JPH0575500A - Howling prevention device - Google Patents

Abstract

PURPOSE:To prevent occurrence of howling while ensuring simultaneous duplex speech by calculating an insertion attenuation based on echo cancellation quantity. CONSTITUTION:An echo elimination means 21 generates a pseudo echo from an output signal of a duplex talking device and subtracts the pseudo echo from an input signal of the duplex talking device to eliminate and echo circulating through a closed loop. Moreover, an attenuation quantity calculation means 22 calculates the attenuation to set the gain of the closed loop to the unity or below based on the acoustic coupling gain of the closed loop, the echo cancellation quantity of the echo elimination means 21 and the amplification gain of the closed loop. Then an attenuation inserting means 23 inserts the attenuation calculated by the attenuation quantity calculation means 22 to the closed loop. Thus, even when the echo cancellation quantity is reduced, since the closed loop gain is kept to the unity or below, no howling is caused. Since the insertion attenuation quantity is calculated based on the echo cancellation quantity, the required minimum attenuation is always inserted.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a howling prevention device for preventing howling in a hands-free type two-way communication device.

In recent years, in a telephone system, a hands-free type telephone has been developed which uses a speaker (received voice output means) and a microphone (transmitted voice input means) instead of a conventional handset type telephone. There is.

Also, in the video conference system, a two-way communication device for making a call using a speaker and a microphone is used.

In such a hands-free type two-way communication device, a closed circuit is formed by acoustically coupling a speaker and a microphone (in a two-way communication device having an anti-sidetone circuit, sidetone coupling is included). It

When such a closed circuit is formed, the received voice output from the speaker may be input to the microphone as an echo, and howling may occur.

Therefore, a hands-free type two-way communication device requires a howling prevention device for preventing the occurrence of howling.

[0007]

2. Description of the Related Art Conventionally, for example, an echo canceller has been used to prevent howling from occurring.

Here, the echo canceller removes the echo flowing through the closed loop by generating a pseudo echo from the output signal of the two-way communication device and subtracting the pseudo echo from the input signal of the two-way communication device. Is.

According to the structure using such an echo canceller, the above-mentioned coupling is canceled, so that howling can be prevented.

Further, since the howling is prevented by eliminating the echo, simultaneous two-way communication can be realized.

However, this echo canceller has a drawback that the echo canceling ability is deteriorated due to double talk or echo path fluctuation.

If the echo canceling ability is lowered, the gain of the closed loop is increased, and the risk of howling is increased.

Therefore, when the echo canceller is used as the howling prevention device, it is necessary to deal with the decrease in the echo cancellation amount due to double talk or echo path variation.

As this method, it is possible to use a voice switch together with an echo canceller.

Here, the voice switch generally means
By inserting a large amount of attenuation in a route that is not in a call state between the transmitting route and the receiving route, howling is prevented.

Therefore, if the voice switch is driven when the echo canceling ability of the echo canceller is lowered, the closed loop gain is maintained below 1, so that howling can be prevented.

However, in such a configuration, when the echo canceling ability is deteriorated, a large amount of attenuation is inserted into the closed loop, so that the simultaneous bidirectional call finally made possible by adopting the echo canceller can be secured. Disappear.

Therefore, when the echo canceller and the voice switch are used together, it is necessary to set the attenuation amount of the voice switch so that the simultaneous two-way communication can be guaranteed.

As a method of calculating such an attenuation amount,
There is a method described in "Adaptive Voice Switch" (Technical Research Report of the Institute of Electronics, Information and Communication Engineers EA87-76).

According to this document, in the speech circuit model of the hands-free telephone shown in FIG. 16, the acoustic coupling gain α required to calculate the loop closed gain is the output R ₀ of the speaker 11 and the input of the microphone 12 in the receiving state. S _i ratio of (S _i / R ₀₎ is expressed as _r.

The sidetone coupling rate β is determined in the transmitting state.
Speaking voice output S₀And received voice input R _iRatio of (R_i/
S₀)_sIs expressed as

[0022]

[Equation 1]

[Equation 2] However, K: transmit signal level / receive signal level G: switch attenuation

Therefore, the acoustic coupling rate α is obtained from both equations as the minimum value of the above equation (1). The sidetone coupling rate β is obtained as the minimum value of the above equation (2).

When both coupling rates α and β are obtained, the minimum necessary switch attenuation G that can guarantee simultaneous two-way communication is given by a reciprocal of the product of both coupling rates α and β with some margin M. It is calculated by the following equation (3). G = M / (αβ) (3)

The theoretical development in this document is rigorous,
It is judged that the switch attenuation amount G determined from the coupling rates α and β obtained by the equations (1) and (2) gives the optimum value.

By the way, this document does not refer to the calculation of the switch attenuation amount G when the echo canceller and the voice switch are used together.

However, it is easy to apply the result obtained in this document to the calculation of the switch attenuation amount G in the combined configuration.

That is, in this case, the switch attenuation amount G may be considered to include the echo cancellation amount by the echo canceller.

However, in the case of this method, although simultaneous two-way communication can be guaranteed, there is a problem in that it is not possible to follow the change in the echo cancellation amount that inevitably occurs in the echo canceller.

In the case of this document, this is because the switch coupling amount G is calculated by estimating the static coupling ratios α and β in the installed state of the two-way communication device, and therefore the echo canceller of the echo canceller This is because the switch attenuation amount G cannot be controlled according to the decrease in the amount.

That is, what is meant by "adaptive control of attenuation amount" in this document is, as expected from that context, to calculate and adapt a fixed coupling rate determined by the installed environment, and double-talk. It is not to control in accordance with the dynamic coupling rate that fluctuates during use due to changes in the echo path.

Therefore, when the switch attenuation amount is calculated by the method of this document, howling may occur if the echo cancellation amount changes during use.

[0033]

As described above, since the conventional voice switch is designed to obtain the switch attenuation amount based on the fixed coupling ratio, it is considered to be used in combination with the echo canceller. However, there is a problem in that it may not be possible to prevent howling even if a simultaneous two-way call can be secured.

Therefore, an object of the present invention is to provide a howling prevention apparatus which can prevent the occurrence of howling while ensuring simultaneous two-way communication in a structure in which an echo canceller and a voice switch are used together. To do.

[0035]

FIG. 1 is a block diagram showing the principle configuration of the present invention.

