JPH07295578A - Method for determining convergence coefficient of adaptive controller, active noise controller, and active vibration controller - Google Patents

Abstract

PURPOSE:To easily set the convergence coefficient, used for probability gradient algorithm, to a proper value without depending upon a skilled person, etc. CONSTITUTION:The processes in the figure are performed at specific timing. Here, a rotational frequency control signal RC is outputted and the rotating speed of an engine is set to a rotational frequency at which the level of the booming sound is a problem (step 101); and gradient (Je/Wi) determining the update quantity of the update arithmetic of LMS algorithm is calculated (step 108) to find the steepest gradient (step 112), the optimum value ¦W(opt)¦<2> of power of an adaptive digital filter is found (step 107), and on the basis of the optimum value, an optimum point (W0(opt), W1(opt)) which is closest to an origin (0,0) is found (step 113). On the basis of the steepest gradient and optimum point, the convergence coefficient alpha is calculated (step 114).

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 even if the behavior of a controlled object is unknown or fluctuates, and a noise source or periodic noise transmitted from the vibration source. Alternatively, the present invention relates to an active noise control device and an active vibration control device for reducing noise or vibration by interfering control sound or control vibration with vibration, and particularly to a filter that generates a drive signal for driving a control sound source or a control vibration source. In an apparatus provided with an adaptive digital filter with variable coefficient and an adaptive processing means for updating the filter coefficient of this adaptive digital filter according to a stochastic gradient algorithm, the convergence coefficient used in the stochastic gradient algorithm is controlled. The value can be easily determined even by an unskilled person.

[0002]

2. Description of the Related Art As a conventional apparatus of this type, there are those described in British Patent No. 2149614 and Japanese Patent Publication No. 1-501344. These conventional devices are noise reduction devices applied to aircraft cabins and similar closed spaces, and a single noise source such as an engine located outside the closed space has 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 component, those frequencies f ₀ ~
The loudspeaker is driven by a signal having a phase opposite to that of the f _n component,
Therefore, 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.

Then, as a method of generating the control sound emitted from the loudspeaker, PROCEEDINGS OF THE IEEE, VOL.
63 PAGE 1692,1975, “ADAPTIVE NOISE CANCELLATION:
PRINCIPLES AND APPLICATIONS ”
An algorithm that applies the DROW LMS 'algorithm to multiple channels is applied. The content of the paper is "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 the algorithms suitable for updating the filter coefficient 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 the transfer function from the loudspeaker to the microphone is set for all combinations of the loudspeaker and the microphone, and a reference signal representing the noise generation state of the noise source is set. Update the filter coefficient of the adaptive digital filter with variable filter coefficient provided for each loudspeaker so that the value of the predetermined evaluation function based on the value processed by the filter and the residual noise detected by each microphone is reduced. is doing.

Here, the LMS algorithm (in particular, Fi
tered-X LMS algorithm) will be specifically described. Note that, here, an active noise control device for reducing noise transmitted to the control space will be described as an example. That is, the assumed active noise control device includes a plurality of M loudspeakers capable of generating a control sound in a control space to which periodic noise is transmitted from a noise source, and a residual noise sound remaining in the control space. The pressure is measured and the residual noise signal e _l (l = 1,
2, ..., L), a plurality of L microphones, a reference signal generation unit that detects a noise generation state of a noise source that periodically emits noise, and generates a reference signal x that represents the generation state, and a reference signal generation unit. The drive signal y _m (m = 1, 2 ,,) for filtering each signal x to drive each loudspeaker
, M) and outputs them to each loudspeaker. M adaptive digital filters W _m with variable filter coefficients and each filter coefficient of the adaptive digital filter W _m according to the LMS algorithm based on the reference signal x and the residual noise signal e _{l. W mi (i = 0,1, ...} , I-1: I is the number of taps of the adaptive digital filter W _m) and a, a filter coefficient updating unit for sequentially updating. The adaptive digital filter W _m and the filter coefficient updating unit are actually realized by a program in the microcomputer.

Then, no control sound is generated from the loudspeaker (that is, noise reduction control is not executed).
The residual noise signal detected by the l-th microphone (that is, the noise transmitted from the noise source) is ep
_l (n), a transfer function filter which is a digital filter obtained by modeling the transfer function between the mth loudspeaker and the lth microphone in the form of a finite impulse response filter, C ^ _lm , and the transfer function filter C ^. _lm j-th (j = 0,1,2, ..., J -1: J is the number of taps of the transfer function filter C ^ _lm) when the filter coefficients of the C ^ _LMJ,

[0008]

[Equation 1]

(1) is established. Note that the terms with (n) all represent sample values at the sampling time n. In the above (1), the right side _{"ΣW mi x (n-j-} i) " term of which is the output when the input of the reference signal x (n) to the adaptive digital filter W _m drive signals y _m (n ), "ΣC ^
_lmj term _{{ΣW mi x (n-j} -i)} "is through the m-th input to the loudspeaker drive signal y _m (n) is outputted to the space as a control sound from which the transfer function C ^ _lm represents a signal when it reaches the l-th microphone, furthermore, since the _{"ΣΣC ^ lmj {ΣW mi x (} n-j-i)} " is adding the signal arriving to the l-th microphone, It represents the total sum of control sounds that reach the l-th microphone.

Then, the evaluation function Je is

[0011]

[Equation 2]

It is defined as (2). Then, it is L to obtain the filter coefficient W _mi that minimizes the evaluation function Je (that is, the sum of squares of residual noise).
This is an MS algorithm, and specifically, a value (∂Je / ∂Je obtained by partially differentiating the evaluation function Je with respect to each filter coefficient W _mi.
W _mi ) to update the filter coefficient W _mi .

Therefore, from the above equation (2),

[0014]

[Equation 3]

(3) From the above equation (1),

[0016]

[Equation 4]

.. (4) Therefore, if the right side of the equation (4) is set to r _lm (n−i), the filter coefficient is updated by the following equation (5).

[0018]

[Equation 5]

(5) Here, the coefficient α is a coefficient called a convergence coefficient, and
It is related to the speed at which the filter converges optimally and its stability. That is, as can be seen from the above formula (5), this LMS
In the algorithm, the instantaneous residual noise signal e
_Since this is an algorithm for updating the filter coefficient W _mi using _l (n) as it is, the update direction (gradient) of each filter coefficient W _mi in each update operation represented by the above formula (3) is not always accurate. It does not always point to the optimum point. The optimum point is the value of the filter coefficient W _mi that can minimize the evaluation function Je represented by the above equation (2).

However, if the number of update operations is increased, the sum of the errors should approach zero infinitely. Therefore, by setting the convergence coefficient α within the range of 1 >>α> 0, the divergence of control, etc. Adaptive digital filter W without inviting
The filter coefficient W _{mi of m} can be converged to the optimum point. In this way, the LMS algorithm is an algorithm that finally finds the optimum point while gradually descending a curved surface (such a curved surface is referred to as an “error curved surface”) having a gradient represented by the above formula (3). It is widely used as a representative algorithm of the stochastic gradient algorithm suitable for the adaptive control device.

[0021]

In a stochastic gradient algorithm such as the LMS algorithm described above, in order to guarantee the convergence and stability of control, the convergence coefficient α is set to what size. Is an important issue. If the convergence coefficient α is large, the amount of update for each update operation is large, so that quick convergence is possible, but the possibility of divergence also increases. On the other hand, if the convergence coefficient is small, the update amount for each update operation is small, so the possibility of divergence is small, but it takes a long time to converge to the optimum value.

However, a specific method for determining the optimum convergence coefficient α has not been established, and it is the current situation that an expert determines it by trial and error based on experience and simulation. When determining the convergence coefficient α, it is necessary to consider the deterioration of the characteristics of devices such as loudspeakers and microphones over time and the manufacturing variations so as not to cause divergence, which requires a lot of man-hours. Will end up. In particular, the characteristics of loudspeakers, microphones, etc. are important factors for determining the convergence coefficient α, but the convergence coefficient α must be determined while checking the variations in the characteristics as needed, so the yield is poor. It was a major cause of high manufacturing cost, and even if a loudspeaker or microphone with small variations in characteristics was used, the equipment itself would be expensive, which would also lead to higher manufacturing costs, so an effective solution. Don't

The present invention has been made by paying attention to the unsolved problem of the conventional technique, and an adaptive control device for reducing periodic noise and vibration by using a stochastic gradient algorithm. It is an object of the present invention to provide a method capable of accurately and easily determining the convergence coefficient used in the stochastic gradient algorithm of. It is also an object of the present invention to provide an active noise control device and an active vibration control device having a function capable of accurately and easily determining the convergence coefficient used in the stochastic gradient algorithm.

