A target orientation estimation method based on sparse bayesian learning

Abstract

The application discloses a target direction estimation method based on sparse Bayesian learning, and belongs to the technical field of sonar passive direction finding. The application solves the problem of low estimation accuracy of the existing direction estimation method under the background of underwater platform self-noise. The technical scheme adopted by the application is as follows: step one, establishing a far-field signal array receiving model under the condition of existing platform self-noise; step two, establishing a sparse Bayesian learning framework under the platform self-noise according to the receiving model in step one, and obtaining a sound source signal power estimation model according to the established sparse Bayesian learning framework; step three, projecting a receiving data covariance matrix to a noise subspace to obtain a platform self-noise covariance matrix estimation model; and step four, obtaining a sound source target direction estimation result according to the sound source signal power estimation model and the platform self-noise covariance matrix estimation model. The method can be applied to target direction estimation.

Description

Technical Field

[0001] This invention belongs to the field of passive sonar direction finding technology, specifically relating to a target orientation estimation method based on sparse Bayesian learning. Background Technology

[0002] During navigation, underwater mobile platforms generate significant self-noise due to internal mechanical vibrations and propeller rotation. For sonar receiving arrays fixed on these platforms, the main component of the received noise is platform self-noise. When using this array to estimate the azimuth of distant targets, the spatial distribution of platform self-noise exhibits a certain directionality under the far-field signal array receiving model, so this self-noise can be approximated as anisotropic noise. Most existing methods assume an isotropic Gaussian white noise model during model construction, leading to a mismatch in the noise model and resulting in decreased accuracy in target azimuth estimation, even the appearance of spurious peaks, severely degrading the performance of target azimuth estimation.

[0003] Existing literature has studied the orientation estimation method under anisotropic noise conditions. Reference 1 (Yang L, Yang Y, Wang Y. Sparse spatial spectral estimation in directional noise environment[J].The Journal of the Acoustical Society of America,2016,140(3):EL263–EL268.) proposed a sparse spectrum fitting method, abbreviated as DN-SpSF, by simulating noise using a finite term Fourier series. However, the regularization parameter of the proposed method is difficult to select, and the inaccuracy of the regularization parameter selection will directly affect the orientation estimation performance. Moreover, the computational cost of the method increases significantly when the array aperture increases. Reference 2 (Liang G, Shi Z, Qiu L, Sun S, Lan T. Sparse Bayesian Learning Based Direction-of-Arrival Estimation under Spatially Colored Noise Using Acoustic Hydrophone Arrays[J]. Journal of Marine Science and Engineering, 2021, 9(2): 127.) uses a long ellipsoidal wavefunction to simulate anisotropic noise. However, the algorithm performance degrades significantly when the noise model differs too much from the actual received noise.

[0004] In summary, under the background of underwater platform self-noise, existing orientation estimation methods still suffer from low estimation accuracy. Therefore, it is essential to propose a new target orientation estimation method to solve the above problems. Summary of the Invention

[0005] The purpose of this invention is to address the problem of low estimation accuracy in existing orientation estimation methods under the background of self-noise on underwater platforms, and to propose an orientation estimation method based on sparse Bayesian learning.

[0006] The technical solution adopted by the present invention to solve the above-mentioned technical problems is as follows:

[0007] A location estimation method based on sparse Bayesian learning, the method specifically includes the following steps:

[0008] Step 1: Establish a far-field signal array reception model with platform self-noise;

[0009] Step 2: Based on the receiving model in Step 1, establish a sparse Bayesian learning framework under platform self-noise, and obtain a sound source signal power estimation model based on the established sparse Bayesian learning framework.

[0010] Step 3: Project the received data covariance matrix onto the noise subspace to obtain the platform's self-noise covariance matrix estimation model;

[0011] Step 4: Obtain the target location estimation result of the sound source based on the sound source signal power estimation model and the platform self-noise covariance matrix estimation model.

