A low-frequency multi-point out-of-plane sound source 3D imaging method

Abstract

This invention discloses a 3D imaging method for low-frequency multi-point heterogeneous sound sources, employing a 3D spherical array for asynchronous measurement. A specific convex cylindrical surface movement method is proposed. Then, the energy spectrum matrix is ​​updated using the time-space-frequency correlation characteristics of signals acquired by a multi-channel microphone array. Simultaneously, a fast iterative threshold shrinkage algorithm is used to further fill the synthesized energy spectrum matrix. Finally, a multi-signal classification algorithm for asynchronous cylindrical surface measurement is proposed, enabling rapid 3D point source imaging of multiple heterogeneous low-frequency sound sources. This method differs from existing sound source localization methods. It belongs to the category of low-frequency multi-point heterogeneous sound source 3D imaging methods based on asynchronous measurement of convex cylindrical surface trajectories. The method is stable and efficient. Compared with traditional methods, it extends from two-dimensional imaging to three-dimensional imaging, specifically three-dimensional imaging of low-frequency point sound sources. It offers good real-time performance, simplicity, and high resolution.

Description

Technical Field

[0001] This invention relates to the field of signal processing technology, and in particular to a method for 3D imaging of low-frequency multi-point heterogeneous sound sources. Background Technology

[0002] Beamforming is an important method for acoustic imaging or reconstruction, and it has been widely used in sound source localization and noise reduction. A beamforming algorithm can be described as all microphones in a plane that simultaneously record the source signal, and then the source location is determined by maximizing the result of the beamformer.

[0003] Asynchronous measurement (NSM) is an effective method for achieving low-frequency acoustic localization performance through sequential scanning of the sound field. There is evidence that asynchronous measurement of a moving microphone array can improve the accuracy of sound source localization in a two-dimensional plane. However, existing methods can only perform sound source imaging at frequencies above 2000 Hz.

[0004] In noise source identification and localization, sound source imaging research generally uses planar arrays to image in a two-dimensional plane. Among them, asynchronous measurement mainly involves planar movement. The asynchronous measurement method of two-dimensional planar arrays sequentially moves the planar array to scan the spatially distributed sound sources, approximates the measurement of large-aperture and high-density microphone arrays, and then uses traditional beamforming algorithms to image and locate the sound source.

[0005] Currently, the asynchronous measurement method of 3D spherical microphone arrays in three-dimensional planar imaging still follows the method of planar arrays in far-field two-dimensional planar imaging. It adopts horizontal movement, which improves the spatial resolution of CBF to a certain extent, but does not make any innovation in 3D space. Furthermore, the existing literature has not yet shown the use of cylindrical surfaces as the movement trajectory. Summary of the Invention

[0006] The purpose of this invention is to provide a method for constructing a 3D asynchronous measurement based on convex cylindrical trajectory using a 3D spherical microphone array, which further improves the resolution at multiple points of non-convexity at low frequencies (below 500Hz), facilitating large-scale and long-distance measurements.

[0007] To achieve the above objectives, the present invention provides the following technical solution: a low-frequency multi-point heterogeneous sound source 3D imaging method, comprising the following steps:

[0008] S01: Using the convex cylindrical surface formula to construct the trajectory of the 3D spherical microphone array, perform asynchronous measurement of convex cylindrical surface-3D.

[0009] S02: Conventional asynchronous measurement achieves a larger and denser aperture array by moving a single planar array. Convex cylindrical-3D asynchronous measurement moves a 3D spherical microphone array along a specific cylindrical trajectory. This convex cylindrical movement trajectory can significantly improve radial positioning capability and is beneficial for 3D spatial sound source positioning.

[0010] S03: Preprocess the acquired signal to cancel out side lobes, eliminate high-frequency interference and energy leakage, and construct the cylindrical asynchronous measurement-energy spectrum matrix in a specific way based on the principle of cylindrical asynchronous measurement-energy spectrum matrix.

