JP2690519B2 - Echo canceller tap coefficient updating method - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION [Summary] For example, regarding a tap coefficient updating method of an echo canceller used in a telephone communication system, a malfunction of tap coefficient updating occurring at the beginning and the end of a signal is prevented, and at the same time, from a transient state. In order to reduce the residual echo by using the value obtained by normalizing the product sum of the residual echo and the input signal on the receiving side with the input power on the receiving side, the echo canceller is designed to reduce the residual echo. In an echo canceller configured to update the tap coefficient, the beginning and end of the signal are detected using a value based on the short-time average of the receiver input signal and a value based on the receiver input power, At the beginning and the end of the input, when the difference between the short-time average of the input signal on the receiving side and the input power on the receiving side is large, the input power on the receiving side is multiplied by a predetermined value. On the other hand, when the difference is small, on the other hand, the tap coefficient is updated by normalizing using the value based on the short time of the input signal on the receiving side.

[Industrial applications]

The present invention relates to a tap coefficient updating method for an echo canceller used in, for example, a telephone communication system.

[Conventional technology]

In a telephone communication system, a hybrid circuit may cause a signal on the receiving side from the other party to leak into the signal on its own transmitting side in a hybrid circuit and return to the other side as an echo. The echo canceller suppresses such echo, and an example of its configuration is shown in FIG.

The function of the echo canceller is to generate a pseudo echo generated by convolution of the echo impulse propulsion impulse response and the input signal on the receiving side with respect to the signal component that becomes an echo of the input signal on the receiving side and wraps around to the transmitting side. It is erased by subtracting from the input signal.

This method of estimating the impulse response of the echo path is called an adaptive algorithm and is an important factor that determines the performance of the echo canceller. Generally, the "learning identification method", which has a relatively simple algorithm and shows good characteristics, is most often used. This learning identification method is described below.

Let the input signal sequence X _n on the receiving side be X _n = (x _n , x _n-1 , ..., x _{n-N + 1} ) and the estimated impulse response (tap coefficient of echo canceller) H _n be H _n = (H ₀ , h ₁ , ..., h _{N + 1} ). However, n is the sample time and N is the number of taps.

At this time, the tap coefficient at time (n + 1) is updated according to the following update equation (1).

In this update equation (1), ‖X _n ² ‖ is the input power on the receiving side, and this input power ‖X _n ² ‖ is the sum of squares of N data in the tap, so The average value is shown.
Therefore, if all N data have almost the same level value, input is made for all of these N data (x _n , x _n-1 , ... x _{n-N + 1} ). The power ‖X _n ² ‖ can be said to be an appropriate value, and the tap coefficient is updated appropriately.

[Problems to be solved by the invention]

On the other hand, when the N data (x _n , x _n-1 , ... x _{n-N + 1} ) have large level fluctuations and large variations, the input power ‖X _n ² ‖ is the respective data. However, in such a case, the update of the tap coefficient is not successful if the update equation (1) is followed. As an example of such a case where the variation of N data is severe, there is a signal start or end.

For example, in the case of the beginning of signal input, as shown in FIG. 6, in the case of the beginning of signal input from the silent state to the voiced state, data is tapped at time t (k−N + 1). During the period from the start of input until all taps are filled with data at time t (k), the data x _n rises sharply, while its input power ‖X _n ² ‖ slowly follows the fluctuation. In the process of being. Therefore, during this period, the input power ‖X _n ² ‖ is too small for the data x _n in the tap.

Similarly, as shown in FIG. 7, in the case of the end portion of the signal from the voiced state to the silent state, the input power ‖X _n ² ‖ gradually falls with respect to the fall of the data x _n. However, in this case, for the oldest data in the tap, the input power ‖X _n ² ‖ becomes too small and becomes an inappropriate value.

