JPH08328569A - Fuzzy active noise erase device - Google Patents

Abstract

PURPOSE: To erase a noise flexibly by estimating error scatter without a C filter and by fuzzy control, related to a noise erase device actively erasing the noise such as an active noise controller, etc., applying a fuzzy theory. CONSTITUTION: This device is the noise erase device applying a noise erase signal to the noise and erasing the noise. Then, the noise erase device is the device provided with an error scatter system that an error erasing the noise is scattered timewisely, and is provided with a fuzzy operation part 8 performing a fuzzy inference for the error scattered by the error scatter system and an input signal and an adaptive filter 2 inputting the operation result of the fuzzy operation part 8 and the input signal and outputting the noise erase signal, and is constituted so that the optimum coefficient of the adaptive filter 2 is obtained based on the operation result of the fuzzy operation part 8, and the noise erase signal is generated.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an active noise canceling device such as an active noise control device to which a fuzzy theory is applied.

In particular, an active noise control device having a transmission system (error scattering system) that scatters errors in the time axis direction from the error detector to the adaptive filter, and an echo canceller for telephones, image ghosting, and control The present invention relates to an active noise canceller such as vibration control and seismic control.

[0003]

2. Description of the Related Art FIG. 12 shows the structure of a conventional noise canceller. In FIG. 12, 210 is an unknown system.

Reference numeral 211 is an adaptive filter. A noise elimination filter 212 generates a signal for eliminating noise.

A weight control unit 213 obtains an optimum filter coefficient of the noise elimination filter 212 by the normalized least squares method (NLMS). A C filter 214 is a filter for estimating the transfer function of the error scattering system.

Reference numeral 215 denotes an adder which adds a disturbance (noise) N _j to the signal h (i) passing through the unknown system 210. Reference numeral 216 denotes an adder, which adds a noise elimination signal to the signal that has passed through the unknown system 210 and added with noise.

Reference numeral 217 denotes an error scattering system, which is a transmission system from an error detection section (not shown) to the adaptive filter 211, and has scattering on the time axis. X _j is the input signal at time j.

H (i) is a signal transmitted through the unknown system 210. H _j is the transfer function of the cancellation filter at time j. H
_{j + 1} is the transfer function of the cancellation filter at time j + 1.

Y _j is the output of the C filter 214. G
_j is the output of the noise elimination filter H _j (212).
E _j is the output of the system and is the input of the error scattering system.

E _j is the output of the error scattering system. N _j is a noise signal. The transfer function of the system of FIG. 12 is expressed as:

H _{j + 1} = H _j + Ke _j Y _j / ΣY _{j -t} (Σ
Is the sum of t = 0 to N−1). However, N is the number of taps of the noise elimination filter 212, and K is a step gain.

The operation of the configuration of FIG. 12 will be described. The input signal X _j is transmitted through the signal transmission system h (i) and the noise N _j is added by the addition unit 215. Meanwhile, the C filter (scatter estimation filter) 214 estimates the error scattering based on the input signals X _j, outputs a Y _j. The weight control unit 213 uses Y _j and e _j.
The optimum filter coefficient is obtained by and. When the transfer function of the noise elimination filter 212 at time j is H _j , H _{j + 1}
= H _j + Ke _j Y _j / ΣY _{j −t} (Σ is the sum of t = 0 to N−1).

The noise elimination filter 212 uses X _j (j = 0).
~ N-1) and H _{j + 1} to calculate the convolution, G _j
Ask for. The adder 216 adds the noise-eliminating signal G _j to the noise-containing signal to generate a noise- _free signal E _j .

Further, at the next time point j + 1, similar processing is performed to obtain G _{j + 1} .

[0015]

In the conventional noise canceller, a C filter for estimating an error scattering system is provided, and the optimum filter coefficient of the adaptive filter is obtained by learning identification. Therefore, a large number of DSPs are used. It took a long time to process.

