Improved FxLMS noise reduction algorithm based on wavelet transform

Abstract

The invention discloses an improved FxLMS noise reduction algorithm based on wavelet transform, and the algorithm comprises the following steps: S1, collecting an automobile noise signal, and carrying out the preprocessing; s2, performing multi-scale wavelet forward transformation decomposition and thresholding processing on the preprocessed noise signal; s3, performing FxLMS main and auxiliary channel adaptive signal filtering on the processed signal; s4, performing n-stage reconstruction wavelet inverse transformation on the output filtering signal, and further reducing noise; and S5, outputting the signal after noise reduction. According to the method, the multi-resolution analysis characteristic of wavelet transform and the self-adaptive filtering advantage of the FxLMS algorithm are combined, and the noise can be effectively removed while the main characteristics of the signal are reserved.

Description

Technical Field

[0001] The present invention relates to the technical field of vehicle noise reduction, and particularly to an improved FxLMS noise reduction algorithm based on wavelet transform. Background Art

[0002] Automobile noise is one of the main pollution sources generated during vehicle operation. It not only affects the comfort experience of passengers but also may have adverse effects on the surrounding environment and human health. The sources of automobile noise are complex and diverse, including engine noise, tire noise, wind noise, etc. Moreover, the frequency distributions and intensities of various noises vary greatly, making it technically difficult to effectively suppress the noise. Currently, traditional automobile noise reduction technologies mainly rely on sound insulation materials and structural optimization, but these methods often have problems such as high costs and limited effects.

[0003] In view of the deficiencies of the existing technologies, in recent years, noise reduction methods based on signal processing have gradually received attention. Especially the wavelet transform technology, due to its unique time-frequency localization characteristics, can decompose signals into different frequency and time components, providing a new solution idea for the accurate identification and suppression of automobile noise. Automobile noise has time-varying characteristics. The ability of single wavelet processing to separate complex coupled noises is limited. Especially when the frequency bands of the useful signal and noise overlap, threshold processing will weaken the high-frequency components without discrimination, resulting in the loss of useful features. Summary of the Invention

[0004] The purpose of this part is to outline some aspects of the embodiments of the present invention and briefly introduce some preferred embodiments. Simplifications or omissions may be made in this part, as well as in the abstract and title of the present application, to avoid obscuring the purpose of this part, the abstract, and the title, and such simplifications or omissions shall not be used to limit the scope of the present invention.

[0005] In view of the above and / or problems existing in the existing noise reduction technologies, the present invention is proposed.

[0006] Therefore, the purpose of the present invention is to provide an improved FxLMS noise reduction algorithm based on wavelet transform, which combines the multi-resolution analysis characteristics of wavelet transform and the adaptive filtering advantages of the FxLMS algorithm, and can effectively remove noise while retaining the main features of the signal.

[0007] To solve the above technical problems, the present invention provides the following technical solution: An improved FxLMS noise reduction algorithm based on wavelet transform, comprising the following steps:

[0008] S1. Collect the automobile noise signal and perform preprocessing;

[0009] S2. Perform multi-scale wavelet forward transform decomposition and thresholding processing on the preprocessed noise signal;

[0010] S3. Perform adaptive signal filtering on the processed signal in the primary and secondary channels of FxLMS;

[0011] S4. Perform an n-level inverse reconstruction wavelet transform on the output filtered signal to further reduce noise;

[0012] S5. Output the signal after noise reduction.

[0013] As a preferred solution of the FxLMS noise reduction algorithm improved based on wavelet transform in the present invention, wherein: in step S1, the specific steps are as follows.

[0014] S101a. Obtain a batch of vehicle noise signals during vehicle driving through a high-sensitivity microphone array, denoted as the original noise data set;

[0015] S101b. Perform data preprocessing on the noise signals in the original noise data set and convert them into digital signals x(t);

[0016] S101c. Use the interpolation method to fill in missing values, eliminate outliers, define a threshold z, z = (x - μ) / σ, where μ is the mean of the original noise data set and σ is the standard deviation of the original noise data set. If the absolute value of z is greater than 3, determine the corresponding data point as an outlier and eliminate it to form a cleaned data set.

[0017] As a preferred solution of the FxLMS noise reduction algorithm improved based on wavelet transform in the present invention, wherein: in step S2, the specific steps are as follows.