In the figure, reference numeral 21 is an echo canceling means for generating a pseudo echo from the output signal of the two-way communication device and subtracting the pseudo echo from the input signal of the two-way communication device to remove the echo flowing through the closed loop. Is.

Reference numeral 22 calculates an attenuation amount for setting the gain of the closed loop to 1 or less based on the acoustic coupling gain of the closed loop, the echo cancellation amount of the echo removing means 21, and the amplification gain of the closed loop. Attenuation amount calculation means.

Reference numeral 23 is an attenuation amount inserting means for inserting the attenuation amount calculated by the attenuation amount calculating means 22 into the closed circuit.

[0039]

In the above structure, since the insertion attenuation amount is calculated based on the echo cancellation amount, when the echo cancellation amount decreases due to double talk or echo path fluctuation, the insertion attenuation amount increases accordingly.

As a result, even if the echo cancellation amount decreases,
Since the loop closed loop gain is maintained at 1 or less, howling does not occur.

Moreover, since the insertion attenuation amount is calculated based on the echo cancellation amount, it is possible to always insert the necessary minimum attenuation amount.

Thus, howling can be prevented while guaranteeing simultaneous two-way communication as much as possible.

[0043]

Embodiments of the present invention will now be described in detail with reference to the drawings.

FIG. 2 is a block diagram showing the configuration of an embodiment of the present invention. Note that FIG. 2 shows a case where the present invention is applied to a howling prevention device for a hands-free telephone as a representative.

In the figure, first, the structure of the communication circuit will be described.

In the figure, 31 is a microphone for inputting the transmitted voice. Reference numeral 32 is an output terminal to which a transmission circuit (not shown) for transmitting a transmission voice signal is connected.

Reference numeral 33 is an input terminal to which a receiving circuit (not shown) for receiving the received voice signal is connected. 34
Is a speaker for outputting the received voice.

The above is the configuration of the communication circuit. Next, the structure of the howling prevention device inserted in this communication circuit will be described.

In the figure, reference numeral 35 is an echo canceller which eliminates an echo flowing through a closed loop constituted by acoustic coupling of a speaker 34 and a microphone 31.

Reference numeral 36 is an attenuation amount calculation unit for calculating the attenuation amount to be inserted in the closed circuit. 37 is the attenuation amount calculation unit 3
6 is a voice switch that inserts the amount of attenuation calculated in 6 into a closed circuit.

The above is the overall structure of the howling prevention apparatus. Next, the configuration of the echo canceller 35 that constitutes the howling prevention device will be described.

In this echo canceller 35, 35
Reference numeral 1 is an adaptive filter that generates a pseudo echo G _j .

The adaptive filter 351 generates the pseudo echo G _j by applying the coefficient data H _j supplied from the coefficient updating circuit 354, which will be described later, to the output signal X _j of the communication circuit supplied to the speaker 34. It has become.

Reference numeral 352 is a subtraction circuit which removes the echo g _j from the closed loop by subtracting the pseudo echo G _j from the input signal Y _{j of the} speech circuit output from the microphone 31.

Reference numeral 353 is a difference calculation circuit for calculating data indicating the difference between the current coefficient data H _j of the adaptive filter 351 and the next coefficient data H _{j + 1} . This difference calculation circuit 3
53 is a residual echo E _j output from the subtraction circuit 352.
The difference data is calculated based on the output signal X _j of the call circuit.

The coefficient updating circuit 354 obtains new coefficient data H _{j + 1} by adding the difference data calculated by the difference calculating circuit 353 to the current coefficient data H _j .

Reference numeral 355 is a double talk detecting circuit for detecting the occurrence of double talk. The double-talk detection circuit 355 is configured to output the output signal of the subtraction circuit 352 and the pseudo echo G.
The occurrence of double talk is detected based on _j .

Reference numeral 356 is an echo path fluctuation detection circuit for detecting fluctuations in the echo path. The echo path fluctuation detection circuit 356 is adapted to detect the echo path fluctuation based on the output signal of the subtraction circuit 352 and the pseudo echo G _j .

357 is based on the detection outputs of the double talk detection circuit 355 and the echo path fluctuation detection circuit 356.
It is a gate circuit that controls the coefficient updating operation of the coefficient updating circuit 354.

The detection output of the double talk detection circuit 355 switches from "0" level to "1" level when double talk is detected. Similarly, the detection output of the echo path change detection circuit 356 switches from "0" level to "1" level when an echo path change is detected.

As a result, the gate output of the gate circuit 357 switches from the "1" level to the "0" level only when double talk is detected, and in other cases,
It is held at the "1" level.

The coefficient updating circuit 354 is the gate circuit 3
When the gate output of 57 is "1" level, the coefficient updating operation is executed, and when it is "0" level, this operation is stopped.

As a result, the coefficient updating operation of the coefficient updating circuit 354 is stopped only when double talk is detected, and is executed in other cases.

The above is the configuration of the echo canceller 35. The coefficient data of the adaptive filter 351 is updated based on, for example, the learning identification method.

According to this learning identification method, the coefficient data H _{j + 1} (m) of the tap m (m = 1 to I) of the adaptive filter 351 at time j + 1 is expressed by the following equation (4).

H _{j + 1} (m) = H _j (m) + {KE _j X _j (m) / ΣX _j (i)} (4) where K: correction coefficient (step gain) ΣX _j (i ): Norm (i = 1 to I)

Here, the first term on the right side shows the current coefficient data, and the second term shows the difference data. This difference data is calculated by the difference detection circuit 353 as described above,
The coefficient update circuit 354 adds the current coefficient data.

Next, the structure of the voice switch 37 will be described.

In the voice switch 37, 371 is an attenuating circuit inserted in the receiving path, and 372 is an attenuating circuit inserted in the transmitting path. Reference numeral 373 is an attenuation amount control circuit that controls the attenuation amounts of the attenuation circuits 371 and 372.

The above is the configuration of the voice switch 37. Next, the configurations of the attenuation amount calculator 36 and the attenuation amount control circuit 373 of the audio switch 37 will be described with reference to FIG.

First, the configuration of the attenuation amount calculation section 36 will be described. In the attenuation amount calculation unit 36, 361 is an echo cancellation amount calculation circuit that calculates the echo cancellation amount P _e of the echo canceller 35.