The present invention does not require a C filter,
It is an object of the present invention to provide a fuzzy active noise canceller capable of estimating error scattering and flexibly canceling noise by fuzzy control.

[0017]

SUMMARY OF THE INVENTION The present invention is an active noise canceller for actively applying a noise canceling signal to noise to cancel the noise. A fuzzy arithmetic unit having an error scattering system that scatters and performing fuzzy inference on the error and the input signal scattered by the error scattering system, and the operation result of the fuzzy arithmetic unit and the input signal are input to cancel noise. An adaptive filter that outputs a signal, and obtains the optimum coefficient of the adaptive filter based on the calculation result of the fuzzy calculation unit.
A noise canceling signal is actively generated.

FIG. 1 is a diagram showing the basic configuration of the present invention.
In FIG. 1, 1 is an unknown system.

Reference numeral 2 is an adaptive filter. Reference numeral 5 denotes a fuzzy arithmetic unit which performs fuzzy inference based on a fuzzy control rule for selecting a membership function of an input value and an output value of the fuzzy arithmetic unit 5 and a membership function to determine an output value.

Reference numeral 6 is an error scattering system. Fuzzy operation unit 5
In the figure, 11 is an error scattering calculation unit, and Y is determined by a sample value a (t) on the time axis of the impulse response of the error scattering system
_It seeks _j . For example, Y _j = Σa (t) X _j
It can be expressed as. Σ is the sum of t, and t =
0 to T-1 (T corresponds to the number of taps).

Reference numeral 16 denotes an error scattering inference unit, which performs fuzzy inference on the sample value a (t) on the time axis of the impulse response of the error scattering system. An evaluation unit 17 evaluates whether or not the noise elimination signal G _j is optimally obtained by fuzzy operation, and selects a fuzzy control rule, membership function, etc. when an appropriate output is not obtained. is there.

Reference numeral 21 is a membership function holding unit. Two
Reference numeral 2 is a fuzzy control rule holding unit. X _j is an input signal.

G _j is the output of the adaptive filter and is a noise elimination signal. E _j is the output. e _j is the output of the error scattering system 6.

H _j and H _{j + 1} are transfer functions of the applied filter.

[0025]

The operation of the basic configuration of the present invention shown in FIG. 1 will be described. Not yet
Input signal X containing noise that has passed through intelligence system 1_jIs the addition unit 8
Noise elimination signal G _jIs added and E_jIs output. Time
At time j, the error scattering system 6 is e_jIs output.

The fuzzy operation section 5 inputs the input values X _j , e _j, and inputs the sample values a (t) of X _j , e _j , K and error scattering.
For each of these, fuzzy inference is performed by the respective membership functions, for example, the MIN-MAX method. Then, based on the result of the fuzzy inference, the control data (step gain K, estimated value Y _{j of} error scattering, input value X _j , error e _j, etc.) of the filter coefficient of the adaptive filter is defuzzified by the centroid method or the like. .

The evaluation unit 17 evaluates e _j and changes the fuzzy rule and the membership function if the noise elimination signal G _j is not appropriate.

[0028]

EXAMPLE 1 FIG. 2 is a diagram showing Example 1 of the present invention. In FIG. 2, 1 is an unknown system.

Reference numeral 2 is an adaptive filter. 3 is a noise elimination filter. A weight control unit 4 determines the filter coefficient by the NLMS method.

Reference numeral 5 denotes a fuzzy arithmetic unit, which performs fuzzy inference based on a fuzzy rule for selecting a membership function and a membership function with respect to an input value and an output value of the fuzzy arithmetic unit 5 for defuzzification. is there.

Reference numeral 6 is an error scattering system. Fuzzy operation unit 5
In the above, reference numeral 10 denotes a defuzzification unit, which calculates the operation data of the weight control unit based on the integrated result obtained as a result of the fuzzy inference.

