Virtual sensing active noise control algorithm based on grey wolf algorithm optimization and VBAP interpolation

Abstract

The invention discloses a virtual sensing active noise control algorithm based on grey wolf algorithm optimization and VBAP interpolation. The method comprises the following steps: 1, measuring and recording the distance between a virtual sensor and a physical sensor; step 2, calculating the weighted weight of the physical sensor by adopting a VBAP (Video Broadband Access Point) method; 3, obtaining a sound pressure signal by using the physical sensor, weighting the sound pressure signal of the physical sensor by using the calculated weight, and estimating to obtain a sound pressure signal at the virtual sensor; and step 4, based on the estimated sound pressure signal at the virtual sensor, optimizing the coefficient of the control filter by adopting a grey wolf algorithm so as to realize active noise control. The method has no requirements on the number and layout of the physical sensors, can effectively estimate the sound pressure, can be applied to any active noise control algorithm, is robust to the characteristic and azimuth change of the noise source position, and is robust to the change of the secondary path, such as the position change of the secondary source, the position change of the virtual sensor and the like.

Description

Technical Field

[0001] This invention belongs to the field of active noise control in the field of speech signal processing, specifically involving a virtual sensing active noise control algorithm based on Grey Wolf algorithm optimization and VBAP interpolation. Background Technology

[0002] Active noise control algorithms, as an effective means of suppressing low-frequency noise, have been widely used in various fields such as headphones, automobiles, and industrial noise reduction. The Filtered-input Least Mean Square (FxLMS) algorithm is widely adopted due to its simple structure, convenient implementation, and good controllability. Although the FxLMS algorithm has demonstrated strong robustness and practicality in engineering practice, it also has some significant limitations: First, the algorithm relies heavily on the modeling of secondary paths; large modeling errors or significant changes in system dynamics can easily lead to decreased control performance or even system instability. Second, because it is essentially a gradient descent algorithm, it is prone to degraded control performance and getting trapped in local optima, especially performing poorly in complex noise fields or nonlinear environments. Furthermore, the step size limit set to improve algorithm stability also restricts its response speed and noise reduction depth to some extent. Traditional active noise control systems typically aim to minimize the sound pressure level at the error microphone, thereby creating a limited quiet zone around it. However, in some practical applications, due to physical constraints or wearing requirements, it is difficult to deploy error sensors in the desired quiet area. To this end, virtual sensing technology is used to estimate the error sound pressure signal at the target location, thereby enabling flexible and portable quiet zone construction.

[0003] Existing virtual sensing technologies can be mainly divided into two categories: The first category is based on observation filters. This method establishes a transfer function between a physical microphone and a virtual microphone, and trains the observation filter offline to achieve sound field mapping. Typical methods include remote microphone technology and adaptive minimum mean square error virtual sensing technology. The second category is based on auxiliary filters. This method places the physical microphone at a preset position of the virtual microphone in the initial stage of the system and records the path characteristics of the reference signal and control signal to infer the sound pressure response at that position. Once training is complete, the temporary microphone is removed, and the controller operates according to a fixed auxiliary filter. Although the above two methods can achieve virtual observation to a certain extent, their performance is highly dependent on the accuracy of the modeling of the main noise field and the control path. When the system environment changes (such as changes in the position of the noise source or speaker), the performance of the observation filter or auxiliary filter may degrade significantly, resulting in reduced noise reduction effect and insufficient system robustness. Summary of the Invention

[0004] This invention proposes a virtual sensing active noise control algorithm based on the Grey Wolf algorithm optimization and Vector Base Amplitude Panning (VBAP) interpolation, and applies it to an active noise control system. This method does not require pre-constructing a sound pressure mapping model between physical and virtual sensors, exhibits good adaptability to the directionality and characteristics of noise sources, and has no strict limitations on the sensor layout and number, demonstrating strong flexibility and robustness. The method includes the following main steps: First, the spatial distance information between the virtual and physical sensors is recorded to provide a geometric basis for subsequent weight calculation; second, the weighting coefficients of the physical sensors are calculated using the VBAP interpolation method. These weights depend only on the sensor's geometric layout and do not require frequency domain modeling; third, the sound pressure signals from the physical sensors are acquired and weighted according to the weighting coefficients to estimate the sound pressure signal at the virtual sensor; finally, using the estimated sound pressure signal at the virtual sensor as the control objective, an optimization model is constructed, and the coefficients of the active control filter are iteratively updated using the Grey Wolf optimization algorithm to achieve noise minimization control. This invention fully combines the geometric advantages of VBAP interpolation with the global optimization capability of the Grey Wolf algorithm, overcoming the problems of traditional FxLMS-type algorithms such as strong dependence on secondary path modeling and easy getting trapped in local optima. It significantly improves the stability and control performance of the system in dynamic sound fields and is suitable for low-frequency noise suppression application scenarios in complex environments.

