CN110599997B - Impact noise active control method with strong robustness - Google Patents

Abstract

An impact noise active control method with strong robustness mainly comprises the following steps: A. acquiring a reference signal, wherein the discrete value x (n) of the noise signal of the current moment n is acquired by a reference microphone, and the input signal vector of a filter is X (n); B. generating a filter coefficient, and filtering the noise input vector X (n) by a filter according to the weight coefficient vector W (n) to obtain an output value y (n) of the filter; C. generating a noise-canceling signal, and obtaining a noise-canceling signal y' (n) after the output value y (n) of the filter passes through a secondary path S (z); D. calculating the average value of p-order moments of the residual signals; E. exponential residual signal p-order moment average c_pAnd (n), F, updating the calculation of the gradient value delta (n), G, and updating to obtain W (n +1) as the weight coefficient vector of the next time n +1, wherein W (n +1) is W (n) + mu delta (n) x (n). The method has strong robustness to impact noise, small steady-state error and good noise reduction performance.

Description

A Robust Active Control Method for Impulse Noise

technical field

The present invention relates to an active noise control method, in particular to an active control method of impulse noise.

Background technique

With the rapid development of modern industry, various forms of noise, such as automobile engine noise, train noise, transformer noise, etc., have gradually become a non-negligible interference factor in people's daily life and production. Excessive noise will affect people's normal life, reduce labor productivity, endanger human health, and even directly lead to the occurrence of certain diseases. Therefore, how to effectively reduce the environmental noise has become an urgent problem to be solved.

The traditional noise reduction technology is passive noise control (PNC), which mostly uses acoustic materials to reflect and absorb sound waves. Although this method is widely used, it only has a good effect on high-frequency noise and low-frequency noise. not so useful. For low-frequency noise such as transformer noise and motor noise, a new noise control method is required.

The principle of active noise control (ANC) is to collect the source noise through the reference microphone at the noise source, and convert the input noise into a digital signal through an analog-to-digital converter, and then through the adaptive control algorithm, the circuit and the speaker send out the noise. A secondary sound wave with the same frequency and opposite phase as the noise amplitude. The secondary sound wave cancels the waveform of the source noise to achieve noise control.

Residual-minimized mean square (FxLMS) filtering algorithm has become the main algorithm for active noise control due to its low computational cost and compact structure, and has been widely used in ordinary Gaussian noise environments. But for the shock noise with large amplitude and drastic change, the Minimizing Residual Mean Square (Second Moment of Residual) (FxLMS) filter algorithm has high sensitivity to shock noise, low stability of the algorithm, and poor noise control effect; The p-norm filtering algorithm that minimizes the residual has a significant improvement in the control effect of impulsive noise, such as document 1 "Leahy R, Zhou Z, Hsu YC. Adaptive filtering of stableprocesses for active attenuation of impulsive noise. on Acoustics, Speech, and Signal Processing, vol. 5; 1995.pp. 2983-2986.", which performs noise reduction by minimizing the p-th moment of the residual (p is a definite number between 1-2). Suppression reduces the sensitivity to shock noise, and the stability of the algorithm is improved to a certain extent; however, for strong shock noise, the algorithm has a limited inhibitory effect, and it is prone to iterative non-convergence and instability. Its robustness needs to be improved.

SUMMARY OF THE INVENTION

The purpose of the present invention is to provide an active control method for impulse noise with strong robustness, which has small steady-state error for impulse noise, strong robustness and good noise reduction effect.

The technical solution adopted by the present invention to achieve the object of the invention is a robust active control method for impulse noise, the steps of which are as follows:

A. Noise signal acquisition

The reference microphone near the noise source collects the noise signal discrete value x(n) at the current time n, and divides the noise signal discrete value x(n), x(n-1), from the current time n to the previous L-1 time. ..,x(n-L+1), constitute the noise signal vector X(n) at the current moment n, X(n)=[x(n),x(n-1),...,x(n -L+1)] ^T ; Wherein L=128, is the tap number of filter, superscript T represents transpose operation;

B, filter coefficient generation

The filter generates the weight coefficients w(n), w(n-1),...,w(n-L+1) of the current moment n and the previous L-1 moments, and these L weight coefficients constitute the current moment n The weight coefficient vector W(n) of , W(n)=[w(n),w(n-1),...,w(n-L+1)]; when the current moment n<129, W( n)=0;

