JPH07231281A - Echo canceller - Google Patents

Abstract

PURPOSE:To form an echo canceller that detects and identifies a double-talking state, an echo path fluctuation state and a howling state with high accuracy. CONSTITUTION:The canceller is provided with frequency band split circuits 16, 17 dividing respectively a reception signal x(n) and an echo signal y(n) into plural frequency bands, plural pseudo echo paths 7k provided to each of plural frequency bands, and each of the plural pseudo echo paths 7k is made up of a digital filter and a coefficient of the digital filter is corrected sequentially based on the algorithm operated to minimize a cancellation error of the echo signal y(n). Then the canceller is provided with plural double-talk detection circuits 11k provided respectively to each of the plural frequency bands, a howling detection circuit 19 monitoring the detection output of the double-talking detection circuits 11k and a frequency band synthesis circuit 18.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an echo canceller, and more particularly to a two-wire to four-wire converter, a loudspeaker communication device and other communication devices having a transmission / reception transmission path, which cause howling and impair hearing. The present invention relates to an echo canceller that cancels a reverberant echo signal.

[0002]

2. Description of the Related Art A conventional example will be described with reference to FIG. With the spread of satellite communication devices, voice conferencing devices, and other communication devices having transmission / reception transmission paths, it has been desired to provide a communication device with excellent simultaneous communication performance and low echo. Echo cancellers have been developed to satisfy this demand. FIG. 2 is a diagram illustrating a conventional example of an echo canceller,
The figure shows a case where it is applied to a voice communication device.

A reception path from a reception input terminal 1 from which a reception signal x (t) is obtained to a speaker 2 which receives the reception signal x (t), and a transmission path from a microphone 3 to a transmission output terminal 4. In the communication device consisting of and, the received signal x (t) is A /
The digital reception signal x (n), which has been digitally converted by the D converter 8 and has been digitally converted, is then supplied to the pseudo echo path 7 and the pseudo echo signal y via the pseudo echo path 7.
Converted to ^ (n). On the other hand, the echo signal y (t) from the speaker 2 obtained via the microphone 3 is A
The echo signal y (n), which is digitally converted by the / D converter 5, is obtained. In the subtractor 9, the echo signal y (n) can be eliminated by subtracting the pseudo echo signal y ^ (n) from the echo signal y (n).

Here, the pseudo echo path 7 needs to follow the temporal change of the echo path. In the conventional example of FIG. 2, the pseudo echo path 7 is composed of a FIR filter which is a digital filter. The filter forming the pseudo echo path 7 is, for example, an LMS method, a learning identification method, an ES method or an RLS method or other adaptive algorithm so that the error e (n) = y (n) -y ^ (n) approaches 0. The filter coefficient is sequentially corrected by the estimation circuit 6 using Since the pseudo echo path 7 is modified, optimum echo cancellation is always maintained.

Next, a conventional filter coefficient correction method will be described. First, the LMS method, the learning identification method and other gradient type adaptive algorithms are expressed as follows. h ^ (n + 1) = h ^ (n) + α [-Δ (n)] (1) However, h ^ (n) = (h ^ ₁ (n), h ^ ₂ (n), ... .., h ^
_L (n)) ^T : Pseudo echo path (FIR filter) coefficient Δ (n): (Mean) Gradient vector of squared error α: Step size (scalar amount) L: Number of taps T: Transposition of vector n: Discretization The gradient vector Δ (n) of the time (mean) squared error is, for example, LM
In the S method, Δ (n) = -e (n) x (n) In the learning identification method, Becomes However, e (n): estimation error (= y (n) -y (n) + s (n)) y ^ (n) = h ^ (n) ^T x (n): pseudo echo signal x (n) = (X (n), x (n-1), ...., x (n-L + 1))
^T : Received signal vector.

The coefficient correction operation by the adaptive algorithm is
It is executed in the receiving state, that is, when s (n) ≈0 and the error signal e (n) can be regarded as the cancellation error y (n) -y ^ (n) of the echo signal. However, the transmission signal s of the near-end speaker
If (n) is present, the error signal e (n) contains the transmission signal s
(n) is included. Since the estimation accuracy of the adaptive filter is limited by the ratio of the echo signal y (n) and the transmitted signal s (n), the coefficient correction operation is performed in the presence of the transmitted signal s (n) of the near-end speaker. If is continued, the filter coefficient will be disturbed. Therefore, as in the case where the transmission signal s (n) is added, the echo signal y
In the double-talk state in which the magnitude of the transmission signal s (n) is not negligible compared to (n), the correction term α [−Δ in Eq.
(n)] is set to 0 to stop the coefficient correction and h ^
It prevents (n) from being disturbed. The above control is called double talk control.

