JP2011205533A - Adaptive equalizer, echo canceler and active noise control apparatus - Google Patents

Abstract

PROBLEM TO BE SOLVED: To obtain an adaptive equalizer, capable of preventing identification accuracy of a linear element from being reduced by the existence of a non-linear element in a transfer function.SOLUTION: In a first signal converter 101 to a second signal converter 106, a filter coefficient represented in a frequency domain is held, and based on the filter coefficient, adaptive learning is performed from an input signal of a target unknown system and an output signal of the unknown system, to identify a transfer function of the unknown system. A frequency weight processor 107 multiplies the filter coefficient or the updated amount of the filter coefficient in a filter processor 102 with a weight, on the basis of non-linear distortion characteristic information about non-linear distortion characteristics of the unknown system held in a non-linear distortion characteristic information holder 108.

Description

  The present invention relates to an adaptive equalizer for identifying a desired transmission coefficient such as a propagation path of a sound wave, an echo canceller using the same, and an active noise control device.

  In the field of acoustic signal processing, an adaptive equalizer (adaptive filter) is used for the purpose of identifying a desired transfer function such as a propagation path of a sound wave. For example, in an acoustic echo canceller, an echo leaking into a microphone is predicted and eliminated by identifying a transfer function of an echo path from an amplifier / speaker that reproduces a received sound to a microphone. Such an acoustic echo canceller is disclosed in Non-Patent Document 1, for example.

  In active noise control, it is used to optimize a filter for controlling noise. For example, in feed-forward active noise control, it is adapted to derive a filter with a transfer function that adds the inverse of the sound wave transfer characteristic from the amplifier to the target position to the sound wave transfer characteristic from the noise source to the target position. An equalizer is used. The derived filter has a characteristic of converting a noise signal picked up in the vicinity of a noise source into a control sound that reduces noise at a target point. Such an active noise control device is disclosed in Non-Patent Document 2, for example.

Shigeo Sakurai, “Adaptive Signal Processing”, Shosodo, May 15, 1995, pp. 48-49 Shiro Ise et al., “Active Noise Control”, Corona, July 7, 2006, pp. 9-14

  It is known that a speaker used for an acoustic echo canceller or an active noise control device generally has a nonlinear element in response characteristics. Also, it is known that an amplifier that drives this generates a non-linear response such as a harmonic with respect to an input exceeding a saturation level. Therefore, in an acoustic echo canceller or active noise control device in which the response characteristics of speakers and amplifiers are included in the transfer function to be identified, it is necessary to use an adaptive equalizer corresponding to a nonlinear transfer function. For example, as such an adaptive equalizer, Meiji Muneyasu et al., “Nonlinear Digital Signal Processing”, Asakura Shoten, March 20, 1999, pp. There is a Volterra adaptive filter as shown in 129-132. However, compared to a general linear adaptive filter, the Volterra adaptive filter has a problem of high calculation cost. For this reason, in practice, nonlinear elements in the response characteristics of speakers and amplifiers are ignored, and these characteristics are considered to be completely linear, and a linear adaptive equalizer with a low calculation cost is often used.

  However, in this case, there is a problem that an error due to a nonlinear element is always included between the transfer function identified by the adaptive equalizer and the actual transfer function. Further, there is a problem that identification of a linear element cannot be performed with sufficient accuracy by identifying a transfer function including a nonlinear element as linear.

  For example, in acoustic echo cancellers, the nonlinear component of the echo generated by the nonlinear component of the speaker amplifier is always included in the residual of the echo cancellation process, so even the linear component of the transfer function of the echo path is sufficient. It becomes impossible to identify, and there arises a problem that the echo canceling effect is lowered.

  Further, in the active noise control device, when noise exists in a frequency band where the speaker distortion rate is relatively large, for example, in the vicinity of the reproduction lower limit frequency of the speaker, the adaptive equalizer ignores this non-linear distortion and controls noise for the control sound. By deriving the filter, unintended nonlinear distortion occurs in the control sound output from the speaker, and an extra sound wave is added by the nonlinear distortion component, resulting in a problem that the noise reduction effect is lowered.

  The present invention has been made to solve the above-described problems. When a transfer function including a nonlinear element is identified by a linear adaptive equalizer, the identification accuracy of the linear element due to the presence of the nonlinear element of the transfer function is improved. An object of the present invention is to prevent the deterioration and to solve the problems caused by the echo canceller and the active noise control device.

  The adaptive equalizer according to the present invention holds the filter coefficient expressed in the frequency domain, and performs adaptive learning from the input signal of the target unknown system and the output signal of the unknown system based on the filter coefficient. The frequency domain adaptive filter means for identifying the transfer function of the unknown system and the frequency by which the filter coefficient of the frequency domain adaptive filter means or the update amount of the filter coefficient is multiplied by the weight based on the nonlinear distortion characteristic information related to the nonlinear distortion characteristic of the unknown system Weighting means.

  The adaptive equalizer according to the present invention multiplies the filter coefficient of the frequency domain adaptive filter means or the update amount of the filter coefficient based on the nonlinear distortion characteristic information related to the nonlinear distortion characteristic of the unknown system. It is possible to prevent a decrease in the identification accuracy of the linear element due to the presence of the nonlinear element.

It is a block diagram which shows the adaptive equalizer by Embodiment 1 of this invention. It is a block diagram which shows the example of a connection of the adaptive equalizer and unknown system by Embodiment 1 of this invention. It is a block diagram which shows the adaptive equalizer by Embodiment 2 of this invention. It is explanatory drawing of the input-output characteristic function in the adaptive equalizer by Embodiment 2 of this invention. It is a block diagram which shows the adaptive equalizer by Embodiment 3 of this invention. It is explanatory drawing which shows the example of a setting of the frequency weight according to the distortion rate of the adaptive equalizer by Embodiment 3 of this invention. It is a block diagram which shows the adaptive equalizer by Embodiment 4 of this invention. It is a block diagram which shows the example of a connection of the adaptive equalizer and unknown system by Embodiment 4 of this invention. It is a block diagram which shows the echo cancellation apparatus by Embodiment 5 of this invention. It is a block diagram which shows the active noise control apparatus by Embodiment 6 of this invention. It is a block diagram which shows the active noise control apparatus by Embodiment 7 of this invention.