[0024]

In order to achieve the above object, the invention according to claim 1 generates a control sound or control vibration that interferes with periodic noise or vibration transmitted from a noise source or a vibration source. A controllable sound source or a controllable vibration source, a reference signal generating means for generating a reference signal representing a generation state of the periodic noise or vibration in the noise source or the vibration source, and the control by filtering the reference signal. An adaptive digital filter having a variable filter coefficient for generating a drive signal for driving a sound source or a controlled vibration source, and residual noise detection means or residual for detecting residual noise or residual vibration after the interference and outputting it as a residual noise signal or residual vibration signal Vibration detection means and a stochastic gradient algorithm to reduce the noise or vibration after the interference based on the reference signal and the residual noise signal or residual vibration signal. A method of determining a convergence coefficient used in the stochastic gradient algorithm of an adaptive control device, comprising: adaptive processing means for updating the filter coefficient of the adaptive digital filter. Noise or vibration is generated at a constant frequency and the control sound or control vibration is not generated from the control sound source or the control vibration source, and the temporary reference signal having a frequency slightly deviated from the constant frequency is applied to the adaptive reference signal. The steepest gradient among the gradients that determines the update amount of the filter coefficient is obtained by sequentially performing the processing of supplying the processing means to update the filter coefficient of the adaptive digital filter, and the steepest gradient and the residual noise. The convergence coefficient is determined based on the signal or the residual vibration signal.

According to a second aspect of the present invention, in the convergence coefficient determination method of the adaptive control device according to the first aspect of the present invention, the filter coefficient that is sequentially executed according to the temporary reference signal is updated continuously. The average value of the gradient in a plurality of processes is calculated, and the steepest gradient is determined from the average value. The invention according to claim 3 is the convergence coefficient determination method for an adaptive control device according to claim 1 or 2, wherein the filter of the adaptive digital filter is based on the residual noise signal or the residual vibration signal. The optimum value of the coefficient is calculated, the optimum value is divided by the steepest gradient, and the result of the division is multiplied by the safety factor to determine the convergence coefficient.

Further, the invention according to claim 4 is the convergence coefficient determining method for an adaptive control device according to any one of claims 1 to 3, wherein the constant frequency is the periodic noise or It is the frequency of vibration. On the other hand, in order to achieve the above object, the invention according to claim 5 is a control sound source capable of generating a control sound that interferes with a periodic noise transmitted from a noise source, and the periodic noise in the noise source. A reference signal generating means for generating a reference signal indicating the generation state of the signal, an adaptive digital filter having a variable filter coefficient for filtering the reference signal to generate a drive signal for driving the control sound source, and residual noise after the interference. Residual noise detecting means for detecting the noise and outputting it as a residual noise signal, and adaptively updating the filter coefficient of the adaptive digital filter according to a stochastic gradient algorithm so as to reduce the noise after the interference based on the reference signal and the residual noise signal. In the active noise control device including a processing unit, the noise source periodically repeats at a predetermined timing for a predetermined time. Noise source control means for generating a sound at a constant frequency, control sound source stopping means for keeping the control sound from the control sound source at the predetermined timing and for a predetermined time continuously, and a slight deviation from the constant frequency A provisional reference signal generating means for generating a provisional reference signal of a frequency and supplying it to the adaptive processing means, and a filter coefficient of the adaptive digital filter which is sequentially executed by the adaptive processing means when the provisional reference signal is supplied. A steepest gradient calculating means for obtaining the steepest gradient among the gradients that determines the update amount in the updating process, and a convergence coefficient for determining a convergence coefficient used in the stochastic gradient algorithm based on the steepest gradient and the residual noise signal. And a determining means.

According to a sixth aspect of the present invention, in the active noise control device according to the fifth aspect of the present invention, the steepest slope computing means sequentially executes filter coefficients according to the temporary reference signal. A gradient average value calculating means for calculating an average value of the gradient in a plurality of continuous processes for updating, and a steepest gradient determining means for making the steepest value among the average values the steepest gradient, It was decided to prepare.

The invention according to claim 7 is the active noise control device according to claim 5 or 6, wherein the convergence coefficient determining means determines the adaptive digital signal based on the residual noise signal. An optimum value calculating means for calculating an optimum value of a filter coefficient of the filter, dividing the optimum value of the filter coefficient by the steepest gradient, and determining a convergence coefficient by multiplying the division result by a safety factor. did.

Further, the invention according to claim 8 is the active noise control device according to any one of claims 5 to 7, wherein the constant frequency is the frequency of the periodic noise in question. It is a thing. On the other hand, in order to achieve the above object, the invention according to claim 9 is a control vibration source capable of generating a control vibration that interferes with a periodic vibration transmitted from a vibration source, and the periodic vibration in the vibration source. Reference signal generating means for generating a reference signal indicating a vibration generation state, an adaptive digital filter having a variable filter coefficient for filtering the reference signal to generate a drive signal for driving the controlled vibration source, and Residual vibration detecting means for detecting residual vibration and outputting it as a residual vibration signal, and updating the filter coefficient of the adaptive digital filter according to a stochastic gradient algorithm based on the reference signal and the residual vibration signal so as to reduce the vibration after the interference. In the active vibration control device including the adaptive processing means for controlling the vibration from the vibration source continuously at a predetermined timing for a predetermined time. Source control means for generating a specific vibration at a constant frequency, control vibration source stopping means for keeping the control vibration from the control vibration source at the predetermined timing and for a predetermined time continuously, and from the constant frequency Temporary reference signal generating means for generating a temporary reference signal having a slightly deviated frequency and supplying it to the adaptive processing means, and the adaptive digital filter sequentially executed by the adaptive processing means when the temporary reference signal is supplied. Of steepest gradients among the gradients that determine the update amount in the updating process of the filter coefficient, and the convergence coefficient used in the stochastic gradient algorithm based on the steepest gradient and the residual vibration signal. Convergence coefficient determining means for determining.

According to a tenth aspect of the invention, in the active vibration control device according to the ninth aspect of the invention, the steepest slope computing means sequentially executes filter coefficients in accordance with the temporary reference signal. A gradient average value calculating means for calculating an average value of the gradient in a plurality of continuous processes for updating, and a steepest gradient determining means for making the steepest value among the average values the steepest gradient, It was decided to prepare.

The invention according to claim 11 is the active vibration control device according to claim 9 or 10, wherein the convergence coefficient determining means is based on the residual vibration signal. An optimum value calculating means for calculating an optimum value of a filter coefficient of the filter, dividing the optimum value of the filter coefficient by the steepest gradient, and determining a convergence coefficient by multiplying the division result by a safety factor. did.

Furthermore, the invention according to claim 12 is the active vibration control device according to any one of claims 9 to 11, wherein the constant frequency is the frequency of the periodic vibration in question. .

[0033]

According to the present invention, the noise source or the vibration source generates periodic noise or vibration of a constant frequency, and the control sound source or the control vibration source does not generate the control sound or the control vibration. Then, the residual noise detecting means or the residual vibration detecting means will detect only the periodic noise or vibration emitted from the noise source or the vibration source. That is, in the above example, only the component of the first term on the right side of the above equation (1) is included in the residual noise signal, and the component of the second term on the right side of the equation is not included in the residual noise signal. .

In such a situation, when the temporary reference signal is supplied to the adaptive processing means to update the filter coefficient of the adaptive digital filter, the frequency of the temporary reference signal slightly deviates from the frequency of noise or vibration. (For example, if the frequency of noise or vibration is 60 Hz, the frequency of the temporary reference signal should be 59 Hz), so it is equivalent to gradually shifting the phase of the temporary reference signal with respect to noise or vibration. become.

Then, unless the phase characteristic of the adaptive digital filter is gradually changed so as to compensate for the phase shift, noise or vibration cannot be canceled by the control sound or control vibration, so the phase of the adaptive digital filter is canceled. The characteristic changes 360 degrees with time 1 / Δf corresponding to the frequency deviation Δf between the frequency of the temporary reference signal and the frequency of noise or vibration as one cycle.

