JPH0821881B2 - Double-talk detection control method of eco-cancera - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION (Industrial field of application) The present invention relates to an echo canceller for canceling an echo signal which causes deterioration of the call quality of satellite communication, a voice call, etc. The present invention relates to a double-talk detection control method for an echo canceller that performs time processing.

(Prior Art) Conventionally, as a technology in such a field, "Application of Digital Signal Processing", 3rd Edition (Showa 58-7-10), IEICE
P.212-221, JP-A-58-219837, and JP-A-58-219837.
There was a thing described in 61-56526. The configuration will be described below with reference to the drawings.

FIG. 2 is a configuration block diagram of the conventional echo canceller described in the above-mentioned document.

Generally, in a long distance telephone line, the subscriber line 1 connected to the telephone at the subscriber's end adopts a two-wire system for transmitting signals in both directions on one line from the viewpoints of economy of the line and convenience of exchange. To be done. On the other hand, the long-distance line 2 to which the subscriber line is connected requires an amplifier for compensating for the line loss, so a four-wire configuration in which a line is prepared for each direction is adopted. . A hybrid coil 3 is connected to the conversion units for mutual conversion between the two-wire system and the four-wire system, and the impedance Zl of the hybrid coil 3 and the impedance of the subscriber line 1 on the two-wire side are matched. ing. However, since the subscriber lines have different line types and line lengths for each subscriber line, the impedances of the subscriber lines 1 are different, and it is difficult to perfectly match the impedance Zl of the hybrid coil 3. . Therefore, the received signal Rin from the far-end talker A into the four long-distance line 2 is the hybrid coil 3
In addition to flowing to the near-end talker B side through the subscriber line 1, the hybrid coil 3 leaks as an echo signal Sin to the transmitting side as an echo path and deteriorates the call quality. Therefore, in order to erase the echo signal Sin, the echo canceller 10 is connected to the 2-line to 4-line conversion section.

The echo canceller 10 includes analog / digital converters (hereinafter referred to as A / D converters) 11 and 12, digital / analog converters (hereinafter referred to as D / A converters) 13 and 14, and adaptive digital filters (Adaptive Digital Filter). , Below, "AD
F ") 15, subtractor 16, and double-talk detector 17
It is composed of

When the reception signal Rin is input from the far-end talker A to the long-distance line 2, the reception signal Rin is transmitted by the A / D converter 11 at time k.
Is sampled into discrete values at and converted into a digital received signal Rin (k). Digital received signal Rin
(K) is converted into an analog reception signal Rout by the D / A converter 13 and transmitted to the near-end talker B side through the hybrid coil 3 and the subscriber line 1, but in the case of impedance mismatch, the analog reception signal Rout A part of Rout wraps around to the transmission side through the echo path C in the form of the echo signal Sin. The sneak echo signal Sin is sampled into discrete numerical values at time k by the A / D converter 12, and the digital echo signal Si
After being converted into n (k), it is given to the subtractor 16.

The ADF 15 estimates the characteristic of the echo path C, and based on the estimated characteristic and the digital received signal Rin (k), the pseudo echo signal in estimated to be an echo signal by calculation.
(K) is generated and this signal in (k) is given to the subtractor 16. The subtractor 16 subtracts the pseudo echo signal in (k) from the digital echo signal Sin (k), and the residual signal Res
(K) is output. The ADF 15 converges this residual signal Res (k) to zero and eliminates the echo signal Sin.

The estimation operation (that is, adaptive operation) of the ADF15 is based on the received signal R
The normal operation is performed in the single talk state in which only in exists, but the double talk state in which the reception signal Rin and the transmission signal N from the near-end speaker B simultaneously exist may disturb the estimation operation of the AFD 15. So, double talk detector
17 compares a fixed double-talk detection threshold value preset therein and the level of the residual signal Res (k) (for example, average voltage, average power, peak voltage, peak power),
When the level of the residual signal Res (k) becomes larger than the double-talk detection threshold, it is determined that the state is the double-talk state, and the prohibition signal INH is output to prohibit the estimation operation of the ADF 15. As a result, the ADF 15 performs only the operation of outputting the pseudo echo signal in (k). Therefore, the transmission signal N and the echo signal Sin of the near-end talker B are converted by the A / D converter 12 into the digital signal Sin.
(K) + N (k) and the subtractor 16 subtracts the difference from the pseudo echo signal in (k) to obtain the echo signal Sin.
After (k) is erased, only the transmission signal N (k) of the near-end speaker B, which is the residual signal Res (k), becomes the D / A converter 1
It is converted into an analog signal in 4 and transmitted to the far-end talker A side in the form of a transmission signal Res.

