JP2924500B2 - Adaptive control device and active noise control device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an adaptive control device capable of controlling the behavior of an object to be controlled even when its behavior is unknown or fluctuating, and to emit a noise transmitted from a noise source from a control sound source. Active noise control device that aims to reduce noise by interfering control sound, especially
An apparatus comprising: a digital filter having a variable filter coefficient for generating a signal for driving a control actuator or a control sound source; and an adaptive processing unit for updating a filter coefficient of the digital filter according to a predetermined adaptive algorithm. The amount of calculation can be reduced without causing a time delay of the drive signal for the control sound source.

[0002]

2. Description of the Related Art Conventional apparatuses of this kind are described in British Patent No. 2149614 and Japanese Patent Application Laid-Open No. 1-501344. These conventional devices are noise reduction devices applied to the cabin of an aircraft or a similar closed space, and a single noise source such as an engine located outside the closed space includes a fundamental frequency f ₀ and its harmonics. It operates under the condition that noise including the waves f _{1 to} f _n is generated.

[0003] More specifically, includes a microphone for detecting a plurality of the installed sound pressure to a location within the closed space, and a plurality of loudspeakers for generating a control sound to the closed space, the frequency f ₀ of the noise source - Based on the f _n components, the frequencies f ₀ to
The loudspeaker is driven by a signal having a phase opposite to that of the f _n component,
Thus, a control sound having a phase opposite to that of the noise transmitted to the closed space is generated from the loudspeaker to cancel the noise.

[0004] As a method for generating a control sound emitted from a loudspeaker, PROCEEDINGS OF THE IEEE, VOL.
63 PAGE 1692, 1975, "ADAPTIVE NOISE CAN CE LLATION:
PRINCIPLES AND APPLICATIONS ”
DROW LMS 'algorithm is applied to multiple channels. The contents are described in a paper by the inventor of the above patent, “A MULTIPLE ERROR LMS ALGORITHM AND
ITS APPLICATION TOTHE ACTIVE CONTROL OF SOUND AND
VIBRATION ”, IEEE TRANS.ACOUST., SPEECH, SIGNAL PRO
CESSING, VOL. ASSP-35, PP. 1423-1434, 1987.

That is, the LMS algorithm is one of suitable algorithms for updating the filter coefficients of the adaptive digital filter, and is, for example, a so-called Filter.
In the ed-X LMS algorithm, a transfer function filter that models a transfer function from a loudspeaker to a microphone is set for all combinations of loudspeakers and microphones, and a reference signal indicating a noise generation state of a noise source is set. Updating the filter coefficients of digital filters with variable filter coefficients provided for each loudspeaker so that the noise in the closed space is reduced based on the value processed by the filter and the residual noise detected by each microphone. I have.

[0006]

Certainly, in the above-mentioned conventional apparatus, the filter coefficient of the digital filter for generating the signal for driving the loudspeaker converges to the optimum value as the control progresses. Although it is possible to ultimately reduce noise in the closed space, a filter that requires a normal LMS algorithm for performing a convolution operation or the like in the time domain, particularly a convolution operation between a reference signal and a transfer function filter is also required.
The amount of operation of the red-X LMS algorithm becomes enormous as the number of taps (the number of filter coefficients included in the digital filter) and the number of channels (the number of reference signals to be taken, the number of loudspeakers, and the number of microphones) increase. Would.

For this reason, the generated noise is a random noise (random noise) such as a road noise of a vehicle, so that a large number of taps are required for a digital filter, and the accuracy of control is increased. For this reason, when it is necessary to increase the number of loudspeakers or microphones, the processing capacity of the processor needs to be greatly improved (in some cases, to an unrealistic level). In particular, since an increase in the number of taps affects an increase in the amount of calculation in the order of the square, update processing of the filter coefficient of the digital filter cannot be performed in time, so that good adaptive control may not be performed.

Note that, unlike the LMS algorithm in the time domain, the LM performs all operations such as convolution in the frequency domain.
S algorithm (for example, ““ Adaptive Filtering in
theFrequency Domain ”, MAURO DENTIN
O, JOHN McCOOL and BERNARD W
IDROW Proc.IEEE.vol 66, pp1658 -1659,1978
”,“ Fast Implementation of LMS Adaptive Filte
rs ", EARL R. FERRARA IEEE Trans.Ac
oust, Speech, and Signal Processing vol ASSP −28, P
P.474-475, 1980], “Unconstrained Frequecy-D
omain Adaptive Filter ”, DAVID MANSOU
R and AUGUSTINE H. GRAY, JR. ,
IEEE Trans.Acoust, Speech, and Signal Processing vol
ASSP-30, PP.726-734, 1982]. )
Since the amount of calculation increases only in the order of “n · log n” with respect to the increase in the number of taps, it is an advantageous algorithm when the number of taps is large as in the active noise control device for reducing random noise described above. is there.

However, in the LMS algorithm in the frequency domain, it is necessary to convert a signal in the time domain into a signal in the frequency domain, that is, generally, it is necessary to use FFT (Fast Fourier Transform). Since it is handled in units of the number of blocks, at least a time delay of the block length occurs to generate a drive signal from the captured reference signal, and only control sounds that have a delay with respect to noise There is a problem that output cannot be performed and a sufficient control effect cannot be expected.

The present invention has been made in view of such unresolved problems of the prior art, and makes use of the advantages of the arithmetic processing in the time domain and the arithmetic processing in the frequency domain. It is an object of the present invention to provide an adaptive control device and an active noise control device that can significantly reduce the amount of arithmetic processing without causing a time delay of a drive signal.

