Passive ultrasonic sound source positioning method and system, computer equipment and medium

Abstract

The invention provides a passive ultrasonic sound source positioning method and system, computer equipment and a medium, and belongs to the field of array signal processing.The method comprises the steps that a planar microphone array is used for collecting multiple sets of snapshot signals, the signals are combined randomly, a sample covariance matrix of each combination is calculated, a Frobenius norm is used for optimizing a correction matrix, a power spectrum is extracted, and a passive ultrasonic sound source is obtained; counting the stability weight in the effective direction and carrying out weighted fusion on the stability weight and the average power to obtain a reconstructed power spectrum; constructing a local grid by taking a peak value of a reconstructed power spectrum as a center and a Rayleigh limit as a range, initializing a sparse Bayesian learning prior variance vector into an all-one vector, and taking a mean value of diagonal elements of a diagonal load shedding matrix as a noise variance estimated value; and executing SBL iterative optimization based on the prior, the local grid and the snapshot signal, and outputting the direction of arrival of the sound source. The method does not depend on an array geometric structure, can distinguish multiple close-range sound sources with the angle interval smaller than the Rayleigh limit, and gives consideration to positioning precision and real-time performance.

Description

Technical Field

[0001] This invention belongs to the field of array signal processing, specifically relating to a method, system, computer equipment, and medium for locating a passive ultrasonic source. Background Technology

[0002] Direction of Arrival (DOA) estimation is a crucial research area in array signal processing, with wide applications in radar, sonar, navigation, wireless communication, and speech processing. Among these, passive sound source localization (DAO) based on microphone arrays, which utilizes acoustic signals generated in specific scenarios (such as ultrasound) combined with DOA methods for direction estimation to locate the sound source, is a key technique in industrial non-destructive testing and fault location. Typical applications include partial discharge detection in electrical equipment, bearing defect detection, and gas pipeline leak location.

[0003] Among numerous DOA estimation methods, beamforming algorithms are widely used due to their computational simplicity and robustness. Conventional Beamforming (CBF) boasts the highest robustness with low computational cost, but it suffers from low azimuth resolution due to Rayleigh constraint, making it difficult to distinguish multiple sound sources at similar angles. Minimum Variance Distortionless Response (MVDR), through adaptive weighting, theoretically achieves higher resolution. However, the performance of both CBF and MVDR heavily depends on the accuracy of the received signal covariance matrix estimation. With a low snapshot number, a significant estimation error exists between the sample covariance matrix estimated by a finite number of snapshots and the ideal matrix. Low signal-to-noise ratio further amplifies this error, thus affecting the performance of subsequent beamforming algorithms.

[0004] Because improving the estimation quality of the sample covariance matrix can effectively improve the performance of subsequent beamforming algorithms, numerous studies have focused on finding better matrix estimates under different conditions. Among these, diagonal offloading methods improve array gain by subtracting diagonal noise terms; however, under low snapshot conditions, off-diagonal elements of the sample covariance matrix introduce estimation errors, and simple diagonal term processing cannot handle these errors. Reconstruction methods utilizing special array structures (such as those based on Hermitian-Toeplitz properties) can effectively utilize array priors, but they heavily rely on regular array geometry and are not applicable to arrays with arbitrary structures. Interference-plus-noise covariance matrix reconstruction methods aim to improve the robustness of the MVDR algorithm under strong interference environments, but are not optimal under interference-free or weak interference environments. Furthermore, while semi-parametric methods such as SPICE and LIKES offer a covariance matrix processing approach based on covariance fitting criteria—which iteratively minimizes the deviation between the ideal covariance and the sample covariance to obtain better DOA estimation results—their iterative process requires frequent calculation of matrix inverses. When the number of snapshots is less than the number of matrix elements, to avoid the singularity of the sample covariance matrix, the optimal master fitting criterion must be abandoned, and an alternative criterion based on the Frobenius norm of the data matrix must be adopted instead. This alternative criterion is statistically suboptimal, and its theoretical performance is clearly inferior to the master criterion used when there are sufficient snapshots.

[0005] In summary, existing methods suffer from significant source signal localization errors when dealing with ultrasonic sources with low snapshot speeds and low signal-to-noise ratios. Summary of the Invention

[0006] To address the aforementioned problems, this invention provides a method, system, computer device, and medium for locating a passive ultrasonic source.

[0007] To achieve the above objectives, the present invention provides a method for locating a passive ultrasonic sound source, comprising: N sets of snapshot sound signals of the target ultrasonic source are acquired using a planar microphone array.