Moreover, in the stochastic gradient algorithm such as the LMS algorithm, the phase characteristic of the adaptive digital filter changes 360 degrees because it converges to the optimum value so as to go down the error curved surface as described above. In the process of, all gradients on the contour line passing through the origin of the error curved surface (points where the filter coefficients are all zero) are sequentially calculated. For example, in the case of the above-mentioned example, {Σ of the second term on the right side of the above equation (3) (or the above equation (5) is used.
The gradient (∂Je / ∂W _mi ) represented by e ₁ (n) r _lm (n−i)} multiplied by 2 will be sequentially calculated.

The gradient determines the updating direction of the filter coefficient of the adaptive digital filter. The steeper the gradient, the shorter the distance to the optimum point, and the gentler the gradient, the shorter the distance to the optimum point. Should be long. And
The convergence coefficient α should be set based on the steepest gradient (hereinafter referred to as the steepest gradient). This is because if a large convergence coefficient α is set with reference to a gentle gradient,
This is because, if the update is performed at a steep slope, the control is likely to diverge. On the other hand, when the convergence coefficient α is set on the basis of a steep gradient, it takes a long time to converge to the optimum value when updating is performed at a gentle gradient. This is because such a problem can be solved by improving the capacity of the arithmetic processing device, and preventing the divergence of control is a top priority issue.

On the other hand, in such a situation that the residual noise detecting means or the residual vibration detecting means measures only the periodic noise or vibration generated from the noise source or the vibration source, the residual noise detecting means or the residual vibration detecting means is detected. The residual noise signal or residual vibration signal output by the means represents the magnitude of the control sound or control vibration required to completely cancel the noise or vibration. The optimum value of the power required for the adaptive digital filter to cancel the vibration is known.

When the optimum value of the power required for the adaptive digital filter is known, the locus of the optimum point of the adaptive digital filter is also known. This is because the formula expressing the power of the adaptive digital filter can be easily derived. The shape of the contour line of the error curved surface is, for example, an ellipse when the number of taps of the adaptive digital filter is two (when the number of taps is greater than two, the shape of the concept corresponding to the ellipse in each dimension). It has been found that the inclination of the ellipse is also uniquely determined according to the ratio (f / f _S ) of the noise or vibration frequency f to the sampling frequency f _S. By the way, considering the filter coefficients of the adaptive digital filter as W ₀ and W _1, and considering the W ₀ -W ₁ plane formed with these filter coefficients W ₀ and W ₁ as Cartesian coordinates, the horizontal axis represents the filter coefficient W ₀ (the right side is Positive), vertical axis is filter coefficient W
_{When 1} (upper side is positive), (f / f _S )> (1 /
4), an ellipse that is long in the first and third quadrant directions and is inclined by 45 degrees with respect to both axes becomes (f / f _S ) <(1
/ 4) is an ellipse that is long in the directions of the second and fourth quadrants and is inclined by 45 degrees with respect to both axes.

Further, if the contour line has an elliptical shape (when the number of taps is greater than 2, the shape of the concept corresponding to the ellipse in each dimension), the minor axis has the steepest slope and its shortest length. The optimum point in the radial direction is the optimum value of the power required for the adaptive digital filter, an expression expressing the power of the adaptive digital filter, and W ₀ = W in the short radial direction.
₁ (when (f / f _S ) <(1/4)) or W ₀ = −W
₁ ((f / f _S )> (1/4)), and the optimum point value represents the distance from the origin to the optimum point for each filter coefficient.

When the distance to the optimum point and the steepest slope are known and the convergence coefficient α is determined based on them, the control is most likely to diverge (when updating is performed at the steepest slope). ), An accurate convergence coefficient α that does not cause divergence is determined. Next, in the invention according to claim 2, when the filter coefficient updating process according to the temporary reference signal is executed a plurality of times, the average value of the gradients that determine the updating amount obtained in those processes is obtained. . And
From the average value, the steepest gradient that is the basis for determining the convergence coefficient α is obtained.

Here, since the stochastic gradient algorithm such as the LMS algorithm is an algorithm for determining the updating direction (gradient) based on the value of the residual noise signal or residual vibration signal as an instantaneous value as described above, An error is included in the gradient obtained in the update calculation of.
Therefore, when the invention according to claim 1 is executed based on the average value of the gradients in a plurality of consecutive processes, as in the invention according to claim 2, the error included in the gradient as an instantaneous value Since the influence is small, a more accurate convergence coefficient α is determined.

On the other hand, the optimum value W _{(opt) of} each filter coefficient representing the distance from the origin to the optimum point and the steepest gradient (∂Je /
If ∂W _mi ) _MAX is known, the convergence coefficient α is α = W _(opt) / (∂Je / ∂ in order to start updating the filter coefficient from the origin and converge to the optimum value in the shortest time. W _mi ) _MAX ... (6) may be set. This is because, with such a convergence coefficient α, the second term on the right side of the above equation (5) matches the optimum value W _(opt) .

However, the steepest gradient (∂Je / ∂W _mi ) _MAX
However, since the error is included as described above, if the convergence coefficient α is determined as in the above equation (6), the stability of the update process is likely to be impaired. Therefore, in the invention according to claim 3, since the convergence factor α is determined by further multiplying the result of the equation (6) by the safety factor, the convergence factor α that enables stable update processing is determined. For example, in an active vibration control device that reduces the engine noise transmitted from the vehicle engine to the vehicle interior, if the filter coefficient converges to the optimum value in about 0.1 seconds immediately after the control is started, It is known that changes in the rotation speed can be followed, and if the sampling frequency is 1 kHz, it is sufficient to converge to an optimum value in 100 update calculations, so the safety factor may be about 1/100. . Even in other adaptive control devices, the safety factor may be appropriately determined from the same viewpoint.

Further, in the invention according to claim 4,
The frequency of the noise or vibration generated from the noise source or the vibration source when determining the convergence coefficient α is the frequency of the noise or vibration in question (for example, the frequency causing the deterioration of the noise level or the vibration level, the occupant if the vehicle. The frequency or the like of noise or vibration that is likely to give an unpleasant feeling is set, so that the convergence coefficient α that is particularly effective is determined in the situation where the noise or vibration of the problem frequency is generated.

The active noise control device according to the invention of claim 5 is, to put it simply, an active noise control device integrally provided with a function for implementing the method of determining the convergence coefficient according to the invention of claim 1. Type noise control device. That is, the noise source control means causes the noise source to generate noise of a constant frequency at a predetermined timing and continuously for a predetermined time, and at the same time, the control sound source stopping means causes the control sound source to generate no control sound. The predetermined timing is
It is a timing arbitrarily selected according to the application target of the active noise control device, for example, if it is an active noise control device applied to a vehicle, immediately after turning on the ignition switch or at the time of regular inspection. It may be. Further, the predetermined time is a time necessary and sufficient for obtaining the steepest gradient by the steepest gradient calculating means described later.

In such a state, the temporary reference signal generating means generates a temporary reference signal having a frequency slightly deviated from the frequency of the noise generated from the noise source, and the temporary reference signal is applied to the adaptive processing means. The filter coefficient is supplied to update the filter coefficient of the adaptive digital filter according to the above equation (5), and the steepest slope among the slopes (∂Je / ∂W _mi ) that determines the update amount at that time is calculated by the steepest slope calculating means. Desired.

The convergence coefficient determining means determines the convergence coefficient used in the stochastic gradient algorithm based on the steepest gradient and the residual noise signal supplied from the residual noise detecting means. As in the invention, an accurate convergence coefficient α will be determined. That is, according to the invention of claim 5, the convergence coefficient α used in the stochastic gradient algorithm is variable and is appropriately set to an appropriate value. Even if the initial convergence coefficient α does not allow stable control due to deterioration of each device after installation in the actual control system, a new convergence coefficient α is set for stable control. To be done.

Further, in the invention according to claim 6, the steepest gradient computing means comprises a gradient average value computing means and a steepest gradient determining means, and the gradient average value computing means
The average value of the gradients in a plurality of continuous processings of the update calculation that is sequentially executed according to the temporary reference signal is calculated, and the steepest gradient determining means calculates the convergence coefficient α from the average value.
The steepest slope that is the basis for determining is determined. Therefore, similarly to the invention according to claim 2, the influence of the error included in the gradient as the instantaneous value is reduced, and the more accurate convergence coefficient α is determined.