[0011]

In order to achieve the above object, according to the present invention, a predetermined reference signal is filtered by a digital filter having a variable filter coefficient to generate a signal for driving a control actuator. An adaptive control apparatus comprising: a driving signal generating unit; and an adaptive processing unit for updating a filter coefficient of the digital filter so that a value of a predetermined evaluation function matches a target value. And the adaptive processing means determines the update amount of the filter coefficient of the digital filter by a calculation process in the frequency domain.

According to another aspect of the present invention, a control sound source capable of generating a control sound in a space to which noise is transmitted from a noise source and a noise generation state of the noise source are detected. Noise generating state detecting means for outputting a signal for driving the control sound source by filtering the reference signal with a digital filter having a variable filter coefficient; and a predetermined position in the space. Residual noise detecting means for detecting the residual noise and outputting the residual noise signal, and adaptive processing for updating a filter coefficient of the digital filter so as to reduce noise in the space based on the reference signal and the residual noise signal Means, the drive signal generating means executes time-domain filter processing, and the adaptive processing means By the processing of several areas to determine the amount of updated filter coefficient of the digital filter.

According to a third aspect of the present invention, in accordance with the second aspect of the present invention, a plurality of residual noise detecting means are provided, and the adaptive processing means is configured to update the plurality of residual noises every time the filter coefficient of the digital filter is updated. Are sequentially selected, and the filter coefficient of the digital filter is updated so that the residual noise detected by the selected one of the residual noise detecting means is reduced.

Further, according to a fourth aspect of the present invention, in the second or third aspect, the adaptive processing means includes a signal converting means for converting the reference signal and the residual noise signal into a frequency domain signal; Updating amount determining means for determining an updating amount of a filter coefficient of the digital filter based on a conversion result of the signal converting means so as to reduce noise in the space; and updating of a filter coefficient determined by the updating amount determining means. Update amount conversion means for converting the amount into a value in the time domain; and a filter coefficient of the digital filter is updated according to a conversion result of the update amount conversion means.

[0015]

According to the first aspect of the present invention, the drive signal generation means generates the drive signal of the control actuator by executing the filtering process in the time domain. It is supplied to the control actuator from the drive signal generation means without delay.

Since the arithmetic processing in the adaptive processing means is performed in the frequency domain, for example, in the case of a convolution operation, the integration of each frequency component can be performed, so that the update amount of the filter coefficient of the digital filter is determined in the time domain. The amount of calculation is greatly reduced as compared with the case where it is determined by the calculation processing. If the update amount of the filter coefficient is calculated in the frequency domain, it is necessary to convert the signal required for the calculation of the evaluation function into a signal in the frequency domain. Therefore, based on the information including the influence of the past signal, Therefore, there is a concern about the influence of a time delay of determining the update amount of the current filter coefficient.

However, when a digital filter composed of a large number of filter coefficients is operated adaptively, the adaptation speed tends to be unstable when trying to increase the adaptation speed. Therefore, the application target of such a filter does not change or the change is extremely gradual. It is often assumed that there is something. In other words,
The change of the filter coefficient of the digital filter does not need to be so abrupt, so that even if there is a slight time delay in the calculation of the update amount of the filter coefficient, convergence to the optimum value of the filter coefficient becomes impossible. There is no.

According to the second aspect of the present invention, the operation of the first aspect is also provided.
The operation of the invention described in the second aspect is essentially the same as that of the invention described in the second embodiment, except that the invention described in the second aspect is applied to the active noise control apparatus for reducing the noise in the space. There is a specific effect of the means provided for this. In other words, according to the second aspect of the present invention, the drive signal generation means generates a drive signal for the control sound source by filtering the reference signal output from the noise generation state detection means in the time domain. Is supplied from the drive signal generation means to the control sound source without time delay with respect to the reference signal.

For this reason, the control sound source generates a control sound correlated with the noise generated at the noise source. However, immediately after the start of the control, the filter coefficient does not always converge to an optimum value, so that the noise does not always occur. Is not necessarily reduced. However, since the adaptive processing means updates the filter coefficient of the digital filter based on the reference signal and the residual noise detected by the residual noise detection means so that the noise in the space is reduced, the filter coefficient converges to an optimum value. The noise is canceled by the control sound emitted from the control sound source, and the noise in the space is reduced.

Since the adaptive processing means determines the update amount of the filter coefficient of the digital filter by the arithmetic processing in the frequency domain, the adaptive processing means determines the update amount of the filter coefficient of the digital filter by the arithmetic processing in the time domain. This greatly reduces the amount of calculation. The invention according to claim 3 is based on a so-called error scanning method (detailed in "Active Noise Control Chair" in March 1990, "Transactions of the Acoustical Society of Japan").
The present invention is applied to the invention according to claim 2, wherein the update amount of the filter coefficient of the digital filter is determined by a frequency domain operation process, wherein one of the residual noises detected by the plurality of residual noise detecting means is determined. Focusing attention, updating the filter coefficient of the digital filter so that the residual noise is reduced, and performing the same processing focusing on other residual noise in the next processing, etc. As a result, the filter coefficients are updated.

That is, when a plurality of residual noise detecting means are provided to enhance the noise reduction effect, the filter coefficients of the digital filter are normally updated based on all the residual noises. Although the amount of calculation for updating the filter coefficient of the digital filter increases in accordance with the number of means, according to the invention of claim 3, one residual noise is used each time the filter coefficient is updated. Therefore, even if a plurality of residual noise detection means are provided,
The amount of calculation does not significantly increase as compared with the case where there is only one residual noise detecting unit.

