JP5293952B2 - Signal processing method, signal processing apparatus, and signal processing program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a signal processing method, a signal processing device and a signal processing program, capable of attaining signal erasure that reduces output distortions under a low operation quantity, without delay with respect to a desired signal. <P>SOLUTION: In the signal processing method, a plurality of coefficients of an adaptive filter are collected to generate a plurality of coefficient groups, a coefficient representative value in the plurality of coefficient groups is found; and a coefficient updating step size of each coefficient in the coefficient groups is controlled by using the coefficient representative value. <P>COPYRIGHT: (C)2010,JPO&amp;INPIT

Description

  The present invention relates to a signal processing method, a signal processing apparatus, and a signal processing program, and in particular, eliminates noise or interference signals mixed in a desired signal input from a microphone, a handset, a communication path, or the like by an adaptive filter, or a desired signal. The present invention relates to a method, an apparatus, and a program for emphasizing the above.

  A voice signal input from a microphone, a handset, or the like is a target of voice encoding or voice recognition processing. In a narrow-band speech encoding device, speech recognition device, or the like with a high degree of information compression, there is a big problem in performing speech encoding and speech recognition of a speech signal on which such background noise is superimposed. As signal processing apparatuses for the purpose of eliminating the acoustically superimposed noise as described above, Patent Documents 1, 2, and 3 disclose a two-input type noise canceling apparatus using an adaptive filter.

  The two-input type noise canceller disclosed in Patent Documents 1, 2, and 3 uses an adaptive filter that approximates an impulse response of an acoustic path (noise path) from a noise source to the voice input terminal. Pseudo noise corresponding to the mixed noise component is generated. Then, the pseudo noise is subtracted from the received signal (main signal) input to the audio input terminal, thereby operating to suppress the noise component. Here, the received signal is a signal in which an audio signal and noise are mixed, and is generally supplied from a microphone or a handset to an audio input terminal. A reference signal supplied to the reference input terminal is input to the adaptive filter. Here, the reference signal is a signal correlated with a noise component in the noise source, and is captured in the vicinity of the noise source. Thus, by capturing the reference signal in the vicinity of the noise source, the reference signal can be regarded as being almost equal to the noise component in the noise source.

  The coefficient of the adaptive filter is corrected by correlating the error obtained by subtracting the pseudo noise from the received signal and the reference signal input to the reference input terminal. As an adaptive filter coefficient correction algorithm, Patent Documents 1, 2, and 3 disclose “LMS algorithm (Least Mean-Square Algorithm)” and “LIM (Learning densification method)”.

  The LMS algorithm and LIM are a kind of algorithm called a gradient method, and the speed and accuracy of coefficient update depend on a constant called coefficient update step size. The filter coefficient is updated by the product of the coefficient update step size and the error, but the desired signal included in the error interferes with the coefficient update, and in order to reduce the influence, the coefficient update step size is set to a very small value. There is a need. However, when the coefficient update step size is small, the followability of the adaptive filter coefficient to the environmental change is degraded. For this reason, an error increases or a desired signal is distorted.

  To deal with this problem, Patent Documents 1, 2, and 3 describe a noise canceller that performs adaptive control of the coefficient update step size. The noise cancellers of Patent Literatures 1, 2, and 3 include two adaptive filters, and the coefficient update of the second adaptive filter is performed using the signal-to-noise ratio in the main signal estimated using the first adaptive filter. The step size is controlled, and a signal from which noise has been eliminated by the second adaptive filter is output. By using a small coefficient update step size when the speech signal is larger than noise and using a large coefficient update step size in the opposite state, the adaptive filter can follow the environmental changes and reduce distortion in the signal after noise cancellation. can do.

  The first adaptive filter operates in the same manner as the second adaptive filter, but the coefficient update step size of the first adaptive filter is set to a value larger than the coefficient update step size of the second adaptive filter. For this reason, the output of the first adaptive filter has high followability to environmental changes, but the noise estimation accuracy is inferior to that of the second adaptive filter. The output of the first adaptive filter operating in this way is pseudo noise that follows changes well, and the difference between the main signal and the first adaptive filter can be regarded as a pseudo desired signal. Thereby, the desired signal-to-noise ratio in the main signal can be estimated based on the above signal. By applying a small coefficient update step size to the adaptive filter 2 when the ratio is a large value, a small coefficient update step size is applied to the adaptive filter 2 because a small disturbance is caused by a desired signal. It is possible to simultaneously achieve a low environmental distortion and low distortion in the signal after noise cancellation.

  The reference signal supplied to the adaptive filter 2 and the main signal used for canceling the pseudo noise that is the output of the adaptive filter are each delayed. This is to compensate for the estimated delay when estimating the desired signal-to-noise ratio using the adaptive filter 1. If this delay is not applied, the estimated value of the desired signal-to-noise ratio in the main signal is obtained later than the correct value, and correct coefficient update step size control cannot be performed. For this reason, residual noise in the signal after noise cancellation increases, and distortion further increases.

  In the description so far, it has been assumed that the reference signal is noise itself by capturing the reference signal in the vicinity of the noise source. However, in reality, it is difficult to satisfy this condition. In such a case, the reference signal is composed of noise and a desired signal mixed therein. Such a mixed component of the desired signal with respect to the reference signal is called crosstalk. Patent Document 4 discloses a configuration for simultaneously achieving sufficient environmental changeability and low distortion in a signal after noise cancellation in noise cancellation when crosstalk exists.

  Similar to noise cancellation, a third adaptive filter for controlling the coefficient update step size and a fourth adaptive filter for canceling crosstalk are introduced. Using the third and fourth adaptive filters that approximate the impulse response of the acoustic path (crosstalk path) from the desired signal source to the reference input terminal, a pseudo desired signal corresponding to the desired signal component mixed at the reference input terminal is obtained. It operates to suppress the desired signal component (crosstalk) by subtracting this pseudo desired signal from the received signal (reference signal) generated and input to the reference input terminal. The output of the third adaptive filter is a desired signal component (crosstalk) that follows the change well, and the difference between the reference signal and the third adaptive filter can be regarded as pseudo noise. Therefore, the desired signal-to-noise ratio in the reference signal can be estimated based on these signals. By applying a small coefficient update step size to the adaptive filter 4 when the ratio is a large value, the coefficient update step size due to noise is large, so that a small coefficient update step size is applied. Thus, sufficient followability to environmental changes and low distortion in the signal after crosstalk elimination can be achieved at the same time.

  In the configuration disclosed in Patent Document 4, “the desired signal-to-noise ratio in the main signal” and “reference” in which the coefficient update step sizes of the first and third adaptive filters are further approximated by the ratio of the main signal and the reference signal. The desired signal-to-noise ratio in the signal is used for control. By controlling the coefficient update step sizes of the first and third adaptive filters in this way, the input signal having a wide range of desired signal-to-noise ratios in the main signal and the reference signal has high followability to environmental changes. In addition, noise cancellation with less output distortion can be achieved.

  On the other hand, non-patent document 1 is intended for the elimination of acoustic echo generated when a reproduction signal from a speaker wraps around to a microphone and circuit echo generated when a transmission signal wraps around from a transmission side to a reception side in two-wire / four-wire conversion of a communication line. Discloses a “PNLMS algorithm (Proportionate Normalized Least Mean-Square Algorithm)”.

  The PNLMS algorithm achieves high-speed convergence by performing coefficient update using a coefficient update step size corresponding to the filter coefficient value. However, the PNLMS algorithm does not discuss the distortion in the signal after echo cancellation at all and does not mention its existence.

  The PNLMS algorithm inherently suppresses coefficient updates on small coefficients and prevents the more important large coefficient updates from being disturbed by disturbances caused by uncertain updates of small coefficients based on the statistical nature of the signal. . For this reason, when the adaptive filter coefficient as a whole is evaluated, a faster convergence is achieved.

However, it can be understood that the PNLMS algorithm does not exhibit desirable characteristics from the perspective of following environmental changes. For example, when a change occurs in a coefficient having a small value, it takes time to follow the change because the coefficient update step size applied to this coefficient is small. For this reason, when the adaptive filter coefficient is evaluated as a whole, a delay in following the environmental change is caused and distortion in the output signal is caused.
JP-A-8-241086 JP-A-10-215193 JP 2000-172299 A WO2005 / 024787 Donald L. Dutweiler, “Proportionate Normalized Least-Mean-Square Adaptation in Echo Cancers,” IEEE TRANSACTIONS ON SPEECH AND AUDIOSING. 8, NO. 5,2000, pp. 508-518

  As described above, in the configuration disclosed in Patent Document 1, the main signal is delayed with respect to the second adaptive filter. This represents that the noise-suppressed desired signal that is the output signal of the noise suppression device is delayed. In two-way communication, a call becomes unnatural due to a delay of a desired signal. Moreover, in order to control the coefficient update step size of the second adaptive filter, the first adaptive filter is required, and the required calculation amount increases. Similarly, in the configuration disclosed in Patent Document 4 that can handle crosstalk, the main signal after noise cancellation is delayed.

  Moreover, in order to control the coefficient update step sizes of the second and fourth adaptive filters, the first and third adaptive filters are required, and the required calculation amount is further increased. Further, the PNLMS algorithm disclosed in Non-Patent Document 1 has a problem that followability to environmental changes is insufficient and causes output distortion.

  An object of the present invention is to provide a signal processing method, a signal processing apparatus, and a signal processing program capable of realizing signal erasure with no delay with respect to a desired signal, low calculation amount, and low output distortion.

Signal processing method of the present invention, the adaptive coefficients of the filter sequentially plurality collectively generate a plurality of coefficients groups, using the coefficients of the adaptive filter in said plurality of coefficient groups, it obtains a coefficient representative value of each coefficient group, The coefficient update step size of each coefficient in the coefficient group is controlled using the coefficient representative value.

The signal processing apparatus of the present invention is sequentially plurality collectively coefficients of the adaptive filter to generate a plurality of coefficients groups, using the coefficients of the adaptive filter in said plurality of coefficient groups, a coefficient representative value of each coefficient group And at least a step size control circuit for controlling a coefficient update step size of each coefficient in the coefficient group using the coefficient representative value, and updating the coefficient of the adaptive filter using the coefficient update step size. Features.