[0011] S04: Using a multi-signal classification algorithm for asynchronous cylindrical measurement, obtain a 3D low-frequency multi-point heterogeneous sound source localization imaging result map. Then, compare the sound source positions obtained from the upper and lower convex surfaces with the actual simulated low-frequency multi-point heterogeneous positions and relative sound source amplitudes.

[0012] As a further description of the above technical solution:

[0013] In step S01, the sound source 3D imaging method is an asynchronous measurement method that uses points arranged to form a cylindrical surface. This is an improvement on asynchronous measurement methods in real-world scenarios. The formula for the trajectory of the convex cylindrical surface is:

[0014]

[0015] Where X represents the x-axis displacement, Y represents the y-axis displacement, R represents the radius, h represents the cylinder height range, θ represents the angle range, and n is a positive integer. The general binary expression for a cylinder is as follows:

[0016] F = (X + R) 2 +Y 2 -R 2

[0017] The formula for determining the lower convex surface is:

[0018] F(λ*X1+(1-λ)X2, λ*Y1+(1-λ)Y2)≤λ*F(X1, Y1)+(1-λ)*F(X2, Y2)

[0019] Where λ∈[0,1], satisfying the above conditions, it is defined as a lower convex cylinder; otherwise, it is an upper convex cylinder.

[0020] As a further description of the above technical solution:

[0021] In step S02, for low-frequency multi-point non-plane measurements, firstly, assuming the sound source is a steady-state sound source, the 3D spherical array is moved sequentially to L positions along convex cylindrical trajectories with equal angles and equal arc lengths, and the source at each position is measured. It is assumed that there is a Q-uncorrelated far-field narrowband signal on the source plane. Therefore, in the asynchronous cylindrical measurement, the {l}th...th The data vector received at positions (l = 1, ..., L) can be represented in the frequency domain as:

[0022] P {l} =G {l} S+E {l}

[0023] Where S = [s1, ..., s Q ] T ∈C Qx1 P is the far-field narrowband uncorrelated signal in the frequency domain; E is the received signal vector; E represents the uncertainty of the model, including noise interference and model approximation, assuming that the noise is independent and identically distributed, and follows a complex Gaussian distribution random variable. This is the guide vector for the spatial array;

[0024]

[0025] in, t and t′ represent the position vector of the sound source and the signal emission time, respectively. Let t and t represent the position vector of the receiver and the signal reception time, respectively, |·| represent the magnitude of the vector, and δ(.) represent the Dirac function.

[0026] As a further description of the above technical solution:

[0027] In step S03, the spectral matrix is ​​filled according to a specific construction and algorithm. The specific steps can be expressed as follows: The second step of asynchronous cylindrical measurement is covariance matrix completion. Using asynchronous cylindrical measurement, the small covariance matrix of each measurement position in the convex cylindrical trajectory can be obtained. The received signal data vector P {l} The covariance matrix is ​​expressed as

[0028]

[0029] in, It is the covariance matrix of uncorrelated sources. q th The power of the source can be directly obtained as a block diagonal matrix. The composite covariance matrix representing missing terms has its off-diagonal blocks empty, replaced by zeros due to the lack of phase relationships between consecutive positions. Then, a complete composite covariance matrix is ​​obtained. The matrix can be obtained. Finally, the covariance matrix needs to be analyzed. The goal of continuing with fast eigenvalue decomposition of sparse symmetric matrices is to enable the algorithm to perform fast and accurate imaging.

[0030] Furthermore, the covariance matrix completion problem can be formulated as a constrained optimization problem. The fast iterative shrinkage threshold analysis algorithm is used to solve this problem. Compared to simultaneous measurement, this algorithm inevitably leads to matrix completion errors. A complete optimization model for the covariance matrix is ​​as follows:

[0031]

[0032]

[0033] Among them ||·|| * The nuclear norm of a matrix is ​​denoted as . It is defined as the eigenvalue λ i The sum, minimizing the objective function It is to find the one with the smallest nuclear norm. A(·) represents the sampling operator that retrieves elements from the diagonal block of a matrix.

