JPS6096055A - Echo signal canceller - Google Patents

Abstract

PURPOSE:To constitute easily an echo canceller cancelling a long echo signal by using a low pass filter eliminating a signal component of a frequency > shift clock X 1/2 in response to a shift clock of a transversal filter. CONSTITUTION:An echo input signal x(t) subject to band limit to 0-4kHz, for example, becomes a digital signal x(k) sampled by 8kHz at an A/D converter 301, inputted to a transversal filter 302, where a signal x (k-N+1) shifted by N-stage in this filter is subject to band limit to, e.g., 0-2kHz by a digital low pass filter 403, becomes x' (k-N+1), which is inputted to a transversal filter 404. Since this transversal filter is, e.g., N'=2,000 tap filter and subject to band limit to 0-2kHz, the signal is shifted by 4kHz sample. Thus, the filter can synthesize the pseudo echo signal y2(k) in the range of 125-500msec.

Description

[Detailed Description of the Invention] [Technical Field of the Invention] The present invention is directed to a system in which the sound emitted from a speaker is reflected on a human-powered wall and enters one channel in the 4th episode of amplification [Technical Background of the Invention] Amplified 1U telephones that allow you to talk from one side to the other without using the handset are very convenient because you can use both hands freely during a call. Indispensable (= considered to be.

In the loudspeaker telephone of C, the received sound is amplified by an amplifier and given to the speaker, so the sound emitted from the speaker is picked up by the microphone and amplified by the quick-connect amplifier. Then, a loop occurs in the signal that is input to the receiving amplifier again. So just a microphone, a speaker.

If amplification is used to construct a loudspeaker, there is a risk that this signal loop will cause - ringing, making Ii/X') impossible.

Conventionally, in order to prevent this - ringing, the voice switch method inserts a loss on the receiving side when transmitting a call, and inserts a loss on the transmitting side when receiving a call to break this loop. This method has been widely used.However, this method does not allow simultaneous calls due to the loss introduced into one of the communication paths, and when the loss is switched (giving the two talkers an unnatural switch feeling). It had drawbacks.

On the other hand, with the advancement of digital signal technology (2), echo cancellers have become relatively easier to implement than LSI (2), and voice switch systems (1) have attracted attention as a technology to replace them.

This echo canceller is used in a loudspeaker telephone as shown in Figure 1 (= the signal n(t) +
Of y(t), only the signal y(t) emitted from the speaker and reflected by walls etc. is canceled, so the signal loop as described above is broken between the speaker and the microphone, and howling can be prevented. Therefore, if this echo canceller is used, there is no need to introduce loss into the communication path, so better speech quality can be obtained compared to the voice switch method.

This echo canceller can have a configuration of 42, for example, as shown in FIG.

In the figure, the transversal filter 201 is configured so that the signal x(t) from the speaker is reflected (= from y(1)
(this is called the transfer function of anti-4Il\18) and a filter with approximate characteristics (this filter is called pseudo-anti-711I11#l!), and in general, the impulse response of the transfer function and the approximation are A transversal filter (consisting of two) with tap coefficients.

This L! ! In one pass, a pseudo echo signal y(t) is generated by approximating the reflected signal y(t) using the 1δ number x(t), and the microphone input signal y<t)+n(t) is subtracted. vessel 20
By subtracting 2 e9<t><Φy(t)),
Only the echo signal y(t) is canceled and the true transmission signal 1d is n(t)
i Extract.

Here, the tap coefficient of the transversal filter 201 is determined by the No. 18 x(t) from the speaker and the output signal e(t) of the subtracter 202. The learning algorithm sequentially (−
I can find it. In the figure, an estimating circuit 203 inputs all tap coefficients from the filter 201, makes corrections according to the learning identification method, returns the tap coefficients to the filter 201, and sequentially inputs the tap coefficients into the filter 201.
1, the tap coefficients of the transversal filter are finally made to approximate the impulse response of the transfer function of the reflected signal.