As described above, the updating process is executed one after another while scanning each residual noise.
As in the case where the filter coefficients of the digital filter are updated so that all the residual noises are reduced, the filter coefficients converge so that the noise in the entire space is reduced. Further, the invention described in claim 4 is the above-described claim 2 or claim 3.
The adaptive processing means in the invention described in the above is more specific, wherein the signal conversion means converts the reference signal and the residual noise signal into a signal in the frequency domain. Therefore, for example, the convolution operation can be performed by integrating the respective frequency components, and the amount of operation is greatly reduced.

The update amount of the filter coefficient determined by the update amount determining means is converted into a value in the time domain by the update amount converting means, and the filter coefficient updating means converts the filter coefficient of the digital filter in accordance with the value. Since the update is performed, only the process of determining the update amount of the filter coefficient in the adaptive processing means is performed in the frequency domain.

[0024]

Embodiments of the present invention will be described below with reference to the drawings. FIG. 1 is a view showing the overall configuration of a first embodiment of the present invention. In this embodiment, a road and a road transmitted from a noise source between a road surface and wheels 2a to 2d into a vehicle interior 6 as a space are shown.
The present invention is applied to an active noise control device 1 for reducing noise.

First, the construction will be described. The vehicle body 3 is supported by suspensions interposed between the vehicle body 3 and the front wheels 2a, 2b, the rear wheels 2c, 2d and the wheels 2a to 2d. The vehicle shown in FIG. 1 is a so-called front-engine front-wheel drive vehicle in which front wheels 2a and 2b are rotationally driven by an engine 4 arranged in the front part of the vehicle body 3. The acceleration sensors 5a, 5b, 5c and 5d as noise generation state detecting means are attached to each of the suspensions interposed between the wheels 2a to 2d and the vehicle body 3, respectively.
Acceleration sensors 5e and 5f as noise generation state detecting means are attached to the member at the intermediate position between the a and 2b and the member at the intermediate position between the rear wheels 2c and 2d, and road noise input from the road surface is also provided. Are supplied to the controller 10 as reference signals x _k (k = 1 to K: K = 6 in this embodiment), which are acceleration signals corresponding to.

Further, in a vehicle room 6 as a space in the vehicle body 3,
Is a control actuator, that is, a control sound source.
Speakers 7a, 7b, 7c and 7d are located in the front seats.
S₁, S _TwoAnd rear seat S_Three, S_FourFacing each of
Door. Furthermore, each seat S₁~ S_Four
The headrest position of
Two microphones 8a to 8h are arranged respectively.
And these microphones 8a to 8h
The measured residual noise signal e₁~ E₈Is the controller 10
Supplied to

[0027] Then, the controller 10 may be composed of a microcomputer and the necessary interface circuits and the like, and the reference signal x _k supplied from the acceleration sensor 5 a to 5 f, the residual noise signal e supplied from the microphone 8a~8h based on the ₁ to e _8, and performs arithmetic processing to be described later, as control sound to cancel the load noise transmitted to the passenger compartment 6 is emitted from the loudspeaker 7a to 7d, they loudspeaker 7a and outputs a drive signal y ₁ ~y ₄ to ～7D.

As shown in FIG. 2, which is a block diagram of the functional configuration, the controller 10 has acceleration sensors 5a to 5a.
An adaptive digital filter W _km as a digital filter with a variable filter coefficient is used as a reference signal x _k supplied from f.
In (specifically, by performing a convolution operation) by filtering, and the drive signal generation unit 11 as a drive signal generating means for generating a drive signal y _m that drives the loudspeaker 7a~7d output, based on the reference signal x _k and the residual noise signal e _l includes an adaptive processing unit 12 as adaptive processing means for updating the filter coefficient W _KMI of the adaptive digital filter W _miles by performing the operation processing described later, the.

In this embodiment, the drive signal generator 1
The first filter processing is performed in the time domain, and the arithmetic processing of the adaptive processing unit 12 is performed in the frequency domain. That is,
In the adaptive processing unit 12, the controller 10 updates the filter coefficient W _kmi of the adaptive digital filter W _km by the update amount Δ
Only W _kmi determination processing is _{filtered in} the frequency domain
-XLMS.

Here, first, the time domain Filtered
-The XLMS algorithm will be described. That is, l
The remaining noise signal detected by the (m = 1, 2,..., L: L = 8 in this embodiment) microphone is e _l (n), and the _l- th noise signal when no control sound is generated from the loudspeaker. The residual noise signal detected by the microphone is represented by d _l (n),
A digital filter C _lm in which a transfer function between an m-th (m = 1, 2,..., M: M = 4 in this embodiment) loudspeaker and an l-th microphone is modeled in the form of a finite impulse response function. J (j = 0, 1, 2,..., J
-1: J is a constant), a filter coefficient of C _lmj , a reference signal of x _k (n), and an adaptive digital filter W _km for driving the m-th loudspeaker to which the reference signal x _k (n) is input.
If the i-th (i = 0, 1, 2,..., I-1: I is a constant) filter coefficient is W _kmi , Holds.

