A total variation regularized direction finding method and system based on acoustic features

Abstract

The present invention provides a total variation regularized direction-finding method and system based on acoustic features, comprising obtaining an output acoustic signal of an incident sound source based on a planar array; calculating an estimated value of the direction of arrival of the incident sound source based on a pre-constructed two-dimensional total variation sparse direction of arrival estimation model; wherein the construction process of the two-dimensional total variation sparse direction of arrival estimation model is specifically as follows: establishing a spatially extended sound source model according to the spatial distribution characteristics of the sound source; constructing a two-dimensional array manifold matrix to obtain an overcomplete representation of the two-dimensional array manifold matrix, and determining a two-dimensional sparse representation of the incident sound source based on the overcomplete representation of the two-dimensional array manifold matrix; constructing a two-dimensional total variation structure matrix, and constructing a two-dimensional total variation sparse direction of arrival estimation model based on the two-dimensional total variation structure matrix; the present invention promotes piecewise constant solutions and embodies spatial smoothness, and thus effectively improves DOA estimation performance in situations involving block sparsity or a mixture of peaks and flat areas.

Description

Technical Field

[0001] The invention belongs to the technical field of acoustic array signal processing, and in particular relates to a total variation regularized direction finding method and system based on acoustic features. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] As an important pre-processing technology in human-computer interaction, direction finding technology is the core of sound source localization and the prerequisite for speech enhancement. It uses a microphone array to receive sound signals and analyze and calculate the location of the sound source, that is, to estimate the direction of arrival (DOA) to achieve directional pickup and enhancement of the sound source.

[0004] At present, the main methods for DOA estimation include: subspace algorithms, maximum likelihood algorithms, beamforming algorithms and sparse reconstruction algorithms. The first three methods require complex statistical values ​​as initial variables, and cannot accurately estimate DOA for low signal-to-noise ratio, small number of snapshots, coherent sources, etc., and are gradually replaced by sparse DOA estimation based on sparse representation and sparse reconstruction. Sparse DOA estimation is usually an underdetermined problem, that is, the number of estimated sound source positions is greater than the number of microphones in the measurement sound array, which will lead to the linear equation between the array output signal and the sound source signal has no solution, and the DOA of the sound source cannot be obtained. To this end, the approximate solution that satisfies this constraint is obtained through regularization processing with additional constraints. Different constraints lead to different approximate solutions. Among them, the LASSO model is formed with sparsity (L1 norm constraint) as the constraint term, and the least squares model is formed with energy minimization (L2 norm constraint) as the constraint. They can also perform high-resolution DOA estimation in the presence of coherent sources or single snapshot data. However, as the number of snapshots decreases, the target density increases, and the practical application needs with strong multi-target correlation, especially for the sparse DOA estimation of extended sound sources with grouped spatial positions, phase linearity and block sparsity in amplitude distribution, these two methods are no longer suitable.

[0005] To this end, the generalized LASSO, or GL, is proposed. By constructing an additional structure matrix, geometric or structural constraints are imposed on the solution to enhance the accuracy of DOA estimation. GL is mathematically equivalent to the total variation model. The total variation model is widely used in image processing because it can retain larger features in the image while suppressing smaller-scale noise components. However, it is rarely mentioned in acoustics and is limited to one-dimensional DOA estimation. Its application in acoustic DOA estimation needs to be expanded. Summary of the invention

[0006] In order to solve the above problems, the present invention proposes a total variation regularized direction finding method and system based on acoustic features. The total variation regularization term constraint created and utilized by the present invention on the basis of general LASSO promotes piecewise constant solution and embodies spatial smoothness. Therefore, in the case of block sparsity or mixture of peaks and flat areas, the DOA estimation performance is effectively improved.

[0007] According to some embodiments, a first solution of the present invention provides a total variation regularized direction finding method based on acoustic features, which adopts the following technical solution:

[0008] A total variation regularized direction finding method based on acoustic features, comprising:

[0009] Acquiring an output acoustic signal of an incident acoustic source based on a planar array;

[0010] Calculate the estimated value of the direction of arrival of the incident sound source based on the pre-built two-dimensional total variation sparse direction of arrival estimation model;

[0011] The construction process of the two-dimensional total variation sparse direction of arrival estimation model is specifically as follows:

[0012] Establish a spatially extended sound source model based on the spatial distribution characteristics of the sound source;

[0013] A two-dimensional array manifold matrix is ​​constructed to obtain an overcomplete representation of the two-dimensional array manifold matrix, and a two-dimensional sparse representation of the incident sound source is determined based on the overcomplete representation of the two-dimensional array manifold matrix;

[0014] A two-dimensional total variation structure matrix is ​​constructed, and a two-dimensional total variation sparse direction of arrival estimation model is constructed based on the two-dimensional total variation structure matrix.

[0015] Furthermore, the obtaining of the output sound signal of the incident sound source based on the planar array is specifically:

[0016] The dimensions of the planar array are M×N and M=N, and the planar arrays are evenly distributed at intervals of 0.5λ, where λ is the wavelength of the incident wave;

[0017] The planar array is used to receive and sample the incident sound source to obtain an output signal of the incident sound source.

