JP2953954B2 - Double talk detector and echo canceller - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a double talk detecting device and an echo canceller suitable for use in a mobile communication network or a long distance telephone network.

[0002]

2. Description of the Related Art In a long-distance telephone line via a submarine cable or a communication satellite, a subscriber line connected to both ends is generally of a two-wire type, and an intermediate long-distance transmission portion has four lines for signal amplification or the like. It is a line type. Similarly, mobile phones (or cellular phon)
In the mobile communication network using e)), the subscriber line of the fixed-side analog telephone is a two-wire type, and the part from the terminal of the mobile telephone to the exchange is a four-wire type. In this case, two lines and four
A hybrid circuit for performing 4-line / 2-line conversion is provided at a connection portion with the line. Although this hybrid circuit is designed to match the impedance of the two-wire circuit, it is difficult to always obtain good matching, so that the received signal arriving at the 4-wire input side of the hybrid circuit is Leaks to the side, producing a so-called echo. Such an echo arrives at the sender at a lower level than the voice of the sender and with a delay of a certain time, so that a call failure occurs. Such a communication failure due to the echo becomes more remarkable as the signal propagation time becomes longer. In particular, in the case of mobile communication using a mobile telephone, various processes are performed in a wireless communication section to an exchange or the like, so that the amount of signal delay is large, and a call failure due to an echo becomes a problem.

There are echo suppressors and echo cancellers as devices for blocking the above echo. FIG. 5 shows a schematic configuration of an echo canceller used in a mobile communication network.
The echo canceller 1 shown here is provided before the hybrid circuit 2. In this figure, a normal analog telephone subscriber is called a near-end talker, and a mobile telephone subscriber is called a far-end talker. The far-end voice signal input to the echo canceller 1 is Rin, the far-end voice signal output from the echo canceller 1 is Rout, the near-end voice signal input to the echo canceller 1 is Sin, and The output near-end audio signal is indicated by Sout.

The echo canceller 1 shown in FIG. 1 comprises an echo path estimation / pseudo echo generation circuit 3, a control device 4, an adder 5, and a nonlinear processing circuit 6. Here, the echo path estimation / pseudo echo generation circuit 3 outputs the far-end voice input R
hybrid circuit 2 based on in and near-end audio input Sin
, The echo path (that is, the line on which the echo propagates) is estimated. Next, an expected echo from the hybrid circuit 2 (ie, a pseudo echo) is generated by a convolution operation of the estimation result and the far-end voice input Rin. In the adder 5, this pseudo echo is subtracted from the near-end voice input Sin, thereby canceling the echo.

A digital signal is transmitted on a transmission line, and a D / A conversion (generally μ) is performed between an echo canceller 1 for processing the digital signal and a hybrid circuit 2 for conversion to an analog line. -LAW conversion) is performed. For this reason, a non-linear characteristic relationship is established between the far-end audio output Rout and the near-end audio input Sin, and the echo path estimation / pseudo-echo generation circuit 3
A complete echo cancellation cannot be performed only by a linear operation using the above method. For this reason, an echo component that cannot be completely canceled occurs. In order to eliminate such an echo component (referred to as “residual echo”), the nonlinear processing circuit 6
Is provided. The nonlinear processing circuit 6 performs a nonlinear switching operation. That is, when the near-end voice output Sout is constituted only by the echo, that is, when only the far-end speaker is in the transmitting state (this case is referred to as “far-end speaker single talk”), the near-end is output. Audio output Sou
The switching operation is performed so as to prevent the transmission of t, or the near-end voice output Sout is replaced with pseudo noise.

The control device 4 controls the echo path estimation / pseudo echo generation circuit 3 and the nonlinear processing circuit 6. That is, detection of the non-transmission state of the far-end person or detection of double talk is performed, and the learning function of the echo path estimation is turned ON / OF.
In addition to performing F control, detection of far-end speaker single talk is performed, and switching operation of the nonlinear processing circuit 6 is controlled.

