Robust active noise control method for parameter self-adjustment

Abstract

The invention discloses a parameter self-adjusting robust active noise control method, and relates to the technical field of noise control. The method comprises the following steps: acquiring a sound pressure signal in real time, processing the sound pressure signal to form a discrete noise signal as a reference signal, and constructing a dynamically updated reference signal sequence; constructing a filter and generating a weight coefficient thereof, and filtering the reference signal to generate an output signal; processing an output signal of the filter through a secondary path to generate a de-noising signal, collecting a residual signal, feeding back the residual signal to the filter, providing error feedback for self-adaptive iteration of the filter, and driving a weight coefficient of the filter to be optimized and updated; determining an optimal parameter combination for the generalized robust adaptive loss function to realize adaptive optimal determination; and updating a filter weight coefficient vector. According to the method, through organic combination of a self-adaptive parameter adjustment mechanism and a robust control strategy, the fast convergence characteristic, the low steady-state error and the high impact resistance are achieved, and the noise suppression efficiency in the complex environment is remarkably improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of noise control, and in particular relates to a robust active noise control method with self-adjusted parameters. Background Art

[0002] With the rapid development of modern industry and transportation, low-frequency noise pollution (such as power transformer hum, internal combustion engine vibration noise, and rail transit wheel-rail coupling noise) has become an environmental hazard that urgently needs to be addressed. According to statistics, approximately 60% of industrial noise in urban environments is concentrated in the 100-500Hz frequency range, typically manifested by transformer electromagnetic noise (125-250Hz), rail transit wheel-rail noise (200-400Hz), and power equipment vibration noise. This type of noise, characterized by concentrated energy, long propagation distance, and strong penetration, not only seriously interferes with human health and well-being but can also cause fatigue damage to industrial equipment. Traditional passive noise control (PNC) technologies use sound-absorbing and sound-insulating materials to attenuate noise. However, their effectiveness decreases exponentially with decreasing frequency. For frequencies below 500Hz, the material thickness must be increased to half the wavelength of the sound wave (approximately 0.34-1.7m). This significantly increases project implementation costs and has poor spatial adaptability, making it difficult to meet the needs of high-precision industrial and urban acoustic environment management.

[0003] Active noise control (ANC) technology is based on the principle of destructive interference of acoustic waves. It uses an adaptive algorithm to generate antiphase acoustic waves to achieve noise cancellation. It has shown unique advantages in the field of low-frequency noise control. Its core lies in the design of adaptive filtering algorithms. The filter-x least mean square (FxLMS) algorithm based on the minimization of the residual mean square criterion is widely used due to its low computational complexity and compact structure. However, its cost function J(n) = E[e 2 (n)] (where the symbol E[·] represents the mathematical expectation) has significant defects in the impact noise environment: when non-Gaussian noise samples such as gear meshing impact and aerodynamic commutation noise obey α-stable distribution (characteristic index α < 2), the heavy-tailed distribution of the impact noise leads to the second-order moment statistical information E[e 2 (n)] diverges, the algorithm is prone to fall into local optimality, weight coefficient oscillation and even system instability.

[0004] For impulsive noise environments, the FxMCC algorithm, based on the Maximum Correlation Entropy (MCC) criterion, effectively suppresses the interference of impulsive noise by maximizing the correlation entropy function of the residual signal, extending the noise statistics to the higher-order moment space. For example, in the paper "Kurian NC, Patel K, George NV. Robust active noise control: An information theoretic learning approach, Applied Acoustics, 2017, 117:180-184," this algorithm adapts to different noise distributions by adjusting kernel function parameters (such as the Gaussian kernel width σ). However, this method suffers from two major technical bottlenecks: First, the value of the kernel parameter directly affects the algorithm's matching accuracy to the noise distribution, and the kernel parameter relies on empirical trial and error or offline optimization, making it unsuitable for real-time noise environments. Second, when the noise amplitude exceeds the kernel function's range, the variance of the gradient estimate of the correlation entropy cost function increases significantly, resulting in a decrease in the algorithm's convergence speed and steady-state accuracy.