Note that it is necessary to estimate this tap coefficient only when the microphone force is the reflected signal y(t). Otherwise, the reflected signal y(t) will be masked from the transmitted signal n(t)l2, and the estimation of the tap coefficients will be inaccurate. Therefore, the double talk detector 204 detects the microphone input signal n(t)
+ y(t) and deer with received signal x(t),
When the negative force of n(s+y(t)) is greater than the negative force of x(t), which is a constant decrease of one force, it is determined that the transmission signal n(t) exists, and the tag of the estimation circuit 203 is determined. Prohibit coefficient update.

[Background technology point]

The above-mentioned echo canceller can be used as a sound-I in a public address telephone.
In order to use it for i-coupling removal, since the impulse response length of the acoustic echo path is as long as 300 ms to 400 ms, it is necessary to use a transpersal type with 4-f/b and a sample of about 8 KHz with a frequency of 3000 to 4000 ms. This resulted in an increase in the number of taps required.

For this reason (2), taking advantage of the tendency that wide band signals attenuate faster in reverberant signals, a full band echo canceller 303 and a low band echo canceller 302 as shown in FIG. By using two types in parallel and lowering the sampling frequency for low bands, the amount of delay per tag is increased, and the amount of delay for anti-# signals that can be hit/A is correspondingly increased (however, only for low bands) Several methods have been proposed in the past.

(For example, literature such as 4+ Telecommunications Society Technical Report EA83-5 (=described)) However, since this conventional method uses two independent echo cancellers, further improvements are desired in the following points. It will be done.

(1) Since two echo cancellers are connected in parallel, if both are operated adaptively at the same time, the echo path:
Adaptive operation of the echo canceller closer to the echo path (2 changes in characteristics may have a negative effect on the adaptive operation of the echo canceller of the song, so to prevent echo cancellation, preset the echo canceller closer to the echo path) Therefore, except during initial training, it becomes necessary to prohibit the adaptive operation or to reduce the speed of the adaptive operation of the echo canceller on the echo path side.

i2) The low-frequency part of the full-band echo canceller overlaps with a part of the low-frequency echo canceller (the initial inno(lus response part)), so it is wasted.

[Purpose of the invention]

The non-invention was made in view of the above-mentioned problem ζ. [Summary of the invention] The gist of the non-invention is that within the same echo canceller, the shift clock of the transversal fill will be significantly delayed. By lowering the frequency by a certain amount, the impulse response length is long and the low frequency region is strengthened.

However, simply changing the shift clock does not work.

Since the input No. 48 of the transversal filter and the residual signal for tag coefficient correction contain a frequency component of X% or more (shift clock), this may cause malfunction.

In this patent, in order to prevent this, according to the shift clock, 1 of the frequency of (shift clock)
A low-pass filter that removes the d component is used.

[Embodiments of the invention]

Embodiment (i) of the present invention is shown in FIG. 4C and will be described in detail below. Greeting device 30
The digital signal x(k) sampled at 1 and 8 KHz (Jt) is input to the transversal filter 302.

This transversal is, for example, a filter with N=1000 taps and a frequency of 8 KHz J)? In order to shift in synchronization with the angle, the initial pseudo-reverse 4
Synthesize Hakugo Buri.

On the other hand, the signal x(k-N
+1) is limited by a digital low-pass filter 403 to, for example, 0 to 2 KHz E-band 1511, and x'(k -
N+1) is input to the 9 transversal filter 404.

For example, this transversal filter has id'N'=20
x'(k-N+1) is 0 to 2K with 00 tap filter
Since it is band limited to Hz, it shifts by 4 KHz samples. Therefore, in this filter, 125~
A pseudo echo signal 92 (0) in the range of 500 m5ec can be synthesized.

Therefore, if we add 9+(k) and yi(tc), we get
4. Can synthesize long pseudo-reverberation signals of O ~ 500 m5ec! (= become.