The above-described echo path estimation / pseudo echo generation circuit 3 has an echo path estimation circuit 3a, an h register 3b, and a pseudo echo generation circuit 3c, as shown in FIG. In this case, the echo path estimation circuit 3a
Generally estimates the echo path by using a learning identification method having a relatively small amount of operation among the adaptive algorithms and having good convergence characteristics, and writes a tap coefficient (described later) corresponding to the estimated echo path to the H register 3b. . The pseudo echo generation circuit 3c is configured by an FIR type adaptive digital filter, and performs a convolution operation with the far-end voice input Rin using the tap coefficient in the H register 3b.
Generates a pseudo echo.

The learning identification method is described in, for example, a paper “Regarding Echo Cancellation Characteristics of an Echo Canceller Using the Learning Identification Method” (Itakura, Nishikawa) in the IEICE Transactions '77 / 11 Vol.J60-A NOo.11. It is a well-known estimation method as shown.
Hereinafter, the outline of the learning identification method described in this paper will be briefly described.

First, it is assumed that the signal propagation characteristic of the echo path is linear, and its impulse response h (t) and input signal x
Using (t), the echo yk at time kT (T is a sampling interval) is

[0010]

## EQU1 ## It is represented by yk = h'xk. However,

[0011]

H = (h ₁ , h ₂ ,..., H _n ) ′, h _j = h ( _j T) x _k = (x _k−1 , x _k−2 ,..., X _kn ) ′ , X _j = x
( _JT ) (where 'is the transpose of a vector). On the other hand, time k
Assuming that the estimated value of h at T is H _k (hereinafter referred to as tap coefficient), the estimated value Y _k of y _k is

[0012]

## EQU3 _## It is given by Y _k = H _{k ′} x _k . Then, the successive correction of H _k by the learning identification method is

[0013]

(Equation 4) Done by Where e _k is

[0014]

Equation 5] a e _k = yk-Yk, the residual echo. This residual echo appears on the output side of the adder 5. Then, as is apparent from Equation 5, the next tap coefficient H is set so that the residual echo is reduced.
_{k + 1} is calculated.

The above-mentioned algorithm is concretely expressed by an operation in a digital circuit as follows. First, the far-end audio signal Rin taken into the echo path estimating circuit 3a is treated as a digital signal Xt (t is a sampling time) having N sample values,

[0016]

X _t = (x (t), x (t−1),..., X
(T- (N-1)). Also, the tap coefficient Ht at time t written to the H register 3b is

[0017]

[Mathematical formula-see original document] If _Ht = ( _ht (0), _ht (1), ..., _ht (N-1)), the pseudo echo generation circuit 3b (FIR filter)
The convolution operation in is

[0018]

(Equation 8) Becomes Here, if the inner product of the vectors is represented by *, the above equation 8 becomes

[0019]

## EQU9 ## Y (t) = x _t * H _t Now, if the residual echo obtained on the output side of the adder 5 is expressed as er (t),

[0020]

Equation 10] er (t) = e (t ) -Y (t) becomes, from the above equations, variation [Delta] H _t of H _t is, if the step gain and g (corresponding to the number of alpha),

[0021]

ΔH _t = g × er (t) × x _t / (x _t * X _t ), and H _{t + 1} becomes

[0022]

The [number 12] _{H t + 1 = H t +} ΔH t. Therefore, the echo path estimating circuit 3a reads the tap coefficient H in the H register 3b, and
By adding ΔHt calculated in step 1, the next tap coefficient Ht _{+ 1} is calculated and written to the H register 3b. Thus, the tap coefficient H in the H register 3b is sequentially updated. The above is the specific calculation in the digital circuit to which the learning identification method is applied. The above equations 6 to 12 are also disclosed in Japanese Patent Application Laid-Open No. 5-129989.

The following conditions are required as conditions for performing the above learning. There is a far-end voice output Rout at a level at which an echo returns as the near-end voice input Sin, in other words, the far-end speaker is in a transmitting state.

The near-end voice input Sin is composed of only echoes (or echoes and white noise), in other words, the near-end speaker is not in a transmitting state.

On the other hand, when the far-end speaker is in the non-transmission state, and when the far-end speaker and the near-end speaker are simultaneously talking (this state is called double talk), erroneous learning of the echo path estimation is performed. Therefore, it is necessary to turn off the learning function.