However, in the double-talk detection control method using the double-talk detector 17 of this type, the double-talk detection threshold is fixed, and therefore, depending on this set value, the estimation operation is unnecessarily prohibited due to fluctuations in the echo path C or the like. It may occur or the double-talk detection accuracy may deteriorate. Therefore, in order to solve this, a technique for changing the double talk detection threshold value is described in the above-mentioned document.

In the technique of the above document, the level ratio between the residual signal Res (k) and the received signal Rin (k) In the double talk detection control method of observing, and determining that there is double talk when -X exceeds the double talk detection threshold -Vt, the double talk detection threshold -Vt (k) at the time of sampling at time k is set to double talk. According to the detection result, it moves as shown in the following (a) and (b).

(A) When the double talk is judged, the adaptive operation of the ADF is prohibited, and as shown in the equation (2), Vt (k + 1) = Vt (k) −δu (2) The amount of correction set in advance by switching the switch. −δu
(Δu> 0) is selected, Vt (k) is reduced by this correction amount −δu, the threshold −Vt is raised, and the direction is moved in the direction in which it is difficult to determine double talk.

(B) Canceling the ADF adaptive operation prohibition at the time of receiving other than the double talk judgment, and as shown in the equation (3), Vt (k + 1) = Vt (k) + δD (3) It is preset by switching the switch. Correction amount + δD
(ΔD> 0) is selected, Vt (k) is increased by this correction amount + δD, the threshold value −Vt is lowered, and it is moved in the direction in which it is easy to determine double talk.

By thus performing the operations of the expressions (2) and (3), the detection accuracy and the detection speed are improved.

(Problems to be Solved by the Invention) However, the technology of the literature has the following problems.

(I) Since the threshold −Vt is expressed by a monotonically increasing formula such as the formula (3) when receiving a voice other than the double talk determination, the double talk detection sensitivity in the initial convergence process in the ADF 15 is low.

(Ii) When selecting the correction amount −δu with a switch, be sure to
Since the ADF15 is accompanied by the prohibition operation, the prohibition operation is unnecessarily generated for a voice or the like in which the instantaneous power change is large. Therefore, when there is a constant fluctuation in the echo path, the echo canceller 10 does not follow well.

(Iii) Since the correction amounts δu and δD in the equations (2) and (3) are set based on an empirical rule, it is not always the case that an appropriate threshold −Vt is obtained depending on the statistical properties of the received signal Rin. Therefore, the double-talk detection sensitivity may deteriorate.

The present invention has the following problems.
(EN) A double-talk detection control method for an echo canceller, which solves the problem that the double-talk detection accuracy deteriorates and the estimation prohibition operation occurs unnecessarily.

(Means for Solving Problems) In order to solve the above problems, the present invention estimates the characteristics of an echo path based on a received signal and a transmitted signal to which an echo signal of this received signal is added, by an ADF. Generates a pseudo-echo signal and subtracts the pseudo-echo signal from the transmission signal to send out a residual signal in which the echo signal has been canceled. During operation of the echo canceller, a double-talk state is detected based on the double-talk detection threshold value. Then
In the double talk detection control method of the echo canceller which prohibits the estimation operation of the ADF, the double talk detection threshold is changed by the following control procedure.

