A sparse bayesian target direction estimation method based on far-field dictionary reconstruction in near-field strong interference environment

Abstract

This invention discloses a sparse Bayesian target azimuth estimation method based on far-field dictionary reconstruction under strong near-field interference. The method includes subspace decomposition of the observation data to construct a signal subspace projection operator; projecting and normalizing the far-field dictionary to obtain a projected reconstruction dictionary, which replaces the far-field dictionary in sparse Bayesian learning, and jointly performing sparse Bayesian iterative estimation with the near-field dictionary; finally, outputting the far-field target azimuth estimation result. This method effectively partitions the subspace and constructs a projection matrix accordingly to suppress the impact of strong near-field interference on far-field azimuth estimation, significantly improving the far-field target azimuth estimation capability of the sparse Bayesian method under strong near-field interference conditions.

Description

Technical Field

[0001] This invention relates to the field of underwater acoustic array signal processing and target localization technology, and in particular to a sparse Bayesian target orientation estimation method based on far-field dictionary reconstruction under near-field strong interference environment. Background Technology

[0002] Sparse representation methods outperform traditional methods in terms of resolution, estimation accuracy, and weak target detection capability. However, in practical applications, the operating environment of the receiving array is usually quite complex, inevitably subject to interference from the self-noise of towed ship engines or the noise of numerous passing vessels. These interference signals are often close to the array, and their energy may be much greater than that of far-field targets, leading to a significant degrade in performance or even complete failure of conventional far-field DOA estimation methods. Therefore, researching a sparse representation-based target location estimation method that can effectively suppress interference under strong near-field interference has significant theoretical and engineering value. Summary of the Invention

[0003] The purpose of this invention is to address the shortcomings of existing technologies by proposing a near-field and far-field dictionary reconstruction concept. By using subspace projection technology for the far-field dictionary, the projection energy of strong near-field interference power on the far-field dictionary can be effectively reduced, thereby reducing false peaks and improving the far-field target azimuth estimation performance.

[0004] The objective of this invention is achieved through the following technical solution: a sparse Bayesian target location estimation method based on far-field dictionary reconstruction under near-field strong interference environment, comprising the following steps: The output signal of the receiving array is acquired and its covariance matrix is ​​calculated. The covariance matrix is ​​decomposed into eigenvectors to obtain its eigenvectors. Discriminant features are extracted based on the eigenvectors. A support vector machine is introduced to classify the discriminant features into near-field or far-field. All eigenvectors identified as near-field are arranged into a near-field subspace matrix by columns, and their orthogonal projection matrix is ​​calculated simultaneously. Grid the near-field and far-field observation spaces and construct ultra-complete near-field and far-field dictionaries; The far-field dictionary is projected using an orthogonal projection matrix to obtain the projected far-field dictionary; the near-field dictionary and the projected far-field dictionary are then combined into a near-far-field hybrid dictionary by column. Based on sparse representation theory, the DOA estimation problem is transformed into a sparse signal reconstruction problem using a near-field and far-field hybrid dictionary. A corresponding sparse Bayesian learning probability model is established to obtain the estimated values ​​of each latent variable. The sparse power spectrum of the signal is obtained by inference, and the DOA estimate of the far-field target is extracted.

[0005] Furthermore, the eigenvalue decomposition of the covariance matrix includes: in, Let the covariance matrix be u1,…,u i …,u M (i=1,…,M) represent the eigenvectors of the covariance matrix, σ1,…, σ i, …, σ M (i=1,…,M) represents the eigenvalues ​​corresponding to each eigenvector.

[0006] Furthermore, the extraction of discriminative features based on the feature vector is as follows: Near-field and far-field dictionary matching degree, phase consistency measure, phase curvature measure and beamwidth measure are extracted from the feature vector. The four features are concatenated into a four-dimensional discriminative feature vector, which is used as the input feature of the support vector machine. The support vector machine classifies samples of different categories into near-field or far-field by finding the maximum margin hyperplane.

