JP3248551B2 - Echo canceler - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a reverberation control apparatus for eliminating or suppressing a reverberation signal which causes howling and impairs hearing in a two-wire / four-wire conversion system and a loudspeaker system.

[0002]

2. Description of the Related Art FIG. 5A is a schematic diagram of a loudspeaker system.
The transmitted voice uttered by the transmitter 10 is sequentially supplied to the receiving speaker 15 through the transmitting microphone 11, the transmitting signal amplifier 12, the transmission line 13a, and the receiving signal amplifier 14, and the reproduced voice from the speaker 15 is received. To the person 16. On the other hand, the voice uttered by the receiver 16 includes a transmission microphone 17, a transmission signal amplifier 18, a transmission line 13b, a reception signal amplifier 19,
The signal is transmitted to the transmitter 10 via the receiving speaker 21 sequentially. This loudspeaker call system does not require a handset as in the conventional telephone call system, so it has the advantage that it is possible to make a call while working or to realize a natural face-to-face call, It is widely used in teleconferences, videophones, loudspeakers, and the like.

On the other hand, as a drawback of this loudspeaker system, the existence of reverberation is a problem. That is, in FIG. 5A, the sound transmitted from the speaker 15 to the receiving side is received by the microphone 17 as shown by the arrow 22, and is reproduced to the transmitting side via the amplifier 18, the transmission path 13b, the amplifier 19, and the speaker 21. . For the transmitter 10, this phenomenon is a reverberation phenomenon in which the voice uttered by the speaker 10 is reproduced from the speaker 21, and is called an acoustic echo or the like. This reverberation phenomenon causes adverse effects such as trouble in communication and discomfort in a loudspeaker system. Further, the sound reproduced from the speaker 21 is received by the microphone 11 to form a closed loop of the signal. If the gain of the loop is greater than 1, a howling phenomenon occurs and the call cannot be made.

[0004] In order to overcome such a problem of the loudspeaker communication system, a loss control device, an echo canceller, and the like are used. FIG. 5B is a schematic diagram illustrating an example of the loss control device.
Parts common to those in FIG. 5A are given the same numbers. Also,
The amplifier is omitted for simplicity, and only the receiver 16 side in FIG. 5A is shown. The transmission line 13a is connected to the loss control circuit 23.
, And an output signal z (n) of the microphone 17 (n is a parameter representing time), and the transmission / reception state is determined based on their magnitudes. As a method of determining the transmission / reception state, for example, each short-time power Px of the reception signal x (n) and the output signal z (n) is used.
(N) and Pz (n) are calculated and their sizes are compared.
Here, assuming that the echo signal y (n) passing through the arrow 22, that is, the echo path, is smaller than the reception signal x (n), when there is no transmission signal s (n) uttered by the receiver 16, , Px (n)> Pz (n). At this time, it is determined that the receiving state is set. On the other hand, the transmission signal s of a certain level or more
If (n) exists, Px (n) <Pz (n), and
At this time, it is determined to be in the transmission state. When it is determined that the receiving state is in the receiving state, the loss is inserted into the transmitting side loss unit 24 inserted on the output side of the microphone 17. As a result, the speaker 15
The reverberation signal y (n) sneaking around from and received by the microphone 17 is attenuated by the loss unit 24 to reduce the reverberation phenomenon. On the other hand, if it is determined that the transmission state is established, the loss of the loss unit 24 on the transmission side is determined to be 0 (dB), and the loss is inserted into the reception side loss unit 25 inserted in front of the speaker 15. As a result, the transmission signal received by the microphone 17 is transmitted without attenuation. Also, by inserting the loss into the receiving side, the closed loop gain can be kept at 1 or less, and the howling phenomenon can be prevented.

As described above, the loss control device 26
It is possible to reduce the reverberation phenomenon by using
In some cases, the insertion loss is fixed to keep the configuration simple. In this case, a large value (for example, 20 dB) is given as the value of the insertion loss amount assuming all situations. However, if the inserted loss exceeds 10 dB, the beginning or end of the voice is cut off due to a time delay in the transmission / reception determination, which causes a problem that the speech quality deteriorates.