[0026]

As a method of detecting double talk performed by the control device 4, conventionally, a power ratio between a far-end voice output Rout and a near-end voice input Sin is used, and this is expected. Echo level (for example,
Maximum echo level -6d determined by CCITT standard
When B) was exceeded, it was determined that double talk had occurred. However, this conventional double talk detection method has a problem that detection is delayed. That is, if there is not a sufficient level difference at the beginning of double talk occurrence, double talk is not detected, and double talk is detected only when the level difference exceeds a certain value. There is a disadvantage that it is not performed well. There is also a problem that double talk cannot be effectively detected even when the transmission levels of the far-end speaker and the near-end speaker are significantly different. That is, when the transmission level of the near end speaker is lower than the transmission level of the far end speaker,
The difference between the echo level and the transmission level of the near-end speaker does not appear sufficiently, and in that case, it is difficult to detect double talk.

The low accuracy of the above double talk detection is as follows.
There is a risk of erroneous learning of the echo path estimation. When such erroneous learning occurs, not only does the echo canceling function deteriorate, but also an erroneous pseudo echo is generated, so that abnormal noise is transmitted to a far-end speaker or the like.

The present invention has been made in view of such a background, and an object of the present invention is to provide a double talk detecting device capable of accurately detecting the presence or absence of double talk. It is another object of the present invention to provide an echo canceller that can accurately detect double talk and secure good pseudo echo transmission.

[0029]

In order to solve the above-mentioned problems, the invention according to claim 1 is directed to a telephone line network for performing transmission between voice transmitted through four lines and voice transmitted through two lines. In a double talk detection device applied to an installed echo canceller, an echo path is estimated by a learning identification method, and a tap coefficient corresponding to the echo path is output.
And echo path estimation means that, of the plurality of patterns of the echo path
Impulse response characteristics and the plurality of patterns of impulses
Predetermined tap coefficient tolerance for each response characteristic
Storage means for storing errors in association with each other ;
The tap coefficient output from the storage estimating means is
For all impulse response characteristics stored by
Double talk depending on whether or not
And determining means for determining the presence or absence of

[0030]

According to the second aspect of the present invention, in an echo canceller provided in a telephone line network for transmitting between voice transmitted through four lines and voice transmitted through two lines, an echo path is estimated by a learning identification method. and the echo path estimation means for outputting a tap coefficient corresponding thereto, and the pseudo echo generation means for generating a pseudo echo by a convolution operation based on the tap coefficients, a plurality patterns
Response characteristics of the echo path of the
Predetermined for each of the impulse response characteristics of the turn
Storage means for storing the associated tap coefficient allowable errors in association with each other, and the tap coefficient output by the echo path estimation means.
The number corresponds to all the impulse response characteristics stored in the storage means.
Whether or not the tap coefficient allowable error for
Further, the apparatus is characterized by comprising determining means for determining the presence or absence of double talk, and control means for stopping the tap coefficient output of the echo path estimating means when the determining means determines that there is double talk.

[0032] In the invention of claim 3, claim
3. The invention according to claim 2, wherein the tap coefficient output by the echo path estimating means can be stored with a plurality of delays.
If the determination means determines that the double talk, the time of the determination
Retains its own memory contents and stores it in the last stage of the plurality of stages.
The stored tap coefficient is sent to the pseudo echo generating means.
It is characterized by comprising a tap coefficient delay storage means to be provided .

According to the first aspect of the present invention, a plurality of patterns
Response characteristics of the echo path and the plurality of patterns
Predefined for each of the impulse response characteristics
The tap coefficient tolerance is stored.
Is compared with the tap coefficient output by the
You. Therefore, if erroneous learning occurs, the result is reflected in the comparison result, and the determination unit determines whether or not double talk is performed based on the comparison.

[0034]

According to the second aspect of the present invention, a plurality of
The impulse response characteristics of the echo path of the turn,
Predetermined for each of the impulse response characteristics of the pattern
The stored tap coefficient tolerance is stored.
The tap coefficient output by the echo path estimating means.
And all impulse response characteristics stored in the storage means.
And the tap coefficient allowable error of
Be compared. As a result of the comparison, the echo path
The tap coefficient output by the setting means is stored in the storage means.
If the tap coefficient allowable error is exceeded, it is determined that double talk has occurred, and the tap coefficient output of the echo path estimating means is stopped.

