Sparse sound source identification method and system for microphone array measurement

Abstract

The present invention provides a sparse sound source identification method for microphone array measurement. First, a discrete Fourier transform is used to convert a signal into a frequency domain to form a frequency domain sound pressure matrix, and a signal component is selected in the frequency domain sound pressure matrix to form a sound pressure vector. Then, a sound source reconstruction area is selected according to a preset rule, and the sound source reconstruction area is divided into discrete grids. A sound propagation model is established according to the position coordinates of array elements in the microphone array and the sound pressure vector. Then, an acoustic inverse problem based on a generalized minimum-maximum concave function is established according to the sound propagation model, an auxiliary matrix is ​​introduced to deform the acoustic inverse problem to obtain a deformed expression, and the deformed expression is solved in an alternating splitting manner to obtain a target sound source intensity to be obtained, and a sound image is formed according to the target sound source intensity to complete sound source identification. In this way, it is convenient to solve the acoustic inverse problem, improve calculation efficiency, and improve identification accuracy.

Description

Technical Field

[0001] The present invention relates to the technical field of array signal processing, and more particularly to a sparse sound source identification method and system for microphone array measurement. Background Art

[0002] A microphone array consists of multiple microphones arranged in a specific spatial distribution. Compared to measurements using independent microphones, array measurements involve synchronous signal acquisition across multiple channels. Due to the disparity in array element positions, the acoustic signals collected by different microphones experience a certain phase delay. Therefore, the differences between channels can reflect the spatial information of the sound source. The human ear can detect the presence of a sound source by hearing it, but it is difficult to accurately estimate the exact location and intensity of a sound source through hearing alone. Therefore, quantitative identification of a sound source requires assistance from other means. Through array signal processing technology, microphone arrays can obtain multi-dimensional characteristics of a sound source, including temporal and spatial domains, transforming audible sound into a visible image, thereby enabling sound source identification. Microphone array-based sound source identification can reconstruct the distribution of sound sources within a specific area and identify the presence and nature of abnormal sound sources. It has broad application prospects in the automotive industry, aerospace, and machinery health monitoring.

[0003] Sparse sound sources are widely present in modern production and life scenes. For example, abnormal noise sources of cars mostly appear in local areas such as the center console and the edge of the window, and the impact noise source caused by mechanical failure only exists near the moving pair. The distribution of these sound sources in the spatial domain is sparse, that is, there are only a small number of non-zero elements in the sound source identification results. If conventional beamforming (Williams J R. Fast Beam-Forming Algorithm [J]. The Journal of the Acoustical Society of America, 1968, 44 (5): 1454-1455.) and adaptive beamforming (CN201610409905, CN201911406837) are directly used to carry out sound source identification, the resolution and accuracy of the obtained results are low, which cannot fully meet the needs of sparse sound source identification in practical applications. In the process of sparse sound source identification, spatial sparsity can be used as prior knowledge to provide the constraints required for identification, thereby improving the accuracy of sound source identification. Currently, the basic idea behind sparse sound source identification is to establish a sound propagation model based on the sound source reconstruction area and array layout, and then solve the inverse problem based on the sound propagation model under the constraints of sparse prior knowledge to reconstruct information such as the sound source location and source intensity. The sparse constraint in sound source identification is represented by the zero-norm of the sound source intensity vector. The non-smooth and non-convex nature of the zero-norm makes solving the acoustic inverse problem difficult, making it impossible to directly solve the acoustic inverse problem. Sparse sound source identification methods can be divided into two categories based on the different approximate solutions to the acoustic inverse problem: methods based on orthogonal matching pursuit and methods based on basis pursuit. The orthogonal matching pursuit (OMP) method is a greedy solution that reconstructs the sound source through a step-by-step iterative calculation. At each iteration, the global optimal solution is replaced by a local optimal solution, attempting to approximate the global optimal solution through cumulative updates of the local optimal solution. This method is computationally efficient and time-saving. However, due to limitations in the solution strategy, the local optimal solution often cannot accurately approximate the global optimal solution. This makes it difficult to ensure the accuracy of the sound source intensity vector obtained when the algorithm converges, weakening the reliability of the sparse sound source identification results. Basis pursuit-based methods replace the zero-norm in the original acoustic inverse problem with an approximate function. This approximate function approximates the sparsity constraint of the zero-norm during the solution process, imposing a sparsity penalty on nonzero elements in the solution vector, thereby obtaining the sparsest possible sound source identification results. Furthermore, the approximate function has better solution properties than the zero-norm, such as convexity and smoothness, which facilitates the smooth solution of the optimization problem during sparse recovery. Therefore, this type of method can produce more accurate results than the orthogonal matching pursuit method. Common approximate functions include the 1-norm and the near-zero-norm. However, due to the differences in basic properties between approximate functions and zero-norms, the sparse sound source identification method based on basis pursuit still has shortcomings in terms of result accuracy.The zero-norm of all non-zero elements is the same value, indicating that non-zero elements of different amplitudes are subject to the same sparsity penalty due to the zero-norm. However, in the above-mentioned approximate function regularized acoustic inverse problem solution process, non-zero elements of different amplitudes in the solution vector are subject to different degrees of sparsity penalty. This difference weakens the reliability of existing basis pursuit-based methods in sound source identification, resulting in the problem of underestimation of sound source intensity. When the dynamic range of the solution result is small, sound sources with severely underestimation of intensity will be mistakenly ignored, resulting in the inability of existing methods to effectively obtain information such as sound source location and sound source intensity. In addition, in sparse sound source identification methods based on basis pursuit, the solution of the acoustic inverse problem is mostly performed in an iterative manner, which requires considering the balance between result accuracy and computational efficiency. Therefore, for specific approximate functions, the construction of an acoustic inverse problem solution algorithm becomes one of the difficulties in sparse sound source identification.

[0004] Therefore, there is an urgent need for a sparse sound source identification method and system for microphone array measurement that can improve the approximation effect of the zero norm, enhance the consistency of the penalty degree of non-zero elements in the sparse recovery process, improve the accuracy of sparse sound source identification results, and improve computational efficiency. Summary of the Invention

[0005] In view of the above problems, the purpose of the present invention is to provide a sparse sound source identification method for microphone array measurement to solve the problem that the sound source intensity is under-estimated in the prior art. When the dynamic range of the solution result is small, the sound source with severely under-estimated intensity will be mistakenly ignored, resulting in the existing method being unable to effectively obtain information such as the sound source position and sound source intensity.