Each of the terms followed by (n) represents a sample value at the sampling time n, J represents the number of filter coefficients (the number of taps) included in the digital filter _Clm , and I represents the adaptive digital filter W. It is the number of filter coefficients (the number of taps) included in _km . In the above equation (1), “ΣΣW _kmix x _k ( _nji )” on the right side
_Represents the output y _m (n) when the reference signal x _k (n) is input to the adaptive digital filter, and “{C _lmj }
The term “W _km i x _k (n−j−i)｝” indicates that the signal y _m (n) input to the m-th loudspeaker is output to the space as a control sound and passes through the transfer function C _lm to the l-th signal. Represents the signal when it reaches the microphone,
_{The term “lmj} {W _km i x _k (n−j−i)}” is the sum of signals reaching the l-th microphone, and thus represents the sum of control sounds reaching the l-th microphone.

Next, the evaluation function Je is And

The LMS algorithm finds the filter coefficient W _kmi that minimizes the evaluation function Je. More specifically, the evaluation function Je is calculated using each filter coefficient W _kmi.
The filter coefficient W _kmi is updated with the value obtained by partially differentiating. Therefore, from the above equation (2), From the above equation (1), Therefore, if the right side of the equation (4) is set to r _klm (n−i), the update of the filter coefficient is as shown in the following equation (5).

[0034] Here, α is a coefficient called a convergence coefficient, and is related to the speed at which the filter converges optimally and its stability.

That is, the adaptive processing unit 12 updates the filter coefficient W _kmi of the adaptive digital filter W _km based on the above equation (5) in the time domain, but in the present embodiment, the adaptive processing unit 12 Is to determine the update amount ΔW _kmi by a frequency domain calculation process. FIG. 3 shows a specific functional configuration of the adaptive processing unit 12 for executing the content of the above equation (5) in the frequency domain. Become like

[0036] That is, the adaptive processing unit 12, a fast Fourier transform unit 13 of the reference signal x _k in the time domain as a signal converting means for generating a reference signal signal X _k of the Fourier transform to the frequency domain, the residual noise signal e a fast Fourier transform unit 14 as a signal conversion means for generating a residual noise signal E _l in the frequency domain by Fourier transformation of the _l, digital filter C _lm representing the digital filter C _lm of the time domain in the frequency domain and (f) , Reference signal X _k and digital filter C
_lm (f) is integrated for each frequency component and the processed signal R
an integrating section 15 of _klm (corresponding to the processed signal r _klm of the time domain) as an update amount determining means for calculating the convergence factor alpha, frequency by integrating the residual noise signal E _l and processing signals R _klm for each frequency component an integrating unit 16 of the update amount determining means for calculating the update amount ΔW _km (f) in the region, and calculates an update amount [Delta] W _KMI in the inverse Fourier transform to the time domain the update amount ΔW _km (f) in the frequency domain Inverse Fourier transform unit 17 as update amount conversion means, and addition as filter coefficient update means for adding new update amount ΔW _kmi to current filter coefficient W _kmi (n) to calculate new filter coefficient W _kmi (n + 1) A part 18.

FIG. 4 is a flowchart showing the outline of the processing executed in the controller 10.
The operation of this embodiment will be described with reference to FIG. First, Step 1
01, clear counter T, then step 1
02, the reference signal x _k (n) and the residual noise signal e _l (n) at the current time point n are read.

Then, the routine proceeds to step 103, where the counter T is incremented.
Is greater than or equal to the number of taps I of the adaptive digital filter W _km , and if not, the process proceeds to step 105. In step 105, the reference signal x _k (n) is filtered by the adaptive digital filter W _km , and more specifically, the adaptive digital filter W _k expressed in the time domain.
performing convolution operation between each filter coefficient W _KMI and the reference signal x _k of _km generates a drive signal y _m, and outputs the drive signal y _m in each loudspeaker 7a~7d in step 106. The processing of these steps 105 and 106
This corresponds to the processing in the drive signal generator 11 in the controller 10.

After the processing of step 106 is executed,
The process proceeds to step 102 and the above-described processing is repeatedly executed.
Then, each time step 103 is executed, the counter T is reset.
Therefore, the processing of step 102 is
The determination in step 104 is "YES"
That is, the processing after step 107 is executed. That is,
The processing after step 107 is performed in the controller 10
This corresponds to the processing in the processing unit 12, and
First, at step 107, the I-number of blocks
Time domain reference signal x stored as lock_k_
= [X_k(N-I + 1), x_k(N−I + 2),..., X
_k(N-2), x_k(N-1), x _k(N)]
D to convert the frequency domain reference signal X_k_ Is calculated.
Note that the attached symbol “_” indicates that it is a vector quantity.
I have.

Next, the routine proceeds to step 108, where the residual noise signal e _l _ = [e _l (n), e _l (n−1), e in the time domain also stored as a block of I units.
_l (n−2),..., _el (n−I + 2), _el (n−I
+1)] to calculate a residual noise signal E _l _ in the frequency domain. Then, the process proceeds to step 109, where the reference signal X _k _ and the digital filter C _lm _ (f) expressed in the frequency domain are integrated for each frequency component to calculate the processing signal R _klm _ Step 11
0, the processing signal R _{klm —} and the residual noise signal E _{l —}
Is further multiplied by a convergence coefficient α for each frequency component to calculate an update amount ΔW _{km —} (f) in the frequency domain.

Further, the processing shifts to step 111 to perform an inverse Fourier transform on the update amount ΔW _{km —} (f) in the frequency domain to obtain an update amount ΔW _km — = [ΔW _km0 , ΔW _km1 , ΔW in the time domain.
_km2 ,..., ΔWkm _(I-1) ], and _proceeds to step 112 to update the filter coefficient _Wkmi in accordance with the above equation (5). After executing the process of step 112, the process returns to step 101 and the above-described process is repeatedly executed.