[0018] Furthermore, the spatially extended sound source model is established according to the spatial distribution characteristics of the sound source, specifically:

[0019] According to the continuity characteristics of the sound source spatial distribution, a spatially extended sound source model is established;

[0020] Among them, the spatial extension sound source model is as follows:

[0021]

[0022] Wherein, d is the packet length, assuming sampling time t, t = 1, 2, ..., L, L is the number of snapshots, there are K far-field narrow-band sound sources, and K = n × d, n is the index.

[0023] Furthermore, a two-dimensional array manifold matrix is ​​constructed to obtain an overcomplete representation of the two-dimensional array manifold matrix. Based on the overcomplete representation of the two-dimensional array manifold matrix, a two-dimensional sparse representation of the incident sound source is determined, specifically:

[0024] Based on the output sound signal sample data of the incident sound source, a two-dimensional array manifold matrix is ​​constructed;

[0025] Based on the gridded observation space, an overcomplete representation of the two-dimensional array manifold matrix is ​​constructed;

[0026] Based on the overcomplete representation of the two-dimensional array manifold matrix, a two-dimensional sparse representation of the incident sound source is obtained.

[0027] Furthermore, the two-dimensional array manifold matrix is ​​constructed based on the output sound signal sample data of the incident sound source, specifically:

[0028] Get the time difference of the incident sound source from the first sensor to the Xth sensor position;

[0029] Based on the time difference, the array manifold matrices of the incident sound source in the x-axis and y-axis directions are obtained respectively;

[0030] According to the definition of two-dimensional generalized direction vector, the two-dimensional direction vector of the plane rectangular array is obtained;

[0031] Convert the two-dimensional direction vector of a planar rectangular array into a two-dimensional array manifold matrix.

[0032] Furthermore, the two-dimensional total variation structure matrix is ​​constructed, and a two-dimensional total variation sparse direction of arrival estimation model is constructed based on the two-dimensional total variation structure matrix, specifically:

[0033] Acquire acoustic signal sample data, analyze and extract block sparse features and correlation features of the acoustic signal;

[0034] Based on the block sparse features and correlation features of the acoustic signal, a total variation structure matrix is ​​constructed;

[0035] According to the total variation structure matrix combined with the sparse direction of arrival estimation model, the final two-dimensional total variation sparse direction of arrival estimation model is obtained, which is as follows:

[0036]

[0037] Where μ1 and μ2 are regularization parameters, the observation vector y is given, and the overcomplete array manifold matrix is Total variation structure matrix D tv , target solution s.

[0038] Furthermore, the estimated value of the direction of arrival of the incident sound source is calculated based on the pre-built two-dimensional total variation sparse direction of arrival estimation model, specifically:

[0039] The matrix optimization calculation method is used to quickly solve the two-dimensional total variation sparse direction of arrival estimation model, and the two-dimensional arrival estimation values ​​of the incident sound source in the pitch angle and azimuth directions are obtained;

[0040] The matrix optimization calculation method is specifically as follows:

[0041] Obtain the array structure, the angle range of the sound source distribution, the sound propagation speed and the wavelength, and construct an overcomplete representation of the array matrix from the two-dimensional planar array manifold;

[0042] Construct a two-dimensional total variation matrix using the generalized lasso matrix and block sparse features;

[0043] Constructing a two-dimensional sparse DOA estimation model from a two-dimensional total variation matrix Assign values ​​to μ1 and μ2;

[0044] The spatial spectrum of the sound source distribution is generated by using the interior point of convex optimization to search for the maximum value;

[0045] The direction of arrival of the sound source is accurately estimated from the spatial spectrogram.

[0046] According to some embodiments, a second solution of the present invention provides a total variation regularized direction finding system based on acoustic features, which adopts the following technical solution:

[0047] A total variation regularized direction finding system based on acoustic features, comprising:

[0048] A sound source acquisition module, configured to acquire an output sound signal of an incident sound source based on a planar array;

[0049] A sound source estimation module is configured to calculate an estimated value of a direction of arrival of an incident sound source based on a pre-built two-dimensional total variation sparse direction of arrival estimation model;

[0050] The construction process of the two-dimensional total variation sparse direction of arrival estimation model is specifically as follows:

[0051] Establish a spatially extended sound source model based on the spatial distribution characteristics of the sound source;

[0052] A two-dimensional array manifold matrix is ​​constructed to obtain an overcomplete representation of the two-dimensional array manifold matrix, and a two-dimensional sparse representation of the incident sound source is determined based on the overcomplete representation of the two-dimensional array manifold matrix;

[0053] A two-dimensional total variation structure matrix is ​​constructed, and a two-dimensional total variation sparse direction of arrival estimation model is constructed based on the two-dimensional total variation structure matrix.

[0054] According to some embodiments, a third aspect of the present invention provides a computer-readable storage medium.

[0055] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of a total variation regularized direction finding method based on acoustic features as described in the first aspect above.

[0056] According to some embodiments, a fourth aspect of the present invention provides a computer device.