[0007] Furthermore, the pre-training process of the support vector machine includes: Calculate the representation of each feature using the simulated array element received data as training data: Near-field and far-field dictionary matching degree: , in, The angles selected for the far-field steering vector and the near-field steering vector. The distance chosen for the near-field guidance vector. Select the angle range for the far-field steering vector and the near-field steering vector. The range of distances selected for the near-field guidance vector. For far-field steering vector, This is the near-field steering vector. , , where is the element index and η is the element spacing. Indicates the signal wavelength. c The speed of sound in water, f For processing frequency; Phase consistency measure: in, For feature vectors The m-th value The phase information, where real() and imag() represent taking the real and imaginary parts of the complex number, This represents the parameters of the linear model after least-squares fitting of the phase information. ; Phase curvature measurement: in, The phase curvature of the eigenvector is specifically calculated as follows: ; Beamwidth measurement: Among them, B w For spatial spectrum The beamwidth is the difference in angles between the left and right sides of the maximum peak power in the spatial spectrum when the power drops to half of the peak power. The selection step size is 1°; The extracted features are fused. , Update using the objective function: ; in, It is the first i The labels of each training sample are set, with the near-field label being -1 and the far-field label being 1. Indicates the first i One training sample; For parameters to be optimized, It is the vector 2 norm.

[0008] Furthermore, the step of constructing a near-field subspace matrix from all the feature vectors determined to be near-field along their columns, and simultaneously calculating its orthogonal projection matrix, includes: The eigenvectors identified as near-field are arranged in columns to form a near-field subspace matrix U. nf where n1,…,n k Indexes for classification as near-field feature vectors: ; Calculate the orthogonal projection matrix P of the near-field subspace matrix. nf ,in I M express The identity matrix, Represents the near-field subspace matrix U nf Conjugate transpose: .

[0009] Furthermore, the construction of the overcomplete near-field dictionary and far-field dictionary includes: Gridded far-field angle observation space, constructing an overcomplete far-field dictionary matrix A F ; to observe the space from the far-field angle within the range The inner part is divided into uniformly spaced 1° intervals to obtain an angular grid point set. K f K is the total number of far-field grid points, and K f >>N f Based on the far-field angle grid point set Construct the far-field dictionary matrix A F : ; Grid the near-field angle and range observation space, and construct an overcomplete near-field dictionary matrix A. N ; Near-field angle observation space within range Inner The angles are evenly divided to obtain the angle grid point set. ,in This represents the total number of grid points in the near-field angle, and The near-field distance observation space is uniformly divided within the range [200m, 1000m] at 100m intervals to obtain a distance grid point set. ,in The total number of grid points in the near field distance, and Based on the near-field angle grid point set and near-field distance point set Construct the near-field dictionary matrix A N Its dimensions are : .

[0010] Furthermore, the transformation of the DOA estimation problem into a sparse signal reconstruction problem using a near-field and far-field hybrid dictionary is as follows: ; in, Represents a near-field and far-field mixed dictionary, where X represents... A 3D signal matrix, where n represents The additive white Gaussian noise matrix of dimension X, since X contains only K non-zero row vectors, X is a sparse matrix, and K f This represents the total number of far-field grid points. This represents the total number of grid points in the near-field angle. This represents the total number of grid points in the near-field distance. To establish a sparse Bayesian probability model, first, a complex Gaussian distribution assumption is made for each column vector of the array output signal Y. Then, the likelihood function of Y is expressed as: ; in, Represents the i-th column vector of matrices Y and X. The noise precision is expressed as the reciprocal of the noise variance. Perform a gamma prior distribution, i.e.: , These are the parameters of the gamma distribution; Next, a hierarchical sparse prior is constructed for the signal matrix X. In the first layer of priors, a set of latent variables is introduced. 3D auxiliary latent variable matrix , K×1 matrix W and K×1 signal mean vector Perform a linear transformation on each column of X. ,in Let the i-th column vector of matrix Z be represented. Let X represent a complex Gaussian vector with zero mean. Then the linear Gaussian prior model is: X follows the following complex Gaussian distribution: ; in, , This represents the operation of generating a diagonal matrix; Let be a scalar, representing the reciprocal of the incident target signal power in the direction of the i-th grid point; using a linear Gaussian prior model allows the autocorrelation component of signal X to be included in the matrix. In this matrix, the cross-correlation components are contained in matrix W; In the second layer of priors, the hyperparameters are respectively... Make a prior hypothesis, assume Each column and All follow a zero-mean complex Gaussian distribution. Each element is independently and identically distributed, and all follow a gamma distribution, i.e.: in, Let W represent the i-th column vector. express The i-th element, , For the parameters of the gamma distribution, For accuracy parameters, This represents the matrix inversion operation; In the third layer prior, regarding hyperparameters For each element, make a gamma prior hypothesis: ; in, Parameters for the gamma distribution; set initial parameter values. ,remember This is called the set of latent variables; The latent variables are calculated using the variational Bayesian inference method. Approximate posterior distribution ,get: That is, the first in X i column vector The posterior distribution is a complex Gaussian distribution with a mean of and variance They are respectively: , ;in, , Let Y and Z represent the i-th column vectors, respectively. This represents the expectation operation. This represents the matrix conjugate transpose operation; That is, the j-th column vector in W The posterior distribution is a complex Gaussian distribution with a mean of and variance They are respectively: in Represents the first j Line 1 i Column elements, express The j One element; That is, the first in Z i Column vector Z i The posterior distribution is a complex Gaussian distribution with a mean of and variance They are respectively: ,Right now The posterior distribution is a complex Gaussian distribution with a mean of and variance They are respectively: ,Right now The Middle i element The posterior distribution of is a gamma distribution, with parameters g and h as follows: ,in For vector 2 norm operations; ,Right now The posterior distribution of is a gamma distribution, and the parameters of this distribution are... c,d for: ,in Represents the trace of a matrix; ,Right now The Middle j element The posterior distribution of is a gamma distribution, and the parameters a and b of this distribution are: W j Let W be the j-th column vector. Represents absolute value; according to The update formula, after setting initial values ​​for the hidden variables, iterates and updates each variable until the convergence condition is met, then stops the iteration; initial values ​​for the hidden variables... , , ,in The variable representing the r-th iteration. express A 1-dimensional matrix of all ones. The F-norm of a matrix is ​​denoted by .

