JPH0771026B2 - Echo signal canceller - Google Patents

Description

Description: TECHNICAL FIELD OF THE INVENTION The present invention relates to an improvement in an echo signal canceling device that cancels an echo signal that causes howling in a public telephone, a two-way repeater or the like.

[Technical background of the invention and its problems]

With the increasing functionality of telephones in recent years, it has been desired to improve the call quality of loudspeakers. Loudspeaker telephones have a larger reception gain acoustic coupling (coupling by the microphone picking up the sound emitted from the speaker) than telephones using a general handset,
Howling is likely to occur. That is, if the voice gain of the loudspeaker is g _T (dB), the voice gain is g _R (dB), the side sound attenuation of the hybrid coil is β (dB), and the acoustic coupling attenuation is α (dB), When there is a relationship of α + β <g _T + g _R (1), howling may occur.

In general, it is considered that a gain of g _T = 50 dBg _R = 30 dB is required to guarantee a call even in the worst line loss. The worst values of acoustic coupling and sidetone attenuation are 40 dB and 6 respectively.
Since it is considered to be dB, α + β = 46 <g _T + g _R = 80 (2), and howling will occur if the transmitter and receiver amplifiers are simply connected to the microphone, speaker, and hybrid coil. Voice as shown in Fig. 4
A public address telephone with a switch is used.

In the figure, the transmission / reception determination circuit 108 inputs the power of the transmission signal from the microphone 101 and the power of the reception signal from the hybrid coil 107, and compares them to determine the transmission state and the reception state. When transmitting, the insertion loss 102 on the transmitting side is set to 0.
dB, the receiving side insertion loss 106 is set to VS dB, and vice versa when receiving. At this time, the insertion loss VS, the gain of the transmitting amplifier 103 is g _T (dB), and the gain of the receiving amplifier 105 is g _T (dB).
_R (dB), acoustic coupling between speaker 104 and microphone 101 is α
(DB), the side sound attenuation of the hybrid coil 107 is β (d
In case of B), howling can be prevented by setting Vs> (g _T + g _R ) − (α + β) (3) so that loss occurs on the receiving side during transmission and on the transmitting side during reception. Therefore, you can make a normal call.

However, this method cannot perform a normal call because a loss occurs in either one of the two talks when the transmission signal and the reception signal exist at the same time.Because the transmission / reception determination involves some delay, Calls are cut off and call quality deteriorates.

The problem of such a voice switch system can be solved by using an echo canceller as shown in FIG.

The echo canceller creates a pseudo echo path by estimating the transfer characteristic of the echo signal y (t) generated from the transmitted signal x (t) (201 in FIG. 5), thereby generating the pseudo echo signal (t). Then, the echo signal is canceled by subtracting the output signal y (t) + n (t) of the hybrid coil including the reception signal n (t) (202 in FIG. 2). Now, the cancellation amount after the echo path estimation of the echo canceller is calculated as ERLE _ope (d
When B), loss of voice switches for preventing howling _{Vs> (y T + y R} ) - (α + β + ERLE ope) ...... (4) made can be set to a value, which is compared with the case of voice switches only ERLE _The amount of loss can be reduced by _ope (generally 20 to 30 dB), and the problem of the voice switch can be improved.

The echo path estimation of the echo canceller is performed by using the training signal, temporarily changing the signal from the training signal generating circuit 203 to the transmission signal by using the switch 204, and transmitting the signal to the line.

As described above, the echo canceller is effective in reducing the insertion loss in a voice call, but has a problem that the training signal needs to be sent to the line and the call is temporarily interrupted. Therefore, as another method of the echo canceller, a method of estimating the echo path by voice widely used in satellite communication etc. can be considered, but this method may have a convergence speed that differs depending on the voice frequency due to the frequency characteristics of the voice. Due to this, low-level frequency components have the property of poorly following line fluctuations, resulting in deterioration of the cancellation characteristics of that frequency.In particular, it is applied to loudspeaker telephones that require constant cancellation over the entire band. Was not suitable for.