[0057] A computer device comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps in the total variation regularized direction finding method based on acoustic features as described in the first aspect above are implemented.

[0058] Compared with the prior art, the present invention has the following beneficial effects:

[0059] 1. The present invention combines the generalized LASSO with the two-dimensional total variation to simultaneously improve the sparsity of the solution coefficients (i.e., there are few active coefficients in the solution) and the sparsity of the continuous coefficient differences (i.e., sparse profile flatness). Compared with other existing DOA estimation methods, the proposed method achieves the purpose of jointly estimating the pitch angle and azimuth of the extended sound source with a higher detection probability in applications involving block sparsity or a mixture of peaks and flat areas.

[0060] 2. The present invention does not require eigendecomposition of the high-dimensional sample covariance matrix, thus avoiding the additional angle pairing required by conventional methods and the decorrelation step for coherent sound sources; through the definition of two-dimensional total variation, a total variation structure matrix is ​​constructed to describe the grouping characteristics of the target sound source, which is conducive to promoting the formation of piecewise constant solutions. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0062] Figure 1 It is a flow chart of a total variation regularized direction finding method based on acoustic features according to the first embodiment of the present invention;

[0063] Figure 2 It is the incident model of the extended sound source in the plane array constructed in the first embodiment of the present invention;

[0064] FIG3( a ) is a spatial spectrum in a case of multiple extended sound sources according to the first embodiment of the present invention;

[0065] FIG3( b ) is a distribution diagram of DOA estimation values ​​in a situation of multiple extended sound sources according to Embodiment 1 of the present invention;

[0066] FIG3( c ) is a distribution diagram of DOA estimation values ​​of a general LASSO in a situation of multiple extended sound sources according to the first embodiment of the present invention;

[0067] FIG4( a ) is a spatial spectrum of a point source and an extended sound source coexisting according to the first embodiment of the present invention;

[0068] FIG4( b ) is a distribution diagram of DOA estimation values ​​when a point source and an extended sound source coexist according to Embodiment 1 of the present invention;

[0069] FIG4( c ) is a distribution diagram of DOA estimation values ​​of a general LASSO under the coexistence of a point source and an extended sound source according to the first embodiment of the present invention;

[0070] Figure 5 is a comparison diagram of the detection probabilities of the estimation results of multiple extended sound sources under different sparsity K conditions between the method described in the first embodiment of the present invention and two existing DOA estimation methods;

[0071] Figure 6 1 is a comparison chart of the detection probability of the method described in the first embodiment of the present invention and the detection probability of the estimation results of multiple extended sound sources under the signal-to-noise ratio condition of two existing DOA estimation methods. DETAILED DESCRIPTION

[0072] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0073] It should be noted that the following detailed descriptions are all illustrative and intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0074] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprising" and / or "including" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0075] In the absence of conflict, the embodiments of the present invention and the features of the embodiments may be combined with each other.

[0076] Embodiment 1

[0077] like Figure 1-Figure 6As shown, this embodiment provides a total variation regularized direction finding method based on acoustic features. In this embodiment, the method includes the following steps:

[0078] Step S1: acquiring an output sound signal of an incident sound source based on a planar array;

[0079] Step S2: Calculate the estimated value of the direction of arrival of the incident sound source based on a pre-built two-dimensional total variation sparse direction of arrival estimation model.

[0080] Among them, in step S2, the construction process of the two-dimensional total variation sparse direction of arrival estimation model is specifically as follows:

[0081] Step S2.1: Establishing a spatially extended sound source model according to the spatial distribution characteristics of the sound source;

[0082] Step S2.2: construct a two-dimensional array manifold matrix, obtain an overcomplete representation of the two-dimensional array manifold matrix, and determine a two-dimensional sparse representation of the incident sound source based on the overcomplete representation of the two-dimensional array manifold matrix;

[0083] Step S2.3: construct a two-dimensional total variation structure matrix, and construct a two-dimensional total variation sparse direction of arrival estimation model based on the two-dimensional total variation structure matrix.

[0084] Specifically, in step S1, the output sound signal of the incident sound source is obtained based on the planar array, specifically:

[0085] The dimensions of the planar array are M×N and M=N, and the planar arrays are evenly distributed at intervals of 0.5λ, where λ is the wavelength of the incident wave;

[0086] The planar array is used to receive and sample the incident sound source to obtain an output sound signal of the incident sound source.

[0087] Step S2.1: Establish a spatially extended sound source model according to the spatial distribution characteristics of the target sound source, specifically:

[0088] Get the acoustic signal sample data. Assume that a planar array with dimensions of M×N and M=N is used, and the array is evenly distributed with an interval of 0.5λ, where λ is the wavelength of the incident wave. Assume that at sampling time t, t=1,2,…,L, where L is the number of snapshots, there are K far-field narrowband sound sources s(t)=[s(t)1,s(t)2,…,s(t) K ] T The two-dimensional incident wave direction of the kth sound source incident on this array is k=1,2,…,K,θ k and Represent the incident azimuth and elevation angle, θ k ∈[0°,90°], The incident sound source is received and sampled using the receiving array. The typical array observation model of the received signal at time t can be expressed as:

[0089] y(t)=As(t)+n o (t) (1)

[0090] Where y(t) is the array output signal vector, A is the two-dimensional array manifold matrix, and n o (t) is the additive noise vector, n o (t) = [n o (t)1,n o (t)2,…,n o (t) MN ] T , the mean square error is σ 2 , has a mean of zero and is independent of the original received sound source. Here T is the transpose.