According to the third aspect of the present invention, the tap coefficients output by the echo path estimating means can be stored after being delayed by a plurality of stages , and the determining means has determined that double talk has occurred.
In this case, the self-stored content at the time of the
The tap coefficient stored in the last of several stages is
Tap coefficient delay storage means provided to the similar echo generation means
A pseudo echo can be generated by a tap coefficient before double talk occurs.

[0037]

【Example】

A: Configuration of Embodiment FIG. 1 is a block diagram showing a configuration of an embodiment of the present invention. The control device 4 in this embodiment, unlike the control device shown in FIG. 5, controls only the nonlinear processing circuit 6. That is,
The detection of the far-end talker single talk is performed, and only the switching operation of the nonlinear processing circuit 6 is controlled. The double talk detection in this embodiment is performed inside the echo path estimation circuit 3a.

Here, FIG. 2 shows an echo path estimating circuit 3a.
FIG. 3 is a block diagram showing the configuration of the main part of FIG. 1.

By the way, the response characteristic of the impulse of the echo path is generally as shown in FIG. 3, and the response characteristic of the echo path directly corresponds to the pattern of the tap coefficient H. To be precise, the echo path estimation circuit 3
If a performs correct learning, the time series pattern of the calculated tap coefficient H (see Equation 7) is similar to the impulse response characteristic of the echo path.

However, the response characteristics of the echo path are slightly different due to the type of the hybrid and variations in the characteristics of each hybrid. In this embodiment, patterns are classified into about 10 to 20 patterns that do not cause any practical problems, and the respective response characteristics are stored in the memories M1 to Mn.

Next, reference numeral 20 denotes a double talk monitoring circuit which compares response characteristics in each memory with the tap coefficient H output from the adder 12 while sequentially selecting the memories M1 to Mn by the signal SEL.

Then, it is monitored how far the tap coefficient H output from the adder 12 deviates from the response characteristic (reference value) in the memory, and the degree is monitored for all the memories M1.
When .about.Mn is out of a predetermined allowable error range, a double talk detection signal DT is output. in this case,
Information indicating an allowable error is also stored in the memories M1 to Mn.

Next, reference numeral 25 denotes a processing register configured in the same manner as the H register 3b, and its output signal is supplied to the storage register 26-1 and the adder 12.
Save registers 26-1 to 26-n are provided, and the configuration of each register is the same as that of the H register 3b. The tap coefficient H is sequentially transferred from the storage register 26-1 to the storage register 26-n, and the output data of the storage register 26-n is supplied to the H register 3b. It has become so.

B: Operation of Embodiment Next, the operation of this embodiment having the above-described configuration will be described. First, when a call is started, the arithmetic unit 10 sets ΔH such that residual echo is reduced based on the learning identification method.
Is calculated and added to the current tap coefficient H (the tap coefficient in the processing register 25) to generate a tap coefficient at the next time point and supplied to the processing register 25. By repeatedly performing the above processing, the tap coefficient H in the processing register 25 is adaptively controlled. The tap coefficient H supplied to the processing register 25 is stored in the storage register 2.
The data is sequentially transferred to 6-1 to 26-n. Therefore,
Tap coefficient H stored in save register 26-n
Is the tap coefficient of several samples before. Since the tap coefficient H in the storage register 26-n is transferred to the H register 3b, the pseudo echo generator 3c generates a pseudo echo based on the tap coefficient several samples before.
At this time, if no double talk has occurred, the tap coefficients H output from the adder 12 are stored in the memories M1 to Mn.
The double-talk monitoring circuit 20 does not output the double-talk detection signal DT because it matches any one of the response characteristics or within the allowable error.