[0006] Therefore, the acoustic coupling in the actual use situation
Adaptive loss control that adaptively controls the loss amount according to the combined amount
A device has been proposed. FIG. 6A shows an adaptive loss control device.
FIG. The same numbers are used for the parts common to FIG. 5B.
Granted. Passing through the loss device 25 to the insertion loss determination circuit 27
Signal L after _Rx (n) and from the microphone 17
Output signal z (n) is input, and based on these levels,
Between the speaker 15 and the microphone 17
The coupling amount G is estimated, and the insertion loss amount is determined according to the estimated amount.
Set. For example, the predicted value G ′ of the acoustic coupling amount is L_Rx
(N) and short-time power PL of z (n)_Rx (n) and P
Using z (n), it can be calculated as follows.

G '= Pz (n) / PL _R x (n) If the predicted coupling amount G' is larger than 1 (0 dB),
The insertion loss determining circuit 27 calculates a loss such that the gain between x (n) and L _T z (n) becomes 1 or less. For example, if G 'is 4 (6 dB), the input signal x
In order to keep the open loop gain between (n) and the transmission signal L _T z (n) at 1 (0 dB) or less, the insertion loss is 0.5
(6 dB in power) or less. Further, when the acoustic coupling amount G changes and G ′ becomes 2 (3 dB), the insertion loss amount is set to 0.7 (3 dB in power) or less. The insertion loss determined in this way is transferred to the loss control circuit 23. The loss control circuit 23 determines the loss
4 and 25 are given insertion loss amounts.

As described above, in the adaptive loss control device, the insertion loss is determined according to the magnitude of the acoustic coupling amount G of the echo path, and the loss is inserted to minimize the reduction in speech quality. It becomes possible to hold down. However, since the estimation of the acoustic coupling amount G is calculated with short-time power, it is difficult to know an accurate value.

As described above, the reverberation phenomenon can be reduced by using the adaptive loss control device.
Under the condition that the acoustic coupling amount G is large, there still remains a problem that the communication quality is deteriorated. In recent years, the echo canceller described below has been introduced as a new technology that does not cause such a problem. FIG. 6B shows an example of a conventional echo canceller 31. In the echo canceller 31, first, the impulse response (echo path transmission characteristic) of the echo path 22 is estimated by the echo path estimating circuit 32, and the estimated value h
(N) 'is transferred to the pseudo echo path 33. Next, in the pseudo echo path 33, a convolution operation of the estimated impulse response h (n) 'and the reception signal x (n) is performed to execute the pseudo echo signal y.
(N) 'is synthesized. Then, in the subtractor 34, the pseudo echo signal y from the output signal z (n) of the microphone 17 is obtained.
(N) 'is subtracted. If the impulse response of the echo path 22 is well estimated, the echo signal y (n) and the pseudo echo signal y (n) 'are almost equal, and the result of the subtraction by the subtractor 34 is as follows. The echo signal y (n) included in the microphone output z (n) is canceled. Note that the echo signal y (n) and the signal s (n) cannot be distinguished by looking only at the output z (n) of the microphone 17, and therefore z (n) may be referred to as an echo signal.

Here, it is necessary for the pseudo echo path 33 to follow the temporal variation of the characteristic h (n) of the echo path 22. Therefore, the echo path estimation circuit 32 estimates the impulse response of the echo path 22 using an adaptive algorithm. This estimating operation is a receiving state, that is, s (n) ≒ 0, and z (n) ≒ y
This is executed when it can be regarded as (n). In the listening state,
The output of the subtractor 34, that is, the error signal e (n) can be regarded as the erasure residual y (n) -y (n) 'of the echo signal y (n). In the following description, this receiving state is assumed.
The adaptive algorithm is composed of the received signal x (n) and the error signal e.
(N) is an algorithm that determines an estimated value h (n) ′ of the impulse response so that the power of e (n) is minimized. Known algorithms include an LMS method, a learning identification method, and an ES method. I have.

Next, a typical adaptive algorithm will be described. When a digital FIR filter is used as the pseudo echo path 33, the impulse response of the filter and the filter coefficient match, so that the estimated value h (n) 'of the echo path impulse response is hereinafter referred to as a filter coefficient. In the adaptive algorithm, the estimation of h (n) 'is performed in a sequential manner. That is, the filter coefficient h (n + 1) ′ at the time n + 1 is calculated by the following equation, which is obtained by modifying the filter coefficient h (n) ′ at the time n.