FIG. 3 is a diagram showing an internal configuration of the double talk detection circuit 11. The reception signal x (n) is set to the reception signal level p _x (n) by the reception signal level detection circuit 12, and the transmission signal y (n) is set to the transmission signal level p _y (n) by the transmission signal level detection circuit 13. The error signal e (n) is set to the error signal level p _e (n) by the error signal level detection circuit 14. These levels are sent to the double-talk determining circuit 15, and the levels are compared to determine whether or not the state is the double-talk state. When it is determined that the state is the double-talk state, a control signal is sent from the double-talk determination circuit 15 to the estimation circuit 6, and the correction term α [−Δ (n)] in the equation (1) is transmitted.
Is set to 0, where the coefficient correction is stopped. In addition,
Pseudo echo path (FIR filter) even if coefficient correction is stopped
The echo is canceled because the coefficients are retained and a pseudo echo is generated.

Next, a method of determining the double talk state will be described. First, the reception signal level p _x (n) and the transmission signal level p _y (n) are compared. As a result, the received signal is smaller than the predetermined threshold Th _x (p _x (n) <
Th _x ), and the transmission signal is larger than a predetermined threshold Th _y ( _py (n)> Th _y ), it can be determined that the one-way single transmission communication state is set. On the contrary,
When the reception signal exists (p _x (n)> Th _x ), the reception signal becomes an echo signal and wraps around the microphone 3.
The value of p _y (n) echo signal exists increases, the which it is difficult to determine the near-end speaker's transmission signal s (n). Therefore, the transmission signal level p _y (n) is compared with the error signal level p _e (n). If the transmission signal s (n) is not added, the error signal e (n) after echo cancellation is extremely small, while if the transmission signal s (n) is added, the error signal e
(n) increases dramatically. As a result, the transmission signal s (n) can be detected.

However, even when the impulse response changes due to the movement of a person or an object, the error signal e (n) increases drastically. Therefore, the double talk state in which the adaptive operation should be stopped and the echo path in which the adaptive operation should be continued. It is not possible to distinguish fluctuations. Further, when the howling occurs, the adaptive operation is stopped. Therefore, once the howling occurs, the howling cannot be canceled and the call becomes impossible.

In the echo canceller, it is extremely important in practice to accurately distinguish the double talk state in which the adaptive operation should be stopped from the echo path variation state and the howling state in which the adaptive operation should be continued.

[0011]

In the conventional example as described above, the double talk state in which the adaptive operation should be stopped and the echo path variation state and the howling state in which the adaptive operation should be continued cannot be accurately distinguished. When once howling occurs, it is judged as a double-talk state and the correction operation is prohibited, so that the howling cannot be resolved and there is a problem that a call becomes impossible.

The present invention provides an anti-erasing device which automatically eliminates howling by detecting and distinguishing a double talk state, an echo path variation state and a howling state with high accuracy, and operates stably. .

[0013]

[Means for Solving the Problem] The received signal x (
n) and the echo signal y (n) after passing through the echo path of the reception signal x (n), the pseudo echo path 7 is generated, and the reception signal x (
n) is input to the pseudo echo path 7 and the pseudo echo signal y (n) obtained by subtracting the pseudo echo signal y ^ (n) obtained from the pseudo echo path 7 from the echo signal y (n) is eliminated. It is composed of a digital filter, and the coefficient of the digital filter is sequentially modified by an adaptive algorithm. The reception signal level p _x (n), the transmission signal level y (n), and the error signal level p _e (n) In the echo canceller that prohibits the correction operation when it is determined that the received signal x (n) and the echo signal y (n) are determined to be in the double-talk state by dividing the frequency band dividing circuit 16 into a plurality of frequency bands. , 17 and a plurality of pseudo echo paths 7 _k provided in each of a plurality of frequency bands,
Each of the plurality of pseudo echo paths 7 _k is composed of a digital filter, and the coefficient of the digital filter is the echo signal y.
It is sequentially corrected by an algorithm that operates so as to minimize the erasure error of (n), and is equipped with a plurality of double-talk detection circuits 11 _k provided in each of a plurality of frequency bands. An echo canceller including a howling detection circuit 19 for monitoring the detection output of the circuit 11 _k and a frequency band synthesis circuit 18 was constructed.

In the echo canceller described above, when the double-talk detecting circuit determines that the double-talk state is in one of the frequency bands but does not determine the double-talk state in other frequency bands, An echo canceller was constructed to judge howling and continue corrective action.

