An active noise control method and system based on generalized hyperbolic tangent

Abstract

This invention discloses an active noise control method and system based on generalized hyperbolic tangent, belonging to the field of active noise control technology. The method involves picking up a primary noise signal and constructing a noise input vector. The noise input vector is then filtered to generate an anti-noise signal, which is played through a secondary speaker. An error microphone is used to collect the residual signal. Based on a generalized hyperbolic tangent robust cost function containing saturation level control parameters, transient characteristic adjustment parameters, and small error penalty intensity adjustment parameters, the residual signal undergoes a nonlinear transformation to obtain a nonlinear error signal. This nonlinear error signal is then used to update the weight coefficient vector, and the above steps are iteratively executed to complete the noise control. The system includes a reference microphone, an adaptive filter, a secondary speaker, an error microphone, a secondary path estimation module, and a weight update module. This invention can flexibly adapt to different noise scenarios and ensure stable system convergence.

Description

Technical Field

[0001] This invention relates to the field of active noise control technology, specifically to an active noise control method and system based on generalized hyperbolic tangent. Background Technology

[0002] Active noise control (ANC) technology, as a highly efficient noise suppression method, works by generating an anti-noise signal with the same amplitude but opposite phase as the primary noise through an adaptive filter. This anti-noise signal is then played through a secondary speaker and interacts destructively with the primary noise, thus eliminating the noise. Among various ANC algorithms, the Filter-x Least Mean Square (FxLMS) algorithm is the most classic and widely used fundamental algorithm due to its simple structure and ease of engineering implementation. This algorithm preprocesses the reference signal through a secondary path estimation module to compensate for the influence of the acoustic path on signal transmission, and then updates the filter weight coefficients based on the least mean square error criterion. It achieves good noise reduction results in ideal Gaussian noise environments and has been successfully applied in consumer electronics fields such as headphones and automotive noise reduction systems. However, its performance limitations become increasingly apparent in complex noise scenarios such as industrial production and construction. In particular, impulsive non-Gaussian noise can cause deviations in the algorithm's gradient estimation, leading to drastic adjustments in the filter coefficients and ultimately causing system divergence.

[0003] The standard FxLMS algorithm is essentially derived based on the assumption that noise follows a Gaussian distribution. However, real-world noise environments often exhibit non-Gaussian characteristics, especially with the presence of impulsive noise and pulse interference, which severely impacts the algorithm's stability and noise reduction performance. In industrial settings such as stamping workshops and mining operations, impulsive noise contains numerous large-amplitude spike pulses, which can cause significant deviations in the gradient estimation of the FxLMS algorithm. This leads to drastic and erroneous adjustments to the adaptive filter coefficients, ultimately causing system divergence, a sharp deterioration in noise reduction performance, or even complete failure. Similarly, pulse interference caused by sensor malfunctions and electromagnetic interference can also disrupt the algorithm's convergence stability. To address this issue, various robustness improvement algorithms have been proposed, such as nonlinear transformation methods based on Huber functions or M-estimation, and outlier suppression methods based on Gaussian kernel functions. However, these solutions generally have significant limitations: Huber function-based algorithms require manual threshold setting, making parameter selection difficult and easily affecting convergence speed; Gaussian kernel function algorithms are sensitive to kernel bandwidth parameters, and exponential operations incur significant computational overhead; while some hyperbolic tangent or exponential cost function algorithms can suppress outliers to a certain extent, they lack adaptability to the dynamic range of errors, have poor parameter adjustment flexibility, and exhibit poor gradient smoothness under strong noise, making it difficult to simultaneously meet the dual requirements of robustness and convergence efficiency.

[0004] To address the robustness issue of ANC systems under non-Gaussian noise, targeted patent research has been conducted in related fields. Chinese patent CN120636360A (A Robust Active Noise Control Method with Self-Adjusting Parameters) proposes a robust active noise control method with self-adjusting parameters. This method constructs a dynamic reference signal sequence by acquiring sound pressure signals in real time to form a discrete noise reference signal; designs a filter and generates weight coefficients; filters the reference signal to generate a denoised signal; and collects the residual signal as error feedback. The core of this method is the adaptive determination of the optimal parameter combination for a generalized robust adaptive loss function. The filter weight coefficient vector is updated through a parameter self-adjustment mechanism, achieving fast convergence, low steady-state error, and strong impact resistance, thus improving noise suppression efficiency in complex environments. However, the loss function in this patent is not constructed based on the generalized hyperbolic tangent, resulting in insufficient optimization of gradient smoothness under impact noise. Furthermore, the parameter self-adjustment mechanism relies on a complex statistical identification process, leading to significant computational overhead. To address the computational complexity issue of the aforementioned patents, patent CN119516991A (A Robust Online Secondary Path Modeling Active Noise Control Method) proposes using a microphone to acquire primary noise signals, updating the weight vector of the primary noise, calculating the error signal using the updated compensation signal and the primary noise, and then using the error signal to update the adaptive filtering weights in secondary path modeling, achieving noise control through iteration. However, this patent still uses a traditional robust cost function, resulting in limited suppression performance against impulse interference. Furthermore, it does not incorporate hyperbolic secant operations to handle nonlinear error signals, and still suffers from insufficient gradient attenuation under extremely strong impulse noise. Additionally, it lacks a multi-parameter adjustment mechanism to achieve a linear-saturation transition, making it difficult to flexibly adapt to noise environments with different statistical characteristics.

