JP3304609B2 - Echo canceller learning method - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method for learning an echo canceller 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 a two-wire type, and an intermediate long-distance transmission portion is used for signal amplification or the like. , Transmission and reception are independent. Mobile phone (or cellular horn, ce
mobile phones using mobile phones,
The subscriber line of the analog telephone is a two-wire type, and a part from a relay station for receiving a signal of the mobile telephone to an exchange or the like is a four-wire type. In these cases, a hybrid circuit for performing 4-wire / 2-wire conversion is provided at a connection portion between the 2-wire system and the 4-wire system.

[0003] Hybrid circuits are designed to match the impedance of a two-wire circuit. However, since the impedance of the subscriber line is various, it is difficult to always obtain perfect matching. For this reason, a part of the signal input to the 4-wire input terminal of the hybrid circuit leaks to the 4-wire output terminal, and a so-called echo is generated. Although the level of the echo is lower than the transmission level, if the signal returns to the transmitter after a delay, the call quality is greatly reduced and a call failure occurs. Therefore, an echo canceller is used as a device for blocking echo.

FIG. 5 shows a two-wire / echo system provided with an echo canceller.
FIG. 2 shows a hardware block diagram of a four-wire conversion circuit. As shown in FIG. 5, the echo canceller 1 is provided before the hybrid circuit 2. A subscriber of a normal analog telephone is called a near-end talker, and a subscriber of a mobile telephone or the like 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 near-end voice signal is output from the echo canceller 1. The near-end audio signal is denoted by Sout.

[0005] The echo canceller 1 performs echo path estimation /
A pseudo echo generation circuit 3, a control circuit 4, an addition circuit 5, and a non-linear processing circuit 6 are provided. The echo path estimation / pseudo echo generation circuit 3 generates the same signal as the echo generated in the near-end audio signal Sin using the far-end audio signal Rout. This signal is called a pseudo echo. The pseudo echo is subtracted from the near-end audio signal Sin by the adding circuit 5. The echo path estimation / pseudo echo generation circuit 3 performs learning for canceling the echo using the output signal Rout to the hybrid circuit 2 and the residual echo that cannot be canceled by the addition circuit 5.

[0006] Digital signals are transmitted on the transmission line on the far-end talker side. D / A conversion (generally, μ-LAW conversion) is performed between the echo canceller 1 that performs digital signal processing and the hybrid circuit 2 that performs conversion to an analog line. Therefore, the relationship between the far-end audio output Rout and the near-end audio input Sin is not linear. Therefore, the echo cannot be completely canceled only by performing the linear operation by the echo path estimation / pseudo echo generation circuit 3.

Therefore, a non-linear processing circuit 6 is provided to cancel the residual echo that cannot be canceled and remains. When only the far-end speaker is in the transmitting state, the near-end voice output Sout
Is constituted only by an echo, the nonlinear processing circuit 6 performs switching so as to block transmission of the near-end audio output Sout, or replaces the near-end audio output Sout with pseudo noise. The operation of the nonlinear processing circuit 6 is performed based on the control from the control circuit 4.

[0008]

It is necessary to stop echo path learning when there is no sound at the far end or during double talk. However, when these states are detected, an error may occur in the detection timing. Such an error causes abnormal noise due to erroneous learning. In order to suppress the influence of the error, it is necessary to set a learning constant so that the designated value changes slowly. However, this is nothing but a reduction in the learning speed.

The decrease in the learning speed causes a problem that the echo immediately after the change of the characteristics of the communication line is rather large. In particular, when a new call is set up, the echo path greatly changes, so that there is a problem that the echo generated on the far-end talker side increases until learning of the echo canceller converges.

[0010] The speech disturbance caused by the echo becomes more pronounced as the signal propagation time becomes longer. In particular, a call through a communication satellite has a large delay and an echo is a serious problem. In the case of mobile communication using a mobile phone, high-efficiency encoding processing may be performed for communication between the base station and the mobile device or between stations, and a lot of switching processing is performed in the relay station. The delay was large and echo was a major problem.

Accordingly, an object of the present invention is to provide an echo canceling method capable of performing high-speed learning when the characteristics of a communication line changes and canceling the echo with high stability. . In particular, an object of the present invention is to provide an echo canceling method capable of effectively reducing the echo of the far-end talker voice generated when a new call is set up.