[0015]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENT An embodiment of the present invention will be described with reference to FIG. In FIG. 1, the received signal x (t) is divided into N signals x _k (m) (k = k
0, 1, ..., N-1). Similarly, the echo signal y (t) is converted into N signals y _k (m) (k = 0, 1, ..., N−) for each frequency band by the frequency band division circuit 17.
It is divided into 1).

Each of the frequency bands has a pseudo echo path 7 _k.
And the pseudo echo signal y ^ _k supplied from the pseudo echo path
The echo signal y _k (m) is eliminated by subtracting (m) from the echo signal y _k (m) in the subtractor 9 _k . here,
The pseudo echo path 7 _k needs to be made to follow the time variation of the echo path, and the error e _k (m) = y _k (m) −y ^ _k (m) is 0.
The estimation circuit 6 _k using, for example, the LMS method, the learning identification method, the ES method, the RLS method, or other adaptive algorithm, is sequentially modified so that the optimum echo cancellation is always maintained.

The error signal e _k (m) in each frequency band is combined by the frequency band combining circuit 18 into the error signal e (t) in all frequency bands. A double talk detection circuit 11 _k is provided in each frequency band, and the double talk state is independently determined for each frequency band. These determination results are sent to the howling detection circuit 19. When howling is detected, by weighting the determination result of the double-talk detection circuit 11 _{k in} each frequency band, more accurate howling can be detected. Looking at the frequency spectrum of the speech, the sound pressure shows a maximum at 200H _Z vicinity regardless of male and female, show a monotonically decreasing characteristic as the frequency is increased than this. From the above, the degree of weighting is appropriately performed in consideration of this frequency characteristic.

By the way, when the transmitted voice is added, the transmitted signal level is relatively high in all of the plurality of frequency bands, and therefore it is determined that the double talk state is simultaneously in all of the plurality of frequency bands. It will be. On the other hand, when howling occurs, it is determined that the double talk state is present in only one frequency band. Therefore, if it is determined that the double-talk state is present only in one frequency band but not the double-talk state in the other frequency bands, it is determined as a howling state and the correction operation is performed. To do. As a result, howling can be immediately eliminated.

[0019]

As described above, according to the present invention, a signal is divided into a plurality of frequency bands, and a plurality of pseudo echo paths provided in each of the plurality of frequency bands respectively include a double talk detection circuit. , If the double talk state is judged independently in each band and it is judged that it is the double talk state in one frequency band, but it is not judged that it is the double talk state in the other frequency band, Is determined to be a howling state, and a correction operation is performed.

With the above-described configuration, the double-talk state, the echo path variation state, and the howling state can be detected and discriminated with high accuracy, and when the howling state is determined, the correction operation is started immediately and the howling is prevented. It is automatically canceled and the anti-erasing device operates stably. Therefore, the call quality can be improved by applying the anti-erasing device of the present invention to a satellite communication device, a voice conference device, and other communication devices having a transmission / reception transmission path.

[Brief description of drawings]

FIG. 1 is a block diagram illustrating an embodiment.

FIG. 2 is a block diagram illustrating a conventional example.

FIG. 3 is a block diagram showing the inside of a double talk detection circuit.

[Explanation of symbols]

7 _k pseudo echo path 11 _k double-talk detection circuit 16 frequency band division circuit 17 frequency band division circuit 18 frequency band synthesis circuit 19 howling detection circuit

Front Page Continuation (72) Inventor Masafumi Tanaka 1-1-6 Uchisaiwaicho, Chiyoda-ku, Tokyo Nihon Telegraph and Telephone Corporation

Claims (2)

[Claims] 1. A pseudo echo signal obtained by generating a pseudo echo path from a received signal to the echo path and an echo signal of the received signal after passing through the echo path, and inputting the received signal to the input of the pseudo echo path. The echo signal is eliminated by subtracting from the echo signal, and the pseudo echo path is composed of digital filters, and the coefficients of the digital filters are sequentially modified by the adaptive algorithm. In the echo canceller that monitors the error signal level and prohibits the correction operation when it is determined to be in the double talk state, the echo canceller is provided with a frequency band division circuit that divides the received signal and the echo signal into a plurality of frequency bands. , Each of which has a plurality of pseudo echo paths, each of which is a digital filter. The coefficient of the digital filter is sequentially modified by an algorithm that operates so as to minimize the cancellation error of the echo signal, and the multiple double-talk detection circuits provided in each of the multiple frequency bands are An echo canceller, comprising: a howling detection circuit for monitoring the detection output of the double-talk detection circuit, and a frequency band synthesis circuit. 2. The echo canceller according to claim 1, wherein the howling detection circuit determines a double talk state in one of the frequency bands even if the double talk detection circuit determines the double talk state in another frequency band. The echo canceller is characterized in that if it is not determined to be, the howling state is determined and the correction operation is continued.