[0005] Therefore, it is urgent to address the technical pain points of existing algorithms, such as insufficient adaptability in complex non-Gaussian noise environments, poor convergence stability, and cumbersome parameter adjustment. Summary of the Invention

[0006] To address the aforementioned technical problems, this application discloses an active noise control method and system based on generalized hyperbolic tangent. The active noise control method based on generalized hyperbolic tangent specifically includes:

[0007] Obtain the primary noise signal generated by the noise source at discrete time nodes, and construct the noise input vector of the filter;

[0008] The filter's weight coefficient vector is initialized, and the noise input vector is filtered based on the weight coefficient vector to generate an anti-noise output signal, which is then played through a secondary speaker.

[0009] The residual signal at the noise cancellation point is acquired by an error microphone. The residual signal is then subjected to a nonlinear transformation based on a preset generalized hyperbolic tangent robust cost function to obtain a nonlinear error signal. The robust cost function includes three adjustable parameters: a saturation level control parameter, a transient characteristic adjustment parameter, and a small error penalty intensity adjustment parameter.

[0010] The weight coefficient vector is updated using the nonlinear error signal to obtain the weight coefficient vector for the next discrete time node;

[0011] Repeat the above steps iteratively until active noise control is complete.

[0012] Preferably, the step of acquiring the primary noise signal generated by the noise source at discrete time nodes and constructing the noise input vector of the filter includes:

[0013] Discrete time nodes are picked up using a reference microphone. Primary noise value Select discrete time nodes and before At that moment Primary noise value Construct the noise input vector , superscript For transpose operation, The number of taps in the filter is denoted as .

[0014] Preferably, the initialization of the filter's weight coefficient vector includes:

[0015] Extract the current time Individual weight coefficient Construct the weight coefficient vector , ;

[0016] When discrete time node At that time, the weight coefficient vector Set all elements to 0.

[0017] Preferably, filtering the noise input vector based on the weight coefficient vector to generate an anti-noise output signal includes:

[0018] Calculate the weight coefficient vector With the noise input vector The inner product is then compared with the impulse response of the secondary path. Perform convolution operation to obtain the anti-noise output signal. , ,symbol This is a convolution operation.

[0019] Preferably, the formula for the generalized hyperbolic tangent robust cost function is:

[0020]

[0021] in, For residual signals, This is the saturation level control parameter. Adjust parameters for transient characteristics. This is the parameter for adjusting the penalty intensity for small errors.

[0022] Preferably, the parameters For positive real numbers, when At that time, the robust cost function degenerates into It still maintains robust suppression characteristics against large-amplitude residual signals.

[0023] Preferably, the residual signal is nonlinearly transformed based on a preset generalized hyperbolic tangent robust cost function to obtain a nonlinear error signal, including:

[0024] Based on the robust cost function, the residual signal The nonlinear error signal is derived by taking the derivative and using the gradient descent method. Its formula is:

[0025]

[0026] in, This is the hyperbolic secant operation.

[0027] Preferably, updating the weight coefficient vector using the nonlinear error signal to obtain the weight coefficient vector for the next discrete time node includes:

[0028] The nonlinear error signal Compared with the reference signal vector filtered by the secondary path estimation model Multiply, then combine the product with the step factor. After multiplication, the weight coefficient vector at the current discrete time node. By superimposing these values, we obtain the weight coefficient vector for the next discrete time node. , Reference signal vector The noise input vector via secondary path estimation model Filtered signal, step size factor This is an adjustable parameter used to adjust the update rate of the weight coefficient vector.

[0029] The active noise control system based on generalized hyperbolic tangent includes a reference microphone, an adaptive filter, a secondary speaker, an error microphone, a secondary path estimation module, and a weight update module.

[0030] The reference microphone is used to pick up the primary noise signal generated by the noise source and to construct the noise input vector of the adaptive filter;

[0031] The adaptive filter has a built-in 128th-order FIR filter unit, which is used to initialize the weight coefficient vector and filter the noise input vector based on the weight coefficient vector to generate an anti-noise output signal.

[0032] The secondary loudspeaker is used to play the anti-noise output signal, which forms a destructive interference with the primary noise;

[0033] The error microphone is used to collect the residual signal at the noise cancellation point and transmit it to the weight update module;

[0034] The secondary path estimation module is used to filter the noise input vector to obtain a reference signal vector and transmit it to the weight update module.

[0035] The weight update module incorporates the generalized hyperbolic tangent robust cost function, which is used to perform nonlinear transformation on the residual signal to obtain a nonlinear error signal, and to iteratively update the weight coefficient vector of the adaptive filter based on the nonlinear error signal and the reference signal vector.

[0036] Compared with the prior art, the technical solution of this application has the following technical effects:

[0037] The active noise control method of this invention relies on a dedicated generalized hyperbolic tangent robust cost function to achieve nonlinear processing of the residual signal. This function has multi-dimensional adjustable parameters, which can flexibly adapt to different types and intensities of noise environments. The parameter configuration can be adjusted according to the noise statistical characteristics of the actual acoustic scene, so that the active noise control system can maintain adaptability in various complex acoustic environments, achieve precise noise suppression, and improve the system's environmental adaptability.