Further, the signal processing program of the present invention causes a computer, and generating a plurality of coefficient groups sequentially plurality collectively coefficients of the adaptive filter using the coefficients of the adaptive filter in said plurality of coefficients groups, each coefficient group A process for obtaining the coefficient representative value, a process for controlling the coefficient update step size of each coefficient in the coefficient group using the coefficient representative value, and a process for updating the coefficient using the coefficient update step size. It is made to perform.

  In the present invention, a plurality of coefficient groups are generated by sequentially combining a plurality of coefficients of the adaptive filter, coefficient representative values in these coefficient groups are obtained, and coefficient update of each coefficient in the coefficient group is performed using these coefficient representative values. Control the step size. The coefficient update step size for each coefficient is equal to the coefficient update step size for other coefficients in the same coefficient group in the vicinity. For coefficient values belonging to coefficient groups with many large coefficient values, large coefficient values and small coefficient values The coefficient values belonging to many coefficient groups are controlled to be small values. Therefore, both the coefficient having an extremely large coefficient value and the coefficient having an extremely small coefficient value belonging to the same coefficient group are updated on average with the same coefficient update step size as the other coefficient group coefficients.

  In the present invention, the reference signal representative value in the coefficient group is obtained, and the coefficient update step size of each coefficient in the coefficient group is controlled using the reference signal representative value and the coefficient representative value. Even when the amplitude of the reference signal sample is large and the coefficient value to be updated is small, the coefficient value change amount due to the coefficient update is prevented from becoming excessive. Is not done.

  Furthermore, in the present invention, the coefficient update step size of the adaptive filter is controlled using an estimated value of the ratio of the desired signal to the erasure target signal in addition to the representative value. A small coefficient update step size is used when the desired signal that is an obstacle to the error used for coefficient update is relatively large, and a large coefficient update step size is used when the desired signal is small. In the present invention, erroneous coefficient updating is not performed even when the interference component for the error cannot be ignored. For this reason, it is possible to realize signal erasure with no delay with respect to a desired signal, a small amount of calculation, and little output distortion.

  The best mode for carrying out the present invention will be described. First, the signal processing device according to the present invention will be described with a form realized as a device for processing an audio signal, in particular, an example realized as a noise canceling device. However, it goes without saying that the signal processing devices of the following embodiments can be used as various signal processing devices other than the noise cancellation device without changing the configuration thereof.

The signal processing apparatus according to the preferred embodiment of the present invention shown in FIG. 1 includes a microphone 1 that inputs a sound, a microphone 2 that inputs a reference signal, an output terminal 7, and a main signal x _P ( The first pseudo signal is generated by inputting the first subtracter 3 that generates the first error by subtracting the first pseudo signal from k) and the reference signal x _R (k) input from the microphone 2. The first adaptive filter 5 that updates the coefficient so that the first error is minimized, and the coefficient update step size is calculated using the coefficient value received from the first adaptive filter 5, and the first adaptive filter is calculated. And a step size control circuit 8 that feeds back to the filter 5. The first error e ₁ (k) from the first subtracter 3 is also output to the output terminal 7 as an audio signal from which noise has been eliminated. Here, k is an index representing time.

Since the operation in the best embodiment is the same as that described in Patent Document 1 except for the step size control circuit 8, details are omitted. The step size control circuit 8 receives the coefficient vector W ₁ (k) from the adaptive filter 5 and obtains the coefficient update step size _{１ 1} (k). The step size control circuit 8 feeds back the obtained coefficient update step size Μ ₁ (k) to the adaptive filter 5.

W ₁ (k) and Μ ₁ (k) are given by the following equations.
<Equation 1>
W ₁ (k) = [w ₁ (0, k), w ₁ (1, k),..., W ₁ (N ₁ −1, k)]
<Equation 2>
_{１ 1} (k) = [μ ₁ (0, k), μ ₁ (1, k),…, μ ₁ (N ₁ −1, k)]
Here, N ₁ is an integer of 1 or more that represents the number of taps of the adaptive filter 5.

The step size control circuit 8 first divides the coefficient vector W ₁ (k) into L subvectors (coefficient groups). In the division, it can be divided into sub-vectors of a predetermined size. Further, it may be equally divided into sub-vectors of equal size, or may be unequally divided based on some index. As this index, for example, the variance of the coefficient absolute value can be used. The division is performed by applying a large subvector size when the coefficient absolute value variance in the same subvector is small, and a small subvector size when large. For such a relationship between the variance and the subvector size, a function or a correspondence table is determined in advance, and the subvector size is calculated according to the coefficient update.

Assuming that the sub-vector generated in this way is W _S1 (j, k), j = 0, 1,..., L−1,
W ₁ (k) = [W _S1 (0, k), W _S1 (1, k), ..., W _S1 (L-1, k)]
It becomes. In the case of L _S equal division, the j-th subvector is given by the following equation.
<Equation 4>
W _S1 (j, k) = [w ₁ (jL / L _S , k), w ₁ (jL / L _S +1, k),..., W ₁ ((j + 1) L / L _S −1, k)] (j = 0, 1, ..., L-1)
Here, L _S is an integer value such that L / L _S is an integer.

Next, a representative value of a coefficient is obtained in each subvector (coefficient group). As the representative value, a maximum value, a minimum value, a median value, an average value, a variance, or the like can be used. The representative value w _REP1 (j, k) for the j-th _subvector , when the maximum value is the representative value,
<Equation 5>
w _REP1 (j, k) = max {| w ₁ (jL / L _S , k) |, | w ₁ (jL / L _S +1, k) |, ..., | w ₁ ((j + 1) L / L _S -1, k) |}
Given in. Here, max {·} is an operator for obtaining the maximum value. Similarly, the minimum value and the median value can also be obtained by the minimum value operator or the median value operator. These operators can be realized by sorting based on the size of the elements and selecting values corresponding to maximum, minimum or center.

となる。ここに、ａｖｅ｛・｝は平均値を求める演算子である。分散についても、同様に定義に従って計算することができる。 When the average value is the representative value w _REP1 (j, k) for the j-th _subvector ,
<Equation 6>

It becomes. Here, ave {·} is an operator for obtaining an average value. Similarly, the variance can be calculated according to the definition.

  In the description so far, the coefficient representative value is obtained for the absolute value of each coefficient value. However, the same effect can be obtained by using a square value instead of an absolute value. In particular, when calculation is performed using a signal processing chip, the square value is more suitable because the calculation of the square value is easier than the calculation of the absolute value.

  FIG. 2 shows coefficient values and coefficient groups (subvectors) when L = 7 and the maximum coefficient group coefficient is selected as a representative value. The coefficient groups 1, 2, 3, and 4 are configured from a single coefficient. The coefficient groups 0, 5, and 6 are composed of a plurality of coefficients. In any coefficient group, the maximum value of the coefficient absolute value in the coefficient group is selected and becomes the coefficient representative value. These coefficient representative values are represented by square marks and can be easily distinguished from general coefficients represented by circles.

The coefficient update step size corresponding to w _REP1 (j, k) is calculated using the sub-vector representative value w _REP1 (j, k), j = 0, 1,..., L−1. To do. In the first example, the maximum value of the representative value of each subvector is obtained, the ratio of each subvector representative value to the maximum value is obtained, and the coefficient update step size μ ₀ is weighted based on this ratio to update the coefficient. This is a step size method. That is, the coefficient update step size μ ₁ (j, k) corresponding to each subvector is obtained by the following equation.
<Equation 7>
μ ₁ (j, k) = w _REP1 (j, k) / max {w _REP1 (j, k)} · μ ₀
Since the ratio takes a value between 0 and 1, the maximum coefficient is μ ₀ , and other coefficients are updated with a coefficient update step size smaller than μ ₀ .

In the second example, the average value of representative values of each subvector is obtained, the ratio of each subvector representative value to the average value is obtained, and the coefficient update step size μ ₀ is weighted based on this ratio to update the coefficient. This is a step size method. That is, the coefficient update step size μ ₁ (j, k) corresponding to each subvector is obtained by the following equation.
<Equation 8>
μ ₁ (j, k) = w _REP1 (j, k) / ave {w _REP1 (j, k)} · μ ₀

  As described above, the best mode for carrying out the present invention updates the coefficient using the coefficient update step size determined using the representative value of the coefficient subgroup to which each coefficient belongs. A small coefficient update step size is applied to a small coefficient, and a large coefficient update step size is applied to a large coefficient. For this reason, even when the interference component with respect to the error cannot be ignored, the erroneous coefficient update is not performed, there is no delay with respect to the desired signal, and it is possible to realize signal erasure with a small amount of calculation and little output distortion.

  Next, a second embodiment for carrying out the present invention will be described.

FIG. 3 is a block diagram showing a second embodiment of the present invention. Compared with FIG. 1 which is a block diagram showing the best embodiment of the present invention, the step size control circuit 8 is replaced with a step size control circuit 9. The step size control circuit 9 receives the coefficient vector W ₁ (k) and the reference signal x _R (k) from the adaptive filter 5 and the microphone 2, respectively, and obtains a coefficient update step size Μ ₁ (k). The step size control circuit 9 feeds back the obtained coefficient update step size Μ ₁ (k) to the adaptive filter 5.

In coefficient update by the LMS and NLMS algorithms, the coefficient update amount is proportional to the amplitude of the reference signal sample. Accordingly, when the amplitude of the reference signal sample is large and the coefficient value to be updated is small, the amount of change in the coefficient value due to the coefficient update is relatively large. When the interference component for the error cannot be ignored, erroneous coefficient updating is performed. It will be. From this point of view, in the second embodiment, the representative value of the reference signal power of each subvector is used in addition to the coefficient representative value used in the best embodiment. Contrary to the case of the coefficient representative value, the basic coefficient update step size μ ₀ is weighted by the reciprocal of the ratio with respect to the reference signal so that the coefficient update step size is reduced when the reference signal sample is large.