That is, the sound and silence of the received signal is determined, when the received signal is silent, the estimation operation of the ADF is prohibited and the double talk detection threshold is stored, and when the received signal is sound, Whether the transmitted signal (including an echo signal) has a sound is determined. If the transmitted signal is silent, the prohibition of the estimation operation of the ADF is released and the double talk detection threshold value is rapidly lowered. If the transmission signal is voiced, a level difference between the reception signal and the residual signal is obtained, and a value obtained by adding a margin to the level difference and the double talk detection threshold value are compared, and when the value is smaller than the latter value. , A double-talk state is determined and the estimation operation of the ADF is prohibited. When the above is larger than the latter, the single-talk state is determined and the inhibition of the ADF estimation operation is released.

When it is determined that the state is the single talk state, the difference between the powers of the received signal and the residual signal is multiplied by the first coefficient δ1 to obtain the first multiplication result, and the past value of the double talk detection threshold is set. The second coefficient (1-δ1) is multiplied to obtain the second multiplication result, and the first and second multiplication results are added to update the double talk detection threshold. On the other hand, when it is determined that the state of double talk, subtract the value obtained by adding a margin to the power difference between the received signal and the residual signal from the past value of the double talk detection threshold,
The subtraction result is multiplied by the third coefficient δ2 to obtain the third multiplication result, and the third multiplication result is subtracted from the past value of the double talk detection threshold to update the double talk detection threshold.

(Operation) According to the present invention, since the double-talk detection control method of the echo canceller is configured as described above, the margin added to the level difference between the received signal and the residual signal corresponds to the minute fluctuation of the echo path. It works to prevent the ADF from presuming to operate. When it is determined that the state is single-talk, the first and second multiplication results are added to update the double-talk detection threshold value. When it is determined that the state is the double-talk state, the third multiplication result is set as the double-talk detection threshold value. By updating the double talk detection threshold value by subtracting from the past value of, the unnecessary inhibition of the ADF estimation operation is prevented. Further, when the transmission signal (including the echo signal) is silent, the ADF estimation operation inhibition can be suppressed by rapidly lowering the double talk detection threshold. By these controls, detection accuracy can be improved and accurate operation can be performed. Therefore, the above problem can be eliminated.

(Embodiment) FIG. 1 is a configuration block diagram of a double talk detector for explaining a double talk detection control method according to an embodiment of the present invention.

This double-talk detector is a circuit provided in the conventional echo canceller of FIG. 2 and receives a received signal Rin (k), a residual signal Res (k), and an echo signal Sin (k), which are digital signals, respectively. It has input terminals 20, 21, and 22 for inputting.

Here, as shown in FIG. 2, the echo signal Sin (k) is obtained by adding the signal leaked to the communication side by the reception signal Rin using the hybrid coil 3 as the echo path C and the transmission signal N of the near-end talker B. It is a signal that has been processed. However, during single talk, the received signal Rin is substantially the same as the hybrid coil 3
Is a signal leaking to the communication side as an echo path C (excluding the transmission signal N of the near-end speaker B), and during double talk,
The reception signal Rin is a signal obtained by adding the signal leaking to the communication side using the hybrid coil 3 as the echo path C and the transmission signal N of the near-end talker B.

Therefore, in the following description of the present embodiment, when the pseudo echo signal in (k) of the ADF 15 is described, the echo signal Si
n (k) is premised on a signal (excluding the transmission signal N of the near-end talker B) in which the reception signal Rin leaks to the communication side using the hybrid coil 3 as the echo path C. In addition, in the description of detecting the double talk, the power level of the signal added to the transmission signal N of the near-end speaker B becomes the comparison target of the double talk, so that the echo signal Sin (k) is the received signal. Rin is a signal obtained by adding the signal leaking to the communication side using the hybrid coil 3 as the echo path C and the transmission signal N of the near-end talker B.

Power detectors 23, 24, 25 are connected to the input terminals 20, 21, 22 respectively, and the power detectors 23, 24, 25 are further connected.
Are connected to logarithmic converters 26, 27 and 28, respectively. The power detectors 23, 24, 25 detect the power (for example, average power, peak power, etc.) of the reception signal Rin (k), the residual signal Res (k), and the echo signal Sin (k), and output the signals.
This is a circuit for obtaining Prin (k), Pres (k), and Psin (k). Each of the logarithmic converters 26, 27, 28 is connected to the signals Prin (k), Pres
(K) and Psin (k) are each the logarithmic domain dB value level (this level corresponds to the signal amplitude or voltage).
It is a circuit for converting into signals Lrin (k), Lres (k) and Lsin (k).