C. Denoising signal generation

Input the noise signal vector X(n) of the current moment n into the filter to obtain the output value y(n) of the current moment n of the filter, y(n)=W ^T (n)X(n);

The output value y(n) of the filter, through the secondary path S(z) composed of D/A, reconstruction filter, power amplifier, noise canceling speaker, and error microphone at the noise canceling point, is obtained at the noise canceling point De-noising signal y'(n), y'(n)=s(n)*y(n); the symbol * represents the convolution operation, and s(n) represents the impulse response of the secondary path S(z);

D. Calculation of the average value of the p-order moment of the residual signal

式中|e(n)|^p表示残差信号e(n)的p阶矩，dp 表示对阶矩p的微分，|e(n)|表示残差信号e(n)的绝对值；The error microphone collects the noise canceling point at the current time n of the noise canceling signal y'(n) and the noise signal discrete value x(n) at the current time n. The sound signal is used as the residual signal e(n) at the current time n. , and send the filter; the filter calculates the average value of the p-order moment of the residual signal at the current time n in the interval p∈[1,2] from the residual signal e(n) of the current time _n ( n),

where |e(n)| ^p represents the p-order moment of the residual signal e(n), dp represents the differential to the order moment p, and |e(n)| represents the absolute value of the residual signal e(n);

E. Exponentialization of the mean value of the p-order moment of the residual signal

The filter calculates the exponential average value g(n) of the p-order moment of the residual signal at the current time n, g(n)=exp{-ηc _p (n)}, where exp( ) represents the exponential operation, η is an indexation parameter, and its value is a positive number less than 10;

F. Calculation of gradient vector

其中，sign(·)表示符号运算；The filter calculates the updated gradient value Δ(n) of the weight of n at the current moment,

Among them, sign( ) represents symbolic operation;

G. Update of the weight coefficient vector

The filter updates the gradient value Δ(n) according to the weight value of n at the current moment, and the updated weight coefficient vector of n+1 at the next moment is W(n+1), w(n+1)=w(n)+μΔ( n)X(n); in the formula, μ is the step factor, and its value ranges from 0.01 to 0.1;

H. Iteration

Let n=n+1, repeat the steps A to G until the noise control ends.

Compared with the prior art, the beneficial effects of the present invention are:

代替文献1的一个p值确定的p阶矩|e(n)|^p；所有p值的p阶矩的平均值能适应大多数种类的冲击噪声，其适应性更强；也不需要先验知识进行p值的选择，从而更易于冲击噪声有源控制的实现。1. In a strong impact noise environment, the residual signal will fluctuate, and noise control has higher requirements on the stability of the control method; compared with the p-order moment determined by a p-value of the residual signal in Reference 1 |e(n)| ^p is used as the basis for estimating noise impact, the present invention adopts the average value of p-order moments of all p-values in the p∈[1,2] interval of the residual signal, c _p (n),

Instead of the p-order moment |e(n)| ^p determined by a p-value in Reference 1; the average value of the p-order moment of all p values can adapt to most types of impact noise, and its adaptability is stronger; it does not require a priori Knowledge of p-value selection makes it easier to implement active control of shock noise.

2. The generalized maximum entropy exp(-η|e(n)| ^p ) of the residual signal e(n) is the higher-order absolute moment of the residual signal including the second-order moment in the mapping feature space. The invention draws on the mapping mechanism of generalized maximum entropy, and performs an operation similar to generalized maximum entropy to the average value C _p (n) of the p-order moment of the residual signal in the interval p∈[1,2], and obtains the average value of the p-order moment. The indexed value g(n) of c _p (n), g(n)=exp{-ηC _p (n)}, and the minimum indexed value g(n) is used as the criterion for noise estimation, and the filter weight is obtained. The update increment value of the value coefficient and the exponential update increment value can not only effectively track the change of the residual signal e(n) in the impact noise environment, Great changes can also be effectively slowed down, which significantly reduces the occurrence of non-convergence and instability of iterations, and has strong robustness; the steady-state error to impact noise is small, and the noise reduction effect is good.