Reference numeral 11 denotes an error scattering calculation unit, which calculates Y _j from the sample value a (t) on the time axis of the impulse response of the error scattering system. Y _j = Σa (t) X _j . Reference numeral 12 is an E _d calculation unit, and E _d = Y _j / ΣY _j ²
Is calculated. Functions that represent error scattering are normal distribution, Poisson distribution, binomial distribution, and the like.

Reference numeral 13 denotes another numerical operation unit for calculating other values (X _j , e _j , K, etc.). 1
5 is a fuzzy inference unit.

Reference numeral 16 denotes an error scattering inference unit, which performs fuzzy inference on the sample value a (t) on the time axis of the impulse response of the error scattering system. An evaluation unit 17 evaluates whether or not the noise elimination signal G _j is optimally obtained by fuzzy operation, and selects a fuzzy control rule, membership function, etc. when an appropriate output is not obtained. is there.

Reference numeral 21 is a membership function holding unit. Two
Reference numeral 2 is a fuzzy control rule holding unit. X _j is an input signal.

N _j is noise. G _j is the output of the noise elimination filter. E _j is the output.

E _j is the output of the error scattering system 6. H _j ,
H _{j + 1} is the transfer function of the noise elimination filter. K is a step gain, which is obtained by defuzzification based on the integrated result of the fuzzy calculation results.

E _d = Y _j / ΣY _j ² . The operation of the first embodiment of the present invention shown in FIG. 2 will be described. In the system of FIG. 2, the input signal X _j is transmitted through the unknown system 1 and is affected by noise N _{j in} the process. Then, a noise elimination signal for eliminating the noise N _j is added to the signal including the noise, and the ideal error signal is E _j , but the error is affected by error scattering and the error is observed at e _j. It represents the system.

In the system of FIG. 2, the input signal X _j is transmitted through the unknown system 1, the noise N _j is added by the adder 7, and the noise canceling signal G _j is added by the adder 8, resulting in an error E _j .
E _j propagates through the error scattering system 6 and is observed as an error e _j .

The fuzzy operation unit 5 inputs the input values X _j and e _j , and performs fuzzy inference on the basis of the membership functions of X _j , e _j , K and a (t). For example, MIN-
The inference result is obtained by the MAX method or the like. Defuzzification unit 1
0 integrates the inference results and obtains X _j , e _j , K, and a (t) by, for example, the centroid method. Then, the error scattering calculator 11 calculates Y _j = Σa from X _j and a (t).
(T) Calculate X _j . The E _d calculation unit 12 calculates E _d = Y _j / ΣY _j ² from the value.

Then, when the transfer function of the noise canceling filter 3 at time j is H _j , the weight controller 4 sets the transfer function H _{j + 1} of the noise canceling filter 3 to H _{j + 1} = H _j + K _{e j.}
Calculate with E _d . Furthermore, the noise elimination filter 3 has X _j
Request (j = 0~N-1) and H _{j + 1} performs a convolution operation by G _j. The addition unit 8 adds the noise elimination signal G _j to the signal containing noise to generate a noise elimination signal E _j . Further, at the next time point j + 1, similar processing is performed, and E
_{Find j + 1} .

The evaluator 17 determines the degree of convergence of e _j (the size of the convergence), the range of the frequency that is effectively erased (frequency dependence), and the stability of the convergence (time change of the convergence) for each membership. Fuzzy inference is performed based on the function, and defuzzification is performed for evaluation. And so that the evaluation value becomes better,
Change membership control rules and membership functions.

FIG. 3 shows a second embodiment of the present invention, which is applied to an active noise control system for eliminating noise in a duct. In FIG. 3, reference numeral 2 is an adaptive filter.

Reference numeral 3 is a noise elimination filter. Reference numeral 4 is a weight control unit. 6 is an error scattering system.

Reference numeral 5 is a fuzzy operation section. 41 'is a duct. 42 'is a sensor microphone for detecting noise entering the duct.

Reference numeral 43 'is an error microphone for detecting noise (error signal) output from the duct. Four
Reference numeral 4'denotes a secondary sound source, which is a speaker that outputs a sound that eliminates noise.