In satellite communication, it is sufficient to lower the power of the reverberation signal, and even if the component of a certain specific frequency is large, if the proportion seen from the whole is small, it will not be a big problem, but in the case of a loudspeaker telephone, the worst cancellation amount Howling can occur at frequencies. This has the same drawback in the two-way repeater.

[Object of the Invention]

The present invention has been made in view of the above-described drawbacks of the conventional method, and an object thereof is to perform a voice call using an echo canceller and a voice switch together without performing initial training of the echo canceller. An object is to provide an echo signal canceling device that prevents howling and enables high-quality calls.

[Outline of Invention]

In the present invention, the voice switch circuit measures the frequency characteristic of the echo canceller cancellation amount and determines the insertion loss amount of the voice switch based on this worst value at each time.

The amount of insertion loss is determined as follows.

Power spectrum XP (f) of received signal x (t) is second
Elimination amount ERL of the echo canceller at time t in Figure (a)
It is assumed that (f) has the characteristics shown in FIG. 2 (b).

At this time, the voice switch according to the present invention has a bandwidth of n.
Divide (n ≧ 1), investigate the cancellation characteristics in each partial band by FFT, etc., find the minimum cancellation amount MINERL _R in the R band, and L _vsi = L _HOWL −MINERL _R + α (5) where L _HOWL ... The amount of cancellation required to prevent howling α ... ... The amount of insertion loss L _vsi of the voice switch is determined by a formula that is a margin that takes into consideration the error during f special measurement of the amount of cancellation.

〔The invention's effect〕

According to the present invention, since the insertion loss amount of the voice switch is determined based on the frequency characteristic of the echo canceller canceling amount, it is possible to eliminate the defect caused by the frequency characteristic.

Figures 3 (a), (b), and (c) show the bands 1, 2, 4 respectively.
Echo canceller cancellation amount and insertion loss amount L when divided
_It shows _vsi (i = 1 to n), and the amount of loss required to prevent howling is 30 dB. At this time, in the case of no band division, L _vsi = 25 dB over the entire band, but division 4
Then, in the band of 0 to 1 KHz, L _vsi = 9 dB can be set, and the insertion loss can be finely controlled to improve the call quality.

In either case, the echo canceller and voice
A loss of 30 dB can be secured in both switches, and howling can be prevented (however, α = 0 in this example).

Example of Invention

An embodiment of the present invention will be described in detail below with reference to FIG. In the figure, received signal x (t) and echo y
A transmission signal S consisting of (t) and a true transmission signal n (t)
(T) is converted into discrete digital values x (k) and S (k) by AD converters (A / D) 119 and 121.

Of these, the transmission signal S (k) is obtained by subtracting the pseudo echo (k) generated by the echo canceller 113 in the subtractor 114 to obtain the residual signal e (k).

e (k) = S (k)-(k) = (y (k)-(k)) + n (k) ...
(6) Here, if the echo path estimation of the echo canceller is completely completed, y (k) = (k), and the echo is completely removed, and the acoustic coupling amount between the speaker and the microphone becomes 0. , Howling does not occur. However, as already mentioned, y (k) ≠
It is (k), and the echo cannot be canceled out completely.
Therefore, in order to prevent howling, appropriate losses 107 and 110 (LOSS) are introduced into the speech path to prevent howling due to this residual echo.

The transmission / reception determination circuit 109 introduces such a loss. This circuit can be used for three types of calls: transmission, reception, and double talk.
The state is judged by the level of the received signal x (k) and the level of the residual signal e (k), and the switches 108 and 123 cause the loss on the transmitting side during receiving, on the receiving side during transmitting, and on both transmitting and receiving during double talk. Add.

The amount of insertion loss is determined by receiving signal x (k) and residual signal e.
This is done by observing (k) to find the cancellation frequency characteristic of the echo canceller.