[0012]

Means to solve the Problems] To achieve the above object, a first aspect of the present invention, in the communication line connected to the far-end talker and near-end talker side, the near of the echo canceller to cancel the arising echo by the connected hybrid to-end talker side, the echo canceler learning method for learning to improve the echo cancellation capability,
A first learning method performed when a call is made from the far-end speaker side, and a second learning method performed when a call is made from the near-end speaker side, wherein the echo canceller is connected. Based on the information given to the echo canceller from the exchange, the first learning method is performed when the call is made from the far-end speaker, and when the call is made from the near-end speaker.
Is characterized by using the second learning method.

[0013] According to a second aspect of the invention, in <br/> echo canceler learning method according to claim 1, wherein, when the is <br/> et calling or near end talker side, the near end speaker using a ring back tone and an echo of the ring back tone is transmitted to the finisher side, and performs the learning.

According to a third aspect of the present invention, there is provided the echo canceller learning method according to the first aspect, wherein a call originated from the far-end talker is set. , by using the echo of the training signal and the training signal is transmitted to the near-end talker side machine from the exchange of the echo canceller is connected, and performs the learning.

Further, according to the present invention, in the communication line connected to the far end talker side and the near end talker side, the near end talk
An echo canceller for canceling the echo generated by the hybrid connected to the user side, in the echo canceller learning method for learning to enhance the echo canceling ability, wherein the near-end is determined based on a signal given from an exchange to which the echo canceller is connected. When it is determined that the call is originated from the speaker side, the learning is performed using the ringback tone transmitted to the near end speaker side and the echo of the ringback tone, and the echo canceller is connected. A training signal transmitted from the exchange to the near-end talker when it is determined based on a signal given from the exchange that a call originated from the far-end talker has been set. The learning is performed by using an echo of a training signal.

According to a fifth aspect of the present invention, there is provided the echo canceller learning method according to the fourth aspect, wherein it is determined that a call is made from the near-end talker and a call is made from the far-end talker. When it is determined that a call has been set, the parameters used for the learning are set to increase the learning speed, and after a certain period of time, the parameters used for the learning are set to increase the accuracy of the learning. It is characterized by the following. Further, in the invention according to Claim 6, in communication line connected to the far-end talker and near-end talker to cancel the echo caused by the hybrid is connected to the near end talker In the echo canceller, a first learning means for executing a first learning method when a call is made from the far-end talker, and a second learning method when the call is made from the near-end talker. A second learning unit to execute, and a selection unit that selects and operates the first learning unit or the second learning unit based on information given to the echo canceller from an exchange connected to the echo canceller. It is characterized by doing.

[0017]

According to the first aspect of the present invention, a first learning method or a fixed telephone which is performed when a call is originated from a mobile telephone based on information given to the echo canceller from an exchange to which the echo canceller is connected. A second learning method to be performed when a call is made from is selected. Therefore, the most appropriate learning method can be used in each case.

According to the second aspect of the present invention, when a call is made from a fixed telephone, learning is performed using a ringback tone and an echo of the ringback tone transmitted to the fixed telephone. Therefore, learning can be converged before a call is set up.

According to the third aspect of the present invention, when a call originated from a mobile telephone is set, a training signal and a training signal transmitted from the exchange connected to the echo canceller to the fixed telephone are set. The learning is performed using the echo of. Therefore, the learning can be converged before the far-end speaker performs transmission.

According to the fourth aspect of the present invention, when a call is made from a fixed telephone, learning is performed using the ringback tone and the echo of the ringback tone transmitted to the fixed telephone, and the mobile telephone moves. When a call originating from a telephone is set, learning is performed using a training signal transmitted from the exchange to the fixed telephone and an echo of the training signal. Therefore, when a call is made from a fixed telephone, the learning converges before the call is set up, and when a call is made from a mobile telephone, the learning converges before the far-end talker starts transmitting. I do.

According to the fifth aspect of the present invention, when it is determined that a call originated from a fixed telephone and when it is determined that a call originated from a mobile telephone has been set, the echo canceling capability is provided at a high convergence speed. Learning to improve,
After a certain period of time, learning is performed with high stability.