Reference numeral 45 'is an addition unit, which is used for the sensor microphone 4
The detection signal of 2 and the output of the wraparound prevention filter 46 are added. Reference numeral 46 denotes a sneak-in prevention filter, which allows the sound generated from the secondary sound source 44 'to be transmitted to the sensor microphone 42.
The system for transmitting the sound is simulated to prevent the sound generated from the secondary sound source 44 'from being detected by the sensor microphone 42.

X _sj is the output of the sensor microphone 42. X
_S is the output of the wraparound prevention filter 46. X _j is the addition result of the addition unit 45.

E_jIs the output of the error microphone 43,
This is the output of the error scattering system. _jEquivalent to). E
_jIs the output of the error scattering system.

K, a (t) are outputs of the fuzzy operation section 5. The output Y _j of the conventional C filter is calculated by a fuzzy operation unit based on a (t) obtained by performing fuzzy inference using a membership function of a distribution representing scattering such as a normal distribution and defuzzification. By the time function a (t) representing the scattering, Y _j = Σa (t) X _jt (Σ is t = 0.
~ Accumulation of T-1 (T is the number of taps of the C filter))). The function that represents the error scattering is the normal distribution, the binomial distribution,
Poisson distribution, etc.

The operation of the configuration of FIG. 3 will be described. Duct 41
The noise X _{sj at} the time j of the _incoming noise is the sensor microphone 4
Detected in 2. The adder 45 adds the signal X _sj and the output X _sj of the wraparound prevention filter 46 to generate a signal X _j .

On the other hand, the secondary sound source 44 'outputs a sound for eliminating noise based on G _j output by the adaptive filter 2. The sound generated by the secondary sound source is added to the noise, scattered by the error scattering system 6, and the error is detected by the error microphone 43. The error microphone 43 outputs the error E _j (e described above).
corresponding to _j ).

The fuzzy operation unit 5 operates at X _j , E _j , K, a
Fuzzy inference is performed based on the membership function of (t). Then, the inference result is defuzzified so that K, a (t),
Ed is calculated and input to the weight control unit 4. Y _j = Σa
(T) Y _jt , and E _d = Y _j / ΣY _{j -i} ² .

The weight controller 4 uses the E _j calculated by the fuzzy calculator 5 and the step gains K and Ed to calculate the time j + 1.
Then, the transfer function H _{j + 1} of the noise elimination filter 3 is obtained. H
_{j + 1} = H _j + Ke _j E.

The noise canceling filter 3 receives the input X _j (j = 0).
~ N-1) and H _j are used to calculate the convolution,
The noise elimination signal G _j is output. The secondary sound source 44 'is G _j
The noise is deleted by adding the noise to the generated noise.

In the fuzzy operation section 5, the output E
_j is evaluated by fuzzy reasoning, and the fuzzy rules and membership are changed to appropriate ones. FIG. 4 shows an embodiment of the fuzzy arithmetic unit of the present invention.

In FIG. 4, reference numeral 5 is a fuzzy arithmetic unit. Reference numeral 10 is a defuzzification section.

Reference numeral 15 is a fuzzy inference unit. In the fuzzy inference section 15, 31 is X a _j inference unit, X _j MIN-M by the input signal of the membership function and X _j of
Fuzzy inference of X _j is performed by the AX method.

[0059] 32 is to be E _j inference unit, the fuzzy inference to E _j by MIN-MAX method by the input signal of the membership function and E _j of E _j. 33 is K
An inference unit, which has a membership function of K and X _j and E
Fuzzy inference of K (step gain) is performed by the MIN-MAX method based on the inference result of _j .

An error scattering inference unit 34 performs fuzzy inference on the error scattering coefficient a (t) by the MIN-MAX method based on the scattering error membership function and the inference results of E _j and X _j. It is a thing.

In the defuzzification section 10, 41 is X _j
An arithmetic unit integrates the fuzzy inference results of X _j, and requests the defuzzification values of X _j by gravity method.

