 20', and the register 21 '. The product-sum operation circuits 13 and 13 'are the product-sum operation circuits of the register 12, the register 18, the register 12', and the register 18 ', respectively, and are connected to the subtracters 15 and 15', respectively. Reference numeral 14 denotes an input end of the signal s _j in which the near end input is superimposed on the echo, and the subtracters 15, 1
It is connected to 5'and computing units 16, 16 '. Subtractor 15,15 ′
Are subtracters for subtracting the outputs of the product-sum operation circuits 13 and 13 'from the signal s _j, respectively, and input to the operation unit 16, the register 21, the first residual echo e _j and the operation units 16' and 21 '. It is connected. The computing units 16 and 16 ′ are computing units that calculate the cancellation amount from the transmission input s _j and the power ratio of the first residual echo and the second residual echo, respectively, and are connected to the control unit 17. The control unit 17 detects double talk, sets a threshold for double talk detection, transfers an estimated impulse response between the register 18 and the register 18 ', stops the estimation calculation of the echo canceller, and detects the impulse response at the time of double talk detection. It is supposed to start estimation calculation, etc.
It is connected to the registers 18, 18 'and the register 19. The register 19 stores the double-talk detection threshold value set by the control unit 17. Registers 18 and 18 'store the estimated impulse response and the estimated impulse response when the estimation is continued when the double talk is detected. The sum-of-products arithmetic circuit 13, the adder 23, the sum-of-products circuit 13', and the adder 23, respectively. 'It is connected to the. The adders 23 and 23 'are respectively register 18 and register 21 and register 18' and register 2
It is for finding the sum of 1 '. 20 and 20 'are registers for storing the sum of squares of the values in the register 12 and the register 12', respectively. Registers 21 and 21 'are registers for storing the correction amounts of register 18 and register 18', respectively, and register 20 and residual echo, respectively.
It is connected to the output 22 of e _j and the register 20 '.
(A) is a control signal for stopping the estimation calculation to the echo canceller, and (B) is a control signal for starting the estimation calculation of the impulse response when the double talk is detected.

Next, this operation will be described. In this embodiment, the echo canceller executes the equations (3) and (4) when the near-end input n (t) has a low level single talk.

(However, k = 0, 1, ..., N−1) The powers of e _j and s _j are E _pj and S _pj , respectively, and the calculator 16 sequentially calculates the cancellation amount E _pj / S _pj . During single talk as described above, the cancellation amount E _pj / S _pj monotonically decreases. The initial value of the register 19 is "2". Deterioration of the amount of cancellation due to double talk or change of echo path is detected by the equation (5).

E _pj / S _pj > 2 ・ E _pM / S _pM , E _pM / S _pM = min (E _pi / S _pi )
(5) (However, i = 0, 1, ..., J) Since the right side of the equation (5) decreases monotonically as the echo canceller converges, the deterioration of the cancellation amount can be detected more quickly as the echo advances. When the deterioration of the cancellation amount is detected, the control unit 17 outputs the estimated impulse response H _j of the register 18 to the register.
After copying the H _M to 18 ', the equation using H _j (3)
While executing only the cancellation operation of and outputting from the output end 22,
On the other hand, the canceling calculation and the correction calculation of the above formulas (3) and (4) are executed for H _M by using the free time due to not executing the correction calculation of H _j . Then, the computing unit 16 ′ sets the cancellation amount E _pj / S _pj when the cancellation amount deterioration is detected to an initial value and the above H _M
Calculate the cancellation amount E _pj ′ / S _pj for.

If such deterioration of the cancellation amount is due to double talk due to an increase in the level of the near-end input n (t), the cancellation amount will not be reduced even if the cancellation calculation and the correction calculation of the equations (3) and (4) are executed. Since it does not improve, the above processing is continued and the near-end input n (t)
When the level of H _j decreases and Eq. (5) does not hold, the single talk process is started and the estimation of H _j is restarted.

On the other hand, if the deterioration of the cancellation amount is due to a sharp change in the echo path, the error increases due to the deviation of the estimated impulse.
Canceling the amount of the H _M for formula (3) and by running the bucking operation and correction calculation equation (4) H _M improves, so equation (6) holds.