Similarly to the best embodiment, the reference signal vector X ₁ (k) of the adaptive filter 5 at time k is defined as follows.
<Equation 9>
X ₁ (k) = [x ₁ (0, k), x ₁ (1, k),..., X ₁ (N ₁ −1, k)]

となる。分散についても、同様に定義に従って計算することができる。係数値の場合と同様に、絶対値の代わりに２乗値を用いても、同様の効果を得ることができる。 Next, a representative value of the reference signal in each subvector is obtained. As the representative value, a maximum value, a minimum value, a median value, an average value, a variance, or the like can be used. The representative value x _REP1 (j, k) for the j-th _subvector , when the maximum value is the representative value,
<Equation 10>
x _REP1 (j, k) = max {| x ₁ (jL / L _S , k) |, | x ₁ (jL / L _S +1, k) |, ..., | x ₁ ((j + 1) L / L _S -1, k) |}
Given in.
Similarly, the minimum value and the median value can also be obtained by the minimum value operator or the median value operator. When the average value is a representative value x _REP1 (j, k) for the j-th _subvector ,
<Equation 11>

It becomes. Similarly, the variance can be calculated according to the definition. Similar to the case of the coefficient value, the same effect can be obtained by using the square value instead of the absolute value.

Reference signal representative value x _REP1 (j, k), j = 0, 1,..., L−1 and coefficient representative value w _REP1 (j, k), j = 0, 1, .., L-1 is used to calculate the coefficient update step size corresponding to x _REP1 (j, k) and w _REP1 (j, k). In the first example, the maximum value of the reference signal representative value and coefficient representative value of each subvector is obtained, the ratio of the reference signal representative value and coefficient representative value of each subvector to the maximum value is obtained, and these ratios are obtained. In this method, the basic coefficient update step size μ ₀ is weighted to obtain the coefficient update step size. That is, the coefficient update step size μ ₁ (j, k) corresponding to each subvector is obtained by the following equation.
<Equation 12>
μ ₁ (j, k) = [w _REP1 (j, k) max {x _REP1 (j, k)}] / [max {w _REP1 (j, k)} x _REP1 (j, k)] · μ ₀
As is apparent from this equation, the maximum multiplication factor mu ₀ as in the case of using only the coefficients representative values, will be updated with a small coefficient update step size than mu ₀ coefficients otherwise.

In the second example, the average value of the reference signal representative value and the coefficient representative value of each subvector is obtained, the ratio of the reference signal representative value and the coefficient representative value of each subvector to the average value is obtained, and these ratios are obtained. In this method, the basic coefficient update step size μ ₀ is weighted to obtain the coefficient update step size. That is, the coefficient update step size μ ₁ (j, k) corresponding to each subvector is obtained by the following equation.
<Equation 13>
μ ₁ (j, k) = [w _REP1 (j, k) ave {x _REP1 (j, k)}] / [ave {w _REP1 (j, k)} x _REP1 (j, k)] · μ ₀

  As described above, in the second embodiment for implementing the present invention, coefficient update is performed using the coefficient update step size determined using the coefficient representative value and the reference signal representative value of the coefficient subgroup to which each coefficient belongs. Therefore, a small coefficient update step size is applied to a coefficient having a small coefficient representative value, and a large coefficient update step size is applied to a large coefficient. A large coefficient update step size is applied to a coefficient with a small reference signal representative value, and a small coefficient update step size is applied to a large coefficient. For this reason, even when the interference component with respect to the error cannot be ignored, the erroneous coefficient update is not performed, there is no delay with respect to the desired signal, and it is possible to realize signal erasure with a small amount of calculation and little output distortion.

  In the above description, the case where the same method is used as the method for determining the coefficient representative value and the reference signal representative value has been described, but these may be different. For example, the coefficient representative value can be determined based on the average value in each subgroup, and the reference signal representative value can be determined based on the maximum value in each subgroup. Further, when the maximum value, the minimum value, the median value, the average value, the variance, and the like are obtained based on the values in each subgroup, it is possible to calculate by excluding extreme outliers. Specifically, those larger or smaller than a predetermined threshold value can be excluded, or an average value can be obtained first, and the threshold value can be set as a ratio to the average value. By calculating by excluding such outliers, adverse effects of extreme outliers with respect to the representative values can be avoided, and signal erasure with less output distortion can be realized.

  Furthermore, in the best embodiment and the second embodiment, the method of controlling the coefficient update step size based on the coefficient representative value and the method of controlling the coefficient update step size based on both the coefficient representative value and the reference signal representative value have been described. However, it is also possible to perform control based only on the reference signal representative value, which has the same effect as in the second embodiment. In particular, the control based on the reference signal representative value is similar to the NLMS algorithm using the reference signal power sum. However, in the NLMS algorithm, the entire reference signal vector is handled at once without being divided into subvectors. For this reason, if there is a bias in the reference signal value in the reference signal vector, there may be a subvector that has an inappropriate value when viewed in units of subvectors. In the present invention, since the reference signal is evaluated in units of subvectors and reflected in the coefficient update step size, it is possible to avoid an adverse effect due to the deviation of the reference signal value.

  Next, a third embodiment for carrying out the present invention will be described.

FIG. 4 is a block diagram showing a third embodiment of the present invention. Compared with FIG. 1 which is a block diagram showing the best embodiment of the present invention, the step size control circuit 8 is replaced with a step size control circuit 10. In addition to the coefficient vector W ₁ (k) and the first error e ₁ (k) from the adaptive filter 5 and the subtracter 3, the step size control circuit 10 includes n ₁ (k) that is the output of the adaptive filter 5. Is supplied. The step size control circuit 10 obtains a coefficient update step size _{１ 1} (k) based on these signals. The step size control circuit 10 feeds back the obtained coefficient update step size Μ ₁ (k) to the adaptive filter 5. The step size control circuit 10 calculates the coefficient update step size based on the estimated value of the ratio of the desired signal and noise in the main signal x _P (k) in addition to the coefficient representative value in the coefficient group.

In the step size control circuit 10, the process of determining the coefficient update step size for a plurality of coefficients in the coefficient group based on the coefficient value is as described in the best embodiment of the present invention. Therefore, here, the process of obtaining the estimated value of the ratio of the desired signal and noise in the main signal x _P (k) will be described in detail.

ここに、Ｒは平均値を求めるために用いるサンプル数を表す自然数である。平均値はまた、次式で表される漏れ積分演算によって求めてもよい。
〈数１５〉
e_１ ^２(k)バー＝γ・e_１ ^２(k-１)バー＋(１―γ)・e_１ ^２(k)
ここに、γは０＜γ＜１を満たす定数である。 FIG. 5 is a block diagram showing the configuration of the step size control circuit 10. The step size control circuit 10 includes power averaging circuits 101 and 102, an SNR estimation circuit 103, and a step size calculation circuit 104. The power average circuit 101 is supplied with a first error e ₁ (k), and the power average circuit 102 is supplied with a _first pseudo signal n ₁ (k) that is an output of the adaptive filter 5. The power averaging circuit 101 squares the supplied first error e ₁ (k) to obtain e ₁ ² (k). The power averaging circuit 101 further calculates the average value e ₁ ² (k) bar and supplies it to the SNR estimation circuit 103. The e ₁ ² (k) bar can be obtained by the following equation.

<Expression 14>

Here, R is a natural number representing the number of samples used for obtaining the average value. The average value may also be obtained by a leakage integral calculation represented by the following equation.
<Expression 15>
e ₁ ² (k) bar = γ · e ₁ ² (k-1) bar + (1−γ) · e ₁ ² (k)
Here, γ is a constant that satisfies 0 <γ <1.

The power averaging circuit 102 squares the supplied first pseudo signal n ₁ (k) to obtain n ₁ ² (k). The power averaging circuit 102 further calculates the average value n ₁ ² (k) bar and supplies it to the SNR estimation circuit 103. The average value can be obtained by the methods shown in Equation 14 and Equation 15. The power averaging circuits 101 and 102 may be replaced with absolute value averaging circuits, respectively. At that time, the calculation in the SNR estimation circuit 103 needs to double the value originally obtained.

The SNR estimation circuit 103 is a ratio e ₁ ² (k) bar / n ₁ ² between the e ₁ ² (k) bar supplied from the power averaging circuit 101 and the n ₁ ² (k) bar supplied from the power averaging circuit 102. The (k) bar is obtained and the logarithm log ₁₀ {e ₁ ² (k) bar / n ₁ ² (k) bar} is supplied to the step size calculation circuit 104 as the desired signal-to-noise ratio SNR1 (k) in the main signal. To do. That is,
<Equation 16>
SNR1 (k) = log ₁₀ {e ₁ ² (k) bar / n ₁ ² (k) bar}
It is.

The step size calculation circuit 104 receives the desired signal-to-noise ratio SNR1 (k) in the main signal and the coefficient vector W ₁ (k) supplied from the adaptive filter 5, and calculates the coefficient update step size Μ ₁ (k). . Here, _{１ 1} (k) is expressed by Equation 17.
<Equation 17>
_{１ 1} (k) = m ₁ (k) · [μ ₁ (0, k), μ ₁ (1, k),…, μ ₁ (N-1, k)]
m ₁ (k) is a term determined by a desired signal-to-noise ratio SNR1 (k) in the main signal, μ ₁ (j, k), j = 0, 1,..., L−1 is a coefficient vector W ₁ ( It is a term determined by a method such as Equation 7 or Equation 8 described in the best embodiment of the present invention based on k).

m ₁ (k) is obtained as a value of a function f ₁ (v) that monotonously decreases when SNR1 _min <SNR1 (k) <SNR1 _max . Here, SNR1 _min and SNR1 _max are constants that satisfy SNR1 _min <SNR1 _max . This relationship can be expressed by Expression 18 to Expression 20.
<Equation 18>
m ₁ (k) = m _{1 max} (SNR 1 (k) <SNR 1 _min )
<Equation 19>
_{m 1 (k) = f 1} (SNR1 (k)) (SNR1 min ≦ SNR1 (k) ≦ SNR1 max)
<Expression 20>
m ₁ (k) = m _{1 min} (SNR 1 (k)> SNR 1 _max )
However, m1 _min and m1 _max are constants that satisfy m1 _min <m1 _max .