Embodiment 1 FIG.
FIG. 1 is a block diagram showing an adaptive equalizer 100 according to Embodiment 1 of the present invention.
FIG. 2 is a configuration diagram showing an example of connection between the adaptive equalizer 100 shown in FIG. 1 and the unknown system 900 to be identified.
The adaptive equalizer 100 shown in FIG. 1 has an input signal x (n) input to the unknown system 900 and an output signal y (n) output from the unknown system 900 as inputs, and these two signals x (n ), Y (n), the transfer function of the unknown system 900 is learned and identified, and an output signal d (n) simulating the output signal y (n) of the unknown system 900 is output. The unknown system 900 is an acoustic transmission system including a speaker 901, an amplifier 902, and a microphone 903.

  As shown in FIG. 1, the adaptive equalizer 100 according to the present embodiment includes a first signal conversion unit 101, a filter processing unit 102, a signal inverse conversion unit 103, a subtractor 104, an adaptive learning processing unit 105, 2 signal conversion unit 106, frequency weight processing unit 107, and nonlinear distortion characteristic information holding unit 108. Here, the first signal conversion unit 101 to the second signal conversion unit 106 constitute a frequency domain adaptive filter unit, and the frequency weight processing unit 107 constitutes a frequency weighting unit.

Next, the operation of the adaptive equalizer 100 according to Embodiment 1 will be described.
The first signal conversion unit 101 receives an input signal x (n) of the unknown system 900 that is a time series signal, converts this into a signal expressed in the frequency domain, and a frequency domain input signal X (ω, k). Is output. Here, n represents the sample order of the time series signal. Further, ω represents the frequency, and k represents the block order of the frequency domain transformed signal. For example, DFT (Discrete Fourier Transform) is used for frequency conversion of the time series signal. Such conversion processing involves an operation of dividing the time series signal by a predetermined block length while performing arbitrary overlapping.

  The filter processing unit 102 receives the frequency domain input signal X (ω, k), multiplies it by the frequency domain filter coefficient H (ω, k), and outputs a frequency domain simulation signal D (ω, k). The frequency domain filter coefficient H (ω, k) is intended to simulate the transfer function of the unknown system 900 and is assumed to be held in the filter processing unit 102. The signal inverse conversion unit 103 receives the frequency domain simulation signal D (ω, k), converts it into a time-series signal, and outputs a simulation signal d (n). The simulation signal d (n) is a simulation of the output signal y (n) of the unknown system 900.

The subtractor 104 is obtained by receiving the simulated signal d (n) and the output signal y (n) of the unknown system 900, and subtracting the simulated signal d (n) from the output signal y (n) of the unknown system 900. An error signal e (n) is output. The second signal converter 106 converts the error signal e (n) into a signal expressed in the frequency domain, and outputs a frequency domain error signal E (ω, k). The conversion of the second signal conversion unit 106 performs conversion similar to that of the first signal conversion unit 101. The adaptive learning processing unit 105 receives the frequency domain input signal X (ω, k) and the frequency domain error signal E (ω, k), and uses the frequency domain filter coefficient H (ω, k) as a transfer function of the unknown system 900. The frequency domain filter coefficient update amount ΔH (ω, k) is output for the purpose of approaching. Upon receiving the frequency domain filter coefficient update amount ΔH (ω, k), the filter processing unit 102 updates the frequency domain filter coefficient H (ω, k) by updating as in the following equation (1).
H (ω, k + 1) = H (ω, k) + ΔH (ω, k) (1)

  As a suitable example of an adaptive filter that performs adaptive learning processing in such a frequency domain, FDAF (Frequency Domain Adaptive Filter) is, for example, Haykin, “Adaptive Filter Theory”, Science and Technology Publication, 2001 1 May 10, pp. 500-541.

  The frequency weight processing unit 107 receives the frequency domain input signal X (ω, k) and the frequency domain error signal E (ω, k), and further transfers the transfer function of the unknown system 900 from the nonlinear distortion characteristic information holding unit 108. Is read out, and the frequency weight W (ω) is output based on the information. In the adaptive equalizer 100 of the present embodiment, the nonlinear distortion information held in the nonlinear distortion characteristic information holding unit 108 is the j-order harmonic included in the output signal of the unknown system 900 with respect to the input signal of the frequency ω. It is assumed that the harmonic power ratio A (ω, j) represents the power ratio of the waves. Actually, the power of the input signal and the higher harmonics is not necessarily a constant ratio, and therefore, here, a value sufficiently larger than the observed power ratio is assumed.

ここで、Ｍは所定の定数であり、高次高調波の計算を適当な次数によって打ち切るためのものである。 The frequency weighting processing unit 107 calculates the predicted harmonic power spectrum B (ω, k) from the frequency domain input signal X (ω, k) based on the harmonic power ratio A (ω, j), for example, ).

Here, M is a predetermined constant, and is used to terminate the calculation of high-order harmonics by an appropriate order.

周波数重み処理部１０７は、この比率Ｐ（ω，ｋ）に従い、比率Ｐ（ω，ｋ）が高いほど、周波数重みＷ（ω）が小さくなるように定める。これには、一例として、比率Ｐ（ω，ｋ）に対して所定の閾値を設け、０＜ａ＜ｂの関係をなす所定の重みａ，重みｂに対し、比率Ｐ（ω，ｋ）が上記閾値を超える周波数ωの重みは重みａ、超えない周波数ωの重みは重みｂと定めるといった方法を用いるが、他の重み付け方法であってもよい。 Further, the frequency weight processing unit 107 obtains a ratio P (ω, k) between the predicted harmonic power spectrum B (ω, k) and the frequency domain error signal E (ω, k) as in the following equation (3). .

The frequency weight processing unit 107 determines that the frequency weight W (ω) becomes smaller as the ratio P (ω, k) is higher in accordance with the ratio P (ω, k). As an example, a predetermined threshold value is provided for the ratio P (ω, k), and the ratio P (ω, k) is set for a predetermined weight a and weight b having a relationship of 0 <a <b. A method is used in which the weight of the frequency ω exceeding the threshold is set as the weight a, and the weight of the frequency ω not exceeding the weight ω is set as the weight b.