[0012] Furthermore, the specific process of step one is as follows:

[0013] If K far-field sound source signals are incident on a uniform linear array composed of M array elements, then the data received by the array at time t can be represented as:

[0014] x(t)=A(θ)s(t)+n(t) (1)

[0015] Where θ = {θ1, θ2, ..., θ} K},θ1,θ2,…,θ K Let K be the incident angles of far-field sound source signals, and let the sound source signal steering vector matrix be A(θ) = [a(θ1), a(θ2), ..., a(θ...]. K )], k=1,2,…,K, the steering vector a(θ) of the k-th sound source signal k ) is: a(θ k )=[1,exp(j(2π / λ)dcos(θ k )),…,exp(j(2π / λ)(M-1)dcos(θ k ))]T λ is the wavelength of the sound source signal, j is the imaginary unit, d is the spacing between adjacent array elements, the superscript T represents transpose, x(t) is the data received by the array at time t, s(t) is the sound source signal received by the array at time t, and n(t) is the platform self-noise received by the array at time t.

[0016] Extending equation (1) to:

[0017] X=A(θ)S+N (2)

[0018] Wherein, the received data X=[x(1),x(2),…,x(T′)], the received sound source signal S=[s(1),s(2),…,s(T′)], the received platform self-noise N=[n(1),n(2),…,n(T′)], and T′ is the total number of snapshots.

[0019] Furthermore, the specific process of step two is as follows:

[0020] Step 2.1. Perform spatial sparse representation on the sound source signal S in equation (2) to obtain the received data model in equation (3):

[0021]

[0022] in, It is an overcomplete dictionary formed by L steering vectors, where L is the number of discrete spatial angles. The zero-padding result corresponding to S, The zero-padding result for s(1) is s(1) = [s1(1), s2(1), ..., s K (1)],s k (1) is the element of s(1) in the k-th direction. s(2) is the zero-padding result corresponding to s(2), and s(T′) is the zero-padding result corresponding to s(T′);

[0023] Step 22: According to the received data model in equation (3), the likelihood function of the received data X is:

[0024]

[0025] in, Let H be the likelihood function of the received data X, where the superscript H represents the conjugate transpose, Σ n The platform's self-noise covariance matrix is ​​represented by the superscript -1, which represents the inverse of the matrix, and |·| represents the determinant operation.

[0026] Based on the sparse Bayesian learning framework, it is assumed that the received sound source signal S follows a zero-mean complex Gaussian distribution in the spatial domain:

[0027]

[0028] in, yes The distribution form in the spatial domain, hyperparameters γ1, γ2, ..., γ L They represent the sound source signals respectively. power, for The element in the l-th direction, l = 1, 2, ..., L, Representing a complex Gaussian distribution, γ = [γ1, γ2, ..., γ] L ]≥0, Γ=diag(γ), diag(·) represents a diagonal matrix;

[0029] Step 2. According to equations (4) and (5), the probability density function p(X; γ, Σ) of the received data X is calculated. n ) is represented as:

[0030]

[0031] Where, Σ x The covariance matrix representing the received data X,

[0032] According to Bayesian theory, using equations (4), (5), and (6), we obtain Posterior probability of X

[0033]

[0034] Where μ represents The expectation, μ=[μ1,μ2,…,μ T′ ],μ1,μ2,…,μ T′ Represent Expectations, Σ s represent The variance;

[0035]

[0036]

[0037] Then based on μ and Σ s A model for estimating the power γ of the sound source signal is obtained.

[0038] Furthermore, the statement based on μ and Σ s The EM algorithm was used to obtain the sound source signal power γ estimation model.

[0039] Furthermore, the estimation model for the sound source signal power γ is as follows:

[0040]

[0041] Where, μ l· For the l-th row of μ, (Σ s ) l,l For Σ s The l-th diagonal element.

[0042] Furthermore, the specific process of step three is as follows:

[0043] By performing a spectral peak search on γ, the locations of K potential sound source signals are obtained. By maximizing the logarithmic form of the likelihood function in equation (4), the expression for the potential sound source signal is obtained:

[0044]

[0045] Where S′ is the potential sound source signal, It is composed of K potential sound source signals in The array manifold matrix composed of the corresponding guiding vectors;

[0046] Substituting equation (11) into equation (2), we obtain equation (12):

[0047]

[0048] in, The projection matrix representing the noise space. I is the identity matrix;

[0049] Then the platform's self-noise covariance matrix Σ n for:

[0050]

[0051] in,

[0052] Furthermore, the specific process of step four is as follows:

