JPH05218904A - Echo canceller controlling method - Google Patents

Abstract

PURPOSE:To attain a pleasant conversation by solving a problem that an echo suppression effect is deteriorated in the case of the presence of a low frequency 2-way input in the echo canceller used for a 2/4-wire conversion circuit for a telephone line and an electronic conference system or the like. CONSTITUTION:The calculation of a level changing steeply and the calculation of a level changing slowly are executed by using transmission input level calculation sections 9, 15 with a different time constant, transmission output level calculation sections 13, 16 and a control section 17 in the echo canceller control correcting an estimate impulse response. The detection of a double talk whose detection speed is emphasized is based on the former level and the detection of the end of double talk and a change in an echo path whose precision is emphasized to restart the correction calculation is based on the latter level, then the echo is suppressed stably without losing the echo suppression effect even in the case of a low frequency 2-way input caused in the usual conversation.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an echo canceller control method used in a two-line to four-line conversion circuit in a telephone line, an electronic conference system, or the like.

[0002]

2. Description of the Related Art FIG. 2 is a functional block diagram of a conventional echo canceller. In FIG. 2, the reception input end 1 is connected to the reception output end 2, the input side of the shift register 3 that stores the reception input sequence, and the input side of the reception input level calculation unit 4. The output side of the shift register 3 is connected to the input side of the convolution calculator 5 and the input side of the multiplier 6. The transmission input terminal 7 is connected to the input side of the subtractor 8 and the input side of the transmission input level calculation unit 9. The output of the convolution calculator 5 is input to the subtractor 8. The output of the multiplier 6 is input to the adder 10. Reference numeral 11 is a memory in which the estimated impulse response of the echo path from the reception output end 2 to the transmission input end 7 is stored, which is input to the adder 10 and is output to the convolution calculator 5. The subtractor 8 outputs to the transmission output end 12 and the transmission output level calculation unit 13. Control unit 14
Is input from the reception input level calculation unit 4, the transmission input level calculation unit 9, and the transmission output level calculation unit 13, and the multiplier 6
Output to.

The operation of the echo canceller configured as described above will be described below. First, when receiving the input x _j is inputted from the receiving input terminal 1 x _j in the shift register 3 with the output from the receiving output terminal 2 vector X _j
Is taken in as. The reception signal x _j output from the reception output end 2 partially enters the transmission input end 7 as an echo y _j . Since the transmission signal n _j is also input to the transmission input end 7, a signal of s _j = y _j + n _j is eventually input to the transmission input end 7, and the transmission signal including ambient noise or the like corresponds to the echo signal. Affect as a disturbance. On the other hand, in the echo canceller, the estimated impulse response of the echo path from the reception output end 2 to the transmission input end 7 is stored in the memory 11 as a vector H _j , and the convolution calculator 5 calculates H _j and the reception input sequence X. A convolution operation of _j is performed to generate a pseudo echo y _j . Subtracter 8 subtracts the echo replica y _j from the transmission input s _j, by suppressing an echo to obtain the residual echo e _j = s _j -y _j, outputs from transmission output terminal 12. Next, the control unit 14 calculates the square norm value ‖X _j ‖ ² of the received input sequence, and further calculates and outputs the correction coefficient αe _j / ‖X _j ‖ ² of the estimated value. The multiplier 6 multiplies the received input sequence vector X _j by the above correction coefficient, and the result is added to the estimated impulse sequence vector by the adder 10
The estimated value is corrected by adding in.

The algorithm executed in this manner is called the learning identification method shown in (Equation 1) and (Equation 2), and is the one most often used in the echo canceller.

[0005]

[Equation 1]

[0006]

[Equation 2]

When the echo canceling operation of (Equation 1) and the estimated value correcting operation of (Equation 2) are alternately executed in the above procedure, H during the period when the level of the disturbance n _j is lower than the level of the echo y _j.
j asymptotically converges to the true impulse response, and the echo suppression effect increases. Further, a period in which the level of the disturbance n _j is higher than the level of the echo y _j is called a double-talk state, and if the correction calculation of (Equation 2) is performed at this time, the estimated value is greatly disturbed, so the correction calculation must be stopped. I won't. On the other hand, when an echo path change occurs, the estimation error increases, but a correction calculation must be executed to suppress the echo again. Transmission input level calculation unit 9, transmission output level calculation unit 1
3. The reception input level calculation unit 4 calculates the levels Sp _j , Ep _j , and Xp _j of the transmission input s _j , the transmission output e _j , and the reception input x _{j according to} (Equation 3), (Equation 4), and (Equation 5), respectively. calculate.