[0018] S201a. Perform wavelet decomposition on the preprocessed noise signal to decompose it into sub-signal components of different frequencies;

[0019] S201b. For the signal x(t), the continuous wavelet transform is W x (a, b) is the continuous wavelet transform function of the x(t) signal, ψ * is the complex conjugate of the mother wavelet function, t is the time scalar, b is the translation factor to control the time position of the wavelet, and a is the scale factor used to control the stretching of the wavelet;

[0020] S201c. The discrete wavelet transform realizes the time-frequency distribution of the signal through binary decomposition. ψ j，k [n] is the discrete mother wavelet function. W x (j, k) is the discrete wavelet transform of the signal x[n], n is the discrete time index, j is the scale level, and k is the time translation index;

[0021] S201d. In the discrete wavelet decomposition, the signal is decomposed into approximation coefficients cAj and the detail coefficient cD j , update the data set x(t), x(t) = ∑ j cA j + ∑jcD j ;

[0022] S201e. Threshold the coefficients obtained by wavelet decomposition, λ is the set threshold;

[0023] S201f. Repeat the above steps m times to obtain multiple groups of signals x(t) = {cA1, cD1, cD2,..., cD m}], where cA1 is the low-frequency approximation signal of the first layer, and cD1 to cD n are the high-frequency detail signals of different layers.

[0024] As a preferred solution of the improved FxLMS noise reduction algorithm based on wavelet transform in the present invention, wherein: in step S3, the filtering step is specifically as follows,

[0025] S301. Taking the high-frequency detail signal of the nth layer as an example, at this time x(t) = cD n , obtain the desired signal d(t) = x(t) * P(z) through the main channel path P(z), and obtain the estimated value of x(t) for x(t) through the secondary channel estimation path S(z)

[0026] S302. The adaptive filter outputs the signal y(t), and through the secondary channel path, obtain the final output signal Y(t) = S(z) * y(t);

[0027] S303. Calculate the error signal e(t), e(t) = d(t) - Y(t), and process it in the weight coefficient updater W(z) and e(t) to obtain the output signal of the adaptive filter

[0028] S304. Further calculate the error signal e(t), give the cost function J(t), E is the expected value, and obtain the descending gradient From the weight coefficient update formula W(t + 1) is the weight coefficient at time t + 1, W(t) is the weight coefficient at time t, and μ is the independent step size parameter; after multiple iterations, output the final filtered signal.

[0029] As a preferred solution of the improved FxLMS noise reduction algorithm based on wavelet transform in the present invention, wherein: in step S4, the steps for performing the n-level inverse wavelet reconstruction are specifically as follows,

[0030] S401. Reconstruct starting from the nth layer. Combine the low-frequency approximation signal cA n and the high-frequency detail signal cD n of the nth layer to obtain the reconstructed signal X n (t) = h n * cA n + g n * cD n , where h n is the wavelet low-pass filter of the nth layer, g n is the wavelet high-pass filter of the nth layer, and * represents the convolution operation;

[0031] S402. Reconstruct starting from the (n - 1)th layer. Use the reconstructed signal x n (t) of the nth layer as the input of the (n - 1)th layer, and perform weighted synthesis to obtain the reconstructed signal X n-1 (t),

[0032] X n-1 (t) = h n-1 * cA n-1 + g n-1 * cD n-1 , where X n-1 (t) is the reconstructed signal of the (n - 1)th layer. Keep reconstructing until the first layer finally recovers to the complete reconstructed signal X(t). Here, h n-1 is the wavelet low-pass filter of the (n - 1)th layer, and g n-1 is the wavelet high-pass filter of the (n - 1)th layer.

[0033] As a preferred solution of the improved FxLMS noise reduction algorithm based on wavelet transform in the present invention, in step S5, the output noise reduction signal is

[0034] Compared with the prior art, the present invention has the following technical effects: By combining the multi-resolution analysis characteristics of wavelet transform and the adaptive filtering advantage of the FxLMS algorithm, the present invention can effectively remove noise while retaining the main features of the signal. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] To more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings. Among them:

[0036] Figure 1 is the flow diagram of the present invention.

[0037] Figure 2 is the schematic diagram of the specific principle of the present invention.

[0038] Figure 3 (a) Comparison graph of the offset output and the original input Figure 3 (b) Noise reduction effect graph using the present invention

[0039] Figure 4 (a) Comparison graph of the convergence speed between the traditional FxLMS noise reduction algorithm and the noise reduction using the present invention Figure 4 (b) For the traditional FxLMS noise reduction algorithm and the noise reduction using the present invention Detailed implementation manners

[0040] To make the above objects, features and advantages of the present invention more obvious and understandable, the following detailed description of the specific implementation manners of the present invention will be given in conjunction with the accompanying drawings of the specification.