[0062] 42 is a E _j arithmetic unit integrates the fuzzy inference results of E _j, and requests a value obtained by defuzzification the E _j by gravity method. 43 is a K operation unit,
The fuzzy inference result of K is integrated, and the value obtained by defuzzifying K by the centroid method is obtained.

Reference numeral 44 denotes an error scattering calculation unit
Fuzzy reasoning result of the reasoning section and X _jFuzzy inference section
Integrate inference results and calculate sample value a (t) of error scattering
Things. Y_j= Σa (t) X_jAsk in.

[0064] 45 is a E _d arithmetic unit, E by Y _j
It calculates _d . FIG. 5 is an embodiment of the membership function of the present invention. FIG. 5 (a) shows the membership function for the input signal X _j , VS (very small), S
(Small), M (medium size), B (large),
There are five types of VB (very large). Horizontal axis Suppo
The unit of rt is voltage.

FIG. 5B is a membership function for the output signal E _j , which is VS (very small), S (small), M (medium size), B (large), VB.
There are five types (very large). Horizontal axis Support
The unit of is voltage.

FIG. 5C shows the membership function with respect to the step gain K. VS (very small), S (small), M (medium size), B (large), VB
There are five types (very large).

FIG. 5 (d) is a membership function for a (t), VS (very small), S (small), M
There are five types: (medium size), B (large), and VB (very large).

FIG. 6 shows an example of the fuzzy control rule of the present invention. FIG. 6A is an example of the control rule of the step gain K, which is a rule for selecting the consequent part K with the membership functions of X _j and E _j as the antecedent part.

For example, when the antecedent part X _j is B and E _j is M, the consequent part K is B. Figure 6 (b) shows the antecedents X _j and E _j.
Is a fuzzy control rule for selecting the sample value a (t) of the impulse response of the error scattering system which is the consequent part.

For example, when the antecedent part X _j is M and E _j is B, the fuzzy control rule of the antecedent part a (t) is M. FIG. 7 is an example of a method of calculating the sample value a (t) of the miscalculation scattering of the present invention, and a case where the membership functions are set to three types of VS, M, and VB will be described.

FIG. 7A shows the membership function of X _j . FIG. 7B shows the membership function of E _j . Figure 7
(c) shows that when the membership function of X _j is VS, its support value is X _j0 , the membership function of E _j is M, and its support value is E _j0 , a (t) is MIN-MAX method. The results obtained by fuzzy inference are shown by.

In FIG. 7D, the membership function of X _j is M.
Then, when the support value is X _j0 , the membership function of E _j is M, and the support value is E _j0 , a (t) is M
The result obtained by fuzzy inference by the IN-MAX method is shown.

FIG. 7 (e) is an integration of the results of the fuzzy inference of FIGS. 7 (c) and 7 (d). Input signal X _j is X
_{It is assumed} that _j0 and the output signal E _j are E _j0 . The fuzzy control rule of the membership function of sample value a (t) of miscalculation is
It is as follows (assuming that the fuzzy control rule of the membership function of a (t) in FIG. 6 is followed).

When the antecedent part X _j is VS and the E _j is M, the antecedent part a (t) is VS ... (1) When the antecedent part X _j is M and E _j is M, the The subject part a (t) is M. (2) In the membership function VS of X _j grade for X _j0 μ (X _j0) is 0.7. Further, Grade mu (X _j0) for X _j0 in the membership function M is 0.3
Is.

[0075] grade for E _j0 in the membership function M of E _j μ (E _j0) is 0.6. So X
_When the membership function VS of _{j is} the antecedent part, the membership function of a (t) is VS according to the above condition (1).
Is. Then, according to the fuzzy operation rule of the MIN-MAX method, the smaller one of μ (X _j0 ) = 0.3 and μ (E _j0 ) = 0.6 is adopted, and the grade μ (a) = a (t) = 0.
3, the membership function of a (t) and the grade μ
The MAX operation with (a) = 0.3 is performed to obtain the result of the fuzzy inference of FIG. 7 (c).