  The frequency weight W (ω) is given to the adaptive learning processing unit 105, and the adaptive learning processing unit 105 multiplies the frequency domain filter coefficient update amount ΔH (ω, k) by this to thereby calculate the ratio of the harmonic components. The update at a high frequency is suppressed, and the accuracy of identification is prevented from being lowered due to misadaptation.

  As described above, in the first embodiment, the coefficient is updated at a frequency having a high ratio P (ω, k) according to the ratio P (ω, k) of the harmonic components included in the frequency domain error signal E (ω, k). By suppressing the above, it is possible to prevent the adaptive equalizer from misadapting to the harmonic component due to the nonlinear element of the unknown system 900, and to improve the identification accuracy.

  As described above, according to the adaptive equalizer 100 of the first embodiment, the filter coefficient expressed in the frequency domain is held, and based on this filter coefficient, the input signal of the target unknown system 900 and the unknown The frequency domain adaptive filter means for performing adaptive learning from the output signal of the system 900 to identify the transfer function of the unknown system 900, and the frequency domain adaptive filter means based on the nonlinear distortion characteristic information regarding the nonlinear distortion characteristics of the unknown system 900. Since the frequency weighting means for multiplying the filter coefficient or the update amount of the filter coefficient by the weight is provided, it is possible to prevent a decrease in the identification accuracy of the linear element due to the presence of the nonlinear element of the transfer function.

  Further, according to the adaptive equalizer 100 of the first embodiment, the nonlinear distortion specifying information indicates the nature of the nonlinear distortion that appears in the output signal of the unknown system 900 with respect to the frequency of the input signal of the unknown system 900. Therefore, it is possible to prevent the identification accuracy from being lowered due to misadaptation.

  Moreover, according to the adaptive equalizer 100 of the first embodiment, the nonlinear distortion characteristic information is obtained by using the power ratio between the input signal of the unknown system 900 and each harmonic component included in the output signal of the unknown system 900 corresponding thereto. Therefore, it is possible to prevent erroneous adaptation of the adaptive equalizer with respect to the harmonic component due to the nonlinear element of the unknown system 900, and to improve the identification accuracy.

  Further, according to the adaptive equalizer 100 of the first embodiment, the frequency weighting means estimates the power of the harmonic contained in the output signal of the unknown system 900 according to the input signal of the unknown system 900 and the nonlinear distortion characteristic information. The frequency domain adaptive filter means is weighted so that the coefficient update amount of the filter coefficient is reduced at frequencies where the harmonic power is large. It is possible to prevent the identification accuracy from being lowered due to misadaptation.

Embodiment 2. FIG.
The adaptive equalizer 100 of the first embodiment multiplies the predicted harmonic power spectrum B (ω, k) by multiplying the frequency domain input signal X (ω, k) by the harmonic power ratio A (ω, j). I was asking. However, in order to suppress the information load of the harmonic power ratio A (ω, j) and the calculation load necessary for obtaining the predicted harmonic power spectrum B (ω, k) to a range that can be practically realized as a device. The order of the predicted harmonics needs to be appropriately cut off by a predetermined constant M. However, by performing this truncation, there is a problem that the higher-order harmonic components affect the adaptive learning, thereby lowering the identification accuracy. In the second embodiment, an adaptive equalizer that has a configuration suitable for implementation and does not require truncation of high-order harmonic components will be described.

  FIG. 3 is a block diagram showing a configuration of adaptive equalizer 200 according to Embodiment 2 of the present invention. Parts that are the same as or equivalent to those in FIG. The difference between the adaptive equalizer 200 and the adaptive equalizer 100 of the first embodiment is that the input signal x (n) passes through the distortion adding unit 109, the subtractor 110, and the third signal converting unit 111 and is predicted harmonic. The point converted to the power spectrum B (ω, k) and input to the frequency weight processing unit 107, and the nonlinear distortion characteristic information holding unit, as nonlinear distortion characteristic information to the distortion adding unit 109, are input / output characteristic function f (x ).

Next, portions of the operation of adaptive equalizer 200 according to Embodiment 2 that are different from adaptive equalizer 100 of Embodiment 1 will be described with reference to FIG.
An input signal x (n) to the unknown system 900 is input to the distortion adding unit 109. The distortion adding unit 109 reads the input / output characteristic function f (x) modeling the nonlinear distortion of the unknown system 900 from the nonlinear distortion characteristic information holding unit 108, and outputs the input signal x ′ (n) with distortion. Here, x ′ (n) = f (x ′ (n)).

  The input / output characteristic function f (x) is a function that determines an output amplitude value with respect to an input amplitude value, and is a sigmoid as shown in FIG. This sigmoid approximates the amplitude saturation characteristics of the speaker / amplifier included in the unknown system 900. By this, the distorted input signal x ′ (n) simulates the harmonic component due to the amplitude saturation characteristics of the speaker / amplifier. Signal is included. Here, as the input / output characteristic function f (x), a function approximated by using polynomial approximation or the like from the measured value of the amplitude saturation characteristic of the speaker / amplifier can be used. In addition, the process which provides such arbitrary input / output characteristics to a signal is generally known as a function called DRP (Dynamic Range Processing) in acoustic signal processing.

In the subtractor 110, the input signal x (n) is subtracted from the distorted input signal x ′ (n), and a predicted harmonic signal q (n) obtained by extracting only the harmonic component is output. The predicted harmonic signal q (n) is input to the third signal converter 111. The third signal converter 111 converts the predicted harmonic signal q (n) into a frequency domain signal, and further converts the power spectrum thereof. The calculated predicted harmonic power spectrum B (ω, k) is output.
The frequency weight processing unit 107 determines the frequency weight W (ω, k) based on the predicted harmonic power spectrum B (ω, k) output from the third signal conversion unit 111.
These processes added in the second embodiment are often used in general acoustic signal processing, and the calculation load is within a scale that can be practically realized as a device.

  As described above, in the second embodiment, by using the input / output characteristic function f (x) simulating the saturation characteristic of the speaker / amplifier, the predicted harmonic power spectrum B (ω, k) is derived, thereby reducing a smaller calculation load. Thus, it is possible to obtain an adaptive equalizer that enables higher-precision identification by predicting the higher-order harmonic components without truncating them and reflecting the frequency weights based on the predictions in the adaptive learning process.