[0022]

【Example】

(Embodiment 1) FIG. 1 shows an echo canceller apparatus embodying the present invention. 1, the same constituent blocks as those in FIG. 5 are denoted by the same reference numerals as those in FIG. Echo path estimation /
The pseudo echo generation circuit 3 detects a response characteristic of the hybrid circuit 2 based on the far-end voice input Rin and the residual echo, and estimates an echo path equivalent to the hybrid circuit 2. Next, the same pseudo echo as the echo generated in the hybrid circuit 2 is generated by a convolution operation of the estimation result and the far-end voice signal Rin. A signal line for transmitting a signal indicating that a call has been made by a near-end speaker and a border signal for indicating that a call has been set are connected from the exchange 13 to the echo path estimation / pseudo echo generation circuit 3.

The adder 5 cancels the echo by subtracting the pseudo echo from the near-end voice input Sin. The control device 4 controls whether the learning of the echo path estimation / pseudo echo generation circuit 3 is performed or not in addition to the control of the nonlinear processing circuit 6. When the far-end speaker is in a non-transmission state, or when the far-end speaker and the near-end speaker are talking at the same time, the echo path estimation / pseudo-echo generation circuit 3 may learn by mistake. Then, the learning of the echo path estimation / pseudo echo generation circuit 3 is stopped.

In this embodiment, a learning identification method is used as a learning method of the echo path estimation / pseudo echo generation circuit 3. The learning identification method has a simpler algorithm and a smaller amount of calculation than other methods, and has relatively good echo canceling characteristics. The echo canceller is composed of a digital circuit. Let T be the sampling time interval of the echo canceller.
Assuming that the signal propagation characteristic of the echo path is linear, and using the impulse response h (j) and the input signal x (t) at the time tT, the echo yimp (k) at the time kT is expressed by the following equation.

[0025]

Yimp (k) = Himp X (k) ^t (1) ( ^t is transposition of vector) Himp = (h (1), h (2),..., H (n)) (2) X (k) = (x (k-1), x (k-2),..., X (kn)) (3)

If the estimated value of Himp at time kT is Hpre (k), the estimated value y of the echo at time kT is y
pre (k) is

Ypre (k) = Hpre (k) X (k) '(4)

Is given by In addition, the successive correction of Hpre (k)

Hpre (k + 1) = Hpre (k) + αe (k) X (k) / X (k) X (k) ′ (5) (X (k) X (k) ′> nε ² Hpre (k + 1) = Hpre (k) (6) (when X (k) X (k) ′ ≦ nε ² ).

Here,

E (k) = y (k) −ypre (k) (where y (k) is an actually measured value of Sin) (7). e (k) is a residual echo remaining without being canceled at time kT. ε is the maximum value of x (t), 1
Choose a size of about ５〜 to 1/10. Hpre (k + 1)
Is used to generate a pseudo echo ypre (k + 1) at the next sample time (k + 1) T.

The following conditions are required for learning the echo path estimation / pseudo echo generation circuit 3. (1) There is a far-end audio output Rout at such a level that the echo returns as the near-end audio input Sin;
In other words, the far-end speaker is transmitting. (2) The near-end voice input Sin is composed of only an echo (or only an echo and white noise), in other words, the near-end speaker is not transmitting.

When a new call is set up, the characteristics of the echo path change much more than during a call, so that the residual echo e (k) increases. For this reason, it is desirable that the far-end speaker performs learning before starting transmission. However, according to the conventional method, the echo path estimation / pseudo-echo generation circuit 3 may learn by mistake when the far-end speaker is in a non-transmission state even when a new call is set. You have to stop. Therefore, in this embodiment, when a new call is set, the echo path estimation / pseudo-echo generation circuit 3 is learned by using a ringback tone or the like before the far-end talker performs transmission.

FIG. 2 shows a communication sequence when a near-end talker makes a call. 2, the same components as those in FIG. 5 or FIG. 1 are denoted by the same reference numerals. Reference numeral 11 denotes a far-end mobile telephone, 12 denotes a base station in a wireless zone where the mobile telephone 11 is located, 13 denotes an exchange to which the echo canceller 1 is connected, and 16 denotes a near-end fixed telephone. is there.
When a call is made from the near-end talker's fixed telephone 16, the call loop on the near-end talker side is closed before the call is set up, and the communication line on the near-end talker side is sent to the echo canceller 1. Connected. A ring back tone is transmitted to the fixed telephone 16 through the connected communication line, and the hybrid circuit 2
Returns the echo of the ring back tone to the echo canceller 1.