According to the invention of claim 7,
When the optimum value calculating means included in the convergence coefficient determining means calculates the optimum value of the filter coefficient based on the residual noise signal,
The optimum value is divided by the steepest slope (see the above equation (6)), and the division result is multiplied by the safety factor to determine the convergence coefficient α. Therefore, similarly to the invention according to the third aspect, the convergence coefficient α that can perform the stable update process is determined.

Further, in the invention according to claim 8,
Similarly to the invention according to claim 4, since the frequency of the noise generated from the noise source when determining the convergence coefficient α is set as the frequency of the problem noise, noise of the problem frequency is generated. A particularly effective convergence coefficient α is determined under such circumstances. Here, while the inventions according to claims 5 to 8 are all directed to noise, claim 9
The invention according to claim 12 is directed to vibration. Therefore, the operation of the invention according to claims 9 to 12 is substantially the same as that of the above-mentioned claim 5 although there is a difference from the vibration of sound.
It is the same as the invention according to claim 8.

[0052]

Embodiments of the present invention will be described below with reference to the drawings. FIG. 1 is a diagram showing an overall configuration of a first embodiment of the present invention. In this embodiment, a "muffled sound" as a periodic noise transmitted from an engine 4 as a noise source into a vehicle interior 6 is shown. The present invention is applied to the active noise control device 1 for reducing the noise.

First, the structure will be described. The vehicle body 3 is supported by front wheels 2a and 2b, rear wheels 2c and 2d, and suspensions interposed between the wheels 2a to 2d and the vehicle body 3. The vehicle shown in FIG. 1 is a so-called front-mounted engine front-wheel drive vehicle in which the front wheels 2a and 2b are rotationally driven by an engine 4 arranged in the front part of the vehicle body 3. A crank angle sensor 5 as a reference signal generating means is attached to the engine 4, and the crank angle sensor 5 is
A reference signal x, which is a sinusoidal signal having the same cycle as the muffled sound generated in the engine 4 and transmitted to the vehicle interior 6, is supplied to the controller 10. It should be noted that the muffled sound is generated, for example, in synchronization with 1/2 rotation of the crankshaft in the case of a reciprocating 4-cylinder engine, and is generated in synchronization with 2/3 rotation of the crankshaft in the case of a reciprocating 6-cylinder engine. Based on the engine type, it is possible to know which order of the crank rotation occurs in synchronization with the engine type. Therefore, a sine wave signal synchronized with the rotation of the crankshaft is appropriately generated according to the engine type, and this is used as a reference signal x. And it is sufficient.

In the passenger compartment 6 of the vehicle body 3, loudspeakers 7a, 7b, 7c and 7d as control sound sources face front seats S ₁ and S ₂ and rear seats S ₃ and S ₄ , respectively. It is located at the door. Furthermore, each seat S ₁
The head restraint position of the to S ₄ is a microphone 8a~8h as residual noise detecting means, have been respectively disposed two by two, the residual noise signal e ₁ to e these microphones 8a~8h was measured as a sound pressure ₈ Are supplied to the controller 10.

Then, the controller 10 executes the arithmetic processing described later based on the reference signal x supplied from the crank angle sensor 5 and the residual noise signals e _{1 to} e ₈ supplied from the microphones 8a to 8h. , So that drive signals y _{1 to} y ₄ are output to the loudspeakers 7a to 7d so that control sounds for canceling the muffled sound transmitted to the vehicle interior 6 are emitted from the loudspeakers 7a to 7d. There is.

The controller 10 is actually constituted by including a microcomputer and necessary interface circuits, but functionally, as shown in FIG. 2 which is a block diagram showing the functional constitution, the reference signal x Of the adaptive digital filter W _m with a variable filter coefficient for generating a drive signal y _m and a transfer function filter C ^ for filtering a reference signal x to generate a processed signal (Filtered-X signal) r _{lm. lm} and a filter coefficient updating unit 11 that updates each filter coefficient W _mi of the adaptive digital filter W _m based on the processed signal r _lm and the residual noise signal e _l .

The adaptive digital filter W _m is a finite impulse response type digital filter composed of I filter coefficients W _mi, and is provided for each of the loudspeakers 7a to 7d. Here, the active noise control device 1 is a device for reducing the muffled noise that is a periodic noise, and the tap number I of the adaptive digital filter W _m may be “2”. Therefore, the adaptive digital filter of this embodiment is used. W _m is W _m (z) = W _m0 + W _m1 z ^-1 (7).

The transfer function filter C ^ _lm is a finite impulse response type digital filter that models the transfer function between the loudspeakers 7a to 7d and the microphones 8a to 8h. Then, M =
It is provided for all combinations (L × M) of 4) loudspeakers 7a to 7d and L (L = 8 in this embodiment) microphones 8a to 8h.

Then, the filter coefficient updating unit 11 updates each filter coefficient W _mi of the adaptive digital filter W _m based on the processed signal r _lm and the residual noise signal e _l according to the LMS algorithm which is one of the probability gradient algorithms. It is supposed to do. The update formula of the filter coefficient W _mi is
It is expressed by the above equation (5) as in the conventional case. Further, as shown in FIG. 1, the controller 10 receives an ignition signal IG from an ignition switch 21, and the controller 10 supplies a rotation speed control signal RC to an engine rotation speed controller 22. It has become so.

That is, the ignition switch 21 outputs the ignition signal IG at the timing when the ignition is turned on. Further, the engine speed controller 22 is a controller capable of controlling the engine speed to a desired value by appropriately adjusting the injection amount of the fuel supplied to the engine 4 irrespective of the operation of the driver. When the speed control signal RC is input, control is performed to match the engine speed with the speed represented by the speed control signal RC.

Further, as shown in FIG. 2, the controller 10 includes a convergence coefficient calculation control unit 12 to which an ignition signal IG is input, a temporary reference signal generation unit 13 for generating a temporary reference signal x ', and a transfer function filter. A changeover switch 14 that can switch the input signal to C ^ _lm between the reference signal x and the temporary reference signal x ′, an intermittent switch 15 that can intermittently output the output of the adaptive digital filter W _m , and an update by the filter coefficient update unit 11. Slope (∂Je / ∂) that determines the amount of calculation update
_{W mi = 2Σe l (n)} r lm (n-i): (3) -
(See equation (5)) and a convergence coefficient calculator 16 that determines the convergence coefficient α based on the residual noise signal e _l .

The convergence coefficient calculation control unit 12 operates from the time when the ignition signal IG is input (predetermined timing) to a predetermined time (for example, a fixed time of 1 second, or until the steepest gradient described below is detected). Until the temporary reference signal generation unit 13 outputs the temporary reference signal x ′, the changeover switch 14 is connected to the temporary reference signal generation unit 13 side, and the intermittent switch 15 is turned off. As a result, the drive signal y _m is not output, and the convergence coefficient calculation unit 16 calculates the convergence coefficient α. On the other hand, after a predetermined time has elapsed, the temporary reference signal generation unit 13 and the convergence coefficient calculation unit are executed. Control of stopping the processing of 16 and connecting the changeover switch 14 to the side to which the reference signal x is input to bring the intermittent switch 15 into a conductive state is executed. .

Further, when the ignition signal IG is input, the convergence coefficient calculation control unit 12 receives the rotation speed control signal RC.
Is output. However, the rotation speed represented by the rotation speed control signal RC output here is an engine rotation speed (for example, 3000 rp) in which the level of muffled noise is a problem.
m). Then, the temporary reference signal generation unit 13
A temporary reference signal x'having a frequency slightly deviated from the frequency of the reference signal x generated by the crank angle sensor 5 (that is, the frequency of the muffled sound actually transmitted to the vehicle interior 6) is generated. There is. For example, the frequency of the reference signal x is 6
If it is 0 Hz, the frequency of the temporary reference signal x'may be 59 Hz. However, the frequency shift between the muffled sound and the temporary reference signal x ′ is not limited to 1 Hz, and the frequency of the temporary reference signal x ′ may be shifted in the direction higher than the frequency of the reference signal x. The temporary reference signal x '
May be generated based on the reference signal x, but the engine speed is known from the engine speed control signal RC, and the frequency of muffled noise is uniquely determined from the engine speed. It may be generated based on.