[0006] The present invention provides a sparse sound source identification method for microphone array measurement, comprising:

[0007] The microphone array with M channels is used to collect the data of length L. a The acoustic signal p Im (m=1,2,...,M), to construct a size of M×L a The original sound pressure matrix The acoustic signal p collected by each channel Im Perform segmentation processing to form a signal segment with a length of L and an overlap rate of R, and obtain the signal containing A quick sound pressure matrix in, Indicates not greater than integer, the submatrix corresponding to the s(s=1,2,...,S)th snapshot The size is M×L, express The L(1-R)(s-1)+1th to L(1-R)(s-1)+Lth elements in ;

[0008] The signal corresponding to the s-th snapshot is converted into the frequency domain using discrete Fourier transform to form the frequency domain sound pressure matrix And select the lth (l=1,2,...,L) frequency f in the frequency domain sound pressure matrix l The corresponding signal components are used to form the sound pressure vector p = [p1(f l ),p2(f l ),...,p M (f l )] T ,in, DFT stands for Discrete Fourier Transform; m (f l )represent Intermediate frequency f l The corresponding signal components;

[0009] The sound source reconstruction area is selected according to the preset rules, and the sound source reconstruction area is divided into N=N1×N2 discrete grids, and the position coordinates of the grid points in the discrete grids are u n =(u nx ,u ny ,u nz )(n=1,2,...,N); According to the position coordinates of the array elements in the microphone array and the sound pressure vector p=[p1(f l ),p2(f l ),...,p M (f l )] T Establish the sound propagation model p=Gq+r, where q is the sound source intensity vector to be solved, r is the noise vector, G is the transmission matrix of size M×N, and the elements of the transmission matrix are Where exp() is the exponential function, c is the speed of sound, is the imaginary unit, u m is the position coordinate of the array element in the microphone array, u n are the position coordinates of the grid points in the discrete grid;

[0010] According to the sound propagation model, an acoustic inverse problem based on a generalized minimum-maximum concave function is established, an auxiliary matrix is ​​introduced to deform the acoustic inverse problem to obtain a deformed expression, and the deformed expression is solved by an alternating splitting method to obtain the target sound source intensity to be determined, and a sound image is formed according to the target sound source intensity to complete sound source identification.

[0011] Preferably, in the process of establishing an acoustic inverse problem based on a generalized minimum-maximum concave function according to the sound propagation model, and introducing an auxiliary matrix to deform the acoustic inverse problem to obtain a deformed expression, the process includes:

[0012] According to the sound propagation model, an acoustic inverse problem based on a generalized minimum-maximum concave function is established. Among them, λ>0 is the regularization parameter that needs to be given, is a generalized minimum-maximum concave function, B is an auxiliary matrix; v is an auxiliary vector and is a variable;

[0013] Given a balance parameter 0<γ<1, let the auxiliary matrix The acoustic inverse problem is deformed to determine a deformed expression of the acoustic inverse problem:

[0014]

[0015] The transformed expression is used as the objective function, and the smooth term of the objective function for the variable q is denoted as f1(q), the non-smooth term of the objective function for the variable q is denoted as f2(q), the smooth term of the objective function for the variable v is denoted as g1(v), and the non-smooth term of the objective function for the variable v is denoted as g2(v).

[0016] Preferably, the process of solving the deformed expression in an alternating splitting manner to obtain the target sound source intensity to be determined, and forming a sound image according to the target sound source intensity includes:

[0017] Initialize the counting variable and randomly initialize the intermediate variables and

[0018] As the counting variable is updated, the value of the sound source intensity vector to be calculated is iteratively updated in sequence according to the deformed expression, and the intensity error between the values ​​of the sound source intensity vectors obtained at two adjacent times is calculated;

[0019] determining whether the intensity error exceeds a preset convergence threshold; if so, continuing to iteratively update and calculate the value of the sound source intensity vector to be calculated; and if the intensity error does not exceed the convergence threshold, using the value of the sound source intensity vector calculated last as the target sound source intensity to be calculated;

[0020] A sound image is generated according to the target sound source intensity.

[0021] Preferably, the convergence threshold is 0.0001.

[0022] Preferably, the alternating splitting mode is the Douglas-Rachford splitting mode.

[0023] Preferably, the process of iteratively updating and calculating the value of the sound source intensity vector to be calculated according to the modified expression as the counting variable is updated, and calculating the intensity error between the values ​​of the sound source intensity vectors obtained at two adjacent times, includes:

[0024] Calculate the value of the current sound source intermediate variable and the value of the current sound source intensity vector under the current count variable; where the current sound source intermediate variable is: The current sound source intensity vector is: in, is the pre-calculated proximal operator, x is the function independent variable, and z is the optimization target variable;

[0025] Let k=k+1 in the current sound source intensity vector, calculate the value of the source intensity vector of the next variable of the current counting variable, and calculate the intensity error e between the value of the source intensity vector of the next variable and the value of the current sound source intensity vector R .

[0026] Preferably, the intensity error is

[0027] Preferably, in the process of calculating the value of the current sound source intermediate variable and the value of the current sound source intensity vector under the current count variable,

[0028] Also calculate the value of the current auxiliary vector and the value of the current auxiliary intermediate variable; where,

[0029] The current auxiliary vector is:

[0030] The current auxiliary intermediate variables are:

[0031] in, is the proximal operator of the budget.

[0032] Preferably, if the intensity error does not exceed the convergence threshold, the value of the sound source intensity vector calculated last time is used as the target sound source intensity to be determined; and the process of generating the sound image according to the target sound source intensity includes:

[0033] The value of the sound source intensity vector calculated last time is used as the value of the target sound source intensity to be determined;

[0034] The target sound source intensity is converted into a sound source intensity matrix of size N1×N2 And the sound source intensity matrix Perform logarithmic transformation on each element in to obtain the target sound source matrix in, is a matrix The elements in Qref is the preset reference sound source intensity;

[0035] A sound image is generated according to the target sound source matrix, and sound source identification is performed according to the sound image.