[0038] The control method of this invention has excellent robustness. In scenarios with large-amplitude impact noise and pulse interference, the cost function will enter a saturation state as the residual signal amplitude increases, and the gradient will approach zero. This can effectively suppress the adverse effects of large-amplitude errors on the updating of filter weight coefficients, avoid drastic and erroneous adjustments to the weight coefficients, ensure that the adaptive filter always works stably, and keep the entire noise control system in a convergent state without system divergence.

[0039] The control method of this invention uses a generalized hyperbolic tangent robust cost function that is a continuous and differentiable smooth function with no discontinuities. This function is compatible with the weight coefficient update logic of the gradient descent method and will not cause convergence oscillation problems due to function discontinuity. At the same time, the penalty intensity can be optimized by adjusting parameters in small error scenarios to ensure the convergence accuracy of the system in the low noise residual stage. This allows the system to maintain a fast convergence rate while operating stably, thereby improving the response efficiency of noise control.

[0040] The core operations of the control method of this invention only include hyperbolic tangent, hyperbolic secant, absolute value and basic power and fraction operations, without complex exponential operations, the calculation logic is simple and the hardware implementation is easy. At the same time, the method can be directly extended to multi-channel active noise control systems without major algorithm reconstruction. The system has good scalability and takes into account the convenience of engineering implementation and the flexibility of practical application, making it easy to implement in various noise control scenarios.

[0041] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the preferred embodiments of this application are described in detail below with reference to the accompanying drawings.

[0042] The above and other objects, advantages and features of this application will become more apparent to those skilled in the art from the following detailed description of specific embodiments in conjunction with the accompanying drawings. Attached Figure Description

[0043] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. In all drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, the elements or parts are not necessarily drawn to scale.

[0044] Based on the description of the figures and their corresponding technical content in the document, the titles of the figures are as follows:

[0045] Figure 1 A flowchart illustrating the steps of an active noise control method based on generalized hyperbolic tangent.

[0046] Figure 2 : An architecture diagram of an active noise control system based on generalized hyperbolic tangent that includes multiple modules working collaboratively;

[0047] Figure 3Time-domain waveforms of impulse noise with three stable α distributions of α values ​​of 1.3, 1.5, and 1.7.

[0048] Figure 4 The figure shows a comparison of the average noise residue of the method of this invention and two existing robust algorithms for processing impact noise of different intensities. Detailed Implementation

[0049] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. In the following description, specific details such as specific configurations and components are provided merely to help fully understand the embodiments of this application. Therefore, those skilled in the art should understand that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this application. In addition, for clarity and brevity, descriptions of known functions and structures are omitted in the embodiments.

[0050] It should be understood that the phrase "an embodiment" or "this embodiment" throughout the specification means that a specific feature, structure, or characteristic related to the embodiment is included in at least one embodiment of this application. Therefore, "an embodiment" or "this embodiment" appearing throughout the specification does not necessarily refer to the same embodiment. Furthermore, these specific features, structures, or characteristics can be combined in any suitable manner in one or more embodiments.

[0051] Furthermore, reference numerals and / or letters may be repeated in different examples within this application. Such repetition is for the purpose of simplification and clarity and does not in itself indicate a relationship between the various embodiments and / or settings discussed.

[0052] In this article, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can mean: A exists alone, B exists alone, and A and B exist simultaneously. The term " / and" in this article describes another type of relationship between related objects, indicating that two relationships can exist. For example, A / and B can mean: A exists alone, and A and B exist alone. In addition, the character " / " in this article generally indicates that the related objects before and after it are in an "or" relationship.

[0053] In this article, the term "at least one" is merely a description of the relationship between related objects, indicating that there can be three relationships. For example, "at least one of A and B" can mean: A exists alone, A and B exist simultaneously, or B exists alone.

[0054] It should also be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion.

[0055] Example 1

[0056] This embodiment mainly describes an active noise control method based on generalized hyperbolic tangent, such as... Figure 1 As shown, it specifically includes:

[0057] Obtain the primary noise signal generated by the noise source at discrete time nodes, and construct the noise input vector of the filter;

[0058] The filter's weight coefficient vector is initialized, and the noise input vector is filtered based on the weight coefficient vector to generate an anti-noise output signal, which is then played through a secondary speaker.

[0059] The residual signal at the noise cancellation point is acquired by an error microphone. The residual signal is then subjected to a nonlinear transformation based on a preset generalized hyperbolic tangent robust cost function to obtain a nonlinear error signal. The robust cost function includes three adjustable parameters: a saturation level control parameter, a transient characteristic adjustment parameter, and a small error penalty intensity adjustment parameter.

[0060] The weight coefficient vector is updated using the nonlinear error signal to obtain the weight coefficient vector for the next discrete time node;

[0061] Repeat the above steps iteratively until active noise control is complete.