  As described above, according to the adaptive equalizer 200 of the second embodiment, since the nonlinear distortion characteristic information is the input / output characteristic of the unknown system 900, the transfer function of the unknown system 900 is identified with high accuracy. Can do.

Embodiment 3 FIG.
The adaptive equalizer 100 according to the first embodiment operates the frequency domain filter coefficient update amount ΔH (ω, k) according to the distortion component ratio, but the frequency domain filter coefficient H (ω, k) itself. The filter gain may not be provided in a specific frequency band. In particular, a loudspeaker is known to cause a large nonlinear distortion with respect to an input signal near the reproduction lower limit frequency. In such a frequency band, it is difficult to identify the transfer characteristic itself. For example, it is better not to identify for the purpose of active noise control. In the third embodiment, an adaptive equalizer that does not have a filter gain in a frequency band with large distortion will be described.

  FIG. 5 is a block diagram showing a configuration of an adaptive equalizer 300 according to Embodiment 3 of the present invention. Parts that are the same as or equivalent to those in FIG. The difference between the adaptive equalizer 300 and the adaptive equalizer 100 of the first embodiment is that the frequency weight processing unit 107 has a frequency domain input signal X (ω, k) and a frequency domain error signal E (ω, k), and the frequency weight W (ω) is given to the filter processing unit 102.

Next, portions of the operation of adaptive equalizer 300 according to Embodiment 3 that are different from adaptive equalizer 100 of Embodiment 1 will be described with reference to FIG.
The information stored in the nonlinear distortion characteristic information holding unit 108 is the distortion rate A (ω) of the unknown system 900, and the frequency weight processing unit 107 uniquely assigns the frequency weight W (ω) to A (ω). Determine. FIG. 6 shows an example of setting frequency weights according to the distortion rate. As shown in FIG. 6, the frequency weight processing unit 107 decreases the weight in a frequency band with a large distortion rate and increases the weight in a large frequency band according to the distortion rate of the unknown system 900. As a specific procedure, for example, a predetermined threshold value is set for the distortion rate, and for a predetermined weight a and weight b having a relationship of 0 <a <b, a frequency band in which a distortion rate exceeding the threshold value exists is as follows. Although the method of determining the weight a and the frequency band not exceeding as the weight b is used, other weighting methods may be used.

  The frequency weight W (ω) is given to the filter processing unit 102, and the filter processing unit 102 multiplies the frequency domain filter coefficient H (ω, k) by the frequency weight W (ω) to thereby obtain the distortion rate of the unknown system 900. The gain in the frequency band having a large value is limited.

  As described above, the third embodiment avoids imperfect identification by limiting the gain of the frequency domain filter in a frequency band in which the unknown system 900 has a large distortion rate and identification accuracy cannot be expected. The effect of improving the identification accuracy is obtained.

  As described above, according to adaptive equalizer 300 of Embodiment 3, since the nonlinear distortion characteristic information is the distortion rate of the output signal of unknown system 900 relative to the input signal of unknown system 900, incomplete identification is performed. Can be avoided, and as a result, the overall identification accuracy can be improved.

  Further, according to adaptive equalizer 300 of the third embodiment, the frequency weighting means, for each frequency, the filter of the frequency domain adaptive filter means so that the gain of the filter at that frequency is reduced as the distortion rate increases. Since the coefficient is multiplied by the weight, the gain of the unknown system 900 in the frequency band with a large distortion rate is limited, and it is possible to prevent the identification accuracy from being lowered due to erroneous adaptation.

Embodiment 4 FIG.
The adaptive equalizer 100 according to the first embodiment assumes that the nonlinear distortion information held in the nonlinear distortion characteristic information holding unit 108 is given in advance, but the adaptive equalizer itself acquires the information. By doing so, for example, when there is a secular change of the speaker, it is possible to always correspond to the latest state. In the fourth embodiment, such an adaptive equalizer will be described.

  FIG. 7 is a block diagram showing a configuration of an adaptive equalizer 400 according to Embodiment 4 of the present invention. Parts that are the same as or equivalent to those in FIG. The adaptive equalizer 400 differs in configuration from the adaptive equalizer 100 of the first embodiment in that it includes a switch 112, a switch 113, and a distortion measurement unit 114.

  FIG. 8 is a block diagram showing an example of connection between the adaptive equalizer 400 and the unknown system 900 according to the fourth embodiment. In FIG. 8, the adaptive equalizer 400 outputs the measurement signal r (n) in addition to the simulation signal d (n), and the measurement signal r (n) is superimposed on the input signal x (n) and unknown. Input to the system.

Next, portions of the operation of adaptive equalizer 400 according to Embodiment 4 that are different from adaptive equalizer 100 of Embodiment 1 will be described with reference to FIGS. 7 and 8.
The strain measuring unit 114 constitutes a non-linear strain characteristic information measuring unit. When the measuring operation is started by an external operation or the like, an instruction is sent to the switch 112 and the switch 113 to change the signal path. That is, the switch 112 opens the circuit when receiving an instruction for the measurement operation, and prevents the input signal x (n) of the unknown system 900 from being input to the first signal conversion unit 101. Similarly, the switch 113 switches the destination of the output signal y (n) of the unknown system 900 from the subtracter 104 to the distortion measurement unit 114. In this way, no filtering process or adaptive learning is performed while the distortion measuring unit 114 is performing a measurement operation.

  The distortion measurement unit 114 outputs a predetermined measurement signal r (n) after switching the signal path in this way. The measurement signal r (n) is superimposed on the input signal x (n) of the unknown system 900 and input to the unknown system 900, as shown in FIG. At this time, in order to perform measurement with higher accuracy, if the input signal x (n) of the unknown system 900 is controllable, x (n) = 0 is set, and only the measurement signal r (n) is obtained. Input to unknown system 900. When control is not possible, the measurement signal r (n) is set to a level sufficiently higher than the input signal x (n) of the unknown system 900.

  While the measurement signal r (n) is input, the output signal y (n) of the unknown system 900 is input to the adaptive equalizer 400, and is input to the distortion measurement unit 114 via the switch 113 inside. . The distortion measurement unit 114 acquires information about distortion from the output signal y (n) with respect to the measurement signal r (n), and stores the information in the nonlinear distortion characteristic information holding unit 108.