[0091] On the basis of the above, we further assume that the sound source is a spatially extended sound source, and the spatial incidence model is as follows: Figure 2 As shown. The spatial extension sound source is formed by arranging multiple simple sound sources according to a certain shape, array element gain, and phase relationship between adjacent array elements, such as arranging multiple simple sources to form a continuous line source. Based on its significant grouping characteristics, that is, grouping of related positions, a spatial extension sound source model can be established:

[0092]

[0093] Among them, d is the group length, obviously we have K=n×d, n is the index, and T is the transpose.

[0094] Further considering the correlation of grouped sound sources, here we take the first group as an example, taking the first sound source s(t)1 in the grouped sound source as a reference, then the kth coherent sound source s(t) in the group of sound sources k It can be expressed as:

[0095]

[0096] Where β k is the amplitude attenuation factor, represents s(t) related to s(t)1 k The phase difference is β without losing the genericity. k =1, j is the imaginary unit.

[0097] In step S2.2, a two-dimensional array manifold matrix is ​​constructed to obtain an overcomplete representation of the two-dimensional array manifold matrix. Based on the overcomplete representation of the two-dimensional array manifold matrix, a two-dimensional sparse representation of the incident sound source is determined, specifically:

[0098] Step S2.2.1: constructing a two-dimensional array manifold matrix based on the output sound signal sample data of the incident sound source;

[0099] Step S2.2.2: Based on the gridded observation space, construct an overcomplete representation of the two-dimensional array manifold matrix;

[0100] Step S2.2.3: Obtain a two-dimensional sparse representation of the incident sound source based on the overcomplete representation of the two-dimensional array manifold matrix.

[0101] In step S2.2.1, the two-dimensional array manifold matrix is ​​constructed based on the output sound signal sample data of the incident sound source, specifically:

[0102] Assume that the time difference of the kth single incident sound source from the first sensor (0,0) to the Xth sensor position (m,n) is:

[0103]

[0104] Where c is the speed of sound waves in air at standard atmospheric pressure.

[0105] The received sound signal is:

[0106]

[0107] In the formula, ω represents the angular frequency of the incident wave, ω=2πf, f is the frequency of the incident wave.

[0108] The direction vector of the k-th sound source in the x-axis and y-axis directions can be represented by the following Vandermonde matrices respectively:

[0109]

[0110]

[0111] Where a x ,a y is a vector. The array manifold matrix of K incident sound sources in the x-axis and y-axis directions can be expressed as:

[0112]

[0113] Where A x , A y is a matrix.

[0114] Since the two-dimensional generalized direction vector can be approximately expressed as the Kronecker product of two one-dimensional direction vectors along the x-axis and y-axis, the array physical structure determines the array manifold matrix, so the two-dimensional direction vector of the planar rectangular array is:

[0115]

[0116] In the formula represents a two-dimensional direction vector, represents the Kronecker product. The corresponding two-dimensional array manifold matrix is ​​constructed as:

[0117]

[0118] In the formula

[0119] In step S2.2.2, an overcomplete representation of the two-dimensional array manifold matrix is ​​constructed based on the gridded observation space, specifically:

[0120]

[0121] in, k=1,2,…,P,is the steering vector.

[0122] is the set of all possible sound source positions, where P>>K. In sparse DOA estimation, is known and has nothing to do with the actual location of the sound source.

[0123] In step S2.2.3, a two-dimensional sparse representation of the incident sound source is obtained based on the overcomplete representation of the two-dimensional array manifold matrix, specifically:

[0124] Combining the idea of ​​sparse representation, the DOA estimation problem is transformed into a sparse signal reconstruction problem;

[0125] The sound source is represented as a P×1-dimensional vector, in which only K elements are non-zero. Therefore, the sound source is sparse in space. The non-zero element position in corresponds to the angular position of the actual sound source. The array model of the sparse array can be expressed as:

[0126]

[0127] The sparse DOA reconstruction problem is to reconstruct the observation vector y(t) and the overcomplete array manifold matrix In the case of , find the position indices of all non-zero elements in the P×1 dimensional vector.

[0128] In step S2.3, the two-dimensional total variation structure matrix is ​​constructed, and a two-dimensional total variation sparse direction of arrival estimation model is constructed based on the two-dimensional total variation structure matrix, specifically:

[0129] Step S2.3.1: Acquire acoustic signal sample data, analyze and extract block sparse features and correlation features of the acoustic signal;

[0130] Step S2.3.2: constructing a total variation structure matrix based on the block sparse features and correlation features of the acoustic signal;

[0131] Step S2.3.3: Based on the total variation structure matrix combined with the sparse direction of arrival estimation model, the final two-dimensional total variation sparse direction of arrival estimation model is obtained.