[0062] Furthermore, the primary noise signal generated by the noise source at discrete time nodes is obtained, and the noise input vector of the filter is constructed. The acoustic signal of the noise source at discrete time nodes is picked up in real time through a reference microphone to obtain the discrete time node. Primary noise value ,in A positive integer representing a discretized time counter node. This discrete time node... Based on the baseline, select consecutive preceding... The initial noise value at each time step, compared with the current time step. The primary noise values ​​together constitute the total. A sequence of primary noise values ​​for elements, specifically the sequence is... ,in This is a fixed number of taps for the filter, adapted to the structural design of a 128th-order FIR adaptive filter. The above... The primary noise values ​​are constructed into column vectors in chronological order to obtain the filter at discrete time nodes. The noise input vector Its mathematical formula is superscript in the formula This is a vector transpose operation that converts a one-dimensional noise value sequence into a column vector form suitable for filter operations, providing a standardized input signal carrier for subsequent filtering operations.

[0063] Further, the weight coefficient vector of the filter is initialized, and the noise input vector is filtered based on the weight coefficient vector to generate an anti-noise output signal, which is then played through a secondary speaker. Specifically:

[0064] Initialization of the filter weight coefficient vector: For a 128th-order FIR adaptive filter, extract its values ​​at discrete time nodes. The weight coefficients below and select consecutive preceding ones The weight coefficients at each time point, where The number of filter taps is consistent with the dimension of the noise input vector, forming a structure containing... The weight coefficient sequence of each element is used to construct the filter at discrete time nodes by arranging this weight coefficient sequence in chronological order. The weight coefficient vector below Its mathematical formula is This vector is in row vector form, which is compatible with the inner product operation rules of the noise input column vectors. Initialization is performed on the weight coefficient vector, and when the discrete time nodes satisfy... When the condition is met, the weight coefficient vector will be... All elements are set to 0 to complete the initial state configuration of the filter and ensure the operational stability of the filter in the early stages of iteration.

[0065] The filtering process generates an anti-noise output signal. Based on the linear filtering operation rules of the FIR adaptive filter, the discrete time nodes are first processed. The weight coefficient row vector With noise input column vector Performing an inner product operation yields the fundamental filter value. This inner product operation involves multiplying corresponding elements sequentially and then summing the results, adapting to the convolution logic of an FIR filter. The result of this inner product operation is then compared with the impulse response of the secondary path. Perform a convolution operation; the convolution operator is... This operation compensates for the influence of the acoustic transmission path from the filter to the secondary speaker on the signal, making the generated signal match the actual acoustic propagation characteristics, and finally obtaining the discrete-time node. The anti-noise output signal is output to the secondary speaker. Its mathematical formula is In the formula The column vector is converted into a row vector by transposing the weight coefficient vector, ensuring dimension matching for the inner product operation. After the anti-noise output signal is generated, it is converted into an acoustic signal by a secondary speaker and played out, providing an acoustic signal basis for destructive interference with the primary noise.

[0066] Furthermore, the residual signal at the noise cancellation point is acquired through an error microphone, and a nonlinear transformation is performed on the residual signal based on a preset generalized hyperbolic tangent robust cost function to obtain a nonlinear error signal, specifically:

[0067] The residual signal is acquired by deploying an error microphone at the target location for noise cancellation and acquiring the acoustic signal at that location in real time. This acoustic signal is the residual acoustic signal after the primary noise and the anti-noise played by the secondary speaker undergo destructive interference. After converting it into an electrical signal, the discrete-time node is obtained. The residual signal below This signal is in scalar form and directly represents the noise reduction effect at the current moment, providing error feedback for the subsequent update of filter weight coefficients.

[0068] The predefined generalized hyperbolic tangent robust cost function is a continuously differentiable smooth function, suitable for optimization operations using gradient descent. Its mathematical formula is as follows: In the formula Discrete time nodes The residual signal below, The absolute value of the residual signal ensures the non-negativity of the error term during the calculation process; This is the saturation level control parameter, and its value satisfies... The constraints are used to control the overall saturation level of the cost function and determine the threshold range for the cost function to enter the saturation state. This is a transient characteristic adjustment parameter, and its value must satisfy... The constraints are used to control the steepness of the transition and the growth rate of the cost function from linear change to saturation. This is a parameter for adjusting the penalty strength for small errors; the value of this parameter satisfies... The constraint condition, and this parameter can be a positive integer or a positive real number, is used to control the penalty strength of the cost function in small error scenarios, adapting to the convergence accuracy requirements under different noise environments. When the transition characteristic adjustment parameter... At that time, the robust cost function degenerates into It still maintains the robust suppression characteristics for large-amplitude residual signals, and is a simplified application form of the cost function.

[0069] The nonlinear error signal is obtained based on the nonlinear transformation of the robust cost function. The gradient descent method is then used to analyze the generalized hyperbolic tangent robust cost function with respect to the residual signal. The first derivative is calculated, and the gradient information of the cost function is obtained through derivative operations. This gradient information is the core basis for updating the weight coefficient vector. The discrete time nodes are then derived using this gradient information. Nonlinear error signal Its mathematical formula is:

[0070]

[0071] In the formula The hyperbolic secant operation is a derivative operation of the hyperbolic tangent function, satisfying the following conditions: , The square of the hyperbolic secant function, whose value decreases exponentially with increasing independent variable; fractional terms In the cost function Part of the information about | The derivative of |, when combined with the sign function sign(e(n)), forms the complete derivative with respect to e(n). As the core error signal for weight coefficient updating, a nonlinear transformation of the residual signal is realized: the small error region is approximately linear to maintain the convergence speed, and the large error region effectively suppresses the interference of outliers on weight updating through saturation and decay characteristics, thereby enhancing the robustness of the algorithm.

