Robust online secondary path modeling active noise control method

Abstract

The invention provides a robust online secondary path modeling active noise control method. The method comprises the following steps: S1, carrying out primary noise signal acquisition by using a microphone; s2, weight vector updating is carried out on the primary noise, and an error signal is obtained through calculation of the updated compensation signal and the main noise; s3, updating an adaptive filtering weight in secondary path modeling by using the error signal; and S4, enabling i to be equal to i + 1, and repeating the steps S1 to S3 until the noise is controlled. Compared with other robust cost functions, the cost function provided by the invention has better tolerance and smoothness, so that the cost function can show excellent performance in the aspect of suppressing pulse interference, and in addition, in consideration that the performance of the algorithm may be reduced when the secondary path changes, adaptive secondary path modeling is adopted to supplement the algorithm.

Description

Technical Field

[0001] The present invention relates to the technical field of active noise control, and in particular to a robust online secondary path modeling active noise control method. Background Art

[0002] In the past two decades, the research and development of active noise control (ANC) has been a topic of interest. Compared with traditional passive noise control technology, ANC has shown significant advantages in low-frequency noise processing. In the delivery method, ANC usually consists of four parts: a speaker for generating noise sound waves and an anti-noise speaker for reducing noise, an adaptive filter for iteratively updating the algorithm, and an error signal for updating the adaptive filter.

[0003] The main challenges facing ANC are that the performance of the algorithm may degrade in high-intensity pulsed background noise environments and when the secondary path changes, the tracking ability of the algorithm decreases accordingly.

[0004] In order to solve this technical problem, some improved methods have been proposed successively. The more mature methods are as follows:

[0005] (1) Active noise cancellation method using logarithmic function transformation (FxlogLMS)

[0006] Reference 1: An active impulsive noise control algorithm with logarithmic transformation. This method eliminates the need for robust ANC algorithm threshold estimation by minimizing the square logarithmic transformation of the error signal, reducing the computational complexity of the algorithm. However, when the error signal |e(i)<1|, the FxlogLMS algorithm has a blind spot.

[0007] (2) Active noise cancellation method using hyperbolic tangent function transformation (FxatanLMS)

[0008] Reference 2: Active Impulsive Noise Control Algorithm Based on Adjustable Hyperbolic Tangent Function. The main idea of ​​this algorithm is to use the adjustable hyperbolic tangent function in the cost function to compress the error signal. Adaptive adjustment of the compression function can achieve the ideal processing effect for impulse noise of different intensities. However, the tracking ability of this algorithm will decrease when the secondary path changes in time. Summary of the invention

[0009] In view of the technical problems of poor tracking ability and algorithm blind spots in the existing technology, the present invention proposes a robust online secondary path modeling active noise control method, which has good effect on eliminating speech noise in the communication system, low computational complexity, fast convergence speed and low steady-state error.

[0010] In order to achieve the above object, the technical solution of the present invention is implemented as follows: a robust online secondary path modeling active noise control method, the steps comprising:

[0011] S1: Use microphone to collect primary noise signal;

[0012] S2: Update the weight vector of the primary noise signal and calculate the updated compensation signal and the main noise.

[0013] to the error signal;

[0014] S3: using the error signal to update the adaptive filter weights in the secondary path modeling;

[0015] S4: Let i=i+1, and repeat steps S1-S3 until the noise is controlled.

[0016] The method for collecting the primary noise signal in step S1 is:

[0017] The microphone is used to collect the primary noise signal x(i) at the current time i. After the primary noise passes through the main path P(z), the main noise d(i) to be eliminated is obtained. The formula of the main noise d(i) is:

[0018] d(i)=x(i)*p(i);

[0019] Among them, p(i) is the time domain representation of P(z).

[0020] The expression of the primary noise signal is:

[0021] x(i)=[x(i),x(i-1),…,x(i-L+1)]T ;

[0022] Where L is the number of taps of the filter, the superscript T represents the transpose, and x(i), x(i-1),…, x(i-L+1) represent the noise signals collected by the microphone.