The switch 13 can determine that a call has been made by a near-end speaker according to a well-known protocol.
The exchange 13 supplies a signal indicating that a call has been made from the near-end talker to the echo canceller 1. Then, the echo path estimation / pseudo echo generation circuit 3 performs learning using the ringback tone and its echo. The time required for the learning to converge is usually sufficiently shorter than the time during which the ringback tone is transmitted, so that the learning can be completed before the call is set up. Note that a ringback tone suitable for estimation of an echo path may be used instead of a conventionally used ringback tone.

FIG. 3 shows a communication sequence when a call is made from the mobile telephone 11 on the far end talker side. At the time of calling from the far-end speaker, since the communication loop on the near-end speaker side is not formed, the echo path estimation / pseudo-echo generation circuit 3 cannot be learned using the ringing tone. When a call is made from a far-end speaker, a call loop on the near-end talker side is formed only when the near-end talker goes off-hook and a call is set up.

The boundary where a call is set, that is, the call border, is determined by the exchange 13 to which the echo canceller 1 is attached.
Can be determined according to a known call control protocol. When the switch 13 detects a call border when a call is made from the far-end talker, it provides a signal indicating the call border to the echo canceller 1. As a result, the echo canceller 1 transmits the training signal of the voice band to the near-end speaker. Then, the echo of the training signal is returned from the hybrid circuit 2 to the echo canceller 1. The echo path estimation / pseudo echo generation circuit 3 can learn the echo estimation path using the training signal and its echo.

The time from the call border to the time when the near-end speaker touches the receiver is usually sufficiently long as compared with the time for converging the learning of the echo estimation path. During the period, the learning is converged and the transmission of the training signal can be terminated.

(Embodiment 2) By changing the parameters used in the learning algorithm, the learning speed of the echo path estimation / pseudo echo generation circuit 3 can be increased when the characteristics of the echo path greatly change. In this embodiment,
The parameter α in the learning identification method used in the first embodiment is changed. Therefore, first, the effect of a change in α on learning will be described.

When α = 1, using equation (5),

Hpre (k + 1) X (k) '= Hpre (k) X (k)' + e (k) = ypre (k) + e (k) = y (k) (8) . That is, when α = 1, Hpre (k + 1) becomes Hpre
Make (k + 1) X (k) 'equal to y (k). If the pseudo echo is calculated using Hpre (k + 1) at time kT, the pseudo echo becomes equal to the actually measured echo, and the residual echo e
(k) becomes zero.

Therefore, in an ideal case where there is no instantaneous fluctuation of line characteristics, line noise, or near-end speaker's voice that is not detected, α = 1 is most preferable because learning can be performed quickly. However, because of these fluctuations and noise,
If α = 1, Hpre greatly fluctuates due to instantaneous fluctuations in line characteristics and line noise. If Hpre is erroneously corrected, ypre (k) becomes too small to cancel the echo, or conversely, ypre (k) becomes too large to add an echo whose polarity is inverted.

The following ERLE (k) is used as an index indicating the high canceling ability of the echo canceller at time kT.
(Echo return loss enhancement).

ERLE (k) = 10 log 10 (power of yreal (k) / power of ereal (k)) (9)

In the above equation, yreal (k) and ereal (k) represent the noise component v from the echo y (k) and the residual echo e (k), respectively.
It is the quantity excluding (k) and is given by the following equation.

[Mathematical formula-see original document] yreal (k) = y (k) -v (k) ereal (k) = e (k) -v (k) Also, S of the true echo yreal (k) and the noise v (k) The / N ratio is defined by the following equation.

S / N (k) = 10 log 10 (power of yreal (k) / power of v (k)) (10)

If α = 1, ypre (k + 1) undergoes the same level of fluctuation as the sum of noise v (k) and echo ereal (k) in the previous step. Therefore, the same level of noise as the noise in the previous step occurs in the pseudo echo of each step. Therefore, as time elapses after learning starts, the average value of ERLE (k) of each step (hereinafter simply referred to as E
RLE) is the average value of S / N (k) (hereinafter simply referred to as S / N (k)).
N). As the ERLE or S / N, for example, each ERLE (k), S / N (k) for 10 ms
Can be considered.