[0036] On the other hand, the present invention further provides a sparse sound source identification system for microphone array measurement, which implements the sparse sound source identification method for microphone array measurement as described above, comprising:

[0037] The sound pressure matrix acquisition unit is used to collect the sound pressure matrix of length L using a microphone array containing M channels. a The acoustic signal p Im (m=1,2,...,M), to construct a size of M×L a The original sound pressure matrix The acoustic signal p collected by each channel Im Perform segmentation processing to form signal segments with a length of L and an overlap rate of R, and obtain the signal segments containing A quick sound pressure matrix in, Indicates not greater than integer, the submatrix corresponding to the s(s=1,2,...,S)th snapshot The size is M×L, express The L(1-R)(s-1)+1th to L(1-R)(s-1)+Lth elements in ;

[0038] The frequency domain conversion unit is used to convert the signal corresponding to the s-th snapshot into the frequency domain using discrete Fourier transform to form a frequency domain sound pressure matrix And select the lth (l=1,2,...,L) frequency f in the frequency domain sound pressure matrix l The corresponding signal components are used to form the sound pressure vector p=[p1(f l ),p2(f l ),...,p M (f l )] T ,in, DFT stands for Discrete Fourier Transform; m (f l )represent The signal component corresponding to the intermediate frequency fl;

[0039] The model building unit is used to select a sound source reconstruction area according to a preset rule, and divide the sound source reconstruction area into N=N1×N2 discrete grids, where the position coordinates of the grid points in the discrete grid are u n =(u nx ,uny ,u nz )(n=1,2,...,N); According to the position coordinates of the array elements in the microphone array and the sound pressure vector p=[p1(f l ),p2(f l ),...,p M (f l )] T Establish the sound propagation model p=Gq+r, where q is the sound source intensity vector to be solved, r is the noise vector, G is the transmission matrix of size M×N, and the elements of the transmission matrix are Where exp() is the exponential function, c is the speed of sound, is the imaginary unit, u m is the position coordinate of the array element in the microphone array, u n are the position coordinates of the grid points in the discrete grid;

[0040] An identification and solving unit is used to establish an acoustic inverse problem based on a generalized minimum-maximum concave function according to the sound propagation model, introduce an auxiliary matrix to deform the acoustic inverse problem to obtain a deformed expression, solve the deformed expression using an alternating splitting method to obtain the target sound source intensity to be determined, and form a sound image according to the target sound source intensity to complete sound source identification.

[0041] From the above technical solutions, it can be seen that the sparse sound source identification method for microphone array measurement provided by the present invention first uses the microphone array to collect sound signals to construct a M×L aThe original sound pressure matrix; the sound signals collected by each channel are segmented to form signal segments, and the sound pressure matrix is ​​obtained according to the original sound pressure matrix, and then the discrete Fourier transform is used to convert the signal to the frequency domain to form a frequency domain sound pressure matrix, and the signal components are selected in the frequency domain sound pressure matrix to form a sound pressure vector, and then the sound source reconstruction area is selected according to the preset rules, and the sound source reconstruction area is divided into discrete grids, and a sound propagation model is established according to the position coordinates of the array elements in the microphone array and the sound pressure vector, and then an acoustic inverse problem based on the generalized minimum and maximum concave function is established according to the sound propagation model, and an auxiliary matrix is ​​introduced to deform the acoustic inverse problem to obtain a deformed expression, and the deformed expression is solved by alternating splitting to obtain the target sound source intensity to be determined, and a sound image is formed according to the target sound source intensity to complete the sound source. Identification; in this way, a sparse sound source identification method is established using the generalized minimum-maximum concave function as a regularization function. The approximate function used has a better approximation effect on the 0 norm, and the degree of penalty received by non-zero elements in the sparse recovery process is more consistent, which can solve the problem of underestimation of sound source intensity in existing technologies; and the objective function of the acoustic inverse problem is convex, which can ensure the accuracy of the sparse sound source identification results. After introducing auxiliary variables, the acoustic inverse problem is solved by alternating splitting, and an explicit variable iterative update formula is established, which can conveniently solve the acoustic inverse problem and improve computational efficiency; and considering the signal overlapping segmentation strategy and different sound source frequencies to adapt to changes in the external environment, the sparse prior information used in the solution process can suppress the interference caused by measurement noise and improve identification accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] By referring to the following description in conjunction with the accompanying drawings, and with a more complete understanding of the present invention, other objects and results of the present invention will become more apparent and easier to understand. In the accompanying drawings:

[0043] Figure 1 is a flow chart of a sparse sound source identification method for microphone array measurement according to an embodiment of the present invention;

[0044] Figure 2 A sound source identification result obtained by the sparse sound source identification method for microphone array measurement according to an embodiment of the present invention;

[0045] Figure 3 is the root mean square error between the sparse sound source identification method for microphone array measurement according to an embodiment of the present invention and the sound source identification results obtained in the prior art;

[0046] Figure 4 4 is a block diagram of a sparse sound source identification system for microphone array measurement according to an embodiment of the present invention. DETAILED DESCRIPTION

[0047] The zero-norm of all non-zero elements is the same value, indicating that non-zero elements of different amplitudes are subject to the same sparsity penalty due to the zero-norm. However, in the existing approximate function regularized solution to the acoustic inverse problem, non-zero elements of different amplitudes in the solution vector are subject to different degrees of sparsity penalty. This difference weakens the reliability of existing basis pursuit-based methods in sound source identification, resulting in underestimation of sound source intensity. When the dynamic range of the solution is small, sound sources with severely underestimation of intensity will be mistakenly ignored, resulting in the inability of existing methods to effectively obtain information such as sound source location and intensity. In addition, in sparse sound source identification methods based on basis pursuit, the solution to the acoustic inverse problem is mostly performed iteratively, which requires considering the balance between result accuracy and computational efficiency. Therefore, for specific approximate functions, the construction of an acoustic inverse problem solution algorithm becomes one of the difficulties in sparse sound source identification.