[0023] The method of obtaining the error signal by calculating the updated compensation signal and the main noise in step S2 is:

[0024] The input signal x(i) is delayed-compensated through the secondary path to obtain the compensation signal r(i). The specific formula is:

[0025] r(i)=x(i)*s(i);

[0026] Where s(i) is the adaptive weight vector in secondary path modeling;

[0027] The compensation signal r(i) obtained by delay compensation is input into the adaptive filter, and the filtered value y(i) is obtained after passing through the adaptive filter:

[0028] y(i)=w T (i)r(i);

[0029] Wherein, w(i) is the weight coefficient of the adaptive filter;

[0030] Subtract the main noise d(i) to be eliminated from the filtered value y(i) output by the adaptive filter to obtain the error signal e(i) of the algorithm:

[0031] e(i)=d(i)-y(i).

[0032] The method of obtaining the error signal by calculating the updated compensation signal and the main noise in step S3 is:

[0033] The LMS algorithm is used to update the secondary path in real time and compensate for the delay noise during the acquisition process;

[0034] Define the random Gaussian white noise g at the current time i respectively c (i) and the adaptive weight vector s(i) of the secondary path; random Gaussian white noise g c (i) Output y through the secondary path s (i)

[0035] y s (i) = g c (i)s(i);

[0036] The error signal e(i) is compared with the output of the auxiliary noise path, that is, the output of the secondary path y s (i) Subtract and obtain the error signal e for regulating the secondary paths (i)

[0037] e s (i) = e(i) - y s (i);

[0038] Through the error signal e s (i) and the auxiliary noise vector g c (i) Realize real-time update of secondary paths.

[0039] The method for implementing real-time update of the secondary path is:

[0040] The update formula is as follows:

[0041] s(i+1)=s(i)+μ s e s (i)g c (i);

[0042] Where s(i) is the adaptive filter weight vector at time i in the secondary path modeling module, μ s is the step size parameter used to update the secondary path modeling.

[0043] The μ s The value range is 0.0001~0.1.

[0044] The method for updating the adaptive filtering weights in the main path modeling described in step S3 is:

[0045] By designing a robust cost function, the cost function is defined as follows:

[0046]

[0047] Among them, the parameter c is the degree of transformation that affects the residual, and erf is the Gaussian error function.

[0048] By calculating the partial derivative of the above formula, we can get:

[0049]

[0050] in,

[0051] According to the gradient descent method, the adaptive filter weight update equation in the main path modeling can be described as follows:

[0052]

[0053] Where μ is the step size of the noise canceller.

[0054] The value range of μ is 0.001 to 0.1.

[0055] The method of repeating steps S1-S3 until the noise is controlled in step S4 is:

[0056] Through the active noise control system, steps S1-S3 are repeated until a noise with the same amplitude and opposite phase as the noise is generated, thereby achieving noise cancellation.

[0057] The beneficial effects of the present invention are as follows: unlike the traditional method, the cost function used in the method proposed in the present application is a robust nonlinear cost function with a logarithmic function structure. Compared with other robust cost functions, the cost function proposed in the present application has better tolerance and smoothness, so it can show superior performance in suppressing pulse interference. The method has good effect on eliminating speech noise in the communication system, low computational complexity, fast convergence speed, and low steady-state error. In addition, considering that the performance of the algorithm may decline when the secondary path changes, adaptive secondary path modeling is used to supplement it. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0059] Figure 1 It is a work flow chart of the present invention;

[0060] Figure 2 It is the average noise reduction level curve of Document 1, Document 2 and the simulation experiment of the present invention under Gaussian white noise.

[0061] Figure 3 It is the average noise reduction level curve of Document 1, Document 2 and the simulation experiment of the present invention under real speech noise.

[0062] Figure 4 It is a spectrum comparison curve of Document 1, Document 2 and the present invention. DETAILED DESCRIPTION

[0063] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0064] like Figure 1As shown, a robust online secondary path modeling active noise control method comprises the following steps:

[0065] S1: Use a microphone to collect primary noise signals.

[0066] The microphone is used to collect the primary noise signal x(i) at the current time i, and the primary noise signal x(i) = [x(i), x(i-1), ..., x(i-L+1)] T ; Wherein, L=256 is the number of taps of the filter, the superscript T represents the transpose, and x(i), x(i-1), ..., x(i-L+1) represent the noise signals collected by the microphone.

[0067] After the primary noise passes through the main path P(z), the main noise d(i) that needs to be eliminated is obtained. The formula of the main noise d(i) is:

[0068] d(i)=x(i)*p(i);

[0069] Among them, p(i) is the time domain representation of P(z).