[0072] Furthermore, the weight coefficient vector is updated using the nonlinear error signal to obtain the weight coefficient vector for the next discrete time node, thus constructing a reference signal vector. This vector represents discrete-time nodes. The noise input vector via secondary path estimation model Filtered signal, secondary path estimation model This model estimates the acoustic path transfer function from the loudspeaker to the error microphone. It uses filtering to compensate for the influence of secondary acoustic paths on the reference signal, ensuring the accuracy of weight coefficient updates. The reference signal vector... With noise input vector These are column vectors of the same dimension. A step size factor is selected. This factor is an adjustable positive parameter used to adjust the update rate of the weight coefficient vector, adapting to the convergence speed requirements under different noise environments. The discrete time nodes... Nonlinear error signal With reference signal vector Perform scalar multiplication to obtain the basic value of the update increment in vector form, and then multiply this basic value by the step size factor. Perform scalar multiplication to obtain the final weight coefficient vector update increment. This updates the discrete time nodes. The weight coefficient vector below Perform vector addition with the above update increment to obtain the next discrete time node. The weight coefficient vector below Its mathematical formula is The update formula is in vector operation form, which ensures that each element in the weight coefficient vector is updated and adjusted accordingly, adapting to the weight coefficient configuration requirements of a 128th-order FIR adaptive filter.

[0073] Iterative execution continues until active noise control is completed, at which point discrete time nodes are obtained. weight coefficient vector Then, update the counts of the discrete time nodes to... Using the updated time point as the new current moment, the entire iterative computation process is repeated sequentially, including constructing the noise input vector, initializing the weight coefficient vector and filtering to generate the anti-noise output signal, acquiring the residual signal and generating the nonlinear error signal, and updating the weight coefficient vector, forming a closed-loop iterative operation. During the iteration, the filter's weight coefficient vector is continuously updated as the time point advances, and the anti-noise output signal is also dynamically adjusted accordingly, achieving real-time adaptive noise control of the noise source. The iterative operation continues until the noise control objective is achieved, or the residual signal at the noise cancellation point reaches a preset stable state, or a noise control stop command is received, at which point the entire iterative process terminates, completing all operations of active noise control.

[0074] This embodiment details how the present application achieves active noise control based on the generalized hyperbolic tangent robust cost function. Through flexible configuration of saturation level control parameters, transient characteristic adjustment parameters, and small error penalty intensity adjustment parameters, it can adapt to different noise scenarios. The continuous differentiability of the cost function ensures the stability of weight coefficient updates, and the hyperbolic secant operation and gradient decay characteristics can effectively suppress the interference of large-amplitude impact noise on the system.

[0075] Example 2 describes in detail an active noise control system based on generalized hyperbolic tangent, including a reference microphone, an adaptive filter, a secondary speaker, an error microphone, a secondary path estimation module, and a weight update module, such as... Figure 2 As shown;

[0076] A reference microphone is used to pick up the primary noise signal generated by the noise source and to construct the noise input vector of the adaptive filter;

[0077] The adaptive filter has a built-in 128th-order FIR filter unit, which is used to initialize the weight coefficient vector and filter the noise input vector based on the weight coefficient vector to generate an anti-noise output signal.

[0078] The secondary speaker is used to play the anti-noise output signal, which cancels out interference with the primary noise.

[0079] The error microphone is used to collect the residual signal at the noise cancellation point and transmit it to the weight update module;

[0080] The secondary path estimation module is used to filter the noise input vector to obtain the reference signal vector and transmit it to the weight update module;

[0081] The weight update module incorporates the generalized hyperbolic tangent robust cost function, which is used to perform nonlinear transformation on the residual signal to obtain a nonlinear error signal, and to iteratively update the weight coefficient vector of the adaptive filter based on the nonlinear error signal and the reference signal vector.

[0082] Furthermore, the reference microphone, serving as the system's noise signal input, possesses real-time acoustic signal pickup capabilities, accurately capturing the primary noise acoustic signals generated by the noise source at continuous discrete time points and converting them into standardized electrical signals. At discrete time points... The reference microphone picks up the primary noise value at the current moment. At the same time, automatic caching before Primary noise values ​​at consecutive time points ( (consistent with the number of taps in the adaptive filter), forming a system containing... A sequence of noise values ​​for each element. Based on this sequence, the reference microphones perform column vector construction operations in chronological order to generate the noise input vector required by the adaptive filter. superscript This is a vector transpose operation, ensuring that the vector dimension matches the computational requirements of the adaptive filter, and providing a standardized input carrier for subsequent filtering processing.

[0083] Furthermore, the adaptive filter incorporates a fixed-configuration 128th-order FIR (Finite Length Unit Impulse Response) filter unit, the structure of which determines the number of taps in the filter. It adapts to the convolution operation requirements of noisy signals at multiple time points. Its core operation process consists of two steps: weight coefficient vector initialization and filtered signal generation.