A subtractor 29 is connected to the output side of the logarithmic converters 26 and 27,
Further, the subtractor 29 is connected to the adder 30 and the threshold controller 31. Subtractor 29 produces signals Lrin (k) and Lres (k)
Level difference (that is, the received signal Rin (k) and the residual signal R
es (k) power difference), and the signal of this level difference
This is a circuit that outputs Acoms (k) and supplies it to the adder 30 and the threshold control unit 31. The adder 30 is a circuit that adds a margin value γ to the signal Acoms (k) to obtain a signal FLG (k) and supplies the signal FLG (k) to the threshold control unit 31. The margin value γ is a margin value for allowing the ADF 15 to follow a slight change in the echo path C in FIG. The threshold controller 31 uses the signal Ac
It is a circuit that inputs oms (k), FLG (k), etc. and generates a variable double-talk detection threshold TRIM (k + 1).

Comparators 32 and 33 are connected to the output sides of the logarithmic converters 26 and 28, respectively. One comparator 32 has a reference signal XTH
And a signal Lrin (k) are compared with each other, a silence of the received signal Rin (k) is detected, and a prohibition signal INH and a control prohibition signal S32 corresponding to the estimation operation are output. That is, the comparator 32 detects Lrin (k) <XTH and outputs the inhibition signal INH of logic "1" to inhibit the update of the estimation operation in the ADF15 of FIG. It has a function of outputting and prohibiting the update of the double talk detection threshold TRIM (k) in the threshold control unit 31 and preventing the estimation operation of the ADF15 from being disturbed by idle noise (silent noise) of the received signal Rin (k). are doing. The other comparator 33 has a reference signal YT.
H is compared with the signal Lsin (k), and when Lsin (k) <YTH, the silence of the transmission signal (including the echo signal Sin (k)) is detected and ADF is detected.
Clear signal CL1 to set the estimated operation prohibition signal INH to zero
And a clear signal for clearing the double talk detection threshold TRIM (k) of the threshold controller 31 to, for example, zero.
This circuit outputs CL2. That is, this comparator 33
Has a function of detecting the switching of the echo path such as a circuit interruption, the case where the echo path C has a flat delay, and the like so that the ADF 15 can be estimated and operated.

Further, a comparator 34 is connected to the output side of the adder 30, and the output side of the comparator 34 is connected to an output terminal 36 via a prohibit clear circuit 35. This output terminal 36 is ADF15 of FIG.
It is connected to the. The comparator 34 includes a threshold value TRIM (k) calculated at the time (k−1) by the threshold value control unit 31 and an adder.
Compare with 30 output signal FLG (k), TRIM (k) ≧ F
The LG (k) judges that it is a double talk and outputs a "1" prohibition signal INH to prohibit the ADF estimation operation, and TRIM (k) <F
LG (k) judges that it is a single talk, and the prohibition signal IN of "0"
H is output and the threshold value TRIM is output according to these judgment results.
The control signal S34 for selecting the control method of (k) is set to the threshold control unit 3
It has a function to output to 1. The prohibit clear circuit 35
For example, it is composed of a switch, and outputs the inhibit signal INH input in the normal state as it is to the output terminal 36, and when the output clear signal CL1 of the comparator 33 is input, it switches to the ground side and becomes "0".
It has a function to output the inhibit signal INH.

FIG. 3 is a circuit diagram showing a configuration example of the threshold value control unit 31 of FIG. The threshold control unit 31 has an input terminal 40 for inputting a signal Akms (k) of a power difference between a received signal Rin (k) and a residual signal Res (k), and an input terminal 4 for inputting a signal FLG (k).
1, and double talk detection threshold TRIM (k), TRIM (k +
It has an output terminal 42 for outputting 1), .... Input terminal
Circuits 50 and 60 are connected to 40 and 41, respectively.
The output sides of 50 and 60 are connected to the output terminal 42 via the selector switch 70 and the limiter 71.