[0070] S2: Update the weight vector of the primary noise, and obtain the error signal through the updated compensation signal and the main noise calculation.

[0071] Considering that the sound waves generated by the anti-noise speaker are not directly received by the error sensor, but are transmitted through the secondary path, the input signal x(i) is delayed compensated through the secondary path to obtain the compensation signal r(i). The specific formula is:

[0072] r(i)=x(i)*s(i);

[0073] Among them, s(i) is the adaptive weight vector in secondary path modeling.

[0074] The compensation signal r(i) obtained by delay compensation is input into the adaptive filter, and the filtered value y(i) is obtained after passing through the adaptive filter:

[0075] y(i)=w T (i)r(i);

[0076] Among them, w(i) is the weight coefficient of the adaptive filter, and w(i) represents the weight vector of i at the current moment.

[0077] Subtract the main noise d(i) to be eliminated from the filtered value y(i) output by the adaptive filter to obtain the error signal e(i) of the algorithm:

[0078] e(i)=d(i)-y(i).

[0079] S3: Using the error signal, the adaptive filter weights in the secondary path modeling are updated, and the adaptive filter weights in the primary path modeling are updated.

[0080] Update the secondary path:

[0081] The LMS algorithm (least mean squares) is used to update the secondary path in real time to compensate for the delay of noise during the acquisition process.

[0082] Define the random Gaussian white noise g at the current time i respectively c (i) and the adaptive weight vector s(i) of the secondary path; random Gaussian white noise g c (i) Output y through the secondary path s (i)

[0083] y s (i) = g c (i)s(i);

[0084] The error signal e(i) is then compared with the output of the auxiliary noise path, i.e., the output of the secondary path y s (i) Subtraction can obtain the error signal e for adjusting the secondary path s (i)

[0085] e s (i) = e(i) - y s (i);

[0086] Through the error signal e s (i) and the auxiliary noise vector g c (i) The real-time update of the secondary path can be achieved, and the update formula is as follows:

[0087] s(i+1)=s(i)+μ s e s (i)g c (i).

[0088] Where s(i) is the adaptive filter weight vector at time i in the secondary path modeling module, μ s It is the step size parameter used to update the secondary path modeling, and its value range is 0.0001 to 0.1.

[0089] The method for updating the adaptive weights in main path modeling is:

[0090] Considering that high-intensity impulse noise interference often occurs in actual scenarios, in order to ensure the denoising performance of the algorithm, a robust cost function is designed and defined as follows:

[0091]

[0092] Among them, the parameter c is the degree of transformation that affects the residual, and erf is the Gaussian error function.

[0093] By calculating the partial derivative of the above formula, we can get:

[0094]

[0095] In order to simplify the representation, we define

[0096] According to the gradient descent method, the adaptive filter weight update equation in the main path modeling can be described as follows:

[0097]

[0098] Wherein, μ is the step size of the noise eliminator, and its value range is 0.001 to 0.1.

[0099] S4: Let i=i+1, repeat steps S1-S3, and generate a noise with the same amplitude and opposite phase as the noise to cancel the noise.

[0100] Through the active noise control system, a noise with the same amplitude and opposite phase as the noise is generated to cancel the noise. The final result will approach a very small constant value (that is, the background noise, which cannot be eliminated). The standard here is to achieve the cancellation of active noise.

[0101] Simulation experiment:

[0102] In order to verify the effectiveness of the present invention, simulation experiments were carried out and compared with the existing algorithms in references 1 and 2. The comparison results are shown in Table 1.

[0103] Table 1 Comparison of simulation results

[0104] FxlogLMS α=1,μ=0.0001 FxatanLMS β=1,μ=0.0001 The present invention <![CDATA[μ=0.0005,k=1,μ s =0.01]]>

[0105] The input signal of the simulation experiment is a real noise signal, the number of filter taps L = 256, and the background noise is a Gaussian pulse mixed noise. Gaussian noise is a Gaussian white noise with a signal-to-noise ratio of 30dB. The impulse noise α(i) is usually modeled as a Bernoulli Gaussian process, that is: α(i) = β(i)ζ(i), and its probability density function is described as γ(β(i) = 1) = γ, γ(β(i) = 0) = γ, ζ(i) is a Gaussian function with a mean and variance of 0 The experiment uses an impulse noise with an amplitude of ρ = 500. The corresponding simulation results are obtained by using MATLAB 2019 on a computer equipped with a Core i7 processor and 32GB memory. Figure 2-4It can be seen that under pulse background noise and real noise input, the present invention has a faster convergence speed and a better average noise reduction level than references 1 and 2.