[0005] This paper aims to construct a parameter adjustment mechanism that is both adaptive and robust, enabling dynamic optimization of control performance in impulsive noise environments. This paper proposes a robust active noise control method with self-tuning parameters. By constructing an online noise signature estimator and an adaptive kernel parameter update law, this method overcomes the traditional algorithm's reliance on manual parameter adjustment, significantly improving the practicality of active noise control systems in complex noise environments. Summary of the Invention

[0006] The purpose of the present invention is to provide a robust active noise control method with self-adjusting parameters, so as to solve the problems of the existing methods proposed in the above background technology, such as fixed kernel parameters leading to limited algorithm robustness, manual experience-based parameter adjustment reducing engineering practicality, and lack of dynamic parameter matching mechanism in time-varying noise environment.

[0007] To achieve the above objectives, the present invention adopts the following technical solutions:

[0008] The present invention proposes a robust active noise control method with self-adjusted parameters, comprising the following steps:

[0009] Acquiring reference signals: Acquire sound pressure signals in real time, convert them into discrete noise signals at the current moment as reference signals through analog-to-digital conversion, and use a forward sliding time window mechanism to construct a dynamically updated reference signal sequence;

[0010] Constructing a filter: Determine the filter and generate its weight coefficients. The weight coefficient vector is updated online through an adaptive algorithm. The current reference signal is filtered to generate an output signal.

[0011] Collecting the residual signal: The filter output signal is processed through the secondary path to generate a noise-cancelled signal. During the secondary path processing, the eliminated residual noise signal is collected and discretized as the residual signal. The residual signal is fed back to the filter to provide error feedback for the filter's adaptive iteration and drive the optimization and update of the filter weight coefficients.

[0012] Design adaptive parameter selection update: During the filter weight coefficient update process, the generalized robust adaptive loss function is used to determine the optimal parameter combination in a statistical sense based on the objective function and the Hampel identifier to achieve adaptive optimal determination;

[0013] Update the filter weight coefficient vector: Calculate the gradient of the generalized robust adaptive loss function with respect to the filter weight coefficient vector based on the residual signal at the current moment; Use the stochastic gradient method to iteratively optimize the filter weight coefficient for noise reduction processing at the next moment.

[0014] Preferably, the construction of the dynamically updated reference signal sequence is specifically as follows:

[0015] Obtain the discrete noise signal x(k) at the current time k, and use the forward sliding time window mechanism to construct the reference signal sequence x(k). x(k) contains the historical noise data of L-1 sampling periods before the current time, which serves as the noise input signal vector of the filter. The expression is:

[0016] x(k)=[x(k),x(k-1),L,x(k-L+1)] T

[0017] Where L is the number of filter taps and T represents the transpose operation.

[0018] Preferably, the filter is determined and its weight coefficients are generated as follows:

[0019] The filter adopts a finite impulse response digital filter structure, and its weight coefficient generation mechanism is as follows: at the current time k, the filter generates a weight coefficient vector of equal length according to the dimension of the noise input signal vector. Its mathematical expression is:

[0020] w(k)=[w0(k),w1(k),L,w L-1 (k)] T

[0021] Among them, each element w of the weight coefficient vector i (k) corresponds to the real-time gain of the i-th tap of the filter, which constitutes the core parameter set of time-domain adaptive filtering;

[0022] At the current time k, the noise input signal vector x(k) is filtered using the weight coefficient vector w(k) at this time, and the filter output value y(k) = w at the current time is obtained. T (k)x(k).

[0023] Furthermore, the weight coefficient vector is updated online through an adaptive algorithm, as follows:

[0024] In the initialization stage, the weight coefficient vector adopts the zero vector initialization strategy, that is, w(0) = [0, 0, L, 0] T ,As the iterative process progresses, the weight coefficient vector is updated online through the ,adaptive algorithm.

[0025] Preferably, the generating of the noise cancellation signal is specifically as follows:

[0026] The filter output signal passes through a secondary path S(z) to obtain the noise-canceling signal; this secondary path includes a digital-to-analog converter, a reconstruction filter, a power amplifier, a noise-canceling speaker, a physical path between the noise-canceling speaker and the error microphone, the error microphone, a preamplifier, an anti-aliasing filter, and an analog-to-digital converter.