[0005] The overall design process is briefly described below:

[0006] The first step involves recording the distances between virtual and physical sensors. By measuring or setting parameters, the distances or coordinate relationships between each virtual sensor and its adjacent physical sensors are obtained, providing fundamental geometric information for subsequent interpolation calculations. The second step calculates the weighting coefficients of the physical sensors using the VBAP method. Based on the spatial distribution of the physical sensors, the VBAP interpolation algorithm determines the weighting of each physical sensor in the virtual location sound pressure estimation. The third step uses the sound pressure signals acquired by the physical sensors and applies the calculated weighting to estimate the sound pressure signal at the virtual sensor location. This step does not require modeling the main noise field or system path and allows for real-time estimation of the error signal at the target location in a dynamic sound field environment. The fourth step uses the estimated sound pressure signal at the virtual sensor location to control and update the filter coefficients using the Grey Wolf algorithm. By constructing an optimization problem with error minimization as the objective function, the Grey Wolf algorithm performs a global search in the parameter space to obtain the optimal filter coefficients, thereby achieving stable and effective active noise suppression.

[0007] The technical solution of this invention is to solve the problem of active noise control in virtual positioning, and mainly consists of the following steps:

[0008] Step 1: Measure and record the distance between the virtual sensor and the physical sensor.

[0009] Step 2: Calculate the weighted weights of the physical sensors using the VBAP method.

[0010] Step 3: Use physical sensors to acquire sound pressure signals, and use the weighting weights calculated in Step 2 to weight the sound pressure signals of the physical sensors to estimate the sound pressure signals at the virtual sensors.

[0011] Step 4: Based on the sound pressure signal at the virtual sensor estimated in Step 3, the filter coefficients are updated using the Grey Wolf algorithm.

[0012] 1. The implementation method of step 1 is as follows: In this embodiment, the spatial layout operation of physical sensors is first performed. Specifically, M physical sensors are arranged in a geometric distribution within the target sound field area to be estimated, and their three-dimensional spatial positions are represented by vectors. The aforementioned sensors can be installed according to a preset geometric structure (e.g., coplanar array, regular polygon, or irregular but uniformly distributed arrangement), and their spatial distribution should meet the requirements of interpolation coverage, preferably forming a three-dimensional convex hull structure. After the physical sensor deployment is completed, a virtual measurement point is randomly selected within the geometric area formed by the M physical sensors, based on the actual application scenario requirements. Its corresponding spatial position vector is denoted as... The virtual measurement point does not need to coincide with the location of any physical sensor. However, it must be located inside the geometric convex hull formed by the M physical sensors to ensure the solvability and non-negativity of the interpolation coefficients during subsequent interpolation calculations. This step allows for the initial setting of the spatial positions of the physical sensor array and the virtual sensor, providing the necessary geometric information for subsequent virtual error signal estimation based on the VBAP algorithm.

[0013] 2. Step 2 is implemented as follows: The system selects the vector basis triplet or multi-element from all M physical sensors that is closest to the physical sensor position v. Assume k microphones are selected, k ≤ M, and the unit vector of their position is denoted as... Each unit vector is obtained by normalizing the position vector of the corresponding sensor, i.e. Arrange the above unit vectors column-wise to form a column vector matrix: Then, based on the basic principles of VBAP interpolation, the goal is to solve for the weight vector while satisfying the constraints. v = L·g, ||g|| = 1, g i≥0. To achieve the above solution process, the least squares solution method based on linear algebra can be adopted. Preferably, the matrix pseudo-inverse or constrained convex optimization method is used to calculate g. When the above conditions are met, the obtained vector g is the weighted weight of the sound pressure contribution of the currently selected k physical sensors to the virtual position, which can be used for subsequent signal interpolation calculation.