Furthermore, the convergence coefficient calculation unit 16 calculates the gradient (∂Je /
A process of obtaining the steepest (∂Je / ∂W _{mi) MAX} from the ∂W _mi), 2 mean square <e _l ^2> virtual noise transmission system from the residual noise signal e _l (engine 4 and the microphone 8
a~8h between noise transmittance system) G power of | W _(opt) | ² and processing for obtaining its power | optimum value W ² and the filter coefficient based on the later-described formula represents power | W _{(opt)
The process of obtaining m0 (opt)} and W _{m1 (opt)} and its optimum value W
_{Set m0 (opt)} (or W _{m1 (opt)} ) to the steepest gradient (∂Je / ∂W)
_mi ) A process of multiplying the result of division by _MAX by the safety factor (1/100) to calculate the convergence coefficient α is executed.

Then, the filter coefficient updating unit 11 is adapted to execute the update calculation using the latest convergence coefficient α calculated by the convergence coefficient calculation unit 16. next,
The operation of this embodiment will be described. 3 and 4 are flowcharts showing the outline of the processing executed in the controller 10. FIG. 3 shows a drive signal y _m generated and supplied to each of the loudspeakers 7a to 7h. FIG. 4 shows an outline of a noise reduction process for reducing noise, and FIG. 4 shows an outline of a process of determining the convergence coefficient α executed at a predetermined timing.

That is, the vehicle ignition switch 21
When is turned on, at the same time when the engine 4 is started, the ignition signal IG is supplied to the controller 10, the convergence coefficient calculation control unit 12 is activated, and the processing shown in FIG. 4 for determining the convergence coefficient α is executed. However, in the processing shown in FIG. 4, a system of one channel (that is, L = 1, M = 1) is considered in order to make the contents easy to understand.

When the processing shown in FIG. 4 is started, first, at step 101, a rotation speed control signal RC is output to the engine rotation speed controller 22. Then, the rotational speed of the engine 4 becomes equal to the rotational speed at which the muffled sound becomes a problem, so that a muffled sound at a problematic frequency is generated in the vehicle interior 6. Although not shown in the flow chart in particular, the intermittent switch 15 shown in FIG.
Is cut off and the drive signal y _m is not output.

Next, in step 102, the transfer characteristic (gain) of the transfer function C between the loudspeaker and the microphone is calculated. The transfer characteristic is represented by | C ^ (f) | ² from the transfer function filter C ^, but the specific calculation may be performed by Fourier transform,
The characteristics may be tabulated in advance. It should be noted that the muffled sound frequency f is easily obtained because there is a fixed relationship between the engine speed and the muffled sound frequency. For example, in the case of a reciprocating four-cylinder engine, the muffled sound is synchronized with 1/2 rotation of the crankshaft, so the value of the muffled sound is the value obtained by dividing the engine speed by 30. That is, if the engine speed is 3000 rpm, 3000/30 = 100 Hz is the muffled sound frequency. However, the engine speed may be calculated from the rotation speed control signal RC output in step 101 or may be calculated from the reference signal x output from the crank angle sensor 5, but more accuracy is pursued. For example, the reference signal x
It is better to be based on.

Then, the process proceeds to step 103, and the number of repetitions N is set. The number of times of repetition N corresponds to the number of data in the calculation for obtaining the average value of the gradient which will be described later. Since the ratio (f _S / f) between the sampling frequency f _S and the muffled sound frequency f is not necessarily an integer, the coefficient q is repeated so that N = q (f _S / f) becomes an integer. It is desirable to set the number of times N. The reason for this is, as will be described later, that the frequency difference is intentionally given between the temporary reference signal x ′ and the muffled sound so that the phase difference between the two gradually changes. This is because it is not preferable that the calculation for obtaining the average value of the gradients is switched in the phase difference).

When the number of repetitions N is set, the process proceeds to step 104, the residual noise signal e supplied from the microphone is read, and then the process proceeds to step 105,
A temporary reference signal x ′ having a frequency slightly deviated from the muffled sound frequency (frequency of the reference signal x) is generated. Therefore, as shown in FIG. 5, there is a relationship between the reference signal x and the provisional reference signal x ′ in which the phase difference gradually increases as time elapses.

Then, the process proceeds to step 106 and the power of the virtual muffled sound transmission system G is calculated according to the following equation (8):
Calculate G | ² . | G | ² = <e ² > (8) However, <e ² > is the root mean square of the residual noise signal e, and the number of data of the average calculation is N in this case.

The reason why the power | G | ² is obtained based on the equation (8) is that no control sound is generated in the passenger compartment 6 when the processing shown in FIG. 4 is being executed. From
The residual noise signal e contains only noise components mainly including muffled sound components. The transmission system of the muffled sound and the control sound between the engine 4 and the residual noise signal e is shown in FIG.
Although the control sound is not generated, the control sound transmission system represented by WC can be ignored. Therefore, since the level of the residual noise signal e is determined by the muffled sound transmission system G, the root mean square of the residual noise signal e measured as the sound pressure <
e ² > can be an estimated value of the power | G | ² of the transmission system G.

Next, the routine proceeds to step 107, where the power | W | ² required for the adaptive digital filter W for canceling the muffled sound, that is, the optimum power value | W _(opt) | ² according to the following equation (9). Is calculated. | W _(opt) | ² = | G | ² / | C ^ (f) | ² (9) Here, in the LMS algorithm, the filter coefficients W ₀ and W ₁ of the adaptive digital filter W and the evaluation are performed. Function J
If e is the axis of the three-dimensional orthogonal coordinate system, the error curved surface ES becomes as shown in FIG. 7, and the bottom height of the error curved surface ES is
It is the value of the evaluation function Je when the level of muffled sound is the lowest, and if the bottom point is projected on the W ₀ -W ₁ plane,
Optimal points (W _{0 (opt)} , W of filter coefficients W ₀ , W _{1
1 (opt)} ) is obtained.

Then, the length from the origin (0, 0) of the filter coefficients W ₀ and W ₁ to the optimum point (W _{0 (opt)} , W _{1 (opt)} ) is the optimum power value | W _(opt) | ² (see FIG. 8), and in the case of the system shown in FIG. 6, the relationship | G | ² = | W | ² | C | ² (10) is necessary to cancel the muffled sound. It is known that the transfer function C is modeled by the transfer function filter C ^.
Therefore, according to the above equation (9), the optimum power value | W
_(opt) | ² is required.

[0075] Then the process proceeds to step 108, the filter coefficients W _0, W ₁ origin (0 according (11) below,
The gradient (∂Je / ∂W _i ) of the error curved surface Je in 0) is calculated. ∂Je / ∂W _i = 2 <e · r (n−i)> (11) It is clear from the above formulas (1) to (5) that the gradient is obtained by the formula (11). However, <er
(N−i)> means the average value of the products of the residual noise signal e and the processed signal r, and the number of data for the average value calculation is N.

The processed signal r is the result of filtering the temporary reference signal x'with the transfer function filter C ^.
It is expressed as in equation (12) below.

[0077]

[Equation 6]

(12) Next, the process proceeds to step 109 and the above step 104
It is determined whether the processes of to 108 have been executed N times, and if it is determined that the number of repetitions has not reached N times, the process returns to step 104 and the above-described processes are executed again. On the other hand, when the determination in step 109 is “YES”, the process proceeds to step 110, and it is determined whether or not the gradient (∂Je / ∂W _i ) calculated this time is the steepest, that is, the optimum point. It is determined whether or not the distance to (W _{0 (opt)} , W _{1 (opt)} ) is the shortest.

Here, when the tap number I of the adaptive digital filter W is 2, as shown in FIG. 7, the shape of the contour line ES 'obtained by projecting the error curved surface ES on the W ₀ -W ₁ plane is an ellipse. I know it will take shape. Moreover, it is known that the inclination of the elliptical shape with respect to the W ₀ axis and the W ₁ axis is uniquely determined according to the ratio (f / f _S ) of the muffled sound frequency f to the sampling frequency f _S.