[0034] As a further description of the above technical solution:

[0035] In step S04, based on the specific construction and algorithm filling of the spectral matrix, the NCSM-MUSIC-3D imaging process can be represented as follows:

[0036]

[0037] in, for The vector form of U is calculated from the measured sound pressure level, which is derived from the actual position of the microphone array at each NCSM location. ss The subspace spanned by the eigenvectors corresponding to the eigenvalues ​​in the equation.

[0038] As a further description of the above technical solution:

[0039] In step S04, the advantage of obtaining low-frequency sound source imaging results through the upper and lower convex surfaces compared to planar low-frequency sound source imaging is achieved through the L2 norm error:

[0040]

[0041]

[0042] The low-frequency multi-point heterogeneous sound source 3D imaging method provided by the present invention, as described above, has the following beneficial effects:

[0043] This low-frequency multi-point non-planar sound source 3D imaging method utilizes cylindrical trajectory movement to approximate the measurement results of large-aperture, high-density microphone arrays for multi-point non-planar sound sources using convex cylindrical trajectory-asynchronous measurement technology. Finally, it can quickly image at low-frequency multi-point non-planar locations through the 3D-NCSM-MUSIC algorithm, successfully improving the current effect that three-dimensional space can only be imaged at high frequencies. It truly achieves high resolution for point sound sources, which is beneficial for large-scale, long-distance measurements. Attached Figure Description

[0044] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0045] Figure 1 This is a flowchart illustrating a low-frequency multi-point heterogeneous sound source 3D imaging method provided in an embodiment of the present invention;

[0046] Figure 2 The actual design of the 600mm diameter 3D spherical array diagram provided for the embodiments of the present invention;

[0047] Figure 3 The distribution diagram of the center point coordinates O(0, 0, 1) of a 600mm diameter 3D spherical array provided in the embodiments of the present invention;

[0048] Figure 4 The distribution diagram of moving points (5 rows and 9 columns, 45 points) for asynchronous measurement of the convex cylindrical surface and the trajectory of the convex cylindrical surface provided in the embodiments of the present invention;

[0049] Figure 5 A distribution diagram showing the vertical movement of the center point of the spherical array along a convex cylindrical surface within the height h∈[-0.2, 0.2], provided in an embodiment of the present invention.

[0050] Figure 6 The diagram shows the vertical movement of the center point of the spherical array provided in the embodiment of the present invention along the convex cylindrical surface in numerical order within the height h∈[-0.2, 0.2].

[0051] Figure 7 A distribution diagram showing the movement of the center point of the spherical array along a convex cylindrical surface in a digital horizontal arc within the angle t∈[-π / 3, π / 3], provided in an embodiment of the present invention.

[0052] Figure 8 A distribution diagram of the movement of the center point of the spherical array along a single horizontal arc on the convex cylindrical surface within the angle t∈[-π / 3, π / 3], provided in an embodiment of the present invention.

[0053] Figure 9 The following diagrams are provided for embodiments of the present invention: the location of the moving point (3 rows and 5 columns, 15 points) of the center point O of the spherical array, the distribution of the real sound sources (2 sound sources on each of the 6 sides, 0.2m apart), and the lower convex front view of the asynchronous measurement movement trajectory of the convex cylindrical surface.

[0054] Figure 10 A schematic diagram of the 500Hz-3D imaging result corresponding to the convex front surface under the asynchronous measurement and movement trajectory of the convex cylindrical surface provided in an embodiment of the present invention;

[0055] Figure 11 The diagram shows the location of the moving point (3 rows and 5 columns, 15 points) of the center point O of the spherical array and the distribution of the real sound sources (2 sound sources on each of the 6 sides, 0.2m apart), as well as the upper convex front view of the asynchronous measurement movement trajectory of the convex cylindrical surface, provided for the embodiments of the present invention.