One circuit 50 is a circuit for integrating the signal Acoms (k) and has a multiplier 51 and an adder 52 connected in series to the input terminal 40, and one input side of this adder 52 has a unit delay element 53.
And the multiplier 54 to the output terminal 42. The multiplier 51 multiplies the signal Acoms (k) by the first coefficient δ1 and gives the first multiplication result Acoms (k) · δ1 to the adder 52. The unit delay element 53 has a delay time Z ⁻¹ . Output terminal 4
This circuit holds the past threshold TRIM (k) of the double talk detection threshold TRIM (k + 1) on the second side. The multiplier 54 is a circuit and an adder which multiply the past threshold value TRIM (k) by the second coefficient (1-δ1) and give the second multiplication result TRIM (k) · (1-δ) to the adder 52. 52 is Acoms (k) ・ δ1 and TRIM (k)
A circuit for adding (1-δ1). The other circuit 60
Is a subtracter 61 and a multiplier 6 connected in series to the input terminal 41.
2 and a subtractor 63, and one input side of the subtractor 63 is connected to the output terminal 42 side via a unit delay element 64.
The unit delay element 64 has a delay time Z ⁻¹ and holds the past value TRIM (k) of the threshold value TRIM (k + 1) on the output terminal 42 side. The subtracter 61 receives the signal FLG (k) from the threshold value TRIM (k). ) To obtain a subtraction value TRIM (k) -FLG (k), the multiplier 62 adds a third coefficient δ to the subtraction value TRIM (k) -FLG (k).
A circuit that obtains a multiplication value (TRIM (k) −FLG (k)) · δ2 by multiplying 2 by a threshold value TRIM (k) and a multiplication value (TRIM
This is a circuit for subtracting (k) -FLG (k)). Delta.2. Each of the circuits 50 and 60 stops the arithmetic operation by the control prohibiting signal S32 of the comparator 32.

The selector switch 70 connected to the output side of these circuits 50 and 60 has a terminal 70a connected to the output side of the adder 52,
It has a terminal 70b connected to the output side of the subtractor 63, a terminal 70c connected to the ground, and a common terminal 70d connected to the input side of the limiter 71, and based on the control signal S34 of the comparator 34, a single talk Has a function of switching to the terminal 70a side in the case of, and to the terminal 70b side in the case of double talk, and further has a function of switching to the terminal 70c side by the clear signal CL2 of the comparator 33. The limiter 71 is a circuit that limits the maximum value and the minimum value of the threshold value TRIM (k).

FIG. 4 is a signal waveform diagram during single talk, FIG. 5 is a signal waveform diagram during double talk, and FIG. 6 is a signal waveform diagram during echo path fluctuation.
The control procedure of FIGS. 1 and 3 will be described. 4 to 6, the horizontal axis represents time (sec) t, and the vertical axis represents the signal level value in dB.

First, in FIG. 1, the received signal Rin (k) and the residual signal R
When es (k) and the echo signal Sin (k) are supplied to the respective input terminals 20, 21, 22, the power of the received signal Rin (k) is detected by the power detector 23 and the signal Prin (k) is detected. )
Logarithm converter 26 converts the logarithmic signal Lrin
It is converted into (k) and given to the subtracter 29 and the comparator 32. Similarly, the residual signal Res (k) is converted into the signal Pres (k) by the power detector 24, and further, the signal Lr is converted by the logarithmic converter 27.
It is converted into es (k) and given to the subtractor 29. Further, the echo signal Sin (k) (this is a signal obtained by adding the signal leaked to the communication side by the reception signal Rin of FIG. 2 with the hybrid coil 3 as the echo path C and the transmission signal N of the near-end talker B). ) Is converted by the power detector 25 into a signal Psin (k),
Further, it is converted into a signal Lsin (k) by the logarithmic converter 28 and given to the comparator 33.

Output signals Lrin (k) and Lres of each logarithmic converter 26, 27
(K) is subtracted by the subtractor 29, and this subtraction signal Acoms
(K) is supplied to the threshold controller 31 and the adder 30. The adder 30 adds the margin value γ to the signal Acoms (k) and supplies the added value signal FLG (k) to the threshold value control unit 31 and the comparator 34.