[0011] Furthermore, by inferring the sparse power spectrum of the signal, the estimated DOA of the far-field target is extracted, including: Based on the obtained estimates of each latent variable, The former The value is the y-coordinate. Plot the spatial range spectrum for the horizontal axis. ; In sequence elements Select all that meet the criteria. , The elements of the condition are rearranged into a new sequence. , where q is a sequence Length, For the preset minimum peak height, if Then take the sequence The largest The angle corresponding to each value is used as the estimated DOA result; if Then take As the estimated DOA result.

[0012] According to another aspect of the specification, a sparse Bayesian learning device based on far-field dictionary reconstruction under strong near-field interference is also provided, including a memory and one or more processors. The memory stores executable code, and when the processor executes the executable code, it implements the sparse Bayesian target orientation estimation method based on far-field dictionary reconstruction under strong near-field interference.

[0013] According to another aspect of the specification, a computer-readable storage medium is also provided, on which a program is stored, which, when executed by a processor, implements the sparse Bayesian target orientation estimation method based on far-field dictionary reconstruction under near-field strong interference environment.

[0014] The beneficial effects of this invention are: This invention aims to reduce the correlation between the far-field dictionary and the near-field interference signal, avoid the spurious peak masking problem caused by energy leakage of strong near-field interference in the far-field dictionary, improve the far-field target orientation estimation performance under the sparse Bayesian framework, and is simple to implement in engineering. Attached Figure Description

[0015] Figure 1 This is a flowchart illustrating the implementation of the method of the present invention.

[0016] Figure 2 This is the classification result of SVM in this embodiment.

[0017] Figure 3 The diagram shows the azimuth spectrum of the method of the present invention, the conventional beamforming method, and the sparse Bayesian learning method in the embodiments.

[0018] Figure 4 This is a schematic diagram of the device in the embodiment. Detailed Implementation

[0019] The specific embodiments of the present invention will be further described in detail below with reference to the accompanying drawings.

[0020] like Figure 1 As shown, this invention processes the array-received signal. First, it classifies the eigenspace of the signal covariance matrix to distinguish between near-field and far-field components. Then, it constructs a near-far-field hybrid dictionary and uses sparse Bayesian learning to infer the sparse power spectrum of the signal, ultimately achieving high-resolution DOA estimation of far-field targets in near-far-field hybrid sources.

[0021] Example 1: Step 1: Obtain the output signal Y of the receiving array and calculate its covariance matrix R.

