JPS62122339A - Echo canceller - Google Patents

Abstract

PURPOSE:To obtain an echo canceller having a high tracking performance to a change in an echo path even with the large number of taps by using an orthogonal transformation such as Fourier transformation effectively. CONSTITUTION:A reception signal x(n) is subject to Fourier transformation by a DFT circuit 110, a signal X(i,N) is obtained, a weight circuit 210 multiplies a weight H(i,n) with the signal X(i,n) to form a Fourier transformation Z(i,n) of a pseudo echo signal. A signal y(n) including a reflection signal being a reception signal is subject to Fourier transformation circuit 120 into a signal Y(i,n), a subtraction circuit 400 subtracts the signal Z(i,n) from the signal Y(i,n) to form a residual signal E(i,n), a correction circuit 220 corrects the weight H(i,n) and the residual signal E(i,n) is transformed into a signal e(n) of a time region by an IDFT circuit 130. A two-way talking detection circuit 300 references a ratio G (i,n) of the amplitudes (absolute value) of the signals X(i,n) and Y(i,n), number of points where the reflection signal Y(i,n) at the transmission side is larger is counted and it is discriminated when the count exceeds a certain value to output a control signal C.

Description

DETAILED DESCRIPTION OF THE INVENTION (Field of Industrial Application) The present invention relates to an improvement in an echo canceller that cancels reflected signals and reverberation signals due to acoustic coupling in two-wire and four-wire conversion.

(Prior Art) Echo cancellers are used to eliminate echoes in long-distance lines and acoustic feedback signals caused by coupling between microphones and speakers in teleconferences and the like. The echo canceller is inserted into the tube where unnecessary reflections occur. In other words, unnecessary signals are removed by measuring the input and output of the system where reflection occurs, estimating the characteristics of that system, creating a reflected signal based on that, and subtracting it from the actual reflected signal. It is something to do. As a conventional example of an echo canceller, a transversal filter has been used as described in Japanese Patent Application No. 56-101938 (Reference 1).

FIG. 3 is a diagram showing an example of a conventional echo canceller. In FIG. 3 the reflection occurs in the echo path 5. More specifically, in long-distance telephone lines, reflections occur in 2-wire and 4-wire conversion circuits, and in teleconferences, reflections occur due to looping from speakers to microphones.

Here, we will continue to explain mainly about telephone lines.

The Rin signal X sent from the far end is converted into a digital signal x(n) by the AD converter 11. Signal x(
n) is stored in the X memory 20. The estimated value h(i) of the impulse response representing the characteristics of the echo path 5 is stored in the H memory 3o. The product-sum circuit 40 calculates a pseudo echo signal z(n) from the contents of the H memory 3o and the X memory 20 according to the following equation.

z(n)=ΣX (n −i)ice h(i) where N is the number of taps.

The received signal from the near end, that is, the signal Sin that has passed through the echo path 5 is converted into a digital signal y(n
) is converted to By subtracting the pseudo echo signal obtained above from this y(n), a residual signal e(n) is obtained. This residual signal e(n) does not include a reflected signal, and after being converted into an analog signal by the DA converter 14, it is sent to the far end. Based on the gradient end method, the correction circuit 6o calculates the contents of the H memory 30 using the residual signal e(n) and the received signal x(ni) according to the following equation.

h(i) = h(i) +g-x(n-i)・e(
n) where g is the modified gain and i is from 0 to N-1
The value is up to. = in the above equation means replacing the left side with the right side, and the correction circuit 6o and H memory 3
Replacement with 0 is performed.

(Problems to be solved by the invention) The above is an overview of the operation of the conventional echo canceller.
If we try to operate the correction algorithm described above stably, as the number of taps N increases, the correction speed becomes slower in inverse proportion to N. Therefore, when N is large, the correction speed slows down. This has had the problem of poor followability to changes.