[0084] Weight coefficient vector initialization: The filter extracts the current discrete time node. and before At that moment Each weight coefficient is used to construct a weight coefficient vector, and at discrete time nodes... At that time, the filter is in the initial startup phase, and the weight coefficient vector... All elements are set to 0 to avoid computational interference in the initial state and ensure system startup stability.

[0085] Filtering to generate anti-noise signals: Based on the FIR filtering principle, the filter first adjusts the weight coefficient vector... With noise input vector Perform inner product operation ( The filtered base signal is obtained; then the base signal is compared with the impulse response of the secondary path. Perform convolution operation (symbol) The convolution operation compensates for the influence of the acoustic transmission path from the filter to the secondary speaker on the signal, ultimately generating an anti-noise output signal. This signal is a prototype of an acoustic signal with amplitude matching and phase opposite to the primary noise, providing a basis for destructive interference.

[0086] Furthermore, the secondary loudspeaker, serving as the system's acoustic signal output, receives the anti-noise output electrical signal transmitted by the adaptive filter and efficiently converts it into an acoustic signal for external playback. This standardized anti-noise signal is precisely released into the noise control area, allowing the anti-noise and primary noise to meet in space and form destructive interference. Through the equal amplitude and opposite phase characteristics of the anti-noise and primary noise, the acoustic energy of the primary noise is canceled out, thereby reducing the overall noise intensity in the noise control area. The loudspeaker's output power and acoustic response characteristics are optimized to ensure that the propagation range and amplitude stability of the anti-noise signal are matched with the distribution characteristics of the primary noise, guaranteeing the uniformity and reliability of the interference cancellation effect.

[0087] Furthermore, an error microphone is deployed in the target area for noise cancellation (i.e., the core location where interference cancellation occurs) to monitor the residual acoustic signal after interference cancellation in real time and convert it into a residual signal in the form of an electrical signal. The residual signal directly characterizes the noise control effect at the current moment and is the core feedback basis for the system's adaptive adjustment. When the residual signal amplitude is large, it indicates that the current anti-noise signal and the primary noise are not well matched, requiring optimization through subsequent weight coefficient updates. When the residual signal amplitude is small and tends to be stable, it indicates that the system has achieved the ideal noise control effect. The error microphone has high sensitivity and low noise characteristics, enabling it to accurately capture weak residual noise signals, ensuring the authenticity and accuracy of the feedback signal, and providing reliable input data for the weight update module.

[0088] Furthermore, the core function of the secondary path estimation module is to process the noisy input vector. Filtering is performed to generate a reference signal vector. Essentially, it estimates and compensates for the transfer function of the secondary acoustic path, from the loudspeaker to the noise cancellation point to the error microphone. The module has a built-in secondary path estimation model. This model models the acoustic characteristics of the secondary acoustic path (such as signal attenuation, time delay, frequency response, etc.) and the noise input vector. Perform filtering operations to simulate the transmission process of the anti-noise signal in the secondary path, and generate a reference signal vector that matches the characteristics of the signal after actual transmission. This vector is used in subsequent weight coefficient update calculations, which can compensate for the influence of secondary paths on the signal, avoid weight coefficient update deviations caused by acoustic path characteristics, and ensure the convergence stability and control accuracy of the system.

[0089] Furthermore, the weight update module incorporates a generalized hyperbolic tangent robust cost function. Through nonlinear transformation and iterative updating of weight coefficients, it enables the system to adaptively adapt to noisy environments. The specific computational process is as follows:

[0090] Robust cost function configuration: The module's built-in generalized hyperbolic tangent robust cost function formula is as follows: ,in The residual signal transmitted by the error microphone. This is the saturation level control parameter. Adjust parameters for transient characteristics. (Can be a positive real number) is the parameter for adjusting the intensity of the small error penalty; when When the cost function degenerates to It still retains its robust properties.

[0091] Nonlinear error signal generation: Based on the gradient descent optimization principle, the module optimizes the robust cost function with respect to the residual signal. By taking the first derivative, the nonlinear error signal is derived. The formula is:

[0092]

[0093] in The hyperbolic secant operation effectively suppresses the influence of large-amplitude residual signals and preserves the feedback effect of small-error signals through this nonlinear transformation.

[0094] Iterative update of weight coefficient vector: The module updates the nonlinear error signal Reference signal vector generated by the secondary path estimation module Multiply, then add the step factor. (Adjustable parameter, adjusting update rate) multiply to obtain the weight coefficient update increment; this increment is then multiplied by the weight coefficient vector at the current time. Superimpose to generate the weight coefficient vector for the next discrete time node. The signal is then fed back to the adaptive filter to update the weighting coefficients. Through continuous iteration of this process, the anti-noise signal is constantly optimized, ensuring that the system maintains good noise control performance even in complex noise environments.

[0095] This embodiment describes in detail how the collaborative work of various modules enables adaptive adaptation to various noise scenarios, ensuring stable system convergence. It features a simple structure, high computational efficiency, and strong adaptability, stably achieving precise noise suppression and completing reliable active noise control without complex debugging.