If the input power ‖X _n ² ‖ becomes an inappropriate value for the input data x _n , the tap coefficient is updated according to the update equation (1). Since N pieces of data in the above are normalized by the input power ‖X _n ² ‖, x _n is normalized by a smaller ‖X _n ² ‖ in the above section. As a result, when normalization is performed by a fixed decimal point calculation, overflow may occur and incorrect coefficient updating may be performed.

Therefore, in order to solve such a problem, the applicant has previously proposed the following method.

That is, as shown in FIG. 8, at the beginning of signal input, x _n · k _L is set as a judgment threshold value for level comparison between the data x _n and the input power ‖X _n ² ‖. Where k _L
Is 1 ≧ k _L > 0. Then, when this judgment threshold value x _n · k _L is larger than the input power ‖ X _n ² ‖, that is, when x _n · k _L ＞ ‖ X _n ² ‖, the input power for the data x _n is ‖X _n ² ‖
Is determined to be too small, and instead of ‖X _n ² ‖, the corrected normalized power P ₀ , P ₀ = ‖X _n ² ‖ ・ k _R (where k _R > 1) is used to calculate the power value. Increase and increase H _{n + 1} = H _n + α · e _n · X _n / P ₀ ... (2) Normalize and update the coefficient. In addition, until the data x _n in the transition period comes out of the tap (N
The same process is performed as the protection time for the frame).

Also, as shown in FIG. 9, at the end of the signal, the oldest data x _{n-N + 1} in the tap and the input power ‖X _n ² ‖
And are compared using the decision threshold x _n · k _L , and if x _{n-N + 1} · k _L > ‖X _n ² ‖, the corrected normalized power P ₀ is also used ( 2)
Normalize the expression.

By adopting such a method, it is possible to eliminate the jarring sound of the residual echo, which is caused by the coefficient being erroneously updated due to the overflow of calculation.

However, in this method, instead of the input power ‖X _n ² ‖ at the beginning and end of the signal, a larger correction normalized power P ₀ = ‖X _n ² ‖ ・ k _R is used, and α e _n
-Since the value of X _n / P ₀ is suppressed to a small value to prevent overflow, the problem is that the tap coefficient update speed is somewhat slower. If the data x _n fluctuates drastically, the decision threshold x _n · k _L is not stable, so either ‖ X _n ² ‖ or ‖ X _n ² ‖ · k _R is used as the corrected normalized power P _0. Since or is not stable, it prevents the update coefficient from converging smoothly. Especially at the beginning of signal input, it is necessary to increase the update speed as much as possible to accelerate convergence.

Therefore, an object of the present invention is to prevent erroneous operation of tap coefficient update occurring at the beginning and end of signal input, and at that time, perform coefficient update as soon as possible to converge from a transient state.

[Means for solving the problem]

 FIG. 1 is a diagram illustrating the principle of the present invention.

The tap coefficient updating method of the echo canceller according to the present invention uses the value obtained by normalizing the sum of products of the residual echo e _n and the reception side input signal x _n by the reception side input power ‖X _n ² ‖.
tap coefficient H _n of the echo canceller to reduce the _n
In an echo canceller that is configured to update the input signal using a value based on the short-time average P _x of the input signal x _n on the receiving side and a value ‖ X _n ² ‖ ・ k _L based on the input power on the receiving side. The beginning and the end are detected, and at the beginning and the end of this signal, when the difference between the short-time average P _x of the receiving side input signal x _n and the receiving side input power ‖ X _n ² ‖ is large, the receiving side input power is The value multiplied by a predetermined value ‖ X _n ² ‖ ・ k _R , on the other hand, when this difference is small, normalization is performed using the value based on the short-time average P _x of the input signal on the receiving side to update the tap coefficient. Is composed of.

[Action]

For example, at the beginning of signal input, as shown in Fig. 1, the short-time average P _x of the input signal x _n on the receiving side is compared in magnitude, and P _x · k _L > ‖ X _n ² ‖ (however, , 1 ≧ k _L > 0), it is determined that the signal starts or ends.