[0048] In order to solve the above problems, the present invention provides a sparse sound source identification method for microphone array measurement. The specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0049] In order to illustrate the sparse sound source identification method and system for microphone array measurement provided by the present invention, Figure 1-Figure 3 The sparse sound source identification method for microphone array measurement according to an embodiment of the present invention is exemplarily illustrated; Figure 4 The sparse sound source identification system for microphone array measurement according to an embodiment of the present invention is exemplarily illustrated.

[0050] The following description of exemplary embodiments is merely illustrative in nature and is in no way intended to limit the present invention, its application, or uses. Techniques and devices known to those skilled in the art may not be discussed in detail, but where appropriate, such techniques and devices should be considered part of the specification.

[0051] like Figure 1 、 Figure 2 、 Figure 3 As shown in the figure, the present invention provides a sparse sound source identification method for microphone array measurement, comprising:

[0052] The microphone array with M channels is used to collect the data of length L. a The acoustic signal p Im (m=1,2,...,M), to construct a size of M×L a The original sound pressure matrix The acoustic signal p collected by each channel Im Perform segmentation processing to form signal segments with a length of L and an overlap rate of R, and obtain the signal segments containing A quick sound pressure matrix in, Indicates not greater than integer, the submatrix corresponding to the s(s=1,2,...,S)th snapshot The size is M×L, express The L(1-R)(s-1)+1th to L(1-R)(s-1)+Lth elements in ;

[0053] The signal corresponding to the s-th snapshot is converted into the frequency domain using discrete Fourier transform to form the frequency domain sound pressure matrix And select the lth (l=1,2,...,L) frequency f in the frequency domain sound pressure matrix l The corresponding signal components are used to form the sound pressure vector p=[p1(f l ),p2(f l ),...,p M (f l )] T ,in, DFT stands for Discrete Fourier Transform; m (f l )represent Intermediate frequency f l The corresponding signal components;

[0054] The sound source reconstruction area is selected according to the preset rules, and the sound source reconstruction area is divided into N=N1×N2 discrete grids, and the position coordinates of the grid points in the discrete grids are u n =(u nx ,u ny ,u nz )(n=1,2,...,N); According to the position coordinates of the array elements in the microphone array and the sound pressure vector p=[p1(f l ),p2(f l ),...,p M (f l )] T Establish the sound propagation model p=Gq+r, where q is the sound source intensity vector to be solved, r is the noise vector, G is the transmission matrix of size M×N, and the elements of the transmission matrix are Where exp() is the exponential function, c is the speed of sound, is the imaginary unit, u m is the position coordinate of the array element in the microphone array, u n are the position coordinates of the grid points in the discrete grid;

[0055] According to the sound propagation model, an acoustic inverse problem based on a generalized minimum-maximum concave function is established, an auxiliary matrix is ​​introduced to deform the acoustic inverse problem to obtain a deformed expression, and the deformed expression is solved by an alternating splitting method to obtain the target sound source intensity to be determined, and a sound image is formed according to the target sound source intensity to complete sound source identification.

[0056] exist Figure 1 、 Figure 2 、 Figure 3 In the embodiment shown in FIG. 1 , step S1 is to use a microphone array including M channels to collect a signal of length L. a The acoustic signal p Im (m=1,2,...,M), to construct a size of M×L a The original sound pressure matrix The acoustic signal p collected by each channel Im Perform segmentation processing to form signal segments with a length of L and an overlap rate of R, and obtain the signal segments containing A quick sound pressure matrix process.

[0057] In a specific embodiment, the number of channels M is selected to be 50, and the signal acquisition length L is selected to be a =96000, the array topology is a planar uniform circular array with an array diameter of 1m. The array is located in the xy plane of the three-dimensional rectangular coordinate system. The array plane is perpendicular to the positive direction of the z axis. The array center is located at the origin of the coordinate system. Then the signal p of each channel is converted into Im Divide it into signal segments with a length of L and an overlap rate of R, and obtain the signal segments containing A quick sound pressure matrix in Indicates not greater than integer, the submatrix corresponding to the s(s=1,2,...,S)th snapshot The size is M×L, express The L(1-R)(s-1)+1th to L(1-R)(s-1)+Lth elements in the signal are selected; and in the specific implementation process, the signal segmentation length L=96000 and the overlap rate R=0.5 are selected.

[0058] exist Figure 1 、 Figure 2 、 Figure 3 In the embodiment shown in FIG. 1 , step S2 is to convert the signal corresponding to the s-th snapshot into the frequency domain using discrete Fourier transform to form a frequency domain sound pressure matrix And select the lth (l=1,2,...,L) frequency f in the frequency domain sound pressure matrix lThe corresponding signal components are used to form the sound pressure vector p = [p1(f l ),p2(f l ),...,p M (f l )] T process.

[0059] In this embodiment, for the signal corresponding to the s-th snapshot, the signal is converted into the frequency domain using discrete Fourier transform to form the frequency domain sound pressure matrix in DFT stands for Discrete Fourier Transform; select the lth (l = 1, 2, ..., L) frequency f in the frequency domain sound pressure matrix l The corresponding signal components constitute the sound pressure vector p=[p1(f l ),p2(f l ),...,p M (f l )] T , where p m (f l )represent Intermediate frequency f l Corresponding elements; in the specific implementation process, choose to analyze the signal corresponding to the 5th snapshot and select the analysis frequency f l =The signal component corresponding to 4000Hz.

[0060] exist Figure 1 、 Figure 2 、 Figure 3 In the embodiment shown in FIG. 2 , step S3 is to select a sound source reconstruction area according to a preset rule, and divide the sound source reconstruction area into N=N1×N2 discrete grids, where the position coordinates of the grid points in the discrete grids are u n =(u nx ,u ny ,u nz )(n=1,2,...,N); According to the position coordinates of the array elements in the microphone array and the sound pressure vector p=[p1(f l ),p2(f l ),...,p M (f l )] T The process of establishing the sound propagation model p=Gq+r.

[0061] In this embodiment, the sound source reconstruction area is selected according to a preset rule, and the sound source reconstruction area is divided into N=N1×N2 discrete grids, and the position coordinates of the grid points are u n =(u nx ,u ny ,u nz)(n=1,2,...,N), according to the position coordinates u of the array elements in the microphone array m =(u mx ,u my ,u mz ) Establish the sound propagation model p = Gq + r, where q is the sound source intensity vector to be solved, r is the noise vector, and G is a transmission matrix of size M × N, where the elements are exp() is the exponential function, c is the speed of sound, Is an imaginary unit.