[0042] Note that the processing in steps 107 to 110 the reference signal X _k _ digital filter C _lm _ (f) in the filtered processed signal R _klm _ the use so-called F
Although the filtered-X LMS algorithm is a filtered-X LMS algorithm in the frequency domain, the complex LMS algorithm (B. Wi
draw, J.M. McCool, and M.C. Ball,
“The complex LMS algorithm”, Proc, IEEE (Lett.
), Vol. 63, pp. 719-720, Apr. 1975). That is, the result of the Fourier transform of the data at the point I is a set of I complex numbers. The first half of the I / 2 point represents the characteristic in the negative frequency domain.
Since the characteristics are conjugate in the positive frequency domain, in the actual calculation, the negative frequency portion of the digital filter C _{lm —} (f), the negative frequency portion of the reference signal X _{k —,} and the residual noise signal E _{l —} With the positive frequency part (all I / 2
) _Is the data in the positive frequency domain of the update amount ΔW _{km —} (f), and the result of expanding the conjugate to the negative frequency domain is subjected to inverse Fourier transform to perform the update amount ΔW _{kmi in the} time domain. _.

When the processing shown in FIG. 4 is executed, the reference signal x
Each time _k is input (that is, at every predetermined sampling clock), the processing of steps 102 to 106 is executed, and as a result, control having a correlation with the reference signal x _k in the vehicle interior 6 from the loudspeakers 7a to 7d. A sound is generated,
Immediately after the start of the control, the respective filter coefficients W _kmi of the adaptive digital filter W _km are not always converged to an optimum value, so that it cannot be said that the road noise transmitted into the vehicle interior 6 is necessarily reduced.

However, when the processing of FIG. 4 is repeatedly executed, the processing of steps 107 to 112 is executed every time I sampling clocks are input, and each of the adaptive digital filters W _km is calculated according to the above equation (5). Filter coefficient W
_{Since kmi} is updated as appropriate, each filter coefficient W _kmi
Converges toward the optimum value, the road noise transmitted into the cabin 6 is canceled by the control sound emitted from the loudspeakers 7a to 7d, and the cabin 6
Noise in the interior is reduced.

In this embodiment, the filter coefficient W
_Since the calculation processing for determining the update amount ΔW _kmi of _kmi is performed in the frequency domain as described in the processing of steps 107 to 110, for example, in the time domain, a convolution calculation is performed for each frequency component. The amount of calculation can be greatly reduced, for example, the integration can be completed. The effect of reducing the calculation amount will be described with reference to a specific example.

That is, the number of taps I of the adaptive digital filter W _km is 128, and the number of taps of the digital filter _Clm is 32. In the configuration of the present embodiment, K = 6, M =
4, L = 8. Considering the case where the update operation of the above equation (5) is performed in the time domain under this condition, first, the convolution operation of the reference signal x _k and the digital filter C _lm for the operation of the processing signal r _klm includes L × M × K × J = 6144 integrations are required. Between the update amount [Delta] W _KMI residual noise signal e _l and the processed signal r _klm for operation, it is necessary K × M × L × I = 24576 times integration. Then, in order to compare with the calculation in the frequency domain in which the data of I units is treated as a block,
During the processing on the I reference signals x _k in the time domain, (6144 + 24576) × 128 = 3932160 integrations are required.

On the other hand, in the configuration of this embodiment, first,
The fast Fourier transform unit 13 performs a Fourier transform on the data of point I × K channel, and a fast Fourier transform unit 14
Performs a Fourier transform on the data of the I point × L channel, and the inverse Fourier transform unit 17 calculates the I point × K channel × M
In order to perform the inverse Fourier transform on the data of the channel, in order to execute the fast Fourier transform and the inverse Fourier transform, the result of the Fourier transform is a set of complex numbers (since a complex number has a real part and an imaginary part, 4 × (I / 2) · log ₂ I × (K + L + K × M) = 680 in consideration of the fact that integration between complex numbers requires four integrations.
96 integrations are required. Further, the processing in step 109 corresponding to the convolution operation of the reference signal x _k and the digital filter C _lm in the time domain requires that the result of the Fourier transform is a set of complex numbers, and that the numerical values of the positive and negative frequency parts are In consideration of the conjugate relationship (integration is required only for the positive frequency portion, that is, the integration is halved), (1/2) × I × K × L × M × 4 = 49152 times Multiplication is required. Then, the update amount ΔW in the frequency domain
The processing of step 110 for calculating _km (f) also takes into account that the result of the Fourier transform is a set of complex numbers and that the numerical values of the positive and negative frequency parts are in a conjugate relationship. / 2) × I × K × M × L × 4 = 49152 times of integration is required. Accordingly, the total number of integrations required to calculate the update amount ΔW _km (f) in the frequency domain calculation processing is 68096 + 49152 + 49152 = 166400 times.

From the above, according to the configuration of the present embodiment in which the update amount of the filter coefficient W _kmi is determined in the frequency domain, 166400/3932160 = 0.423177 ≒ 4 compared to the case where the update amount is performed in the time domain.
% Of the calculation amount. For this reason, for example, in order to improve the control effect, the number of taps I and J, the number of channels L,
Even if M, K, and the like are increased, it is possible to avoid an extreme increase in the amount of calculation required for determining the update amount ΔW _kmi , and to achieve sufficient practical use without significantly increasing the processing capability of the processor. There are advantages. In addition, according to the configuration of the present embodiment, step 105 corresponding to the drive signal generation unit 11 is performed.
Since the generation of the drive signal y _m is performed by the filtering process in the time domain as described in the processes of and 106,
Can be supplied to each loudspeaker 7a~7d without temporally delayed that the reference signal x _k a drive signal y _m.