[0027] Preferably, the residual noise signal eliminated by the acquisition is specifically as follows:

[0028] The error microphone in the secondary path is used to collect the sound signal at the noise cancellation point. After discretization processing, the signal at the current moment k is obtained, which is the residual signal e(k) at the current moment; this residual signal is formed by the superposition of the noise cancellation signal and the original noise signal.

[0029] Preferably, the determination of the optimal parameter combination is as follows:

[0030] Adaptively select the free parameters γ and η to be selected in the generalized robust adaptive loss function J(k);

[0031] The generalized robust adaptive loss function J(k) is as follows:

[0032]

[0033] For the parameter γ, the objective function is constructed as follows:

[0034] h(e(k),γ,η)=-logP(e(k),γ,η)=log(ηA(γ))+J(k)

[0035] Where e(k) represents the residual signal at the current moment; the objective function is a shifted form of the logarithmic division of the loss function J(k); the adaptive selection of the parameter γ is achieved by statistically minimizing the expectation of h(e(k),γ,η);

[0036] The parameter η is used to define the interval in which the derivative of the loss function J(k) with respect to the filter weight coefficient vector w(k) has an approximately linear relationship with the error signal, and this interval assumes that the noise signal is not contaminated by impact noise. By introducing a Hampel identifier to process the outliers of the noise signal, and using the mean absolute value of the preprocessed signal to estimate η, the parameter η is adaptively selected based on the noise characteristics.

[0037] Preferably, the parameter η is adaptively selected based on noise characteristics, specifically as follows:

[0038] The Hampel identifier is introduced to process the outliers of the noise signal. The specific rules are as follows:

[0039]

[0040]

[0041] Where M(k) is the length L w The local median of the sliding data window, t is the threshold, σ sk (k) represents the estimated standard deviation, κ = 1.4826; by preprocessing the signal x H The mean absolute value of (k) estimates η.

[0042] Preferably, the gradient of the filter weight coefficient vector is calculated as follows:

[0043]

[0044] The gradient is solved to obtain the gradient information reflecting the characteristics of the residual signal, providing a direction basis for updating the weight coefficient vector.

[0045] Preferably, the iterative optimization filter weight coefficients are specifically as follows:

[0046] The filter updates the gradient based on the current weight coefficient and generates the weight coefficient vector at the next moment through the weight coefficient update formula to achieve dynamic optimization of the filter weight coefficient; the weight coefficient update formula is as follows:

[0047]

[0048] x′(k)=[x′(k),x′(k-1),L,x′(k-L+1)] T

[0049] x′(k)=x(k)*s(k)

[0050] Where μ is the iteration step size; s(k) represents the impulse response of the secondary path S(z).

[0051] Compared with the prior art, the present invention has the following beneficial effects:

[0052] (1) The method of the present invention designs an active noise control filtering algorithm based on a generalized robust cost function. First, the algorithm can quantify the robustness through a continuous parameter that is adaptively adjusted according to the residual distribution to improve the robustness to impact noise. Second, the method automatically determines the proportional parameter based on the noise signal preprocessed by the Hampel filter, realizing the self-adjustment capability of the parameter. Therefore, the present invention overcomes the problems of limited algorithm robustness due to fixed kernel parameters, reduced engineering practicality of manual experience parameter adjustment, and the lack of a dynamic parameter matching mechanism in a time-varying noise environment.

[0053] (2) In a strong impact noise environment, the residual signal is prone to violent fluctuations, causing the adaptive update process of the control filter based on the residual to oscillate, and even causing the system to become unstable. Therefore, strict requirements are placed on the stability of the control method. Existing technologies such as Reference 1 use a single correlation entropy criterion to construct a loss function. Its generalization ability is limited by the selection of fixed kernel parameters, and it is easy to cause the control filter update process to diverge under strong impact conditions. The present invention innovatively proposes a generalized robust loss function framework. Through the dynamic adjustment of kernel parameters, the framework can be equivalently degenerated into typical robust criteria such as Welsch, Huber, Cauchy and correlation entropy. This design gives the method stronger environmental adaptability and robustness, and is more flexible in application than the control strategy of a single loss function.