[0053] Step 4.1 Initialize the sound source signal power γ to γ 1 Initialize the platform's self-noise covariance matrix Σ n For Σ n 1 Set the maximum number of iterations to i. max Set the iteration termination parameter to ε;

[0054] Step 42: Let the iteration number i = 1;

[0055] Step 43, γ i and Σ n i Substituting into equations (8) and (9), we obtain μi and Σ s i ;

[0056] Step 44: Put μ i and Σ s i Substituting into equation (10), we obtain γ i+1 ;

[0057] Steps four and five: Based on γ i+1 After obtaining the potential sound source target location, substitute the potential sound source target location into equation (13) to obtain Σ. n i+1 ;

[0058] Step 46: Determine whether the iteration stopping condition is met;

[0059] If the iteration stopping condition is met, the iteration stops, and the sound source target location estimation result is obtained based on the sound source signal power output in the last iteration.

[0060] If the iteration stopping condition is not met, let i = i + 1 and return to step 43.

[0061] Furthermore, the iteration stopping condition is: at least one of condition 1) and condition 2) is satisfied;

[0062] Condition 1) The maximum number of iterations i has been reached. max ;

[0063] Condition 2), |γ i+1 -γ i ||2 / |γ i+1 ||2<ε, where ||·||2 is the 2-norm.

[0064] Furthermore, the step of obtaining the sound source target location estimation result based on the sound source signal power output in the last iteration is specifically as follows:

[0065] The azimuth angles corresponding to the K maximum powers in the sound source signal power are used as the sound source target azimuth estimation results.

[0066] The beneficial effects of this invention are:

[0067] This invention first establishes a far-field signal array reception model in the presence of platform self-noise. Then, based on a sparse Bayesian learning framework under platform self-noise, it obtains a sound source signal power estimation model. Finally, it obtains a self-noise covariance matrix estimation model by projecting the received data covariance matrix onto the noise subspace. Based on the sound source signal power estimation model and the self-noise covariance matrix estimation model, the sound source target azimuth estimation result is obtained. This invention's method achieves higher azimuth estimation accuracy in the presence of underwater mobile platform noise and suppresses spurious peaks caused by platform self-noise, thus improving azimuth estimation performance. Attached Figure Description

[0068] Figure 1 This is a flowchart of the method of the present invention;

[0069] Figure 2a Spatial spectra of the proposed method and the comparative method under isotropic noise and a signal-to-noise ratio of 5dB;

[0070] Figure 2b Spatial spectra of the proposed method and the comparative method under the condition of platform self-noise and signal-to-noise ratio of 5dB;

[0071] Figure 2c Spatial spectra of the proposed method and the comparative method under isotropic noise and a signal-to-noise ratio of -5dB;

[0072] Figure 2d Spatial spectra of the proposed method and the comparative method under the condition of platform self-noise and signal-to-noise ratio of -5dB;

[0073] Figure 3 The graph shows the root mean square error of orientation estimation as a function of signal-to-noise ratio for the proposed method and the comparative method.

[0074] Figure 4 The graph shows the root mean square error of orientation estimation for the proposed method and the comparative method as a function of the number of snapshots. Detailed Implementation

[0075] The present application will now be described in further detail with reference to specific embodiments and accompanying drawings. Obviously, the described embodiments are merely a part of the embodiments of the present invention, and not all of them. Other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are all within the scope of protection of the present invention.

[0076] Specific Implementation Method 1: Combination Figure 1 This embodiment describes a location estimation method based on sparse Bayesian learning, which specifically includes the following steps:

[0077] Step 1: Establish a far-field signal array reception model with platform self-noise;

[0078] Step 2: Based on the receiving model in Step 1, establish a sparse Bayesian learning framework under platform self-noise, and obtain a sound source signal power estimation model based on the established sparse Bayesian learning framework.

[0079] Step 3: Project the received data covariance matrix onto the noise subspace to obtain the platform's self-noise covariance matrix estimation model;

[0080] Step 4: Obtain the target location estimation result of the sound source based on the sound source signal power estimation model and the platform self-noise covariance matrix estimation model.