The variation of ereal (k) at each step is
Since it is much smaller than the fluctuation of the noise component v (k), if α is reduced and the influence of the noise v (k) in the previous step is reduced, the canceling ability of the echo canceller increases, and the ERLE can be further increased. . But,
As α decreases, the amount of correction of Hpre (k) also decreases,
The learning speed of the echo path estimation / pseudo echo generation circuit 3 is reduced.

When ERLE is smaller than S / N by a certain degree or more, α = 1, and ERLE approaches S / N or S / N
If α gradually decreases as / N becomes larger,
The canceling ability can be increased without significantly reducing the learning speed of the echo canceller. However, true echo y
Since real (k) and noise component v (k) cannot be measured separately, ERLE and S / N cannot be calculated from actual measurement values. Therefore, α is changed as follows using a value having a correlation with ERLE.

(1) When the characteristics of the echo path change, ERLE becomes small. Here, the change in the characteristics of the echo path when a new call is set is much larger than the change in the characteristics of the echo path during a call. When a call is set up, that is, the call border can be determined by the information given from the exchange 13, so that α is set to a value close to 1 for a certain period of time from the call border, and the value of α is gradually reduced. For example, by changing as follows, the canceling ability can be increased without lowering the learning speed when the characteristics of the echo path change.

Α = 0.9 (t ≦ T0) (T0 is a time constant) (11) α = 0.5 (T0 <t ≦ T1) (T1 is a time constant) (12) α = 0.1 ( T1 <t) (13)

When the far-end speaker is in a non-transmission state, or when the far-end speaker and the near-end speaker are talking at the same time, the echo path estimation / pseudo-echo generation circuit 3 may learn by mistake. Therefore, the echo path estimation / pseudo echo generation circuit 3
You have to stop learning. Equation (11)-
As the time t in (13), the time during which only the far-end speaker after the call border speaks is used. Alternatively, when a call is made from a far-end speaker as described in the first embodiment, a training signal for learning may be transmitted with t = 0 when the near-end speaker goes off-hook. Further, when a near-end talker makes a call, t = 0 may be set when the ringback tone starts to be transmitted, and learning may be performed using the ringback tone.

When the near-end talker is transferred during a call, for example, the characteristics of the echo path change and the ERLE rapidly decreases. In such a case, it is necessary to increase the learning speed of α. Therefore, α may be changed based on the equations (11) to (13), where t = 0 even when ERLE suddenly decreases, and t is the time during which only the far-end speaker speaks thereafter.

(2) ERLE can be calculated by substituting the standard noise component magnitude for v (k). By using this ERLE and changing α as in the following equation 10, for example, the canceling ability can be increased without significantly lowering the learning speed. here,
Even when α is set by using ERLE, it is basically preferable to increase α at the start of a call and gradually reduce α. That is, the smallest one of α obtained in the past according to Equation 10 may be set as the actually used α.

Α = 0.9 (ERLE ≦ 10 dB) α = 0.5 (10 dB <ERLE ≦ 20 dB) α = 0.1 (20 dB <ERLE)

(3) The progress of learning is based on the residual echo e
It can also be determined from the average value e of (k). The average value is used because the level of the residual echo e (k) in each step is greatly influenced by the transmission level x (k) and the noise v (k) and is unstable. The average value e can be obtained from each residual echo e (k) during, for example, 10 ms. Using the average value e of the residual echo, α can be switched as follows. α = 0.9 (e ≦ 10 dB) α = 0.5 (10 dB <e ≦ 20 dB) α = 0.1 (20 dB <e)

(4) Instead of changing α stepwise, as shown in FIG. 4, it is also possible to continuously change α with respect to time t, ERLE, or residual echo e. In this case, α is represented by the following equation. (4-1) When the time t is used as a parameter, the time constant T
Using 0,

Α = 0.8 (1−t / T0) +0.1 (t <T0) α = 0.1 (T0 ≦ t) or α = 0.9exp (−t / T0)

(4-2) When ERLE is used as a parameter, a constant E0 is used to

Α = 0.8 (1−ERLE / E0) +0.1 (ERLE <T0) α = 0.1 (T0 ≦ ERLE) or α = 0.9exp (−ERLE / E0)

(4-3) When the residual echo e is used as a parameter, a constant A0 is used

Α = 0.8 (1−e / A0) +0.1 (e <A0) α = 0.1 (A0 ≦ e) or α = 0.9exp (−e / A0)

(5) In any of the methods (2) to (4), similarly to the first embodiment, when the far-end speaker calls, the training signal is output when the near-end speaker goes off-hook. You may transmit and learn. When a near-end talker makes a call, learning may be performed using a ringback tone and its echo.