[0054] (3) Traditional methods have the defect of relying on experience to adjust parameters. The generalized robust criterion of the present invention does not need to rely on prior knowledge of noise. It can adaptively match the optimal parameters for the algorithm based on the real-time characteristics of the impact noise. This adaptive adaptation mechanism completely avoids the complex process of manual parameter adjustment and realizes dynamic optimization of control parameters. Regardless of the characteristics of the impact noise environment, the system can maintain good noise reduction performance through parameter self-adjustment, significantly improving the practicality and flexibility of the active noise control system in complex noise scenarios, ensuring that the noise reduction target can be efficiently achieved in different impact noise environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 Flowchart of the robust active noise control method with self-adjusted parameters in the present invention;

[0056] Figure 2 This is the time domain waveform of the impact noise simulated by the α stable distribution used in the simulation experiment of the present invention ( Figure 2 (a) is α=1.4, Figure 2 (b) is α=1.6, Figure 2 (c) is α = 1.8);

[0057] Figure 3This is the average noise residual curve after the impact noise in the simulation experiment of the present invention is processed by the simulation experiment of the document 1 and the method of the present invention ( Figure 3 (a) Figure 2 (a) The curve after impact noise processing; Figure 3 (b) Figure 2 (b) The curve after impact noise processing; Figure 3 (c) Figure 2 (c) Curve diagram after impact noise processing). DETAILED DESCRIPTION

[0058] The following will provide a clear and complete description of the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0059] Example 1:

[0060] like Figure 1 As shown in FIG, the robust active noise control method with self-adjusting parameters has the following steps:

[0061] Step 1: Collect reference signals.

[0062] A reference microphone placed near the noise source collects the sound pressure signal in real time, and after analog-to-digital conversion, the discrete noise signal x(k) at the current time k is obtained. The input signal vector is defined as the reference signal sequence within the sliding time window, and its mathematical expression is:

[0063] x(k)=[x(k),x(k-1),L,x(k-L+1)] T

[0064] Here, L is the number of filter taps, and the superscript T represents the transpose operation. This vector constructs an input feature space based on a time-domain sliding window, which contains historical noise data from L-1 sampling periods before the current moment, providing time-related input features for the adaptive algorithm.

[0065] Furthermore, by setting a fixed-length sliding time window (corresponding to the number of filter taps), the nonstationary noise signal is converted into a time-correlated vector sequence, providing a standardized input format for the subsequent robust control algorithm based on kernel parameter self-adjustment. The time window uses a forward sliding mechanism, updating the windowed data once every sampling period, ensuring that the input signal always contains the latest noise characteristics.

[0066] Step 2: Generate filter coefficients.

[0067] The adaptive filter adopts a finite impulse response (FIR) digital filter structure. Its weight coefficient generation mechanism is as follows: at the current time k, the filter generates a weight coefficient vector of equal length according to the dimension of the input signal vector to ensure the temporal and spatial consistency of the filtering operation. Its mathematical expression is:

[0068] w(k)=[w0(k),w1(k),L,w L-1 (k)] T

[0069] Each element w of the weight coefficient vector i (k) corresponds to the real-time gain of the i-th tap of the filter, which constitutes the core parameter set of time-domain adaptive filtering.

[0070] In the initialization phase of the algorithm, the weight coefficient vector adopts the zero vector initialization strategy, that is, w(0) = [0, 0, L, 0] T This initialization method ensures that there is no initial phase offset when the system starts up, avoiding undesired secondary sound source excitation. As the iterative process progresses, the weight coefficient vector is updated online through an adaptive algorithm to achieve dynamic modeling of the noise signal.

[0071] Step 3: Generate a noise cancellation signal.