[0062] In the specific implementation process, the sound source is located at (-0.1m, 0.15m, 1m) and (0.2m, 0.3m, 1m), and the sound source intensity is 100dB. The sound source intensity is defined as the sound pressure level at a distance of 1m from the sound source, and the reference value is 2×10 -5 Pa, the reconstruction area is a 1m×1m square area parallel to the xy plane, the distance between the reconstruction area and the microphone array plane is 1m, the center of the reconstruction area is located at (0m, 0m, 1m), select N1=41, N2=41, and divide the sound source reconstruction area into discrete grids evenly. The sound speed c=340m / s.

[0063] exist Figure 1 、 Figure 2 、 Figure 3 , in the embodiment shown together, step S4 is to establish an acoustic inverse problem based on a generalized minimum-maximum concave function according to the sound propagation model, introduce an auxiliary matrix to deform the acoustic inverse problem to obtain a deformed expression, and solve the deformed expression in an alternating splitting manner to obtain the target sound source intensity to be determined, and form a sound image according to the target sound source intensity to complete the sound source identification process; wherein,

[0064] In the process of establishing an acoustic inverse problem based on a generalized minimum-maximum concave function according to the sound propagation model, and introducing an auxiliary matrix to deform the acoustic inverse problem to obtain a deformed expression, the process includes:

[0065] S411: Establish an acoustic inverse problem based on the generalized minimum-maximum concave function according to the sound propagation model Among them, λ>0 is the regularization parameter that needs to be given, is a generalized minimum-maximum concave function, B is an auxiliary matrix; v is an auxiliary vector and is a variable;

[0066] S412: Given a balance parameter 0<γ<1, let the auxiliary matrix The acoustic inverse problem is deformed to determine a deformed expression of the acoustic inverse problem:

[0067]

[0068] S413: Take the transformed expression as the objective function, and record the smooth term of the objective function for variable q as f1(q), the non-smooth term of the objective function for variable q as f2(q), the smooth term of the objective function for variable v as g1(v), and the non-smooth term of the objective function for variable v as g2(v).

[0069] The process of solving the deformed expression in an alternating splitting manner to obtain the target sound source intensity to be determined, and forming a sound image according to the target sound source intensity includes:

[0070] S421: Initialize the counting variable and randomly initialize the intermediate variable and

[0071] S422: As the counting variable is updated, the value of the sound source intensity vector to be calculated is iteratively updated in sequence according to the modified expression, and the intensity error between the values ​​of the sound source intensity vectors obtained at two adjacent times is calculated;

[0072] S423: Determine whether the intensity error exceeds a preset convergence threshold; if so, continue iteratively updating and calculating the value of the sound source intensity vector to be calculated; if the intensity error does not exceed the convergence threshold, use the value of the sound source intensity vector calculated last as the target sound source intensity to be calculated;

[0073] S424: Generate a sound image according to the target sound source intensity.

[0074] In a specific embodiment, the convergence threshold is 0.0001; the alternating splitting method is a Douglas-Rachford splitting method.

[0075] In this specific embodiment, the process of iteratively updating and calculating the value of the sound source intensity vector to be calculated according to the modified expression as the counting variable is updated, and calculating the intensity error between the values ​​of the sound source intensity vectors obtained at two adjacent times includes:

[0076] S4221: Calculate the value of the current sound source intermediate variable and the value of the current sound source intensity vector under the current count variable; wherein the current sound source intermediate variable is: The current sound source intensity vector is: in, is the pre-calculated proximal operator, x is the function independent variable, and z is the optimization target variable;

[0077] S4222: Let k=k+1 in the current sound source intensity vector, calculate the value of the source intensity vector of the next variable of the current counting variable, and calculate the intensity error e between the value of the source intensity vector of the next variable and the value of the current sound source intensity vector R ; Among them, the intensity error is

[0078] In the process of calculating the value of the current sound source intermediate variable and the value of the current sound source intensity vector under the current count variable,

[0079] Also calculate the value of the current auxiliary vector and the value of the current auxiliary intermediate variable; where,

[0080] The current auxiliary vector is:

[0081] The current auxiliary intermediate variables are:

[0082] in, is the proximal operator of the budget.

[0083] If the intensity error does not exceed the convergence threshold, the value of the sound source intensity vector calculated last time is used as the target sound source intensity to be determined. The process of generating a sound image according to the target sound source intensity includes:

[0084] S4231: Using the value of the sound source intensity vector calculated last time as the target sound source intensity value to be calculated;

[0085] S4232: Convert the target sound source intensity into a sound source intensity matrix of size N1×N2 And the sound source intensity matrix Perform logarithmic transformation on each element in to obtain the target sound source matrix in, is a matrix The elements in Q ref is the preset reference sound source intensity;

[0086] S4233: Generate a sound image according to the target sound source matrix, and perform sound source identification based on the sound image.

[0087] More specifically, in a more specific embodiment, step S4 includes the following steps:

[0088] (1) First, according to the sound propagation model, the acoustic inverse problem based on the generalized minimum-maximum concave function is established. Where λ>0 is the regularization parameter that needs to be given, is a generalized minimum-maximum concave function, and B is an auxiliary matrix. In the specific implementation process, the regularization parameter λ is selected as 0.1.

[0089] (2): Given a balance parameter 0<γ<1, let the auxiliary matrix The further expression of the acoustic inverse problem is obtained:

[0090] Let the smooth term f1(q) of the objective function for the variable q be f1(q), let the non-smooth term f2(q) of the objective function for the variable q be f2(q), let the smooth term g1(v) of the objective function for the variable v be g1(v), let the non-smooth term g2(v) of the objective function for the variable v be g2(v); in the specific implementation process, the balance parameter γ = 0.7 is selected, and the smooth term of the objective function for the variable q is The non-smooth term of the objective function for variable q is f2(q) = λ||q||1, and the smooth term of the objective function for variable v is The non-smooth term of the objective function for the variable v is g2(v)=-λ||v||1.