In this embodiment, α is changed as a parameter of the learning method. However, it is also possible to change other parameters that affect the learning speed by using the fact that fast learning is required immediately after the call border and high-precision learning is required thereafter. Even when other learning algorithms are used in addition to the learning identification method, the parameters used in those algorithms can be changed in the same manner as in the present embodiment.

(Embodiment 3) In addition to the learning identification method, for example, an RLS (Recursive Le
ast Square) can be used. Since the RLS algorithm is well known, the description is omitted. The RLS has a faster convergence speed than the learning identification method, but has a drawback that the amount of calculation is large because an inverse matrix calculation is required for the calculation. Therefore,
When prompt estimation of the echo path is required at the start of a call, the RLS algorithm is used, and when the estimation has progressed to some extent, the method is switched to the learning identification method. This makes it possible to quickly estimate an echo path while suppressing an increase in the amount of calculation of the entire exchange.

Since the computational power of the general-purpose DSP at present is the computational power required for the learning identification method <the computational power of the DSP <the computational power required for the RLS, the DSP which is high in speed for the initial input after the start of the call. By performing calculation by RLS using a plurality of DSPs and switching to the learning identification method halfway using the obtained learning results, it is possible to quickly estimate an echo path while effectively utilizing resources in the entire switch. .

The switching of the learning method can be performed using time t, ERLE, or echo level e, for example, as in the case of switching α, for example, as follows.

[Equation 14]

After switching to the learning identification method, the value of α can be further changed by using the method described in the second embodiment. As in the first embodiment, when a call is made from a far-end speaker, a training signal may be transmitted to perform learning when the near-end speaker goes off-hook. Further, when a near-end speaker makes a call, learning may be performed using a ringback tone.

(Embodiment 4) In the method described in the embodiment 2 or 3, since the noise component v (k) is mixed in the echo, it is accurately determined how far the learning of the echo canceller has advanced. Can not do. Then, true echo y
In order to measure a value close to real (k), a method of analyzing the voice spectrum of x (k) and measuring the level of the echo y (k) with respect to the obtained spectrum component can be considered. Also,
If the echo level is not uniform with audio frequency,
It is necessary to analyze the echo spectra of x (k) and y (k).

The speech spectrum analysis methods are roughly classified into non-parametric analysis (NPA) and parametric analysis (PA). Non-parametric analysis includes (1) short-time autocorrelation analysis, (2) short-time spectrum analysis, (3) cepstrum analysis, (4) bandpass filter bank analysis, and (5) zero-crossing number analysis. On the other hand, as parametric analysis, (1) analysis by synthesis (A-b
-S method), (2) linear prediction analysis (LPC), (3) maximum likelihood spectrum estimation (autocorrelation method, inverse filter method), (4)
Non-variance method, (5) PARCOR analysis, (6) LSP analysis. As in the case of switching the learning method in the second embodiment,
Using the time t, ERLE, or echo level e, the spectral analysis method used can be switched.

[0060]

As is apparent from the above description, according to the first aspect of the present invention, the first learning method performed when a call is made from a mobile telephone or the case where a call is made from a fixed telephone. Since the second learning method to be performed is appropriately selected, a learning method most suitable for each sequence can be used.

According to the second aspect of the present invention, when a call is made from a fixed telephone, learning is performed using the ringback tone and the echo of the ringback tone transmitted to the fixed telephone. Learning converges before the call is set up. Therefore, it is possible to effectively reduce the echo generated immediately after the call originated from the fixed telephone is set.

According to the third aspect of the present invention, when a call originated from a mobile telephone is set, a training signal and a training signal transmitted from the exchange connected with the echo canceller to the fixed telephone are set. Since the learning is performed using the echo of, the learning is converged before the far-end speaker transmits. Therefore, it is possible to effectively reduce the echo generated immediately after the call originated from the mobile telephone is set.