[0072] At the current time k, the noise input vector x(k) is filtered using the weight coefficient vector w(k) at this time, thereby obtaining the output value y(k) = w T (k)x(k). The filter output passes through a secondary path S(z), which consists of a digital-to-analog converter (D / A), a reconstruction filter, a power amplifier, a noise cancellation speaker, a physical channel between the noise cancellation speaker and the error microphone, the error microphone, a preamplifier, an anti-aliasing filter, and an analog-to-digital converter (A / D), ultimately yielding the de-noised signal y′(k). The de-noised signal can be convolved with the secondary path impulse response to yield y′(k) = y(k) * s(k), where * represents the convolution operation and s(k) represents the impulse response of the secondary path S(z).

[0073] Step 4: Collect the residual signal.

[0074] An error microphone collects the sound signal at the noise cancellation point. After discretization, the signal at the current time k is obtained. This signal is formed by superimposing the denoised signal y′(k) with the original noise signal, and serves as the residual signal e(k) at the current time. The original noise signal is the noise signal to be cancelled and serves as the input signal to the system. The residual signal e(k) is fed into the filter, providing error feedback for the filter's adaptive iterations and driving the optimization and update of its weight coefficients.

[0075] Step 5: Adaptively select parameters.

[0076] For generalized robust adaptive loss function It contains the free parameters γ and η to be selected.

[0077] To achieve adaptive optimal determination of the parameter γ, we construct the objective function h(e(k),γ,η) = -logP(e(k),γ,η) = log(ηA(γ)) + J(k). This function is a shifted form of the logarithmic partition of the loss function J(k). Adaptive selection of the parameter γ is achieved by statistically minimizing the expectation of h(e(k),γ,η).

[0078] The parameter η is used to define the interval in which the derivative of the loss function J(k) with respect to the filter coefficient vector w(k) is approximately linearly related to the error signal, and this interval assumes that the noise signal is not contaminated by impact noise. To this end, the Hampel identifier is introduced to process outliers in the noise signal. The specific rules are:

[0079]

[0080] Where M(k) is the length L w The local median of the sliding data window, t is the threshold, and the estimated standard deviation κ=1.4826. Finally, using the preprocessed signal x H The mean absolute value of (k) is used to estimate η, and adaptive selection of η is achieved based on noise characteristics.

[0081] Step 6: Calculate gradient information.

[0082] Substitute the adaptively determined parameters γ and η in step 5 and calculate the gradient according to the defined rules:

[0083]

[0084] The gradient of the loss function with respect to the filter weight coefficient is solved to obtain the gradient information reflecting the characteristics of the residual signal, providing a direction basis for weight update.

[0085] Step 7: Update the filter weight coefficient vector.

[0086] The filter updates the gradient based on the current weight, through the weight update formula Calculate and generate the weight coefficient vector for the next moment, where x′(k)=[x′(k),x′(k-1),L,x′(k-L+1)] T , x′(k)=x(k)*s(k), μ is the iteration step size. Through this mechanism, dynamic optimization of the filter weights is achieved to improve the noise control efficiency.

[0087] Step 8: Iteration.

[0088] Let k=k+1, and repeat steps 1 to 7 until the noise control is completed.

[0089] Simulation experiment:

[0090] In order to verify the effectiveness of the present invention, a simulation experiment was conducted and compared with the method in Reference 1.

[0091] The simulation platform uses FIR filters to construct the transfer functions of the main path and the secondary path, where the transfer function of the main path is P(z)=z -5 +0.2z -6 +0.5z -7 -0.9z -8 , the transfer function of the secondary path is S(z)=z -1 +1.5z -2 -z -3 The control filter order L of the active noise control system is set to 10.

[0092] In order to verify the adaptability of the proposed method to impact noise, three α-stable distributions with different impact characteristics are selected as primary noise sources (α = 1.4, 1.6, 1.8). The time domain waveform characteristics are as follows: Figure 2 (a) Figure 2 (b) Figure 2 It should be noted that the α parameter is negatively correlated with the noise impact, that is, when the α value decreases, the noise impact characteristics are significantly enhanced.

