CN110335582B - An active noise reduction method suitable for active control of impulse noise - Google Patents

Abstract

The invention provides an active noise reduction method suitable for impulse noise active control, which comprises the following steps: step one, adopting a reference microphone to collect impulse noise emitted by a noise source as a reference signal, and adjusting step length parameters by using a Gaussian distribution function with the reference signal as an independent variable; collecting noise after cancellation by an error microphone as an error signal, and updating a prediction filter weight coefficient according to a continuously updated reference signal, the error signal and the adjusted step length parameter; and thirdly, calculating a counteracting signal through the updated weight coefficient of the prediction filter, and sending the counteracting signal through a loudspeaker to counteract the reference signal so as to perform active noise reduction.

Description

An active noise reduction method suitable for active control of impulse noise

Technical field

The invention relates to the field of active noise control, and also relates to an active noise control method. Specifically, it relates to an active noise reduction method suitable for active control of pulse noise.

Background technique

With the in-depth development of modern industry, construction, transportation and other fields, the problem of noise pollution has become increasingly prominent, seriously affecting residents' study, work and life.

Existing noise control technologies are mainly divided into two categories: Passive Noise Control (PNC) and Active Noise Control (ANC). Passive noise reduction technology uses traditional sound absorption, sound isolation and silencing to achieve target noise suppression. It can achieve ideal results in mid- and high-frequency noise control, but it is unsatisfactory in low-frequency noise control. In contrast, active noise reduction technology (also known as active noise control technology) uses the principle of destructive interference of sound waves and uses an adaptive control algorithm to achieve cancellation of the secondary sound source and the target noise with approximately the same phase and opposite amplitude. Signal. Active noise control systems are composed of controllers, speakers and microphones. They are usually small in size and highly versatile, and can be arranged flexibly in specific application scenarios. Therefore, they are widely used in industry, transportation, military and other fields.

With the in-depth development of adaptive filtering technology and signal processing systems, active noise reduction technology has developed rapidly in the past 30 years. Domestic and foreign research teams have made many breakthroughs in the core control algorithm and hardware design of active noise reduction. However, there are still some problems in active noise reduction technology that directly limit the practical application of active noise control systems. Among them, active impulse noise control (Active impulse noise control) is one of the core problems.

The classic active noise reduction algorithm has an ideal control effect on noise signals that obey Gaussian distribution. However, impulse noise usually obeys α steady-state distribution rather than Gaussian distribution. Therefore, when using classical algorithms such as FxLMS to control impulse noise, phenomena such as failure and system divergence often occur. Impulse noise is usually modeled by α steady-state distribution, whose characteristic function is Among them, α is the characteristic index of impulse noise. The smaller the α value, the more significant the pulse characteristics of the signal and the more intense the pulse intensity.

In order to achieve effective control of impulse noise, many researchers have conducted relevant explorations and proposed a series of active impulse noise reduction algorithms. Typical representatives of these algorithms include: FxLMP (Filtered-x least mean p-norm) algorithm, which achieves active control of impulse noise by minimizing the p-order moment of the error signal. Compared with the classic FxLMS algorithm, the FxLMP algorithm has better pulse suppression performance, but its shortcoming is that the algorithm relies on the prior acquisition of the impulse noise characteristic index α. Therefore, it is universal for different intensity impulse noises. The performance is poor; Akhtar et al. proposed the Th-FxLMS (Thresholding-based FxLMS) algorithm, which uses a limiting function to suppress the amplitude of pulse samples in the reference signal and error signal, thereby improving the pulse noise reduction performance of the system. However, , the necessary estimation of the limiting threshold in the limiting function is an obvious shortcoming of this algorithm, because the limiting threshold needs to be re-estimated for different intensities of impulse noise; Wu Lifu et al. proposed FxlogLMS (Filtered-xlogarithmic error LMS) Algorithm, this algorithm uses the mean square value of the logarithmic transformation of the error signal as the cost function to adaptively update the weight coefficient of the filter, thereby achieving effective control of impulse noise. The FxlogLMS algorithm does not require a priori acquisition of the characteristic index α, and does not require A limiting threshold was introduced, and at the same time, the algorithm applied for a relevant patent and was authorized (impulse noise active control method based on logarithmic transformation, ZL201010017642.4). However, its disadvantage is that when the error signal amplitude is less than 1, there is a dead zone in the algorithm's filter coefficient update, causing the algorithm's noise reduction performance and convergence speed to slightly deteriorate.