At this time, the desired distortion characteristic information is information on the power ratio of the j-th harmonic included in the output signal of the unknown system 900 with respect to the input signal of frequency ω used in the first embodiment, or in the third embodiment. If it is the distortion rate used, the measurement signal r (n) uses a frequency sweep signal that sweeps a desired frequency band, the spectrum analysis of the output signal y (n) of the unknown system 900 is performed, and the harmonic signal These can be easily obtained by measuring power.
In the case of the input / output characteristics used in the second embodiment, a sine wave having a predetermined frequency is used as the measurement signal r (n), and this is monotonically increased in a desired amplitude range to obtain the measurement signal r ( n) and the power ratio of the output signal y (n) of the unknown system 900 can be easily obtained.

  As described above, the fourth embodiment has the function of measuring the nonlinear distortion characteristic information, so that even if the nonlinear distortion characteristic of the unknown system changes, it is not necessary to provide the nonlinear distortion characteristic information again, and the adaptation can be performed by itself. There is an effect that it becomes possible.

  As described above, according to adaptive equalizer 400 of Embodiment 4, a measurement signal is input to unknown system 900, and nonlinear distortion characteristic information is collected based on the output signal of unknown system 900 corresponding thereto. Since the nonlinear distortion characteristic information measuring means is provided, it is possible to always deal with the latest state even when the unknown system 900 changes with time.

Embodiment 5 FIG.
In the fifth embodiment, a method for using an adaptive equalizer suitable for an acoustic echo canceller apparatus will be described. FIG. 9 is a block diagram showing a configuration of an acoustic echo canceller apparatus according to the fifth embodiment of the present invention. The acoustic echo canceller apparatus shown in FIG. 9 includes an adaptive equalizer 500. The adaptive equalizer 500 includes an adaptive equalizer 100 according to the first embodiment, an adaptive equalizer 200 according to the second embodiment, In adaptive equalizer 400 according to the fourth embodiment, whether the distortion characteristic information is information on the power ratio of the j-th harmonic included in the output signal of unknown system 900 with respect to the input signal of frequency ω used in the first embodiment. Of the input / output characteristics used in the second embodiment, those of any of the embodiments can be used.

The acoustic echo canceller of FIG. 9 collects and collects an amplifier 902 that amplifies the received signal x (n) with a predetermined gain, a speaker 901 driven by the amplifier 902, and a user's conversational voice. A microphone 903 for converting to a sound signal y (n) and a subtractor 501 described later are provided.
The speaker 901, the amplifier 902, and the microphone 903 constitute an unknown system 900. The received signal x (n) is an input signal to the unknown system 900, and the collected sound signal y (n) is an output from the unknown system 900. It is a signal. In the unknown system 900, the received voice output from the speaker 901 is picked up by the microphone 903 as an echo, so the picked-up signal y (n) includes both the conversation voice and the echo of the user. ing.

  Therefore, the adaptive equalizer 500 calculates the transfer function of the unknown system 900 from the received signal x (n) that is an input signal to the unknown system 900 and the collected sound signal y (n) that is an output signal from the unknown system 900. A simulation signal d (n) that identifies and simulates an echo is output. As described above, the collected sound signal y (n) includes not only the echo that is the response of the unknown system 900 to the received signal x (n) but also the conversation voice of the user, so that accurate identification is possible. In order to perform the above, it is necessary to execute adaptive learning in a state where only the echo is included in the collected sound signal y (n). For this reason, generally, the execution / stop of learning of the adaptive equalizer 500 is controlled using a determiner that determines this state, but this is a known technique and is not a part related to the essence of the present invention. Therefore, in this embodiment, the description is omitted in 5.

  The subtractor 501 constitutes an echo removal means, and subtracts the simulated signal d (n) from the collected sound signal y (n) and outputs a transmission signal y ′ (n). Since the simulation signal d (n) is a signal simulating an echo, the subtraction process removes the echo and realizes a function as an acoustic echo canceller.

  In this acoustic echo canceller, the adaptive equalizer 500 uses the nonlinear distortion characteristic information of the speaker 901 and the amplifier 902 of the unknown system 900 to predict a frequency band with a high ratio of harmonic components in the collected sound signal y (n). However, by suppressing the update amount of the frequency domain adaptive filter coefficient in such a frequency band, it is possible to prevent erroneous adaptation due to the influence of the harmonic component and to increase the identification accuracy.

  As described above, in the fifth embodiment, the frequency domain adaptive filter coefficient update amount is suppressed in the frequency band having a high ratio of harmonic components, and the identification accuracy is improved by preventing erroneous adaptation, thereby improving the echo cancellation effect. Is obtained.

  As described above, according to the echo canceller of Embodiment 5, the speaker 901 that reproduces the received signal as the received sound, the amplifier 902 that drives the speaker 901, and the microphone 903 that acquires the transmitted sound as the transmitted signal. And the non-linear distortion characteristic of the speaker 901 and the amplifier 902 is adapted as non-linear distortion characteristic information using the adaptive equalizer 500 of the first or second embodiment. The transfer function of the echo path is identified by using the received signal as the input signal of the unknown system 900 and the transmitted signal as the output signal of the unknown system 900, outputting a pseudo echo signal, and from the transmitted signal. Since the echo removing means for removing the pseudo echo signal is provided, an echo canceling device having a high echo canceling effect can be obtained.

  Further, according to the echo canceling apparatus of the fifth embodiment, adaptive equalizer 500 uses the power of the harmonic component of the echo included in the transmitted signal according to the received signal and the nonlinear distortion characteristic information of speaker 901 and amplifier 902. The frequency domain adaptive filter means is multiplied by a weight so that the coefficient update amount of the filter coefficient is reduced at frequencies where the harmonic power is large. Can be suppressed, and the accuracy of identification can be prevented from deteriorating due to misadaptation.

Embodiment 6 FIG.
In the sixth embodiment, a method of using an adaptive equalizer suitable for a feedforward active noise control apparatus will be described. FIG. 10 is a block diagram showing the configuration of an active noise control apparatus for illustrating an example of usage of an adaptive equalizer according to Embodiment 6 of the present invention. The active noise control apparatus shown in FIG. 10 includes an adaptive equalizer 600. The adaptive equalizer 600 includes the adaptive equalizer 300 according to the third embodiment or the adaptive equalizer 400 according to the fourth embodiment. In this case, any of the strain characteristic information that is the distortion rate used in the third embodiment can be used.

