JP2814777B2 - Echo canceller control method - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

The present invention relates to a two-wire telephone line.
The present invention relates to an echo canceller control method used for a four-wire conversion circuit, an electronic conference system, and the like.

[0002]

2. Description of the Related Art FIG. 3 shows a functional block diagram of a conventional echo canceller. In FIG. 3, a reception input terminal 1 is connected to a reception output terminal 2, a shift register 3 for storing a reception input sequence, and a control unit 8. The shift register 3 is connected to the convolution unit 5 and the multiplier 9 in addition to the blocks 1, 2 and 8. The transmission input terminal 4 is connected to a subtractor 6 and a control unit 8. The convolution unit 5 is connected to a shift register 3, a subtractor 6, a memory 7, a multiplier 9, and an adder 10. The subtractor 6 is connected to the control unit 8 and the transmission output terminal 11 in addition to the blocks 4 and 5. The memory 7 storing the estimated impulse response of the echo path from the reception output terminal 2 to the transmission input terminal 4 is connected to the arithmetic unit 5 and the adder 10. The control unit 8 controls each of the blocks 1, 2, 3, 4, 6, 9, 1, 1
1 connected. The multiplier 9 is connected to the above blocks 3, 5,
8 and is connected to an adder 10.

The operation of the apparatus shown in FIG. 3 is as follows. First, when a reception input xj is input from the reception input terminal 1, xj is output from the reception output terminal 2 and is taken into the shift register 3 as a vector Xj. The reception signal xj output from the reception output terminal 2 is partially input into the transmission input terminal 4 as an echo yj. Since the transmission signal nj is also input to the transmission input terminal 4, sj = yj + n
The transmission signal including the ambient noise and the like when the signal j is input affects as a disturbance to the echo signal. On the other hand, inside the echo canceller, the memory 7 stores the data from the reception output terminal 2 to the transmission input terminal 4.
The estimated impulse response of the echo path leading to
The convolution operation unit 5 performs a convolution operation on Hj and the received input sequence Xj to generate a pseudo echo yj. The subtractor 6 calculates a pseudo echo yj from the transmission input sj.
And the residual echo ej = sj-yj
And outputs it from the transmission output terminal 11. Next, the control unit 8 calculates a square norm value {Xj} ² of the received input sequence, and further calculates and outputs a correction coefficient αej / {Xj} ² of the estimated value. The multiplier 9 corrects the estimated value by multiplying the received input column vector Xj by the correction coefficient and adding the result to the estimated impulse column vector by the adder 10.

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

[0005]

(Equation 1)

[0006]

(Equation 2)

[0007] When the echo canceling operation of (Equation 1) and the estimation value correcting operation of (Equation 2) are alternately performed in the above procedure, when the level of the disturbance nj is lower than the level of the echo yj, H
j converges asymptotically to the true impulse response, increasing the echo suppression effect. Further, a period in which the level of the disturbance nj is higher than the level of the echo yj is called a double talk state. At this time, if the correction operation of (Equation 2) is performed, the estimated value is greatly disturbed. The level of the transmission input sj and the level of the transmission output ej are compared to detect a double talk state, and the correction calculation of (Equation 2) is stopped during double talk.

[0008]

However, in the echo canceller using the above-mentioned conventional learning identification method, even if the estimated value once becomes a good value, the estimated value is disturbed by the level fluctuation, and sufficient echo cannot be obtained in ordinary conversation. There was a problem that a suppression effect could not be obtained. This can be explained as follows. Let the true impulse response of the echo path be h
(0),. . . . , H (n−1), and the error from the estimated value is uj (m) = h (m) −hj (m), (Equation 3) is obtained from (Equation 2).

[0009]

(Equation 3)

(Equation 3) can be further arranged as (Equation 4) using (Equation 1).

[0011]

(Equation 4)

Here, the estimation is successfully performed, and the estimation error u
j (m), m = 0,. . , N-1 are almost zero,
The first and third terms on the right side are negligible. If the second term is small, u
j (m) does not change significantly, and is a desirable operation. It can also be seen that, conversely, the level of the disturbance nj limits the estimation error. However, if the level of the disturbance nj is constant and a large level change occurs sharply, such as at the beginning or end of the speech, a signal sequence in which the pulse-like amplitude distribution is offset is stored in the received input sequence. For a signal xm having a large amplitude, xm ² / {xj} ² ≒ 1, and the value of the second term (equation 4) becomes uj + 1 (m) as it is, and the estimation error increases instantaneously.

The present invention solves the above-mentioned conventional problem, and provides an echo canceller manufacturing method capable of preventing deterioration of an estimated value due to level fluctuation and obtaining a sufficient echo suppression effect even in ordinary conversation. Aim.

[0014]

According to the present invention, in order to achieve the above object, a square norm value {xj} ² of a received input sequence and a peak amplitude value of a signal in the received input sequence or its approximate value x
p is obtained, and when xp / {Xj} ² is equal to or larger than a certain threshold value, the correction operation is not executed.

[0015]

With this configuration, the level fluctuation is detected, and when the level fluctuation is detected, the correction of the estimated value is stopped to prevent the deterioration of the estimated value at the beginning and end of the speech, so that the echo suppression effect is impaired even in ordinary conversation. A comfortable call can be made without any interruption.