In this embodiment, since the update amount of the filter coefficient W _kmi is calculated by the frequency domain calculation, one update process requires data of I units. It is conceivable that the convergence of the filter coefficient W _kmi is not guaranteed due to the influence of the time delay of the update operation processing. However, if the attempt to reduce the load noise is strong noise of non-periodic nature as in the present embodiment, the reference signal x _k underlying the control period information of load noise, the phase information and amplitude information Since it is included, that is, the reference signal x _k accurately represents the generation state of the road noise, a sufficient noise reduction effect can be obtained without abruptly changing the filter coefficient W _kmi of the adaptive digital filter W _km. be able to. In other words, the filter coefficient W
_Since the change in _kmi is gradual, even if there is some time delay in the calculation of the update amount ΔW _kmi of the filter coefficient W _kmi , it does not become impossible to converge the filter coefficient W _{km to} the optimum value. is there.

Further, in the present embodiment, the reference signal x at the same point I as the number of taps I of the adaptive digital filter W _km is used.
_k, residual noise signal e _l to Fourier transform, the result reduce very significantly the amount of computation as described above for seeking the appropriate integration to update amount ΔW _km (f) by integrating unit 15, 16 in FIG. However, since the fast Fourier transform (FFT) assumes a repetitive waveform, the integration of the FFT result results in a so-called cyclic convolution, which does not exactly match the convolution in the time domain (linear convolution). However, road noise has a certain amount of periodic components due to unevenness of the road surface, etc., so that the problem caused by the circular convolution is not so large. It is better to obtain reduction as an effect.

FIG. 5 is a block diagram showing a second embodiment of the present invention, and shows a functional configuration of the adaptive processing unit 12. As shown in FIG.
The same components as those in the first embodiment are denoted by the same reference numerals, and the description thereof will not be repeated. That is, even in the present embodiment, the point of determining the update amount [Delta] W _KMI filter coefficients W _KMI by calculation in the frequency domain is the same as the first embodiment, different from the result of the integration in the frequency domain The point is that it is devised to match the result of strict linear convolution.

Here, the fast Fourier transform is based on the assumption that given data is one cycle of a signal having a cycle of the data length and in which the same waveform appears repeatedly. Because it is a method of converting to a signal,
In particular, there is a drawback that the result of the front part considered in the time domain includes data with low reliability. Then, in order to solve such a drawback, an overlap save method (AV Openhein and RW Sc) is used.
hafer, Digital Signal Processing. Englewood Cl
iffs, NJ: Prentice-Hall, 1975) has been proposed in the past, and this embodiment is an application of the overlap save method to the present invention.

First, in the present embodiment, the reference signal x _k fetched by the fast Fourier transform unit 13 and the residual noise signal el _fetched by the fast Fourier transform unit 14 are 2 · I times that of the first embodiment. Is a block in units of. However, the determination in step 104 of FIG. 4 is simply made (T> 2 · I).
Thus, instead of continuously reading 2 · I data, a series of processes for reading the I data is counted by the counter P with respect to the reference signal x _k , and one process before the current P is performed. of connecting the I pieces of data read in the process of (P-1) to generate blocks with 2 · I number of units, with respect to the residual noise signal e _l, 2 · by connecting the I-number of 0 data Generate a block in I units.

Therefore, the reference signal x _kP _ and the residual noise signal e _lP _ are as follows. x _kP — = [x _k ((n−I) −I + 1), x _k ((n−
I) −I + 2),..., X _k ((n−I) −2), x
_k ((n-I) -1), x _k (n-I), x _k (n-I
+1), x _k (n−I + 2),..., X _k (n−2), x
_k (n−1), x _k (n)] e _lP — = [0, 0,..., 0, e _l (n), e _l (n−
1),..., E _l (n−I + 2), e _l (n−I + 1)] Therefore, the reference signal X _kP output from the fast Fourier transform unit 13 and the residual noise signal output from the fast Fourier transform unit 14 Each of the E _lP becomes 2 · I data strings.

The reference signal X _kP is integrated by the integration unit 15 with the digital filter C _lm (f) for each frequency component, and the resulting processed signal R _klm has the disadvantages of the fast Fourier transform described above. As a result, information with low reliability is included in the front part in the time domain. Therefore, in the present embodiment, the processed signal R _klm is temporarily returned to a signal in the time domain in the inverse Fourier transform unit 20 to generate a processed signal r _klmP composed of 2 · I data, and the processed signal r _klmP is _subjected to signal replacement. Unit 21 where the processing signal r
_{In addition} to removing temporally the first half of I data of _klmP, the signal replacement unit 21 in the previous (P-1) processing.
Of the signal r _{klm (P-1)} generated as an output of the above is connected to the front part of the processed signal r _klmP to form a new I / I data consisting of 2 · I data. Processing signal r
_{Generate klmP} .

[0056] Then, the new processing signal r _KlmP into a processed signal R _klm in the frequency domain in a fast Fourier transform unit 22, the frequency components and processing signals R _klm residual noise signal E _l by integrating unit 16 Convergence coefficient α
To calculate an update amount ΔW _km (f) in the frequency domain. Then, the update amount ΔW _km (f) is inverse Fourier transformed by the inverse Fourier transform 17 to calculate an update amount ΔW _kmi in the time domain. The update amount ΔW _kmi is _composed of 2 · I data, In addition, since the latter half of the data is also unreliable due to the above-described disadvantage of the fast Fourier transform, the first half of the data in which the last half of the I data is discarded is added as the update amount ΔW _kmi. The filter coefficient W _kmi is supplied to the section 18 to update the filter coefficient W _kmi .