[0022] Step 1.1: Sample the received signal sequence of a uniform linear array with 128 array elements. 30000, sampling frequency of 5000 Hz, assuming there are N in space f A far-field narrowband coherent signal and N n A near-field narrowband interference signal with a signal-to-interference ratio of -20dB, wherein the far-field signal is expressed as an angle... Incident, near-field interference signals The incident signal was divided into 11 subsequences. Each subsequence has a length of 5000 points, the overlap length between data segments is 2500, and the element spacing is 1.2m. The final array output signal is denoted as: in, This indicates that the array outputs a signal at time t, and L=11 is the number of snapshots; Step 1.2: Calculate the covariance matrix R of the received signal; Among them, Y H This represents the conjugate transpose of the received signal matrix.

[0023] Step 2: Perform eigenvalue decomposition on the covariance matrix R from Step 1 to obtain its eigenvectors; Step 2.1: Perform eigenvalue decomposition on the covariance matrix obtained in Step 1. The decomposition is expressed as: Where u1,…,u i …,u M (i=1,…,M) represent the eigenvectors of the covariance matrix, σ1,…, σ i, …, σ M (i=1,…,M) represents the eigenvalues ​​corresponding to each eigenvector.

[0024] Step 3: Introduce a Support Vector Machine (SVM) to classify features into near-field or far-field based on the discriminative features extracted from the feature vectors. For example... Figure 2 As shown, the feature vectors in the training set are distributed across the four discriminative feature dimensions.

[0025] Step 3.1: Establish a support vector machine model for feature vector classification. For each feature vector u obtained in Step 2... i For each i=1,…,M, calculate the following four discriminant features: near-field and far-field dictionary matching degree R1, phase consistency measure R2, phase curvature measure R3, and beamwidth measure R4. Concatenate the four features into a four-dimensional discriminant feature vector ψ. i =[R1(u i ),R2(u i ), R3(u i ),R4(u i )] T The data serves as the input features for the support vector machine. The model uses simulated array element received data as training data and establishes the support vector machine by maximizing the classification margin. The simulated array element data consists of 1000 data entries, with the far-field target angle randomly selected between 0 and 180°, the distance randomly selected between 200 and 1000 m, and the noise set to additive white Gaussian noise with a signal-to-interference ratio of -20 dB. The feature representations of each feature vector are as follows: (a) Near-field and far-field dictionary matching degree: in, Select the angle range for the far-field steering vector and the near-field steering vector. The range of distances selected for the near-field guidance vector. and The step lengths are 1° and 20m respectively; Here, η is the far-field steering vector, and η is the element spacing. This is the near-field steering vector. , , The wavelength represents the signal, c=1540m / s is the speed of sound in water, and f=200Hz is the processing frequency.

[0026] (b) Phase coherence measure: in, For feature vectors The m-th value The phase information, where real() and imag() represent taking the real and imaginary parts of the complex number, The parameters represent the linear model after least-squares fitting of the phase information, specifically calculated as follows: The solution process is achieved by calling the linear_model.LinearRegression class of Python's scikit-learn machine learning library, setting whether to fit the intercept term (fit_intercept) to True.

[0027] (c) Phase curvature metric: in, The phase curvature of the eigenvector is specifically calculated as follows: .

[0028] Beamwidth measurement: Among them, B w For spatial spectrum The beamwidth is the difference in angles between the left and right sides of the maximum peak power in the spatial spectrum when the power drops to half of the peak power. The step size is selected as 1°.

[0029] Step 3.2: After completing feature extraction and fusion, use the fused feature vector. Using the input data, a Support Vector Machine (SVM) model is trained to perform near-field and far-field feature vector classification. SVM effectively classifies samples of different classes by finding the maximum margin hyperplane. The objective function of the SVM model is: in, It is the first i The labels of each training sample are set, with the near-field label being -1 and the far-field label being 1. Indicates the first i One training sample; For parameters to be optimized, The vector norm is 2. This paper uses the Sequence Minimum Optimization (SMO) algorithm to solve the above support vector machine, implemented by calling the svm.SVC class in the scikit-learn machine learning library of Python. The key parameters of the model are configured as follows: the kernel function is set to linear basis function; the regularization penalty term C is set to 1.0; the kernel coefficient (gamma) is set to scale, i.e., automatically calculated based on the feature variance; the convergence tolerance (tol) of the optimization algorithm is set to 0.001; and the maximum number of iterations (max_iter) is set to -1, i.e., no limit is set until convergence. Finally, the optimal weight vector is obtained. , This completes the training of the classifier.