H (n + 1) ′ = h (n) ′ + αΔ (n) (1) where h (n) ′ = (h1 (n) ′, h2 (n) ′,..., HL (n) ′) ^T : filter coefficient (vector representing the impulse response of the pseudo echo path 33 at time n) Δ (n): vector representing the correction direction of the coefficient α: step size (parameter representing the magnitude of correction of the coefficient: scalar amount) L : Number of taps (number of filter coefficients) ^T : Transposition of vector n: Discretization time Δ (n) varies depending on the algorithm. For example, in the LMS method, Δ (n) = e (n) × (n) (2) In the learning identification method, Δ (n) = e (n) X (n) / (X (n) ^TX (n)) (3) Where e (n): error signal (= y (n) -y (n) ') y (n)' = h (n) ' ^TX (n) X (n) = (x (n), x (n-1),..., x (n-L + 1)) ^T : received signal vector Here, the characteristic of the pseudo echo path 33 is close to the characteristic of the true echo path 22, and the pseudo echo signal y (n) ' Is referred to as having converged when it has become substantially equal to the echo signal y (n).

In various adaptive algorithms, the step size parameter α is usually set to 1. The step size parameter α is an amount that affects the state of convergence. When the value is 1, the convergence speed is maximized. When the value is set to a value smaller than 1, the convergence speed is reduced. Is smaller than one. For this reason, a variable step size parameter type adaptive algorithm has been proposed in which the step size parameter α is set to 1 in the initial state, and the step size parameter α is reduced when the convergence to some extent is achieved. In such an algorithm, it is necessary to correctly determine the convergence state.

Further, the algorithm for estimating the characteristics of the echo path 22 is based on the following.
This is regarded as an error, and the pseudo echo path 33 is set in the wrong direction. Therefore, it is necessary to stop the estimation of the echo path in the double talk state by detecting the presence / absence (double talk) of the reception signal x (n) and the transmission signal s (n) of the near end speaker.

As a method for solving the problem of stopping the adaptive estimation at the time of double talk, an FG / BG (foreground / background) system for performing good echo estimation without explicitly detecting a double talk is proposed. Have been. FIG. 7A shows an echo canceller 36 configured using the FG / BG method, and the same reference numerals are given to portions common to FIG. 6B. In the FG / BG type echo canceller 36, first, the impulse response of the echo path 22 is estimated in the echo path estimation circuit 32, and the estimated value hb (n) 'is transferred to the pseudo echo path 33 on the BG side. Next, in the pseudo echo path 33 on the BG side, a convolution operation of hb (n) 'and the received signal x (n) is executed to synthesize a pseudo echo signal yb (n)'.
Then, the subtractor 34 subtracts the pseudo echo signal yb (n) ′ from the output signal z (n) of the microphone 17.
If the echo path impulse response is well estimated,
The echo signal y (n) and the pseudo echo signal yb (n) 'are almost equal. If the characteristics of the pseudo-echo path 33 on the BG side are close to the characteristics of the true echo path 22, the pseudo-echo path 33 on the BG side
Is transferred to the coefficient of the pseudo echo path 37 on the FG side. Generally, the fact that the pseudo echo path 33 on the BG side approaches the characteristic of the true echo path 22 means that the difference eb (n) between the microphone output signal z (n) and the pseudo echo signal yb (n) ′ and z (n). This is performed by comparing the power with n). Here, the power is a time integration value of the signal, and when dealing with a discretized signal, for example, Px (n) = {x ² (ni) (} is i
= 0 to k-1). Here, k represents the integration time. The transfer of the coefficients is performed as follows. That is, when the input signal x (n) is equal to or more than the set threshold value in the input determination circuit 38, the power of the error signal eb (n) is determined by the power comparison circuit 39 to be smaller than the power of z (n) to some extent. And the error comparison circuit 41
, The power of the BG-side error signal eb (n) is equal to the FG-side error signal ef (n) (the output of the FG-side pseudo echo path 37 and z
(N), the estimated coefficient hb (n) 'is compared with the FG-side pseudo-echo path coefficient hf (n)', and the estimated coefficient hb (n) 'is smaller than that of the actual echo path 22. Since it is considered that the impulse response is simulated better, B
The coefficient hb (n) ′ of the G side pseudo echo path 33 is transferred to the FG side pseudo echo path 37. Coefficient hf of FG side pseudo echo path 37
Since (n) 'is updated only when the above condition is satisfied, the above condition is not satisfied when the BG-side echo path estimation circuit 32 makes an erroneous estimation at the time of double talk.
The G side coefficient is not transferred to the FG side, and good echo cancellation before double talk is maintained. The pseudo echo signal yf (n) 'is subtracted from z (n) by the subtractor 42 from the FG side pseudo echo path 37, and the subtraction output ef (n) is output to the transmission path 13b.