[0132] Specifically, in step S2.3.2, based on the block sparse features and correlation features of the acoustic signal, a total variation structure matrix is ​​constructed, specifically:

[0133] Construct the total variation structure matrix D tv ;

[0134] Total variation regularization relies on the local smoothness of spatially extended sound sources or the sparsity of discrete gradients, which greatly promotes the block-sparse characteristics of the sound source solution and gives a piecewise constant fit for the setting where coordinates in the true model are closely related to their neighbors (local correlation), while suppressing single noise peaks.

[0135] Assume that the sound source Sorting the columns into a matrix S i,j represents any element in the matrix, i is the row vector index, j is the column vector index, i, j = 1,..., M, the variation of adjacent elements on any row and column in the matrix is ​​defined as:

[0136]

[0137] ΔS h , ΔS v They represent the first-order x-axis variation and y-axis variation of adjacent elements of the matrix S, respectively. The two-dimensional total variation is defined as:

[0138]

[0139] From formula (16), we get the total variation structure matrix

[0140]

[0141] Let S Z =[SS T ], then for the sound source matrix S, we have:

[0142] D tv S Z =D tv [SS T ] (18).

[0143] In step S2.3.3, the final two-dimensional total variation sparse direction of arrival estimation model is obtained based on the total variation structure matrix combined with the sparse direction of arrival estimation model, which is specifically:

[0144] Construct a two-dimensional total variation sparse DOA estimation model;

[0145] Introducing equation (18) into the sparse DOA estimation model of equation (14), we get the two-dimensional total variation DOA estimation model:

[0146]

[0147] In order to ensure the sparsity of the target solution s itself, we combine the general LASSO and the two-dimensional total variation on the basis of equation (19), thereby improving the sparsity of the coefficients (i.e., there are few active coefficients in the solution) and the sparsity of the continuous coefficient differences (i.e., sparse contour flatness), making the proposed method suitable for applications involving block sparsity or a mixture of peaks and flat areas. The final two-dimensional total variation sparse DOA estimation model is expressed as:

[0148]

[0149] Where μ1 and μ2 are regularization parameters.

[0150] In step S2, the estimated value of the direction of arrival of the incident sound source is calculated based on the pre-built two-dimensional total variation sparse direction of arrival estimation model, specifically:

[0151] The matrix optimization calculation method is used to quickly solve the two-dimensional total variation sparse arrival direction estimation model and obtain the two-dimensional arrival estimation value of the incident sound source in the pitch angle and azimuth direction.

[0152] Specifically, the matrix optimization calculation algorithm is:

[0153] Step 1: Given the calculation parameters: array structure, sound source distribution angle range, sound propagation speed, wavelength, etc., construct an overcomplete representation of the array matrix from the two-dimensional planar array manifold

[0154] Step 2: Use the generalized lasso matrix and block sparse features to construct a two-dimensional total variation matrix D tv ;

[0155] Step 3: D tv Constructing a two-dimensional sparse DOA estimation model Assign values ​​to μ1 and μ2;

[0156] Step 4: Use the convex optimization internal point to perform maximum value search and generate the spatial spectrum of the sound source distribution;

[0157] Step 5: Accurately estimate the direction of arrival (DOA) of the sound source from the spatial spectrogram.

[0158] Calculate the DOA estimate of the incident sound source;

[0159] According to the two-dimensional total variation sparse DOA estimation model of formula (20), we use convex optimization to obtain Pitch Angle is the horizontal coordinate, azimuth angle θ k Draw a spatial orientation spectrum for the ordinate, and calculate the DOA estimate of the incident sound source by two-dimensional search in the grid point angle area where the peak point in the spectrum is located.

[0160] The present invention is further described in detail below in conjunction with specific practical applications:

[0161] Take M = 20, that is, the antenna array consists of a 20×20 uniform planar rectangular array in the x-axis direction and the y-axis direction, assuming that the sound sources are all coherent sound sources. The azimuth and elevation angles between 0° and 90° are evenly divided into 91 parts with an angle interval of 1° to construct an overcomplete array manifold matrix.

[0162] In order to verify the practicality and robustness of the method of the present invention, the Monte Carlo experiment is used to compare and analyze the method 2DTV-CES of the present invention with the general LASSO method and the improved ESPRIT method. The number of snapshots L = 100, SNR = 20dB. Since the azimuth accuracy obtained by the DOA estimation method based on sparse representation depends largely on the complete space, this paper uses the performance evaluation index of the DOA estimation method based on sparse representation, that is, the probability of correctly detecting the signal, to evaluate the performance of different algorithms, rather than the traditional mean square error method. When the absolute value of the difference between the estimated signal incident angle and the actual angle is less than or equal to 2°, it is defined as a correct detection.

[0163] Detection probability P C Defined as:

[0164]

[0165] Among them, q is the experimental number index, Q is the number of Monte Carlo experiments, K is the number of sound sources, and k c is the number of correctly estimated sound sources.