Similarly, when the membership function M of X _{j is} the antecedent part, the membership function of a (t) is M according to the above condition (2). Then, according to the fuzzy operation rule of the MIN-MAX method, μ (X _j0 ) = _0.7 and μ
_Adopting the smaller one of (E _j0 ) = 0.6, the grade μ (a) of a (t) = 0.6, and the membership function of a (t) and the grade μ (a) = 0.6. MAX calculation with and is performed, and FIG. 7 (d) is obtained as a result of fuzzy inference.

Further, according to the operation rule of the MIN-MAX method, the fuzzy inference results of FIGS. 7 (c) and 7 (d) are integrated to obtain MAX and the fuzzy inference result for a (t) is obtained as shown in FIG. get e).

The sample value a _{0 at} that time is, for example, as shown in FIG.
The center of gravity of is calculated, and the value is set as the sample value a ₀ . The present invention
Observing the change of the output E _j , the fuzzy control rule of the membership function is changed so that E _j quickly approaches the target value.

FIG. 8 shows an example of the method of changing the fuzzy control rule of the present invention. FIG. 8A shows the case where the degree of convergence is small. (a) -1 is a method of changing the control rule of the step gain K. Move back one direction to the left.

For example, when X _j is VS, E _j is B, and the step gain K is M, K is one diagonally lower left S,
X _j is changed to S and E _j is changed to B. (a) -2 is a rule for changing the sample value a (t) of the impulse response. Advance one to the left.

For example, when X _j is VS, E _j is VS, and the sample value a (t) of the impulse response is VS, a (t) selects a VS one diagonally higher to the left, and X _j is S , E _j is changed to S.

FIG. 8B shows a case where the erase frequency bandwidth is narrow. (a) -3 is a method of changing the control rule of the step gain K. Proceed in the y-axis direction (up).

(A) -4 is a sample value a of the impulse response
It is a control rule of (t). Move backward in the downward direction. FIG. 8 (c) shows the case where the convergence is not stable.

(A) -5 is a method of changing the control rule of the step gain K. Proceed in the y-axis direction (down). (a)-
6 is a method of changing the control rule of the sample value a (t) of the impulse response. Move backward in the upward direction.

FIG. 9 is an explanatory diagram of an output fuzzy evaluation method of the present invention. FIG. 9A is an explanatory diagram of changes in E _j before and after erasing. The vertical axis represents E _j and the horizontal axis represents frequency f. Observe the difference between before and after erasing, and evaluate the degree of reduction.

In the evaluation, the degree of convergence is judged, the judgment is made in the frequency domain, and the stability of convergence is judged. Judgment of Convergence Degree The judgment of the degree of convergence will be described with reference to FIG.

The determination of the degree of convergence is performed by collecting the sampled values of E _j several times, for example, 16 to 64 times, and averaging them to eliminate the suddenly large E _j and average them. Make a decision.

Judgment in Frequency Domain The judgment in the frequency domain will be described with reference to FIG.
The vertical axis of FIG. 9B is E _j (f), which is a function of E _j with respect to frequency. The horizontal axis is frequency.

F _L is the minimum frequency that can be erased _,
f _U is the maximum frequency that can be erased. By FFT (Fast Fourier Transform) E
_j (f) is obtained and the erase amount is evaluated.

Convergence stability is determined with reference to FIG. 9C. In FIG. 9C, the vertical axis is the output time function E _j (t). The horizontal axis is time.

When the minimum value of E _j is detected, it is sampled several times with that value as a reference, and if it is still near the minimum value, it is determined that the convergence state is stable. FIG. 10 is an explanatory diagram of the error evaluation method of the present invention.

FIG. 10A shows a method for comprehensively evaluating the present invention.
Show. In the present invention, the degree of convergence E_j, Convergence stability E_j
(T), convergence frequency band E _jAbout (f)
Evaluate and find the state that minimizes each.