Each of the comparators 32, 33, 34 performs the determination operation in the order of 32, 33, 34. First, the comparator 32 detects the reference signal XTH and the signal Lrin.
(K) is compared, and if Lrin (k) <XTH, the received signal R
It is determined that in (k) is silent, and the inhibition signal INH of "1" and the control inhibition signal S32 are output. The inhibition signal INH of "1" is given to the ADF 15 of FIG. 2 through the selector switch 35 and the output terminal 36, and the estimation operation of this ADF 15 is stopped. further,
The control prohibition signal S32 prohibits the double-talk detection threshold TRIM (k) from being updated in the threshold controller 31. At this time, the other comparators 33 and 34 do not execute the comparison operation. Comparator 32 is Lrin
When it is determined that (k) ≧ XTH, the comparator 33 executes the comparison operation, and if Lsin (k) <YTH, the transmission signal (echo signal S
(including in (k)) determines that there is no sound and outputs clear signals CL1 and CL2. Output terminal 3
6 The inhibit signal INH on 6 is cleared to zero, for example, and ADF15 enters the presumed operation state, and the other clear signal CL2
Causes the selector switch 70 in the threshold control unit 31 to move to the terminal 70.
The threshold value TRIM (k) is cleared to, for example, zero by switching to the c side. At this time, the comparator 34 does not execute the comparison operation. When the comparator 33 determines that Lsin (k) ≧ YTH, the comparator 34
Performs a comparison operation, and if TRIM (k) ≥ FLG (k) determines double talk, outputs a prohibition signal INH of "1", and TRIM (k) ≥ FLG (k)
If (k) <FLG (k), it is determined to be single talk, and the inhibition signal INH of "0" is output. The inhibit signal INH of “1” is
It is given to the ADF 15 through the selector switch 35 and the output terminal 36, and the estimation operation of this ADF 15 is prohibited. This comparator 3
The determination result of 4 is given to the threshold control unit 31 in the form of the selection signal S34, and in the case of single talk, the selector switch 70
Is switched to the terminal 70a side, and in the case of double talk, the selector switch 70 is switched and connected to the terminal 70b side.

In the threshold controller 31 of FIG. 3, two different signals Acoms
The threshold TRIM (k) is updated by (k) or FLG (k), and the updated threshold TRIM (k + 1) is output via the selector switch 70.

When it is determined that the single talk state (TRIM (k) <FLG (k)), the selector switch 70 is connected to the terminal 70a side, and the double talk detection threshold TRIM (k + 1) updated by the circuit 50 with the signal Acoms (k). ) Is selected. That is, in the circuit 50, the multiplier 51 calculates Acoms (k) · δ1 and the multiplier 54 calculates TRIM (k) of the unit delay element 53.
And the second coefficient (1-δ1) multiplied by TRIM (k) · (1
When −δ1) is obtained, the adder 52 obtains a threshold value TRIM (k + 1) TRIM (k + 1) = Acoms (k) · δ1 + TRIM (k) (1-δ1) (4), which is the selector switch 70 and the limiter 71. , And output terminal 42. Here, if the first coefficient δ1 and the second coefficient (1-δ1) are appropriately set, the threshold value TRIM
(K) becomes an integral curve of the signal Acoms (k) as shown in FIG. The signals FLG (k) and TRIM (k) always maintain a margin value γ (dB) and satisfy FLG (k)> TRIM (k). Therefore, the detection sensitivity is kept constant and ADF1
5 is not unnecessarily prohibited from presuming motion.

When it is determined that the double talk state (TRIM (k) ≧ FLG (k)), the select switch 70 is connected to the terminal 70b side, and the double talk detection threshold TRIM (k + 1) updated by the circuit 60 with the signal FLG (k). ) Is selected. That is, in the circuit 60, the subtracter 61 is configured to receive the received signal Rin (k).
Signal Acoms (k) of the power difference between the residual signal and the residual signal Res (k)
The signal FLG (k) having a value added with a margin is subtracted from the threshold value TRIM (k) of the past value of the double talk detection threshold value output from the unit delay element 64. This subtracted value TRIM (k)
-FLG (k) and the third coefficient δ2 are multiplied by the multiplier 62. The subtractor 63 uses the multiplication value (TR
IM (k) −FEG (k)) · δ2 is subtracted from the threshold value TRIM (k) of the past value of the double talk detection threshold value to obtain the threshold value TRIM.
(K + 1) is calculated (updated).