[0014] 3. Step 3 is implemented as follows: M sound pressures are measured by M physical sensors, denoted as e1,…,e M Select k physical sensors used to construct the vector basis in the preceding steps, and combine them with the weighted vector g = [g1, g2, ..., g] obtained by the VBAP interpolation method. k ] T According to the formula The sound pressure signal e0 at the virtual sensor v is estimated using a weighted average. The above formula achieves interpolation reconstruction of the sound pressure at the virtual measurement point. The weight g used satisfies normalization and non-negativity constraints, thus ensuring the validity and stability of the estimated value in a physical sense. This virtual sensor signal e0 can be used for subsequent updates to the filter coefficients of the active noise controller. Through this step, the system can obtain virtual error signals at any point inside the sound field without actually installing physical sensors, thereby improving the spatial flexibility and deployment freedom of the active noise control system.

[0015] 4. The implementation of step 4 is as follows: The main function of the adaptive filter is to dynamically adjust the control signal according to the noise signal, so that after the control signal is emitted by the loudspeaker and propagated through the secondary path, it effectively interferes with the original noise signal at the target position, thereby achieving noise cancellation. Its performance directly affects the stability and effectiveness of the entire noise reduction system. The adaptive filter design of this invention does not rely on the traditional gradient descent mechanism, but adopts an optimization update strategy oriented towards multiple candidate filters. In each iteration, the performance of multiple candidate filter coefficients is evaluated in parallel, and the candidate filter with the best performance is selected. The coefficient vector of this filter is determined as the final design result of the adaptive filter. In this embodiment, the number of candidate filters is set to p, and these p candidate filters form a search space. Each candidate filter corresponds to a weight coefficient vector, which represents the position of the candidate filter in the search space. The order N of the filter corresponds to the dimension of the position vector. Therefore, the entire search space can be modeled as p points distributed in an N-dimensional real space. In the initial stage, the system completes the initialization process by randomly generating p initial filter coefficient vectors in the search space:

[0016]

[0017] Each vector It is a potential optimal solution, and represents the coefficient vector of one of the p adaptive filters.

[0018] During system operation, adaptive filters are selected sequentially at intervals of λ samples. At the start of system operation, W1 is selected first, and the noise signal x(n) is input into the signal buffer: x = (x(n), x(n-1), ..., x(n-N+1)). The noise signal x is filtered by W1 to generate the control signal y(n), which can be expressed as:

[0019] y(n)=W1·x T #(2)

[0020] The control signal y(n) is input into the control signal buffer to obtain y = (y(n), y(n-1), ..., y(n-L+1)), where L is the length of the secondary path. The control signal y is filtered by the secondary path S and superimposed with the signal d(n) collected by the physical sensor to obtain the error signal e(n):

[0021] y(n)=S·y T #(3)

[0022] e(n)=d(n)-y(n)#(4)

[0023] in, This represents the acoustic path between the loudspeaker and the physical sensor.

[0024] The signal d(n) reaching the physical sensor is obtained by filtering the noise signal x through the main path P:

[0025] d(n)=P·x T #(5)

[0026] in, This represents the acoustic path between the noise signal and the physical sensor.

[0027] e(n) is stored in the processor. When the next noise signal sample x(n+1) is input into the noise signal buffer, the above steps are repeated to obtain the next error signal e(n+1). This process is repeated λ times to obtain λ error signals, e(n), e(n+1), ..., e(n+λ-1). These λ sampling points are used to calculate the appropriate value of W1. The appropriate value measures the noise reduction performance of each candidate filter coefficient vector under the current sound field conditions and is the basis for the optimization algorithm to rank and update the positions of candidate filters. Then, W2, ..., W... are selected. p Repeat the above calculation steps.

[0028] In this optimization algorithm, a single error signal cannot provide sufficient information to update the filter coefficients. In practical applications, the optimal value is generally defined as the sum of the mean square error signals of a data block's sampling points:

[0029]

[0030] Using the sampling points of the pλ error signals obtained above, calculate the fitness value for each adaptive filter:

[0031]

[0032] Arrange the obtained fitness values ​​in descending order, and select the three candidate filters with the smallest fitness values. The corresponding adaptive filter coefficients are W. α W β W δ These three filters are designated as α (optimal), β (second best), and δ (third best) candidate filters, respectively. Each of these three acts as a leader in the algorithm, guiding the search direction of the remaining candidate filters. This role-sharing mechanism constitutes the core leadership control structure of the algorithm. By simulating the hierarchical behavior of a wolf pack in nature, it further controls the update direction and scope during the search process, achieving global optimization of the filter parameters.