[0030] Step 3.3: Based on the trained support vector machine, classify all feature vectors into near-field or far-field, and denote the near-field feature vector index set as... The far-field feature vector index set is .

[0031] Step 3.3: Using the weight vector w = [1.48, 0.32, -0.31, -4.97] and bias b = -2.70 obtained in Step 3.2, for each feature vector u obtained in Step 2... i (i = 1,…,M) are classified. First, the four-dimensional discriminant feature vector r(u) is calculated according to step 3.1. i ) = [R1(u i ), R2(u i ), R3(u i ), R4(u i Then substitute it into the decision function g(u)], i ) = w·r(u i ) + b, classify according to the decision function value: if g(u_i) ≥ 0, then determine u i Corresponding far-field component; if g(u i If ) < 0, then determine u i This corresponds to the near-field components. Let N be the set of near-field eigenvector indices. un = {i | g(u i The set of far-field feature vector indices is N, where 0 < 0. uf = {i | g(u i ) ≥ 0}.

[0032] Step 4: Arrange all eigenvectors identified as near-field features into a near-field subspace matrix U. nf Simultaneously calculate its orthogonal projection matrix P nf .

[0033] Step 4.1: Assemble the near-field subspace matrix U from the eigenvectors identified as near-field in step (3) column-wise. nf Specifically, it is expressed as equation (5), where n1,…,n k This is the index for classifying near-field feature vectors.

[0034] Step 4.2, calculate the orthogonal projection matrix P of the near-field subspace matrix. nf ,in I M express The identity matrix, Let U be the near-field subspace matrix nf The conjugate transpose of .

[0035] Step 5: Grid the near-field and far-field observation spaces and construct an overcomplete near-field dictionary A N and far-field dictionary A F .

[0036] Step 5.1: Grid the far-field angle observation space and construct an overcomplete far-field dictionary matrix A. F ; to observe the space from the far-field angle within the range The inner part is divided into uniformly spaced 1° intervals to obtain an angular grid point set. ,in K is the total number of far-field grid points, and K f >>N f Based on the far-field angle grid point set Construct the far-field dictionary matrix A F : Step 5.2: Grid the near-field angle and distance observation space, and construct an overcomplete near-field dictionary matrix A. N ; Near-field angle observation space within range Inner The angles are evenly divided to obtain the angle grid point set. ,in The total number of near-field angle grid points; the near-field distance observation space is uniformly divided within the range [200m, 1000m] at 100m intervals to obtain the distance grid point set. ,in The total number of grid points at near-field distance; based on the grid point set at near-field angle. and near-field distance point set Construct the near-field dictionary matrix A N Its dimensions are : Step 6: Use the orthogonal projection matrix P from Step 4 nf Projecting the far-field dictionary yields the projected far-field dictionary A'. F ;The near-field dictionary A in step 5 N And the far-field dictionary A' after projection F A near-field and far-field hybrid dictionary A is constructed by column.

[0037] Step 6.1, use the orthogonal projection matrix obtained in step 4. For the far-field dictionary A in step 5 F Project the image to obtain the projected far-field dictionary A. F ': Step 6.2, use the near-field dictionary from step 5. And the far-field dictionary after projection in step 6.1 The near-field and far-field hybrid dictionary A is constructed column-wise, with dimensions of... : Step 7: Based on sparse representation theory, the DOA estimation problem is transformed into a sparse signal reconstruction problem. A corresponding sparse Bayesian learning probability model is established, and the sparse power spectrum of the signal is obtained by inference, using the near-field and far-field hybrid dictionary A from Step 6.

[0038] Step 7.1: Transform the DOA estimation problem into a sparse signal reconstruction problem by solving the following sparse matrix equations: Where X represents A 3D signal matrix, which in this embodiment is A 3D signal matrix, where n represents The dimensional additive white Gaussian noise matrix, in this embodiment is The additive white Gaussian noise matrix of dimension X, since X contains only X is a sparse matrix with 1810 non-zero row vectors.