At the beginning or the end of this signal, the short-term average P _x and the input power on the receiving side ‖ X _n ² ‖ are compared in magnitude.
A value based on the short-term average P _x and the input power of the receiving side ‖ X _n ² ‖ ・ k
_R and are used, where k _R > 1.

Then, for example, if ‖X _n ² ‖ ・ k _R ≤P _x , it is determined that the difference between this short-time average P _x and the input power ‖X _n ² ‖ of the receiving side is large, and the value ‖X _n ² ‖ ・The coefficient update is normalized using k _R.

On the other hand, for example, if ‖X _n ² ‖ · k _R > P _x , it is determined that the difference is small, and normalization is performed using the short-time average P _x .
As a result, the normalization of the tap coefficient updating formula can be always performed properly.

〔Example〕

 Hereinafter, embodiments of the present invention will be described with reference to the drawings.

FIG. 2 is a block diagram showing an echo canceller using a tap coefficient updating method as an embodiment of the present invention. In the figure, the received signal power calculation circuit 1 for obtaining the received signal power ‖X _n ² ‖ received signal x _n, short-time average calculating circuit 2 to calculate the short-term average P _x of the received signal x _n, 3 is received Signal power ‖ X _n ²
‖ And the short-time average P _x are compared to determine the state such as the start or end of signal input, 4 is a correction that selects the correction normalized power P ₀ based on the determination result of the state determination circuit 3. Normalized power selection circuit, 5 is a received signal series circuit for extracting received signal series X _n , 6 is a coefficient based on received signal power ‖ X _n ² ‖, corrected normalized power P ₀ , received signal series X _n , etc. A coefficient correction circuit for correction, 7 is an echo path estimation impulse response circuit for obtaining an echo path estimation impulse response based on the coefficient correction result, and 8 is a pseudo echo generation for generating a pseudo echo _n based on a received signal sequence and an echo path estimation impulse response The circuit 9 is a subtraction circuit for suppressing the echo by subtracting the pseudo echo _n from the input signal on the transmission side. Further, 10 is a comparison circuit, 11 is a comparison judgment circuit, 12 is a power comparison circuit, 13 is an echo power circuit, and 14 is a hybrid circuit.

 The operation of this echo canceller will be described below.

In this embodiment, a short-term average P _x of x _n is used in place of the conventional decision threshold x _n · k _L to detect the start or end of the signal, and the detection We are trying to stabilize. That is, even when the data x _n fluctuates greatly, the short-term average P _x is moderated in degree of fluctuation, so that the corrected normalized power P ₀ is ‖X _n ² ‖ and ‖X _n ² ‖.
・ Instability is less likely to occur when k _R is used.

This short-time average P _x is calculated by the short-time average calculation circuit 2, and this is compared with the received signal power ‖X _n ² ‖ in the state determination circuit 3 to determine the state of the beginning or end of signal input.

FIG. 3 is a diagram for explaining this detection state at the beginning of signal input. Here, when P _x · k _L > ‖X _n ² ‖, it is determined that the signal has started.

Similarly, FIG. 4 is a diagram for explaining this detection state at the end of the signal. Here, the short-time average of x _{n-N + 1} is P
_{Let x} ′ be the end of the signal if P _x ′ · k _L > ‖X _n ² ‖.

To prevent further increases in the Figure 8 and 9 in protection time during the correction period, as shown in FIG unnecessarily corrected normalized power P _0, the correction normalized power P ₀ in this embodiment It is determined by the following algorithm to prevent the tap coefficient modification amount from becoming too small.

First, at the beginning of signal input in Fig. 3, that is, during the period of P _x · k _L > ‖X _n ² ‖, if ‖X _n ² ‖ ・ k _R ≤P _x , the corrected normalized power P ₀ is set. , P ₀ = ‖X _n ² ‖ · k _R is used to increase the value of the corrected normalized power P ₀ to prevent overflow and the like. Then, if the protection time after that, that is, ‖X _n ² ‖ ・ k _R > P _x , becomes the corrected normalized power P ₀ and the short-time average.
Use P _x . That is, the corrected normalized power P ₀ is prevented from becoming too large by setting P ₀ = P _x .