[0033] Calculate W for each filter i With W α W β W δ Distance between:

[0034] D α =|C1·W α -W i |#(8)

[0035] D β =|C2·W β -W i |#(9)

[0036] D δ =|C3·W δ -W i |#(10)

[0037] Based on W α W β W δ Updated candidate solutions:

[0038] W i1 =W α -A1·D α #(11)

[0039] W i2 =W β -A2·D β #(12)

[0040] W i3 =W δ -A3·Dδ #(13)

[0041] Where: A1 = 2a·r1-a, C1 = 2r2, A2 = 2a·r1-a, C2 = 2r2, A3 = 2a·r1-a, C3 = 2r2. During the iteration process, a decreases linearly from 2 to 0, and r1 and r2 are N-dimensional random vectors in the range [0,1].

[0042] The filter W is updated using three candidate solutions. i coefficient:

[0043]

[0044] Following the iterative process described above, the iteration process stops when all noise signal sampling points have been input, and the filter coefficient W with the smallest appropriate value is selected. α These serve as control filter coefficients in an active noise control system. They are used for real-time control signal generation.

[0045] Beneficial effects

[0046] This invention proposes an active noise control method based on the Grey Wolf algorithm optimization and VBAP interpolation virtual sensing. The method utilizes vector basis amplitude omnidirectional interpolation technology to effectively estimate sound pressure signals at arbitrary spatial locations using arbitrarily distributed physical sensors, without strict limitations on the number and geometric layout of sensors, thus constructing a highly flexible virtual sensor structure. In this invention, the estimated virtual error signal is used in the controller update process of the active noise control system. Compared to traditional filter update methods based on the FxLMS algorithm, this invention introduces the Grey Wolf optimization algorithm to perform global search and dynamic update of the control filter parameters, overcoming the shortcomings of traditional algorithms such as reliance on secondary path modeling, sensitivity to changes in the sound field environment, and susceptibility to local optima. This method exhibits strong robustness to changes in the spatial orientation of noise sources and can adapt to complex application scenarios such as changes in the location of secondary sound sources, adjustments to the virtual sensor positions, and non-ideal sensor arrangements, significantly improving the adaptability, stability, and noise reduction effect of the active noise control system in real-world complex environments. Attached Figure Description

[0047] Figure 1 This is an algorithm block diagram of the control method of the present invention.

[0048] Figure 2 This is the flowchart of the Grey Wolf algorithm. Detailed Implementation

[0049] In this embodiment, to verify the effectiveness of the virtual sensing active noise control method based on the Grey Wolf Algorithm optimization and VBAP interpolation described in this invention, the following acoustic simulation environment and control system are constructed: The experimental environment is a closed free-field approximation space, in which a virtual sensor point is set as the target location for active noise control. Noise must be minimized at this virtual location to construct the desired quiet area. One noise source is set, arranged on the same horizontal plane as the virtual sensor, 1.5 meters away, to simulate external noise interference. A secondary sound source (control loudspeaker) is configured in the control system, also located on the same horizontal plane, 1 meter away from the virtual sensor, to generate a control signal that cancels out the primary noise. Several physical sensors (i.e., actual microphones) are used to collect sound pressure signals, installed at a height of 1 meter above the ground, and on the same horizontal plane as the sound source and secondary sound source. The microphones are arranged in a circular, uniform pattern around the virtual sensor, with the virtual sensor located at the geometric center of this circular array. The circular arrangement facilitates balanced spatial interpolation and improves the accuracy of sound pressure estimation at the virtual location. In terms of algorithm implementation, the VBAP method is used to calculate interpolation weights based on the geometric arrangement of the microphone array. The estimated virtual error signal is used to construct the optimization objective, and then the Grey Wolf optimization algorithm is used to iteratively update the control filter coefficients to minimize the estimation error energy. This experiment uses MATLAB 2023a as the signal processing and algorithm simulation platform, setting the noise signal sampling rate to 44.1kHz, the control filter order to 64, the Grey Wolf population size to 20, and the maximum number of iterations to 100. All simulated acoustic paths are implemented using digital filters, and the control signal is applied to the speaker port after being modeled and filtered by secondary paths. Simulation results show that, under this configuration, the method described in this invention can effectively estimate the sound pressure signal at the virtual location, and achieves stable and fast filter coefficient convergence through Grey Wolf algorithm optimization. It significantly suppresses sound pressure interference from the main noise source at the target virtual location, verifying the practicality and superior performance of this invention in virtual sensing and control scenarios.