This echo cancellation amount calculation circuit 361
Circuit input signal Y_jAnd residual echo E _jPower ratio with (Y
_j/ E_j) To approximate the above echo
Erase amount P_eIs calculated.

Hereinafter, the echo cancellation amount P _e obtained by this approximation processing will be referred to as the approximate echo cancellation amount P _E.

Reference numeral 362 denotes the speaker 34 and the microphone 3.
It is an acoustic coupling gain calculation circuit for calculating the acoustic coupling gain P _g between 1 and 2.

The acoustic coupling gain calculating circuit 362 is adapted to approximately calculate the acoustic coupling gain P _g by converting the coefficient data H _j after the convergence of the coefficient estimating operation into power.

Hereinafter, the acoustic coupling gain P _g obtained by this approximation processing will be referred to as the approximate acoustic coupling gain P _G.

Reference numeral 363 is an attenuation amount calculation circuit for calculating the insertion attenuation amount P _s . The attenuation amount calculation circuit 363 calculates the insertion attenuation amount P _s based on the following equation (5) from the approximate echo cancellation amount P _E , the approximate acoustic coupling gain P _G, and the amplification gain P _{a of the} closed loop. It has become.

P _s ≈P _G + P _a −P _E (5)

However, before the coefficient estimation operation converges, the attenuation amount calculation circuit 363 calculates the insertion attenuation amount P _s by using the acoustic coupling gain P _GX under the expected worst condition. There is.

This acoustic coupling gain PP _GX is registered in advance in the internal memory of the attenuation amount calculation circuit 363.

Reference numeral 364 is a convergence determination circuit for determining whether or not the coefficient estimation operation has converged. This convergence determination circuit 3
64 determines whether or not the coefficient estimation operation has converged by determining whether or not the approximate echo cancellation amount P _E has converged to a certain value.

When the convergence determination circuit 364 determines that the coefficient estimation operation has converged, the acoustic coupling gain calculation circuit 362 described above calculates the acoustic coupling gain P based on the coefficient data H _j.
It is designed to calculate _G.

Before the convergence determining circuit 364 determines that the coefficient estimating operation has converged, the attenuation calculating circuit 363 determines the insertion attenuation P _s based on the acoustic coupling gain P _GX.
Is calculated, and if it is determined that it has converged, the insertion attenuation amount P _s is calculated based on the acoustic coupling gain P _G output from the acoustic coupling gain calculation circuit 362.

The above is the configuration of the attenuation amount calculating section 36. Next, the configuration of the attenuation amount control circuit 373 will be described.

In this attenuation amount control circuit 373, 1A
Is an attenuation amount setting circuit that sets the insertion attenuation amount P _s calculated by the attenuation amount calculation circuit 363 in the attenuation circuits 371 and 372.

Reference numeral 2A is an attenuation amount determination circuit for determining whether or not the insertion attenuation amount P _s calculated by the attenuation amount calculation circuit 363 is, for example, 10 dB or more.

When the attenuation amount determination circuit 2A determines that the insertion attenuation amount P _s is 10 dB or less, the attenuation amount setting circuit 1A determines the insertion attenuation amount P _s regardless of the calculation result of the attenuation amount calculation circuit 363. It is fixed at infinity.

Reference numeral 3A is an attenuation amount determination circuit for determining whether or not the insertion attenuation amount P _s calculated by the attenuation amount calculation circuit 363 is less than 6 dB, for example.

When the attenuation amount determining circuit 3A determines that the insertion attenuation amount P _s is 6 dB or less, the attenuation amount setting circuit 1A is irrelevant to the calculation result of the attenuation amount calculating circuit 363.
The insertion attenuation amount P _s is fixed at 6 dB.

Reference numeral 4A is a transmission determination circuit for determining whether or not the transmission state is set, based on the input / output signal of the attenuation circuit 371, for example.

Reference numeral 5A is a transmission determination circuit for determining whether or not the reception state is established based on the input / output signal of the attenuation circuit 372, for example.

If the attenuation amount setting circuit 1A is in the transmitting state based on the judgment results of the judgment circuits 4A and 5A,
The insertion attenuation amount P _s is set in the attenuation circuit 372 on the receiving side.

On the other hand, in the receiving state, the insertion attenuation amount P _s is set in the attenuation circuit 371 on the transmitting side.

Further, in the simultaneous two-way communication state, the attenuation circuit 371 which has already set the insertion attenuation amount P _s.
The insertion attenuation amount P _s is directly applied to (or 372).

When the echo path variation detection circuit 356 detects an echo path variation, the attenuation amount setting circuit 1A is irrespective of the calculation result of the attenuation amount calculation circuit 363.
The insertion attenuation amount P _s is fixed to infinity.

In this state, the attenuation amount calculation circuit 36
The insertion attenuation amount P _s calculated by 3 is maintained until it reaches 10 dB.

In the above structure, the echo canceling operation of the echo canceller 35 will be described first.

In the initial state of the coefficient estimating operation, the coefficient data H _j output from the coefficient updating circuit 354 shows a small value. As a result, the pseudo echo G _j output from the adaptive filter 351 also has a small value.

Therefore, the speaker 34
Echo g input to computer 31_iIs almost never removed
Yes. As a result, the residual error output from the subtraction circuit 352 is output.
Coe E _jIndicates a large value.

From this state, when the coefficient estimating operation progresses, the coefficient data H output from the coefficient updating circuit 354.
_j gradually increases. Thereby, the adaptive filter 351
The pseudo echo G _j output from the output also gradually increases.

As a result, the echo cancellation amount P _e in the subtraction circuit 352 gradually increases, and the residual echo E _j gradually decreases.

As a result of such an operation, when the pseudo echo G _j increases to almost the same value as the echo g _i , the residual echo E
_The value is such that _j can be ignored. As a result, the difference data is not output from the difference calculation circuit 353, and the coefficient estimation operation has converged.

When double talk occurs from this state, the apparent value of the residual echo E _j changes by the amount of the transmitted voice S _j (including the ambient noise N _j ). This allows
Difference data comes to be output again from the difference calculation circuit 353.

However, when double talk occurs, the double talk detection circuit 355 detects this double talk. As a result, the detection output of the double talk detection circuit 355 changes from the "0" level to the "1" level.

When the double talk detection output changes from the "0" level to the "1" level, the output of the gate circuit 357 changes from the "1" level to the "0" level. As a result, the coefficient updating circuit 354 stops the coefficient updating operation.