However, in the FG / BG system, even if the characteristics of the background-side pseudo echo path 33 are different from those of the true echo path 22, the transfer determination circuit 43 satisfies the above-described conditions, and an erroneous BG-side pseudo echo path. In some cases, the coefficient of the path 33 is transferred to the FG-side pseudo echo path 37. This occurs especially when the input signal x (n) is a vowel having a sinusoidal waveform, because the power of the error signal e (n) becomes small even if the coefficient of the pseudo echo path is wrong.

The echo canceling device 31 described above
In addition, the FG / BG type echo canceller 36 has a problem that a sufficient amount of echo cancellation cannot be obtained for an initial stage or a change of the echo path 22 during use, and howling occurs. The solution to this problem is to do initial learning or
Alternatively, when the echo canceller does not converge, it is conceivable to insert a loss to the extent that no howling occurs by the loss control device 28. Hereinafter, an example in which the echo canceller and the above-described adaptive loss controller are combined is shown in FIGS. 7B and 8. In FIG. 7B, the echo canceller 3 is used.
An adaptive loss control device 26 is inserted between the transmission line 1 and the transmission lines 13a and 13b. FIG. 8 shows an example in which the adaptive loss control device 26 is inserted between the echo canceller 36 using the FG / BG system and the transmission lines 13a and 13b. In these figures, the same parts as those in FIGS. 6A, 6B and 7A are given the same numbers. In these devices, the adaptive loss control device 26 inserts a loss corresponding to the amount of acoustic coupling between the speaker 15 and the microphone 17 including the echo canceller 31 or 36. For this reason, when the echo canceling amount of the echo canceller 31 or 36 is small in the initial stage,
When the amount of insertion loss of the adaptive loss control device 26 is increased to suppress reverberation and the reverberation signal is removed to some extent by the reverberation canceller 31 or 36, the adaptive loss control device 2
6 can be reduced in insertion loss.

As described above, it can be understood that howling suppression, which is a problem in the echo canceller when the echo canceling amount is small, can be solved by combining the echo canceller and the adaptive loss control device. However, if the echo canceller does not correctly determine how much echo is removed, for example, the insertion loss may be changed to a value equal to or less than the insertion loss required to prevent howling.

[0019]

As described above, in the conventional echo canceller, even when the error signal is smaller than the echo signal, the pseudo echo path may not simulate the true echo path. is there. In such a case, in the adaptive algorithm of the variable step size parameter type, it is determined that convergence has been achieved even if convergence has not occurred, so that the step size parameter is reduced and convergence is slowed down. . Further, in the FG / BG type echo canceller, erroneous transfer such as transferring a non-converged coefficient on the BG side to a coefficient on the FG side occurs. Further, in a device in which the echo canceller and the adaptive loss control device are combined, there is a possibility that the insertion loss may be changed to be less than or equal to the insertion loss necessary for preventing howling. In order to solve such a problem, an object of the present invention is to provide an echo canceller in which a pseudo echo path correctly simulates a true echo path.

[0020]

According to the first aspect of the present invention, the error signal peak hold means controls the error signal e.
Continuously with the peak value of the power of (n) within a predetermined time
Detected and output as an error signal moving peak value until the next peak value is detected , and the echo signal peak hold means
Of the power of the reverberation signal z (n) within the predetermined time.
Peak values are detected continuously until the next peak value is detected.
In output by the echo signal moves the peak value, the convergence judgment unit, the ratio of the echo signal moves the peak value for the error signal moves the peak value, consistent with the echo path of the pseudo echo path is determined You.

[0021] provided invention <br/> adaptive loss control device according to claim 1 According to the invention of claim 2, the loss control amount is above
It performed based on the determination of the serial convergence judgment means. According to the third aspect of the present invention, in the first or second aspect of the present invention, the echo cancellation is performed by the BG / FG system, and the BG / FG system determines that the pseudo-echo path on the BG side matches the echo path well. the output of the convergence determination unit is used.

The above in any one of claims 1 to 3 of the invention According to the invention of claim 4, the step size parameter determines the convergence in the adaptive algorithm of the echo path estimation
It is controlled according to the output of the serial convergence judgment means.