[0081] Underwater mobile platforms generate strong self-noise during navigation. Therefore, when using the sonar receiving array installed on them to estimate the azimuth of distant targets, the estimation results are often affected by the platform noise, resulting in decreased estimation accuracy and even spurious peaks, severely degrading the azimuth estimation performance. The method of this invention solves this practical problem.

[0082] Specific Implementation Method Two: This implementation method differs from Specific Implementation Method One in that the specific process of step one is as follows:

[0083] If K far-field sound source signals are incident on a uniform linear array composed of M array elements, then the data received by the array at time t can be represented as:

[0084] x(t)=A(θ)s(t)+n(t) (1)

[0085] Where θ = {θ1, θ2, ..., θ} K},θ1,θ2,…,θ K Let K be the incident angles of far-field sound source signals, and let the sound source signal steering vector matrix be A(θ) = [a(θ1), a(θ2), ..., a(θ...]. K )], k=1,2,…,K, the steering vector a(θ) of the k-th sound source signal k ) is: a(θ k )=[1,exp(j(2π / λ)dcos(θ k )),…,exp(j(2π / λ)(M-1)dcos(θ k ))] T λ is the wavelength of the sound source signal (equal for each sound source signal), j is the imaginary unit, d is the spacing between adjacent array elements, the superscript T represents transpose, x(t) is the data received by the array at time t, s(t) is the sound source signal received by the array at time t, and n(t) is the platform self-noise received by the array at time t.

[0086] For multi-shot data, equation (1) can be further extended to:

[0087] X=A(θ)S+N (2)

[0088] Wherein, the received data X=[x(1),x(2),…,x(T′)], the received sound source signal S=[s(1),s(2),…,s(T′)], the received platform self-noise N=[n(1),n(2),…,n(T′)], and T′ is the total number of snapshots.

[0089] The other steps and parameters are the same as in Specific Implementation Method 1.

[0090] Specific Implementation Method Three: This implementation method differs from Specific Implementation Method One or Two in that the specific process of step two is as follows:

[0091] Step 2.1. Perform spatial sparse representation on the sound source signal S in equation (2) to obtain the received data model in equation (3):

[0092]

[0093] in, It is an overcomplete dictionary formed by L steering vectors, where L is the number of discrete spatial angles, and its magnitude satisfies L >> M > K. The zero-padding result corresponding to S, The zero-padding result for s(1) is s(1) = [s1(1), s2(1), ..., s K (1)],s k (1) is the element of s(1) in the k-th direction. Zero-padding is performed on the parts from 1 to L that do not correspond to the sound source signal. The zero-padding result corresponding to s(2), This is the zero-padding result corresponding to s(T′);

[0094] Step 22: Assuming that noise and signal are uncorrelated and that the received data from each snapshot are independent, according to the received data model in equation (3), the likelihood function of the received data X is:

[0095]

[0096] in, Let H be the likelihood function of the received data X, where the superscript H represents the conjugate transpose, Σ n The platform's self-noise covariance matrix is ​​represented by the superscript -1, which represents the inverse of the matrix, and |·| represents the determinant operation.

[0097] Based on the sparse Bayesian learning framework, it is assumed that the received sound source signal S follows a zero-mean complex Gaussian distribution in the spatial domain:

[0098]

[0099] in, yes The distribution form in the spatial domain, hyperparameters γ1, γ2, ..., γ L These represent the sound source signal s1(t), power, for The element in the l-th direction, l = 1, 2, ..., L, Representing a complex Gaussian distribution, γ = [γ1, γ2, ..., γ] L ]≥0, Γ=diag(γ), diag(·) represents a diagonal matrix;

[0100] Step 2. According to equations (4) and (5), the probability density function p(X; γ, Σ) of the received data X is calculated. n ) is represented as:

[0101]

[0102] Where, Σ x The covariance matrix representing the received data X,

[0103] According to Bayesian theory, using equations (4), (5), and (6), we obtain Posterior probability of X

[0104]

[0105] Where μ represents The expectation, μ=[μ1,μ2,…,μ T′ ],μ1,μ2,…,μ T′ Represent Expectations, Σ s represent The variance;

[0106]

[0107]

[0108] Then based on μ and Σ s A model for estimating the power γ of the sound source signal is obtained.

[0109] Other steps and parameters are the same as in specific implementation method one or two.