The monotone decreasing function f ₁ (v) can be expressed by, for example, Expression 21 to Expression 23.
<Expression 21>
f1 (x) = − A · x + B
<Equation 22>
_{A = (m1 max -m1 min)} / (SNR1 max -SNR1 min)
<Equation 23>
_{B = {m1 max + m1 min} + A · (SNR1 max + SNR1 min)} / 2

The purpose of the monotonic decreasing function is to obtain a small m ₁ (k) when the SNR1 (k) is large and a large m ₁ (k) when the SNR 1 (k) is small. A more complicated function expressed by a nonlinear function such as a trigonometric function or a combination thereof may be used. With the above procedure, the step size calculation circuit 104 outputs Μ ₁ (k) obtained by Equation 17 as the step size.

As described above, in the third embodiment for carrying out the present invention, the coefficient representative value of the coefficient subgroup to which each coefficient belongs and the estimated value of the desired signal to noise ratio in the main signal x _P (k) are obtained. Since coefficient update is performed using the coefficient update step size determined by using the coefficient update step size, a small coefficient update step size is applied to a coefficient having a small coefficient representative value, and a large coefficient update step size is applied to a large coefficient. Also, common to all coefficients, a large coefficient update step size is applied when the estimated value of the ratio of the desired signal to noise is small, and a small coefficient update step size is applied when it is large. For this reason, even when the interference component with respect to the error cannot be ignored, the erroneous coefficient update is not performed, there is no delay with respect to the desired signal, and it is possible to realize signal erasure with a small amount of calculation and little output distortion.

  Next, a fourth embodiment for carrying out the present invention will be described.

FIG. 6 is a block diagram showing a fourth embodiment of the present invention. Compared with FIG. 4 which is a block diagram showing the third embodiment of the present invention, the step size control circuit 10 is replaced with a step size control circuit 11. The step size control circuit 11 receives the coefficient vector W ₁ (k) and the output n ₁ (k) from the adaptive filter 5, the first error e ₁ (k) from the subtractor 3, and the reference signal x _R from the microphone 2. (k) is supplied. The step size control circuit 11 performs a coefficient update step based on the coefficient vector W ₁ (k), the first error e ₁ (k), the output n ₁ (k) of the adaptive filter 5 and the reference signal x _R (k). Find the size Μ ₁ (k). The step size control circuit 11 feeds back the obtained coefficient update step size Μ ₁ (k) to the adaptive filter 5.

In addition to the coefficient representative value in the coefficient group, the step size control circuit 10 calculates the coefficient update step size Μ ₁ (based on the reference signal representative value and the estimated value of the desired signal to noise ratio in the main signal x _P (k). k) is calculated. Here, _{１ 1} (k) is expressed by Equation 24.
<Equation 24>
_{１ 1} (k) = m ₁ (k) · [μ ₁ (0, k), μ ₁ (1, k),…, μ ₁ (N ₁ −1, k)]
m ₁ (k) is a term determined by a desired signal-to-noise ratio SNR1 (k) in the main signal, and is determined by a method such as Equation 18 to Equation 20 described in the third embodiment of the present invention. μ ₁ (j, k), j = 0, 1,..., L−1 is based on the coefficient vector W ₁ (k) and the reference signal vector X ₁ (k) in the second embodiment of the present invention. This is a term determined by the method such as Equation 12 or Equation 13 described above.

As described above, in the fourth embodiment for carrying out the present invention, the coefficient representative value and reference signal representative value of the coefficient subgroup to which each coefficient belongs, and the desired signal and noise in the main signal x _P (k) The coefficient is updated using the coefficient update step size determined using the estimated value of the ratio. Therefore, a small coefficient update step size is applied to a coefficient with a small coefficient representative value, and a large coefficient update step size is applied to a large coefficient. Is done. A large coefficient update step size is applied to a coefficient with a small reference signal representative value, and a small coefficient update step size is applied to a large coefficient. Further, in common with all coefficients, a large coefficient update step size is applied when the estimated value of the ratio of the desired signal to noise is small, and a small coefficient update step size is applied when it is large. For this reason, even when the interference component with respect to the error cannot be ignored, erroneous coefficient updating is not performed, there is no delay with respect to the desired signal, and it is possible to realize signal erasure with a low calculation amount and little output distortion.

  Next, a fifth embodiment for carrying out the present invention will be described.

FIG. 7 is a block diagram showing a fifth embodiment of the present invention. The signal processing apparatus according to the fifth embodiment of the present invention shown in FIG. 7 includes a microphone 1 that inputs sound, an input terminal 17 that inputs a reference signal, an output terminal 7, and a main signal x input from the microphone 1. _A first subtracter 3 that generates a first error by subtracting the pseudo echo from _P (k) and a reference signal x _R (k) input from the input terminal 17 are input to generate a pseudo echo. The coefficient updating step size is calculated using the first adaptive filter 5 that updates the coefficient so that the error of 1 is minimized, and the coefficient value received from the first adaptive filter 5. And a step size control circuit 8 for feedback. The first error e ₁ (k) from the first subtracter 3 is also output to the output terminal 7 as an audio signal from which echo has been eliminated. The reference signal x _R (k) input from the input terminal 17 is supplied to the speaker 16. The speaker 16 converts the reference signal x _R (k) into an acoustic signal and radiates it to the acoustic space. This acoustic signal propagates through the acoustic space and reaches the microphone 1 and is supplied to the subtractor 3 as a part of the main signal x _P (k). Such wraparound of the acoustic signal from the speaker to the microphone is known as acoustic echo. In addition, it is also known as line echo that a similar phenomenon occurs on the 4-wire side of 2-wire 4-wire conversion in a communication path. That is, the fifth embodiment of the present invention operates as an echo canceller.

FIG. 7 which is the fifth embodiment for carrying out the present invention is the same as FIG. 1 which is the best embodiment except that the reference signal x _R (k) is supplied from the input terminal 17 instead of the microphone 2. The configuration is exactly the same, and the operation is also the same. For this reason, detailed description of the operation is omitted.

  Next, a sixth embodiment for carrying out the present invention will be described.

FIG. 8 is a block diagram showing a sixth embodiment of the present invention. The signal processing apparatus according to the sixth embodiment of the present invention shown in FIG. 8 includes a microphone 1 that inputs sound, an input terminal 17 that inputs a reference signal, an output terminal 7, and a main signal x that is input from the microphone 1. _A first subtracter 3 that generates a first error by subtracting the pseudo echo from _P (k) and a reference signal x _R (k) input from the input terminal 17 are input to generate a pseudo echo. The coefficient update step size is calculated using the first adaptive filter 5 that updates the coefficient so that the error of 1 is minimized, the coefficient value received from the first adaptive filter 5 and the reference signal received from the input terminal 17. And a step size control circuit 9 that feeds back to the first adaptive filter 5. The first error e ₁ (k) from the first subtracter 3 is also output to the output terminal 7 as an audio signal from which echo has been eliminated. The reference signal x _R (k) input from the input terminal 17 is supplied to the speaker 16. The speaker 16 converts the reference signal x _R (k) into an acoustic signal and radiates it to the acoustic space. This acoustic signal propagates through the acoustic space and reaches the microphone 1 and is supplied to the subtractor 3 as a part of the main signal x _P (k). That is, the sixth embodiment of the present invention operates as an echo canceller.

FIG. 8, which is the sixth embodiment for carrying out the present invention, shows that the reference signal x _R (k) is supplied from the input terminal 17 instead of the microphone 2 as compared with FIG. 3 which is the second embodiment. Except for this, the configuration is exactly the same, and the operation is also the same. For this reason, detailed description of the operation is omitted.

  Next, a seventh embodiment for carrying out the present invention will be described.

FIG. 9 is a block diagram showing a seventh embodiment of the present invention. The signal processing apparatus according to the seventh embodiment of the present invention shown in FIG. 9 includes a microphone 1 that inputs sound, an input terminal 17 that inputs a reference signal, an output terminal 7, and a main signal x input from the microphone 1. _A first subtracter 3 that generates a first error by subtracting the pseudo echo from _P (k) and a reference signal x _R (k) input from the input terminal 17 are input to generate a pseudo echo. The coefficient update step size is calculated using the first adaptive filter 5 that updates the coefficient so that the error of 1 is minimized, the coefficient value received from the first adaptive filter 5 and the reference signal received from the input terminal 17. And a step size control circuit 10 that feeds back to the first adaptive filter 5. The first error e ₁ (k) from the first subtracter 3 is also output to the output terminal 7 as an audio signal from which echo has been eliminated. The reference signal x _R (k) input from the input terminal 17 is supplied to the speaker 16. The speaker 16 converts the reference signal x _R (k) into an acoustic signal and radiates it to the acoustic space. This acoustic signal propagates through the acoustic space and reaches the microphone 1 and is supplied to the subtractor 3 as a part of the main signal x _P (k). That is, the seventh embodiment of the present invention operates as an echo canceller.

FIG. 9 which is the seventh embodiment for carrying out the present invention is similar to FIG. 4 which is the third embodiment, in that the reference signal x _R (k) is supplied from the input terminal 17 instead of the microphone 2. Except for this, the configuration is exactly the same, and the operation is also the same. For this reason, detailed description of the operation is omitted.

  Next, an eighth embodiment for carrying out the present invention will be described.

FIG. 10 is a block diagram showing an eighth embodiment of the present invention. The signal processing apparatus according to the eighth embodiment of the present invention shown in FIG. 10 includes a microphone 1 that inputs sound, an input terminal 17 that inputs a reference signal, an output terminal 7, and a main signal x input from the microphone 1. _A first subtracter 3 that generates a first error by subtracting the pseudo echo from _P (k) and a reference signal x _R (k) input from the input terminal 17 are input to generate a pseudo echo. The first adaptive filter 5 that updates the coefficient so that the error of 1 is minimized, the coefficient value received from the first adaptive filter 5, the reference signal received from the input terminal 17, and the first signal received from the subtractor 3 And a step size control circuit 11 that calculates a coefficient update step size using the output of the adaptive filter 5 and feeds back to the first adaptive filter 5. The first error e ₁ (k) from the first subtracter 3 is also output to the output terminal 7 as an audio signal from which echo has been eliminated. The reference signal x _R (k) input from the input terminal 17 is supplied to the speaker 16. The speaker 16 converts the reference signal x _R (k) into an acoustic signal and radiates it to the acoustic space. This acoustic signal propagates through the acoustic space and reaches the microphone 1 and is supplied to the subtractor 3 as a part of the main signal x _P (k). That is, the eighth embodiment of the present invention operates as an echo canceller.