[0008] The N sets of snapshot sound signals are randomly combined in multiple groups without repetition; the covariance matrix of each combination is obtained, and the power spectrum of each covariance matrix is ​​extracted in a preset scanning direction; for each scanning direction, the number of times the power value in all power spectra in that direction is greater than a set power threshold is counted. When the count is greater than the set threshold, the scanning direction is determined as a valid direction, and the count is used as the stability weight of the corresponding scanning direction; the average power of each valid direction is calculated; based on the stability weight and the average power, a reconstructed power spectrum is generated.

[0009] The first M peaks of the reconstructed power spectrum are extracted, and multiple local grids are constructed with each peak as the center and Rayleigh limit as the range. Based on each local grid, the prior variance vector of the sparse Bayesian learning algorithm is defined. The noise variance estimate of the covariance matrix is ​​obtained. Based on the multiple prior variance vectors, the noise variance estimate, and N sets of snapshot sound signals, the sparse Bayesian learning algorithm is iteratively optimized. When the iteration condition is met, the direction of arrival estimate of the target ultrasonic source is output according to the optimization result. The direction of arrival estimate is used to locate the target ultrasonic source.

[0010] Preferably, before extracting the power spectrum of each covariance matrix in the preset scanning direction, the method further includes: using a diagonal load reduction method based on Frobenius norm optimization to separate the diagonal load reduction matrix from each covariance matrix to obtain multiple corrected covariance matrices, and then extracting the power spectrum of each corrected covariance matrix.

[0011] Preferably, before counting the number of times the power value in all power spectra in each scanning direction is greater than a set power threshold, the method further includes shielding the interference in the negative power direction in each power spectrum, wherein the negative power direction is determined to be a non-signal direction.

[0012] Preferably, the total number of points in the local grid does not exceed 200; the Rayleigh limit is calculated from the coordinates of the array elements of the planar microphone array and the signal frequency of the target ultrasonic source; the array elements of the planar microphone array are irregularly distributed in space.

[0013] Preferably, extracting the first M peaks of the reconstructed power spectrum specifically includes: sorting the power values ​​from high to low in the reconstructed power spectrum, and selecting the first M peaks with the largest power values, where M is a preset estimated value of the number of sound sources or a value automatically determined based on the significance of the peaks.

[0014] Preferably, the iterative optimization of the sparse Bayesian learning algorithm specifically includes: cyclically calculating the posterior mean and posterior covariance based on the prior variance vector and the noise variance estimate; updating the sparse weights in the prior variance vector and the noise variance in the noise variance estimate based on the posterior mean and posterior covariance; and terminating the iteration when the change in noise variance before and after the update is less than a preset convergence threshold.

[0015] Preferably, the total number of snapshots in the N groups of snapshot sound signals does not exceed 200.

[0016] The present invention also provides a positioning system for a passive ultrasonic sound source, comprising: The data acquisition module is used to acquire N sets of snapshot sound signals from the target ultrasonic source using a planar microphone array.

[0017] The reconstruction module is used to perform multiple non-repeating random combinations of the N sets of snapshot sound signals; obtain the covariance matrix of each combination, and extract the power spectrum of each covariance matrix in a preset scanning direction; for each scanning direction, count the number of times the power value in all power spectra in that direction is greater than a set power threshold. When the count is greater than the set threshold, the scanning direction is determined as a valid direction, and the count is used as the stability weight of the corresponding scanning direction; calculate the average power of each valid direction; and generate the reconstructed power spectrum based on the stability weight and the average power.

[0018] The localization module is used to extract the first M peaks of the reconstructed power spectrum, construct multiple local grids with each peak as the center and Rayleigh limit as the range; define the prior variance vector of the sparse Bayesian learning algorithm based on each local grid; obtain the noise variance estimate of the covariance matrix; and perform iterative optimization of the sparse Bayesian learning algorithm based on the multiple prior variance vectors, the noise variance estimate, and N sets of snapshot sound signals. When the iteration condition is met, the direction of arrival estimate of the target ultrasonic source is output according to the optimization result, and the localization of the target ultrasonic source is achieved by using the direction of arrival estimate.

[0019] The present invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement any of the steps in the passive ultrasonic source localization method.

[0020] The present invention also provides a computer-readable storage medium storing a computer program that, when loaded by a processor, can execute any of the steps in the passive ultrasonic source localization method.