[0110] Specific Implementation Method Four: This implementation method differs from Specific Implementation Methods One to Three in that the step of basing the calculations on μ and Σ...s The sound source signal power γ estimation model was obtained using the EM algorithm (Expectation Maximization).

[0111] The other steps and parameters are the same as those in one of the specific implementation methods one to three.

[0112] Specific Implementation Method Five: This implementation method differs from Specific Implementation Methods One to Four in that the sound source signal power γ estimation model is as follows:

[0113]

[0114] Where, μ l· For the l-th row of μ, (Σ s ) l,l For Σ s The l-th diagonal element.

[0115] Its steps and parameters are the same as those in one of the specific implementation methods one to four.

[0116] Specific Implementation Method Six: This implementation method differs from Specific Implementation Methods One to Five in that the specific process of step three is as follows:

[0117] A spectral peak search is performed on γ to obtain the locations of K potential sound source signals. Given the self-noise covariance matrix Σ n Given the potential sound source target location, by maximizing the logarithmic form of the likelihood function in equation (4), we obtain the expression for the potential sound source signal:

[0118]

[0119] Where S′ is the potential sound source signal, It is composed of K potential sound source signals in The array manifold matrix composed of the corresponding guiding vectors;

[0120] Substituting equation (11) into equation (2), we obtain equation (12):

[0121]

[0122] in, The projection matrix representing the noise space. I is the identity matrix;

[0123] Then the platform's self-noise covariance matrix Σ n for:

[0124]

[0125] in,

[0126] The other steps and parameters are the same as those in one of the specific implementation methods one to five.

[0127] Specific Implementation Method Seven: This implementation method differs from Specific Implementation Methods One to Six in that the specific process of step four is as follows:

[0128] Step 4.1 Initialize the sound source signal power γ to γ 1 Initialize the platform's self-noise covariance matrix Σ n For Σ n 1 Set the maximum number of iterations to i. max Set the iteration termination parameter to ε;

[0129] Step 42: Let the iteration number i = 1;

[0130] Step 43, γ i and Σ n i Substituting into equations (8) and (9), we obtain μ i and Σ s i ;

[0131] Step 44: Put μ i and Σ s i Substituting into equation (10), we obtain γ i+1 ;

[0132] Steps four and five: Based on γ i+1 After obtaining the potential sound source target location, substitute the potential sound source target location into equation (13) to obtain Σ. n i+1 ;

[0133] Step 46: Determine whether the iteration stopping condition is met;

[0134] If the iteration stopping condition is met, the iteration stops, and the sound source target location estimation result is obtained based on the sound source signal power output in the last iteration.

[0135] If the iteration stopping condition is not met, let i = i + 1 and return to step 43.

[0136] The other steps and parameters are the same as those in one of the specific implementation methods one to six.

[0137] The method of this embodiment is shown in Table 1:

[0138] Table 1

[0139]

[0140] Specific Implementation Method Eight: This implementation method differs from any of Specific Implementation Methods One to Seven in that the iteration stopping condition is: at least one of condition 1) and condition 2) is satisfied;

[0141] Condition 1) The maximum number of iterations i has been reached. max ;

[0142] Condition 2), |γ i+1 -γ i ||2 / ||γ i+1 ||2<ε, where ||·||2 is the 2-norm.

[0143] The other steps and parameters are the same as those in any of the specific implementation methods one to seven.

[0144] Specific Implementation Method Nine: This implementation method differs from Specific Implementation Methods One through Eight in that the sound source target location estimation result is obtained based on the sound source signal power output in the last iteration; specifically:

[0145] The azimuth angles corresponding to the K maximum powers in the sound source signal power are used as the sound source target azimuth estimation results.

[0146] The other steps and parameters are the same as those in one of the specific implementation methods one to eight.

[0147] Implementation section:

[0148] The effectiveness of the proposed method is verified through computer simulation, and the performance of the algorithm is evaluated and analyzed. Since the proposed method relies on a sparse Bayesian learning framework and is applicable to platform self-noise backgrounds, it can be abbreviated as UNSBL (Underwater maneuvering platformNoise Sparse Bayesian Learning). It is then compared with conventional beamforming (CBF), the MUSIC method in subspace methods, the SBL method in sparse methods, the DN-SpSF method, and the SCN-MSBL method.