[0008]

[Equation 3]

[0009]

[Equation 4]

[0010]

[Equation 5]

The control unit 14 detects double talk and echo path change by using these levels, stops the correction calculation of (Equation 2) at the time of double talk, and even when the transmission output level is high at the time of echo path change. Execute correction calculation.

[0012]

However, in the echo canceller control method using the above-mentioned conventional learning identification method,
If the received input x _j and the disturbance n _j are both low in frequency, even if the estimated value H _j is not correct, the positive and negative values of the transmitted input s _j and the pseudo echo y _j calculated by the echo canceller match for a long time. The level of the output e _j (= s _j −y _j ) becomes low. Therefore, the control of the echo canceller such as the detection and termination of double talk and the detection of the echo path change is governed by the positive or negative coincidence or non-coincidence of the echo input s _j and the pseudo echo y _j , and the canceling amount (Sp _j / Ep) as it should be. Correct control cannot be performed by increasing or decreasing _j ). As a result, the disturbance n _j is input, and the equation (2)
However, since the correction calculation cannot be detected and the correction is executed, the estimated value H _j is disturbed and the level of the transmission output e _j fluctuates unstablely.

The present invention solves the above-mentioned conventional problems and is excellent in that the deterioration of the estimated value can be prevented even when there is a low-frequency bidirectional input, and a sufficient echo suppression effect can be obtained even in ordinary conversation. It is an object of the present invention to provide an echo canceller control method.

[0014]

In order to achieve the above object, the present invention executes level calculation with a large time constant in addition to the above level calculation, restarts the correction calculation, and detects double talk termination and echo path change. In addition, a level value that slowly changes even with a low frequency input is used.

[0015]

With this method, when the double talk ends, the echo path change is accurately detected and the correction calculation is restarted by using the level value that changes slowly even for the low frequency input, the case where there is a low frequency bidirectional input In addition, since the estimated value is prevented from deteriorating, even in a normal conversation, the echo suppression effect is not impaired and a comfortable call can be performed.

[0016]

1 is a functional block diagram of an embodiment of an echo canceller control method according to the present invention. In FIG.
13 has the same function as 1 to 13 in FIG. A second transmission input level calculator 15 has a large time constant and is input from the transmission input terminal 7. Reference numeral 16 is a second transmission output level calculator 16 having a large time constant, to which the output of the subtractor 8 is input. Reference numeral 17 denotes a control unit, which is a reception input level calculation unit 4, a transmission input level calculation unit 9, a transmission output level calculation unit 13, and a second transmission input level calculation unit 1.
5, input from the second transmission output level calculator 16 and output to the multiplier 6.

The operation of the echo canceller configured as described above will be described below. First, when the receiving input x _j is input from the receiving input end 1, x _j is received by the receiving output end 2.
And the vector X to the shift register 3
_Ingested as _j . The reception signal x _j output from the reception output end 2 partially enters the transmission input end 7 as an echo y _j . Since the transmission signal n _j is also input to the transmission input end 7, a signal of s _j = y _j + n _j is eventually input to the transmission input end 7, and the transmission signal including ambient noise is a disturbance to the echo signal. Affect. On the other hand, in the echo canceller, the estimated impulse response of the echo path from the reception output end 2 to the transmission input end 7 is stored in the memory 11 as a vector H _j , and the convolution calculator 5 calculates H _j and the reception input sequence X. A convolution operation of _j is performed to generate a pseudo echo y _j .
Subtracter 8 subtracts the echo replica y _j from the transmission input s _j, by suppressing an echo to obtain the residual echo e _j = s _j -y _j, outputs from transmission output terminal 12.

Next, the control unit 17 calculates the square norm value ‖X _j ‖ ² of the received input sequence, and further the correction coefficient αe _j / of the estimated value.
Calculate and output ‖X _j ‖ ² . The multiplier 6 multiplies the received input sequence vector X _j by the above correction coefficient and adds the result to the estimated impulse sequence vector by the adder 10,
Correct the estimated value. The algorithm executed in this manner is the learning identification method shown in (Equation 1) and (Equation 2).