Therefore, with the occurrence of double talk,
Even if difference data is output from the difference calculation circuit 353,
Number data H_jIs not updated. This allows echo cancellation
Amount P _eIs kept in the state it was in before double talk
It

However, it takes some time for the double talk detection circuit 355 to detect double talk. Therefore, in reality, the coefficient data H _j is disturbed during this detection delay time.

As a result, when double talk occurs, the echo cancellation amount P _e decreases, and the howling is likely to occur. This problem is solved by the attenuation amount inserting operation, which is a feature of the present invention, as described later.

When the double talk state is resolved, the detection output of the double talk detection circuit 355 returns to "0" level,
Since the output of the gate circuit 357 returns to "1" level, the coefficient updating operation is restarted.

As a result, the coefficient estimating operation as described above is restarted, and the coefficient data H _j returns to its original value. Therefore, the echo g _j is removed to a value that can be ignored.

When the echo path changes, the acoustic coupling gain P _g between the speaker 34 and the microphone 31 changes. As a result, the speaker 34 to the microphone 31
The echo g _j input to is changed.

As a result, the echo cancellation amount P _e decreases and the residual echo E _j increases. As a result, the coefficient estimation operation is restarted, and the coefficient data H _j is modified by the amount corresponding to the change in the echo g _j .

As a result, even if the echo path changes, the echo g _j is finally removed.

When the fluctuation state of the echo path is eliminated, the coefficient estimating operation is restarted again, and the coefficient data H _j returns to the original state.

Even if double talk occurs, if the echo path fluctuation occurs, the coefficient updating operation is not stopped because the echo g changes when the echo path fluctuation occurs.
_{This is because j} itself changes.

The above is the echo removing operation of the echo canceller 35. Next, the attenuation amount inserting operation, which is a feature of the present invention, will be described.

When double talk or echo path fluctuation occurs, the echo canceling amount P _e of the echo canceller 35 decreases as described above.

When the echo cancellation amount P _e decreases, the gain of the one-cycle closed circuit becomes larger than 1, and the risk of howling is increased.

Therefore, in this embodiment, the voice switch 3
7 is provided, and when the echo cancellation amount P _e of the echo canceller 35 decreases, the voice switch 37 attenuates the closed loop gain so that the closed loop gain is maintained at 1 or less.

Here, the problem is that the insertion attenuation amount P _{s is}
And the insertion delay of this attenuation amount P _s .

That is, if the insertion attenuation amount P _s is too small, howling occurs, and conversely if it is too large, it becomes impossible to secure the simultaneous two-way communication state.

Further, even if the insertion attenuation amount (hereinafter referred to as "optimal insertion attenuation amount") P _s that can prevent howling from occurring and also secure simultaneous two-way communication, the insertion attenuation If delayed, howling will occur.

Therefore, in the attenuation amount insertion processing,
It is important to find the optimum insertion attenuation amount P _s and how quickly to insert it into the loop circuit.

Therefore, first, a method of calculating the optimum insertion attenuation amount P _s will be described.

In FIG. 2, if the gain of the path from the input terminal 33 of the communication device to the output terminal 32 via the echo path is A (dB), this gain A is expressed by the following equation (6). ..

A = P _g + P _a −P _e −P _s (6)

Here, the condition for preventing howling is A ≦ 0. Therefore, in equation (6),
By setting A = 0, the optimum insertion attenuation amount P _s can be obtained.

That is, A = 0 is the minimum necessary condition for preventing the occurrence of howling. Therefore,
Be calculated insertion attenuation P _s on the basis of this condition, the insertion attenuation P _s can be minimized. As a result, the simultaneous two-way communication state can be secured as much as possible.

This optimum insertion attenuation amount P _s can be expressed by the following equation:
It becomes like the following formula (7). P _s = P _g + P _a −P _e (7)

In this embodiment, the above equation (5) is used.
As described above, the approximate acoustic coupling gain P _G calculated from the coefficient data H _j after the convergence of the coefficient estimation operation is used instead of the actual acoustic coupling gain P _g .

Further, instead of the actual echo cancellation amount P _e , the approximate echo cancellation amount P _E calculated from the input signal Y _j and the residual echo E _j is used.

Further, in the initial state of the coefficient estimating operation, the approximate acoustic coupling gain P _GX under the expected worst condition is used instead of the actual approximate acoustic coupling gain P _G.

As the amplification gain a, the one obtained by measurement in advance is used.

When the insertion attenuation amount P _s is calculated based on the equation (5), this insertion attenuation amount P _s is calculated as the approximate echo cancellation amount P s.
_It changes following the change in _E.

Therefore, howling can be prevented as much as possible even if the actual echo cancellation amount P _e decreases.

That is, in the initial state of the coefficient estimating operation, since the actual echo cancellation amount P _e is small, the residual echo E _j output from the adding circuit 352 has a large value.

As a result, the echo cancellation amount calculation circuit 361
The approximate echo cancellation amount P _E calculated by means of a small value. As a result, the insertion attenuation amount P _s calculated by the attenuation amount calculation circuit 363 has a large value.

Therefore, in this state, howling does not occur even if the actual echo cancellation amount P _e is small.

From this state, as the coefficient estimating operation progresses, the actual echo cancellation amount P _e gradually increases. As a result, the residual echo E _j gradually decreases.

Since the residual echo E _j gradually decreases, the approximate echo cancellation amount P _E output from the echo cancellation amount calculation circuit 361 gradually increases. As a result, the insertion attenuation amount P _s output from the attenuation amount calculation circuit 363 gradually decreases.

Therefore, when the coefficient estimation operation proceeds,
The call state shifts to the simultaneous two-way call state without howling.

As the acoustic coupling gain, P _GX is used in the course of the coefficient estimation operation as described above, and P _G is used when the coefficient estimation operation converges.

Here, P _G is never larger than P _GX . Therefore, the insertion attenuation amount P _s after the coefficient estimation operation has converged does not become larger than the insertion attenuation amount P _s in the progress process of the coefficient estimation operation.

When double talk occurs in this state, the coefficient data H _j is disturbed until the double talk is detected by the double talk detection circuit 355 as described above.

As a result, the echo cancellation amount P _e is reduced by the amount corresponding to the change in the coefficient data H _j , and the residual echo E _{j is} reduced.
Will increase. As a result, there is a risk of howling.