[0149] Assume a 10-element uniform linear array with an array spacing of half a wavelength is used to receive a sound source signal. A far-field narrowband sound source signal is incident from a 120° direction. The number of snapshots is 100, and the spatial discrete angle interval is set to 1°. Figure 2a , Figure 2b , Figure 2c as well as Figure 2d The image shows the spatial spectra of the proposed method and the comparative method under different noise sources and signal-to-noise ratios. From... Figure 2a , Figure 2b , Figure 2c as well as Figure 2dAs can be seen, the sparse methods produce sharper spectral peaks compared to the MUSIC and CBF methods. A horizontal comparison of Figures 2(a), 2(c), 2(b), and 2(d) shows that the noise spectra of the tested methods are relatively flat under isotropic noise conditions. UNSBL and SBL have similar spatial spectra. Under platform self-noise conditions, although the CBF and SBL methods can accurately estimate the target's azimuth, they are also affected by the platform self-noise. Especially when the signal-to-noise ratio is -5dB, the noise spectrum of the CBF method increases significantly near 0° and 180°, where the self-noise spectral level is relatively high, while the SBL method shows a significant spurious peak near the self-noise peak position, i.e., around 160°. In contrast, the noise spectra of the MUSIC, UNSBL, SCN-MSBL, and DN-SpSF methods are relatively flat under different signal-to-noise ratios.

[0150] The statistical performance of the tested method is analyzed below through Monte Carlo experiments, mainly including the variation of the root mean square error (RMSE) of the orientation estimation results with the signal-to-noise ratio and the number of snapshots.

[0151] In the root mean square error statistics, two uncorrelated far-field sound source targets with the same power were incident from 120.3° and 140.3° respectively. First, the signal-to-noise ratio was set to vary from -5dB to 15dB, the spatial discrete angle interval was 2°, and the number of snapshots was 100. 200 independent Monte Carlo experiments were conducted under each condition. Figure 3 This shows the variation of the root mean square error of orientation estimation with signal-to-noise ratio for the proposed method and the comparative method. From... Figure 3 As can be seen, under conditions of low signal-to-noise ratio, the UNSBL method proposed in this invention has the highest azimuth estimation accuracy, which is significantly better than other comparative methods.

[0152] Next, the signal-to-noise ratio was set to 0dB, the number of snapshots was varied from 20 to 200, and the other simulation conditions remained unchanged. The root mean square error of the tested method was observed and analyzed as a function of the number of snapshots. 200 independent Monte Carlo experiments were conducted under each condition. Figure 4 This demonstrates how the root mean square error of orientation estimation varies with the number of snapshots for the proposed method and the comparative method. From Figure 4 As can be seen, the DN-SpSF method exhibits a large root mean square error when the number of snapshots is only 20. The CBF method has a slightly higher overall root mean square error than the other four methods. However, the proposed UNSBL method has significantly higher azimuth estimation accuracy than the other comparative methods when the number of snapshots is low.

[0153] The above examples of the present invention are merely illustrative of the computational model and process of the present invention, and are not intended to limit the implementation of the present invention. Those skilled in the art will recognize that other variations or modifications can be made based on the above description. It is impossible to exhaustively list all possible implementations here. Any obvious variations or modifications derived from the technical solutions of the present invention are still within the scope of protection of the present invention.