With such a configuration, the fast Fourier transform is performed using 2 · I data, which is always twice the amount of I data required for processing, and the low reliability portion is used. Since the data is sequentially removed and the next arithmetic processing is executed, unlike the first embodiment, this is equivalent to performing strict linear convolution in the time domain, and high-precision control can be performed. .

In this embodiment as well, since the calculation processing for determining the update amount ΔW _kmi of the filter coefficient W _kmi is performed in the frequency domain, the calculation amount can be greatly reduced. That is, under the same conditions as in the first embodiment, to perform the fast Fourier transform and the inverse Fourier transform, 4 × (2I / 2) · log ₂ 2I × (K + 2 × K × L × M +
(L + K × M) = 1728512 times of integration is required. Further, in consideration of the same points as in the first embodiment, each of the integrating units 15 and 16 needs (1/2) × 2I × K × L × M × 4 = 98302 times of integration. Become. Therefore, in the present embodiment, the total number of integrations required for calculating the update amount ΔW _km (f) in the calculation processing in the frequency domain is 1728512 + 98302 × 2 = 19252516 times. From the above, according to the configuration of the present embodiment in which the update amount of the filter coefficient W _kmi is determined in the frequency domain, 1925116/3932160 = 0.4895823 {
Only 49% of computation is required.

However, since the amount of computation increases as compared with the first embodiment, it is not advisable to apply the configuration of this embodiment to reduce the noise exhibiting a certain periodicity. Therefore, it is preferable to apply the present invention to a case of reducing a noise showing almost no periodicity such that a sufficient noise reduction effect cannot be exhibited. Other functions and effects are the same as those of the first embodiment.

FIG. 6 is a block diagram showing a third embodiment of the present invention, and shows a functional configuration of the adaptive processing unit 12. As shown in FIG.
The same components as those in the first and second embodiments are denoted by the same reference numerals, and the description thereof will not be repeated. That is, even in the present embodiment, the update amount ΔW of the filter coefficient W _kmi
kmi is determined by the calculation in the frequency domain.
The second embodiment is the same as the second embodiment, and is different from the second embodiment in that a contrivance is made to obtain the same result as the linear convolution. The difference is that the error scanning method described above is applied. That is.

Here, the error scanning method is to select one of the microphones 8a to 8h as a plurality of noise generation state detecting means, and reduce the residual noise detected by the selected microphone. Fil
Update the filter coefficients W _kmi based on the Tered-X LMS algorithm, then select another microphone and filter coefficients W _kmi to reduce noise as well
Is updated for each of the microphones 8a to 8h in order.
Even _if the value of the optimum filter coefficient W _kmi for 8h is different, it is pursued in order, and as a result, the residual noise detected by each of the microphones 8a to 8h is considered at the same time to obtain Fi.
Since the filter coefficient W _kmi converges to the optimum value as in the case of executing the iterated-XLMS algorithm, the effect of the noise reduction control is not particularly reduced.

When the error scanning method is used, since only one microphone output is focused on one process, it is basically equivalent to L = 1. The expression (5) becomes the following expression (6), and the amount of calculation can be greatly reduced. W _kmi (n + 1) = W _kmi (n) + αe _l (n) r _klm ( _ni ) (6) Moreover, as in the present embodiment, the error scanning method is applied to the Filtered-XLMS algorithm in the frequency domain. When applied, the same method can be applied to the time domain Filter.
An effect of further reducing the amount of computation is obtained as compared with the case where the present invention is applied to the ed-XLMS algorithm.

The reason is that in the time domain, the adaptive digital filter W _km is naturally expressed in the time domain, and the update amount ΔW _kmi is also directly expressed in the time domain value. (L = 1, 2,..., L)
If the processing signal r _klm is not calculated for, the portion related to (ni) in the second term on the right side of the above equation (6) is missing and the update calculation becomes impossible. This is because such a problem does not occur in the configuration of the present embodiment that calculates ΔW _kmi .

In the second embodiment, in order to obtain the same result as the strict linear convolution, the signal replacement unit 21 uses the _immediately preceding processed signal _{rkm (P-1)} . in this example, the result of applying the error scanning method, time (P-1) and a microphone that has output the residual noise signal e _l read in, and a microphone that has output the residual noise signal e _l read at time P differ Therefore, the signal replacement unit 21 cannot perform the same processing as in the second embodiment based on the processing signal _rk.mP and the processing signal _{rkm (P-1)} . Therefore, the inverse Fourier transform unit 20
In this case, since the calculation for the next microphone at the present time is also required, the calculation amount in the inverse Fourier transform unit 20 is not simply (1 / L) as compared with the second embodiment, and
(1 / L) × 2.

That is, under the same conditions as in the first embodiment, 4 × (2I / 2) · log ₂ 2I × (K + 2 × K × M + 1 +) in order to perform the fast Fourier transform and the inverse Fourier transform.
2 × K × M) = 421888 times of integration is required. In addition, in the integrating units 15 and 16, it is necessary to perform (1/2) × 2I × K × M × 4 = 12288 times of integration in consideration of the same points as in the first embodiment. Therefore, in this embodiment, the total number of integrations required to calculate the update amount ΔW _km (f) in the frequency domain calculation processing is 421888 + 12288 × 2 = 446,564 times. From the above, according to the configuration of the present embodiment in which the update amount of the filter coefficient W _kmi is determined in the frequency domain, compared with the case of performing the update in the time domain, 446,64 / 3932160 = 0.37847222 ≒ 3.
Only 8% of computation is needed.