[0021] The passive ultrasonic sound source localization method provided by this invention has the following beneficial effects: By performing multiple non-repeating random combinations on N sets of snapshot signals, estimation distortion under limited snapshots is reduced. Then, by statistically filtering effective directions and combining stability weights with average power, noise interference is suppressed for low snapshots and low signal-to-noise ratios, generating a robust reconstructed power spectrum. A grid is constructed based on the reconstructed power spectrum to ensure fair angle search within the local grid, resolving optimization bias caused by inaccurate priors in traditional SBL. Simultaneously, the search range is narrowed through the local grid, balancing accuracy and receiver positioning efficiency. The reconstructed power spectrum and SBL prior optimization can resolve multiple near-field sound sources with angular intervals smaller than the Rayleigh limit, breaking through the resolution bottleneck of traditional methods. The local grid replaces the full-space grid, reducing the computational load of SBL iterations and achieving effective localization of passive sound sources with low snapshots and low signal-to-noise ratios. Attached Figure Description

[0022] To more clearly illustrate the embodiments and design schemes of the present invention, the accompanying drawings required for this embodiment will be briefly described below. The drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 This is a flowchart illustrating a method for locating a passive ultrasonic sound source according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the array element distribution of a planar microphone array according to an embodiment of the present invention; wherein, Figure 2 (a) is a three-dimensional schematic diagram. Figure 2 (b) is the main view; Figure 3 This is a schematic diagram of the coordinate system and azimuth and elevation angles according to an embodiment of the present invention; Figure 4 This is a comparison chart of the mean peak-to-sidelobe ratio under different signal-to-noise ratios according to an embodiment of the present invention; Figure 5 This is a comparison chart of MAE estimation for DOA under different power thresholds in this embodiment of the invention. Figure 6 This is a comparison diagram of power spectrum diagrams of embodiments of the present invention; wherein, Figure 6 (a) is the CBF power spectrum, and (b) is the fusion power spectrum. Figure 7 The figures show the positioning results of azimuth and pitch angles at different signal-to-noise ratios according to an embodiment of the present invention; wherein, Figure 7 (a) is a comparison chart of azimuth MAE, and (b) is a comparison chart of pitch MAE. Figure 8 This is a schematic cross-sectional view of a long-spaced dual-source beammap according to an embodiment of the present invention; Figure 9 This is a schematic cross-sectional view of a near-spaced dual-source beammap according to an embodiment of the present invention. Detailed Implementation

[0024] To enable those skilled in the art to better understand and implement the technical solutions of the present invention, the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. The following embodiments are only used to more clearly illustrate the technical solutions of the present invention and should not be construed as limiting the scope of protection of the present invention.

[0025] This method first employs a diagonal underloading technique to separate and remove noise power from the sample covariance matrix, obtaining a pre-corrected sample covariance matrix. Then, conventional beamforming is performed using the diagonally underloaded covariance matrix to achieve global negative power shielding. Next, conventional beamforming is performed using different combinations of finite snapshots, and a robust power spectrum is reconstructed based on statistical properties to reliably determine the main regions where the sound source exists. Finally, the reconstructed power spectrum is used to determine the credible region where the sound source exists. Within this region, data-driven super-resolution estimation is achieved through sparse Bayesian learning, thus balancing the algorithm's robustness and high resolution.

[0026] This invention provides a method for locating a passive ultrasonic sound source, specifically as follows: Figure 1 As shown, it includes the following steps: S1. Use a planar microphone array to collect N sets of snapshot sound signals from the target ultrasonic source.

[0027] Within a short period of time, N sets of snapshot sound signals from the target ultrasonic source are acquired using a planar microphone array. Random far-field sound sources are set up. For microphone array aperture, The wavelength of the sound wave is , and the distance between the sound source and the center of the microphone array is . ,when When the sound source is in the near field, it is considered to be in the far field; otherwise, it is considered to be in the far field. A rectangular coordinate system is established with the center of the planar microphone array as the origin. The schematic diagram of the model is shown below. Figure 2 As shown in (a), it comes from A schematic diagram for viewing along the positive axis is shown below. Figure 2 (b)

[0028] Coordinate system definition as follows Figure 3 As shown, the plane wave far-field assumption is adopted. Let be the distance from the sound source to the center of the planar array. Then, the direction vector of the sound source can be expressed as: Pitch angle The direction vector of the sound source and Angle along the positive axis, azimuth angle The direction vector of the sound source is in Plane projection and The angle between the positive axis and the positive axis.