It can be seen that proposing an active impulse noise control method with strong adaptive ability, excellent noise reduction performance and fast convergence speed is crucial for the effective suppression of impulse noise in actual scenarios.

Contents of the invention

The present invention designs and develops an active noise reduction method suitable for active control of pulse noise. This method does not require a priori acquisition of characteristic indexes when performing active control of pulse noise; there is no need to introduce limiting thresholds, and more cumbersome thresholds are avoided. Estimation; at the same time, when the error signal amplitude is less than 1, there is no dead zone in the prediction filter weight iteration.

The technical solution provided by the invention is:

An active noise reduction method suitable for active control of impulse noise, including the following steps:

Step 1: Use a reference microphone to collect the pulse noise emitted by the noise source as a reference signal, and adjust the step size parameter using the Gaussian distribution function of the reference signal as an independent variable;

Step 2: Use an error microphone to collect the canceled noise as an error signal, and update the prediction filter weight coefficient according to the continuously updated reference signal, the error signal and the adjusted step parameter;

Step 3: Calculate a cancellation signal through the updated prediction filter weight coefficient, and send the cancellation signal through the speaker to cancel the reference signal, thereby performing active noise reduction.

Preferably, in step one, the adjustment formula of the step parameter is:

In the formula, n is the time series, x(n) is the reference signal, μ[x(n)] is the step parameter with the reference signal as the independent variable, is the basic step parameter, and σ is the standard deviation of the distribution function.

Preferably, in step two, the weight coefficient of the prediction filter is updated through a weight adaptive iterative formula.

Preferably, the weight adaptive iteration formula is expressed as:

w(n+1)=w(n)-μ·▽J(n);

In the formula, w(n) is the weight coefficient of the prediction filter at time n, w(n+1) is the weight coefficient of the prediction filter at time n+1, J(n) is the cost function of the adaptive iteration formula, ▽J(n) is the gradient of J(n).

Preferably, the cost function of the adaptive iteration formula is expressed as:

J(n)＝E{f ² [e(n)]}≈f ² [e(n)];

In the formula, e(n) is the error signal, and f[e(n)] is the nonlinear transformation function of the error signal.

Preferably, the nonlinear transformation function of the error signal is expressed as:

Preferably, the cost function gradient is expressed as:

In the formula, x _f (n) is the reference signal after passing through the secondary path.

Preferably, the cancellation signal obtained based on the weight coefficient of the prediction filter is expressed as:

y(n)=w ^T (n)·x(n);

In the formula, y(n) is the cancellation signal at time n, and T is the matrix transpose.

Compared with the prior art, the present invention has the following beneficial effects:

1. The present invention proposes an active noise reduction method suitable for active control of impulse noise. This method does not require the introduction of a limiting threshold, thus avoiding the more cumbersome threshold estimation; there is no need to obtain a priori the characteristic index of the impulse noise. ;At the same time, there is no dead zone in the prediction filter weight iteration.

2. The present invention uses the error signal nonlinear transformation function to perform robust processing on the error signal participating in the prediction filter weight iteration, attenuating the impact of the pulse samples in the error signal on the system noise reduction performance and stability; at the same time, using the reference The signal is a Gaussian distribution function of the independent variable. The step size parameter is adaptively adjusted, which weakens the impact of pulse samples in the reference signal on the system noise reduction performance and stability. This further improves the pulse suppression performance of the control system and is better than traditional pulse noise reduction. The method has better comprehensive performance.

Description of drawings

Figure 1 is a schematic diagram of the principle of an active noise reduction method suitable for active control of impulse noise according to the present invention.