Next, at the end of the signal in Fig. 4, that is, during the period of P _x ′ ・ k _L > ‖X _n ² ‖, if ‖X _n ² ‖ ・ k _R ≤P _x ′, the corrected normalized power P _{As 0} , P ₀ = ‖X _n ² ‖ · k _R is used to increase the value of the corrected normalized power P ₀ to prevent overflow and the like. Then, if the protection time after that, that is, ‖X _n ² ‖ ・ k _R > P _x ′, becomes data x as corrected normalized power P _0.
It uses the short-term average P _x ′ of _{n-N + 1} . That is, the corrected normalized power P ₀ is prevented from becoming too large by setting P ₀ = P _x ′.

The state determination circuit 3 performs the above-described state determination, and based on the determination result, the correction / normalization power selection circuit 4 sets ‖X _n
² ‖, ‖X _n ² ‖ ・ k _R , P _x , or P _x ′,
It is sent to the coefficient correction circuit 6.

The echo path estimation impulse response circuit 7 calculates an estimated impulse response based on the correction result of the coefficient correction circuit 6,
The pseudo echo generating circuit 8 generates a pseudo echo based on this estimated impulse response.

〔The invention's effect〕

According to the present invention, by using the short-time average of the received signal for the state determination, the normalized power at the beginning and the end of the signal is stabilized and the update coefficient converges smoothly. Further, since the short-time average is used as the correction normalization power during the so-called protection time, the normalization power is not excessively corrected, so that the convergence time is further shortened.

[Brief description of the drawings]

FIG. 1 is an explanatory view of the principle of the present invention, FIG. 2 is a block diagram showing an echo canceller by a tap coefficient updating method as one embodiment of the present invention, and FIG. 3 is an operation explanation of the embodiment apparatus at the beginning of signal input. FIG. 4 is a diagram for explaining the operation of the embodiment apparatus at the end of a signal, FIG. 5 is a block diagram showing a configuration example of a conventional echo canceller, and FIG. 6 is a relationship between an input signal and an input power at the beginning of signal input. FIG. 7 is a diagram for explaining the relationship between the input signal and the input power at the end of the signal, and FIG. 8 is a diagram for explaining the operation at the beginning of signal input according to the previously proposed method, and FIG. The figure is an illustration of operation at the end of a signal according to the previously proposed method. In the figure, 1 ... Received signal power detection circuit 2 ... Short-time average calculation circuit 3 ... State determination circuit 4 ... Correction normalization power selection circuit 5 ... Received signal series circuit 6 ... Integer correction circuit 7 ... Echo path estimation impulse response circuit 8 …… Pseudo echo generation circuit 9 …… Subtraction circuit 10 …… Comparison judgment circuit 11 …… Double talk detection circuit 12 …… Power comparison circuit 13 …… Echo power detection circuit 14 …… Hybrid circuit

Claims (1)

(57) [Claims] 1. A residual echo (e _n ) and a receiving side input signal (x _n )
An echo canceller configured to update the tap coefficient (H _n ) of the echo canceller so as to reduce the residual echo using a value obtained by normalizing the sum of products of the received input power (‖X _n ² ‖) In, the beginning and end of the signal are detected using the value based on the short-time average (P _x ) of the input signal on the receiving side and the value based on the input power on the receiving side. When the difference between the short-time average of the input signal on the receiving side and the input power on the receiving side is large, a value obtained by multiplying the input power on the receiving side by a predetermined value is used. On the other hand, when this difference is small, it is based on the short-time average of the input signal on the receiving side. A tap coefficient update method for an echo canceller configured to perform normalization using values to update tap coefficients.