However, as the residual echo E _j increases, the approximate echo cancellation amount P _E obtained by the power ratio between the residual echo E _j and the input signal Y _j decreases.

As a result, the insertion attenuation amount P _s is increased by the reduction amount of the approximate echo cancellation amount P _E. As a result, the closed loop gain is maintained at 1 or less, and howling is prevented from occurring.

After that, when double talk is detected by the double talk detection circuit 355, the coefficient updating operation is stopped. As a result, the insertion attenuation amount P _s is held at the value when double talk was detected.

When the double talk state is resolved, the stopped state of the coefficient updating operation is released. As a result, the coefficient estimating operation described above is executed again. As a result, the call state shifts to the simultaneous two-way call state without causing howling.

Even when the echo path change occurs, the echo cancellation amount P _e decreases, so that the approximate echo cancellation amount P _E decreases. As a result, the insertion attenuation amount P _s increases,
Howling is prevented from occurring.

After that, when the coefficient estimating operation is started, the echo cancellation amount P _e gradually increases, so that the approximate echo cancellation amount P _E also gradually increases. As a result, the insertion attenuation amount P _s
, The call state shifts to the simultaneous two-way call state without causing howling.

By the way, if the insertion attenuation amount P _s is larger than 10 dB, it is difficult to secure the simultaneous bidirectional communication state.

Therefore, as described above, the insertion attenuation amount P
_{Even if s} is finely calculated according to the fluctuation of the echo cancellation amount P _E , if the insertion attenuation amount P _s is larger than 10 dB, the effect obtained by the calculation process is small.

Therefore, in this embodiment, the attenuation determination circuit 2A determines whether or not the insertion attenuation amount P _s is, for example, 10 dB or more. When the insertion attenuation amount P _s is 10 dB or more, the insertion attenuation amount P _s is compulsorily, for example, I set it to infinity.

Here, the value of 10 dB is set as a value capable of ensuring simultaneous two-way communication.

According to such a configuration, the insertion attenuation amount P _s
Until a value at which the simultaneous two-way communication state can be secured is obtained, the insertion attenuation amount P for small fluctuations in the approximate echo cancellation amount P _E
The effect is that the delay in changing _s can be ignored.

The above is the method of calculating the optimum insertion attenuation amount P _s . Next, a method of coping with the calculation delay and accuracy of the insertion attenuation amount P _s will be described.

A delay is inevitable in calculating the powers of the input signal Y _j and the residual echo E _j necessary for calculating the approximate echo cancellation amount P _E. This calculation delay causes howling especially when the actual echo cancellation amount P _e decreases.

Further, in the echo canceller 35,
Fine fluctuations in the actual acoustic coupling gain P _g and the echo cancellation amount P _e occur constantly, and it is very difficult to calculate these accurately and quickly.

On the other hand, when the insertion attenuation amount P _s is several dB, simultaneous two-way communication can be realized without a feeling of strangeness. Therefore, if the insertion attenuation amount P _s is several dB, even if the insertion attenuation amount P _s is set to be always inserted, the simultaneous two-way communication state can be reliably ensured.

Therefore, in this embodiment, the attenuation amount determination circuit 3A determines whether or not the insertion attenuation amount P _s is less than 6 dB, and if it is less than 6 dB, it is forced to 6 d.
It is fixed to B.

According to such a configuration, the insertion attenuation amount P _s
The calculation delay and the calculation error to some extent are absorbed within 6 dB, and the stability of the call is secured.

By the way, when the variation of the echo cancellation amount P _e is small, this variation is absorbed by the above-mentioned 6 dB offset amount (minimum insertion attenuation amount), and howling occurs due to the insertion delay of the attenuation amount P _s. It can be suppressed.

However, when the fluctuation of the echo cancellation amount P _e is large, this fluctuation cannot be absorbed by the offset amount of 6 dB described above.

In this case, it is certain that the allowable insertion delay is until the time when howling occurs.

Therefore, in order to confirm how much time is left until howling occurs, the rise characteristic of the howling is determined by setting the loop closed loop gain at 0 dB to 42 dB.
I calculated it at 6 dB intervals.

The results are shown in FIG. However, this Figure 4
Shows the result of the one-cycle closed circuit represented by an equivalent circuit composed of the reverberation circuit 41 and the amplification circuit 42 as shown in FIG. 5, and the reverberation characteristic given by the reverberation circuit 41 being calculated as the characteristic shown in FIG. Is.

In the reverberation characteristic of FIG. 6, the impulse response is assumed to decrease exponentially to −60 dB for 1024 sample points, and
It is assumed that there is no memory (transmission delay) in the closed loop.

From the results shown in FIG. 4, in the vicinity of 40 dB, which is the most common closed loop gain in the design of a hands-free telephone, howling is extremely rapid, and delay in insertion of the insertion attenuation amount P _s is hardly permitted. I understand.

As described above, if the insertion delay is hardly allowed, it is better to consider howling will occur if the echo cancellation amount P _e is greatly reduced.

In other words, it is better to consider a countermeasure in the direction of ending the generated howling as quickly as possible.

For this purpose, it is necessary to investigate the relationship between the magnitude of the insertion attenuation amount P _s and the howling end time.

Therefore, the howling ending time was compared while variously changing the insertion attenuation amount P _s .

The results are shown in FIGS. 7 to 10 show changes in the signal level at point A (corresponding to the position of the microphone 31) in FIG. 5 when the howling is stopped by inserting the attenuation amount P _s .

First, in FIG. 7, after applying an impulse having an amplitude of 1 to the closed loop of FIG. 5, eight sampling cycles later, an attenuation amount of 45 dB which exceeds the closed loop gain 42 dB by 3 dB is inserted in the closed loop to stop the howling. The case where an attempt is made to do so is shown.

Normally, the insertion attenuation amount P _s of the voice switch 37 is
Is set to slightly above the open loop gain, but the result of FIG. 7 indicates that howling may not be terminated with such an insertion attenuation amount P _s .

Therefore, in order to terminate the howling, it is necessary to insert an attenuation amount that considerably exceeds the closed loop gain.

Therefore, the insertion attenuation amount P _s is set to 66 dB. FIG. 8 shows the transition of the signal level in this case.

In this case, the howling ends, but the ending speed is slow. Therefore, in order to shorten the howling ending time, it is conceivable to insert a larger attenuation amount.