On the other hand, if a double talk state occurs during a conversation, the voice signal of the near end speaker is superimposed on the adder 5, so that
The calculation unit 10 enters an erroneous learning state, and the value of ΔH to be calculated is
It does not correspond to the impulse response of the echo path. As a result, the value of the tap coefficient H output from the adder 12 deviates from the response characteristics in the memories M1 to Mn. When the deviation exceeds the allowable error for all the memories M1 to Mn, the double talk monitoring circuit 2
0 outputs the double talk detection signal DT.

When the double talk detection signal DT is output, the H register 3b has the storage register 26
−n the tap coefficient H stored several samples before,
That is, since the correct tap coefficient H before the double talk state is transferred, the pseudo echo generation circuit 3c
Generates a pseudo echo based on the tap coefficient H calculated immediately before the double talk state without generating an unnecessary pseudo echo. Also, the storage register 26-1
When the double-talk detection signal DT is supplied, the input coefficients H to -26-n do not accept a new input and retain the internal data. Therefore, the tap coefficient H supplied to the H register 3b becomes Hold the value.

On the other hand, the tap coefficient written in the processing register 25 is continuously supplied to the adder 12 and
In this case, since the state is a double talk state, the value of ΔH becomes an erroneous value, and the tap coefficient H remains a value deviated from the reference value (response characteristic in each memory).

Next, when the double talk state ends and only the transmission signal of the far end talker is used, the signal Sin becomes only the echo signal, so that ΔH calculated by the arithmetic unit 10 gradually becomes a correct value. . As a result, the value of the tap coefficient H output from the adder 12 gradually approaches the reference value and falls within the allowable value. As a result, the double talk monitoring circuit 20 stops sending the double talk detection signal DT.

When the transmission of the double talk detection signal DT is stopped, the saving registers 26-1 to 26-n resume the shift of the tap coefficient H. However, the storage register 26-n keeps its input end closed until the contents of the processing register 25 at the time when the transmission of the double talk detection signal DT is stopped are supplied.
The contents of the H register 3b are not updated until a correctly updated tap coefficient comes. In this way, the state returns to the state before double talk is detected, and the H register 3
The updated tap coefficient H is supplied to b.

As described above, even when double talk is detected, the tap coefficient several samples before that is held in the H register 3b, so that the erroneous learning is not adversely affected. When the double talk state ends, the tap coefficient H reflecting the learning result is transferred to the H register 3b again.

C: Modification In the embodiment, the allowable error is stored in the memories M1 to Mn. However, the allowable error may be stored in the double talk monitoring circuit 20.

In the above-described embodiment, double talk is detected by comparing with the impulse response characteristics in the memory. However, other detection methods can be used. For example, double talk may be determined when the value of ΔH greatly deviates from a normally expected range or when any signal occurs during the delay time.

The latter method will be described with reference to FIG. Assuming that the impulse response detected after the start of the call has a delay time DT1 as shown in FIG. 4A, if the echo path estimation circuit 3a operates normally, No signal should occur. However, if the tap coefficient H having an error is calculated by erroneously detecting the double talk, the pseudo echo generation circuit 3c
, An unnecessary echo is generated. As a result, as shown in FIG. 4B, a signal is detected even within the delay time DT1. Therefore, the delay time is measured and stored in advance, and if any signal is detected during this time, it can be determined that double talk has occurred. Other methods are also optional, but the point is that it is only necessary to determine that double talk has occurred when a state that cannot normally occur appears.

As is clear from the description of the above embodiment,
According to the present invention, a value with a large deviation is not transferred to the H register 3b. For this reason, the requirement for the stability of the arithmetic unit 10 against erroneous learning is reduced, and the learning speed of the echo path can be set correspondingly faster. That is, α in Equation 4 or g in Equation 11 is set large, and
By increasing the correction amount of the tap coefficient H, it is possible to increase the following speed of the tap coefficient H. Then, the learning speed is set sufficiently high, and the memories M1 to M1 are set.
If the accuracy of the reference value based on Mn is increased, double talk can be detected more quickly.
Even if the number of −1 to 26-n is reduced, a value that is largely shifted to the H register 3b is not transferred. Further, in a situation where the learning speed is sufficiently high and the reference value accuracy is high, if there is no particular problem in the usage environment, the saving registers 26-1 to 26-n are not provided, and the tap coefficient immediately after the double talk detection is performed. H may be held in the H register 3b, and the double talk may be used continuously during detection.