[0056] Figure 12 This is a schematic diagram of the 500Hz-3D imaging result corresponding to the convex front side on the asynchronous measurement movement trajectory of the convex cylindrical surface provided in an embodiment of the present invention. Detailed Implementation

[0057] To enable those skilled in the art to better understand the technical solution of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings.

[0058] like Figures 1-12 As shown, a 3D imaging method for low-frequency multi-point skew-face sound sources includes the following steps:

[0059] S01: Using the convex cylindrical surface formula to construct the trajectory of the 3D spherical microphone array, perform asynchronous measurement of convex cylindrical surface-3D.

[0060] S02: Conventional asynchronous measurement achieves a larger and denser aperture array by moving a single planar array. Convex cylindrical-3D asynchronous measurement moves a 3D spherical microphone array along a specific cylindrical trajectory. This convex cylindrical movement trajectory can significantly improve radial positioning capability and is beneficial for 3D spatial sound source positioning.

[0061] S03: Preprocess the acquired signal to cancel out side lobes, eliminate high-frequency interference and energy leakage, and construct the cylindrical asynchronous measurement-energy spectrum matrix in a specific way based on the principle of cylindrical asynchronous measurement-energy spectrum matrix.

[0062] S04: Using a multi-signal classification algorithm for asynchronous cylindrical measurement, obtain a 3D low-frequency multi-point heterogeneous sound source localization imaging result map. Then, compare the sound source positions obtained from the upper and lower convex surfaces with the actual simulated low-frequency multi-point heterogeneous positions and relative sound source amplitudes.

[0063] This low-frequency multi-point non-planar sound source 3D imaging method utilizes cylindrical trajectory movement to approximate the measurement results of large-aperture, high-density microphone arrays for multi-point non-planar sound sources using convex cylindrical trajectory-asynchronous measurement technology. Finally, it can quickly image at low-frequency multi-point non-planar locations through the 3D-NCSM-MUSIC algorithm, successfully improving the current effect that three-dimensional space can only be imaged at high frequencies. It truly achieves high resolution for point sound sources, which is beneficial for large-scale, long-distance measurements.

[0064] In step S01, the 3D imaging method for the sound source utilizes a non-synchronous measurement method that constructs a cylindrical surface from points. This is an improvement on the non-synchronous measurement method used in real-world scenarios. The trajectory formula for the convex cylindrical surface is:

[0065]

[0066] Where X represents the x-axis displacement, Y represents the y-axis displacement, R represents the radius, h represents the cylinder height range, θ represents the angle range, and n is a positive integer. The general binary expression for a cylinder is as follows:

[0067] F = (X + R) 2 +Y 2 -R 2

[0068] The formula for determining the lower convex surface is:

[0069] F(λ*X1+(1-λ)X2, λ*Y1+(1-λ)Y2)≤λ*F(X1, Y1)+(1-λ)*F(X2, Y2)

[0070] Where λ∈[0,1], satisfying the above conditions, it is defined as a lower convex cylinder; otherwise, it is an upper convex cylinder.

[0071] In step S02, for low-frequency multi-point non-plane signals, firstly, assuming the sound source is a steady-state sound source, the 3D spherical array is moved sequentially to L positions along a convex cylindrical trajectory with equal angles and equal arc lengths, and the source at each position is measured. It is assumed that there is a Q-uncorrelated far-field narrowband signal on the source plane. Therefore, in the asynchronous cylindrical measurement, the {l}th... th The data vector received at positions (l = 1, ..., L) can be represented in the frequency domain as:

[0072] P {l} =G {l} S+E {l}

[0073] Where S = [s1, ..., s Q ] T ∈C Qx1P is the far-field narrowband uncorrelated signal in the frequency domain; E is the received signal vector; E represents the uncertainty of the model, including noise interference and model approximation. It is assumed that the noise is independent and identically distributed, and follows a complex Gaussian distribution. This is the guide vector for the spatial array;

[0074]

[0075] in, t and t′ represent the position vector of the sound source and the signal emission time, respectively. Let t and t represent the position vector of the receiver and the signal reception time, respectively, |·| represent the magnitude of the vector, and δ(.) represent the Dirac function.