Specifically, in W ₀ -W ₁ , the horizontal axis represents the filter coefficient W ₀ (right side is positive) and the vertical axis represents the filter coefficient W _1.
If (f / f _S )> (1/4) if (upper side is positive), it is long in the first quadrant and the third quadrant direction and 4 for both axes.
An ellipse inclined by 5 degrees, and if (f / f _S ) <(1/4), for example, as shown in FIG. 9, an ellipse long in the second and fourth quadrant directions and inclined by 45 degrees with respect to both axes. It becomes.

Then, at the center of the contour line ES 'which is an ellipse, the optimum points (W _{0 (opt)} , W 1 ₎ of the filter coefficients W ₀ , W ₁ are obtained.
_{1 (opt)} ) exists, the optimum point is that there is a slight deviation between the frequency of the reference signal x (the muffled sound) and the frequency of the temporary reference signal x ′ as shown in FIG. Due to the frequency shift, the reference signal x increases as the cycles overlap.
And the phase difference between the temporary reference signals x'will gradually increase. However, compensation must be made somewhere to compensate for the phase difference, and as a result, the phase characteristics of the filter coefficients W ₀ and W ₁ of the adaptive digital filter W change so as to compensate for the change in the phase difference. Become.

Therefore, the optimum points (W _{0 (opt)} and W _{1 (opt)} ) of the filter coefficients W ₀ and W ₁ are the period 1 around the origin (0, 0) as shown by the broken line in FIG. / Δf (Δf is the difference between the frequency f and the sampling frequency f _S ) makes one round, and the locus of the optimum point is the same ellipse as the contour line ES ′. The reason is that since the adaptive digital filter W is expressed by the above expression (7), its characteristic at the frequency f is expressed by the following expression (13), where j is the imaginary unit.

W (f) = (W ₀ + W ₁ cos (2πf / f _S )) − j (W ₁ sin (2πf / f _S )) (13) Therefore, its power | W (f) | ² is, | to become ^{_{2 = W 0 2 + W 1}} 2 + 2W 0 W 1 cos (2πf / f S) ...... (14) | W (f). Since the optimum power value | W _(opt) | ² is a constant value when the frequency f is the same as expressed by the above expression (9), the left side of the above expression (14) is set as the optimum value. Then, the locus drawn by the point (W ₀ , W ₁ ) becomes an ellipse,
Since the shape and inclination of the contour line ES 'do not change, the trajectory of the optimum point and the contour line ES' eventually have the same elliptical shape.

If the locus of the optimum point is elliptical, the optimum point closest to the origin should exist in the minor axis direction,
Since it is known that the inclination of the ellipse is 45 degrees as described above, the optimum point in the minor axis direction is (f / f _S )
If _{the> (1/4), W 0 (} opt) = illustrates the relationship _{-W 1 (opt) ...... (15} ), also if _{(f / f S) <(} 1/4) W 0 _(opt) = W _{1 (opt)} ... (16).

Then, the gradient in the minor axis direction (∂J
e / ∂W _i ), in the case of the relation of the above formula (15), (∂Je / ∂W ₀ ) / (∂Je / ∂W ₁ ) = 1 (17), Further, in the case of the relationship as in the above formula (16), (∂Je / ∂W ₀ ) / (∂Je / ∂W ₁ ) = − 1 (18).

Therefore, if the minor axis direction of the ellipse is known based on the ratio (f / f _S ) of the muffled sound frequency f and the sampling frequency f _S , in step 110, the above (1
By judging whether or not the expression (7) or the expression (18) is satisfied, it is possible to know whether or not the gradient (∂Je / ∂W _i ) calculated this time is the steepest. However, step 1
Even if the average value is obtained in 08, the gradient (∂Je / ∂W _i )
Since some error remains in the equation, whether the judgment of the above formula (17) or (18) is substantially satisfied (that is,
It is necessary to judge by whether it is approximately 1 or -1).

If the determination in step 110 is "NO", the process proceeds to step 111 and the counter for counting the number of repetitions N is cleared, and then the above step 1
Returning to 04, the above-mentioned processing is executed again. However, since the optimum value of power | W _(opt) | ² obtained by the processing of steps 106 and 107 is a constant value if the frequency f is the same, the processing of steps 106 and 107 need not be executed. .

However, the determination in step 110 is "YE
In the case of "S", the process proceeds to step 112, and the steepest slope (∂Je / ∂W _i ) _MAX is set to the slope (∂J) calculated this time.
e / ∂W _i ). That is, (∂Je / ∂W _i ) _MAX = (∂Je / ∂W _i ) ... (19)

Then, the process proceeds to step 113, and the optimum points (W _{0 (opt)} , W _{1 (opt)} ) are calculated. In particular,
(14) left the optimum value of the power of expression | W _(opt) | with ² and back, (14) the right-hand side of equation _{(f / f S)> (} 1
In case of / 4), the relationship of the above equation (15) is substituted, and (f / f
_{If S} ) <(1/4), the relationship of the above equation (16) is substituted, and if W _{0 (opt)} or W _{1 (opt)} is solved, those optimum points (W _{0 (opt)} , W _{1 (opt)} ) can be obtained.

Then, the routine proceeds to step 114, where the convergence coefficient α is calculated based on the steepest gradient (∂Je / ∂W _i ) _MAX and the optimum point W _{0 (opt)} (or W _{1 (opt)} ). . Here, since the value W _{0 (opt)} or W _{1 (opt)} of the optimum point represents the distance from the origin to the optimum point for each filter coefficient, it is said that the fastest convergence is possible immediately after the start of control. If the convergence coefficient α is determined from the viewpoint, the convergence coefficient α may be set to a value of α = W _{0 (opt)} / (∂Je / ∂W _mi ) _MAX according to the equation (6). However, this tends to impair the stability of the update process as described above.

Therefore, in this embodiment, the convergence coefficient α is calculated according to the following equation (20). α = | (W _{0 (opt)} / (∂Je / ∂W _mi ) _MAX ) | · SF (20) However, the safety factor SF is a filter after control is started in the process shown in FIG. 3 described later. It is necessary to appropriately determine based on the time until the coefficients W ₀ and W ₁ converge to the optimum values and the sampling frequency f _S. For example, in the case of the active noise control device 1 for muffled sound as in the present embodiment, the filter coefficients W ₀ and W ₁ converge to the optimum values in about 0.1 seconds immediately after the control is started. For example, it is known that the control that follows changes in the engine rotation speed is executed and does not give an occupant an uncomfortable feeling. Therefore, if the sampling frequency f _{S is set} to 1 kHz, it is sufficient to converge to an optimum value in 100 update calculations. Therefore, it is sufficient to set SF = 1/100.

When the convergence coefficient α is obtained in step 114, the convergence coefficient α is stored in a predetermined storage area in the controller 10, and then the processing shown in FIG. 4 this time is ended. When the process shown in FIG. 4 is completed, the process shown in FIG. 3 is subsequently executed. That is, when the process of FIG. 3 is started, first, in step 001, the reference signal x supplied from the crank angle sensor 5 and the residual noise signals e _{1 to} e ₈ supplied from the microphones 8a to 8h are read, and then the step In step 002, the reference signal x and each filter coefficient C ^ _mij of the digital filter C ^ _lm are convoluted to calculate the processed signal r _lm .

Then, the process proceeds to step 003, the convergence coefficient α stored in a predetermined storage area is read, then the process proceeds to step 004, and the filter coefficient W _mi of the adaptive digital filter W _m is calculated according to the above equation (5). To update. When the processing of steps 001 to 004 is completed, the process proceeds to step 005, and the reference signal x and each filter coefficient W _mi of the adaptive digital filter W _m are convoluted to drive the drive signal y.
_{1 to} y ₄ are calculated, and these drive signals y _{1 to} y ₄ are output to the loudspeakers 7a to 7d in step 006.

[0094] Then, the control sound into the passenger compartment 6 from loudspeakers 7a~7d occurs immediately after start of control is not necessarily the filter coefficient W _mi of the adaptive digital filter W _m is converged to an optimum value Therefore, it cannot be said that the muffled sound transmitted to the vehicle interior 6 is necessarily reduced. However, as a result of repeatedly executing the processing of steps 001 to 006, each filter coefficient W _mi converges toward the optimum value, and the muffled sound transmitted into the vehicle interior 6 is transmitted to the loudspeaker 7.
It is canceled by the control sounds emitted from a to 7d, noise in the vehicle interior 6 is reduced, and a comfortable vehicle interior space is realized.