[0016]

FIG. 1 is a flowchart showing one embodiment of an echo canceller control method according to the present invention. In FIG. 1, ST1 indicates the start of the echo canceller control in the sample processing. At this time, a new reception input xj is taken in, and the echo cancellation calculation shown in (Equation 1) has been completed. Next, in ST2, the square norm {Xj} ² of the received input column vector is calculated using the received input xj.
Since the square norm value {Xj} ² is also used in the correction operation of (Equation ² ), it is a value calculated also in a normal echo canceller. Further, in ST3, the maximum amplitude value xp of the signal in the received input sequence is approximately obtained. This detailed procedure will be described later with reference to FIG. Here, when the reception input is at a substantially constant level, the value of xp ² / {Xj} ² is almost the reciprocal of the number of taps N. When the power is concentrated, xp ² / {Xj} ² becomes close to 1, and when the correction operation is performed at this time, the estimated value is disturbed as shown in (Equation 4). Therefore, at ST4, it is detected whether or not xp ² / {Xj} ² exceeds a certain threshold value K.
The normal echo canceller control is performed in step 5, and if it exceeds, the correction calculation is stopped in step ST6. ST7 is the end of the echo canceller control in the sample processing.

FIG. 2 is a flowchart of an embodiment of a procedure for approximately obtaining the maximum amplitude value xp of the signal in the received input sequence. In FIG. 2, flag indicates whether the peak amplitude xp of the signal in the received input sequence is in the increasing state (flag = 0) or decreasing state (flag = 1), and CNT indicates the peak amplitude xp. Is a counter that indicates where in the received input string. And f
The lag, CNT, and xp are cleared once at the start. In the case of this embodiment, when the peak amplitude xp is in the increasing state (flag = 0), a correct peak value xp is obtained, but when the peak amplitude xp is in the decreasing state (flag = 1), xp is obtained as an approximate value. I have. The procedure is described below.
ST8 indicates the start of the sample processing. When the process is started, first, the flag is checked in ST9, and the peak amplitude xp of the signal in the received input sequence is increased (flag =
0) or a decreasing state (flag = 1). S in case of increase (flag = 0)
Proceeding to T10, the amplitude of the newly input reception input | xj
Compare | with xp. When | xj | ≧ xp, ST1
In step 1, xp is updated to | xj |, and the counter CNT is cleared in ST12 to indicate that xp is at the latest time position in the received input sequence. Furthermore, the peak amplitude x
Since p continues to increase, fl
The process proceeds to ST23 with g = 0, and ends the sample processing. When | xj | <xp, it indicates that the position of xp has been shifted by one sample, and the counter CN is set in ST14.
T is increased by 1, and the number of taps N is compared with CNT in ST15. If CNT <xp, since xp exists in the received input sequence, the process proceeds to ST13 and sets flag = 0 to ST2.
Proceed to 3 to complete the sample processing. | Xj | <xp
Since it means that xp has been removed from the received input sequence, flag = 1 is set in ST16 and ST23
To end the sample processing. On the other hand, when the peak amplitude xp is in the decreasing state (flag = 1), since the peak amplitude xp detected in the past is excluded from the reception input sequence, the correct peak value cannot be determined, and the peak value detected in the past in ST17 is determined. With xp as an initial value, a weighted average with the signal amplitude value | xj-N | successively removed from the received input sequence is performed to obtain an approximate value xp of the peak value. ST1
8 compares the approximate value xp with the amplitude of the newly input reception input | xj |, and when | xj | <xp, ST19.
In step ST23, flag = 1 is set, and the process proceeds to ST23 to finish the sample processing. When | xj |
p is updated to | xj |, the counter CNT is cleared in ST21 to indicate that xp is at the latest time position in the received input string, and flag = 0 is set in ST22, and then S is set.
Proceed to T23 to end the sample processing.

As described above, according to the echo canceller control method of the embodiment of the present invention, the peak amplitude x in the reception input sequence
Using the ratio of p and the square norm {xj} ^2, the degree of concentration of the signal amplitude in the received input sequence is detected. When the degree of concentration is large, the correction operation of the estimated value is stopped. An excellent effect is obtained in that the echo suppression effect is not impaired even for a large signal level fluctuation.

[0019]

As is clear from the above embodiments, according to the present invention, the processing for calculating the square norm value of the received input sequence,
A process for detecting or approximating a peak amplitude value in a reception input sequence, a process for calculating the ratio of the peak value / square norm value are provided, and a steep and large level change of the reception input is detected. It is an object of the present invention to provide an echo canceller control method capable of preventing the echo suppression effect from being impaired even in ordinary conversation by stopping the calculation.

[Brief description of the drawings]

FIG. 1 is a flowchart of an echo canceller control method according to an embodiment of the present invention;

FIG. 2 is a flowchart showing one embodiment of a method for calculating an approximate peak amplitude xp in the same method.

FIG. 3 is a functional block diagram of a conventional echo canceller.

[Explanation of symbols]

ST2 Processing for calculating the square norm value of the reception input string ST3 Processing for detecting or approximating the peak amplitude in the reception input string ST4 Processing for comparing the peak value / square norm value with the threshold value ST6 Peak value / square norm value To stop the correction of the echo canceller estimated value when the value exceeds a predetermined threshold

──────────────────────────────────────────────────続 き Continued on the front page (58) Field surveyed (Int.Cl. ⁶ , DB name) H04B 3/00-3/23 H04B 7/005-7/015 H03H 17/00-21/00 H04M 1 / 58-1/60 JICST File (JOIS)

Claims (1)

(57) [Claims] 1. A convolution operation of a received input sequence and an estimated value of an impulse response of an echo path to generate a pseudo echo, and the echo is suppressed by subtracting the pseudo echo from the echo signal. In an echo canceller that successively corrects the above-mentioned estimated value so that the square value of the input sequence approaches zero, a process of calculating a square norm value of a received input sequence used for a convolution operation, and calculating a peak amplitude value in the received input sequence Detecting or approximating; calculating the ratio of the peak value / square norm value;
An echo canceller control method comprising a step of stopping the correction of the estimated value of the echo canceller when the ratio of the square norm value exceeds a certain threshold value.