[0041] In the following description, many specific details are set forth in order to fully understand the present invention. However, the present invention can also be implemented in other ways different from those described herein. Those skilled in the art can make similar generalizations without departing from the connotation of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed below.

[0042] Secondly, the so-called "one embodiment" or "embodiment" herein refers to a specific feature, structure or characteristic that can be included in at least one implementation manner of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it an individual or alternative embodiment that is mutually exclusive with other embodiments.

[0043] Embodiment 1

[0044] As Figure 1 shown

[0045] An improved FxLMS noise reduction algorithm based on wavelet transform includes the following steps:

[0046] S1. Collect the vehicle noise signal and perform preprocessing. The specific steps are as follows

[0047] S101a. Through a high-sensitivity microphone array, obtain a batch of vehicle noise signals during the driving of the vehicle, denoted as the original noise data set;

[0048] S101b. Perform data preprocessing on the noise signals in the original noise data set and convert them into digital signals x(t), which can be expressed as the superposition of the real noise signal s(t) and the interference signal n(t), x(t) = s(t) + n(t);

[0049] S101c, use interpolation method to fill missing values, remove outliers, define threshold z, z = (x-μ) / σ, μ is the mean of the original noise data set, σ is the standard deviation of the original noise data set, if the absolute value of z is greater than 3, the corresponding data point is judged as an outlier and removed to form a cleaned data set;

[0050] S2, performing wavelet decomposition on the pre-processed noise signal to decompose it into 4-6 sub-signal components with different frequencies;

[0051] S201a, for the signal x(t), the continuous wavelet transform is W x (a, b) is the continuous wavelet transform function of the x(t) signal, ψ * It is the complex conjugate of the mother wavelet function, t is the time scalar, b is the translation factor, which controls the time position of the wavelet, and a is the scale factor, which is used to control the expansion and contraction of the wavelet;

[0052] S201b, discrete wavelet transform realizes the time-frequency distribution of the signal through binary decomposition, ψ j，k [n] is the discrete mother wavelet function, W x (j,k) is the discrete wavelet transform of the signal x[n], n is the discrete time index, j is the scale level, and k is the time shift index;

[0053] S201c, in discrete wavelet decomposition, the signal is decomposed into approximation coefficients cA j (keep low frequency components) and detail coefficient cD j (keep high frequency components), update the data set x(t), x(t) = ∑ j cA j +∑ j cD j ;

[0054] S201d, thresholding the coefficients obtained by wavelet decomposition, λ is the set threshold;

[0055] S201e, the low-frequency signal retains the main features of the noise signal and is used to analyze the main causes of the low-frequency noise. The high-frequency signal is used to capture the complex details of the noise. For the low-frequency component, repeat the above steps n times to obtain multiple groups of signals x(t) = {cA1, cA2, ... cA n , cD1, cD2, …, cD n}, where cA1 is the low-frequency approximate signal of the first layer, cD1~cD n It is a high-frequency detail signal at different levels;

[0056] S3. Perform adaptive signal filtering on the processed x(t) signal through the FxLMS main and sub-channels. The specific steps are as follows:

[0057] S301. Taking the high-frequency detail signal of the nth layer as an example, at this time x(t) = cD n , and the desired signal d(t) = x(t) * P(z) is obtained through the main-channel path P(z). The estimated value of x(t) is obtained by passing x(t) through the sub-channel estimation path S(z).

[0058] S302. The adaptive filter outputs the signal y(t), and through the sub-channel path, the final output signal Y(t) = S(z) * y(t) is obtained;

[0059] S303. Calculate the error signal e(t), e(t) = d(t) - Y(t). Due to iteration, and e(t) are processed in the weight coefficient updater W(z) to obtain the output signal of the adaptive filter

[0060] S304. Further calculate the error signal e(t), give the cost function J(t), E is the expected value, and the descending gradient is obtained From the weight coefficient update formula Simplify to obtain the final form So that the filter of each layer can dynamically adjust the parameters according to the noise characteristics of this layer; W(t + 1) is the weight coefficient at time t + 1, W(t) is the weight coefficient at time t, and μ is the independent step size parameter; after multiple iterations, the final filtered signal is output;