[0166] Application Example 1:

[0167] This application example verifies the spatial angle estimation capability of the present invention: the sound source types are divided into two categories:

[0168] (1) When all sound sources are extended sound sources. Assume that the source distribution contains two extended sound sources: a. The pitch angle covers the DOA interval of 5°:14°, and the corresponding azimuth angles are 6°:15°; b. The pitch angle covers the DOA interval of 18°:28°, and the corresponding azimuth angles are 25°:35°.

[0169] The two-dimensional DOA estimated spatial spectrum of the extended sound source is shown in Figure 3(a). It can be seen from the figure that under the action of the total variation regularization term, the present invention promotes the piecewise constant solution, so that the solution coefficients are closely related to their neighborhoods, and the solution coefficients are divided into two groups. Both spatial extended sound sources are modeled as a series of monopoles with equal complex amplitudes (deterministic amplitude and uniformly distributed phase), which reflects the block sparsity characteristics of the extended sound source.

[0170] The distribution of the two-dimensional DOA estimation value of the spatially extended sound source is shown in Figure 3(b) and Figure 3(c). As can be seen from the figure, the present invention with appropriate regularization term parameters can accurately estimate 15 of the 21 point sources in the continuous source for the source distribution composed of multiple extended sound sources. Although the estimation of the remaining 6 point sources has a 1° deviation, it is within the allowable error range of ±2° and is also considered to be correctly estimated. In addition, it effectively separates the two extended sound sources, reflecting an excellent DOA estimation effect. In comparison, the estimation effect of the general LASSO method is poor, and the continuous source is estimated as a collection with fewer monopoles (only 9 of the 21 point sources in the continuous source range are correctly estimated).

[0171] (2) When both point sources and extended sound sources exist, assuming that the incident angles of the two point sources are (3°, 5°) and (6°, 8°), respectively, the extended sound source elevation angle covers a DOA interval of 15°:33°, and the corresponding azimuth angles are 14°:32°.

[0172] The DOA estimated spatial spectrum when a point source and an extended sound source exist at the same time is shown in FIG4( a ). It can be seen from the figure that when a point source and an extended sound source exist at the same time, the present invention effectively utilizes the l1 norm to separate the constant solution into three groups, and utilizes the total variation regularization term to promote the formation of a piecewise constant, thereby modeling the spatial extended sound source as a series of monopoles with equal complex amplitudes.

[0173] The distribution of two-dimensional DOA estimation values ​​in complex situations is shown in Figure 4(b) and Figure 4(c). As can be seen from the figure, the present invention with appropriate regularization term parameters can accurately estimate 13 of the 21 sound sources, 6 sound sources have a deviation of 1°, and 1 sound source has a deviation of 2°, which can all be regarded as correct estimates; while the general LASSO method can only correctly estimate 13 sound sources. Both methods can correctly estimate two point sources within the allowable ±2° deviation, but compared with the actual distribution, the present invention can better realize the DOA estimation of extended sound sources compared to the general LASSO method, and can reconstruct more point sound sources within the continuous source range.

[0174] In summary, since the total variation regularization term constraint introduced and utilized by the present invention on the basis of the general LASSO promotes a piecewise constant solution and embodies spatial smoothness, it effectively improves the DOA estimation performance in situations involving block sparsity or a mixture of peaks and flat areas, proving the good spatial angle estimation capability of the present invention.

[0175] Application Example 2:

[0176] This application example verifies the impact of sparsity K on algorithm performance: considering that the extended sound source generally contains multiple continuous point sources at the same time, the 2DTV-CES method of the present invention is compared with the general LASSO method and the improved ESPRIT method to compare their respective detection probabilities. The first sound source angle of the extended sound source is (8°, 15°), then the extended sound source pitch angle covers the DOA interval position of 8°: (8+Δk)°, and the corresponding azimuth angle is 15°: (15+Δk)°, where Δk is the increment in the number of sound sources, which varies in the range of 1 to 9, and the corresponding K varies in the range of 2 to 10. Each point is obtained from 100 Monte Carlo experiments. The relationship between the detection probability and the sparsity K is as follows: Figure 5 As shown. It can be seen from the figure that the present invention has better recognition performance when the sparsity K of the extended sound source increases. When K is low, that is, in the case of a small number of sources with K<6, the estimation performance of the general LASSO method is significantly better than that of the present invention and the improved ESPRIT method, and the detection probability fluctuates slightly between 97% and 100%. However, as K increases, the performance of the 2DTV-CES method is significantly better than the other two methods, and the detection probability gradually increases to about 98% and tends to be stable, proving that the present invention still has good estimation performance when the sparsity K is large.

[0177] Application Example 3:

[0178] This application example verifies the influence of different signal-to-noise ratios on the performance of the present invention: there are two extended sound sources, whose angle coverage DOA range is the same as that of application example 1 (1). The signal-to-noise ratio varies in the range of -5 to 20 dB.