[0029] Assuming that in the far-field plane wave, there exists The frequency is For a single-frequency sound source, the spatial arrival angle of the array receiver is... The signal vector can be represented as: (1) In the formula, for The signal vector received by each array element, superscript Indicates matrix transpose. , For the first The signal when the sound source signal arrives at the array element The noise is additive white Gaussian noise that is uncorrelated with the signal, and the noise of each array element is independent of each other. Let be the guiding vector matrix, where To correspond to the steering vectors of different sound sources; for a planar array with irregularly distributed array elements, its phase difference can be expressed as shown in equation (2): (2) In the formula, Corresponding to different sound sources For the first The spatial coordinates of each array element and The first The azimuth and elevation angles of each sound source. Let be the wavelength of the sound wave, then the first... The steering vector corresponding to each sound source can be expressed as: .

[0030] The covariance matrix of the signal received by the planar microphone array can be expressed as shown in equation (3): (3) In the formula, This indicates the conjugate transpose. Let be the signal covariance matrix, where ; Let be the noise covariance matrix, where For noise power, for An identity matrix of order 1. In practical applications, matrices... The variance is generally unknown, so the sample covariance matrix is ​​used instead, as shown in equation (4):

[0031] (4) In the formula, This represents the number of snapshots. The power spectrum reconstruction and sparse Bayesian orientation estimation method based on statistical properties is elaborated as follows.

[0032] S2. Perform multiple non-repeating random combinations on the N groups of snapshot sound signals; obtain the covariance matrix of each combination, and extract the power spectrum of each covariance matrix in a preset scanning direction; for each scanning direction, count the number of times the power value in all power spectra in that direction is greater than a set power threshold. When the count is greater than the set threshold, the scanning direction is determined as a valid direction, and the count is used as the stability weight of the corresponding scanning direction; calculate the average power of each valid direction; generate a reconstructed power spectrum based on the stability weight and the average power.

[0033] Diagonal load reduction preprocessing and negative power shielding. Diagonal load reduction is a preprocessing method for the sample covariance matrix. It takes advantage of the fact that uncorrelated noise is mainly located on the diagonal of the covariance matrix when the number of snapshots is large. After estimating the noise power, the noise component on the diagonal is subtracted to improve the received signal-to-noise ratio. Common diagonal load reduction methods include uniform diagonal load reduction [8], that is, different diagonal elements have the same load reduction amount. However, since the power of the received noise of each array element is not equal, non-uniform diagonal load reduction is proposed, that is, each diagonal element has a different load reduction amount. This invention adopts the diagonal load reduction method based on Frobenius norm optimization as the preprocessing step. Its core lies in solving the following optimization problem.

[0034] (5) In the formula, For signal subspace estimation, This is the diagonal load reduction matrix. This optimization criterion achieves effective separation of the signal and diagonal noise in the maximum likelihood sense. Under finite snapshots, the sample covariance matrix... Unable to approximate the ideal covariance matrix Negative Gaussian white noise will introduce estimation errors at off-diagonal element positions. ,Right now Diagonal load reduction is achieved through stripping The diagonal noise term in the equation (5) is the result of solving for it. After being corrected .

[0035] (6) And use The power spectrum obtained by conventional beamforming will show negative power because if... There is an overestimation, which makes The diagonal elements have negative terms, and The non-diagonal terms are complex-valued random errors, and their superposition disrupts the... The positive semi-definiteness of the power spectrum leads to negative values.

[0036] Analyzing the direction where negative power occurs, let the steering vector in any spatial direction be... At this time, the power spectrum of conventional beamforming is expressed as: (7) In the formula, if If the signal direction is... Assuming the signal power is non-negative, middle This is the diagonal noise term separated under the constraint of preserving signal components. The optimization process dictates that its estimated value cannot be large enough to cancel out the signal power; therefore, this term is non-negative. because To conform to a zero-mean complex Gaussian random error, its sign is determined by the random phase, but its amplitude exceeding the signal power is a low-probability event. Therefore, even if the signal directional power value in the power spectrum after diagonal load reduction is not the maximum value, it is still a positive value with a certain amplitude.

[0037] like If it is a non-signal direction, then If the value is 0, the power value in that direction is entirely determined by the error term and the residual term after load reduction, so its sign is random.

[0038] To ensure its reliability, the sample covariance matrix was calculated using finite full-shot data and diagonally underloaded. The negative power direction in the power spectrum obtained by conventional beamforming can be determined to be a non-true signal direction. These directions can be completely shielded in subsequent processing to eliminate interference from erroneous directions in subsequent steps.

[0039] Power spectrum reconstruction based on statistical properties. Although the estimation error of the single-sample covariance matrix is ​​large under low snapshot conditions, the power values ​​in the signal direction of the power spectrum in different snapshot combinations exhibit statistical stability, while the error peaks formed by the coupling of estimation error and noise show random fluctuation characteristics. Based on this statistical property, this invention proposes a power spectrum reconstruction method.

