JPH0720072B2 - Eco-Cancer - Google Patents

Description

TECHNICAL FIELD The present invention relates to an echo canceller used in a communication line or a room sound field control device.

(Prior Art) Generally, an echo canceller is a device used in a long-distance telephone line using a communication satellite and a submarine cable to cancel an echo generated by impedance mismatch of a 2-wire 4-wire converter. , A variable coefficient filter for generating a pseudo echo, and a subtraction circuit. The outline of the operation of the echo canceller will be briefly described below.

FIG. 3 is a block diagram of a conventional echo canceller, in which a variable coefficient filter, a subtraction circuit, a correction amount calculation circuit, a reception signal input terminal, a reception signal output terminal, a transmission signal input terminal, Transmission signal output terminal,
Is a reception signal input register, is a pseudo impulse response register, and is a sum-of-products circuit (convolution circuit).

Now, at time t = jT (T is the sampling frequency), if the echo signal at the transmission signal input terminal is Yj and the pseudo echo created in the echo canceller is Yj ', this pseudo echo Yj' is converted from the echo signal Yj. Deduction (residual echo ej),
The echo is canceled by controlling the electric power of the residual echo to the minimum. Therefore, ej = Yj-Yj ', and the pseudo echo Yj' is represented by a transversal filter output with the received input signal Xj as an input, where n is the number of taps.

Here, hj (k) is a filter coefficient and corresponds to the impulse response of the echo path.

Also, in the echo canceller, when there is a voice signal in the reception signal input and there is no voice signal in the transmission signal input and only the echo is included, the echo canceling operation is performed as the adaptive operation state. Specifically, the filter coefficient hj (k) is successively corrected and updated by the sample residual echo ej and the received input signal Xj. Generally, the least mean square (LMS) algorithm is used for this adaptive operation algorithm. Among them, the learning identification method (Noda: "Effects of noise signal and parameter fluctuation in learning identification method" Measurement and Control Vol.8, No. .
5,) is often used. The internal calculation operation of the correction amount calculation circuit adopting this control method is performed as described below.

When the received input signal vector is Xj, the coefficient vector of the variable coefficient filter is hj, the amount of modification when updating is Δhj, and the loop gain is α, the coefficient modification is hj ₊₁ = hj + Δhj Δhj = α ・ [ej / (ej here ^{Xj T Xj)] · Xj,} Xj = (Xj, Xj -1, Xj - n +1) T hj = (hj (1), hj (2), ..., hj (n)) T Δhj = ( Δhj (1), Δhj (2), ..., Δhj (n)) ^T. However, it represents the transpose of the ^T vector.

The variable coefficient filter is composed of a reception signal input register, a pseudo impulse response register, and a product-sum circuit, and the variable coefficient hj is stored in the content of the pseudo impulse response register.

(Problems to be Solved by the Invention) In this learning identification method, the echo cancellation characteristic depends on the loop gain and the near-end noise. That is, the echo cancellation amount ER
LE is expressed by the following equation.

ERLE = S / N + 10log (1 / α-1) (dB) where S / N is the received input signal to near-end noise power ratio.
From this, it can be seen that in order to increase the amount of echo cancellation, α should be made as small as possible. On the other hand, the convergence time required for the echo cancellation amount to reach the steady value is shorter as α is larger, and the convergence time is minimum when α = 1. Therefore, the learning identification method cannot simultaneously improve the echo cancellation amount and the convergence time, and as a method for overcoming this, a variable loop gain method in which the loop gain α is variable according to the output error signal is proposed. ing. (Japanese Patent Application No. 59-29612) "Echo canceller") However, in the present method, for a color signal having an extremely strong correlation such as an audio signal,
The convergence time becomes extremely long compared to the case of white noise,
For the purpose of canceling the echo due to the reverberation time in the room, the number of taps in the filter increases, so the convergence time further increases.

A second simulation example of the convergence characteristic for the correlation signal of the conventional variable loop gain method is shown. In the figure, the first-order correlation function of the input signal is set to 0, 0.8, and 0.9.
From this, it is apparent that the conventional method has a drawback in that the convergence characteristic deteriorates as the correlation between the input signals becomes stronger.

An object of the present invention is to provide an echo canceller which has a small amount of hardware and can improve the convergence time for a color signal having a very strong correlation such as voice.

(Structure and Operation of the Invention) Generally, the echo cancellation characteristic of the echo canceller largely depends on the statistical property of the input signal and the coefficient updating algorithm of the variable coefficient filter. Since the input signal is intended for the voice signal now, it is desirable to make use of its statistical properties in the algorithm. In speech, the first-order autocorrelation coefficient in the speech section is considerably large at 0.80 to 0.95. Therefore, if this correlation-removed signal is used for coefficient updating,
It can be expected that the echo cancellation characteristics for audio signals will be improved.