[0076] In step S03, the spectral matrix is ​​filled according to a specific construction and algorithm. The specific steps can be expressed as follows: The second step of the asynchronous cylindrical measurement is covariance matrix completion. The small covariance matrix of each measurement position in the convex cylindrical trajectory can be obtained using the asynchronous cylindrical measurement. The received signal data vector P {l} The covariance matrix is ​​expressed as

[0077]

[0078] in, It is the covariance matrix of uncorrelated sources. q th The power of the source can be directly obtained as a block diagonal matrix. The composite covariance matrix representing missing terms has its off-diagonal blocks empty, replaced by zeros due to the lack of phase relationships between consecutive positions. Then, a complete composite covariance matrix is ​​obtained. The matrix can be obtained. Finally, the covariance matrix needs to be analyzed. The goal of continuing with fast eigenvalue decomposition of sparse symmetric matrices is to enable the algorithm to perform fast and accurate imaging.

[0079] Furthermore, the covariance matrix completion problem can be formulated as a constrained optimization problem. The fast iterative shrinkage threshold analysis algorithm is used to solve this problem. Compared to simultaneous measurement, this algorithm inevitably leads to matrix completion errors. A complete optimization model for the covariance matrix is ​​as follows:

[0080]

[0081]

[0082] Among them ||·|| * The nuclear norm of a matrix is ​​denoted as . It is defined as the eigenvalue λ iThe sum, minimizing the objective function It is to find the one with the smallest nuclear norm. A(·) represents the sampling operator that retrieves elements from the diagonal block of a matrix.

[0083] In step S04, based on the specific construction and algorithm for filling the spectral matrix, the NCSM-MUSIC-3D imaging process can be represented as follows:

[0084]

[0085] in, for The vector form of U is calculated from the measured sound pressure level, which is derived from the actual position of the microphone array at each NCSM location. ss The subspace spanned by the eigenvectors corresponding to the eigenvalues ​​in the equation.

[0086] In step S04, the advantage of obtaining low-frequency sound source imaging results through the upper and lower convex surfaces compared to planar low-frequency sound source imaging is achieved through the L2 norm error:

[0087]

[0088]

[0089] The foregoing has only described certain exemplary embodiments of the present invention by way of illustration. Undoubtedly, those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the foregoing drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

Claims

1. A 3D imaging method for low-frequency multi-point skew-face sound sources, characterized in that, Includes the following steps: S01. Using the convex cylindrical surface formula to construct the trajectory of the moving 3D spherical microphone array, perform asynchronous measurement of convex cylindrical surface-3D. S02: Conventional asynchronous measurement achieves a larger and denser aperture array by moving a single planar array. Convex cylindrical-3D asynchronous measurement moves a 3D spherical microphone array along a specific cylindrical trajectory. This convex cylindrical movement trajectory can significantly improve radial positioning capability and is beneficial for 3D spatial sound source positioning. S03: Preprocess the acquired signal to cancel out side lobes, eliminate high-frequency interference and energy leakage, and construct the cylindrical asynchronous measurement-energy spectrum matrix in a specific way based on the principle of cylindrical asynchronous measurement-energy spectrum matrix. S04: Using a multi-signal classification algorithm for asynchronous cylindrical measurement, obtain a 3D low-frequency multi-point heterogeneous sound source localization imaging result map. Then, compare the sound source positions obtained from the upper and lower convex surfaces with the actual simulated low-frequency multi-point heterogeneous positions and relative sound source amplitudes.