[0179] The relationship between detection probability and signal-to-noise ratio is as follows: Figure 6 As shown in the figure, it can be seen that with the increase of the signal-to-noise ratio, the performance of the 2DTV-CES method of the present invention and the general LASSO method is significantly better than the improved ESPRIT method (the detection probability is up to 1%), among which the performance of the present invention is the best. In the case of exceeding the array element limit, with the gradual increase of the signal-to-noise ratio, there is still a high detection probability.

[0180] The present invention constructs an extended sound source model by utilizing the grouping and correlation characteristics of the sound source; realizes the two-dimensional sparse representation of the sound source by constructing a two-dimensional array manifold matrix and an overcomplete representation; realizes the structural constraint on the grouping characteristics of the sound source by introducing a two-dimensional total variation and constructing a total variation structure matrix, and successfully transforms the sparse constraint of the solution into the sparsity of the gradient space of the solution, thereby promoting the block sparsity of the solution; and simultaneously applies the block sparsity and pure sparsity constraints of the solution by combining the total variation regularization term with the l1 regularization term. The proposed algorithm has a certain robustness, and avoids the eigendecomposition, the deallocation step and the angle pairing of the high-dimensional sample covariance matrix in the traditional method. Through specific application examples, it is proved that compared with the existing two-dimensional DOA estimation method, the 2DTV-CES method of the present invention with an appropriate regularization parameter significantly improves the resolution and detection probability in DOA estimation in the case of extended sound sources.

[0181] Embodiment 2

[0182] This embodiment provides a total variation regularized direction finding system based on acoustic features, including:

[0183] A sound source acquisition module, configured to acquire an output sound signal of an incident sound source based on a planar array;

[0184] A sound source estimation module is configured to calculate an estimated value of a direction of arrival of an incident sound source based on a pre-built two-dimensional total variation sparse direction of arrival estimation model;

[0185] The construction process of the two-dimensional total variation sparse direction of arrival estimation model is specifically as follows:

[0186] Establish a spatially extended sound source model according to the spatial distribution characteristics of the target sound source;

[0187] A two-dimensional array manifold matrix is ​​constructed to obtain an overcomplete representation of the two-dimensional array manifold matrix, and a two-dimensional sparse representation of the incident sound source is determined based on the overcomplete representation of the two-dimensional array manifold matrix;

[0188] A two-dimensional total variation structure matrix is ​​constructed, and a two-dimensional total variation sparse direction of arrival estimation model is constructed based on the two-dimensional total variation structure matrix.

[0189] The examples and application scenarios implemented by the above modules and corresponding steps are the same, but are not limited to the contents disclosed in the above embodiment 1. It should be noted that the above modules as part of the system can be executed in a computer system such as a set of computer executable instructions.

[0190] The description of each embodiment in the above embodiments has different emphases. For parts not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0191] The proposed system can be implemented in other ways. For example, the system embodiment described above is only illustrative, and the division of the modules is only a logical function division. In actual implementation, there may be other division methods, such as multiple modules can be combined or integrated into another system, or some features can be ignored or not executed.

[0192] Embodiment 3

[0193] This embodiment provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the steps in the total variation regularized direction finding method based on acoustic features as described in the first embodiment above are implemented.

[0194] Embodiment 4

[0195] This embodiment provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the steps in the total variation regularized direction finding method based on acoustic features as described in the first embodiment above are implemented.

[0196] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of hardware embodiments, software embodiments, or embodiments combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage and optical storage, etc.) containing computer-usable program code.

[0197] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0198] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0199] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0200] A person skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program, and the program can be stored in a computer-readable storage medium, and when the program is executed, it can include the processes of the embodiments of the above-mentioned methods. The storage medium can be a disk, an optical disk, a read-only memory (ROM) or a random access memory (RAM), etc.

[0201] Although the above describes the specific implementation mode of the present invention in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art on the basis of the technical solution of the present invention without creative work are still within the scope of protection of the present invention.

Claims

1. A total variation regularized direction finding method based on acoustic features, characterized in that: include: Acquiring an output acoustic signal of an incident acoustic source based on a planar array; Calculate the estimated value of the direction of arrival of the incident sound source based on the pre-built two-dimensional total variation sparse direction of arrival estimation model; The construction process of the two-dimensional total variation sparse direction of arrival estimation model is specifically as follows: Establish a spatially extended sound source model based on the spatial distribution characteristics of the sound source; A two-dimensional array manifold matrix is ​​constructed to obtain an overcomplete representation of the two-dimensional array manifold matrix, and a two-dimensional sparse representation of the incident sound source is determined based on the overcomplete representation of the two-dimensional array manifold matrix; Construct a two-dimensional total variation structure matrix, and construct a two-dimensional total variation sparse direction of arrival estimation model based on the two-dimensional total variation structure matrix, specifically: Acquire acoustic signal sample data, analyze and extract block sparse features and correlation features of the acoustic signal; Based on the block sparse features and correlation features of the acoustic signal, a total variation structure matrix is ​​constructed; According to the total variation structure matrix combined with the sparse direction of arrival estimation model, the final two-dimensional total variation sparse direction of arrival estimation model is obtained, which is as follows: Where μ1 and μ2 are regularization parameters, the observation vector y is given, and the overcomplete array manifold matrix is Total variation structure matrix D tv , target solution s.