[0040] Power spectrum generation using multiple combinations. An index set is constructed by using non-repeating combinations of different snapshots within a finite number of snapshots. , representing the corresponding snapshot index within different combinations, the total number of combinations is denoted as . Each set stores different combinations of snapshots, for each The sample covariance matrix of the snapshot combination is calculated as shown in equation (8):

[0041] (8) In the formula, This refers to a multi-channel array signal composed of different snapshots within a snapshot combination. Diagonal load reduction is obtained Then, the CBF power spectrum was calculated. .

[0042] Stability statistics and weighted reconstruction. Even when affected by noise, the signal directional power remains a positive value at its non-maximum value in the power spectrum, and it remains stable across different snapshot combinations. However, the signal directional power varies across different snapshot combinations. The results obtained The estimation error is random, which is reflected in the fact that the peak value in the power spectrum fluctuates in the power spectrum of different snapshot combinations. Therefore, a power threshold can be set. To count the results in each combination Exceeding the threshold The number of occurrences in the non-negative direction As a stability counter, and through statistical properties, the power in non-signal directions can be further zeroed out. Then, the stability and snapshot combination is applied to each... Perform weighted fusion.

[0043] (9) In the formula, As a stability weight, it measures the stability of this direction. The stability exhibited on the power spectrum of the combination indicates that the high-power direction with high stability is more likely to be the signal direction. The average power represents the sum of the power in the corresponding direction for different snap combinations.

[0044] Stability is used to weight and superimpose the power spectra of different snapshot combinations. A stable direction that consistently exceeds a threshold is considered to have a higher probability of being a signal direction, while a few high values ​​caused by random noise are suppressed by the stability weights. The selection process first involved negative power masking in the current full-fast-shot method, considering only the fusion of non-negative power directions. This resulted in a reconstructed power spectrum with significantly enhanced robustness. .

[0045] Sparse Bayesian. In single-source scenes or multi-source scenes with long-distance spacing, reconstructed power spectrum can already achieve good DOA estimation results in low snapshot and low signal-to-noise ratio environments. However, its angular resolution is still limited by the Rayleigh limit of CBF. To achieve close-range multi-source super-resolution, this invention introduces a sparse Bayesian learning framework and utilizes... Provide it with high-quality priors to solve its computational complexity and sensitivity to initial values.

[0046] S3. Extract the first M peaks of the reconstructed power spectrum, and construct multiple local grids with each peak as the center and Rayleigh limit as the range. Define the prior variance vector of the sparse Bayesian learning algorithm based on each local grid. Obtain the noise variance estimate of the covariance matrix. Based on the multiple prior variance vectors, the noise variance estimate, and N sets of snapshot sound signals, perform iterative optimization of the sparse Bayesian learning algorithm. When the iteration condition is met, output the direction of arrival estimate of the target ultrasonic source according to the optimization result. Use the direction of arrival estimate to locate the target ultrasonic source.

[0047] Local fine-grained mesh construction. Traditional SBL operates on a uniform mesh across the entire space, and the computational cost increases dramatically with the number of mesh points. This invention utilizes reconstructed fine-grained meshes... Before extraction Each peak has a specific value. Construct a space of size centered on . ( Local fine mesh (Rayleigh limit) The first M peaks of the reconstructed power spectrum are extracted. With each peak as the center, half a Rayleigh limit is extended in both the azimuth and elevation dimensions to construct multiple local fine grids. If there is overlap between local grids, they are merged into a continuous region.

[0048] Prior information settings. In the local mesh. The received signal model can be written as follows:

[0049] (10) In the formula, For local guided dictionaries, Let be the sparse signal matrix to be determined. The key to implementing SBL lies in the initialization of the prior variance vector. Initialization of noise variance estimates The accuracy of the reconstructed power spectrum is affected by Rayleigh limiting, as the beam direction corresponding to the local grid cannot provide accurate source location information due to peak fusion. In this case, directly initializing the sparse weights with the reconstructed power spectrum values ​​is problematic. This would cause the SBL method to converge prematurely to a local optimum, i.e., the fusion peak. The initial values ​​are set to all 1s, indicating that each direction within the current local grid has an equal probability of being a signal direction. The noise variance estimate is then calculated using a diagonal offload matrix. The mean of the diagonal elements is used as an initial estimate of the noise variance.

[0050] Based on the above three prior combinations, the SBL iteration process is shown in Table 1.