Figure 2 is a schematic diagram of a high-intensity pulse noise signal according to an embodiment of the present invention.

Figure 3 is a schematic diagram of a low-intensity pulse noise signal according to an embodiment of the present invention.

FIG. 4 is a schematic diagram of the amplitude-frequency response curves of the primary channel and the secondary channel used in the embodiment of the present invention.

FIG. 5 is a schematic diagram of the phase-frequency response curves of the primary channel and the secondary channel used in the embodiment of the present invention.

Figure 6 is a schematic diagram comparing the noise reduction effects under high-intensity pulse noise signals according to the embodiment of the present invention.

Figure 7 is a schematic diagram comparing noise reduction effects under low-intensity pulse noise signals according to the embodiment of the present invention.

Detailed ways

The present invention will be further described in detail below with reference to the accompanying drawings, so that those skilled in the art can implement it with reference to the text of the description.

When the present invention performs active noise reduction for impulse noise, the noise reduction process is as follows:

Step 1. Use the reference microphone to collect the pulse noise emitted by the noise source as the reference signal x(n), and adjust the step size parameter using the Gaussian distribution function of the reference signal x(n) as the independent variable;

As shown in Figure 1, in another embodiment, the step parameter is adjusted by the step parameter adaptive adjustment sub-module, and the adjustment method is:

In the formula, n is the time series, μ[x(n)] is the step parameter with the reference signal as the independent variable, is the basic step parameter, and σ is the standard deviation of the distribution function.

Step 2: Use an error microphone to collect the canceled noise as the error signal e(n), and adaptively update the prediction filter coefficients based on the constantly updated reference signal, error signal and adjusted step parameter;

In another embodiment, the weight update submodule uses the weight adaptive iteration formula to update the prediction filter coefficients in real time based on the step parameter, the reference signal and the error signal. The weight adaptive iteration formula is based on the steepest descent principle, indicating for:

w(n+1)=w(n)-μ·▽J(n);

In the formula, w(n) is the weight coefficient of the prediction filter at time n, w(n+1) is the weight coefficient of the prediction filter at time n+1, J(n) is the cost function of the adaptive iteration formula, ▽J(n) is the gradient of J(n).

In another embodiment, the cost function of the adaptive iteration formula is expressed as:

J(n)＝E{f ² [e(n)]}≈f ² [e(n)];

In the formula, f[e(n)] is the nonlinear transformation function of the error signal.

In another embodiment, the nonlinear transformation function of the error signal is expressed as:

In another embodiment, the gradient of the cost function is expressed as:

Furthermore, the weight adaptive iteration formula is:

In the formula, x _f (n) is the reference signal filtered by the estimated secondary path.

Step 3: Calculate the cancellation signal through the prediction filter weight coefficient, and send the cancellation signal through the speaker to actively cancel the reference signal.

In another embodiment, the cancellation signal is calculated and sent by the prediction filter sub-module. The cancellation signal and the reference signal have approximately equal amplitudes and opposite phases. In practical applications, the cancellation signal is sent by the loudspeaker. The cancellation signal is expressed as:

y(n)=w ^T (n)·x(n);

In the formula, y(n) is the cancellation signal, and T is the matrix transpose.

In another embodiment, the reference signal x(n) is the signal d(n) after passing through the primary path P(z), and the reference signal x(n) is estimated to be the secondary path The resulting signal x _f (n) is calculated as follows:

d(n)=P(z)*x(n);

Among them, * is the convolution operation, P(z) is the primary path between the reference microphone and the error microphone, is the estimated secondary pathway.

In another embodiment, the signal after the cancellation signal y(n) emitted by the prediction filter sub-module passes through the secondary path is calculated as follows:

y _s (n)=S(z)*y(n);

In the formula, * is the convolution operation, S(z) is the secondary path between the error microphone and the speaker, here we set S(z) and The results are consistent.

In another embodiment, the error signal can also be expressed as:

e(n)=d(n)+ _ys (n);

In the formula, y _s (n) is the signal after the cancellation signal y (n) passes through the secondary path.