[0096] Based on Embodiment 1 or 2, this embodiment, in order to fully verify the effectiveness and superiority of the active noise control method based on generalized hyperbolic tangent of this application, designed targeted simulation experiments. By comparing with existing robust algorithms, the algorithm performance is evaluated from two core dimensions: convergence speed and steady-state noise residue. Specifically:

[0097] The acoustic path modeling in the simulation experiment uses high-order FIR filters, with both the main and secondary paths using 256-order filters to accurately simulate the signal transmission delay and frequency response characteristics in a real acoustic environment. The active noise controller uses a 128th-order FIR adaptive filter with L=128 taps, consistent with the dimensions of the weight coefficient vector and the noise input vector to ensure dimensionality matching of the filtering operation. The key parameters involved in the experiment are uniformly configured as follows: step size factor. =0.0001, the saturation level control parameter of the generalized hyperbolic tangent robust cost function. =1.0, transient characteristic adjustment parameter α=0.05, small error penalty intensity adjustment parameter p=2.0; the impulse response s of the secondary path was obtained by measuring the acoustic path in the industrial environment, the sampling frequency was set to 16kHz, and the total length of the experimental data was 10,000 discrete time nodes to ensure the statistical reliability of the experimental results.

[0098] like Figure 3 As shown, the experiment selected three different intensities of α-stable distribution impulse noise as primary noise to cover three typical non-Gaussian noise scenarios: light, medium, and heavy. The α values ​​were set to 1.3, 1.5, and 1.7, respectively. The smaller the characteristic index α of the α-stable distribution, the stronger the impulse of the noise and the larger the amplitude of the spike pulse. α=1.3 corresponds to high-intensity impulse noise, α=1.5 corresponds to medium-intensity impulse noise, and α=1.7 corresponds to low-intensity impulse noise. The specific time-domain characteristics of the three primary noises are presented in three figures: the time-domain waveform of the high-intensity impulse noise with α=1.3 is shown below. Figure 3 a. As can be seen, the waveform contains a large number of large-amplitude spike pulses, with the amplitude ratio of the pulse peak value to the background noise exceeding 20:1, and the pulse duration concentrated in 1-3 sampling points; the time-domain waveform of moderate-intensity impulse noise with α=1.5 is... Figure 3 b, The amplitude and density of the spike pulse are relatively... Figure 3 α decreases somewhat, and the amplitude ratio of the pulse peak to the background noise is approximately 15:1; the time-domain waveform of low-intensity impulse noise with α=1.7 is... Figure 3 c. The amplitude of the spike pulse is further reduced, and the ratio of the peak pulse amplitude to the background noise amplitude is about 10:1. The waveform is closer to Gaussian noise but still contains obvious non-Gaussian pulse components.

[0099] like Figure 4 As shown, to verify the superiority of the method in this application, two existing typical robust algorithms are selected as comparison objects: one is the robust FxLMS algorithm based on Huber function (corresponding to reference 1 Nithin V. George and GanapatiPanda. A robust filtered-s LMS algorithm for nonlinear active noise control[J]. Applied Acoustics, vol. 73, pp: 836–841, 2012), and the other is the FxEHSAF algorithm based on exponential hyperbolic secant (corresponding to reference 2 Tanveer Alam Khan and Somanath Pradhan. Robustexponential hyperbolic secant algorithm for active control against impulsivenoise environments[J]. IEEE Signal Processing Letters, vol. 33, 2026). The parameters of both comparison algorithms are set according to the optimal configuration of their original references to ensure the fairness of the comparison. Three algorithms were applied to process primary noise of three different intensities. Experimental results were evaluated using the mean noise residual ratio (the ratio of the root mean square value of the filtered residual signal to the root mean square value of the unfiltered primary noise, in dB). Figure 4 Presentation of processing results: Figure 4 'a' represents the processing results for high-intensity impact noise with α=1.3. It can be seen that the average noise residual ratio of the method in this application is stable at around -32dB, while the Huber-FxLMS algorithm in Reference 1 is stable at around -20dB, and the FxEHSAF algorithm in Reference 2 is stable at around -24dB. The steady-state noise residual ratio of the method in this application is 8-12dB lower than that of the comparative algorithms, and the convergence speed is faster, reaching a stable state after 2000 time nodes, while the comparative algorithms require more than 3500 time nodes to stabilize. Figure 4b represents the processing results for moderate-intensity impact noise with α=1.5. The average noise residual ratio of the method in this application is stable at around -35dB, the algorithm in Reference 1 is stable at around -25dB, and the algorithm in Reference 2 is stable at around -28dB. The steady-state noise residual advantage of the method in this application is still maintained at 7-10dB, and the convergence stabilization time is shortened to 1800 time nodes. Figure 4 c represents the processing results for low-intensity impact noise with α=1.7. The average noise residual ratio of the method in this application is stable at around -38dB, while the algorithm in Reference 1 is stable at around -30dB and the algorithm in Reference 2 is stable at around -32dB. At the same time, the convergence speed of the method in this application is further improved, reaching stability after 1500 time nodes, and the fluctuation range of steady-state noise residual is less than ±0.5dB, which is more stable than the comparative algorithms.