That is, the received signal x (k) and the residual signal e (k) are L
First Fourier transform circuit 101,106 (FFT ₁
FFT ₂ ) extracts L power spectra of Xf ₁ ... Xf _i ... Xf _L , Ef _L ... Ef _i ... Ef _L. Then, the averaging circuits 102 and 105 (MEAN ₁ and MEAN ₂ ) take a time average for a certain period of time, and the cancellation amount estimation circuit (EST) echoes the frequency fi.
Canceller cancellation amount Is calculated. The estimation of the cancellation amount is performed by the transmission / reception determination circuit 10
9 (T / R), only when receiving a call.

Next, the minimum value detection circuit 104 (MIN) will be described. Here, the minimum cancellation amount of the k-th band when the band is divided into n Is calculated, the insertion loss amount L _K of the kth band is calculated according to L _K = L _HOWL −MINERL _K + α (9), and is sent to the insertion loss circuits 107 and 110 (LOSS). Note that during double talk, L _K / 2 is sent to the insertion loss circuits 107 and 110 because loss occurs in both transmission and reception.

As shown in FIG. 1, the insertion loss circuit includes a band division circuit 116 including a filter group that divides a band into n and an insertion loss unit 117 that gives a flat loss of L _K (dB) to each band.
−1 to 117 to _n (LOSS _{1 to} LOSS _n ) and a band synthesizing circuit 118 for synthesizing the signals in the respective bands that have suffered loss.

The above is the description of the embodiment of the present invention.
Is L = 16 to 64 in consideration of the nature of voice. In addition, the number of band divisions n should naturally be large, but a significant improvement can be achieved even if n = 2 to 8. Also, α is set to 6 to 10 dB. Then, the device of the present embodiment includes several to 10 digital signal processing processors (DSP) which are widely used at present.
It can be easily realized with only about one piece.

In this embodiment, the FFT is used to calculate the frequency characteristic, but a filter bank may be used as another implementation example. Further, it is also possible to extract the frequency characteristic by using a method suitable for extracting a non-strikeable envelope such as a cepstrum. At this time, since the detection accuracy is increased, α can be reduced.

Further, although the present invention has been described with the acoustic echo canceller, it is obvious that the echo path is not limited to the sound field and can be applied to a line echo canceller such as a bidirectional repeater.

[Brief description of drawings]

FIG. 1 is a diagram showing an embodiment of the present invention, in which FIG. 1 (a) is a diagram showing the overall configuration, and FIG. 1 (b) is the same diagram (a).
FIG. 2 is a diagram showing the configuration of the insertion loss circuits 107 and 110 in FIG.
It is a figure for explaining determining the amount of insertion loss,
FIG. 3A shows an example of a long-time spectrum of x (t), FIG. 3B shows a cancellation amount of an echo canceller in the convergence process, and FIG. 3 shows an example of insertion loss. FIG. 4A is a diagram showing a case where band division is not performed, and FIG. 7B is a diagram showing a case where band division is performed.
FIG. 4C is a diagram showing a case where the band is divided into four,
FIG. 5 is a diagram showing a configuration of a conventional public address telephone, and FIG. 5 is a diagram showing a configuration of a conventional public address telephone using an echo canceller. 101,106 …… FFT 104 …… Minimum value detection circuit 116 …… Band division circuit 117 …… Insertion loss section 118 …… Band synthesis circuit

Claims (2)

[Claims] 1. An echo signal canceller in which an echo canceller and a voice switch are continuously connected, and the insertion loss amount of the voice switch is determined based on the minimum value of the frequency characteristic of the cancel amount of the echo canceller. An echo signal canceling device characterized by the above. 2. The band of the voice switch is divided into a plurality of sub-bands, and the voice switch in each sub-band is divided.
The echo signal cancellation according to claim 1, wherein the insertion loss amount of the switch is determined based on the minimum value of the frequency characteristics of the echo canceller cancellation amount in each of the corresponding subbands. apparatus.