As a preference, σ takes a value of 2 in this example.

As an option, the weight update sub-module continuously repeats the update process to achieve effective control of impulse noise in the target scene.

In order to test the pulse noise reduction performance of the method proposed in the present invention, the following experiments are carried out:

Impulse noise usually obeys α steady-state distribution. According to the α steady-state distribution characteristics, random signals are used to generate target impulse noise.

Example

As shown in Figures 2 and 3, in order to fully test the effectiveness of the present invention, the method proposed by the present invention is used to perform active detection on high-intensity pulse noise signals (characteristic index is 1.4) and low-intensity pulse noise signals (characteristic index is 1.8) respectively. Noise reduction. At the same time, the Th-FxLMS algorithm, FxLMP algorithm and FxlogLMS algorithm among the classic impulse noise control algorithms were selected for noise reduction comparison.

The primary path P(z) and the secondary path S(z) used in this embodiment are both represented by FIR filters with an order of 300, and the amplitude-frequency response curves and phase-frequency response curves of the primary path and the secondary path are As shown in Figure 4 and Figure 5 respectively.

As shown in Figure 6 and Figure 7, the experimental results of this embodiment are shown. Among them, Figure 6 shows the noise reduction effect of the four algorithms for high-intensity pulse noise signals (α = 1.4), and Figure 7 shows the noise reduction effects of the four algorithms for low-intensity pulse noise signals (α = 1.8). Compare Figure 6, It can be seen from Figure 7 that compared with the classic impulse noise control algorithm, the noise reduction method proposed by the present invention has better noise reduction amount and faster convergence speed.

Although the embodiments of the present invention have been disclosed above, they are not limited to the applications listed in the description and embodiments. They can be applied to various fields suitable for the present invention. For those familiar with the art, they can easily Additional modifications may be made, and the invention is therefore not limited to the specific details and illustrations shown and described herein without departing from the general concept defined by the claims and equivalent scope.

Claims (3)

1. An active noise reduction method suitable for active control of impulse noise, characterized by including the following steps: Step 1: Use a reference microphone to collect the pulse noise emitted by the noise source as a reference signal, and adjust the step size parameter using the Gaussian distribution function of the reference signal as an independent variable; Step 2: Use an error microphone to collect the canceled noise as an error signal, and update the prediction filter weight coefficient based on the continuously updated reference signal, the error signal and the adjusted step parameter; Step 3: Calculate the cancellation signal through the updated prediction filter weight coefficient, and send the cancellation signal through the speaker to cancel the reference signal, and then perform active noise reduction; In the step one, the adjustment formula of the step parameter is: In the formula, n is the time series, x(n) is the reference signal, μ[x(n)] is the step parameter with the reference signal as the independent variable, is the basic step parameter, σ is the standard deviation of the distribution function; In the second step, the weight coefficient of the prediction filter is updated through the weight adaptive iterative formula; The weight adaptive iteration formula is expressed as: In the formula, w(n) is the weight coefficient of the prediction filter at time n, w(n+1) is the weight coefficient of the prediction filter at time n+1, J(n) is the cost function of the adaptive iteration formula, is the gradient of J(n); The cost function of the adaptive iteration formula is expressed as: J(n)＝E{f ² [e(n)]}≈f ² [e(n)]; In the formula, e(n) is the error signal, f[e(n)] is the nonlinear transformation function of the error signal; The nonlinear transformation function of the error signal is expressed as: 2. The active noise reduction method suitable for active control of impulse noise according to claim 1, characterized in that the cost function gradient is expressed as: In the formula, x _f (n) is the reference signal filtered by the estimated secondary path. 3. The active noise reduction method suitable for active control of impulse noise according to claim 1 or 2, characterized in that the cancellation signal obtained based on the weight coefficient of the prediction filter is expressed as: y(n)=w ^T (n)·x(n); In the formula, y(n) is the cancellation signal at time n, and T is the matrix transpose.