FIG. 10 (b) shows an example of the membership function for evaluating the error of the present invention. The horizontal axis is the error evaluation (E _j , E _j (f), E _j (t)). The vertical axis represents the grade.

Very Bad (very bad), Sos
Based on the membership function of o (acceptable) and Excellent (very good), fuzzy inference is performed by the MIN-MAX method, etc., and the obtained inference result is defuzzified by the centroid method, etc. Obtain an evaluation value.

FIG. 11 shows an example of a membership function changing method according to the present invention. FIG. 11A shows an example of the VS membership function and its arithmetic expression. By changing the values of S ₂ and Sm, it is possible to make a parallel translation in the lateral direction.

FIG. 11B shows an example of the membership function of M and its arithmetic expression. By changing the values of S ₁ , S ₃ and Sm, it is possible to perform parallel translation in the lateral direction. FIG.
1 (c) is an example of the membership function of VB and its arithmetic expression. It can be moved in parallel in the lateral direction by changing the value of S _4, Sm.

[0097]

According to the present invention, the error scattering system can be estimated by a program without using the C filter, and the sample value a (t) of the impulse response when the error of the error scattering system is scattered in the time axis direction. ) Can be shortened. Therefore, learning of the adaptive filter can be speeded up. In addition, the fuzzy control rule of the membership function and the membership function can be corrected even while the system is operating, and the erasing effect by learning can be greatly improved.

Further, since the error of noise elimination is regarded as the stochastic process of the error scattering system and controlled by the fuzzy mapping, the noise elimination can be flexibly performed.

[Brief description of drawings]

FIG. 1 is a diagram showing a basic configuration of the present invention.

FIG. 2 is a diagram showing a first embodiment of the present invention.

FIG. 3 is a diagram showing a second embodiment of the present invention.

FIG. 4 is a diagram showing an embodiment of a fuzzy operation unit of the present invention.

FIG. 5 is a diagram showing an example of a membership function.

FIG. 6 is a diagram showing an example of a fuzzy control rule.

FIG. 7 is a diagram illustrating an example of a method of calculating a sample value a (t) of error scattering.

FIG. 8 is a diagram showing an example of a method for changing a fuzzy rule.

FIG. 9 is an explanatory diagram of a fuzzy evaluation method.

FIG. 10 is a diagram showing an error evaluation method of the present invention.

FIG. 11 is a diagram showing an example of a membership function changing method of the present invention.

FIG. 12 is a diagram showing a conventional technique.

[Explanation of symbols]

 1: Unknown system 2: Adaptive filter 5: Fuzzy operation unit 6: Error scattering system 11: Error scattering operation unit 16: Error scattering inference unit 17: Evaluation unit 21: Membership function holding unit 22: Fuzzy control rule holding unit

Claims (4)

[Claims] 1. In an active noise canceller for canceling noise by actively applying a noise canceling signal to noise, the active noise canceller has an error scattering system in which an error canceling noise is scattered over time. A fuzzy operation unit for performing fuzzy inference on the error scattered by the error scattering system and the input signal, and an adaptive filter for inputting the operation result of the fuzzy operation unit and the input signal and outputting a noise cancellation signal. A fuzzy active noise eliminator, which is characterized in that it obtains an optimum coefficient of an adaptive filter based on a calculation result of a fuzzy calculator and generates a noise canceling signal. 2. The fuzzy operation unit is a sample value a of error scattering.
(T) is fuzzy calculated and Y _j = Σa (t) X _j (where X _j is the input signal, Σ is the sum for t = 0 to T−1, and T corresponds to the number of taps. The fuzzy active noise canceller according to claim 1, wherein the error scattering system is estimated by the value of 3. A fuzzy active noise canceller, characterized in that the fuzzy arithmetic unit comprises an evaluation unit for evaluating an error of noise cancellation and a membership function for error evaluation, and fuzzy evaluates the error. 4. The fuzzy active noise canceller according to claim 1, wherein the fuzzy active noise canceller is for canceling noise in the duct.