JPS6097732A - Echo signal eliminating device - Google Patents

Abstract

PURPOSE:To improve the detecting accuracy of an echo suppressor by using an echo canceller only for a detection section of the echo suppressor so as to eliminate or decrease an echo signal at the detection. CONSTITUTION:A switch SW302 is thrown to the position A at reception, a reception signal Rin is given to an amlifier 303 with no loss and outputted from a speaker 304. An SW311 of the transmission line side is thrown to the position A so as to give a loss of 4dB to the transmission signal from a microphone 314 so as to be set to vale of L>=GT+GR+gamma-alpha-beta, where GT, GR are gains of transmission and reception amplifiers, gamma is howling margin, and alpha and beta are the worst values of acoustic bypassing coupling. A loss of LdB is given to the reception side by the SW302, 311 at transmission and the reception side is given no loss, and a loss of L/2dB is given to the both by the SW302, 311 at double talking. A pseudo echo signal generated by an echo canceller 307 at the input side of a discriminator 308 is subtracted (310) so as to eliminate the echo signal from a discrimination input signal thereby preventing howling.

Description

[Detailed description of the invention] [Technical portion of the invention! l'ffi'] The present invention relates to an echo No. 1H removal device for removing echo signals that cause deterioration in call quality in long-distance communications such as satellite communications or cause howling in loudspeaker telephones. 3
[Technical Background of the Invention] With the scarcity of next-generation intercom devices, there is an increasing demand for high-performance intercom devices that provide various new services.

Among these, the loudspeaker telephone, which allows you to make calls without having to hold the handset, improves call quality. X is rare. This is because most of the current loudspeaker systems use the voice switch system, which causes problems such as truncated speech and difficulty in communicating in both directions at the same time.

The reason why a voice switch is indispensable in a public address chat machine is as follows. Since public address telephones use speakers, the degree of acoustic coupling caused by the sound emitted from the speakers entering the microphone is greater than in general telephones. Therefore, when the loudspeaker telephone is connected to a two-wire line, the microphone → transmitting amplifier → sidetone of the hybrid coil →
A loop of signals such as receiving amplifier → speaker → audio combination → microphone is formed, and when the loop gain becomes 1 or more, the possibility of occurrence of a ring becomes large. In addition, in a public address phone call between two parties A and B, A's microphone → A's transmitting amplifier → line → B's receiving amplifier → B's speaker → B's acoustic coupling → B's microphone → B's transmitting amplifier → A signal loop is created such as line → A's receiving amplifier → A's speaker → A's acoustic coupling → A's microphone, and the impedance matching of the hybrid coil is good, so even if there is little sidetone or a 4-wire line is used, howling occurs. Furthermore, even if howling does not occur, if the other party's telephone is a loudspeaker telephone, the acoustic coupling of the other party will cause your own voice to be reflected to a large extent and interfere with communication.

This will become a serious problem in the future when a father switching fM is introduced, such as a packet switching device, which has a large signal delay, or when the transmission path itself has a large delay, such as a satellite communication line. It becomes a point.

In order to avoid the above-mentioned problems, the voice switch method uses uke ﾾ 古竑 1 for sōdō 1 yo, and sends ﾆ 古芽 1 to send om 古廽 11 when speaking fatherly.
This method inserts a loss into the 111i sound and removes the sound-sand combination and the combination caused by the l111i sound, but as will be described later, it takes y time to detect the send/receive judgment, which may result in truncations at the beginning of words.

When two-way simultaneous calls are made, there is always a loss on one side, resulting in problems such as audio being cut off on one side.

Echo is a method that solves the problem of this voice switch.
There is one type of canceller.

Figure 1 shows an echo canceller for removing acoustic coupling.
Below 1j57. I will clarify.

Loudspeaker telephone (NaheQ) received signal X'Ct) I is amplified by the receiving amplifier (101) and input to the speaker (106). Then, the output of this speaker is reflected by the nine-piece person and is attenuated by αdB, and is input to the microphone (107) as a reverberation signal y(tl) caused by the acoustic coupling J.