[0051] Table 1 SBL Iteration Steps The effectiveness of the proposed method was verified through simulation. No interference assumptions were made in the simulation, and the noise was complex Gaussian white noise. The simulation parameters are as follows: the array uses a planar spiral array as in the signal model, and the number of array elements is... The speed of sound is 64. 340 signal frequency 40 The azimuth and elevation scanning ranges are both 30°. Up to 150 The scan step size is 1. Quick shot number It is 50.

[0052] Verification of the effectiveness of diagonal load reduction preprocessing. To quantitatively evaluate its performance under different signal-to-noise ratio conditions, this experiment sets up a single sound source scenario, with the sound source orientation randomly generated within the scanning grid. 500 Monte Carlo experiments were conducted, with the signal-to-noise ratio varying from -5dB to 5dB. The peak-to-sidelobe ratio (PSLR) of the conventional beamforming (CBF) power spectrum before and after diagonal load reduction processing was compared.

[0053] Figure 4 The results show that, across the entire test signal-to-noise ratio range, the PSLR of the CBF power spectrum after diagonal underloading preprocessing is consistently lower than that of the original CBF result. This verifies that the method can remove some noise effects without affecting positioning accuracy, making it suitable as a robust preprocessing step in the subsequent power spectrum reconstruction process.

[0054] Sensitivity analysis of power threshold on reconstruction performance. In power spectrum reconstruction methods, the power threshold... This parameter is used to select beam directions participating in stability statistics, and its value directly affects the reconstruction quality. To evaluate the sensitivity of this parameter, this experiment, under the conditions of a signal-to-noise ratio of 0 dB and a snapshot number of 50, compared the results with those obtained by... The azimuth estimation accuracy of the reconstructed power spectrum was measured when the azimuth was set to -1dB, -3dB, and -6dB, respectively. 1000 Monte Carlo tests were conducted, and the mean absolute error (MAE) of the azimuth and elevation angles was used as the evaluation metric. The results are as follows: Figure 5 As shown.

[0055] Figure 5 power threshold The positioning accuracy of the reconstructed power spectrum is affected by non-monotonicity. Its value essentially reflects the trade-off between signal directional stability and noise randomness. When a lower power threshold is set for filtering (i.e....),... When the threshold is -6dB, most of the non-negative power directions participate in the stability statistics, resulting in a large amount of random fluctuations caused by noise being included in the reconstruction process. This dilutes the contribution of the true signal direction, causing a significant increase in the mean absolute error (MAE) of the azimuth and elevation angles, and a decrease in positioning reliability. Conversely, if the threshold is set too high (i.e., ... If the power threshold is -1dB, it will excessively discard some of the true signal directions affected by noise, resulting in insufficient stability counting and thus increasing the positioning error. Simulation results show that, under the current simulation settings, the power threshold... Setting it to -3dB effectively suppresses noise fluctuations while preserving the stable characteristics of the signal direction. This threshold was used in all subsequent experiments.

[0056] Robustness verification of power spectrum reconstruction under low signal-to-noise ratio (SNR) conditions. First, assuming a single source, the SNR is set to -4, and the snapshot combination is set to 100. Comparisons of conventional beamforming power spectrum and fused power spectrum are shown below. Figure 6 As shown. 1000 Monte Carlo tests were conducted to compare the performance of conventional beamforming and MVDR, and the fused power spectrum method, at different signal-to-noise ratios ranging from -5dB to 5dB in 1-step increments. With a positioning accuracy of 50, the comparison results are as follows: Figure 7 As shown.

[0057] Figure 6 This indicates that under low snapshot and low signal-to-noise ratio conditions, the robust CBF method has already made a positioning error due to the influence of estimation error and large noise power. However, the power spectrum fusion method combined with diagonal load reduction can correct the positioning error of the original CBF method through statistical characteristics and the stability of signal directional power. Figure 7 The results showed that, under low-speed shooting conditions (CBF, MVDR, and azimuth and elevation angles MAE of the proposed reconstruction method), the signal-to-noise ratio (SNR) varied from -5dB to 5dB. The MVDR algorithm suffers complete performance degradation due to severe distortion of the sample covariance matrix; while CBF remains robust, its estimation error is large; whereas the proposed method maintains the highest positioning accuracy under all signal-to-noise ratios. This fully demonstrates that, under low snapshot and low signal-to-noise ratio conditions, the proposed power spectrum reconstruction method can provide a more reliable prior for the sound source's location region in subsequent super-resolution processing.