[0100] As can be seen from the simulation experiments and corresponding results in the figures above, regardless of whether the impact noise environment is high-intensity, medium-intensity, or low-intensity, the active noise control method based on generalized hyperbolic tangent in this application can maintain a faster convergence speed and a lower steady-state noise residue, which fully verifies its robustness and superiority in non-Gaussian impact noise environment. At the same time, in the low-intensity impact noise scenario, its advantages in convergence speed and stability are further highlighted, indicating that the method can flexibly adapt to noise environments with different statistical characteristics.

[0101] The above are merely preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. For those skilled in the art, the present invention can have various modifications and variations. Any changes, modifications, substitutions, integrations, and parameter changes made to these embodiments within the spirit and principles of the present invention, without departing from the principles and spirit of the present invention, through conventional substitutions or to achieve the same function, fall within the scope of protection of the present invention.

Claims

1. An active noise control method based on generalized hyperbolic tangent, characterized in that, include: Obtain the primary noise signal generated by the noise source at discrete time nodes, and construct the noise input vector of the filter; The filter's weight coefficient vector is initialized, and the noise input vector is filtered based on the weight coefficient vector to generate an anti-noise output signal, which is then played through a secondary speaker. The residual signal at the noise cancellation point is acquired by an error microphone. The residual signal is then subjected to a nonlinear transformation based on a preset generalized hyperbolic tangent robust cost function to obtain a nonlinear error signal. The robust cost function includes three adjustable parameters: a saturation level control parameter, a transient characteristic adjustment parameter, and a small error penalty intensity adjustment parameter. The weight coefficient vector is updated using the nonlinear error signal to obtain the weight coefficient vector for the next discrete time node; Repeat the above steps iteratively until active noise control is complete.

2. The active noise control method based on generalized hyperbolic tangent according to claim 1, characterized in that, The step of acquiring the primary noise signal generated by the noise source at discrete time nodes and constructing the noise input vector of the filter includes: Discrete time nodes are picked up using a reference microphone. Primary noise value Select discrete time nodes and before At that moment Primary noise value Construct the noise input vector , superscript For transpose operation, The number of taps in the filter is denoted as .

3. The active noise control method based on generalized hyperbolic tangent according to claim 1, characterized in that, The initialization of the filter's weight coefficient vector includes: At discrete time nodes Extract the current time Individual weight coefficient Construct the weight coefficient vector , ; When discrete time nodes At that time, the weight coefficient vector Set all elements to 0.

4. The active noise control method based on generalized hyperbolic tangent according to claim 2, characterized in that, The noise input vector is filtered based on the weight coefficient vector to generate an anti-noise output signal, including: Calculate the weight coefficient vector With the noise input vector The inner product is then compared with the impulse response of the secondary path. Perform convolution operation to obtain the anti-noise output signal. , ,symbol This is a convolution operation.

5. The active noise control method based on generalized hyperbolic tangent according to claim 1, characterized in that, The formula for the generalized hyperbolic tangent robust cost function is as follows: in, For residual signals, This is the saturation level control parameter. Adjust parameters for transient characteristics. This is the parameter for adjusting the penalty intensity for small errors.

6. The active noise control method based on generalized hyperbolic tangent according to claim 5, characterized in that, The parameters For positive real numbers, when At that time, the robust cost function degenerates into It still maintains robust suppression characteristics against large-amplitude residual signals.

7. The active noise control method based on generalized hyperbolic tangent according to claim 6, characterized in that, The residual signal is nonlinearly transformed based on a preset generalized hyperbolic tangent robust cost function to obtain a nonlinear error signal, including: Based on the robust cost function, the residual signal The nonlinear error signal is derived by taking the derivative and using the gradient descent method. Its formula is: in, This is the hyperbolic secant operation.

8. The active noise control method based on generalized hyperbolic tangent according to claim 7, characterized in that, The weight coefficient vector is updated using the nonlinear error signal to obtain the weight coefficient vector for the next discrete time node, including: The nonlinear error signal Compared with the reference signal vector filtered by the secondary path estimation model Multiply, then multiply the product by the step factor. After multiplication, the weight coefficient vector at the current discrete time node is obtained. By superimposing these values, we obtain the weight coefficient vector for the next discrete time node. , Reference signal vector The noise input vector via secondary path estimation model Filtered signal, step size factor This is an adjustable parameter used to adjust the update rate of the weight coefficient vector.

9. An active noise control system based on generalized hyperbolic tangent, characterized in that, It includes a reference microphone, an adaptive filter, a secondary speaker, an error microphone, a secondary path estimation module, and a weight update module; The reference microphone is used to pick up the primary noise signal generated by the noise source and to construct the noise input vector of the adaptive filter; The adaptive filter has a built-in 128th-order FIR filter unit, which is used to initialize the weight coefficient vector and filter the noise input vector based on the weight coefficient vector to generate an anti-noise output signal. The secondary loudspeaker is used to play the anti-noise output signal, which forms a destructive interference with the primary noise; The error microphone is used to collect the residual signal at the noise cancellation point and transmit it to the weight update module; The secondary path estimation module is used to filter the noise input vector to obtain a reference signal vector and transmit it to the weight update module. The weight update module incorporates the generalized hyperbolic tangent robust cost function, which is used to perform nonlinear transformation on the residual signal to obtain a nonlinear error signal, and to iteratively update the weight coefficient vector of the adaptive filter based on the nonlinear error signal and the reference signal vector.

Images