In addition to this, the active noise control device of FIG. 10 is installed at a reference microphone 601, a speaker 602 that outputs a control sound for controlling noise, and a target position. Error microphone 603 for observing noise, control filter 604 for generating control sound, amplifier 605 for driving speaker 602, and frequency domain filter coefficient H (ω, k) converted to time domain filter coefficient h (i, k) And a filter coefficient transfer means 606 for transferring to the control filter 604, a secondary path characteristic filter 607, and a phase inverter 608 for inverting the phase of the signal.
A noise source 904 in FIG. 10 indicates a noise generation source, a primary path 905 is a sound wave path from the noise source 904 to the target position, and a secondary path 906 is a sound wave path from the speaker 602 to the target position. Is shown.

Next, the operation of the active noise control apparatus of FIG. 10 and the method of using the adaptive equalizer according to Embodiment 6 will be described. Noise generated by the noise source 904 is collected by the reference microphone 601 and becomes a reference signal. The reference signal is filtered by the secondary path characteristic filter 607 simulating the transfer function of the secondary path 906, and further phase-inverted by the phase inverter 608 to become the target system input signal x (n), and the adaptive equalizer 600 Is input. Further, the noise generated from the noise source 904 is collected by the error microphone 603 through the primary path 905 and input to the adaptive equalizer 600 as the target system output signal y (n). The adaptive equalizer 600 identifies the transfer function of the unknown system using the target system input signal x (n) as the input signal of the unknown system and the target system output signal y (n) as the output signal of the unknown system. When the transfer function of the unknown system is G as the transfer function of the primary path and C as the transfer function of the secondary path,
Transfer function of unknown system = -G / C
It is expressed.

This transfer function is known to be an optimal transfer function of a control filter in a feedforward type active noise control apparatus. A method for identifying the transfer function of the virtual unknown system using an adaptive equalizer using the target system input signal x (n) and the target system output signal is called a filtered-x method.
Therefore, the filter coefficient transfer means 606 converts the frequency domain filter coefficient H (ω, k) representing the transfer function identified by the adaptive equalizer 600 into the time domain filter coefficient h (i, k), and controls the control filter. 604. Here, i represents the filter order. For the conversion from the frequency domain filter coefficient H (ω, k) to the time domain filter coefficient h (i, k), an inverse conversion of the frequency conversion, for example, an inverse DFT may be used. The control filter 604 receives the reference signal, filters it using the time domain filter coefficient h (i, k) given from the filter coefficient transfer means 606, and outputs a control sound signal. The control sound signal is output from the speaker 602 through the amplifier 605, propagates to the target position through the secondary path 906, and the noise and the control sound are completely out of phase at the target position, thereby canceling each other and reducing the noise.

  At this time, if the noise emitted from the noise source 904 includes a noise component having a frequency near or lower than the reproduction lower limit frequency of the speaker 602, the control sound for canceling this noise is distorted by the speaker 602, and the noise is generated. In addition to being unable to reduce sufficiently, there is a problem that extra sounds such as harmonics are added. For this reason, the adaptive equalizer 600 has a gain in a frequency band having a high distortion rate and lower than the vicinity of the reproduction lower limit frequency based on the distortion rate information held therein as the nonlinear distortion characteristic information of the speaker. The frequency domain filter coefficient H (ω, k) is weighted so as not to be present. By doing so, the control filter 604 has no gain in the frequency band, and as a result, the control sound does not include the signal component of the frequency band, and the speaker 602 can be prevented from distorting the control sound. .

  Note that in the method of using the adaptive equalizer of the sixth embodiment, the adaptive equalizer 400 of the fourth embodiment is used, and the measurement signal r (n) is output from the adaptive equalizer 400 to obtain nonlinear distortion characteristic information. Is measured, the measurement signal r (n) is added to the control sound signal input to the amplifier 605. At this time, the control sound signal may be muted.

  As described above, in the sixth embodiment, the frequency domain filter coefficient H (ω, k) is weighted inside the adaptive equalizer 600 so that there is no gain in the filter in the frequency band where the speaker distortion is large. Thus, the effect of suppressing the distortion of the control sound and enhancing the noise reduction effect of the active noise control device can be obtained.

  As described above, according to the active noise control apparatus of the sixth embodiment, the reference microphone 601 that collects the noise source signal in the vicinity of the noise source and outputs it as the reference signal, and the target position for reducing the noise are installed. An error microphone 603 that picks up a noise signal at this position as a target system output signal, a control filter 604 that outputs a control sound for controlling noise using the noise source signal as an input, and a speaker 602 that reproduces the control sound An amplifier 605 for driving the speaker 602, a secondary path characteristic filter 607 having a secondary path transfer function from the speaker 602 to the error microphone 603, filtering the reference signal and outputting it as an intermediate signal, and an intermediate signal A phase inverter 608 that inverts the phase and outputs it as a target system input signal, and nonlinear distortion characteristics of the speaker 602 and the amplifier 605. Is applied as nonlinear distortion characteristic information, and the transfer function of an unknown system that identifies the target system input signal as an input and outputs the target system output signal from the target system input signal and the target system output signal, etc. is applied. And a filter coefficient transfer means 606 that converts the transfer function derived by the adaptive equalizer 600 into a filter coefficient and outputs the filter coefficient to the control filter 604. The control filter 604 controls the control sound based on the filter coefficient. Therefore, an active noise control device with a high noise reduction effect can be obtained.

  Further, according to the active noise control device of the sixth embodiment, the frequency domain adaptive filter means reduces the gain of the control filter 604 at a frequency with a large distortion according to the distortion rate of the speaker 602 and the amplifier 605 for each frequency. Since the filter coefficient is multiplied by the weight, the gain in the frequency band having a large distortion rate is limited, and it is possible to prevent the identification accuracy from being lowered due to erroneous adaptation.