In the present embodiment, the convergence coefficient α used for the update calculation of the filter coefficient W _mi of the adaptive digital filter W _m is not fixed to a predetermined value, but every time the ignition is turned on. The processing shown in FIG. 4 is executed (that is, every time the engine 4 is started), the optimum convergence coefficient α is calculated at that time, and it is used until the convergence coefficient α is calculated next time. As a result, good update calculation can always be performed. This means that in the design stage of the active noise control device 1, the loudspeakers 7a ...
This means that it is not necessary to determine the convergence coefficient α in consideration of deterioration of the 7d and the microphones 8a to 8h over time, etc., so that there is an advantage that the yield is improved as compared with the conventional case.

In particular, when the convergence coefficient α is determined, the phase characteristic of the adaptive digital filter W is set to 360 by using the temporary reference signal x'which is intentionally given a frequency shift from the muffled sound.
The gradient (∂
Je / ∂W _i ) finds the steepest gradient and sets the convergence coefficient α based on the steepest gradient. Therefore, in the update calculation of the filter coefficient of the actual noise reduction control performed thereafter, Regardless of the value of the phase difference between the reference signal x and the muffled sound, stable noise reduction control can be performed without causing divergence of control.
That is, if the convergence coefficient α is determined according to the process shown in FIG. 4, it is possible to determine the optimum convergence coefficient α regardless of who does it.

Further, in this embodiment, steps 104 to 1
08 is repeated N times while the average value of the gradient is 2 <e.
r (n−i)> is calculated and the gradient is calculated (∂Je / ∂W _mi ).
And the steepest gradient (∂Je / ∂
Since W _mi ) _MAX is obtained, even with an LMS algorithm that is premised on that the instantaneous value includes an error, the effect of the error is reduced and the convergence coefficient α can be accurately determined. . That is, since the instantaneous gradient is obtained based on the residual noise signal e _l at each time, it has a waveform including an error as shown in FIG.
If the average value is obtained in a temporally continuous range, the influence of the error becomes extremely small as shown in FIG. 11, and an accurate gradient can be obtained.

In this embodiment, the convergence coefficient α is calculated according to the equation (20) including the safety factor SF, and the safety factor SF is also theoretically obtained, so that the stability is very excellent. There is an advantage that the update calculation can be performed. Further, in the present embodiment, the convergence coefficient α is determined in a situation in which a muffled sound of a frequency at which the noise level is a problem is generated. Therefore, the convergence coefficient particularly effective when actual noise reduction control is performed. It is possible to obtain α.

In the present embodiment, the process of determining the convergence coefficient α is performed every time the ignition switch 21 is turned on, and noise reduction control is not performed during that period, which is considered to give an occupant an unpleasant feeling. Maybe, but there is no such defect. This is because the process of determining the convergence coefficient α is surely completed within the time when the phase characteristic of the adaptive digital filter W changes by 360 degrees, and its cycle is 1 / Δf. For example, if Δf is 1 Hz, the processing shown in FIG. 4 is completed in 1 second. Moreover, the determination in step 110 of FIG. 4 is actually “YES”.
Since the processing is substantially completed at the time of, and the two short-side directions of the ellipse exist at positions separated by 180 degrees, the time required until the determination in step 110 becomes “YES” is , Δf is 1 Hz, the maximum is 0.5 seconds, which is not a problem.

Here, in this embodiment, particularly in correspondence with the invention according to claim 5, the adaptive processing means is constituted by the transfer function filter C ^ _lm , the filter coefficient updating unit 11 and the processing of steps 002 to 004. , Convergence coefficient calculation control unit 12, engine speed controller 22 and step 1
The noise source control means is constituted by the processing of 01, the control sound source stopping means is constituted in that the drive signal y _m is not output in the intermittent switch 15 and FIG. 4, and the temporary reference signal generation section 13 and the processing of step 105 The reference signal generating means is configured, and the steepest slope computing means is configured by the processing in steps 108 and 110, and steps 104 and 10 are performed.
Convergence coefficient determining means is constituted by 6, 107, 113 and 114.

Further, in the present embodiment, corresponding to the invention according to claim 6, the processing of steps 108 and 109 constitutes the average value calculating means, and the processing of step 110 constitutes the steepest slope determining means. Corresponding to the invention according to claim 7, the optimum value calculating means is constituted by the processing of step 113. In the above embodiment, the case where the active noise control device according to the present invention is applied to the vehicle active noise control device for reducing the muffled noise transmitted to the vehicle interior of the vehicle has been described. The application target of is not limited to this, and may be a device that reduces noise other than muffled noise as long as it is periodic noise.

The target of the reduction is not limited to noise. For example, an engine mount (control vibration source) that generates an active control force is interposed between the engine 4 and the member and the member is mounted on the member side. An acceleration sensor (residual vibration detection means) for detecting residual vibration is provided, and
If such an engine mount is controlled based on the same reference signal x and output signal (residual vibration signal) of the acceleration sensor as in the above embodiment and according to the same LMS algorithm,
The active vibration control device can actively reduce the vibration transmitted from the engine 4 to the vehicle body side. Also in the case of the active vibration control device, if the processing for determining the convergence coefficient α is executed as in the above-described embodiment, the same operational effect as in the above-described embodiment can be obtained. However, even when the active vibration control device is used, the application target is not limited to the vehicle, and may be, for example, a support device for a machine tool.

In the above embodiment, the convergence coefficient α is
The case of incorporating the processing for determining the above into the active noise control device 1 has been described, but the present invention has a significant effect by itself, and therefore may be executed by itself. That is, if the process shown in FIG. 4 is executed to determine the convergence coefficient α at the design stage of the active noise control device and the active vibration control device, the yield is improved and the manufacturing cost is significantly reduced. Is possible.

Further, in the above embodiment, the processing for determining the convergence coefficient α is executed at the timing when the ignition switch is turned on, but the execution timing is arbitrary. For example, if the process shown in FIG. 4 is executed by pressing a predetermined switch, the user may press the predetermined switch at any timing to determine the convergence coefficient α, or The determination of the convergence coefficient α may be one item at the time of maintenance inspection, and the convergence coefficient α may be determined by pressing a predetermined switch at every periodic inspection.

In the above embodiment, the case where the Filtered-XLMS algorithm is applied as the stochastic gradient algorithm has been described, but the present invention is not limited to this. For example, a synchronous Filtered-X LM is used.
S-algorithm (Proceedings of the Acoustical Society of Japan, March 1992)
See pages 515-516 of the month for details. ) And other stochastic gradient algorithms.

In the above embodiment, the four loudspeakers 7a to 7d as the control sound source and the eight microphones 8a to 8h as the residual noise detecting means are provided, but the number of these is arbitrary. Further, in the above embodiment, the case where the tap number I of the adaptive digital filter W is set to 2 has been described, but the tap number I may be larger than 2. In that case, since each filter coefficient W _mi and the evaluation function Je form a coordinate system of four dimensions or more, it is not possible to particularly illustrate and explain, but instead of the above ellipse, a corresponding ellipse in each dimension is used. The concept corresponding to is the shape of the contour line ES 'of the error curved surface ES.
Therefore, if the idea of the above-described embodiment is expanded, the optimum convergence coefficient α can be similarly determined.

[0107]

As described above, according to the invention of claim 1, claim 5 or claim 9, the convergence coefficient α used in the stochastic gradient algorithm is set to a theoretical and appropriate value. As a result, the convergence coefficient α can be determined without the need for a skilled person to think and error, so that the yield is improved and the manufacturing cost is reduced.

Particularly, in the invention according to claim 5 or claim 9, since the convergence coefficient α is set to the optimum value at that time at a predetermined timing, the adaptive digital filter of the adaptive digital filter is always in accordance with the lowest convergence coefficient α. There is an effect that the update calculation of the filter coefficient is performed and favorable noise reduction control and vibration reduction control are executed. In other words, at the designing stage, it is not necessary to consider the deterioration of the device over time, etc., so that it can be said that the yield can be significantly improved.

Further, claim 2, claim 6 or claim 10
According to the invention of (1), the influence of the error due to the stochastic gradient algorithm can be made extremely small, so that there is an effect that a more appropriate convergence coefficient α can be determined. Further, according to the invention of claim 3, claim 7, or claim 11, since the convergence coefficient α having excellent stability can be determined, there is an effect that more stable noise reduction control and vibration reduction control are executed. is there.