FIG. 1 shows a block configuration of the present invention.

First, the received input signal sequence Xj and its primary sending signal sequence Xj _-1
From the following equation, the primary transmission correlation coefficient Rj is calculated. That is, Next, the received input signal Xj is decorrelated using this correlation coefficient Rj. That is, the coefficient correction formula based on Zj = Xj-RjXj- ₁ decorrelation signal sequence Zj is as follows.

ej = Yj-Yj ', and the pseudo echo Yj' is represented by a transversal filter output with the received input signal Xj as input, and the number of taps is n.
Then, the following formula is obtained.

Here, hj (k) is a filter coefficient and corresponds to the impulse response of the echo path.

When the received input signal vector is represented by Xj and the coefficient vector of the variable coefficient filter is represented by hj, Yj ′ is the inner product of the vectors.

Yj '= hj ^T · Xj where _{Xj = (Xj, Xj -1,} Xj - n +1) T hj = (hj (1), hj (2), ..., hj (n)) T The variable coefficient filter The correction amount when updating the coefficient of Δhj,
When the loop gain is α, the correction of the coefficient is hj ₊₁ = hj + Δhj Δhj = α · ej · Zj α = F (ej) where Zj = (Zj, Zj _-1 , ..., Zj _- n ₊₁ ) ^{T Δhj = (Δhj (1)} , Δh (2) j, ..., Δhj (n)) T ej = (ej, ej -1, ..., ej - p), p is given by a non-negative integer.

Further, α is given as a function of ej or a low-pass or band-pass filter output of ej. For example, α = αk Ek ₋₁ <‖ej‖ <Ek (k = 1, ..., m) where ‖ej‖ = ( ej ^T j) ^1/2 0 <α ₁ = αmin <... <αk _-1 <αk <αk ₊₁ <... <αm
= Αmax 0 <E ₁ <... <Ek _-1 <Ek <Ek ₊₁ <... <Em _-1 , it can be easily realized by using an absolute value circuit, a smoothing filter, a ROM and the like. FIG. 4 shows the relationship of the above equation and its circuit configuration example. In the figure, after the error signal ej is input to the absolute value circuit, the smoothing filter output E _K is converted into the ROM address by the decoder, and α can be generated by outputting α from the ROM as shown in FIG. . The smoothing filter and decoder can be configured by an accumulator or the like. Further, in FIG. 10A, an example in which α has a stepwise characteristic with respect to Ek is shown, but it is also possible to make a larger or more continuous α correspond to a large value of ‖ej∥. In the present embodiment, an example in which α depends only on ej is shown, but it is possible to use a variable that is normalized by using Rin's ‖Xj‖ and ‖Zj‖.

Further, in the calculation of the aforementioned Derutahj, by normalizing with ‖Xj‖ ² and ‖Zj‖ ² or (Xj ^T Zj), it can facilitate the setting of the loop gain alpha. For example, when Δhj = α · [ej / (Zj ^T Xj)] · Zj, αmax ≦ 1.

Next, FIG. 5 shows an embodiment in which the polar correlation is used to calculate the primary correlation coefficient Rj. Here, a polar series SXj having ± 1 is created corresponding to the sign (polarity) of the received input signal series Xj,
The primary correlation coefficient Rj is calculated. That is, However, Since this expression is a value obtained by counting only the number of polarities and dividing the value by n, it can be realized by simple hardware without the need for a complicated multiply-add circuit.

FIG. 6 shows an embodiment of the coefficient correction circuit according to the present invention. ......
X register, ... exclusive NOR gate ,, 26 ... Subtraction circuit, ... Division circuit ,,
,, 22,23 …… Multiplication circuit, …… Z register, 21 ……
Inverse circuit or ROM, 24,27,28 …… Accumulator, 25
... It is the H register.

To operate this, input to the X register that stores n + 1 sampled received signals, and exclude only the polarity (sign) bits (polarity output +,-corresponds to theory 0.1) of the adjacent samples. Sive Noah Gate and n
Input the polarity (sign) bits of adjacent samples with sample delay to the exclusive NOR gate,
Further, the difference between the outputs is calculated by a subtraction circuit, and the accumulator 28 accumulates the difference to output the sum of products. By dividing the counted output by the number of taps n in the division circuit, the polarity correlation count can be obtained.

On the other hand, a predictive reception signal is obtained by a division circuit that multiplies each sample value of the X register by the above-described polarity correlation coefficient, and further, a 1
The sample-delayed received signal is subtracted, and the result is stored in the Z register.