Echo cancellation calculation of (Equation 1) according to the above procedure
If the estimated value correction calculation of
n_jIs the level of echo y_jIs lower than the level of
_jAsymptotically converges to the true impulse response, and the echo suppression effect
Will increase. Also, the disturbance n_jIs the level of echo y_jLevel
It is called a double talk state during the period when it is higher than
If the correction calculation of (Equation 2) is performed, the estimated value will be greatly disturbed.
Therefore, the correction calculation must be stopped. On the other hand
Error occurs, the estimation error increases, but
Correction must be performed to suppress
Yes. Transmission input level calculation unit 9, transmission output level calculation unit 1
3, the reception input level calculation unit 4 is the above (Equation 3), (Equation 3)
4), the transmission input s according to (Equation 5)_j, Transmission output e_j, Receive
Input x _jCalculate the level of. And a second send input
The level calculator 15 and the second transmission output level calculator 16 are
According to (Equation 6) and (Equation 7), (Equation 3), (Equation 4),
Transmission input s with a larger time constant than (Equation 5)_j， Transmission output
e_jCalculate the level of.

[0020]

[Equation 6]

[0021]

[Equation 7]

The control unit 17 uses the rapidly changing levels Sp _j , Ep _j , and Xp _{j according} to (Equation 3), (Equation 4), and (Equation 5) to detect double talk, and ends double talk and echo path. Change is detected by (Equation 6) and (Equation 7)

[0023]

[Outer 1]

[0024]

[Outside 2]

Using the value, the correction calculation of (Equation 2) is stopped at the time of double talk, and the correction calculation is performed at the time of echo path change even if the transmission output level is high.

As described above, according to the present embodiment, the double-talk in which the detection speed is more important is detected at a level that changes abruptly, and the correction calculation is restarted. Since termination and echo path change detection are performed at a level that slowly changes even for low-frequency input, the excellent effect of stable echo suppression is achieved even for low-frequency bidirectional input that occurs in normal conversation. can get.

[0027]

As described above, according to the present invention, the level calculation which changes sharply and the level which changes slowly are executed, and the double talk which is more important in the detection speed is detected at the former level, and the correction calculation is performed. Since the double talk end and echo path change detection, where accuracy is more important for resuming speech, are performed at the latter level, the echo suppression effect is impaired even in the case of low-frequency bidirectional input that occurs in normal conversation. It is possible to realize an excellent echo canceller control method that can stably perform echo suppression without fail.

[Brief description of drawings]

FIG. 1 is a functional block diagram of an embodiment of an echo canceller control method of the present invention.

FIG. 2 is a functional block diagram of a conventional echo canceller.

[Description of Reference Signs] 3 shift register (reception input sequence) for storing reception input 4 reception input level calculation unit (calculation of input / output signal level with small time constant) 5 convolution operation unit (convolution operation) 8 subtractor (Subtract) 9 First transmission input level calculator (calculates input / output signal level with small time constant) 11 Memory storing estimated impulse response of echo path (estimated value of impulse response of echo path) 13 First transmission output level calculation unit (calculates input / output signal level with small time constant) 15 Second transmission input level calculation unit (calculates input / output signal level with large time constant) 16 Second transmission Output level calculator (calculates the input / output signal level with a large time constant)

 ─────────────────────────────────────────────────── ─── Continuation of the front page (72) Inventor Koichi Honma 4-3-1, Tsunashima-higashi, Kohoku-ku, Yokohama-shi, Kanagawa Matsushita Communication Industrial Co., Ltd.

Claims (1)

[Claims] 1. A pseudo echo is generated by performing a convolution operation of an estimated value of an impulse response of a received input sequence and an echo path,
In the echo canceller control method in which the estimated value is sequentially corrected so that the echo is suppressed by subtracting this from the echo signal, and the square value of the residual echo obtained at this time is made closer to zero. 1. A process of calculating the input / output signal level with a small time constant; 2. A process of calculating the input / output signal level with a large time constant, 3. A process of making a determination using the rapidly changing level of 1 for double-talk detection, which requires high speed, and reception input drop detection, An echo canceller control method characterized by executing double talk end detection, which requires accuracy, and determination processing using the above-mentioned two levels with little fluctuation in echo path change detection.