For this reason, it has been proposed to use a high-speed algorithm such as a Kalman filter as a correction algorithm, but this has the drawback that the amount of calculation increases significantly as the number of taps increases. There is also a method of dividing the band as in the above-mentioned document 1, but there is a limit to the improvement in correction speed.

An object of the present invention is to provide an echo canceller that effectively uses orthogonal transformation so that the correction speed does not decrease even when the number of taps is large, that is, the echo canceller has a high ability to follow changes in the echo path.

(Means for Solving the Problems) According to the present invention, a first conversion circuit that orthogonally transforms a received signal from a far end, a second conversion circuit that orthogonally transforms a received signal from a near end, a weighting circuit that multiplies the output of the first conversion circuit by a weight; a subtraction circuit that subtracts the output of the weighting circuit from the output of the second conversion circuit; an adaptive circuit for modifying weights; a circuit for inverse orthogonal transformation of the output of the subtraction circuit;
a two-way call detector that monitors each element of the output of the first conversion circuit and each element of the output of the second conversion circuit in correspondence, detects a two-way call, and controls the adaptive circuit. You can get an echo canceller.

(Operation) In the present invention, a signal is orthogonally transformed and the result is used to perform echo cancellation. Since the orthogonally transformed signals are mutually independent or nearly independent, the orthogonalized results can be easily processed independently. There are various types of orthogonal transformation, but here we will explain a method using Fourier transformation. The received signal x(n) from the far end and the echo signal y(n) which is the received signal from the near end
When the Fourier transform is applied to the signals X(xt n)+Y(i, n), frequency domain signals X(xt n)+Y(i, n) are obtained according to the following equations.

X(i, n)=Σx(n-N+1+k) zero WikY(
i, n)=Σy(n-N+1+k)tcWik, where i is the i-th frequency, n is the sampling time, Wik
=exp(-j2rrik/N). Convolution on the time axis can be performed by the product of the frequency domain signals obtained in this way.

In other words, H which is the Fourier transform of the impulse response h(i)
By multiplying (i, n) by X(i, n), Z corresponding to the Fourier transform of the pseudo echo signal z(n) is obtained.
(i, n) is obtained.

Z (i, n) = X (i, n) H (i, n) However, the product here is a complex multiplication, and this complex multiplication is performed for each i (0≦i≦N-1). It will be done. Even if the result obtained in this way is subtracted from the actual echo signal in the frequency domain, the echo component can be eliminated. By inverse Fourier transforming this residual, the signal e(n
) is obtained. In other words, in this respect, the operation is the same as that of a conventional echo canceller on the time axis.

In order to estimate the characteristics of the echo path, H(i,n) must be modified. An example of the algorithm is shown below.

H(i, n+ 1)=H(i, n)−gXc(i,
n) E(i, n) where 0≦i≦N-1, g is the modified gain, and Xc(i, n) is X(i.

Indicates the conjugate complex number of n).

Also, if we assume that the echo path does not increase the signal, then X(
i, n) should always have a larger amplitude than Y(i, n) even for i. Therefore, if the reflected signal becomes larger at a certain frequency, it can be determined that some kind of signal has entered the echo path. This condition is commonly referred to as two-way communication, and the estimation of the characteristics of the echo path becomes unsuccessful. Therefore, at this time, the correction using the above formula must be stopped. In conventional embodiments, two-way communication was detected based on the power ratio between the transmitting side and the receiving side, so the detection sensitivity was not very good. Incoming calls can be detected with high sensitivity.

(Example) FIG. 1 is a diagram showing an example according to the present invention. Figure 2 (
a), (b), (c), (d), (e), (
f) is a diagram showing waveforms of various parts in this example. However, in this embodiment, all input and output signals are digital signals.

That is, it is assumed that an AD converter and a DA converter are located outside of FIG.