Embodiment 7 FIG.
In the seventh embodiment, a method of using an adaptive equalizer suitable for a feedback type active noise control apparatus will be described. FIG. 11 is a block diagram showing the configuration of an active noise control apparatus for illustrating an example of usage of an adaptive equalizer according to Embodiment 7 of the present invention. The active noise control apparatus shown in FIG. 11 includes an adaptive equalizer 700. The adaptive equalizer 700 includes the adaptive equalizer 300 according to the third embodiment or the adaptive equalizer 400 according to the fourth embodiment. In this case, any of the strain characteristic information that is the distortion rate used in the third embodiment can be used.
In addition to this, the active noise control apparatus of FIG. 11 includes a speaker 701, an amplifier 702, a reference microphone 703, a first secondary path characteristic filter 704, a subtractor 705, a control filter 706, a filter coefficient transfer means 707, a second second A next path characteristic filter 708 is provided.
In FIG. 11, a noise source 904 indicates a noise generation source, and a secondary path 906 indicates a sound wave path from the speaker 701 to the target position.

  Next, the operation of the active noise control apparatus in FIG. 11 and the method of using the adaptive equalizer according to Embodiment 7 will be described. Noise generated by the noise source 904 is collected by a reference microphone 703 installed at a target position, and becomes a reference signal. This reference signal is subjected to subtraction processing described later in the subtractor 705, and then input to the adaptive equalizer 700 as the target system output signal y (n). The target system output signal y (n) is also input to the second secondary path characteristic filter 708, and the second secondary path characteristic filter 708 filters the target system input signal x (n). Output. The second secondary path characteristic filter 708 performs the same operation as the secondary path characteristic filter 607 described in the sixth embodiment. The adaptive equalizer 700 identifies the transfer function using the target system input signal x (n) as an unknown system input signal and the target system output signal y (n) as an unknown system output signal. It is known that the identified transfer function becomes a transfer function for generating a control sound from a reference signal in a feedback type active noise control apparatus.

Therefore, the filter coefficient transfer unit 707 converts the frequency domain filter coefficient H (ω, k) of the adaptive equalizer 700 into a time domain filter coefficient h (i, k), and supplies the time domain filter coefficient h (i, k) to the control filter 706. Filters the target system output signal y (n) using the time domain filter coefficient h (i, k) and outputs a control sound signal. The control sound signal is output as a control sound through the amplifier 702 and the speaker 701, and this control sound propagates to the target position via the secondary path 906, has an opposite phase to the incoming noise at the target position, and cancels each other to generate noise. To reduce.
At this time, since not only the incoming noise but also the control sound is picked up by the reference microphone 703, the first secondary path characteristic filter 704 filters the control sound signal output from the control filter 706, and cancels the signal. Is output. Here, the operation of the first secondary path characteristic filter 704 is the same as the operation of the second secondary path characteristic filter 708. The cancellation signal is input to the subtracter 705, and the subtractor 705 subtracts the cancellation signal from the reference signal, thereby removing the control sound component included in the reference signal.

  At this time, as in the feedforward active noise control apparatus of the sixth embodiment, the noise emitted from the noise source 904 includes a noise component having a frequency near or lower than the reproduction lower limit frequency of the speaker 701. Then, the control sound that cancels the distortion is distorted by the speaker 701, and not only the noise cannot be sufficiently reduced, but also an extra sound such as a harmonic is added. Further, in the feedback active noise control apparatus handled in the seventh embodiment, there is a problem that the control sound distortion component cannot be removed even in the control sound component subtraction process in the subtractor 705. As a result, when the distortion component that could not be removed is input to the control filter 706, a more incorrect control sound is generated, causing a distortion component loop, which prevents the active noise control device from performing its function. There is.

  For this reason, the adaptive equalizer 700 has a gain in a frequency band having a high distortion rate and lower than the vicinity of the reproduction lower limit frequency based on the distortion rate information held therein as the nonlinear distortion characteristic information of the speaker. The frequency domain filter coefficient H (ω, k) is weighted so as not to be present. By doing so, the control filter 706 has no gain in the frequency band, and as a result, the control sound does not include the signal component of the frequency band, and the distortion of the control sound in the speaker 701 and the resulting distortion component It is possible to prevent a loop from occurring.

  Note that in the method of using the adaptive equalizer of the seventh embodiment, the adaptive equalizer 400 of the fourth embodiment is used, and the measurement signal r (n) is output from the adaptive equalizer 400 to obtain nonlinear distortion characteristic information. Is measured, the measurement signal r (n) is added to the control sound signal input to the amplifier 702. At this time, the control sound signal may be muted.

  As described above, in the seventh embodiment, the frequency domain filter coefficient H (ω, k) is weighted inside the adaptive equalizer 700 so as not to have a gain in a filter in a frequency band with a large speaker distortion. Thus, the effect of suppressing the distortion of the control sound and enhancing the noise reduction effect of the active noise control device can be obtained.

  As described above, according to the active noise control apparatus of the seventh embodiment, the reference microphone 703 is installed at the target position where noise is reduced, and the noise signal at this position is collected as the reference signal. The subtractor 705 for subtracting the canceling signal of the target system and outputting the target system output signal, the control filter 706 for outputting the control sound signal for controlling the noise with the reference signal subtracted by the subtractor 705 as input, A first path that has a secondary path transfer function from the speaker 701 to the reference microphone 703 and outputs a predetermined canceling signal by filtering the control sound signal. The secondary path characteristic filter 704 has a secondary path transfer function from the speaker 701 to the reference microphone 703, and the subtracted reference signal is filtered. A second secondary path characteristic filter 708 that outputs the target system input signal and the nonlinear distortion characteristics of the speaker 701 and the amplifier 702 are given as nonlinear distortion characteristic information. From the target system input signal and the target system output signal, The adaptive equalizer 700 of Embodiment 3 that identifies the transfer function of an unknown system that receives the target system input signal and outputs the target system output signal, and the transfer function derived by the adaptive equalizer 700 as filter coefficients Filter coefficient transfer means 707 for converting and outputting to the control filter 706, and the control filter 706 generates the control sound based on the filter coefficient, so that an active noise control device having a high noise reduction effect can be obtained. Can do.