[0050] In implementation, this invention embeds the algorithm into the software to automate the execution of each process. The following is a further explanation of the invention with reference to the accompanying drawings and specific implementation steps: The specific workflow is as follows:

[0051] Step 1: First, the physical sensors are deployed spatially. Specifically, M physical sensors are arranged geometrically within the target sound field area to be estimated, and their three-dimensional spatial positions are represented as vectors. The aforementioned sensors can be installed according to a preset geometric structure (e.g., coplanar array, regular polygon, or irregular but uniformly distributed arrangement), and their spatial distribution should meet the requirements of interpolation coverage, preferably forming a three-dimensional convex hull structure. After the physical sensor deployment is completed, a virtual measurement point is randomly selected within the geometric area formed by the M physical sensors, based on the actual application scenario requirements. Its corresponding spatial position vector is denoted as... The virtual measurement point does not need to coincide with the location of any physical sensor. However, it must be located inside the geometric convex hull formed by the M physical sensors to ensure the solvability and non-negativity of the interpolation coefficients during subsequent interpolation calculations. This step allows for the initial setting of the spatial positions of the physical sensor array and the virtual sensor, providing the necessary geometric information for subsequent virtual error signal estimation based on the VBAP algorithm.

[0052] Step 2: The system selects the vector basis triplet or multi-element from all M physical sensors that is closest to the physical sensor's position v. Assuming k microphones are selected, k ≤ M, the unit vector of its position is denoted as... Each unit vector is obtained by normalizing the position vector of the corresponding sensor, i.e. Arrange the above unit vectors column-wise to form a column vector matrix: Then, based on the basic principles of VBAP interpolation, the goal is to solve for the weight vector g∈R while satisfying the constraints. k : v = L·g, ||g|| = 1, g i ≥0. To achieve the above solution process, the least squares solution method based on linear algebra can be adopted. Preferably, the matrix pseudo-inverse or constrained convex optimization method is used to calculate g. When the above conditions are met, the obtained vector g is the weighted weight of the sound pressure contribution of the currently selected k physical sensors to the virtual position, which can be used for subsequent signal interpolation calculation.

[0053] Step 3: The M sound pressures measured by the M physical sensors are denoted as e1,…,e M Select k physical sensors used to construct the vector basis in the preceding steps, and combine them with the weighted vector g = [g1, g2, ..., g] obtained by the VBAP interpolation method. k ] T According to the formula The sound pressure signal e0 at the virtual sensor v is estimated using a weighted average. The above formula achieves interpolation reconstruction of the sound pressure at the virtual measurement point. The weight g used satisfies normalization and non-negativity constraints, thus ensuring the validity and stability of the estimated value in a physical sense. This virtual sensor signal e0 can be used for subsequent updates to the filter coefficients of the active noise controller. Through this step, the system can obtain virtual error signals at any point inside the sound field without actually installing physical sensors, thereby improving the spatial flexibility and deployment freedom of the active noise control system.

[0054] Step 4: The main function of the adaptive filter is to dynamically adjust the control signal based on the noise signal. This control signal, after being emitted by the loudspeaker and propagated through a secondary path, effectively interferes with the original noise signal at the target location, thereby achieving noise cancellation. Its performance directly affects the stability and effectiveness of the entire noise reduction system. The adaptive filter design of this invention does not rely on the traditional gradient descent mechanism. Instead, it adopts an optimization and update strategy oriented towards multiple candidate filters. In each iteration, the performance of multiple candidate filter coefficients is evaluated in parallel, and the candidate filter with the best performance is selected. The coefficient vector of this filter is determined as the final design result of the adaptive filter. In this embodiment, the number of candidate filters is set to p, and these p candidate filters form a search space. Each candidate filter corresponds to a weight coefficient vector, representing the position of the candidate filter in the search space. The order N of the filter corresponds to the dimension of the position vector. Therefore, the entire search space can be modeled as p points distributed in an N-dimensional real space. In the initial stage, the system completes the initialization process by randomly generating p initial filter coefficient vectors in this search space.

[0055]

[0056] Each vector It is a potential optimal solution, and represents the coefficient vector of one of the p adaptive filters.