[0058] Sparse Bayesian super-resolution performance based on reconstruction prior. Assuming a dual-source configuration, the theoretical Rayleigh limit is calculated to be approximately 4 using the current array coordinates and signal frequency. Set up a control experiment, set the signal-to-noise ratio to 0dB, and the number of snapshots. Still 50. First, verify the long-distance separation of two sources, setting the azimuth and elevation angles of the sound sources to (100) and (100) respectively. 105 ) and (123 105 Fixed pitch angle 105° The beam pattern cross-section results are as follows Figure 7 As shown, then verify two sources at close range, setting the azimuth and elevation angles of the sound sources to (125° and 125° respectively). 105 ) and (123 105 ), beam pattern cross-section as Figure 8 As shown.

[0059] Figure 8 In the 0dB test, all three methods could find two corresponding sound sources, and the main lobe and overall side lobes of the fused power spectrum were lower than those of the original CBF power spectrum. Figure 9 In this case, the distance between the two sound sources is 2. The peak values ​​of both the CBF power spectrum and the fused power spectrum have merged into a single peak, making them indistinguishable. Based on this, SBL can further distinguish the two sources, improving resolution. However, the computational complexity of element-domain SBL is limited by the number of angle grid points, making it difficult to adapt to multi-element, large-space scanning scenarios. Using the simulation parameters of this invention (azimuth angle 30°),... ~150 Pitch angle 30 ~150 Step size 1 For example, the original full mesh needs to cover 14641 angle points, and the time complexity of inverting the covariance matrix reaches [value missing]. This invention employs complex multiplication; however, it constructs a local grid (with only about 200 points) guided by the peak value of the fused power spectrum, reducing the computational load to one ten-millionth of the original full grid. This significantly improves the real-time performance of the algorithm while maintaining super-resolution performance. Furthermore, it utilizes CBF for power spectrum fusion as a priori, eliminating complex and sensitive parameter settings and relying entirely on statistical characteristics.

[0060] Based on the same inventive concept, the present invention also provides a positioning system for a passive ultrasonic sound source, comprising: The data acquisition module is used to acquire N sets of snapshot sound signals from the target ultrasonic source using a planar microphone array.

[0061] The reconstruction module is used to perform multiple non-repeating random combinations of the N sets of snapshot sound signals; obtain the covariance matrix of each combination, and extract the power spectrum of each covariance matrix in a preset scanning direction; for each scanning direction, count the number of times the power value in all power spectra in that direction is greater than a set power threshold. When the count is greater than the set threshold, the scanning direction is determined as a valid direction, and the count is used as the stability weight of the corresponding scanning direction; calculate the average power of each valid direction; and generate the reconstructed power spectrum based on the stability weight and the average power.

[0062] The localization module is used to extract the first M peaks of the reconstructed power spectrum, construct multiple local grids with each peak as the center and Rayleigh limit as the range; define the prior variance vector of the sparse Bayesian learning algorithm based on each local grid; obtain the noise variance estimate of the covariance matrix; and perform iterative optimization of the sparse Bayesian learning algorithm based on the multiple prior variance vectors, the noise variance estimate, and N sets of snapshot sound signals. When the iteration condition is met, the direction of arrival estimate of the target ultrasonic source is output according to the optimization result, and the localization of the target ultrasonic source is achieved by using the direction of arrival estimate.

[0063] This invention also provides a computer device. At the hardware level, the computer device includes a processor, an internal bus, a network interface, memory, and non-volatile memory, and may also include other hardware required for various operations. The processor reads the corresponding computer program from the non-volatile memory into the memory and then runs it to implement the passive ultrasonic source localization method described above.

[0064] The present invention also provides a computer-readable storage medium storing a computer program that can be used to execute the above-described method for locating a passive ultrasonic source.

[0065] Specific limitations regarding the computational system for the localization method of passive ultrasonic sources can be found in the limitations of the localization method for passive ultrasonic sources described above, and will not be repeated here. Each module in the aforementioned localization system for passive ultrasonic sources can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in hardware or independently of the processor in a computer device, or stored in software in the memory of a computer device, so that the processor can call and execute the corresponding operations of each module.