Here, if an echo canceller or voice switch is not used, this signal yftl will be directly transmitted to the transmitting amplifier (1
(However, in the case of a two-wire line, the signal is sent out after being converted to 4-wire to 241 by a hybrid coil.) This causes various problems as described above.

Here, the echo canceller (103) uses a control signal from the double talk detector (105) to control the echo path characteristic, which is a characteristic in which the received signal x (t) becomes an echo signal y ttl when there is no transmitted signal x (tl). x (tl and y (transmission signal including echo signal estimated based on tl and applied within its own circuit by approximation of echo path characteristics)'(#l + n
subtracted from (it) the pure transmit signal tl (
Extract tl.

L, the echo canceller returns the echo signal y (
It becomes possible to remove only the tl, and the number of acoustic components can be reduced without causing problems like the voice switch method.

Note that the double talk detector (105) is for example.

Compare the magnetic forces of X (t) and Y (tl + n (tl), and calculate (x (magnetic force of tl) - αmin (y (tl
+ n (magnetic force of tl (however, αmin is the worst value of acoustic coupling), double talk is determined and the estimation operation of the echo canceller is stopped.

This also applies to satellite communications.

In satellite communications, as shown in FIG.
The echo signal generated in the 4-wire-28 converter 202 in the terrestrial line)' (cancels tl.

[Problems with background technology]

Now, the above-mentioned echo signal removal technology.

Echo suppressors have problems that are one meter below.

a) Problems with echo suppressors (voice switches) (1) Since each level is used to determine the transmission, reception, and double talk states, there is a possibility that there is an echo signal with a delay of several tens of milliseconds, for example. If you have this anti-11
In order to prevent malfunctions caused by signals, it is necessary to set a high threshold value for valley judgment or take a long integration time for level detection.This makes it difficult to cut off the beginning of a word or interrupt during double talk, making it difficult to make calls. This may lead to quality deterioration (.

(2) In the howling prevention device, a certain amount of loss is inserted into the communication path even when determining double talk (in order to prevent howling, the loss of the howling loop must always be kept above a certain value). The level of the signal or received signal temporarily decreases. In echo cancellers for satellite communication, when double talk is determined, the insertion loss is very small (because) 1. Although the level does not drop as in the case of a ring prevention device, an echo can be heard temporarily.

In addition, echo cancellers also have the following specific problems.

(Problems with echo cancellers (1) Most of the noise is generated using digital signal processing, so an A/D converter and a D/A converter are required, but for howling prevention etc. In this case, if high-quality speech is desired, the word length of the converter becomes thicker, which reduces cost.

(21 When used to prevent howling, the impulse response length of the echo path is long (several 100 msec), so in order to completely cancel howling, the tap length of the echo canceller must be set to a non-standard value (1000 taps or more). In addition, when high quality is required, such as in video conferencing, the sampling frequency needs to be about twice the normal rate (16KHz), and the tap length must also be doubled accordingly. (more than 2000 taps) is required, resulting in high cost.

(The 31 echo φ canceller estimates the echo path only when there is one echo path input signal, so if the echo path changes when the input signal level is low, it will not be able to follow it.
When used to prevent howling, there is a possibility that howling or quasi-sounding noise may occur.

(4) During double talk, echo path estimation is prohibited using a double talk detector, but since the detection unit requires time, the estimated echo path may be disturbed during this time.

At this time, the tap coefficient of the pseudo echo path in the echo canceller becomes a random value, the frequency characteristics are disturbed, and this distorts the pseudo echo signal, which may distort the cancellation residual signal.

Furthermore, in the case of howling prevention, there is a possibility that howling may occur due to a decrease in the amount of cancellation at this time.

[Purpose of the invention]

The present invention has been made with these circumstances in mind, and its object is to provide an echo signal canceling device that eliminates the above-mentioned problems peculiar to echo cancellers.

[Summary of the invention]

As mentioned above, problem (1) of echo suppressors is caused by the presence of echo signals. If there is no echo signal, the threshold for judgment can be set low, so one word can be cut off,
The difficulty of interrupting during double talk is reduced.