[0093] For the three types of impact noise mentioned above, noise control simulation was carried out using the method in Reference 1 and the method of the present invention. The energy ratio of the residual signal after control processing and the original impact noise at the noise elimination point was used as the average noise residual evaluation index, and a comparison curve was drawn as shown in FIG. Figure 3 (a) Figure 3 (b) Figure 3 (c) As shown in the figure, the curve connected by the symbol "○" is the average noise residual result of the method in Reference 1, and the curve connected by the symbol "☆" is the result of the method of the present invention. Figure 3 (a) Figure 3 (b) Figure 3The results in (c) show that the residual curve of the proposed method is always below the curve of the method in Reference 1, indicating that the proposed method significantly outperforms the method in Reference 1 in controlling impact noise, with a significant reduction in the average noise residual value. Furthermore, the proposed method possesses intelligent parameter adjustment, with its core parameters dynamically optimized according to the noise characteristics. Specifically, when α = 1.4, the parameters are adaptively adjusted to η = 2.05 and γ = 0.42; when α = 1.6, the parameters are updated to η = 1.86 and γ = 0.76; and when α = 1.8, the parameters are further adjusted to η = 1.70 and γ = 0.83. This shows that as the noise impact weakens (as the α value increases), η gradually decreases and γ increases, forming an adaptive match with the noise characteristics.

[0094] The simulation results above fully demonstrate the dual advantages of this invention: First, the parameter self-optimization mechanism effectively avoids the complexity of manual parameter adjustment and achieves dynamic optimization of control parameters. Second, the novel robustness criterion enhances the algorithm's ability to suppress impulsive noise. This synergistic mechanism of adaptability and robustness enables the invention to achieve excellent noise reduction in impulsive noise environments.

[0095] The above description is only used to help understand the method and core essence of the present invention, but the scope of protection of the present invention is not limited thereto. For those skilled in the art, equivalent replacements or modifications based on the technical solutions and inventive concepts of the present invention within the technical scope disclosed by the present invention should be included in the scope of protection of the present invention. In summary, the contents of this specification should not be understood as limiting the present invention.

Claims

1. A robust active noise control method with self-adjusting parameters, characterized in that: The steps include: Acquiring reference signals: Acquire sound pressure signals in real time, convert them into discrete noise signals at the current moment as reference signals through analog-to-digital conversion, and use a forward sliding time window mechanism to construct a dynamically updated reference signal sequence; Constructing a filter: Determine the filter and generate its weight coefficients. The weight coefficient vector is updated online through an adaptive algorithm. The current reference signal is filtered to generate an output signal. Collecting the residual signal: The filter output signal is processed through the secondary path to generate a noise-cancelled signal. During the secondary path processing, the eliminated residual noise signal is collected and discretized as the residual signal. The residual signal is fed back to the filter to provide error feedback for the filter's adaptive iteration and drive the optimization and update of the filter weight coefficients. Design adaptive parameter selection update: During the filter weight coefficient update process, the generalized robust adaptive loss function is used to determine the optimal parameter combination in a statistical sense based on the objective function and the Hampel identifier to achieve adaptive optimal determination; Update the filter weight coefficient vector: Calculate the gradient of the generalized robust adaptive loss function with respect to the filter weight coefficient vector based on the residual signal at the current moment; Use the stochastic gradient method to iteratively optimize the filter weight coefficient for noise reduction processing at the next moment.

2. The robust active noise control method with self-adjusted parameters according to claim 1, characterized in that: The construction of the dynamically updated reference signal sequence is specifically as follows: Obtain the discrete noise signal x(k) at the current time k, and use the forward sliding time window mechanism to construct the reference signal sequence x(k). x(k) contains the historical noise data of L-1 sampling periods before the current time, which serves as the noise input signal vector of the filter. The expression is: x(k)=[x(k),x(k-1),L,x(k-L+1)] T Where L is the number of filter taps and T represents the transpose operation.