[0023]

FIG. 1 shows an embodiment in which the present invention is applied to a device in which an echo canceller 31 and an adaptive loss control device 26 are combined, and the same parts as those in FIG. 7B are denoted by the same reference numerals. In the present invention, the signals z (n) and e (n) are input to the peak hold circuit 51. Peak hold circuit 51
Is to continuously detect the peak value of each of the signals z (n) and e (n) within a certain time of the sample power, so that it can be called a moving peak hold circuit.
It can also be said that a moving peak hold is detected. In this peak hold circuit, for example, the following calculation is performed.

Phz (n) = max [z (n) ^× z (n), βphz (n−1)] phe (n) = max [e (n) ^× e (n), βphe (n−1) Here, phz (n) and phe (n) are z
(N) and e (n) peak hold values (echo signal
Peak value, error signal movement peak value) , β is an attenuation constant, ma
x [a, b] is a function that compares a and b and outputs the larger value. The decay constant β is a value smaller than 1,
It is desirable to be within the range of 0.9999 ± 0.00005. In this example, the power is calculated using only the square value of one sample.

The output phz of the peak hold circuit 51
(N) and phe (n) are input to the convergence determination circuit 52, and the ratio is calculated as, for example, A (n) = 10log (phz (n) / phe (n). That is, it is determined that the larger the A (n), the better the convergence state.

FIG. 2 is a graph showing the effect of the present invention.
This shows the state of convergence. The horizontal axis is time, the vertical axis
Represents the closeness between the simulated reverberation path 33 and the true reverberation path 22
This value indicates the state of convergence.
You. This is the result of the simulation, the dotted line 54
The difference between the microphone output signal z (n) and the error signal e (n)
The ratio of each power Pz (n), Pe (n) (10 log (P
z (n) / Pe (n))).
Is the value that has been used for However, when calculating the power
If the integration time is long, the integration time
Because there is a time lag from the actual convergence state,
Here, a short value is used. Solid line 55 is the peak hole
Phz (n) and phe (n)
Indicates the value of the bundle. Here, the damping constant β is set to 0.9999.
Specified. The chain line 56 indicates the reverberation path 22 and the pseudo reverberation path.
33 shows a state of true convergence between the two. In other words, Simule
Impulse response h of the reverberation path 22
Since (n) is known, the pseudo echo path 3 for this is
3 indicates the degree of approximation of the impulse response.
g _Tenh_i(N)^Two/ ((H_i(N) -h_i(N) '))
^Two(Σ is from i = 1 to L).

From FIG. 2, it can be seen that in the case of the conventional power, the value indicating the convergence state may be large even when the characteristic of the pseudo echo path 33 is not close to the true echo path 22. On the other hand, when the peak hold value according to the present invention is used, the variance is reduced, which is close to the convergence of the pseudo echo path 33. Therefore, it is better to evaluate using the peak hold value than to evaluate the state of convergence by the conventional power comparison.
It can be seen that the state of true convergence is more accurately shown. If β is made too large, the power peak value is maintained even if each of the signals z (n) and e (n) becomes zero, for example, and the next appearing z (n) and e (n)
Cannot know its state. On the other hand, if β is too small, the state follows the instantaneous fluctuations of z (n) and e (n), that is, approaches the conventional art, and is not preferable. For example, if the sample frequency is 8 kHz, 1 / (1
−β) becomes 10000, that is, β = 0.9999
, So that the influence of the small power portion before and after the vowel is removed.

In the present invention, since the peak hold value is used, the calculated value of the convergence value in the state where the microphone output signal z (n) is a large value is held after that time, and therefore, the output signal z (n) is held. Inaccurate calculation of the convergence value, which occurs when (n) is temporarily small, can be eliminated, and variations are reduced. Although vowels and the like may show fast convergence in spite of not converging, if this does not converge at the beginning of the vowel, the large e (n) at that time is retained, so that accurate It can be calculated as a value close to the value.

The first embodiment in which the present invention is applied to a device in which the echo canceller 31 and the adaptive loss control device 26 are combined.
In the embodiment, as is apparent from FIG. 2, it is possible to calculate a value close to the value that has actually converged. Adaptive loss control device 26
Can reduce the insertion loss as the echo reverberation amount obtained by the reverberation canceller 31 increases, but in the conventional power comparison between z (n) and e (n), In some cases, it may be determined that echo cancellation has been obtained even if echo cancellation has not been performed, and the insertion loss may be reduced more than necessary. On the other hand, in the present invention, since the peak hold ratio (curve 55) calculated by the convergence determination circuit 52 is close to the actually converged value (curve 56), the result is sent to the insertion loss amount determination circuit 27, so that an accurate value can be obtained. The amount of insertion loss can be determined.