The adder 407 performs this addition, and this addition is 8
Since it needs to be done with a KHz sample,'? 2(k
) is converted from 4 KHz samples to 8K by interpolation circuit 406.
Hz samples (two transforms f.

The pseudo echo signal y(k) synthesized by this adder is ! In a calculator 409, the echo signal y(k) sampled by the A/D converter 408 is subtracted, and one is the residual signal and D
It is output to the outside via the /A converter 412.

The other one is a tap coefficient estimation circuit 4 that updates the tap coefficients of the transversal filter and estimates the echo path.
05, 410 + two people. Note that the estimation circuit 410, which is a tap coefficient estimating circuit for the transversal filter 2, may malfunction if a signal component of 4 KHz or more is added to the a-pass filter 41.
0-4KH2 in 1! :Band+ttlJ 1ifi and supply.

Note that the interpolation circuit 406 has an 8K
A switch circuit 501 that switches alternately at Hz and 0 to 2
The KHz digital low-pass filter 502 can be easily configured.

〔Effect of the invention〕

According to the above, if the present invention is used, approximately 2000 taps
It can be configured as an echo canceller that cancels the long echo No. 18 of 500 m5ec. Also, unlike the conventional system shown in FIG. 3, the C1, which is essentially a dual echo canceller, has the same performance as the conventional single echo canceller.

For example, at 500 m5ec or more, I KHz
In the case of a reverse four-way filter in which most of the above band signals are attenuated, the broken and lined sections in Figure 4 should be roughly connected (two vertical threads) (however, the band of the low-pass filter is from 0 to IK output). .

The sampling frequency is 2 KHz) and 5T.

・The method for realizing the above embodiment is to convert the % part into LSI etc.
From 2, it is possible to realize it in hardware, or it can be realized in software (from 2) using a digital signal processor.

Furthermore, in this embodiment, the delayed signal x (k=N+1) is
Although the output of the register in the transversal filter at the previous stage is used (c), it is also possible to provide a delay circuit and use this.

[Brief explanation of the drawing]

Figure 1 is a diagram showing the configuration of a conventional example, Figure 2 is a diagram showing the configuration of an echo canceller, Figure 3 is a diagram showing a conventional configuration of an echo canceller for acoustic coupling, and Figure 4 is a diagram showing an uninvented embodiment. ,
FIG. 5 is a diagram showing an example of an interpolation circuit. 101...Microphone 104...Speaker 301, 307...Low pass filter 304...
・Bypass filter

Claims (1)

[Claims] 11) The low frequency band (compared to the two, the high frequency/shore area is installed quickly, but the echo signal canceling device that cancels the echo signal by kneeling) is 4 signal cancellation-J device inputs the echo path input signal which has been canceled at frequency f, and has a first transformer canceling filter which forms a first pseudo anti/# signal. Input a signal that is delayed by a certain period of time from the input signal and further passes through a first low-pass filter that extracts the entire frequency range lower than the counter-route input No. 13, and f) input the input signal at a sampling frequency of f'. and the second
At least one that produces the #1 West issue! 44
2 transversal filter and 1 word of pseudo-contradiction of 42 punishment (=yoυ, 4<L1 of 41.1
at least one interpolation circuit for converting the signal into a signal having the same sampling frequency f as the number 44; an output of the interpolation circuit and an output of the l-th transversal filter t-/1III;
a subtraction circuit that subtracts the output No. 13 of the adder circuit from the output signal of the echo path; and a tap coefficient of the first transversal filter +>fJ by the output of the subtraction -# valley. input the output of the tap coefficient setting circuit of the first tap to be set and the tc calculation circuit into the first low-pass filter, and a signal passed through a second low-pass filter having equal or similar characteristics; 1. A reverberation signal canceling device comprising: a second tap coefficient setting circuit for setting tap coefficients of a second transversal filter.