[0091] (3): Initialize the counting variable k=0 and randomly initialize the intermediate variables and

[0092] In the specific implementation process, the intermediate variables and They are randomly initialized according to a Gaussian distribution with a mean of 0 and a variance of 1, that is, all elements are randomly generated.

[0093] (4): Using the Douglas-Rachford splitting method, calculate the updated value of the sound source intensity vector Calculate the updated value of the intermediate variable Calculate the updated value of the auxiliary vector Calculate the updated value of the intermediate variable

[0094] Among them, the proximal operator is In the specific implementation process, the updated value of the sound source intensity vector Updated values ​​of intermediate variables Updated value of the auxiliary vector Updated values ​​of intermediate variables I is with G T G is the same size as the identity matrix.

[0095] (5): Let k = k + 1, and calculate the relative error e of the sound source intensity vector before and after the update R , if e R Exceeds the pre-set convergence threshold e T , go to step (6); otherwise, return to step (4).

[0096] In the specific implementation process, the relative error Select the convergence threshold e T =0.0001.

[0097] (6): Let q = q (k) , transform the sound source intensity vector q into a matrix of size N1×N2 And the matrix Each element in is logarithmically transformed to form a sound image Q for sound source identification.

[0098] In the specific implementation process, the logarithmic transformation in is a matrix The elements in Q ref is the reference sound source intensity, select the reference sound source intensity Q ref =0.00002.

[0099] In this way, a sound source intensity matrix is ​​obtained to form a sound image, which is then used for sound source identification.

[0100] like Figure 2 、 Figure 3 As shown in FIG, through the sparse sound source identification method for microphone array measurement in the embodiment of the present invention, two sound sources 1m away from the microphone array can be identified, the sound source positions in the obtained sound image are consistent with the actual sound source positions, and the sound source intensity in the obtained sound image is very close to the actual sound source intensity. This result shows that the sparse sound source identification method for microphone array measurement in the embodiment of the present invention can achieve the goal of sparse sound source identification; Figure 3 As shown, the technical solution in the embodiment of the present invention is compared with the prior art, and the mean square error between the sound source intensity vector obtained by different methods and the true sound source intensity vector is calculated. Compared with the prior art, the mean square error of the sound source identification result obtained by the technical solution in the embodiment of the present invention is smaller. This result shows that the sparse sound source identification method for microphone array measurement in the embodiment of the present invention can improve the accuracy of sparse sound source identification. In addition, the calculation time of the above methods is recorded under the same hardware conditions. The calculation time of the technical solution of the sparse sound source identification method for microphone array measurement in the embodiment of the present invention is 0.61 seconds, the calculation time of the method based on the 1 norm is 0.44 seconds, the calculation time of the method based on the approximate 0 norm is 0.72 seconds, and the calculation time of the method based on orthogonal matching pursuit is 0.41 seconds. This shows that the sparse sound source identification method for microphone array measurement in the embodiment of the present invention can obtain the most accurate sound source identification result while maintaining high computational efficiency, and achieve a better balance in terms of result accuracy and computational efficiency. Therefore, the present invention has strong practicality.

[0101] In summary, the sparse sound source identification method for microphone array measurement provided by the present invention first uses the microphone array to collect sound signals to construct a M×L a The original sound pressure matrix; the sound signals collected by each channel are segmented to form signal segments, and the sound pressure matrix is ​​obtained according to the original sound pressure matrix, and then the discrete Fourier transform is used to convert the signal to the frequency domain to form a frequency domain sound pressure matrix, and the signal components are selected in the frequency domain sound pressure matrix to form a sound pressure vector, and then the sound source reconstruction area is selected according to the preset rules, and the sound source reconstruction area is divided into discrete grids, and a sound propagation model is established according to the position coordinates of the array elements in the microphone array and the sound pressure vector, and then an acoustic inverse problem based on the generalized minimum and maximum concave function is established according to the sound propagation model, and an auxiliary matrix is ​​introduced to deform the acoustic inverse problem to obtain a deformed expression, and the deformed expression is solved by alternating splitting to obtain the target sound source intensity to be determined, and a sound image is formed according to the target sound source intensity to complete the sound source. Identification; in this way, a sparse sound source identification method is established using the generalized minimum-maximum concave function as a regularization function. The approximate function used has a better approximation effect on the 0 norm, and the degree of penalty received by non-zero elements in the sparse recovery process is more consistent, which can solve the problem of underestimation of sound source intensity in existing technologies; and the objective function of the acoustic inverse problem is convex, which can ensure the accuracy of the sparse sound source identification results. After introducing auxiliary variables, the acoustic inverse problem is solved by alternating splitting, and an explicit variable iterative update formula is established, which can conveniently solve the acoustic inverse problem and improve computational efficiency; and considering the signal overlapping segmentation strategy and different sound source frequencies to adapt to changes in the external environment, the sparse prior information used in the solution process can suppress the interference caused by measurement noise and improve identification accuracy.

[0102] like Figure 4 As shown, the present invention also provides a sparse sound source identification system 100 for microphone array measurement, which implements the sparse sound source identification method for microphone array measurement as described above, including:

[0103] The sound pressure matrix acquisition unit 101 is used to collect the sound pressure matrix of length L using a microphone array including M channels. a The acoustic signal p Im (m=1,2,...,M), to construct a size of M×L a The original sound pressure matrix The acoustic signal p collected by each channel Im Perform segmentation processing to form a signal segment with a length of L and an overlap rate of R, and obtain the signal containing A quick sound pressure matrix in, Indicates not greater than integer, the submatrix corresponding to the s(s=1,2,...,S)th snapshot The size is M×L, express The L(1-R)(s-1)+1th to L(1-R)(s-1)+Lth elements in ;

[0104] The frequency domain conversion unit 102 is used to convert the signal corresponding to the s-th snapshot into the frequency domain using discrete Fourier transform to form a frequency domain sound pressure matrix And select the lth (l=1,2,...,L) frequency f in the frequency domain sound pressure matrix l The corresponding signal components are used to form the sound pressure vector p=[p1(f l ),p2(f l ),...,p M (f l )] T ,in, DFT stands for Discrete Fourier Transform; m (f l )represent Intermediate frequency f l The corresponding signal components;