TRIM (k + 1) = TRIM (k) − (TRIM (k) −FLG (k)) · δ2 ...
(5) The calculated threshold value TRIM (k + 1) is the selector switch
It is output through the 70, the limiter 71, and the output terminal 42.

In this way, the circuit 60 has the threshold TRIM (k) and the signal FLG.
The threshold value TRIM (k + 1) is decreased in proportion to the level difference of (k). Threshold value TRIM (k) when coefficient δ2 is set appropriately
And the movement of the signal FLG (k) is shown in FIG. Fifth
As shown in the figure, when the double-talk detection threshold TRIM (k + 1) is updated as shown in equation (5), the double-talk detection threshold decreases following the signal FLG (k). For the increase of k) (single talk state),
The direction of double-talk detection cancellation can be dealt with quickly.

That is, when double talk occurs, the signal Lrin (k)
And Lres (k) level difference (that is, received signal Rin
The difference between (k) and the power of the residual signal Res (k)) rapidly decreases, and TRIM (k) ≧ FLG (k) holds, so that the adaptive operation of the ADF 15 is prohibited. However, when double talk ends, the signal FLG (k) increases, so TRIM (k) <FLG
(K) is reached, and the ADF 15 restarts the adaptive operation. Note that the amount of decrease in the threshold value TRIM (k) changes depending on how to select the coefficient δ2, and the estimation prohibited section of the ADF 15 can be set.

When a rapid change in the echo path C occurs, as shown in FIG. 6, TRIM (k) ≧ FLG (k) holds and ADF15
However, since the threshold value TRIM (k) decreases, TRIM (k) <FLG (k) holds after a certain time, and A
DF15 starts estimation operation again.

Further, in the threshold control unit 31 of FIG. 3, the clear signal CL
When receiving 2, the selector switch 70 is connected to the terminal 70c, and the unit delay elements 53 and 64 and the output are cleared to, for example, zero. When receiving the control inhibit signal S32 from the comparator 32, the threshold value TRIM (k) is stored in the unit delay elements 53 and 64 without any updating, and the threshold value TRIM (k) is output as it is.

 The advantages of this embodiment can be summarized as follows.

Almost the same advantages as those described in the above literature are
It can also be obtained in this embodiment.

That is, the received signal Rin (k), the residual signal Res (k),
Also, since the power of the echo signal Sin (k) is detected and converted into the logarithmic domain, it is easy to grasp each signal level. Since the attention is paid to the signal Acoms (k) of the level difference between the reception signal Rin (k) and the residual signal Res (k), the double talk detection sensitivity is high. Even if the estimation operation of the ADF 15 is prohibited due to double talk or echo path fluctuation, the threshold value TRIM (k) decreases with time and the adaptive operation of the ADF 15 is restarted.

Since the margin value γ is provided, the estimation operation of the ADF 15 is not prohibited even if the echo path C changes slightly.

Since the calculation of the threshold value TRIM (k) is realized by the integration circuit 50 for the signal Acoms (k), the threshold value TRIM (k) is calculated as the signal Acoms.
Unnecessary AD guaranteed to move along (k)
The estimation prohibition action of F15 does not occur.

Since the comparator 33 for detecting the silence of the transmission signal (including the echo signal Sin (k)) is provided, the transmission signal (echo signal Sin) is received even though the reception signal Rin (k) is input.
If there is no input (including (k)), the echo path C is switched, it is judged as flat delay, infinite loss, and the threshold value TRIM is immediately set.
Since (k) is cleared, the estimation operation of ADF15 is not prohibited.

It should be noted that the present invention is not limited to the illustrated embodiment, and various modifications such as changing the circuit configurations of FIGS. 1 and 3 to other ones are possible.