[0061] S4. Perform an n-level inverse wavelet transform on the output filtered signal. The specific steps are as follows:

[0062] S401. Start the reconstruction from the nth layer. Combine the low-frequency approximation signal cA n and the high-frequency detail signal cD n of the nth layer to obtain the reconstructed signal X n (t) = h n * cA n + g n * cD n , h n is the nth layer wavelet low-pass filter, g n is the nth layer wavelet high-pass filter, and * represents the convolution operation;

[0063] S402. Start the reconstruction from the (n - 1)th layer. Take the reconstructed signal x n (t) of the nth layer as the input of the (n - 1)th layer and perform weighted synthesis to obtain the reconstructed signal Xn-1 (t), X n-1 (t) = h n-1 *cA n-1 +g n-1 *cD n-1 , X n-1 (t) is the reconstructed signal of the (n - 1)-th layer, until the first layer is finally restored to the complete reconstructed signal X(t), h n-1 is the wavelet low-pass filter of the (n - 1)-th layer, g n-1 is the wavelet high-pass filter of the (n - 1)-th layer;

[0064] S403. Recursively repeat the above process until the first layer is reconstructed, and finally obtain the original layer reconstructed signal X0(t) = h0 * x1(t) + g0 * cD0;

[0065] S404. Summarize the n-level inverse wavelet reconstruction as a recursive formula X j+1 (t) is the reconstructed signal of the previous layer, until the first layer is finally restored to the complete reconstructed signal X(t);

[0066] S5. Output the denoised signal

[0067] This application effectively realizes noise suppression at different frequency levels through the combination of multi-resolution analysis of wavelet transform and the FxLMS adaptive filtering algorithm. It can retain the key features of the signal while denoising, improving the signal quality. By using the adaptive nature of the FxLMS algorithm, this method can dynamically adjust the filter coefficients, providing efficient noise reduction processing for complex noise environments, significantly improving the signal-to-noise ratio and reducing the mean square error. The inverse wavelet transform ensures the accuracy and integrity of the denoised signal by reconstructing the signal layer by layer, with high recovery efficiency and applicability.

[0068] Embodiment 2

[0069] Refer to Figure 3 and Figure 4 , this embodiment uses scientific experiments to verify the noise reduction performance of the present invention.

[0070] Simulate the in-vehicle noise signal in the MATLAB environment, and conduct a simulation comparison experiment between the present invention and the traditional FxLMS algorithm. The parameter values involved in the simulation experiment are: μ = 0.1 in the weight coefficient update formula of the FxLMS noise reduction algorithm improved by wavelet transform, and μ = 0.1 in the weight coefficient update formula of the traditional FxLMS algorithm.

[0071] Use the present invention to denoise the simulated in-vehicle noise signal as the sampled noise signal, and visualize the noise before and after denoising, as shown in Figure 3As shown. It can be seen from the figure that when the noise reduction is about 0.25 s by the improved FxLMS algorithm, there is an obvious noise reduction effect. After 1 s, the signal after noise reduction tends to be stable, indicating that the present invention has a good noise reduction effect.

[0072] The present invention and the traditional FxLMS algorithm (the traditional FxLMS algorithm is a classic adaptive noise control method. Its principle is to collect the noise signal x(n) through a reference sensor, generate an anti-sound wave through an adaptive filter, and the residual noise is detected by an error sensor after being superimposed with the primary noise. The filter weights are updated according to the weight coefficient update formula. The traditional FxLMS algorithm is prior art and will not be elaborated in detail) are used for simulation comparison experiments. The same section of in-vehicle noise signal is input, and the comparison is made from two aspects: the convergence speed and the final noise reduction effect. The performance comparison diagram is as follows. From Figure 4 it can be concluded that the FxLMS algorithm improved based on wavelet transform has a great improvement in both the convergence speed and the noise reduction performance compared with the traditional FxLMS algorithm.

[0073] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and they should all be covered within the scope of the claims of the present invention.

Claims

1. An improved FxLMS noise reduction algorithm based on wavelet transform, characterized in that: The steps include: S1, collect automobile noise signals and perform preprocessing; S2, performing multi-scale wavelet forward transform decomposition and thresholding processing on the pre-processed noise signal; S3, performing FxLMS main and sub-channel adaptive signal filtering on the processed signal; S4, performing n-level reconstruction wavelet inverse transform on the output filtered signal to further reduce noise; S5, output the signal after noise reduction.