Therefore, the insertion attenuation amount P _s is set to 122 dB. FIG. 9 shows the transition of the signal level in this case.

In FIG. 9, the level difference observed just after j = 1024 at which the signal level has decreased by 60 dB after the occurrence of howling is the level difference caused by the reverberation characteristic being cut off at j = 1024 in the simulation.

Therefore, in the actual reverberation characteristic in which such truncation is not performed, the above-described step difference does not occur, and the signal level shows a continuous decreasing tendency.

FIG. 10 shows the transition of the signal level when the insertion attenuation amount P _s is further increased to infinity.

From the results shown in FIGS. 8 to 10, the howling end time does not change even if the insertion attenuation amount P _s is increased,
It can be seen that it only exhibits a decrease corresponding to the reverberation characteristic.

On the other hand, if the insertion attenuation amount P _s is 20 dB, it is unavoidable that the communication state becomes the one-way communication state. Therefore, even if the insertion attenuation amount P _s is 20 dB or more, the call characteristics may change. Absent.

Therefore, when the echo cancellation amount P _e is greatly reduced and howling occurs, the insertion attenuation amount P _s required for ending howling is not adaptively controlled according to the closed loop gain, but It's easier to set it to infinity and maintain this state until the howling ends or gets out of the way.

Therefore, in this embodiment, as shown in FIG. 3, when the attenuation amount determining circuit 2A is provided and the echo cancellation amount P _e is greatly reduced, the insertion attenuation amount P _s is forced to be infinite. It is set to.

However, it is considered that the echo cancellation amount P _e does not always decrease greatly. Then, FIG. 4 also shows that if the closed loop gain is 12 dB or less, there is some time until howling occurs.

If the reduction amount of the echo cancellation amount P _e is about this amount, the reduction amount of the echo cancellation amount P _e is detected during this margin time, and the attenuation amount P _{s at} which the closed loop gain does not exceed 1 is detected during that time.
Can be inserted.

Therefore, the echo cancellation amount P _e is decreased,
The decrease characteristic of the echo cancellation amount P _e due to double talk, which is one of the causes of the need to insert the attenuation amount P _s , was calculated, and the margin time that can be secured as the attenuation amount insertion time was investigated.

FIG. 11 shows a decrease characteristic of the echo cancellation amount P _e calculated when double-talk is considered when the echo-to-ambient noise ratio shifts from 40 dB to 10 dB at j = 4096.

Incidentally, in the figure, the solid line with large fluctuation is
The transition of the echo cancellation amount P _e obtained by the simulation is shown, and the smooth solid line shows the theoretical value. The details of FIG. 11 are described in Japanese Patent Application No. 3-130653 filed by the applicant of the present patent application on May 2, 1991.

As shown in the result of FIG. 11, when double talk occurs, the echo cancellation amount P _e instantly decreases to near the saturation value. Therefore, it is highly possible that the echo cancellation amount P _e has already decreased significantly by the time double-talk is detected and the coefficient updating process is stopped.

It is considered that the reason why the echo cancellation amount P _e is rapidly deteriorated is that the correction constant (step gain) K in the learning identification method adopted as the coefficient estimation algorithm is set to 1.

Therefore, the deterioration characteristics of the echo cancellation amount P _e were calculated for various correction constants K. The result is shown in FIG.

FIG. 12 shows that even if the correction constant is selected to be small at the expense of the convergence speed, the deterioration of about 10 dB will occur in an instant.

Therefore, if double talk occurs,
The insertion attenuation amount P _s needs to be an amount corresponding to the echo cancellation amount P _e that is reduced until howling occurs.

As described above, the decrease amount of the echo cancellation amount P _e due to the double talk can be approximated by the decrease amount of the average power ratio P _E between the input signal Y _j and the residual echo E _j shown in the following equation (8). it can.

P _E = σ _E ² / σ _Y ² (8) where σ _E ² : short-time power of residual echo E _j σ _Y ² : short-time power of input signal Y _j

Approximate echo cancellation amount P obtained by the above equation
_{When the} amount of decrease in _E (corresponding to the insertion attenuation amount P _s ) exceeds the insertion attenuation amount P _s that guarantees the simultaneous bidirectional communication state by several dB, the control of the insertion attenuation amount P _s may be started. ..

Therefore, in this embodiment, as described above, in principle, the insertion attenuation amount P _s is calculated based on the approximate echo cancellation amount P _E , and the calculation result is 10 dB or less that can guarantee the simultaneous two-way communication state. In this case, the insertion attenuation amount P _s is set to infinity, and the calculation result is used as the insertion attenuation amount P _s between 10 dB and 6 dB.

Another factor that reduces the echo cancellation amount P _e is the echo path fluctuation, as described above. If this variation is also small, control similar to double talk is possible.

Therefore, in order to confirm the degree of reduction of the echo cancellation amount P _e due to the echo path variation, only the polarity of the first sampled value of the impulse response that gives the echo path characteristic is changed, and the change amount changes the residual echo E _j . I investigated how much it changes.

The results are shown in FIG. However, in FIG. 13, the echo-to-ambient noise ratio is about 41 dB.

The ratio of this echo path variation to the total echo path gain is only 0.2 dB in terms of power.

Even with such a slight variation, the residual echo E _j rapidly increases as shown in FIG. This increase gives the same effect as double talk, the coefficient estimation error sharply increases, and the echo cancellation amount P _e sharply decreases.

FIG. 14 is a diagram showing an impeller for giving an echo path characteristic.
Samples giving polarity changes to the loose response were set to 1-4
In some cases, echo cancellation obtained when the echo path changes
Amount P _eThe convergence characteristics of are shown together with the normal convergence characteristics.
It is a thing.

From the results shown in FIGS. 13 and 14, it is understood that the echo cancellation amount P _e is greatly reduced even if the echo path variation is small, which is almost equal to the return of the coefficient estimation operation to the initial state. ..

Therefore, considering together with the result of FIG. 4, when the echo path changes, this change is detected quickly,
It is clear that it is necessary to insert the attenuation amount P _s .

The attenuation amount that needs to be inserted when the echo path is changed may be such that the closed loop gain is less than 1, for example, 40 dB before howling occurs.

However, it is apparent that the insertion of such a large attenuation amount P _s makes the communication state a one-way communication state. Therefore, in this case, an infinite amount of attenuation P _s may be inserted as in the case of howling.