As described in detail above, according to the present invention, since a margin is added to the level difference between the received signal and the residual signal,
ADF estimation operation is not prohibited even with a slight change in the echo path. When it is determined that the state is single-talk, the first and second multiplication results are added to update the double-talk detection threshold value. When it is determined that the state is the double-talk state, the third multiplication result is set as the double-talk detection threshold value. Since the double-talk detection threshold value is updated by subtracting from the past value of, the unnecessary ADF estimation prohibition operation does not occur. Furthermore, since the silence of the transmission signal (including the echo signal) is detected and the double talk detection threshold value is rapidly lowered, the estimation operation of the ADF is not prohibited. Therefore, stable double-talk detection control with high detection accuracy is possible.

[Brief description of drawings]

FIG. 1 is a configuration block diagram of a double talk detector for explaining a double talk detection control method according to an embodiment of the present invention, FIG. 2 is a configuration block diagram for explaining a conventional echo canceller, and FIG. Is a circuit diagram of the threshold control unit in FIG. 1, FIG. 4 is a signal waveform diagram during single talk in FIG. 1,
FIG. 5 is a signal waveform diagram during double talk in FIG. 1, and FIG. 6 is a signal waveform diagram during echo path fluctuation in FIG. 3 ... Hybrid coil, 10 ... Echo canceller,
15 …… adaptive digital filter (ADF), 23,24,25 ……
Power detector, 26, 27, 28 ... Logarithmic converter, 31 ... Threshold control unit, 32, 33, 34 ... Comparator, 35 ... Clear circuit, INH ...
… Inhibition signal, Rin (k) …… Received signal, Res (k) …… Residual signal, Sin (k) …… Echo signal, TRIM (k) …… Double talk detection threshold.

 ─────────────────────────────────────────────────── ─── Continuation of front page (72) Inventor Yoshikazu Nakano 1-7-12 Toranomon, Minato-ku, Tokyo Oki Electric Industry Co., Ltd. (56) Reference JP-A-62-299120 (JP, A)

Claims (1)

[Claims] 1. An adaptive digital filter estimates a characteristic of an echo path based on a reception signal and a transmission signal to which an echo signal of the reception signal is added to generate a pseudo echo signal, and the pseudo echo signal is generated by the transmission signal. When the echo canceller that outputs the residual signal in which the echo signal is canceled by subtracting from the above is detected, the double talk state is detected with the double talk detection threshold as a reference, and the double operation of the echo canceller that prohibits the estimation operation of the adaptive digital filter is performed. In the talk detection control method, the presence or absence of sound of the received signal is determined, and when the received signal is silent, the estimation operation of the adaptive digital filter is prohibited and the double talk detection threshold is stored, and the received signal is When there is a sound, the presence or absence of the sound of the transmission signal is judged and the transmission signal is judged. If there is no sound, the prohibition of the estimation operation of the adaptive digital filter is released and the double talk detection threshold value is rapidly lowered. If the transmitted signal is sound, the level difference between the received signal and the residual signal is obtained. , Comparing the value obtained by adding a margin to the level difference with the double-talk detection threshold value, and when the above is smaller than the latter, it is determined to be in the double-talk state and the estimation operation of the adaptive digital filter is prohibited, Is larger, the single-talk state is determined, the prohibition of the estimation operation of the adaptive digital filter is released, and when the single-talk state is determined, the first coefficient is added to the power difference between the received signal and the residual signal. δ1 is multiplied to obtain the first multiplication result, and the past value of the double talk detection threshold is multiplied by the second coefficient (1-δ1) to obtain the second value.
Of the power of the received signal and the residual signal when the double-talk detection threshold is updated by adding the first and second multiplication results A value obtained by adding a margin to the difference is subtracted from the past value of the double-talk detection threshold value, the subtraction result is multiplied by a third coefficient δ2 to obtain a third multiplication result, and the third multiplication result is multiplied by the double value. A double-talk detection control method for an echo canceller, which is characterized in that the double-talk detection threshold is updated by subtracting from the past value of the talk-detection threshold.