[0057] During system operation, adaptive filters are selected sequentially at intervals of λ samples. At the start of system operation, W1 is selected first, and the noise signal x(n) is input into the signal buffer: x = (x(n), x(n-1), ..., x(n-N+1)). The noise signal x is filtered by W1 to generate the control signal y(n), which can be expressed as:

[0058] y(n)=W1·x T #(2)

[0059] The control signal y(n) is input into the control signal buffer to obtain y = (y(n), y(n-1), ..., y(n-L+1)), where L is the length of the secondary path. The control signal y is filtered by the secondary path S and superimposed with the signal d(n) collected by the physical sensor to obtain the error signal e(n):

[0060] y(n)=S·y T #(3)

[0061] e(n)=d(n)-y(n)#(4)

[0062] in, This represents the acoustic path between the loudspeaker and the physical sensor.

[0063] The signal d(n) reaching the physical sensor is obtained by filtering the noise signal x through the main path P:

[0064] d(n)=P·x T #(5)

[0065] in, This represents the acoustic path between the noise signal and the physical sensor.

[0066] e(n) is stored in the processor. When the next noise signal sample x(n+1) is input into the noise signal buffer, the above steps are repeated to obtain the next error signal e(n+1). This process is repeated λ times to obtain λ error signals, e(n), e(n+1), ..., e(n+λ-1). These λ sampling points are used to calculate the appropriate value of W1. The appropriate value measures the noise reduction performance of each candidate filter coefficient vector under the current sound field conditions and is the basis for the optimization algorithm to rank and update the positions of candidate filters. Then, W2, ..., W... are selected. p Repeat the above calculation steps.

[0067] In this optimization algorithm, a single error signal cannot provide sufficient information to update the filter coefficients. In practical applications, the optimal value is generally defined as the sum of the mean square error signals of a data block's sampling points:

[0068]

[0069] Using the sampling points of the pλ error signals obtained above, calculate the fitness value for each adaptive filter:

[0070]

[0071] Arrange the obtained fitness values ​​in descending order, and select the three candidate filters with the smallest fitness values. The corresponding adaptive filter coefficients are W. α W β Wδ These three filters are designated as α (optimal), β (second best), and δ (third best) candidate filters, respectively. Each of these three acts as a leader in the algorithm, guiding the search direction of the remaining candidate filters. This role-sharing mechanism constitutes the core leadership control structure of the algorithm. By simulating the hierarchical behavior of a wolf pack in nature, it further controls the update direction and scope during the search process, achieving global optimization of the filter parameters.

[0072] Calculate W for each filter i With W α W β W δ Distance between:

[0073] D α =|C1·W α -W i |#(8)

[0074] D β =|C2·W β -W i |#(9)

[0075] D δ =|C3·W δ -W i |#(10)

[0076] Based on W α W β W δ Updated candidate solutions:

[0077] W i1 =W α -A1·D α #(11)

[0078] W i2 =W β -A2·D β #(12)

[0079] W i3 =W δ -A3·D δ #(13)

[0080] Where: A1 = 2a·r1-a, C1 = 2r2, A2 = 2a·r1-a, C2 = 2r2, A3 = 2a·r1-a, C3 = 2r2. During the iteration process, a decreases linearly from 2 to 0, and r1 and r2 are N-dimensional random vectors in the range [0,1].

[0081] The filter W is updated using three candidate solutions. i coefficient:

[0082]

[0083] Following the iterative process described above, the iteration process stops when all noise signal sampling points have been input, and the filter coefficient W with the smallest appropriate value is selected. α These serve as control filter coefficients in an active noise control system. They are used for real-time control signal generation.

[0084] Beneficial effects

[0085] The specific embodiments described herein are merely illustrative of the spirit of the invention. Those skilled in the art to which this invention pertains may make various modifications or additions to the described specific embodiments or use similar methods to substitute them, without departing from the spirit of the invention or exceeding the scope defined by the appended claims.

Claims

1. A virtual sensing active noise control algorithm based on gray wolf algorithm optimization and VBAP interpolation, characterized in that... Includes the following steps: Step 1: Measure and record the distance between the virtual sensor and the physical sensor; Step 2: Calculate the weights of the physical sensors using the VBAP method; Step 3: Use physical sensors to acquire sound pressure signals, and use the weighting weights calculated in Step 2 to weight the sound pressure signals of the physical sensors to estimate the sound pressure signals at the virtual sensors. Step 4: Based on the sound pressure signal at the virtual sensor estimated in Step 3, the filter coefficients are updated using the Grey Wolf algorithm.