FIG. 10, which is the eighth embodiment for carrying out the present invention, shows that the reference signal x _R (k) is supplied not from the microphone 2 but from the input terminal 17 as compared to FIG. 6 which is the fourth embodiment. Except for this, the configuration is exactly the same, and the operation is also the same. For this reason, detailed description of the operation is omitted.

  Next, a ninth embodiment for carrying out the present invention will be described.

FIG. 11 is a block diagram showing the ninth embodiment of the present invention. The signal processing apparatus according to the ninth embodiment of the present invention shown in FIG. 11 includes a microphone 1 that inputs sound, a microphone 2 that inputs a reference signal, an output terminal 7, and a main signal x _P that is input from the microphone 1. A first subtracter 3 that generates a first error by subtracting the first pseudo signal from (k), and a second pseudo signal from the reference signal x _R (k) input from the microphone 2 The second subtractor 4 that generates the second error and the first error e ₁ (k) that is the output of the subtractor 3 are input to generate the second pseudo signal, and the second error is minimized. The first pseudo signal is generated by inputting the second adaptive filter 6 for updating the coefficient so that the second error e ₂ (k), which is the output of the subtractor 4, is input, and the first error is minimized. Using the first adaptive filter 5 that updates the coefficient so that the following is obtained, and the coefficient value received from the first adaptive filter 5 The coefficient update step size of 1 is used to calculate the second coefficient update step size using the coefficient value received from the second adaptive filter 6, and each is fed back to the first adaptive filter 5 and the second adaptive filter 6. A step size control circuit 12. The first error e ₁ (k) from the first subtracter 3 is also output to the output terminal 7 as an audio signal from which noise has been eliminated.

  FIG. 11, which is a ninth embodiment for carrying out the present invention, has an adaptive filter 6 and a subtracter 4 added to FIG. 1, which is the best embodiment, and a step size control circuit 8 which is a step size control circuit 12. Except for having been replaced with For this reason, only the operation related to these differences will be described in detail, and description of common parts will be omitted.

The adaptive filter 6 and the subtracter 4 are introduced in order to eliminate the mixed component of the desired signal with respect to the reference signal, that is, the crosstalk. Regarding the elimination of noise and crosstalk using two adaptive filters, Patent Document 4 describes detailed operations. As described in the best mode for carrying out the present invention, the adaptive filter 5 for canceling noise includes the first error e ₁ (k) including an interference signal other than the error. Must be robust against interference. Similarly, the adaptive filter 6 for eliminating crosstalk must be robust against interference even when the second error e ₂ (k) includes interference signals other than errors. Therefore, the step size control circuit 12 calculates the coefficient update step size for the adaptive filter 6 in addition to the coefficient update step size for the adaptive filter 5. Note that the method for calculating the coefficient update step size for the adaptive filter 5 has been described in the best mode for carrying out the present invention, and is therefore omitted here.

The step size control circuit 12 receives the coefficient vector W ₂ (k) from the adaptive filter 6 and obtains the step size Μ ₂ (k). W ₂ (k) and Μ ₂ (k) are given by the following equations.
<Equation 25>
W ₂ (k) = [w ₂ (0, k), w ₂ (1, k), ..., w ₂ (N ₂ -1, k)]
<Equation 26>
Μ ₂ (k) = [μ ₂ (0, k), μ ₂ (1, k),…, μ ₂ (N ₂ −1, k)]
Here, N ₂ is an integer of 1 or more that represents the number of taps of the adaptive filter 6.

The step size control circuit 12 first divides the coefficient vector W ₂ (k) into L subvectors (coefficient groups). In the division, it can be divided into sub-vectors of a predetermined size. The dividing method is as described in the best embodiment.

The subvector generated in this way is designated as W _S2
(j, k), j = 0, 1, ..., L-1
<Expression 27>
W ₂ (k) = [W _S2 (0, k), W _S2 (1, k), ..., W _S2 (L-1, k)]
It becomes. In the case of L _S equal division, the j-th subvector is given by the following equation.
<Equation 28>
W _S2 (j, k) = [w ₂ (jL / L _S , k), w ₂ (jL / L _S +1, k),..., W ₂ ((j + 1) L / L _S −1, k)] (j = 0, 1, ..., L-1)
Here, L _S is an integer value such that L / L _S is an integer.

Next, a representative value of a coefficient is obtained in each subvector (coefficient group). As the representative value, a maximum value, a minimum value, a median value, an average value, a variance, or the like can be used. The representative value w _REP2 (j, k) for the j-th subvector, when the maximum value is the representative value,
<Expression 29>
w _REP2 (j, k) = max {| w ₂ (jL / L _S , k) |, | w ₂ (jL / L _S +1, k) |, ..., | w ₂ ((j + 1) L / L _S -1, k) |}
Given in. Similarly, the minimum value and the median value can also be obtained by the minimum value operator or the median value operator.

となる。分散についても、同様に定義に従って計算することができる。既に最良の実施形態で説明したとおり、各係数の絶対値の代わりに２乗値を用いて、係数代表値を求めてもよい。 When the average value is the representative value w _REP2 (j, k) for the j-th subvector,
<Equation 30>

It becomes. Similarly, the variance can be calculated according to the definition. As already described in the best embodiment, the coefficient representative value may be obtained using a square value instead of the absolute value of each coefficient.

The coefficient update step size corresponding to c _REP2 (j, k) is calculated using the sub-vector representative value w _REP2 (j, k), j = 0, 1,..., L−1. To do. In the first example, the maximum value of the representative value of each subvector is obtained, the ratio of each subvector representative value to the maximum value is obtained, and the coefficient update step size μ ₀ is weighted based on this ratio to update the coefficient. This is a step size method. That is, the coefficient update step size μ ₂ (j, k) corresponding to each subvector is obtained by the following equation.
<Expression 31>
μ ₂ (j, k) = w _REP2 (j, k) / max {w _REP2 (j, k)} · μ ₀
Since the ratio takes a value between 0 and 1, the maximum coefficient is μ ₀ , and other coefficients are updated with a coefficient update step size smaller than μ ₀ .

In the second example, the average value of representative values of each subvector is obtained, the ratio of each subvector representative value to the average value is obtained, and the coefficient update step size μ ₀ is weighted based on this ratio to update the coefficient. This is a step size method. That is, the coefficient update step size μ ₂ (j, k) corresponding to each subvector is obtained by the following equation.
<Equation 32>
μ ₂ (j, k) = w _REP2 (j, k) / ave {w _REP2 (j, k)} · μ ₀

  As described above, the ninth embodiment for carrying out the present invention uses the representative values of the coefficient subgroups belonging to both the adaptive filter for noise cancellation and the adaptive filter for crosstalk cancellation. Determine the coefficient update step size. Since coefficient update is performed using these coefficient update step sizes, a small coefficient update step size is applied to a coefficient with a small representative value, and a large coefficient update step size is applied to a large coefficient. For this reason, even when the interference component with respect to the error cannot be ignored, erroneous coefficient updating is not performed, there is no delay with respect to the desired signal, and it is possible to realize signal erasure with a low calculation amount and little output distortion.

  Next, a tenth embodiment for carrying out the present invention will be described.

FIG. 12 is a block diagram showing the tenth embodiment of the present invention. Compared with FIG. 11 which is a block diagram showing the ninth embodiment of the present invention, the step size control circuit 12 is replaced with a step size control circuit 13. The step size control circuit 13 receives the coefficient vector W ₁ (k) and the second error e ₂ (k) from the adaptive filter 5 and the subtracter 4, respectively, and obtains the coefficient update step size _{１ 1} (k). The step size control circuit 13 receives the coefficient vector W ₂ (k) and the first error e ₁ (k) from the adaptive filter 6 and the subtracter 3, respectively, and obtains the coefficient update step size Μ ₂ (k). Here, the first and second errors which are the outputs of the subtracters 3 and 4 are reference input signals of the adaptive filters 6 and 5, respectively. The step size control circuit 13 feeds back the obtained coefficient update step sizes _{１ 1} (k) and Μ ₂ (k) to the adaptive filters 5 and 6, respectively. The step size control circuit 13 calculates the coefficient update step sizes of the adaptive filters 5 and 6 based on the coefficient representative value and the reference signal representative value in the coefficient group.

  FIG. 12, which is the tenth embodiment for carrying out the present invention, is the same as that of FIG. 3, which is the second embodiment, except that an adaptive filter 6 and a subtracter 4 are added, and the step size control circuit 9 is a step size control circuit. Except for being replaced with 13, the configuration is exactly the same. For this reason, only the operation related to these differences will be described in detail, and description of common parts will be omitted.

  Similar to FIG. 11 which is the ninth embodiment for carrying out the present invention, the step size control circuit 13 includes a coefficient update step for the adaptive filter 6 in addition to the coefficient update step size for the adaptive filter 5. Also calculate the size. Note that the method for calculating the coefficient update step size for the adaptive filter 5 has been described in the second embodiment for carrying out the present invention, and is therefore omitted here.

Similar to the second embodiment, the reference signal vector X ₂ (k) of the adaptive filter 6 at time k is defined as follows.
<Expression 33>
X ₂ (k) = [x ₂ (0, k), x ₂ (1, k), ..., x ₂ (N ₂ -1, k)]

Next, a representative value of the reference signal in each subvector is obtained. As the representative value, a maximum value, a minimum value, a median value, an average value, a variance, or the like can be used. The representative value x _REP2 (j, k) for the j-th subvector, when the maximum value is the representative value,
<Expression 34>
x _REP2 (j, k) = max {| x ₂ (jL / L _S , k) |, | x ₂ (jL / L _S +1, k) |, ..., | x ₂ ((j + 1) L / L _S -1, k) |}
Given in. Similarly, the minimum value and the median value can also be obtained by the minimum value operator or the median value operator.