As the earliest method of detecting echo path fluctuation, there is a method described in Japanese Patent Application No. 2-63133 filed by the applicant of the present application on March 14, 1990. However, even with this method, the detection time takes about 8 sampling cycles.

On the other hand, according to the rising characteristics of the howling shown in FIG. 11, it can be considered that howling has already occurred at the detection delay of about 8 sampling periods.

It has already been described that the insertion attenuation amount P _s should be infinite if the howling is likely to occur.

[0215] According to the patent applicant who applied for a patent on February 16, 1991, claiming the priority right under the provisions of Article 42-2, paragraph 1 of the Patent Act, whether or not howling is occurring. It takes 128 ms to detect ka.

Therefore, if the insertion attenuation amount P _s is made infinite after waiting for howling detection, it is a little slow.

Therefore, in this embodiment, the insertion attenuation amount P _s is set to infinity from the beginning. That is, when the echo path fluctuation detection circuit 356 detects the echo path fluctuation, the attenuation amount setting circuit 1A forcibly sets the insertion attenuation amount P _s to infinity regardless of the calculation result of the insertion attenuation amount calculation circuit 363. Is set to.

According to this embodiment described in detail above, the following effects can be obtained.

(1) First, since the insertion attenuation amount P _s is calculated based on the approximate acoustic coupling gain P _G , the approximate echo cancellation amount P _E , and the amplification gain P _a , the insertion attenuation amount P _s is echoed. It is possible to control by following the change of the erase amount P _e .

Thus, it is possible to prevent howling from occurring while ensuring the simultaneous two-way communication state.

(2) Since the acoustic coupling gain P _g is approximately calculated from the convergence value of the coefficient data H _j ,
This acoustic coupling gain P _g can be calculated quickly and accurately.

That is, the acoustic coupling gain P_gHow to calculate
As a method, the output signal X of the communication circuit _jAnd input signal Y_jWhen
A method of calculating by integrating the ratio of

However, in such a method, since the voice signal has a non-quantitative characteristic, the integration time becomes long and the acoustic coupling gain P _g cannot be accurately calculated.

On the other hand, in this embodiment, the acoustic coupling gain P _g is calculated based on the convergence value of the coefficient data H _j , so that the characteristics of the audio signal are not affected.
The acoustic coupling gain P _g can be calculated quickly and accurately.

(3) Further, the calculation result of the insertion attenuation amount calculation circuit 363 is 10 dB capable of ensuring the simultaneous bidirectional communication state.
In the following cases, since the insertion attenuation amount P _s is forcibly set to infinity, the delay in changing the insertion attenuation amount P _{s with} respect to the fine fluctuation of the approximate echo cancellation amount P _E can be ignored.

(4) Also, in this way, the insertion attenuation amount P _s
According to the configuration in which is set to infinity, the echo cancellation amount P _e is greatly reduced and the howling is generated as compared with the case where the insertion attenuation amount P _s is adaptively controlled according to the closed loop gain. In this case, the howling end time can be easily controlled.

(5) Further, the calculation result of the insertion attenuation amount calculation circuit 363 can surely secure the simultaneous two-way communication state 6
If it is less than dB, the insertion attenuation amount P _s is forcibly set to 6 dB, so that some calculation delay and calculation error of the insertion attenuation amount P _s are absorbed in this 6 dB, and the stability of the call is improved. Secured.

(6) Further, the echo path fluctuation detecting circuit 35
When the echo path fluctuation is detected by 6, the insertion attenuation amount P _s is forcibly set to infinity, so that even a slight fluctuation may cause howling. Changes can be dealt with quickly.

Next, a second embodiment of the present invention will be described.

In the above embodiment, the case where the insertion attenuation amount P _s is forcibly set to infinity when the echo path variation is detected has been described.

However, in the two-way communication device, echo path variation frequently occurs. Therefore, if the insertion attenuation amount P _{s is set} to infinity every time the echo path variation occurs, the call state may frequently become the one-way call state.

When the one-way communication state frequently occurs in this way, the effect of introducing the echo canceller 35 is halved.

Therefore, this embodiment is based on the attenuation circuit 3 of FIG.
71 and 372, the transmission band of the audio signal is divided into a large number of bands, and when the echo path changes, the insertion attenuation amount P _s is controlled independently for each divided band.

With such a structure, even if the echo path changes, the simultaneous two-way communication state can be secured as much as possible.

Such an effect can also be obtained by subjecting the audio signal to fast Fourier transform and controlling the insertion attenuation amount P _s independently for each frequency component.

Regarding the band division control and the fast Fourier transform control as described above, the applicant of the present patent application,
The detailed description is omitted here because it is described in detail in Japanese Patent Application No. 1-131593 filed on March 26.

Next, a third embodiment of the present invention will be described with reference to FIG. In FIG. 15, the same parts as those in FIG. 2 are designated by the same reference numerals.

FIG. 15 is different from FIG. 2 described above in the following four points.

(1) A delay circuit 358 is provided between the difference calculation circuit 353 and the coefficient update circuit 354 of the echo canceller 35.
The point where is inserted.

The delay time of the delay circuit 358 is set to at least the time required for the double talk detection circuit 355 to detect double talk.

(2) When double talk is detected by the double talk detection circuit 355, the attenuation amount setting circuit 1A (see FIG. 3) of the attenuation amount control circuit 373 forcibly inserts the insertion attenuation amount P _s into the insertion attenuation amount. A point fixed to the calculation result of the calculation circuit 363.

(3) The delay circuit 38 is inserted in the closed circuit.

FIG. 15 shows as a representative the case where the delay circuit 38 is inserted between the attenuation circuit 371 and the output terminal 32.

The delay time of the delay circuit 38 is set to at least the value required for the echo path fluctuation detection circuit 356 to detect the echo path fluctuation minus the transmission delay time of the closed loop.

(4) Even if the echo path variation is detected by the echo path variation detection circuit 356, the insertion attenuation amount P _s is not forcibly set to infinity based on the detection output, but the insertion attenuation amount is not changed. The insertion attenuation amount P _s is set based on the calculation result of the amount calculation circuit 363.

However, in this case, the attenuation amount determination circuits 2A, 3
Utilizing the determination result of A is the same as that of the first embodiment.

In the above structure, first, the operation when double talk occurs will be described.