According to the fourth aspect of the present invention, when a call is made from a fixed telephone, learning is performed using a ringback tone and its echo, and a call made from a mobile telephone is set. In this case, learning is performed using the training signal transmitted from the exchange to the fixed telephone and its echo, so if a call is made from a fixed telephone, the learning converges before the call is set up, and the When a call is made, the learning converges before the far-end speaker transmits. Therefore, in any case, the echo generated immediately after the call is set up can be effectively reduced.

According to the fifth aspect of the present invention, when it is determined that a call originated from a fixed telephone and when it is determined that a call originated from a mobile telephone has been set, the echo canceling capability is increased at a high convergence speed. Learning to improve,
Learning is performed with high convergence accuracy after a certain time has elapsed.

According to the present invention, the learning speed of the echo canceller when a new call is set is increased, and the learning accuracy in other cases can be improved. Therefore, echoes at the time of setting a call and at the time of a call can be effectively reduced. Further, according to the present invention, the echo is effectively reduced, so that it is possible to improve the disconnection of the talk head by the nonlinear processing circuit 6.

[Brief description of the drawings]

FIG. 1 is a hardware block diagram showing a configuration of an echo canceller according to the present invention.

FIG. 2 is a sequence diagram showing a learning sequence when a near-end talker makes a call.

FIG. 3 is a sequence diagram showing a learning sequence when a far-end talker makes a call.

FIG. 4 is a graph showing a continuously changed alpha.

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

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Echo canceller 2 Hybrid circuit 3 Echo path estimation / pseudo echo generation circuit 4 Controller 5 Adder 6 Non-linear processing circuit 11 Mobile telephone 12 Base station 13 Exchange 16 Telephone

 ──────────────────────────────────────────────────続 き Continuation of the front page (56) References JP-A-5-14246 (JP, A) JP-A-4-207824 (JP, A) JP-A-57-3441 (JP, A) JP-A-63-1988 236422 (JP, A)

Claims (6)

(57) [Claims] In 1. A far-end talker side and the communication line connected to the near-end talker, the arising echo by the connected hybrid <br/> to the near-end talker An echo canceller learning method for improving the echo canceling ability of an echo canceller to cancel, a first learning method performed when a call is made from the far end speaker, and a call made from the near end speaker and a second learning method performed when it is, on the basis of the exchange which echo canceller is connected to the information given to the echo canceller, the far-end talker
The first learning method when it is calling from Shah side, front
An echo canceller learning method, wherein the second learning method is used when a call is made from the near end speaker . " 2. The echo canceller learning method according to claim 1, wherein when a call is made from the near-end speaker, a ring-back tone transmitted to the near-end speaker and the ring-back tone. An echo canceller learning method, wherein the learning is performed using the echo of (1). 3. The echo canceller learning method according to claim 1, wherein a call originated from the far-end talker is set.
An echo canceller learning method, wherein the learning is performed using a training signal transmitted from the exchange connected to the echo canceller to the near-end talker side device and an echo of the training signal. 4. A hybrid connected to the near-end talker in a communication line connected to the far-end talker and the near- end talker
In the echo canceller learning method of performing the learning for enhancing the echo canceling ability of the echo canceller for canceling the echo generated by the echo canceller, the near-end talker is called based on a signal given from a switch to which the echo canceller is connected. When it is determined that the echo is canceled, the learning is performed by using the ringback tone transmitted to the near end speaker and the echo of the ringback tone, and based on the signal given from the exchange to which the echo canceller is connected. When it is determined that the call originated from the far-end speaker has been set, a training signal transmitted from the exchange to the near-end speaker and an echo of the training signal are used. An echo canceller learning method, wherein the learning is performed. 5. The echo canceller learning method according to claim 4, wherein it is determined that a call originated from the near-end speaker and that a call originated from the far-end speaker has been set. In this case, the parameter used for the learning is set so as to increase the learning speed, and after a certain period of time, the parameter used for the learning is set so that the accuracy of the learning is increased. Learning method. In 6. far-end talker's side and the communication line connected to the near-end talker side, the echo canceler to cancel the echo caused by the hybrid is connected to the near end speaker side, the far-end talker First learning means for executing a first learning method when a call is made from a speaker side, second learning means for executing a second learning method when a call is made from the near end speaker side, Selecting means for selecting and operating the first learning means or the second learning means based on information provided to the echo canceller from the exchange connected to the echo canceller. .