[0066] The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as the combination of these technical features does not contradict each other, it should be considered within the scope of this specification. Furthermore, the above embodiments only illustrate several implementation methods of this application, and their descriptions are relatively specific and detailed, but they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make several modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. A method for locating a passive ultrasonic sound source, characterized in that, include: N sets of snapshot sound signals of the target ultrasonic source are acquired using a planar microphone array; The N sets of snapshot sound signals are randomly combined in multiple groups without repetition; the covariance matrix of each combination is obtained, and the power spectrum of each covariance matrix is ​​extracted in a preset scanning direction; for each scanning direction, the number of times the power value in all power spectra in that direction is greater than a set power threshold is counted. When the count is greater than the set threshold, the scanning direction is determined to be a valid direction, and the count is used as the stability weight of the corresponding scanning direction; the average power of each valid direction is calculated respectively. A reconstructed power spectrum is generated based on stability weights and average power. The first M peaks of the reconstructed power spectrum are extracted, and multiple local grids are constructed with each peak as the center and Rayleigh limit as the range. Based on each local grid, the prior variance vector of the sparse Bayesian learning algorithm is defined. The noise variance estimate of the covariance matrix is ​​obtained. Based on the multiple prior variance vectors, the noise variance estimate, and N sets of snapshot sound signals, the sparse Bayesian learning algorithm is iteratively optimized. When the iteration condition is met, the direction of arrival estimate of the target ultrasonic source is output according to the optimization result. The direction of arrival estimate is used to locate the target ultrasonic source.

2. The method for locating a passive ultrasonic sound source according to claim 1, characterized in that, Before extracting the power spectrum of each covariance matrix in the preset scanning direction, the method further includes: using a diagonal load reduction method based on Frobenius norm optimization to separate the diagonal load reduction matrix from each covariance matrix to obtain multiple corrected covariance matrices, and then extracting the power spectrum of each corrected covariance matrix.

3. The method for locating a passive ultrasonic sound source according to claim 1, characterized in that, Before counting the number of times the power value in all power spectra in each scanning direction is greater than a set power threshold, the method further includes shielding the interference in the negative power direction in each power spectrum, where the negative power direction is determined to be a non-signal direction.

4. The method for locating a passive ultrasonic sound source according to claim 1, characterized in that, The total number of points in the local grid does not exceed 200; the Rayleigh limit is calculated from the coordinates of the array elements of the planar microphone array and the signal frequency of the target ultrasonic source; the array elements of the planar microphone array are irregularly distributed in space.

5. The method for locating a passive ultrasonic source according to claim 1, characterized in that, Extracting the top M peaks of the reconstructed power spectrum specifically includes: sorting the power values ​​from high to low in the reconstructed power spectrum, and selecting the top M peaks with the largest power values, where M is a preset estimated number of sound sources or a value automatically determined based on the significance of the peaks.

6. The method for locating a passive ultrasonic sound source according to claim 1, characterized in that, The iterative optimization of the sparse Bayesian learning algorithm specifically includes: cyclically calculating the posterior mean and posterior covariance based on the prior variance vector and the noise variance estimate; updating the sparse weights in the prior variance vector and the noise variance in the noise variance estimate based on the posterior mean and posterior covariance; and terminating the iteration when the change in noise variance before and after the update is less than a preset convergence threshold.

7. The method for locating a passive ultrasonic sound source according to claim 1, characterized in that, The total number of snapshots in the N groups of snapshot sound signals does not exceed 200.

8. A positioning system for a passive ultrasonic sound source, characterized in that, include: The data acquisition module is used to acquire N sets of snapshot sound signals from the target ultrasonic source using a planar microphone array; The reconstruction module is used to perform multiple non-repeating random combinations of the N sets of snapshot sound signals; obtain the covariance matrix of each combination, and extract the power spectrum of each covariance matrix in a preset scanning direction; for each scanning direction, count the number of times the power value in all power spectra in that direction is greater than a set power threshold. When the count is greater than the set threshold, the scanning direction is determined as a valid direction, and the count is used as the stability weight of the corresponding scanning direction; calculate the average power of each valid direction. A reconstructed power spectrum is generated based on stability weights and average power. The localization module is used to extract the first M peaks of the reconstructed power spectrum, construct multiple local grids with each peak as the center and Rayleigh limit as the range; define the prior variance vector of the sparse Bayesian learning algorithm based on each local grid; obtain the noise variance estimate of the covariance matrix; and perform iterative optimization of the sparse Bayesian learning algorithm based on the multiple prior variance vectors, the noise variance estimate, and N sets of snapshot sound signals. When the iteration condition is met, the direction of arrival estimate of the target ultrasonic source is output according to the optimization result, and the localization of the target ultrasonic source is achieved by using the direction of arrival estimate.

9. A computer device, comprising a memory, a processor, and a computer program stored in the memory, characterized in that, The processor executes the computer program to implement the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is loaded by the processor, it is able to perform the steps of the method according to any one of claims 1 to 7.

Images