In this embodiment, the present invention is applied to a device in which an adaptive loss control device and an echo canceling device are combined.
6 and an FG / BG type echo canceller 36 can be similarly applied. Next, as a second embodiment of the present invention, an example in which the present invention is applied to an echo path estimation adaptive algorithm of an echo canceller is shown in FIG. In FIG. 3, the same parts as those in FIG. 6B are given the same numbers. In various adaptive algorithms such as LMS, a learning identification method, a projection algorithm, and RLS, a step size parameter is set to 1 in order to increase the convergence speed when the convergence is not achieved, and when the convergence is achieved to some extent, the final steady state is set. There has been proposed a variable step size parameter type adaptive algorithm in which the step size parameter is reduced in order to reduce the amount of erasure. In this variable step size parameter type algorithm, depending on the degree of convergence,
In order to make the step size parameter variable, the convergence state must be accurately calculated. As shown in FIG. 2, according to the present invention, the current dB convergence can be calculated close to the actual value. For this reason, in this embodiment, the output of the convergence determination circuit 52 is
2, for example, in FIG. 2, the output of the convergence determination circuit 52, that is, the step size parameter is 1 when the convergence value is 10 dB or less, 0.5 when the convergence value is between 10 and 15 dB, and 0. 0 when the convergence value is 15 dB or more. Set to 1. The step size parameter controlled in this way is used as the step size of the adaptive algorithm in the estimation circuit 32. As can be seen from the curve 55, according to the present invention, the step size can be changed with little variation with time. On the other hand, in the conventional determination of the degree of convergence, since the convergence value with time is different from the actual convergence state (curve 56), the step size parameter is reduced to 0.1 or the like even if the convergence value does not actually converge. Sometimes.

Next, as a third embodiment of the present invention, FG
A case where the present invention is applied to an echo canceller using the / BG system will be described. FIG. 4 shows an embodiment of the present invention, and the same parts as those in FIGS. 7A and 1 are denoted by the same reference numerals. Conventional FG / B
In the G system, as described above, the transfer condition that the pseudo echo path on the BG side is close to the true echo path is set to the error signal eb (n).
Is determined to be smaller than the power of the microphone output signal z (n) by a certain degree or more. On the other hand, when the present invention is applied to the FG / BG system, as described with reference to FIG. 7A, when the input signal x (n) is equal to or larger than the threshold and the power of the BG-side error signal eb (n) is F
In addition to being smaller than the power of the G-side error signal ef (n),
The transfer condition that the pseudo echo path 33 on the BG side is close to the true echo path 22 is determined by the peak hold circuit 51 by the error signal eb.
The determination is made based on whether the peak hold value of (n) has become smaller than the peak hold value of z (n) by a certain degree or more, and the determination is made based on the amount corresponding to the state of the curve 55 in FIG. As described with reference to FIG. 2, the difference between the echo path 22 and the pseudo echo path 33 can be more accurately represented by the peak hold value comparison than by the conventional power comparison. Therefore, when the BG-side pseudo-echo path 33 that occurs during the conventional power comparison is different from the true echo path 22, the BG-side pseudo-echo path 33
Is transferred to the FG-side pseudo echo path 37 is solved by comparison using the peak hold value.

The estimation circuit 32 shown in FIGS.
As described above, each of the step size parameter variable circuits 58 may be provided, and the step size may be changed by the output of the convergence determination circuit 52.

[0033]

As described above, according to the present invention, the evaluation of how close the characteristic of the pseudo echo path is to the true echo path is based on the moving peak hold value of the power of the echo signal (microphone output) z (n). By comparing the moving peak hold value of the power of the error signal e (n) between the reverberation signal and the pseudo reverberation signal, it is possible to perform the comparison as a value closer to the actual value than the comparison with the simple power.