3. The robust active noise control method with self-adjusted parameters according to claim 2, characterized in that: The filter is determined and its weight coefficients are generated as follows: The filter adopts a finite impulse response digital filter structure, and its weight coefficient generation mechanism is as follows: at the current time k, the filter generates a weight coefficient vector of equal length according to the dimension of the noise input signal vector. Its mathematical expression is: w(k)=[w0(k),w1(k),L,w L-1 (k)] T Among them, each element w of the weight coefficient vector i (k) corresponds to the real-time gain of the i-th tap of the filter, which constitutes the core parameter set of time-domain adaptive filtering; At the current time k, the noise input signal vector x(k) is filtered using the weight coefficient vector w(k) at this time, and the filter output value y(k) = w at the current time is obtained. T (k)x(k).

4. The robust active noise control method with self-adjusted parameters according to claim 3, characterized in that: The weight coefficient vector is updated online through an adaptive algorithm, as follows: In the initialization stage, the weight coefficient vector adopts the zero vector initialization strategy, that is, w(0) = [0, 0, L, 0] T ,As the iterative process progresses, the weight coefficient vector is updated online through the ,adaptive algorithm.

5. The robust active noise control method with self-adjusting parameters according to claim 1, characterized in that: The noise cancellation signal is generated as follows: The filter output signal passes through a secondary path S(z) to obtain the noise-canceling signal; this secondary path includes a digital-to-analog converter, a reconstruction filter, a power amplifier, a noise-canceling speaker, a physical path between the noise-canceling speaker and the error microphone, the error microphone, a preamplifier, an anti-aliasing filter, and an analog-to-digital converter.

6. The method for robust active noise control with parameter self-adjustment according to any one of claims 1 to 5, characterized in that: The residual noise signal eliminated by the acquisition is specifically as follows: The error microphone in the secondary path is used to collect the sound signal at the noise cancellation point. After discretization processing, the signal at the current moment k is obtained, which is the residual signal e(k) at the current moment; this residual signal is formed by the superposition of the noise cancellation signal and the original noise signal.

7. The robust active noise control method with parameter self-adjustment according to claim 6, characterized in that: The optimal parameter combination is determined as follows: Adaptively select the free parameters γ and η to be selected in the generalized robust adaptive loss function J(k); For the parameter γ, the objective function is constructed as follows: h(e(k),γ,η)=-logP(e(k),γ,η)=log(ηA(γ))+J(k) Where e(k) represents the residual signal at the current moment; the objective function is a shifted form of the logarithmic division of the loss function J(k); the adaptive selection of the parameter γ is achieved by statistically minimizing the expectation of h(e(k),γ,η); The parameter η is used to define the interval in which the derivative of the loss function J(k) with respect to the filter weight coefficient vector w(k) has an approximately linear relationship with the error signal, and this interval assumes that the noise signal is not contaminated by impact noise. By introducing a Hampel identifier to process the outliers of the noise signal, and using the mean absolute value of the preprocessed signal to estimate η, the parameter η is adaptively selected based on the noise characteristics.

8. The robust active noise control method with parameter self-adjustment according to claim 7, characterized in that: The parameter η is adaptively selected based on noise characteristics, specifically as follows: The Hampel identifier is introduced to process the outliers of the noise signal. The specific rules are as follows: Where M(k) is the length L w The local median of the sliding data window, t is the threshold, σ sk (k) represents the estimated standard deviation, κ = 1.4826; by preprocessing the signal x H The mean absolute value of (k) estimates η.

9. The robust active noise control method with parameter self-adjustment according to claim 8, characterized in that: The gradient of the filter weight coefficient vector, the gradient calculation rule is as follows: The gradient is solved to obtain the gradient information reflecting the characteristics of the residual signal, providing a direction basis for updating the weight coefficient vector.

10. The robust active noise control method with parameter self-adjustment according to claim 9, characterized in that: The iterative optimization filter weight coefficients are specifically as follows: The filter updates the gradient based on the current weight coefficient and generates the weight coefficient vector at the next moment through the weight coefficient update formula to achieve dynamic optimization of the filter weight coefficient; the weight coefficient update formula is as follows: x′(k)=[x′(k),x′(k-1),L,x′(k-L+1)] T x′(k)=x(k)*s(k) Where μ is the iteration step size and s(k) represents the impulse response of the secondary path S(z).

Images