After obtaining the contents of the X and Z registers,
Each (Xj, Zj), (Xj -, Z J-1), ... (Xj - n +1, Zj - n +1) by the multiplication circuit and for obtaining a set of sum of products, XjZj and
Xj _- nzj _- n is calculated, the difference is calculated by the subtraction circuit 26, this is accumulated by the accumulator 27, the product sum is output, and the calculation circuit for calculating the reciprocal or a table of its reciprocal is stored. Calculate the reciprocal by using ROM21.

The product of this reciprocal and the residual signal ej _-1 is calculated by the multiplication circuit,
The result of this operation and the contents of the Z register are multiplied by the multiplication circuit 22, and the result of this multiplication is multiplied by the contents of the Z register by the multiplication circuit 23, and the result is accumulated by the accumulator 24,
This content is stored in the H register 25.

Since the X register and the H register 25 are the same as the reception signal register and the pseudo impulse response register for obtaining the pseudo echo Y'in FIG. 1, they can be shared as hardware.

Furthermore, the configuration of FIG. 6 has a coefficient processing in the counter,
Also, it is possible to handle the sum of products of the X register and Z register as a partial sum.
If there are many samples to be stored, it is possible to divide them into multiple groups to make a cascade connection configuration, which can be realized by a method similar to the cascade connection method described in JP-B-57-84633 "Echo canceller". .

According to the simulation using the present invention,
Shown in the figure. In FIG. 7, conditions such as an input signal are the same as those in FIG. From the comparison with FIG. 2, the characteristic of the present invention is that for a white signal having no correlation (in the case of r = 0, the same echo cancellation characteristic as the conventional configuration is shown, but a signal with a strong correlation r = 0.8 or 0.9). Even with the configuration of the present invention, however, it is understood that the deterioration of the convergence characteristic is small, and the convergence time can be shortened for a chromatic signal having a strong correlation as compared with the conventional variable loop gain method.

(Effect of the invention) As described above, by using the variable loop gain method and the decorrelation signal calculated from the first-order autocorrelation coefficient or the polar correlation coefficient of the input signal, it is possible to add a small amount of hardware. It has an effect that the convergence time can be significantly shortened as compared with the echo canceller having the configuration.

[Brief description of drawings]

FIG. 1 is a block diagram of an echo canceller of the present invention, FIG. 2 is an echo canceling characteristic by a conventional algorithm, FIG. 3 is a block diagram of a conventional echo canceller, FIG. 4 is a loop gain α generation circuit, FIG. 5 is a block diagram of an embodiment of the present invention, FIG. 6 is an embodiment of the coefficient correction circuit of the present invention, and FIG. 7 is a convergence characteristic of the embodiment of the present invention. ...... Variable coefficient filter ...... Addition / subtraction circuit ...... Correction amount calculation circuit ...... Reception signal input terminal ...... Reception signal output terminal ...... Sending signal input terminal ...... Sending signal output terminal ...... Received signal input register, ... pseudo impulse response register, ... sum of products circuit, ... X register, ... exclusive NOR gate,
,, 26 …… Subtraction circuit, …… Division circuit ,,,,
22, 23 ... Multiplier circuit, ... Z register, 21 ... Inverse circuit or ROM, 24, 27, 28 ... Accumulator, 25 ... H register.

Claims (2)

[Claims] 1. A transmission channel (Sin-Sout) and a reception channel (Rin-Rou)
t) and a variable coefficient filter that receives the receiving channel signal as an input, a first register that stores consecutive (N + 1) sample data of the receiving channel signal, and an echo path from the receiving channel output end. A transmission signal including the echo component of the reception signal that circulates to the transmission input terminal as the first input, and the output of the variable coefficient filter as the second input, and a subtractor for each coefficient of the variable coefficient filter. And a coefficient updating circuit for updating the coefficient by adding the correction amount for each sample, and using the coefficient updating circuit for the coefficient of the variable coefficient filter so that the echo component does not appear at the transmission output terminal. In an echo canceller that corrects the error, a first multiplier that receives the error signal output from the subtractor as a first input, and a loop gain generation circuit that generates and outputs a loop gain using the error signal as an input. , A second multiplier which calculates the product of the output of the first multiplier and the loop gain and is input to the coefficient updating circuit, and a correlation coefficient which calculates a correlation function between the reception signal and its first-order delayed signal A calculation circuit, a decorrelation circuit for subtracting a signal obtained by multiplying the primary delay signal by the correlation coefficient from the received signal, and a second register for storing a primary prediction residual signal which is an output of the decorrelation circuit. The echo canceller having the first prediction residual signal as a second input of the first multiplier, and updating the coefficient. 2. The correlation coefficient is calculated by using +1 for a positive signal and -1 for a negative signal corresponding to the polarities of the two signals. The echo canceller according to item 1.