Further, claim 4, claim 8 or claim 1
According to the second aspect of the present invention, the convergence coefficient α is determined in the situation where noise or vibration of a frequency that is a particular problem is generated, so that it is particularly effective when performing actual noise reduction control and vibration reduction control. It is possible to obtain a different convergence coefficient α.

[Brief description of drawings]

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

FIG. 2 is a block diagram showing a functional configuration of a controller.

FIG. 3 is a flowchart showing an outline of processing executed in a controller.

FIG. 4 is a flowchart showing an outline of other processing executed in the controller.

FIG. 5 is a waveform diagram illustrating a relationship between a reference signal x and a temporary reference signal x ′.

FIG. 6 is a block diagram showing a transmission system of muffled sound and control sound.

FIG. 7 is an explanatory diagram of an error curved surface.

FIG. 8 is an explanatory diagram of an optimum power value of an adaptive digital filter.

FIG. 9 is an explanatory diagram of optimum points of filter coefficients.

FIG. 10 is a waveform diagram showing a temporal change in a gradient as an instantaneous value.

FIG. 11 is a waveform diagram showing a temporal change in a gradient as an average value.

[Explanation of symbols]

DESCRIPTION OF SYMBOLS 1 Active noise control device 4 Engine (noise source) 5 Crank angle sensor (reference signal generation means) 6 Cabin 7a-7d Loudspeaker (control sound source) 8a-8h Microphone (residual noise detection means) 10 Controller 11 Filter coefficient update Part 12 Convergence coefficient calculation control part 13 Temporary reference signal generation part 14 Changeover switch 15 Intermittent switch 16 Convergence coefficient calculation part W _m Adaptive digital filter x Reference signal y _m Drive signal e _l Residual noise signal α Convergence coefficient

Continuation of the front page (51) Int.Cl. ⁶ Identification number Office reference number FI technical display location H03H 21/00 8842-5J // G05B 13/02 S 7531-3H

Claims (12)

[Claims] 1. A control sound source or a control vibration source capable of generating a control sound or a control vibration that interferes with a periodic noise or vibration transmitted from a noise source or a vibration source, and the periodic pattern in the noise source or the vibration source. Signal generating means for generating a reference signal representing the generation state of various noises or vibrations, and an adaptive digital filter with a variable filter coefficient for filtering the reference signal to generate a drive signal for driving the control sound source or the controlled vibration source. And a residual noise detecting means or residual vibration detecting means for detecting the residual noise or residual vibration after the interference and outputting the residual noise signal or residual vibration signal as a residual noise signal or residual vibration signal, and based on the reference signal and the residual noise signal or residual vibration signal. Adaptive processing for updating the filter coefficient of the adaptive digital filter according to a stochastic gradient algorithm so as to reduce noise or vibration after interference A method of determining a convergence coefficient used in the stochastic gradient algorithm of an adaptive control device, comprising: generating the periodic noise or vibration at a constant frequency from the noise source or the vibration source; The control sound or the controlled vibration is not generated from the sound source or the controlled vibration source, and a temporary reference signal having a frequency slightly deviated from the constant frequency is supplied to the adaptive processing means to set the filter coefficient of the adaptive digital filter. The updating process is sequentially executed, and the steepest gradient among the gradients that determines the update amount of the filter coefficient is obtained, and the convergence coefficient is calculated based on the steepest gradient and the residual noise signal or the residual vibration signal. A method for determining a convergence coefficient of an adaptive control device, characterized by: 2. An average value of the gradients in a plurality of consecutive processes for updating a filter coefficient that is sequentially executed according to the temporary reference signal is calculated, and the steepest gradient is obtained from the average values. A method for determining a convergence coefficient of the adaptive control device according to claim 1. 3. The optimum value of the filter coefficient of the adaptive digital filter is calculated based on the residual noise signal or the residual vibration signal, the optimum value is divided by the steepest slope, and a safety factor is added to the division result. The convergence coefficient determination method for an adaptive control device according to claim 1, wherein the convergence coefficient is determined by multiplication. 4. The constant frequency is the frequency of the periodic noise or vibration in question.
A method for determining a convergence coefficient of the adaptive control device according to any one of claims 1 to 5. 5. A control sound source capable of generating a control sound that interferes with a periodic noise transmitted from a noise source, and a reference signal generation for generating a reference signal representing a generation state of the periodic noise in the noise source. Means, an adaptive digital filter having a variable filter coefficient for generating a drive signal for driving the control sound source by filtering the reference signal, and a residual noise detecting means for detecting the residual noise after the interference and outputting it as a residual noise signal. And an adaptive processing means for updating the filter coefficient of the adaptive digital filter according to a stochastic gradient algorithm so that the noise after interference is reduced based on the reference signal and the residual noise signal. A noise source control means for generating the periodic noise at a constant frequency from the noise source at a predetermined timing and continuously for a predetermined time. A control sound source stopping means for keeping the control sound from being generated from the control sound source at the predetermined timing and continuously for a predetermined time, and a temporary reference signal having a frequency slightly deviated from the constant frequency to generate the adaptive processing. The temporary reference signal generating means for supplying the means and the gradient for determining the update amount in the update processing of the filter coefficient of the adaptive digital filter which is sequentially executed by the adaptive processing means when the temporary reference signal is supplied. A steepest gradient calculating means for obtaining a steepest gradient, and a convergence coefficient determining means for determining a convergence coefficient used in the stochastic gradient algorithm based on the steepest gradient and the residual noise signal. Active noise control device. 6. The steepest gradient computing means comprises gradient average value computing means for computing an average value of the gradient in a plurality of consecutive processes for updating a filter coefficient that is sequentially executed according to the temporary reference signal. 6. The active noise control device according to claim 5, further comprising: a steepest gradient determining means that sets the steepest value among the average values to the steepest gradient. 7. The convergence coefficient determining means has an optimum value calculating means for calculating an optimum value of a filter coefficient of the adaptive digital filter based on the residual noise signal, and the optimum value of the filter coefficient is the steepest. 7. The active noise control device according to claim 5, wherein the convergence coefficient is determined by dividing the result by a safety factor. 8. The active noise control device according to claim 5, wherein the constant frequency is a frequency of the periodic noise in question. 9. A control vibration source capable of generating a control vibration that interferes with the periodic vibration transmitted from the vibration source, and a reference signal generating a reference signal representing a generation state of the periodic vibration in the vibration source. Generating means, an adaptive digital filter having a variable filter coefficient for filtering the reference signal to generate a drive signal for driving the controlled vibration source, and residual vibration for detecting the residual vibration after the interference and outputting it as a residual vibration signal. Active vibration control including detection means and adaptive processing means for updating the filter coefficient of the adaptive digital filter according to a stochastic gradient algorithm so as to reduce vibration after interference based on the reference signal and the residual vibration signal. In the device, a vibration source for generating the periodic vibration at a constant frequency from the vibration source at a predetermined timing and continuously for a predetermined time. Control means, a control vibration source stopping means for making the control vibration source not generate the control vibration continuously at the predetermined timing for a predetermined time, and a temporary reference signal having a frequency slightly deviated from the constant frequency. Then, the provisional reference signal generation means to be supplied to the adaptive processing means and the update amount in the update processing of the filter coefficient of the adaptive digital filter which is sequentially executed by the adaptive processing means when the provisional reference signal is supplied are determined. And a convergence coefficient determining means for determining a convergence coefficient used in the stochastic gradient algorithm based on the steepest gradient and the residual vibration signal. An active vibration control device characterized by the above. 10. The steepest gradient computing means comprises gradient average value computing means for computing an average value of the gradient in a plurality of consecutive processes for updating a filter coefficient that is sequentially executed according to the temporary reference signal. 10. The active vibration control device according to claim 9, further comprising: a steepest gradient determining means for determining the steepest value among the average values as the steepest gradient. 11. The convergence coefficient determination means has an optimum value calculation means for calculating an optimum value of a filter coefficient of the adaptive digital filter based on the residual vibration signal, and the optimum value of the filter coefficient is the steepest. And divide by
The active vibration control device according to claim 9 or 10, wherein a convergence factor is determined by multiplying the division result by a safety factor. 12. The active vibration control device according to claim 9, wherein the constant frequency is a frequency of the periodic vibration in question.