となる。分散についても、同様に定義に従って計算することができる。 When the average value is a representative value x _REP2 (j, k) for the j-th subvector,
<Equation 35>

It becomes. Similarly, the variance can be calculated according to the definition.

  In the description so far, the reference signal representative value is obtained for the absolute value of each reference signal value. However, similar to the case of the coefficient value, the same effect can be obtained by using the square value instead of the absolute value.

Reference signal representative value x _REP2 (j, k), j = 0, 1,..., L−1 and coefficient representative value w _REP2 (j, k), j = 0, 1, ..., L-1 is used to calculate the coefficient update step size corresponding to x _REP2 (j, k) and w _REP2 (j, k). In the first example, the maximum value of the reference signal representative value and coefficient representative value of each subvector is obtained, the ratio of the reference signal representative value and coefficient representative value of each subvector to the maximum value is obtained, and these ratios are obtained. In this method, the basic coefficient update step size μ ₀ is weighted to obtain the coefficient update step size. That is, the coefficient update step size μ ₂ (j, k) corresponding to each subvector is obtained by the following equation.
<Equation 36>
μ ₂ (j, k) = [w _REP2 (j, k) max {x _REP2 (j, k)}] / [max {w _REP2 (j, k)} x _REP2 (j, k)] · μ ₀

As is apparent from this equation, the maximum multiplication factor mu ₀ as in the case of using only the coefficients representative values, will be updated with a small coefficient update step size than mu ₀ coefficients otherwise.

In the second example, the average value of the reference signal representative value and the coefficient representative value of each subvector is obtained, the ratio of the reference signal representative value and the coefficient representative value of each subvector to the average value is obtained, and these ratios are obtained. In this method, the basic coefficient update step size μ ₀ is weighted to obtain the coefficient update step size. That is, the coefficient update step size μ ₂ (j, k) corresponding to each subvector is obtained by the following equation.
<Expression 37>
μ ₂ (j, k) = [w _REP2 (j, k) ave {x _REP2 (j, k)}] / [ave {w _REP2 (j, k)} x _REP2 (j, k)] · μ ₀

  As described above, in the tenth embodiment for carrying out the present invention, the reference signal representative values of the coefficient subgroups belonging to both the adaptive filter for noise cancellation and the adaptive filter for crosstalk cancellation The coefficient update step size is determined using the coefficient representative value. Since coefficient update is performed using these coefficient update step sizes, a small coefficient update step size is applied to a coefficient with a small coefficient representative value, and a large coefficient update step size is applied to a large coefficient. A large coefficient update step size is applied to a coefficient with a small reference signal representative value, and a small coefficient update step size is applied to a large coefficient. For this reason, even when the interference component with respect to the error cannot be ignored, erroneous coefficient updating is not performed, there is no delay with respect to the desired signal, and it is possible to realize signal erasure with a low calculation amount and little output distortion.

  Next, an eleventh embodiment for carrying out the present invention will be described.

FIG. 13 is a block diagram showing an eleventh embodiment of the present invention. Compared with FIG. 11 which is a block diagram showing the ninth embodiment of the present invention, the step size control circuit 12 is replaced with a step size control circuit 14. The step size control circuit 14 is supplied with the coefficient vector W ₁ (k) and the output n ₁ (k) from the adaptive filter 5 and the first error e ₁ (k) from the subtractor 3. The step size control circuit 14 obtains a coefficient update step size _{１ 1} (k) based on these signals. The step size control circuit 14 is supplied with the coefficient vector W ₂ (k) and the output n ₂ (k) from the adaptive filter 6 and the second error e ₂ (k) from the subtractor 4. The step size control circuit 14 obtains a coefficient update step size Μ ₂ (k) based on these signals. The step size control circuit 14 feeds back the obtained coefficient update step sizes _{１ 1} (k) and Μ ₂ (k) to the adaptive filters 5 and 6, respectively. In addition to the coefficient representative value in the coefficient group, the step size control circuit 14 estimates the ratio of the desired signal to noise in the main signal x _P (k) or the ratio of the desired signal to noise in the reference signal x _R (k). Based on the estimated value, the coefficient update step size of the adaptive filter 5 or 6 is calculated.

  FIG. 13, which is an eleventh embodiment for carrying out the present invention, is the same as that of FIG. 4, which is the third embodiment, except that an adaptive filter 6 and a subtracter 4 are added, and the step size control circuit 10 is replaced by a step size control circuit. Except for being replaced with 14, the configuration is exactly the same. For this reason, only the operation related to these differences will be described in detail, and description of common parts will be omitted.

  Similar to FIG. 12, which is the tenth embodiment for carrying out the present invention, the step size control circuit 14 includes a coefficient update step for the adaptive filter 6 in addition to the coefficient update step size for the adaptive filter 5. Also calculate the size. Note that the method for calculating the coefficient update step size for the adaptive filter 5 has been described in the third embodiment for carrying out the present invention, and is therefore omitted here.

The process of determining the coefficient update step size for the plurality of coefficients in the coefficient group based on the coefficient value in the step size control circuit 14 is as already described in the ninth embodiment of the present invention. Therefore, here, the process of obtaining the estimated value of the ratio of the desired signal and noise in the reference signal x _R (k) will be described in detail.

  FIG. 14 is a block diagram showing the configuration of the step size control circuit 14. The step size control circuit 14 includes power averaging circuits 101, 102, 141 and 142, SNR estimation circuits 103 and 143, and step size calculation circuits 104 and 144. The operations of the power averaging circuits 101 and 102, the SNR estimation circuit 103, and the step size calculation circuit 104 have already been described with reference to FIG. Just like these components, power averaging circuits 141 and 142, SNR estimation circuit 143, and step size calculation circuit 144 operate.

平均値はまた、次式で表される漏れ積分演算によって求めてもよい。
〈数３９〉
e_２ ^２(k)バー＝γ・e_２ ^２(k-１)バー＋(１―γ)・e_２ ^２(k) The power average circuit 141 is supplied with the second error e ₂ (k), and the power average circuit 142 is supplied with the second pseudo signal n ₂ (k) that is the output of the adaptive filter 6. The power averaging circuit 141 squares the supplied second error e ₂ (k) to obtain e ₂ ² (k). The power averaging circuit 141 further calculates the average value e ₂ ² (k) bar and supplies it to the SNR estimation circuit 143. The e ₂ ² (k) bar can be obtained by the following equation.

<Equation 38>

The average value may also be obtained by a leakage integral calculation represented by the following equation.
<Equation 39>
e ₂ ² (k) bar = γ · e ₂ ² (k-1) bar + (1−γ) · e ₂ ² (k)

The power averaging circuit 142 squares the supplied second pseudo signal n ₂ (k) to obtain n ₂ ² (k). The power averaging circuit 142 further calculates the average value n ₂ ² (k) bar and supplies it to the SNR estimation circuit 143. The average value can be obtained by the method shown in Equation 38 and Equation 39. The power averaging circuits 141 and 142 may be replaced with absolute value averaging circuits, respectively. At that time, the calculation in the SNR estimation circuit 143 needs to double the value originally obtained.

The SNR estimation circuit 143 has a ratio e ₂ ² (k) bar / n ₂ ² between the e ₂ ² (k) bar supplied from the power averaging circuit 141 and the n ₂ ² (k) bar supplied from the power averaging circuit 142. (k) bar is obtained, and its logarithmic log ₁₀ {e ₂ ² (k) bar / n ₂ ² (k) bar} is supplied to the step size calculation circuit 144 as the second desired signal-to-noise ratio SNR2 (k). To do. That is,
<Equation 40>
SNR2 (k) = log ₁₀ {e ₂ ² (k) bar / n ₂ ² (k) bar}
It is.

The step size calculation circuit 134 receives the desired signal-to-noise ratio SNR2 (k) in the reference signal and the coefficient vector W ₂ (k) supplied from the adaptive filter 6, and calculates the coefficient update step size Μ ₂ (k). . Here, Μ ₂ (k) is expressed by Equation 41.
<Equation 41>
Μ ₂ (k) = m ₂ (k) · [μ ₂ (0, k), μ ₂ (1, k),…, μ ₂ (N ₂ −1, k)]
m ₂ (k) is a term defined by a desired signal-to-noise ratio SNR2 (k) in the reference signal, μ ₂ (j, k), j = 0, 1,..., L−1 is a coefficient vector W ₂ ( This is a term determined by a method such as Equation 31 or Equation 32 described in the ninth embodiment of the present invention based on k).

m ₂ (k) is obtained as a value of a function f ₂ (v) that monotonically increases with SNR _{2 min} <SNR 2 (k) <SNR 2 _max . _Here, SNR2 min, _{SNR2 max are} constants satisfying the _SNR2 min <SNR2 _max. This relationship can be expressed by Equations 42 to 44.
<Equation 42>
m ₂ (k) = m _{2 min} (SNR 2 (k) <SNR 2 _min )
<Equation 43>
m ₂ (k) = f ₂ (SNR2 (k)) (SNR2 _min ≦ SNR2 (k) ≦ SNR2 _max )
<Equation 44>
m ₂ (k) = m _{2 max} (SNR 2 (k)> SNR 2 _max )
_However, m @ 2 _{min, m @ 2 max are} constants satisfying the _m2 min _{<m2 max.}

The monotonically increasing function f ₂ (v) can be expressed by, for example, Expressions 45 to 47.
<Equation 45>
f2 (x) = C · x + D
<Equation 46>
_{C = (m2 max -m2 min)} / (SNR2 max -SNR2 min)
<Formula 47>
_{D = {m2 max + m2 min} + C · (SNR2 max + SNR2 min)} / 2

The purpose of the monotonically increasing function is to obtain a large m ₂ (k) when the SNR 2 (k) is large and a small m ₂ (k) when the SNR ₂ (k) is small. A more complicated function expressed by a nonlinear function such as a trigonometric function or a combination thereof may be used. With the above procedure, the step size calculation circuit 144 outputs Μ ₂ (k) obtained by Equation 41 as the step size.