E _pj > 2 · E _pj ′ (6) Equation (6) shows that the cancellation amount of H _M that stopped the correction calculation was stopped.
It shows that the value is 6 dB or more better than the cancellation amount of H _j . At this time, the control unit 17 copies H _{M of the} register 18 ′ to the register 18 and makes it H _j, and stores the cancellation amount E _pj ′ / S _pj at this time in the register 19 and makes it E _pM / S _pM , Move on to talk.

In this way, the correction calculation can be stopped during double talk due to the increase in the level of the near-end input n (t), and when a sharp change in the echo path occurs, this can be detected and the correction calculation can be executed. Also, the amount of cancellation E _pM in the register 19
If / S _pM is mistakenly set to a significantly low value, equation (5)
Although the correction calculation is stopped by the above and the convergence is delayed, the correction calculation can be restarted by the equation (6) in the above-described embodiment, so that there is no large delay.

In the above embodiment, H _j of register 18 and register 18 'of register 18' are
Both H _M have a correction calculation function, but if the output to the transmission output end 22 at the time of detecting the deterioration of the cancellation amount is obtained from H _M , the correction calculation only needs to be executed for H _j. it can.

(Advantages of the Invention) As described above, according to the present invention, it is possible to quickly detect the deterioration of the cancellation amount due to the change in the double talk or the echo path according to the progress of the convergence of the echo canceller. Further, when the deterioration is caused by a sharp change in the echo path, it is possible to follow the cancellation amount by comparing the cancellation amount when the correction is continued and when the correction is stopped and restarting the correction calculation. Further, by this processing, even if the threshold value for detecting the deterioration of the cancellation amount is erroneous, the correction calculation can be restarted, and the echo canceller can be stably operated without significantly impairing the convergence speed.

[Brief description of drawings]

FIG. 1 is a block diagram of an embodiment of the present invention, and FIG. 2 is a schematic block circuit diagram of a system using a general digital echo canceller. 11 ... Far end input end, 12, 12 '... Register, 13, 13' ... Sum of products arithmetic circuit, 15, 15 '... Subtractor, 16, 16' ... Subtractor,
17 ... Control unit, 18,18 '... Register, 19 ... Register, 20,20' ... Register, 21,21 '... Register, 22 ...
… Output end.

Claims (1)

[Claims] 1. A method of controlling an echo canceller for removing an echo associated with 2-wire to 4-wire conversion, comprising: (1) pseudo by a convolution operation of a sampled value sequence of a far-end input and an estimated impulse response inside the echo canceller. Echo is generated, and the process of sequentially estimating the estimated impulse response so that the squared value of the residual echo obtained by subtracting this pseudo echo from the echo signal approaches zero, (2) Estimated impulse response of the echo path and the far end A process of generating a pseudo echo by the input sample value, canceling the echo by subtracting the pseudo echo from the echo signal, and calculating a first residual echo, (3) Far-end input and residual echo, or echo signal and residual By calculating the power ratio of the echo, the cancellation amount of the echo canceller is sequentially calculated, and the double talk detection is performed as the cancellation amount increases. (4) In the process of (3) above, a double is set when the calculated amount of cancellation is smaller than the threshold set in the process of (3) above. Talk detection processing, and (5) during double talk detection, the first residual echo is calculated and output using the estimated impulse response at the time of detection, and the second residual echo is continuously estimated.
(6) During the double talk detection, the power of the second residual echo is a constant multiple of the power of the first residual echo (≤
1) When it is below, it is judged that the estimated impulse response when the estimation is continued is valid, and thereafter, the threshold value for double talk detection is reset based on the cancellation amount of the estimated impulse response when the estimation is continued. By setting and setting the impulse response used in the calculation of the second residual echo as an initial value to estimate the impulse response and calculate the first residual echo, the impulse response is estimated during double talk. When the echo path is changed, the echo response is followed to make an impulse response so that the echo can be eliminated. Furthermore, the threshold for double talk detection set in the processing of (3) above is the true cancellation amount. An echo canceller control method characterized in that it is possible to avoid a decrease in the convergence speed that occurs when the value is far below.