2. A total variation regularized direction finding method based on acoustic features as claimed in claim 1, characterized in that: The method of obtaining the output sound signal of the incident sound source based on the planar array is specifically as follows: The dimensions of the planar array are M×N and M=N, and the planar arrays are evenly distributed at intervals of 0.5λ, where λ is the wavelength of the incident wave; The planar array is used to receive and sample the incident sound source to obtain an output sound signal of the incident sound source.

3. A total variation regularized direction finding method based on acoustic features as claimed in claim 1, characterized in that: The spatially extended sound source model is established according to the spatial distribution characteristics of the sound source, specifically: According to the continuity characteristics of the spatial distribution of the sound source, a spatially extended sound source model is established; Among them, the spatial extension sound source model is as follows: Wherein, d is the packet length, assuming sampling time t, t = 1, 2, ..., L, L is the number of snapshots, there are K far-field narrow-band sound sources, and K = n × d, n is the index.

4. A total variation regularized direction finding method based on acoustic features as claimed in claim 1, characterized in that: A two-dimensional array manifold matrix is ​​constructed to obtain an overcomplete representation of the two-dimensional array manifold matrix. Based on the overcomplete representation of the two-dimensional array manifold matrix, a two-dimensional sparse representation of the incident sound source is determined, specifically: Based on the output sound signal sample data of the incident sound source, a two-dimensional array manifold matrix is ​​constructed; Based on the gridded observation space, an overcomplete representation of the two-dimensional array manifold matrix is ​​constructed; Based on the overcomplete representation of the two-dimensional array manifold matrix, a two-dimensional sparse representation of the incident sound source is obtained.

5. A total variation regularized direction finding method based on acoustic features as claimed in claim 4, characterized in that: The output sound signal sample data based on the incident sound source is used to construct a two-dimensional array manifold matrix, specifically: Get the time difference of the incident sound source from the first sensor to the Xth sensor position; Based on the time difference, the array manifold matrices of the incident sound source in the x-axis and y-axis directions are obtained respectively; According to the definition of two-dimensional generalized direction vector, the two-dimensional direction vector of the plane rectangular array is obtained; Convert the two-dimensional direction vector of a planar rectangular array into a two-dimensional array manifold matrix.

6. A total variation regularized direction finding method based on acoustic features as claimed in claim 1, characterized in that: The estimated value of the direction of arrival of the incident sound source is calculated based on the pre-built two-dimensional total variation sparse direction of arrival estimation model, specifically: The matrix optimization calculation method is used to quickly solve the two-dimensional total variation sparse arrival direction estimation model and obtain the two-dimensional arrival estimation value of the incident sound source in the pitch angle and azimuth direction. The matrix optimization calculation method is specifically as follows: Obtain the array structure, the angle range of the sound source distribution, the sound propagation speed and the wavelength, and construct an overcomplete representation of the array matrix from the two-dimensional planar array manifold; Construct a two-dimensional total variation matrix using the generalized lasso matrix and block sparse features; Constructing a two-dimensional sparse DOA estimation model from a two-dimensional total variation matrix Assign values ​​to μ1 and μ2; The spatial spectrum of the sound source distribution is generated by using the interior point of convex optimization to search for the maximum value; The direction of arrival of the sound source is accurately estimated from the spatial spectrogram.

7. A total variation regularized direction finding system based on acoustic features, characterized in that: include: A sound source acquisition module, configured to acquire an output sound signal of an incident sound source based on a planar array; A sound source estimation module is configured to calculate an estimated value of the direction of arrival of an incident sound source based on a pre-built two-dimensional total variation sparse direction of arrival estimation model; The construction process of the two-dimensional total variation sparse direction of arrival estimation model is specifically as follows: Establish a spatially extended sound source model based on the spatial distribution characteristics of the sound source; A two-dimensional array manifold matrix is ​​constructed to obtain an overcomplete representation of the two-dimensional array manifold matrix, and a two-dimensional sparse representation of the incident sound source is determined based on the overcomplete representation of the two-dimensional array manifold matrix; Construct a two-dimensional total variation structure matrix, and construct a two-dimensional total variation sparse direction of arrival estimation model based on the two-dimensional total variation structure matrix, specifically: Acquire acoustic signal sample data, analyze and extract block sparse features and correlation features of the acoustic signal; Based on the block sparse features and correlation features of the acoustic signal, a total variation structure matrix is ​​constructed; According to the total variation structure matrix combined with the sparse direction of arrival estimation model, the final two-dimensional total variation sparse direction of arrival estimation model is obtained, which is as follows: Where μ1 and μ2 are regularization parameters, the observation vector y is given, and the overcomplete array manifold matrix is Total variation structure matrix D tv , target solution s.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps in the total variation regularized direction finding method based on acoustic features as described in any one of claims 1 to 6 are implemented.

9. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the steps in the total variation regularized direction finding method based on acoustic features as described in any one of claims 1 to 6 are implemented.

Images