As described above, in the eleventh embodiment for carrying out the present invention, the coefficient representative values and the main values of the coefficient subgroups belonging to both the adaptive filter for noise cancellation and the adaptive filter for crosstalk cancellation are the same. Since coefficient update is performed using the coefficient update step size determined using the estimated value of the ratio of the desired signal and noise in the signal x _P (k) or the reference signal x _R (k), the coefficient representative value is reduced to a smaller coefficient. A small coefficient update step size is applied to a large coefficient, and a large coefficient update step size is applied to a large coefficient. The adaptive filter 5 for eliminating noise is applied with a large coefficient update step size when the estimated value of the desired signal to noise ratio is small, and with a small coefficient update step size when it is large. The adaptive filter 6 for eliminating crosstalk is applied with a small coefficient update step size when the estimated value of the desired signal to noise ratio is small and a large coefficient update step size when it is large. For this reason, even when the interference component with respect to the error cannot be ignored, erroneous coefficient updating is not performed, there is no delay with respect to the desired signal, and it is possible to realize signal erasure with a low calculation amount and little output distortion.

  Next, a twelfth embodiment for carrying out the present invention will be described.

  FIG. 15 is a block diagram showing a twelfth embodiment of the present invention. Compared with FIG. 14 which is a block diagram showing the eleventh embodiment of the present invention, the step size control circuit 14 is replaced with a step size control circuit 15.

The step size control circuit 15 is supplied with the coefficient vector W ₁ (k) and the output n ₁ (k) from the adaptive filter 5 and the first error e ₁ (k) from the subtractor 3. The step size control circuit 15 is supplied with a coefficient vector W ₂ (k) and an output n ₂ (k) from the adaptive filter 6, and a second error e ₂ (k) from the subtractor 4. The step size control circuit 15 includes a coefficient vector W ₁ (k), a first error e ₁ (k), an output n ₁ (k) of the adaptive filter 5, and a second error that is a reference input signal of the adaptive filter 5. A coefficient update step size Μ ₁ (k) is obtained based on e ₂ (k). The step size control circuit 15 also includes a coefficient vector C ₂ (k), a second error e ₂ (k), an output n ₂ (k) of the adaptive filter 6, and a first reference input signal of the adaptive filter 6. The coefficient update step size Μ ₂ (k) is obtained based on the error e ₁ (k). The step size control circuit 15 feeds back the obtained coefficient update step sizes _{１ 1} (k) and Μ ₂ (k) to the adaptive filters 5 and 6, respectively.

In addition to the coefficient representative value and the reference signal representative value in each coefficient group of the adaptive filter, the step size control circuit 15 estimates the ratio of the desired signal to noise in the main signal x _P (k) or the reference signal x _R (k The coefficient update step sizes _{１ 1} (k) and Μ ₂ (k) of the adaptive filters 5 and 6 are calculated based on the estimated value of the ratio of the desired signal to noise in).

  FIG. 15 which is a twelfth embodiment for carrying out the present invention is the same as the fourth embodiment shown in FIG. 6 except that an adaptive filter 6 and a subtractor 4 are added, and the step size control circuit 11 is replaced with a step size control circuit. Except for being replaced with 15, the configuration is exactly the same. For this reason, only the operation related to these differences will be described in detail, and description of common parts will be omitted.

  Similar to FIG. 13 which is the eleventh embodiment for carrying out the present invention, the step size control circuit 15 includes a coefficient update step for the adaptive filter 6 in addition to the coefficient update step size for the adaptive filter 5. Also calculate the size. Note that the method for calculating the coefficient update step size for the adaptive filter 5 has been described in the fourth embodiment for carrying out the present invention, and is therefore omitted here.

In addition to the coefficient representative value in the coefficient group, the step size control circuit 15 calculates the coefficient update step size Μ ₂ (based on the reference signal representative value and the estimated value of the desired signal to noise ratio in the reference signal x _R (k). k) is calculated. Here, Μ ₂ (k) is expressed by Formula 48.
<Formula 48>
_{Μ 2 (k) = m 2} (k) · [μ 2 (0, k), μ 2 (1, k), ..., μ 2 (N2 1 -1, k)]
m ₂ (k) is a term defined by a desired signal-to-noise ratio SNR2 (k) in the reference signal, μ ₂ (j, k), j = 0, 1,..., L−1 is a coefficient vector W ₂ ( This is a term determined by a method such as Equation 36 or Equation 37 described in the tenth embodiment of the present invention based on k) and the reference signal vector X ₂ (k).

As described above, in the twelfth embodiment for carrying out the present invention, the coefficient representative values of the coefficient subgroups belonging to both the adaptive filter for noise cancellation and the adaptive filter for crosstalk cancellation are referred to. The coefficient update is performed using the coefficient update step size determined using the signal representative value and the estimated value of the desired signal to noise ratio in the main signal x _P (k). A large coefficient update step size is applied to a coefficient having a large coefficient update step size. A large coefficient update step size is applied to a coefficient with a small reference signal representative value, and a small coefficient update step size is applied to a large coefficient. Further, in common with all the coefficients, the adaptive filter 5 for canceling noise is applied with a large coefficient update step size when the estimated value of the desired signal to noise ratio is small and a small coefficient update step size when large. . On the other hand, a small coefficient update step size is applied to the adaptive filter 6 for eliminating crosstalk when the estimated value of the desired signal to noise ratio is small, and a large coefficient update step size is applied when it is large. For this reason, even when the interference component with respect to the error cannot be ignored, erroneous coefficient updating is not performed, there is no delay with respect to the desired signal, and it is possible to realize signal erasure with a low calculation amount and little output distortion.

  In the twelfth embodiment to the twelfth embodiment described so far, the coefficient update step size determined in advance is used until the coefficient of the adaptive filter converges, and the coefficient update step size after the adaptive filter converges. Can also be controlled by the method described in each embodiment. In that case, the convergence of the adaptive filter must be determined. Information regarding the coefficient vector can be used to determine convergence.

  For example, for all taps, convergence can be determined by obtaining the norm of coefficient vectors, that is, the sum of coefficient squares and evaluating the rate of change of the norm. When the adaptive filter converges, the norm of the coefficient vector becomes a substantially constant value, and the increase rate becomes a value very close to zero. In particular, when the coefficient update is started after the coefficient initial value is set to zero, the norm starts from zero, increases, and eventually saturates. Therefore, the rate of change is equal to the rate of increase. This rate of increase is initially high and decreases as it approaches convergence. For this reason, when this increase rate becomes smaller than a predetermined value, it can be determined that the adaptive filter has converged.

  The convergence determination of the adaptive filter coefficient can be performed independently for each coefficient group. At this time, the convergence of the corresponding coefficient group can be determined by evaluating the change rate or the increase rate of the norm of the subvector consisting of the coefficients in the coefficient group instead of the coefficient vector. When the adaptive filter approximates the transfer function of the acoustic space, the larger the delay of the tap where the coefficient is, the smaller the value of the coefficient and the faster the convergence. For this reason, by performing convergence determination independently for each coefficient group, it is possible to perform accurate convergence determination in accordance with the nonuniformity of the convergence time due to the coefficients.

  In the above-described adaptive filter convergence determination using coefficient values, outlier coefficients having extremely large or small values can be excluded. By excluding outliers, it is possible to avoid an unfavorable influence due to the coefficient values of these extreme values and perform an accurate convergence determination.

  In addition to the convergence determination based on the coefficient values described here, the convergence determination of the adaptive filter also includes output error reduction, error autocorrelation reduction, error and reference signal correlation reduction, error and main signal correlation. It is widely known throughout the literature that reduction can be used.

  The signal processing apparatus according to the present invention described above can also be realized by software. That is, a program used for signal processing can be configured by configuring the processing operation of each circuit in the signal processing apparatus of each embodiment described above as a step or procedure in software. Such a program is executed by a processor such as a DSP (Digital Signal Processor) constituting a signal processing device or a noise canceling device.

  Furthermore, a program product including such a program or a storage medium storing such a program is also included in the scope of the present invention.

It is a block diagram which shows the structure of the signal processing apparatus of the best embodiment of this invention. It is a figure showing the relationship between a coefficient value, a coefficient group, and a coefficient representative value. It is a block diagram which shows the structure of the signal processing apparatus of the 2nd Embodiment of this invention. It is a block diagram which shows the structure of the signal processing apparatus of the 3rd Embodiment of this invention. It is a figure showing the 1st structure of a step size control circuit. It is a block diagram which shows the structure of the signal processing apparatus of the 4th Embodiment of this invention. It is a block diagram which shows the structure of the signal processing apparatus of the 5th Embodiment of this invention. It is a block diagram which shows the structure of the signal processing apparatus of the 6th Embodiment of this invention. It is a block diagram which shows the structure of the signal processing apparatus of the 7th Embodiment of this invention. It is a block diagram which shows the structure of the signal processing apparatus of the 8th Embodiment of this invention. It is a block diagram which shows the structure of the signal processing apparatus of the 9th Embodiment of this invention. It is a block diagram which shows the structure of the signal processing apparatus of the 10th Embodiment of this invention. It is a block diagram which shows the structure of the signal processing apparatus of the 11th Embodiment of this invention. It is a figure showing the 2nd structure of a step size control circuit. It is a block diagram which shows the structure of the signal processing apparatus of the 12th Embodiment of this invention.