Claims

1. A target orientation estimation method based on sparse Bayesian learning, characterized in that, The method specifically includes the following steps: Step 1: Establish a far-field signal array reception model with platform self-noise; The specific process of step one is as follows: If K far-field sound source signals are incident on a uniform linear array composed of M array elements, then the data received by the array at time t can be represented as: (1) in, , Let K be the incident angles of far-field sound source signals, and let the sound source signal steering vector matrix be... , The steering vector of the kth sound source signal for: , The wavelength of the sound source signal. The imaginary unit, This represents the spacing between adjacent array elements, with the superscript T indicating transpose. The data received by the array at time t. Let be the sound source signal received by the array at time t. Let be the platform self-noise received by the array at time t; Extending equation (1) to: (2) Among them, the received data Received sound source signal Received platform self-noise , This represents the total number of snapshots. Step 2: Based on the receiving model in Step 1, establish a sparse Bayesian learning framework under platform self-noise, and obtain a sound source signal power estimation model based on the established sparse Bayesian learning framework. The specific process of step two is as follows: Step 2.1, the sound source signal in equation (2) Spatial sparse representation yields the received data model in equation (3): (3) in, It is an overcomplete dictionary formed by L steering vectors, where L is the number of discrete spatial angles. for The corresponding zero-padding result, , for The corresponding zero-padding result, , for In the Elements in each direction, for The corresponding zero-padding result, for The corresponding zero-padding result; Step 2: Receive data according to the data receiving model in equation (3). The likelihood function is: (4) in, To receive data The likelihood function, where the superscript H represents the conjugate transpose. This represents the platform's self-noise covariance matrix, with the superscript -1 representing the matrix inverse. Represents determinant operations; Based on the sparse Bayesian learning framework, assuming the receiver is a sound source signal It follows a complex Gaussian distribution with zero mean in the spatial domain: (5) in, yes Distribution pattern in the spatial domain, hyperparameters They represent the sound source signals respectively. , , ..., power, for In the Elements in each direction, , Represents a complex Gaussian distribution. , , Represents a diagonal matrix; Step 2 and 3: According to equations (4) and (5), receive the data. probability density function Represented as: (6) in, Represents receiving data The covariance matrix, ; According to Bayesian theory, using equations (4), (5), and (6), we obtain about posterior probability : (7) in, represent Expectations , Represent Expectations represent The variance; (8) (9) Then according to and Obtain the power of the sound source signal Estimation model; Step 3: Project the received data covariance matrix onto the noise subspace to obtain the platform's self-noise covariance matrix estimation model; Step 4: Obtain the target location estimation result of the sound source based on the sound source signal power estimation model and the platform self-noise covariance matrix estimation model.

2. The target orientation estimation method based on sparse Bayesian learning according to claim 1, characterized in that, According to and Obtain the power of the sound source signal The estimation model uses the EM algorithm.

3. The target orientation estimation method based on sparse Bayesian learning according to claim 2, characterized in that, The power of the sound source signal The estimation model is: (10) in, for The OK, for The One diagonal element.

4. The target orientation estimation method based on sparse Bayesian learning according to claim 3, characterized in that, The specific process of step three is as follows: right Perform a peak search to obtain By maximizing the logarithmic form of the likelihood function in equation (4) for the location of each potential sound source signal, we obtain the expression for the potential sound source signal: (11) in, For potential sound source signals, It is by A potential sound source signal in The array manifold matrix composed of the corresponding guiding vectors; Substituting equation (11) into equation (2), we obtain equation (12): (12) in, The projection matrix representing the noise space. , It is the identity matrix; Then the platform's self-noise covariance matrix for: (13) in, .

5. The target orientation estimation method based on sparse Bayesian learning according to claim 4, characterized in that, The specific process of step four is as follows: Step 4.1 Initialize the sound source signal power for Initialize the platform's self-noise covariance matrix for Set the maximum number of iterations to Set the iteration termination parameter to ; Step 42: Let the number of iterations be... ; Step 43, and Substituting into equations (8) and (9), we get and ; Step 44, and Substituting into equation (10), we get ; Steps four and five, according to After obtaining the potential sound source target location, substitute the potential sound source target location into equation (13) to obtain ; Step 46: Determine whether the iteration stopping condition is met; If the iteration stopping condition is met, the iteration stops, and the sound source target location estimation result is obtained based on the sound source signal power output in the last iteration. If the iteration stopping condition is not met, then let Return to step four three.

6. The target orientation estimation method based on sparse Bayesian learning according to claim 5, characterized in that, The iteration stopping condition is: at least one of condition 1) and condition 2) must be satisfied; Condition 1) The maximum number of iterations set has been reached. ; Condition 2) ,in, It is a 2-norm.

7. The target orientation estimation method based on sparse Bayesian learning according to claim 6, characterized in that, The sound source target location estimation result is obtained based on the sound source signal power output from the last iteration; specifically: The azimuth angles corresponding to the K maximum powers in the sound source signal power are used as the sound source target azimuth estimation results.

Images