In the embodiment, a plurality of memories for storing the impulse response characteristics are provided. However, if the hybrid to be used is specified to some extent and the characteristics are almost the same, one memory may be used. .

Although the above-described embodiment is an embodiment in which the present invention is applied to signal transmission between a mobile telephone and a fixed telephone, the application of the present invention is not limited to this. The present invention is applicable to all communication networks that perform signal transmission between them.

[0057]

As described above, the present invention provides a plurality of patterns.
Storing the impulse response characteristics of the echo path in storage means,
The tap coefficient output by the echo path estimating means and the memory
Compare all impulse response characteristics of multiple patterns in the stage
By comparing, the presence or absence of double talk is determined
In this way, a more reliable
Bulltalk detection can be performed (claims 1 and 2).

Further , according to the present invention, an echo path estimating means is provided.
Tap coefficients to store the delayed tap coefficients by delaying them by multiple stages
A delay storage unit is provided. This tap coefficient delay storage means
Before and after a double talk is determined
Provides better tap coefficients for similar echo generation means
The present invention is concerned with erroneous learning.
Spurious echoes without using tap coefficients
(Claim 3).

[Brief description of the drawings]

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

FIG. 2 is a block diagram showing a configuration of a main part of the device shown in FIG.

FIG. 3 is a graph showing a typical impulse response of an echo path.

FIG. 4 is a graph showing the impulse response of an echo path with a delay.

FIG. 5 is a block diagram showing a configuration of a conventional echo canceller.

[Explanation of symbols]

REFERENCE SIGNS LIST 1 echo canceller 2 hybrid circuit 3 a echo path estimating circuit 3 b H register 3 c pseudo echo generating circuit (pseudo echo generating means) 4 controller 5 adder 6 nonlinear processing circuit 10 operation unit (echo path estimating means) 20 double talk monitoring circuit (determining means) 25) Processing register 26-1 to 26-n Storage register (tap coefficient delay storage unit) M1 to Mn memory (storage unit)

──────────────────────────────────────────────────続 き Continued on the front page (58) Field surveyed (Int.Cl. ⁶ , DB name) H04B 3/23

Claims (3)

(57) [Claims] A double talk detecting apparatus applied to an echo canceller provided in a telephone line network for transmitting between voice transmitted through four lines and voice transmitted through two lines, wherein an echo path is estimated by a learning identification method. And respond to this
Echo path estimating means for outputting the tap coefficients obtained , impulse response characteristics of the echo paths of a plurality of patterns,
For each of the multiple patterns of impulse response characteristics,
Storage means for storing the predetermined tap coefficient allowable error in association with each other , the tap coefficient output by the echo path estimating means,
For all impulse response characteristics stored in the storage means,
Depending on whether the tap coefficient allowable error is exceeded,
A double-talk detection device, comprising: a determination unit that determines the presence or absence of double-talk. 2. An echo canceller provided in a telephone line network for transmitting between voice transmitted through four lines and voice transmitted through two lines, an echo path is estimated by a learning identification method, and a tap coefficient corresponding to the echo path is estimated. Echo path estimating means for outputting a pseudo echo by a convolution operation based on the tap coefficient ; impulse response characteristics of a plurality of echo paths;
For each of the multiple patterns of impulse response characteristics,
Storage means for storing the predetermined tap coefficient allowable error in association with each other , the tap coefficient output by the echo path estimating means,
For all impulse response characteristics stored in the storage means,
Depending on whether the tap coefficient allowable error is exceeded,
Determining means for determining the presence or absence of double talk, and when the determining means determines that double talk,
Control means for stopping output of tap coefficients of the echo path estimating means. 3. The tap coefficient output by the echo path estimating means can be stored after being delayed by a plurality of stages ,
Is determined to be double talk,
Holds the self-stored contents of the point, and the last stage of the plurality of stages
The tap coefficient stored in the pseudo echo generating means
3. The echo canceller according to claim 2 , further comprising a tap coefficient delay storage unit for providing the echo canceller.