When this is applied to a device combining an echo canceller and an adaptive loss insertion device, it is possible to calculate the extent to which the echo is canceled by the echo canceller with a value close to the actual value, The amount of insertion loss can be accurately changed, which leads to improvement in speech quality. Also,
In the echo path estimation adaptive algorithm in which the step size parameter is made variable in accordance with the state of echo cancellation, the moving peak hold value of e (n) and z
By using the peak hold value of (n), the step size parameter can be changed accurately, and in the stage where the convergence is not converged, which is the purpose of the adaptive algorithm of the variable step size parameter type, the step size is increased in order to increase the convergence speed. When the size parameter is set to 1 and converges to some extent, the effect of reducing the step size parameter to reduce the final steady erase amount can be maximized.

Further, when the present invention is applied to an echo canceller using the FG / BG system, which has been difficult to calculate correctly, an erroneous coefficient on the BG side, which has occurred in the conventional power comparison, is calculated by the FG / BG method. The call is not transferred to the side, and the call quality is improved.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a first embodiment in which the present invention is applied to a device in which an adaptive loss control device and an echo canceling device are combined.

FIG. 2 is a diagram showing a change state of an amount indicating various convergences for explaining an effect of the present invention.

FIG. 3 is a block diagram showing a second embodiment in which the present invention is applied to a variable step size parameter type algorithm.

FIG. 4 is a block diagram showing a third embodiment in which the present invention is applied to an FG / BG type echo canceller.

FIG. 5A is a schematic diagram of a loudspeaker system, and FIG. 5B is a block diagram showing a conventional loss insertion device.

FIG. 6A is a block diagram showing a conventional adaptive loss control device, and FIG. 6B is a block diagram showing a conventional echo canceller.

FIG. 7A is a block diagram showing a conventional FG / BG type echo canceller, and FIG. 7B is a block diagram showing an apparatus combining a conventional adaptive loss control device and an echo canceller.

FIG. 8 is a block diagram showing a device in which a conventional adaptive loss control device and an FG / BG system echo canceller are combined.

────────────────────────────────────────────────── ─── Continuation of front page (72) Inventor Masafumi Tanaka 1-6, Uchisaiwaicho, Chiyoda-ku, Tokyo Nippon Telegraph and Telephone Corporation (56) References JP-A-62-154824 (JP, A) 2-25194 (JP, A) JP-A-2-134030 (JP, A) JP-A-61-187425 (JP, A) WO 93/6679 (WO, A1) (58) Fields investigated (Int. . ⁷ , DB name) H04B 3/00-3/44

Claims (4)

(57) [Claims] And sending signals to claim 1] echo path, with estimation means from the error signal to estimate the transfer Itarutoku of the echo path, it sets the estimated transfer characteristic estimated echo path, above the estimated echo path A reverberation canceling device for generating a pseudo-echo signal through the transmission signal, subtracting the pseudo-echo signal from the reverberation signal passing through the reverberation path by subtracting the pseudo-echo signal, and setting the remainder as the error signal; The peak value of the power of the
An error signal peak hold means for continuously detecting and outputting an error signal as a moving peak value until a next peak value is detected; and a peak value of the power of the echo signal within the predetermined time.
Continuously detected and echoes until the next peak value is detected
Reverberation signal peak hold means for outputting as a signal movement peak value , and error signal movement peaks from the error signal peak hold means.
For over click value, than the echo signal peak hold means
And a convergence determining means for determining the coincidence of the pseudo echo path with the echo path based on the ratio of the echo signal movement peak value of the echo signal . 2. A method according to claim 1, wherein a loss applied to said transmission signal and a loss applied to said error signal are controlled in reverse according to said transmission signal and said echo signal, and a control amount thereof is controlled by said echo of said pseudo echo path. An adaptive loss control means for adaptively controlling according to the coincidence with the road is provided, and the adaptive control of the control amount is performed based on the judgment of the convergence judgment means. 2. The echo canceller according to 1. 3. The transmission signal is passed through a second pseudo echo path to obtain a second pseudo echo signal, and the second pseudo echo signal is subtracted from the echo signal to obtain a second error signal as an output signal.
A BG for detecting a state where the second error signal is larger than the error signal and the pseudo echo path matches well with the echo path, and transfers the characteristics of the pseudo echo path to the second pseudo echo path.
3. An echo canceller according to claim 1, wherein said convergence determining means is used for determining said coincident state in said BG / FG system. 4. The method according to claim 1, wherein a step size parameter for determining a degree of convergence in an adaptive algorithm for echo path estimation in said estimating means is controlled in accordance with an output of said convergence determining means. The described echo canceller.