  100, 200, 300, 400, 500, 600, 700 Adaptive equalizer, 101 First signal conversion unit, 102 Filter processing unit, 103 Signal inverse conversion unit, 104 Subtractor, 105 Adaptive learning processing unit, 106 2 signal conversion unit, 107 frequency weighting processing unit, 108 nonlinear distortion characteristic information holding unit, 109 distortion addition unit, 110,705 subtractor, 111 third signal conversion unit, 112 switch, 113 switch, 114 distortion measurement 601 703 Reference microphone 602 701 901 Speaker 603 Error microphone 604 706 Control filter 605 702 902 Amplifier 606 707 Filter coefficient transfer means 607 Secondary path characteristic filter 608 Phase inversion 704, first secondary path characteristic filter, 708 second secondary path characteristic Filter, 900 an unknown system, 903 a microphone, 904 a noise source, 905 the primary path, 906 secondary path.

Claims (13)

Holds the filter coefficient expressed in the frequency domain, and based on the filter coefficient, performs adaptive learning from the input signal of the target unknown system and the output signal of the unknown system, and identifies the transfer function of the unknown system Frequency domain adaptive filter means for
An adaptive equalizer comprising frequency weighting means for multiplying a weight of the filter coefficient of the frequency domain adaptive filter means or an update amount of the filter coefficient based on nonlinear distortion characteristic information related to the nonlinear distortion characteristic of the unknown system.   2. The adaptive equalizer according to claim 1, wherein the nonlinear distortion specifying information indicates a property of nonlinear distortion appearing in an output signal of the unknown system with respect to a frequency of an input signal of the unknown system.   3. The adaptation or the like according to claim 1, wherein the nonlinear distortion characteristic information is a power ratio between an input signal of the unknown system and each harmonic component contained in the output signal of the unknown system. Generator.   The adaptive equalizer according to claim 1, wherein the nonlinear distortion characteristic information is an input / output characteristic of an unknown system.   3. The adaptive equalizer according to claim 1, wherein the nonlinear distortion characteristic information is a distortion rate of an output signal of the unknown system with respect to an input signal of the unknown system.   The frequency weighting means estimates the power of the harmonic contained in the output signal of the unknown system according to the input signal of the unknown system and the nonlinear distortion characteristic information, and at a frequency where the power of the harmonic is large compared to the frequency domain adaptive filter means. 5. The adaptive equalizer according to claim 1, wherein a weight is multiplied so that a coefficient update amount of the filter coefficient is reduced.   The frequency weighting means, for each frequency, multiplies the filter coefficient of the frequency domain adaptive filter means by a weight so that the gain of the filter of the frequency decreases as the distortion rate increases. 6. An adaptive equalizer according to any one of claims 2 and 5.   8. A non-linear distortion characteristic information measuring unit for inputting a measurement signal to an unknown system and collecting non-linear distortion characteristic information based on an output signal of the unknown system corresponding to the measurement signal. The adaptive equalizer as described in any one of them. An echo path having a speaker that reproduces a received signal as a received sound, an amplifier that drives the speaker, and a microphone that acquires a transmitted sound as a transmitted signal is an unknown system,
Using the adaptive equalizer according to any one of claims 1 to 4 and claim 6, nonlinear distortion characteristics of the speaker and the amplifier are stored in the adaptive equalizer as nonlinear distortion characteristic information. And identifying the transfer function of the echo path using the received signal as an input signal of the unknown system and the transmitted signal as an output signal of the unknown system, and outputting a pseudo echo signal,
An echo canceller comprising echo removing means for removing the pseudo echo signal from the transmission signal.   The adaptive equalizer estimates the power of the harmonic component of the echo included in the transmitted signal according to the received signal and the non-linear distortion characteristic information of the speaker and the amplifier, and supplies the harmonic power to the frequency domain adaptive filter means. 10. The echo canceller according to claim 9, wherein a weight is multiplied so that a coefficient update amount of the filter coefficient is reduced at a frequency having a large value. A reference microphone that picks up a noise source signal in the vicinity of the noise source and outputs it as a reference signal;
An error microphone that is installed at a target position for reducing noise and picks up a noise signal at the position as a target system output signal;
A control filter for outputting a control sound for controlling the noise, using the noise source signal as an input;
A speaker for reproducing the control sound;
An amplifier for driving the speaker;
A secondary path characteristic filter having a transfer function of a secondary path from the speaker to the reference microphone, and filtering the reference signal to output as an intermediate signal;
A phase inverter that inverts the phase of the intermediate signal and outputs it as a target system input signal;
An unknown system in which nonlinear distortion characteristics of the speaker and the amplifier are given as nonlinear distortion characteristic information, and the target system input signal is input from the target system input signal and the target system output signal, and the target system output signal is output. An adaptive equalizer according to claim 5 for identifying a transfer function of:
Filter coefficient transfer means for converting a transfer function derived by the adaptive equalizer into a filter coefficient and outputting the filter coefficient to the control filter;
The active noise control device, wherein the control filter generates a control sound based on the filter coefficient. A reference microphone installed at a target position for reducing noise and collecting a noise signal at the position as a reference signal;
A subtractor that subtracts a predetermined cancellation signal from the reference signal and outputs a target system output signal;
A control filter that outputs a control sound signal for controlling the noise, using the reference signal subtracted by the subtractor as an input;
A speaker for reproducing the control sound signal;
An amplifier for driving the speaker;
A first secondary path characteristic filter having a transfer function of a secondary path from the speaker to the reference microphone and filtering the control sound signal and outputting the predetermined cancellation signal;
A second secondary path characteristic filter having a transfer function of a secondary path from the speaker to the reference microphone and filtering the subtracted reference signal to output a target system input signal;
An unknown system in which a nonlinear distortion characteristic of the speaker and the amplifier is given as nonlinear distortion characteristic information, and the target system output signal is output from the target system input signal and the target system output signal. An adaptive equalizer according to claim 5 for identifying a transfer function of:
Filter coefficient transfer means for converting a transfer function derived by the adaptive equalizer into a filter coefficient and outputting the filter coefficient to the control filter;
The active noise control device, wherein the control filter generates a control sound based on the filter coefficient.   13. The filter coefficient of the frequency domain adaptive filter means is weighted so as to reduce the gain of the control filter at a frequency with a large distortion according to the distortion rate of the speaker and amplifier for each frequency. The active noise control apparatus as described.

Images