Explanation of symbols

1, 2 Microphone 3, 4 Subtractor 5, 6 Adaptive filter 7 Output terminal 8, 9, 10, 11, 12, 13, 14, 15 Step size control circuit 16 Speaker 17 Input terminal 101, 102, 141, 142 Power average Circuit 103,143 SNR estimation circuit 104,144 Step size calculation circuit

Claims (33)

Generate multiple coefficient groups by grouping multiple adaptive filter coefficients in order,
Using the coefficients of the adaptive filter in the plurality of coefficient groups , a coefficient representative value of each coefficient group is obtained,
A signal processing method comprising: controlling a coefficient update step size of each coefficient in the coefficient group using the coefficient representative value, and updating the coefficient using the coefficient update step size. The coefficient representative value selected for each coefficient group is any one of a maximum value, a minimum value, a median value, an average value, and a variance value of coefficient values in each group. Signal processing method. Obtaining a reference signal representative value in the plurality of coefficient groups;
Controlling the coefficient update step size of each coefficient in the coefficient group using the coefficient representative value and the reference signal representative value;
Update coefficients using the coefficient update step size
The signal processing method according to claim 1, wherein the signal processing method is a signal processing method. Normalize the coefficient representative value to obtain a plurality of normalized coefficient representative values,
Controlling the coefficient update step size for the coefficients in the coefficient group using the plurality of normalized coefficient representative values;
Update coefficients using the coefficient update step size
The signal processing method according to any one of claims 1 to 3, wherein: Normalizing the reference signal representative value to obtain a plurality of standardized reference signal representative values;
Controlling a coefficient update step size for a coefficient in the coefficient group using the plurality of standardized reference signal representative values;
Update coefficients using the coefficient update step size
The signal processing method according to claim 3 or 4, wherein Controlling the coefficient update step size after the adaptive filter has converged;
Update coefficients using the coefficient update step size
The signal processing method according to any one of claims 1 to 5, wherein: Detecting a rate of change of a coefficient value of the adaptive filter, and controlling the coefficient update step size when the rate of change is smaller than a predetermined value;
Update coefficients using the coefficient update step size
The signal processing method according to claim 6. An estimated value of the first signal is generated by subtracting the estimated value of the second signal from the main signal in which the first signal (desired signal) and the second signal (noise) are mixed;
An estimated value of the second signal included in the main signal is obtained by an adaptive filter using a reference signal correlated with the second signal as an input,
A signal erasing method, wherein the coefficient update step size of the adaptive filter is controlled by the method according to any one of claims 1 to 7. Obtaining an estimate of the ratio of the first signal to the second signal in the main signal based on the estimate of the first signal and the estimate of the second signal;
The coefficient update step size of the adaptive filter is controlled by the method according to any one of claims 1 to 7, using an estimate of a ratio of the first signal to the second signal in the main signal,
A signal erasing method, wherein a coefficient is updated using the coefficient update step size. The first estimation of the first signal is performed by subtracting the first estimated value (estimated noise) of the second signal from the main signal in which the first signal (desired signal) and the second signal (noise) are mixed. Generate a value,
A second estimated value of the second signal is generated by subtracting a second estimated value (estimated crosstalk) of the first signal from a reference signal in which the second signal and the first signal are mixed. ,
Using a second estimated value of the second signal as an input, a second signal included in the main signal is estimated by a first adaptive filter, and a first estimated value (estimated noise) of the second signal is obtained. Seeking
Using the first estimated value of the first signal as an input, the first signal included in the reference signal is estimated by a second adaptive filter, and the second estimated value (estimated crosstalk) of the first signal is estimated. Seeking
The step size of the first adaptive filter and the second adaptive filter is controlled by the method according to any one of claims 1 to 7,
A signal erasing method comprising updating coefficients of the first and second adaptive filters using the coefficient update step size. An estimate of a first signal to second signal ratio in the main signal based on a first estimate of the first signal and a first estimate of the second signal;
Further, an estimated value of the first signal to second signal ratio in the reference signal is obtained based on the second estimated value of the first signal and the second estimated value of the second signal, respectively.
Controlling the step size of the first adaptive filter with the method of any one of claims 1 to 7 using an estimate of a first signal to second signal ratio in the main signal;
Controlling the step size of the second adaptive filter by the method according to any one of claims 1 to 6, using an estimate of a first signal to second signal ratio in the reference signal;
Update the coefficients of the first and second adaptive filters using the coefficient update step size
A signal erasing method characterized by the above. The first signal is a near-end signal and the second signal is an echo
10. The signal erasing method according to claim 8 or 9, The first signal is a desired signal and the second signal is noise
The signal erasing method according to claim 8, wherein the signal erasing method is a signal erasing method. Generate multiple coefficient groups by grouping multiple adaptive filter coefficients in order,
Using the coefficients of the adaptive filter in the plurality of coefficient groups, a coefficient representative value of each coefficient group is obtained,
Comprising at least a step size control circuit for controlling a coefficient update step size of each coefficient in the coefficient group using the coefficient representative value;
A signal processing apparatus that updates a coefficient using the coefficient update step size. The coefficient representative value selected for each coefficient group is any one of a maximum value, a minimum value, a median value, an average value, and a variance value of coefficient values within each group. Signal processing device. Obtaining a reference signal representative value in the plurality of coefficient groups;
A step size control circuit for controlling a coefficient update step size of each coefficient in the coefficient group using the coefficient representative value and the reference signal representative value;
The signal processing apparatus according to claim 14, wherein the coefficient is updated using the coefficient update step size. Normalize the coefficient representative value to obtain a plurality of normalized coefficient representative values,
A step size control circuit for controlling a coefficient update step size for a coefficient in the coefficient group using the plurality of normalized coefficient representative values;
Update coefficients using the coefficient update step size
The signal processing apparatus according to claim 14, wherein the signal processing apparatus is a signal processing apparatus. Normalizing the reference signal representative value to obtain a plurality of standardized reference signal representative values;
A step size control circuit for controlling a coefficient update step size for a coefficient in the coefficient group using the plurality of standardized reference signal representative values;
Update coefficients using the coefficient update step size
The signal processing apparatus according to claim 16 or 17, characterized in that: Comprising a step size control circuit for controlling a coefficient update step size after the adaptive filter has converged;
Update coefficients using the coefficient update step size
The signal processing apparatus according to claim 14, wherein the signal processing apparatus includes: A step size control circuit that detects a change rate of a coefficient value of the adaptive filter and determines that the change rate has converged when the change rate becomes smaller than a predetermined value;
Update coefficients using the coefficient update step size
The signal processing apparatus according to claim 19. A first subtractor that subtracts an estimated value of the second signal from a main signal in which the first signal (desired signal) and the second signal (noise) are mixed, and obtains an estimated value of the first signal;
A first adaptive filter that obtains an estimated value of a second signal included in the main signal with a reference signal correlated with the second signal as an input;
A step size control circuit according to any one of claims 14 to 20, comprising:
Update coefficients of the first adaptive filter using the coefficient update step size
A signal processing apparatus. The first estimated value of the first signal is obtained by subtracting the first estimated value (estimated noise) of the second signal from the main signal in which the first signal (desired signal) and the second signal (noise) are mixed. A first subtractor for obtaining
A second estimation value of a second signal is obtained by subtracting a second estimated value (estimated crosstalk) of the first signal from a reference signal in which the second signal and the first signal are mixed. A subtractor of
A first adaptive filter that receives a second estimated value of the second signal and estimates a second signal included in the main signal to obtain a first estimated value (estimated noise) of the second signal When,
Second adaptation for obtaining a second estimated value (estimated crosstalk) of the first signal by estimating a first signal included in the reference signal by using a first estimated value of the first signal as an input Filters,
A step size control circuit according to any one of claims 14 to 20, comprising:
The signal processing apparatus updates the coefficients of the first and second adaptive filters using the first and second coefficient update step sizes, respectively. The first signal is a near-end signal and the second signal is an echo
The signal processing apparatus according to claim 21, wherein: The first signal is a desired signal and the second signal is noise
The signal processing apparatus according to claim 21 or 22, On the computer,
A process of generating a plurality of coefficient groups by sequentially combining a plurality of coefficients of the adaptive filter;
A process of obtaining a coefficient representative value of each coefficient group using coefficients of the adaptive filter in the plurality of coefficient groups;
Processing for controlling the coefficient update step size of each coefficient in the coefficient group using the coefficient representative value;
Processing to update the coefficient using the coefficient update step size;
A signal processing program characterized in that 26. The coefficient representative value selected for each coefficient group is any one of a maximum value, a minimum value, a median value, an average value, and a variance value of coefficient values within each group. Signal processing program. On the computer,
Processing for obtaining a reference signal representative value in the plurality of coefficient groups;
Processing for controlling a coefficient update step size of each coefficient in the coefficient group using the coefficient representative value and the reference signal representative value;
26. The signal processing program according to claim 25, wherein: The process of obtaining the coefficient representative value is as follows:
28. The signal processing program according to claim 25, wherein the signal processing program is a process of normalizing the coefficient representative value to obtain a plurality of normalized coefficient representative values. 29. The signal processing program according to claim 27, wherein the process of obtaining the reference signal representative value is a process of obtaining a plurality of standardized reference signal representative values by normalizing the reference signal representative value. The process for controlling the coefficient update step size is:
Detecting the convergence of the adaptive filter,
30. The signal processing program according to claim 25, which is executed after the adaptive filter has converged. The process of detecting convergence of the adaptive filter includes:
A process for evaluating the rate of change of the coefficient value of the adaptive filter;
Including a process of determining that the rate of convergence has converged when the rate of change becomes smaller than a predetermined value.
The signal processing program according to claim 30, wherein: On the computer,
A process of subtracting an estimated value of the second signal from a main signal in which the first signal (desired signal) and the second signal (noise) are mixed to obtain an estimated value of the first signal;
A process of obtaining an estimated value of the second signal included in the main signal by an adaptive filter using a reference signal correlated with the second signal as an input;
A step size control process according to any one of claims 25 to 31,
Processing to update the coefficient of the adaptive filter with the coefficient update step size;
A signal processing program characterized in that On the computer,
The first estimated value of the first signal is obtained by subtracting the first estimated value (estimated noise) of the second signal from the main signal in which the first signal (desired signal) and the second signal (noise) are mixed. Processing for
Processing for subtracting a second estimated value (estimated crosstalk) of the first signal from a reference signal in which the second signal and the first signal are mixed to obtain a second estimated value of the second signal; ,
Using the second estimated value of the second signal as an input, the first adaptive filter estimates the second signal included in the main signal and obtains the first estimated value (estimated noise) of the second signal. The required processing,
Using the first estimated value of the first signal as an input, the second signal is estimated by the second adaptive filter to estimate the first signal included in the reference signal (estimated crosstalk). Processing for
A step size control process according to any one of claims 25 to 31,
Updating the coefficients of the first and second adaptive filters, respectively, with the proxy and second coefficient update step sizes;
A signal processing program characterized in that

Images