2. The improved FxLMS noise reduction algorithm based on wavelet transform as claimed in claim 1, characterized in that: In step S1, the specific steps are: S101a, using a high-sensitivity microphone array, obtaining a batch of automobile noise signals during the driving of the automobile, and recording them as an original noise data set; S101b, performing data preprocessing on the noise signal in the original noise data set and converting it into a digital signal x(t); S101c, use interpolation method to fill missing values, remove outliers, define threshold z, z = (x-μ) / σ, μ is the mean of the original noise data set, σ is the standard deviation of the original noise data set, if the absolute value of z is greater than 3, the corresponding data point is judged as an outlier and removed to form a cleaned data set.

3. The improved FxLMS noise reduction algorithm based on wavelet transform as claimed in claim 1, characterized in that: In step S2, the specific steps are: S201a, performing wavelet decomposition on the preprocessed noise signal to decompose it into sub-signal components of different frequencies; S201b, for the signal x(t), the continuous wavelet transform is W x (a, b) is the continuous wavelet transform function of the x(t) signal, ψ * It is the complex conjugate of the mother wavelet function, t is the time scalar, b is the translation factor, which controls the time position of the wavelet, and a is the scale factor, which is used to control the expansion and contraction of the wavelet; S201c, discrete wavelet transform realizes the time-frequency distribution of the signal through binary decomposition, ψ j，k [n] is the discrete mother wavelet function, W x (j,k) is the discrete wavelet transform of the signal x[n], n is the discrete time index, j is the scale level, and k is the time shift index; S201d, in discrete wavelet decomposition, the signal is decomposed into approximate coefficients cA j and detail coefficient cD j , update the data set x(t), x(t) = Σ j cA j +Σ j cD j ; S201e, thresholding the coefficients obtained by wavelet decomposition, λ is the set threshold; S201f, repeat the above steps m times to obtain multiple groups of signals x(t)={cA1, cD1, cD2, ..., cD m }, where cA1 is the low-frequency approximate signal of the first layer, cD1~cD n It is a high-frequency detail signal at different levels.

4. The improved FxLMS noise reduction algorithm based on wavelet transform as claimed in claim 3 is characterized in that: In step S3, the filtering step is specifically as follows: S301, taking the nth layer high frequency detail signal as an example, at this time x(t) = cD n , the expected signal d(t) = x(t)*P(z) is obtained through the main channel path P(z), and the estimated value of x(t) is obtained through the secondary channel estimation path S(z) S302, the adaptive filter output signal y(t) passes through the secondary channel path to obtain the final output signal Y(t)=S(z)*y(t); S303, calculate the error signal e(t), e(t) = d(t) - Y(t), and process it in the weight coefficient updater W(z) and e(t), the output signal of the adaptive filter is obtained S304, further calculating the error signal e(t) and giving the cost function J(t), E is the expected value, and the descent gradient is obtained Update formula by weight coefficient W(t+1) is the weight coefficient at time t+1, W(t) is the weight coefficient at time t, and μ is the independent step size parameter; After multiple iterations, the final filtered signal is output.

5. The improved FxLMS noise reduction algorithm based on wavelet transform according to any one of claims 1 to 4, characterized in that: In step S4, the steps of performing n-level reconstruction inverse wavelet transform are specifically as follows: S401, reconstructing from the nth layer, converting the low-frequency approximate signal cA of the nth layer n and high frequency detail signal cD n Combine and get the reconstructed signal X n (t) = h n *cA n +g n *cD n ,h n is the nth layer of wavelet low-pass filter, g n is the nth layer of wavelet high-pass filter, * indicates convolution operation; S402, start reconstruction from the n-1th layer, and convert the reconstructed signal x of the nth layer n (t) is used as the input of the n-1th layer, and weighted synthesis is performed to obtain the reconstructed signal X n-1 (t), X n-1 (t) = h n-1 *cA n-1 +g n-1 *cD n-1 , X n-1 (t) is the reconstructed signal of the n-1th layer, until the first layer is finally restored to the complete reconstructed signal X(t), h n-1 is the n-1th layer of wavelet low-pass filter, g n-1 It is the n-1th layer of wavelet high-pass filter.

6. The improved FxLMS noise reduction algorithm based on wavelet transform as claimed in claim 5, characterized in that: In step S5, the output noise reduction signal is

Images