[0106] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A robust online secondary path modeling active noise control method, characterized in that the steps include: S1: Use microphone to collect primary noise signal; S2: Update the weight vector of the primary noise signal, and calculate the error signal through the updated compensation signal and the main noise; S3: using the error signal to update the adaptive filter weights in the secondary path modeling; S4: Repeat steps S1-S3 until the noise is controlled.

2. The robust online secondary path modeling active noise control method according to claim 1, characterized in that: The method for collecting the primary noise signal in step S1 is: The microphone is used to collect the primary noise signal x(i) at the current time i. After the primary noise passes through the main path P(z), the main noise d(i) to be eliminated is obtained. The formula of the main noise d(i) is: d(i)=x(i)*p(i); Among them, p(i) is the time domain representation of P(z).

3. The robust online secondary path modeling active noise control method according to claim 2, characterized in that: The expression of the primary noise signal is: x(i)=[x(i),x(i-1),…,x(i-L+1)] T ; Where L is the number of taps of the filter, the superscript T represents the transpose, and x(i), x(i-1), ..., x(i-L+1) represent the noise signals collected by the microphone.

4. The robust online secondary path modeling active noise control method according to claim 3, characterized in that: The method of obtaining the error signal by calculating the updated compensation signal and the main noise in step S2 is: The input signal x(i) is delayed-compensated through the secondary path to obtain the compensation signal r(i). The specific formula is: r(i)=x(i)*s(i); Where s(i) is the adaptive weight vector in secondary path modeling; The compensation signal r(i) obtained by delay compensation is input into the adaptive filter, and the filtered value y(i) is obtained after passing through the adaptive filter: y(i)=w T (i)r(i); Wherein, w(i) is the weight coefficient of the adaptive filter; Subtract the main noise d(i) to be eliminated from the filtered value y(i) output by the adaptive filter to obtain the error signal e(i) of the algorithm: e(i)=d(i)-y(i).

5. The robust online secondary path modeling active noise control method according to claim 3 or 4, characterized in that: The method of obtaining the error signal by calculating the updated compensation signal and the main noise in step S3 is: The LMS algorithm is used to update the secondary path in real time and compensate for the delay noise during the acquisition process; Define the random Gaussian white noise g at the current time i respectively c (i) and the adaptive weight vector s(i) of the secondary path; random Gaussian white noise g c (i) Output y through the secondary path s (i) y s (i)=g c (i)s(i); The error signal e(i) is compared with the output of the auxiliary noise path, that is, the output of the secondary path y s (i) Subtract and obtain the error signal e for regulating the secondary path s (i) yes s (i)=e(i)-y s (i); Through the error signal e s (i) and the auxiliary noise vector g c (i) Realize real-time update of secondary paths.

6. The robust online secondary path modeling active noise control method according to claim 5, characterized in that: The method for implementing real-time update of the secondary path is: The update formula is as follows: s(i+1)=s(i)+μ s e s (i)g c (i); Where s(i) is the adaptive filter weight vector at time i in the secondary path modeling module, μ s is the step size parameter used to update the secondary path modeling.

7. The robust online secondary path modeling active noise control method according to claim 6, characterized in that: The μ s The value range is 0.0001~0.

1.

8. The robust online secondary path modeling active noise control method according to claim 7, characterized in that: The method for updating the adaptive filtering weights in the main path modeling described in step S3 is: By designing a robust cost function, the cost function is defined as follows: Among them, the parameter c is the degree of transformation that affects the residual, and erf is the Gaussian error function. By calculating the partial derivative of the above formula, we can get: in, According to the gradient descent method, the adaptive filter weight update equation in the main path modeling can be described as follows: Where μ is the step size of the noise canceller.

9. The robust online secondary path modeling active noise control method according to claim 8, characterized in that: The value range of μ is 0.001 to 0.

1.

10. The robust online secondary path modeling active noise control method according to any one of claims 7 to 9, It is characterized in that The method of repeating steps S1-S3 until the noise is controlled in step S4 is: Through the active noise control system, steps S1-S3 are repeated until a noise with the same amplitude and opposite phase as the noise is generated, thereby achieving noise cancellation.

Images