That is, with the configuration of the present embodiment, when a plurality of microphones are provided to improve the noise reduction effect, the amount of calculation increases by the number of microphones. There is an advantage that there is no. Other functions and effects are the same as those of the first and second embodiments. In each of the above embodiments, the case where the active noise control device according to the present invention is applied to the device for reducing the noise transmitted to the passenger compartment 6 has been described. It may be a device for aiming.

Further, in each of the above embodiments, the case of reducing the road noise transmitted into the vehicle interior 6 has been described. However, the noise that can be reduced by the present invention is not limited to this. For example, a pickup signal of a wind noise near a door mirror, a pickup signal for a case vibration of a differential gear or a transmission gear (a signal correlated with noise caused by a case vibration of a driving force transmission system), and an output shaft of a transmission for measuring a vehicle speed. If a configuration is adopted in which a pulse signal (a signal correlated with noise due to meshing of a differential gear or a transmission) according to rotation is detected, it is possible to provide a device for reducing each of those noises.

In the above embodiment, the four loudspeakers 7a to 7d as control sound sources and the eight microphones 8a to 8h are provided as residual noise detecting means, but the numbers are arbitrary. Further, in each of the above embodiments, the case where the adaptive control device according to the present invention is applied to an active noise control device for a vehicle has been described. However, the application target of the present invention is limited to a device for reducing noise. However, the present invention can be applied to other adaptive control systems.

[0069]

As described above, according to the present invention,
While the drive signal is generated by time-domain filter processing, the update amount of the filter coefficient of the digital filter is determined by frequency-domain arithmetic processing. There is an effect that a great reduction can be achieved.

In particular, according to the third aspect of the present invention, in addition to the above-described effects, there is an effect that a smaller amount of calculation is required as compared with a case where the same number of residual noise detecting means are simply provided.

[Brief description of the drawings]

FIG. 1 is a diagram showing an overall configuration of a first embodiment.

FIG. 2 is a block diagram illustrating a functional configuration of the entire controller.

FIG. 3 is a block diagram illustrating a functional configuration of an adaptive processing unit according to the first embodiment.

FIG. 4 is a flowchart showing an outline of processing in the first embodiment.

FIG. 5 is a block diagram illustrating a functional configuration of an adaptive processing unit according to a second embodiment.

FIG. 6 is a block diagram illustrating a functional configuration of an adaptive processing unit according to a third embodiment.

[Explanation of symbols]

REFERENCE SIGNS LIST 1 active noise control device (adaptive control device) 2 a to 2 d noise source 5 a to 5 f acceleration sensor (noise generation state detecting means) 6 cabin (space) 7 a to 7 d loudspeaker (control actuator, control sound source) 8 a to 8 h microphone (Residual noise detecting means) 10 Controller 11 Drive signal generating section (Drive signal generating section) 12 Adaptive processing section (Adaptive processing section) 13, 14 Fast Fourier transform section (Signal converting section) 15, 16 Accumulating section (Update amount determining section) 17) Inverse Fourier transform unit (update amount converting means) 18 Adder unit (filter coefficient updating means)

──────────────────────────────────────────────────続 き Continued on the front page (58) Field surveyed (Int.Cl. ⁶ , DB name) G10K 11/178 G05B 13/02

Claims (4)

(57) [Claims] And a drive signal generating means for generating a signal for driving a control actuator by filtering a predetermined reference signal with a digital filter having a variable filter coefficient so that a value of a predetermined evaluation function matches a target value. Adaptive processing means for updating a filter coefficient of the digital filter;
In the adaptive control device, the drive signal generation unit executes a filter process in a time domain, and the adaptive process unit determines an update amount of a filter coefficient of the digital filter by a calculation process in a frequency domain. Adaptive control device. 2. A control sound source capable of generating a control sound in a space to which noise is transmitted from a noise source; a noise generation state detecting means for detecting a noise generation state of the noise source and outputting it as a reference signal; Signal driving means for generating a signal for driving the control sound source by filtering the digital signal with a digital filter having variable filter coefficients, and a residual noise detecting means for detecting residual noise at a predetermined position in the space and outputting it as a residual noise signal An active noise control device comprising: an adaptive processing unit that updates a filter coefficient of the digital filter based on the reference signal and the residual noise signal so as to reduce noise in the space. The generating means executes a filtering process in a time domain, and the adaptive processing means performs a filtering process of the digital filter by an arithmetic processing in a frequency domain. Active noise control apparatus characterized by determining the update amount of filter coefficients. 3. The apparatus according to claim 1, further comprising a plurality of residual noise detecting means, wherein the adaptive processing means sequentially selects one of the plurality of residual noise detecting means for each process of updating a filter coefficient of the digital filter. 3. The active noise control device according to claim 2, wherein the filter coefficient of said digital filter is updated so that the residual noise detected by one selected residual noise detecting means is reduced. 4. An adaptive processing unit includes: a signal conversion unit configured to convert a reference signal and a residual noise signal into a signal in a frequency domain; and an adaptive processing unit configured to reduce noise in the space based on a conversion result of the signal conversion unit. Update amount determining means for determining the update amount of the filter coefficient of the digital filter; update amount converting means for converting the update amount of the filter coefficient determined by the update amount determining means into a time domain value; 3. A filter coefficient updating means for updating a filter coefficient of the digital filter according to a conversion result.
Or the active noise control device according to claim 3.