[0105] The model building unit 103 is used to select a sound source reconstruction area according to a preset rule, and divide the sound source reconstruction area into N=N1×N2 discrete grids, where the position coordinates of the grid points in the discrete grids are u n =(u nx ,u ny ,u nz )(n=1,2,...,N); According to the position coordinates of the array elements in the microphone array and the sound pressure vector p=[p1(f l ),p2(f l ),...,p M (f l )] T Establish a sound propagation model p=Gq+r, where q is the sound source intensity vector to be solved, r is the noise vector, G is a transmission matrix of size M×N, and the elements of the transmission matrix are Where exp() is the exponential function, c is the speed of sound, is the imaginary unit, u m is the position coordinate of the array element in the microphone array, u n are the position coordinates of the grid points in the discrete grid;

[0106] The identification and solving unit 104 is used to establish an acoustic inverse problem based on a generalized minimum-maximum concave function according to the sound propagation model, introduce an auxiliary matrix to deform the acoustic inverse problem to obtain a deformed expression, solve the deformed expression using an alternating splitting method to obtain the target sound source intensity to be determined, and form a sound image according to the target sound source intensity to complete sound source identification.

[0107] The specific implementation method is not particularly limited and will not be described in detail here. Please refer to the specific embodiment of the sparse sound source identification method for microphone array measurement.

[0108] As described above, the sparse sound source identification system for microphone array measurement provided by the present invention first collects sound signals using a microphone array through a sound pressure matrix acquisition unit 101 to construct an original sound pressure matrix; the sound signals collected by each channel are segmented to form signal segments, and a sound pressure matrix is ​​obtained according to the original sound pressure matrix, and then the signal is converted into a frequency domain by a discrete Fourier transform through a frequency domain conversion unit 102 to form a frequency domain sound pressure matrix, and signal components are selected in the frequency domain sound pressure matrix to form a sound pressure vector, and then a sound source reconstruction area is selected according to a preset rule through a model establishment unit 103, and the sound source reconstruction area is divided into discrete grids, and a sound propagation model is established according to the position coordinates of the array elements in the microphone array and the sound pressure vector, and then an acoustic inverse problem based on a generalized minimum-maximum concave function is established according to the sound propagation model through the identification solution unit 104, and an auxiliary matrix is ​​introduced to deform the acoustic inverse problem to obtain a deformed expression, and an alternating subtraction method is used. The deformation expression is solved in a splitting manner to obtain the target sound source intensity to be determined, and a sound image is formed according to the target sound source intensity to complete the sound source identification; in this way, a sparse sound source identification method is established using a generalized minimum-maximum concave function as a regularization function. The approximate function used has a better approximation effect on the 0 norm, and the degree of penalty suffered by non-zero elements in the sparse recovery process is more consistent, which can solve the problem of underestimation of sound source intensity in the existing technology; and the objective function of the acoustic inverse problem is convex, which can ensure the accuracy of the sparse sound source identification results. After introducing auxiliary variables, the acoustic inverse problem is solved in an alternating splitting manner, and an explicit variable iterative update formula is established, which can conveniently solve the acoustic inverse problem and improve the computational efficiency; and considering the signal overlapping segmentation strategy and different sound source frequencies to adapt to changes in the external environment, the sparse prior information used in the solution process can suppress the interference caused by measurement noise and improve the identification accuracy.

[0109] The sparse sound source identification method and system for microphone array measurement according to the present invention have been described above by way of example with reference to the accompanying drawings. However, those skilled in the art will appreciate that various improvements may be made to the sparse sound source identification method and system for microphone array measurement according to the present invention without departing from the scope of the present invention. Therefore, the scope of protection of the present invention shall be determined by the contents of the appended claims.

Claims

1. A sparse sound source identification method for microphone array measurement, characterized in that: include: The microphone array with M channels is used to collect the data of length L. a The acoustic signal p Im (m=1,2,...,M), to construct a size of M×L a The original sound pressure matrix The acoustic signal p collected by each channel Im The segmentation process is performed to form a signal segment with a length of L and an overlap rate of R, and the signal containing A quick sound pressure matrix in, Indicates not greater than integer, the submatrix corresponding to the sth (s=1,2,...,S) snapshot The size is M×L, express The L(1-R)(s-1)+1th to L(1-R)(s-1)+Lth elements in ; The signal corresponding to the sth snapshot is converted into the frequency domain using discrete Fourier transform to form the frequency domain sound pressure matrix And select the lth (l=1,2,...,L) frequency f in the frequency domain sound pressure matrix l The corresponding signal components are used to form the sound pressure vector p = [p1(f l ),p2(f l ),...,p M (f l )] T ,in, DFT stands for Discrete Fourier Transform; m (f l )represent Middle frequency f l The corresponding signal components; The sound source reconstruction area is selected according to a preset rule, and the sound source reconstruction area is divided into N=N1×N2 discrete grids, and the position coordinates of the grid points in the discrete grids are u n =(u nx ,u ny ,u nz )(n=1,2,...,N); according to the position coordinates u of the array elements in the microphone array m (m=1,2,...,M) and the sound pressure vector p=[p1(f l ),p2(f l ),...,p M (f l )] T Establish the sound propagation model p=Gq+r, where q is the sound source intensity vector to be solved, r is the noise vector, G is a transmission matrix of size M×N, and the elements of the transmission matrix are Where exp( ) is an exponential function, c is the speed of sound, is the imaginary unit, u m is the position coordinate of the array element in the microphone array, u n are the position coordinates of the grid points in the discrete grid; According to the sound propagation model, an acoustic inverse problem based on a generalized minimum-maximum concave function is established, an auxiliary matrix is ​​introduced to deform the acoustic inverse problem to obtain a deformed expression, and the deformed expression is solved by an alternating splitting method to obtain the target sound source intensity to be obtained, and a sound image is formed according to the target sound source intensity to complete sound source identification.