The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

Description of drawings

FIG. 1( a ) is the shock noise of α stable distribution with α=1.6 adopted in the simulation experiment of the present invention.

Fig. 1(b) is the strong impact noise with α stable distribution of α=1.3 adopted in the simulation experiment of the present invention.

Fig. 1(c) is the super-strong shock noise with α stable distribution of α=1.1 adopted in the simulation experiment of the present invention.

FIG. 2 a is a graph of the average noise residual after the impulse noise of FIG. 1 a is processed by the simulation experiments of Document 1 and the method of the present invention.

FIG. 2b is a graph of the average noise residual after the primary noise of FIG. 1b is processed by the simulation experiments of Document 1 and the method of the present invention.

FIG. 2c is a graph of the average noise residual after the primary noise of FIG. 1c is processed by the simulation experiments of Literature 1 and the method of the present invention.

Detailed ways

Example

A specific embodiment of the present invention is a robust impulse noise active control method, the steps of which are as follows:

A. Noise signal acquisition

The reference microphone near the noise source collects the noise signal discrete value x(n) at the current time n, and divides the noise signal discrete value x(n), x(n-1), from the current time n to the previous L-1 time. ..,x(n-L+1), constitute the noise signal vector X(n) at the current moment n, X(n)=[x(n),x(n-1),...,x(n -L+1)] ^T ; Wherein L=128, is the tap number of filter, superscript T represents transpose operation;

B, filter coefficient generation

The filter generates the weight coefficients w(n), w(n-1),...,w(n-L+1) of the current moment n and the previous L-1 moments, and these L weight coefficients constitute the current moment n The weight coefficient vector W(n) of , W(n)=[w(n),w(n-1),...,w(n-L+1)]; when the current moment n<129, W( n)=0;

C. Denoising signal generation

Input the noise signal vector X(n) of the current moment n into the filter to obtain the output value y(n) of the current moment n of the filter, y(n)=W ^T (n)X(n);

The output value y(n) of the filter, through the secondary path S(z) composed of D/A, reconstruction filter, power amplifier, noise canceling speaker, and error microphone at the noise canceling point, is obtained at the noise canceling point De-noising signal y'(n), y'(n)=s(n)*y(n); the symbol * represents the convolution operation, and s(n) represents the impulse response of the secondary path S(z);

D. Calculation of the average value of the p-order moment of the residual signal

式中|e(n)|^p表示残差信号e(n)的p阶矩，dp 表示对阶矩p的微分，|e(n)|表示残差信号e(n)的绝对值；The error microphone collects the noise canceling point at the current time n of the noise canceling signal y'(n) and the noise signal discrete value x(n) at the current time n. The sound signal is used as the residual signal e(n) at the current time n. , and send the filter; the filter calculates the average value of the p-order moment of the residual signal at the current time n in the interval p∈[1,2] from the residual signal e(n) of the current time _n ( n),

where |e(n)| ^p represents the p-order moment of the residual signal e(n), dp represents the differential to the order moment p, and |e(n)| represents the absolute value of the residual signal e(n);

E. Exponentialization of the mean value of the p-order moment of the residual signal

The filter calculates the exponential average value g(n) of the p-order moment of the residual signal at the current time n, g(n)=exp{-ηc _p (n)}, where exp( ) represents the exponential operation, η is an indexation parameter, and its value is a positive number less than 10;

F. Calculation of updated gradient values

其中，sign(·)表示符号运算；The filter calculates the updated gradient value Δ(n) of the weight of n at the current moment,

Among them, sign( ) represents symbolic operation;

G. Update of the weight coefficient vector

The filter updates the gradient value Δ(n) according to the weight value of n at the current moment, and the updated weight coefficient vector of n+1 at the next moment is W(n+1), w(n+1)=w(n)+μΔ( n)X(n); in the formula, μ is the step factor, and its value ranges from 0.01 to 0.1;

H. Iteration

Let n=n+1, repeat the steps A to G until the noise control ends.

Simulation:

In order to verify the effectiveness of the present invention, a simulation experiment is carried out, and the method is compared with the method in Reference 1.

The primary and secondary paths of the simulation experiment are modeled by high-order FIR filters. The order L of the filter is set to 128.