The present invention focuses on this point and uses an echo-1 canceller only in the detection section of the echo suppressor, and
The detection accuracy of the echo suppressor is improved by removing or reducing the echo signal. Then, in one communication path, echo signals are removed by an echo suppressor.

〔Effect of the invention〕

Therefore, according to the present invention, the following new effects can be obtained over conventional echo suppressors and echo cancellers.

(1) Since the thresholds for determining sending/receiving and double talk can be set low, the call quality is improved compared to conventional echo suppressors.

(Since the 21 communication path is an echo suppressor, analog processing is possible and there is no need to use expensive A/D and D/A. Also, the echo canceller for detection completely eliminates echo. Since there is no need to remove the tail part of the impulse response of one echo path, the number of taps is reduced compared to using an echo canceller alone to prevent howling.

In addition, even in high-quality communication that requires gKHz sampling in the case of an echo canceller, 4KHz sampling is required for detection.
The sampling frequency of the canceller should be 8 KHz.

(31) According to the present invention, since there is always a loss in the communication path, howling does not occur unlike an echo canceller even if one echo path fluctuates when there is no echo path input signal.

(4) During double talk, the echo canceller for detection may be disturbed, but in this case, the echo signal is added to the 1 judgment signal due to the reduction in the canceling ability, and even after double talk is finished, a small amount of noise may be generated. During.

Double talk may continue, but no signal distortion or howling occurs due to the echo canceller.

However, since the communication path is superimposed on the echo suppressor method, problem 2) of the echo suppressor (loss occurs during double talk) is not improved, but the frequency of double talk is not that high. There is no significant deterioration in call quality.

[Embodiment of the Invention] FIG. 3 shows an embodiment in which the present invention is applied to the removal of acoustic coupling in loudspeakers.

In the same figure, the operation of this loud voice speaking system will be explained in each call mode.

a) Receive mode When receiving a call, the switch 302 selects the person from the control signal C11 and sends the receive signal Rin to the amplifier 302 without loss.
and output from the speaker 304.

On the other hand, the microphone 314. The transmission signal from the amplifier 313 is sent to the switch 311 by the split signal C2.
A is selected and transmitted with a loss of L bB.

This loss L dB is the gain of the transmitting amplifier GTdB, the gain of the receiving amplifier G RdB, and the howling margin γ
dB, sound 4 condensation volume coupling amount value αdB, 1llll sound.

The worst value of the acoustic coupling sensitivity due to communication with the other party is set to βdB (However, in this embodiment, considering that the other party has a similar circuit, let β (α) be L> GT + GR + γ - α - β. .

b) Sending mode When transmitting, switches 302 and 311 turn L on the receiving side.
dB loss, and the transmitting side transmits without loss.

C) Double talk mode For double talk, use switches 302 and 311.

A loss of L 72 dB is given to both transmitting and receiving calls.

The above is the operation of one communication path in this embodiment, and this part is similar to the conventional system.

The present invention is characterized by transmission/reception determination that controls this loss. Conventionally, six signals for transmitting and receiving were input directly to the determination circuit 30B that performs this transmission/reception determination, but since an echo signal is added to the transmitting signal 1δ, the determination threshold is set to prevent malfunctions caused by this. It has become necessary to raise the temperature, and it has already been mentioned that this is the cause of the persimmon problem.

In the present invention, in order to remove this echo signal.

The input ζ on the transmitting side of the determination circuit 30B is placed in the echo canceller 307, and the pseudo anti-# signal generated here is sent to the subtracter 31.
0, the signal is subtracted from the transmit signal containing the echo signal (possibly removing the echo signal from the decision input signal).

Therefore, the echo signal added to the transmitted signal is, for example, -30
If it can be suppressed to below dBm, the transmitting signal can be detected if a signal of -25 dBm or higher is added to the determination input on the transmitting side with a margin of about 5 dB.