2. The sparse sound source identification method for microphone array measurement according to claim 1, characterized in that: In the process of establishing an acoustic inverse problem based on a generalized minimum-maximum concave function according to the sound propagation model, and introducing an auxiliary matrix to deform the acoustic inverse problem to obtain a deformed expression, the process includes: According to the sound propagation model, an acoustic inverse problem based on a generalized minimum-maximum concave function is established. Among them, λ>0 is the regularization parameter that needs to be given, is a generalized minimum-maximum concave function, B is an auxiliary matrix; v is an auxiliary vector and is a variable; Given a balance parameter 0<γ<1, let the auxiliary matrix The acoustic inverse problem is deformed to determine the deformed expression of the acoustic inverse problem: The deformed expression is used as the objective function, and the smooth term of the objective function for variable q is denoted as f1(q), the non-smooth term of the objective function for variable q is denoted as f2(q), the smooth term of the objective function for variable v is denoted as g1(v), and the non-smooth term of the objective function for variable v is denoted as g2(v).

3. The sparse sound source identification method for microphone array measurement according to claim 2, characterized in that: The process of solving the deformed expression in an alternating splitting manner to obtain the target sound source intensity to be obtained, and forming a sound image according to the target sound source intensity includes: Initialize the counting variable and randomly initialize the intermediate variables and As the counting variable is updated, the value of the sound source intensity vector to be calculated is iteratively updated in sequence according to the deformed expression, and the intensity error between the values ​​of the sound source intensity vectors obtained at two adjacent times is calculated; Determine whether the intensity error exceeds a preset convergence threshold; if the intensity error exceeds the preset convergence threshold, continue to iteratively update and calculate the value of the sound source intensity vector to be calculated; if the intensity error does not exceed the convergence threshold, use the value of the sound source intensity vector calculated for the last time as the target sound source intensity to be calculated; A sound image is generated according to the target sound source intensity.

4. The sparse sound source identification method for microphone array measurement according to claim 3, characterized in that: The convergence threshold is 0.0001.

5. The sparse sound source identification method for microphone array measurement according to claim 3, characterized in that: The alternating splitting mode is the Douglas-Rachford splitting mode.

6. The sparse sound source identification method for microphone array measurement according to claim 5, characterized in that: The process of iteratively updating and calculating the value of the sound source intensity vector to be calculated according to the deformed expression in sequence as the counting variable is updated, and calculating the intensity error of the values ​​of the sound source intensity vectors obtained at two adjacent times includes: Calculate the value of the current sound source intermediate variable and the value of the current sound source intensity vector under the current count variable; where the current sound source intermediate variable is: The current sound source intensity vector is: in, It is the proximal operator of the budget; Let k=k+1 in the current sound source intensity vector, calculate the value of the source intensity vector of the next variable of the current counting variable, and calculate the intensity error e between the value of the source intensity vector of the next variable and the value of the current sound source intensity vector R .

7. The sparse sound source identification method for microphone array measurement according to claim 6, The intensity error is 8. The sparse sound source identification method for microphone array measurement according to claim 7, characterized in that: In the process of calculating the value of the current sound source intermediate variable and the value of the current sound source intensity vector under the current count variable, The value of the current auxiliary vector and the value of the current auxiliary intermediate variable are also calculated; wherein, The current auxiliary vector is: The current auxiliary intermediate variables are: in, It is the proximal operator of the budget.

9. The sparse sound source identification method for microphone array measurement according to claim 8, characterized in that: If the intensity error does not exceed the convergence threshold, the value of the sound source intensity vector calculated last time is used as the target sound source intensity to be calculated; The process of generating a sound image according to the target sound source intensity includes: The value of the sound source intensity vector calculated last time is used as the value of the target sound source intensity to be calculated; The target sound source intensity is converted into a sound source intensity matrix of size N1×N2 And the sound source intensity matrix Logarithmically transform each element in to obtain the target sound source matrix in, For the matrix The elements in Q ref is the preset reference sound source intensity; A sound image is generated according to the target sound source matrix, and sound source identification is performed according to the sound image.

10. A sparse sound source identification system for microphone array measurement, characterized in that: The method for sparse sound source identification for microphone array measurement according to any one of claims 1 to 9 is implemented, comprising: The sound pressure matrix acquisition unit is used to collect the sound pressure matrix with a length of L using a microphone array containing M channels. a The acoustic signal p Im (m=1,2,...,M), to construct a size of M×L a The original sound pressure matrix The acoustic signal p collected by each channel Im The segmentation process is performed to form a signal segment with a length of L and an overlap rate of R, and the signal containing A quick sound pressure matrix in, Indicates not greater than integer, the submatrix corresponding to the sth (s=1,2,...,S) snapshot The size is M×L, express The L(1-R)(s-1)+1th to L(1-R)(s-1)+Lth elements in ; The frequency domain conversion unit is used to convert the signal corresponding to the s-th snapshot into the frequency domain using discrete Fourier transform to form a frequency domain sound pressure matrix And select the lth (l=1,2,...,L) frequency f in the frequency domain sound pressure matrix l The corresponding signal components are used to form the sound pressure vector p = [p1(f l ),p2(f l ),...,p M (f l )] T ,in, DFT stands for Discrete Fourier Transform; m (f l )represent Middle frequency f l The corresponding signal components; The model building unit is used to select a sound source reconstruction area according to a preset rule, and divide the sound source reconstruction area into N = N1 × N2 discrete grids, where the position coordinates of the grid points in the discrete grids are u n =(u nx ,u ny ,u nz )(n=1,2,...,N); According to the position coordinates of the array elements in the microphone array and the sound pressure vector p=[p1(f l ),p2(f l ),...,p M (f l )] T Establish the sound propagation model p=Gq+r, where q is the sound source intensity vector to be solved, r is the noise vector, G is a transmission matrix of size M×N, and the elements of the transmission matrix are Where exp() is the exponential function, c is the speed of sound, is the imaginary unit; u m is the position coordinate of the array element in the microphone array, u n are the position coordinates of the grid points in the discrete grid; An identification and solving unit is used to establish an acoustic inverse problem based on a generalized minimum-maximum concave function according to the sound propagation model, introduce an auxiliary matrix to deform the acoustic inverse problem to obtain a deformed expression, solve the deformed expression in an alternating splitting manner to obtain a target sound source intensity to be obtained, and form a sound image according to the target sound source intensity to complete sound source identification.

Images