The calculation result of the difference calculation circuit 353 is delayed by the delay circuit 358 by at least the double talk detection time and supplied to the coefficient update circuit 354.

As a result, even if double talk occurs,
The coefficient data H _j of the adaptive filter 351 is not disturbed until the double talk is detected by the double talk detection circuit 355.

When double talk is detected by the double talk detection circuit 355, the insertion attenuation amount P _s is forcibly fixed to the calculation result of the attenuation amount calculation circuit 363 based on this detection output.

Thus, even if double talk occurs,
The insertion attenuation amount P _s can be held at a value before double talk occurs.

Next, the operation when the echo path variation occurs will be described.

As described above, when the echo path fluctuation occurs, the echo cancellation amount P _e is greatly reduced. This allows
The risk of howling is high.

However, it takes some time to detect the echo path fluctuation. Therefore, as in the first embodiment, the insertion attenuation amount P is detected when the echo path variation is detected.
_{Even if s} is set to infinity, it may not be possible to avoid howling.

On the other hand, in this embodiment, the delay circuit 38 having a delay time of at least a value obtained by subtracting the transmission delay time of the closed loop from the detection time of the echo path fluctuation is inserted in the closed loop.

According to this structure, howling does not occur even if the echo path change occurs until it is detected.

After the echo path fluctuation is detected, the insertion attenuation amount P _s is set based on the calculation result of the insertion attenuation amount calculation circuit 363. Accordingly, also in this case, howling can be prevented from occurring.

Although the three embodiments of the present invention have been described in detail above, the present invention is not limited to such embodiments, and various modifications can be made without departing from the scope of the invention. Of course, of course.

[0259]

As described in detail above, according to the present invention,
It is possible to provide a howling prevention apparatus capable of suppressing occurrence of howling while ensuring simultaneous two-way communication even if the amount of echo cancellation is reduced due to double talk or echo path fluctuation.

[Brief description of drawings]

FIG. 1 is a block diagram showing a principle configuration of the present invention.

FIG. 2 is a block diagram showing a configuration of a first exemplary embodiment of the present invention.

3 is a block diagram showing an example of a specific configuration of an attenuation amount calculation unit and an attenuation amount control circuit in FIG.

FIG. 4 is a characteristic diagram showing a rising characteristic of howling.

FIG. 5 is a block diagram showing an equivalent circuit of a closed circuit.

FIG. 6 is a characteristic diagram showing reverberation characteristics.

FIG. 7 is a characteristic diagram showing a transition of a signal level when an attenuation amount of 45 dB is inserted in a closed loop.

FIG. 8 is a characteristic diagram showing a transition of a signal level when an attenuation amount of 66 dB is inserted in a closed loop.

FIG. 9 is a characteristic diagram showing a transition of a signal level when an attenuation amount of 122 dB is inserted in a closed loop.

FIG. 10 is a characteristic diagram showing a transition of a signal level when an infinite amount of attenuation is inserted in a closed circuit.

FIG. 11 is a characteristic diagram showing a reduction characteristic of an echo cancellation amount.

FIG. 12 is a characteristic diagram showing a deterioration characteristic of an echo cancellation amount.

FIG. 13 is a characteristic diagram showing a change characteristic of a residual echo.

FIG. 14 is a characteristic diagram showing a convergence characteristic of an echo cancellation amount.

FIG. 15 is a block diagram showing a configuration of a third exemplary embodiment of the present invention.

FIG. 16 is a block diagram showing a call circuit model of a hands-free telephone.

[Explanation of symbols]

 21 Echo Removal Means 22 Attenuation Calculation Means 23 Attenuation Insertion Means

Claims (8)

[Claims] 1. A howling prevention device for a two-way communication device having a closed loop constituted by acoustic coupling of a transmission voice input means for inputting a transmission voice and a reception voice output means for outputting a reception voice. In, the echo canceling means (21) for generating a pseudo echo from the output signal of the two-way communication device and subtracting this pseudo echo from the input signal of the two-way communication device to remove the echo flowing through the one-cycle closed circuit. Attenuation for calculating an attenuation amount for setting the gain of the one-cycle closed circuit to 1 or less based on the gain of the acoustic coupling, the echo cancellation amount of the echo removing means (21), and the amplification gain of the one-cycle closed circuit. A howling prevention apparatus comprising: an amount calculating means (22); and an attenuation amount inserting means (23) for inserting the attenuation amount calculated by the attenuation amount calculating means (22) into the one-cycle closed circuit. 2. The echo removing means (21) is configured to calculate a coefficient by a coefficient estimating operation, and apply the calculated coefficient to an output signal of the two-way communication device to generate the pseudo echo. The howling prevention apparatus according to claim 1, wherein the howling prevention apparatus is provided. 3. The howling prevention apparatus according to claim 2, wherein the acoustic coupling gain is approximately calculated based on a convergence value of the coefficient of the echo removing means (21). 4. The howling prevention according to claim 1, wherein the echo cancellation amount is approximately calculated by obtaining a power ratio between an input signal of the echo removing means and a residual echo thereof. apparatus. 5. The attenuation amount inserting means (23) includes the attenuation amount calculating means (23) until the attenuation amount calculated by the attenuation amount calculating means (22) decreases to a value that enables simultaneous two-way communication. 2
The structure is configured to forcibly insert an infinite amount of attenuation regardless of the calculation result of 2).
The described howling prevention device. 6. The attenuation amount inserting means (23), when the attenuation amount calculated by the attenuation amount calculating means (22) is less than a predetermined value smaller than a value enabling simultaneous two-way communication,
The howling prevention device according to claim 1, wherein the howling prevention device is configured to forcibly insert the attenuation amount of the predetermined value regardless of the calculation result of the attenuation amount calculating means (22). 7. The attenuation amount inserting means (23) determines a transmission band of an audio signal when the attenuation amount calculated by the attenuation amount calculating means (22) is larger than a value enabling simultaneous two-way communication. The howling prevention apparatus according to claim 1, wherein the howling prevention apparatus is configured to be divided into a plurality of bands and set the insertion attenuation amount independently for each divided band. 8. The attenuation amount inserting means (23) Fourier transforms a voice signal when the attenuation amount calculated by the attenuation amount calculating means (22) is larger than a value enabling simultaneous two-way communication. The howling prevention apparatus according to claim 1, wherein the insertion attenuation amount is independently set for each frequency component obtained by the Fourier transform.