Traditionally, anti? If it can be determined that the outgoing call was detected at a level higher than the worst value of the signal (for example, 6 dB below the received signal level), the detection accuracy will greatly increase and the call quality will improve by one level.

FIG. 3 shows a configuration example of the determination circuit 308.

The level of the transmitting signal S from which the echo signal has been removed by the receiving device No. 4 and the echo canceller is detected by the detectors 401 and 404, and the Yohiro bass filters 402 and 405, and is added to the comparators 403 and 407.

Normally, the switch 406 selects A11Ill as the reference input of the comparators 403 and 407, and the reference voltage from the reference voltage generator 409 is applied.

This voltage distribution defines a threshold value for determining a transmitted signal and a received signal, and if the threshold value is, for example, -25 dBm, it has a value equal to the voltage of the level detection section corresponding thereto.

Therefore, in the receive mode, the input of the comparator 403 is higher than the reference voltage, and the logic value "1" becomes the output of the comparator 403.

Also, in the transmit mode, the output of the comparator 407 is "1".
Therefore, during double talk, the outputs of both comparators become rlJ, so that all inputs of the AND gate 408 (normally, however, the control signal C3 is "1") become "1".
”, so the output of this gate is.

It becomes "1" only during double talk.

These respective outputs are: lI+t m+ No. 18 C1, C2, and are given as signals to the switches 302 and 311f.

Nao: 11 &= When I was jealous of the opening, the learning ν of the echo canceller 307 was completed, and -I (the anti-No. At this time.

Until the learning of the echo canceller is completed, it is determined to be tuple talk, and although no ringing occurs, 1. There will be a loss of /2, and each level will go down.

The ftjlJ 1iI11 signal C3 prevents this, and when the start of a call is detected by the call start switch 316 and the timer 309 is activated, the control signal c3 is turned off for a while (for example, several seconds) until the learning of the echo canceller is completed. -Oj. At this time, the switch 406 selects the B side, and when the transmitting signal level is higher than the receiving signal, the transmitting mode is set, which is one conventional amplification episode.

In addition, if a branch etc. occurs and the echo path suddenly changes during a call, it will temporarily enter a double talk state and the transmitting/receiving level will drop (several seconds without J occurring). It is not a big problem because it is a big problem.

The above are the tables and columns of the present invention.

In addition, in the case of one-star landing, howling does not occur even if the loss is zero during double talk, so the loss 305°312 is set to 0 or several dB for the same reason as the call start switch 316. The timer 309 and the control signal C3 can basically be realized with the same configuration as the actual flow side, except that they become unstable.

Finally, the present invention can also be implemented in the following manner, if necessary. The present invention is characterized by transmission/reception and double toll determination, and one communication path is a normal echo-zapresa system. Therefore, for example, by estimating the attenuation of the echo path, C
5 Conventional echo/echo that optimally sets insertion loss
Improvements to the suppressor can be applied as is. Further, the threshold value for determining transmission/reception can also be changed according to the cancellation performance of the detection echo canceller, for example, by monitoring the residual false alarm level.

[Brief explanation of the drawing]

A diagram showing a conventional example using the echo canceller of JTX Koehoroka Kasumi.
Figure 3 shows an example of the use of Canceller in one tumor (Fig. 3 of the present invention)
FIG. 4 is a diagram showing an example of the configuration of a determination circuit. 203...telephone

Claims (1)

[Claims] (1) An echo signal removal device that inputs a received signal and a transmission signal including an echo signal of the reception signal, and in a reception state adds loss to the transmitter signal to remove the echo signal, and in a transmission state , in the receiving state, the detection of a double talk state in which both the true transmitted signal without the echo signal and the received signal are present at the same time is based on the received signal level and an echo cancellation device. (2: Detection of the transmission state, reception state, and double talk state is performed when the level of the transmission signal is above a predetermined value, and the level of the transmission signal of the device is below a predetermined value. The receiving state is set when the device is in the receiving state, and the transmitting state is set when the opposite state occurs, and the double talk state is determined when the receiving signal and the transmitting signal of the device are both at a predetermined level or higher. The echo signal removing device according to item 1.