The received signal x(n) from the far end is Fourier transformed by the DFT circuit 110 to obtain X(i, N). This situation is as shown in FIG. 2(a). However, in the figure, it is expressed as the absolute value (amplitude) of X(i, N). X
(i, n) is the weighting circuit 210 and the two-way conversation detector 3
00 is input. In the weighting circuit 210, the signal
(i, n) is multiplied by the weight H(i, n) in Figure 3 (d
) Create the Fourier transform Z(i, n) of the pseudo-echo signal.

y(n) including the reflected signal which is the received signal from the near end is D
It is Fourier transformed by the FT circuit 120 and becomes Y(i,n) (FIG. 3(b)). The subtraction circuit 400 subtracts Z(i, n) from Y(i, n) to produce a residual signal E(i, n) (FIG. 3(e)). The modification circuit 220 uses weights H(i, n
) will be corrected. The residual signal E(i,n) is converted into a time domain signal e(n) by the IDFT circuit 130 (second
Figure (f)).

The above processing is performed according to the formulas already described. Note that in normal DFT, N pieces of data (samples)
When , N results are obtained, which are also N
Processing is performed every N data, such as returning to N data. However, in the present invention, since adaptive processing is performed for each piece of data, it is necessary to perform N-dimensional DFT every time one piece of data is input. However, this results in unnecessary calculations. In order to perform such calculations efficiently, the document 2 titled "Transmission channel adaptive equivalent S using frequency sampling filter" by Mr. Fujii et al. (Transactions of the Institute of Electronics and Communication Engineers, Vol. 'J65-B, No. ,
1982) may be used. The method of Document 2 is also advantageous in terms of device implementation. On the other hand, when performing IDFT, only one point,
All you need to do is find e(n). ” The two-way conversation detection circuit 300 detects X(i, n) and Y(i
, n), the ratio of the amplitudes (absolute values) of each of them G(i, n)
(Fig. 2 (C)), the reflected signal Y on the transmitting side
The number of points where (i, n) is larger is counted, and if it is greater than a certain value, it is determined that a two-way call is being made and a control signal C is output. The correction circuit 220 performs control according to the control signal C to determine whether or not to perform a correction operation.

Note that although Fourier transform is used as the orthogonal transform in this embodiment, other transforms may be used.

(Effects of the Invention) As described above, in the present invention, since the characteristics representing the echo path are expressed in an orthogonally transformed area,
Correction can be performed independently for each transformed coefficient, and the ability to follow changes in the echo path is greatly improved. Furthermore, since bidirectional communication is detected based on the results of orthogonal transformation, the detection ability is greatly improved.

[Brief explanation of drawings]

FIG. 1 is a block diagram showing an embodiment of an echo canceller according to the present invention, and FIGS. 2(a), (b), (c),
(d), (e), and (f) are diagrams showing waveforms of various parts in an embodiment according to the present invention, and FIG. 3 is a diagram showing an embodiment of a conventional echo canceller. In the figure, 1.2.3.4 is a terminal, 5 is a reverberation path, and 11
゜ 13 is an AD converter, 12.14 is a DA converter, 20 is an X memory, 30 is an H memory, 40 is a product-sum circuit, 50 is a subtraction circuit, 60 is a correction circuit, 110. IDFT circuit, 210 is a weight circuit, 2
20 is a correction circuit, 300 is a two-way conversation detector, and 400 is a subtracter. 1, :]... Teen p! ! 1st person FX 'W',...
−ichi〆／

Claims (1)

[Claims] a first conversion circuit that orthogonally transforms the received signal from the far end;
a second conversion circuit that orthogonally transforms the received signal from the near end;
a weighting circuit that multiplies the output of the first conversion circuit by a weight;
a subtraction circuit that subtracts the output of the weighting circuit from the output of the second conversion circuit; an adaptive circuit that adaptively modifies the weight so that the output signal of the subtraction circuit becomes smaller; and an adaptive circuit that inverts the output of the subtraction circuit. a circuit that performs orthogonal transformation, and a bidirectional circuit that monitors each element of the output of the first conversion circuit and each element of the output of the second conversion circuit in correspondence, detects a two-way conversation, and controls the adaptive circuit. An echo canceller with a call detector.