2. The virtual sensing active noise control algorithm based on VBAP interpolation as described in claim 1, characterized in that: Step 2 is described in detail below: The system selects the vector basis triplet or multi-element from all M physical sensors that is closest to the physical sensor's position v. Assuming k microphones are selected, k ≤ M, the unit vector of their positions is denoted as... Each unit vector is obtained by normalizing the position vector of the corresponding sensor, i.e. ||m i || represents vector m i The modulus; arrange the above unit vectors column-wise to form a column vector matrix: Then, based on the basic principles of VBAP interpolation, the goal is to solve for the weight vector while satisfying the constraints. v=L·g,‖g‖=1,g i ≥0. To achieve the above solution process, the least squares solution method based on linear algebra is adopted. The matrix pseudo-inverse or constrained convex optimization method is used to calculate g. The obtained vector g is the weighted weight of the sound pressure contribution of the currently selected k physical sensors to the virtual position.

3. The virtual sensing active noise control algorithm based on VBAP interpolation as described in claim 1, characterized in that: Step 4 is as follows: The number of candidate filters is set to p, and these p candidate filters form a search space. Each candidate filter corresponds to a weight coefficient vector, representing the position of the candidate filter in the search space. The order N of the filter corresponds to the dimension of the position vector. Therefore, the entire search space can be modeled as p points distributed in an N-dimensional real space. In the initial stage, the system completes the initialization process by randomly generating p initial filter coefficient vectors in the search space. Each vector It is a potential optimal solution, and represents the coefficient vector of one of the p adaptive filters; Adaptive filters are selected sequentially with an interval of λ samples. First, W1 is selected, and the noise signal x(n) is input into the signal buffer: x = (x(n), x(n-1), ..., x(n-N+1)). The noise signal x is filtered by W1 to generate the control signal y(n), expressed as: y(n)=W1·x T #(2) The control signal y(n) is input into the control signal buffer to obtain y = (y(n), y(n-1), ..., y(n-L+1)), where L is the length of the secondary path; the control signal y is filtered by the secondary path S and superimposed with the signal d(n) collected by the physical sensor to obtain the error signal e(n): y(n)=S·y T #(3) e(n)=d(n)-y(n)#(4) in, This represents the acoustic path between the loudspeaker and the physical sensor. The signal d(n) reaching the physical sensor is obtained by filtering the noise signal x through the main path P: d(n)=P·x T #(5) in, V represents the length of the main path; it represents the acoustic path between the noise signal and the physical sensor. e(n) is stored in the processor. When the next noise signal sample x(n+1) is input into the noise signal buffer, the above steps are repeated to obtain the next error signal e(n+1). λ sampling points are calculated to obtain λ error signals, e(n), e(n+1), ..., e(n+λ-1). These λ sampling points will be used to calculate the appropriate value of W1; then W2, ..., W... p Repeat the above calculation steps; The appropriate value is defined as the sum of the mean square error signals of the sampling points of a data block: Using the sampling points of the pλ error signals obtained above, calculate the fitness value for each adaptive filter: Arrange the obtained fitness values ​​in descending order, and select the three candidate filters with the smallest fitness values. The corresponding adaptive filter coefficients are W. α W β W δ These three filters are respectively considered as α (optimal), β (second best), and δ (third best) candidate filters, and serve as leaders in the algorithm to guide the search direction of the remaining candidate filters. Calculate W for each filter i With W α W β W δ Distance between: D α =|C1·W α -W i |#(8) D β =|C2·W β -IN i |#(9) D δ =|C3·W δ -IN i |#(10) Based on W α W β W δ Updated candidate solutions: W i1 =W α -A1·D α #(11) W i2 =W β -A2·D β #(12) IN i3 =In δ -A3 D δ #(13) Where: A1=2a·r1-a,C1=2r2,A2=2a·r1-a,C2=2r2,A3=2a·r1-a,C3=2r2;During the iteration process, a decreases linearly from 2 to 0, and r1 and r2 are random vectors of dimension N in the range [0,1]. The filter W is updated using three candidate solutions. i coefficient: Following the iterative process described above, the iteration process stops when all noise signal sampling points have been input, and the filter coefficient W with the smallest appropriate value is selected. α As control filter coefficients in an active noise control system, they are used for real-time control signal generation.

Images