2. The low-frequency multi-point skew-surface sound source 3D imaging method according to claim 1, characterized in that: The sound source 3D imaging method in step S01 is an asynchronous measurement method that uses points arranged to form a cylindrical surface. It is an improvement on the asynchronous measurement method in real-world scenarios. The trajectory formula for the convex cylindrical surface is: Where X represents the x-axis displacement, Y represents the y-axis displacement, R represents the radius, h represents the cylinder height range, θ represents the angle range, and n is a positive integer. The general binary expression for a cylinder is as follows: The formula for determining the lower convex surface is: F(λ*X1+(1-λ)X2, λ*Y1+(1-λ)Y2)≤λ*F(X1, Y1)+(1-λ)*F(X2, Y2) Where λ∈ [0, 1], satisfying the above conditions, it is defined as a lower convex cylinder; otherwise, it is an upper convex cylinder.

3. The low-frequency multi-point skew-surface sound source 3D imaging method according to claim 1, characterized in that: In step S02, for low-frequency multi-point non-plane signals, firstly, assuming the sound source is a steady-state sound source, the 3D spherical array is moved sequentially to L positions along convex cylindrical trajectories with equal angles and equal arc lengths, and the source at each position is measured. It is assumed that there is a Q-uncorrelated far-field narrowband signal on the source plane. Therefore, in the asynchronous cylindrical measurement... The data vector received at the location can be represented in the frequency domain as: in, is the far-field narrowband uncorrelated signal in the frequency domain; P is the received signal vector; E represents the uncertainty of the model, including noise interference and model approximation. It is assumed that the noise is independent and identically distributed, and follows a complex Gaussian distribution. This is the guide vector for the spatial array; in, and These represent the position vector of the sound source and the signal emission time, respectively. and t represent the position vectors of the receiving end, respectively. The magnitude and signal reception time are given by the vector, |·| represents the magnitude of the vector, and δ(.) represents the Dirac function.

4. The low-frequency multi-point skew-surface sound source 3D imaging method according to claim 3, characterized in that: In step S03, the spectral matrix is ​​filled according to a specific construction and algorithm. The specific steps can be expressed as follows: The second step of asynchronous cylindrical measurement is covariance matrix completion. Asynchronous cylindrical measurement can be used to obtain the small covariance matrix at each measurement position in the convex cylindrical trajectory. Difference matrix The covariance matrix of the received signal data vector P{l} is represented as follows: in, It is the covariance matrix of uncorrelated sources. Let represent the power of the qth source; therefore, a block diagonal matrix can be directly obtained. The composite covariance matrix representing missing terms has its off-diagonal blocks empty, replaced by zeros due to the lack of phase relationships between consecutive positions. Then, a complete composite covariance matrix is ​​obtained. The matrix can be obtained. Finally, the covariance matrix needs to be analyzed. The goal of continuing with fast eigenvalue decomposition of sparse symmetric matrices is to enable the algorithm to perform fast and accurate imaging. Furthermore, the covariance matrix completion problem can be formulated as a constrained optimization problem. The fast iterative shrinkage threshold analysis algorithm is used to solve this problem. Compared to simultaneous measurement, this algorithm inevitably leads to matrix completion errors. A complete optimization model for the covariance matrix is ​​as follows: Where || · ||* denotes the nuclear norm of a matrix. It is defined as the sum of eigenvalues ​​λi. Minimizing the objective function is to find the function with the smallest nuclear norm. A(·) represents obtaining the diagonal of the matrix. The sampling operator for elements in a block.

5. A low-frequency multi-point skew-surface sound source 3D imaging method according to claim 4, characterized in that... In S04, based on the specific construction and algorithmic filling of the spectral matrix, the NCSM-MUSIC-3D imaging process can be represented as follows: in, for The vector form of is calculated from the measured sound pressure, which is calculated from the actual position of the microphone array at each NCSM position. U is the subspace spanned by the eigenvectors corresponding to the eigenvalues ​​in Rss.

6. The low-frequency multi-point skew-surface sound source 3D imaging method according to claim 1, characterized in that: In step S04, the advantage of obtaining low-frequency sound source imaging results through the upper and lower convex surfaces compared to planar low-frequency sound source imaging is achieved through the L2 norm error:

Images