The shock noise is modeled by the standard Alpha stable distribution noise model. The characteristic function of the Alpha stable distribution is in the form of φ(t)=e ^-|t|α , and the smaller the value of α, the stronger the shock noise.

The experiment adopts the shock noise of three α stable distributions (α=1.6, 1.3, 1.1), as shown in Fig. 1(a), Fig. 1(b), and Fig. 1(c).

For the three impulse noises shown in Figure 1(a), Figure 1(b), and Figure 1(c), the average noise residual after the simulation noise control is performed by the method of Document 1 and the present invention (at the noise elimination point, the noise control process Figure 2a, Figure 2b, Figure 2c. In Figure 2a, Figure 2b, Figure 2c, the curves connected by the symbol "○" are the average noise residual curve of the method of document 1, and the curve connected by the symbol "☆" is the average noise residual curve of the method of the present invention.

It can be seen from the simulation results of Fig. 2a, Fig. 2b and Fig. 2c that the method of the present invention has better control over the impact noise, and the average noise residual is much lower than that of the method in Reference 1; the stronger the impact noise, the more obvious the advantages of the method of the present invention are ; In Figure 2b, Figure 2c, the method of Reference 1 cannot even converge. It can be seen that the method of the present invention has strong robustness, good stability, small steady-state error and good noise reduction effect against impact noise.

Claims (1)

1. A robust impulse noise active control method, the steps of which are as follows: A. Noise signal acquisition The reference microphone near the noise source collects the noise signal discrete value x(n) at the current time n, and divides the noise signal discrete value x(n), x(n-1), from the current time n to the previous L-1 time. ..,x(n-L+1), constitute the noise signal vector X(n) at the current moment n, X(n)=[x(n),x(n-1),...,x(n -L+1)] ^T ; Wherein L=128, is the tap number of filter, superscript T represents transpose operation; B, filter coefficient generation The filter generates the weight coefficients w(n), w(n-1),...,w(n-L+1) of the current moment n and the previous L-1 moments, and these L weight coefficients constitute the current moment n The weight coefficient vector W(n) of , W(n)=[w(n),w(n-1),...,w(n-L+1)]; when the current moment n<129, W( n)=0; C. Denoising signal generation Input the noise signal vector X(n) of the current moment n into the filter to obtain the output value y(n) of the current moment n of the filter, y(n)=W ^T (n)X(n); The output value y(n) of the filter, through the secondary path S(z) composed of D/A, reconstruction filter, power amplifier, noise canceling speaker, and error microphone at the noise canceling point, is obtained at the noise canceling point Denoising signal y'(n), y'(n)=s(n)*y(n); the symbol * represents the convolution operation, and s(n) represents the impulse response of the secondary path S(z); D. Calculation of the average value of the p-order moment of the residual signal The error microphone collects the noise canceling signal y′(n) at the current time n at the noise elimination point and the sound signal after the action of the noise signal discrete value x(n) at the current time n, as the residual signal e(n) at the current time n , and send the filter; the filter calculates the average value of the p-order moment of the residual signal at the current time n in the interval p∈[1,2] from the residual signal e(n) of the current time _n ( n),

where |e(n)| ^p represents the p-order moment of the residual signal e(n), dp represents the differential to the order moment p, and |e(n)| represents the absolute value of the residual signal e(n); E. Exponentialization of the mean value of the p-order moment of the residual signal The filter calculates the exponential average value g(n) of the p-order moment of the residual signal at the current time n, g(n)=exp{-ηc _p (n)}, where exp( ) represents the exponential operation, is an indexation parameter, and its value is a positive number less than 10; F. Calculation of gradient vector The filter calculates the updated gradient value Δ(n) of the weight of n at the current moment,

Among them, sign( ) represents symbolic operation; G. Update of the weight coefficient vector The filter updates the gradient value Δ(n) according to the weight value of n at the current moment, and the updated weight coefficient vector of n+1 at the next moment is W(n+1), w(n+1)=w(n)+μΔ( n)X(n); in the formula, μ is the step factor, and its value ranges from 0.01 to 0.1; H. Iteration Let n=n+1, repeat the steps A to G until the noise control ends.

Images