[0039] Step 7.2, establish a sparse Bayesian probability model. First, assume a complex Gaussian distribution for each column vector of the array output signal Y. Then, the likelihood function of Y is expressed as: in, Represents the i-th column vector of matrices Y and X. The noise precision is expressed as the reciprocal of the noise variance. Perform a gamma prior distribution, i.e.: in, Let X be the parameter of the gamma distribution; next, a hierarchical sparse prior is constructed for the signal matrix X. Introducing a set of latent variables 3D matrix , dimensional matrix W and dimensional vector In this embodiment for A 3D matrix, W is 3D matrix for A 3D vector; performing a linear transformation on each column of X. ,in Let the i-th column vector of matrix Z be represented. Let X be a complex Gaussian vector with zero mean. Then X follows a complex Gaussian distribution as follows: Introduce the following latent variables: a D×L dimensional auxiliary latent variable matrix Z, whose column vector Z i They independently follow a D-dimensional complex Gaussian distribution with zero mean, i.e. I D It is a D×D dimensional identity matrix. The signal represents a complex Gaussian distribution; the K×D dimensional signal cross-correlation matrix W, where... The total number of grid points is given, and each column of W independently follows a K-dimensional zero-mean complex Gaussian distribution; the signal mean vector is K×1 dimensional. It follows a K-dimensional zero-mean complex Gaussian distribution. In this embodiment... for A 3D matrix, W is 3D matrix for A dimensional vector; representing each column of X through a linear transformation as... , Let X be a complex Gaussian vector with zero mean. Then X follows a complex Gaussian distribution as follows: in, This is the power hyperparameter vector. , Let be a scalar, representing the reciprocal of the incident target signal power in the direction of the i-th grid point, when When the signal strength is low, the signal power in the corresponding direction is high, and the probability of a target being present is high, thus enabling target location estimation. This represents the operation of generating a diagonal matrix; using the linear Gaussian prior model shown in formula (14), the autocorrelation component of the signal X can be included in the matrix. In this matrix, the cross-correlation components are contained in matrix W; In the second layer of priors, the hyperparameters are respectively... Make a prior hypothesis, assume Each column and All follow a zero-mean complex Gaussian distribution. Each element is independently and identically distributed, and all follow a gamma distribution, i.e.: Where D is 128, It is 1810. Indicates the first W i Column vector, express The i-th element, , For the parameters of the gamma distribution, For accuracy parameters, This represents the matrix inversion operation; In the third layer prior, regarding hyperparameters For each element, make a gamma prior hypothesis: in, Parameters for the gamma distribution; set initial parameter values. ,remember This is called the set of latent variables; Step 7.4: Calculate each latent variable using the variational Bayesian inference method. Approximate posterior distribution ,get: That is, the first in X i column vector The posterior distribution is a complex Gaussian distribution with a mean of and variance They are respectively: , .in, , Let Y and Z represent the i-th column vectors, respectively. This represents the expectation operation. This represents the matrix conjugate transpose operation; That is, the first in W j column vector The posterior distribution is a complex Gaussian distribution with a mean of and variance They are respectively: in Represents the first j Line 1 i Column elements, express The j Each element.

[0040] That is, the first in Z i Column vector Z i The posterior distribution is a complex Gaussian distribution with a mean of and variance They are respectively: ,Right now The posterior distribution is a complex Gaussian distribution with a mean of and variance They are respectively: ,Right now The Middle i element The posterior distribution of is a gamma distribution, with parameters g and h as follows: ,in This refers to the 2-norm operation for vectors.

[0041] ,Right now The posterior distribution of is a gamma distribution, and the parameters of this distribution are... c,d for: ,in Represents the trace of a matrix.

[0042] ,Right now The Middle j element The posterior distribution of is a gamma distribution, and the parameters of this distribution are... a,b They are respectively: W j Indicates the first W j Column vector, Represents absolute value; according to The update formula, after setting initial values ​​for the hidden variables, iterates and updates each variable until the convergence condition is met, then stops the iteration; initial values ​​for the hidden variables... , , ,in The variable representing the r-th iteration. Represents a matrix of all 1s. Let F be the norm of the matrix, and let convergence condition be... Step 8: Extract the DOA estimate of the far-field target. For example... Figure 3 The figure shows the spatial spectrum of the far-field target DOA obtained by the method of this patent (SSBL), conventional beamforming (CBF), and sparse Bayesian learning (SBL). The pentagrams in the figure mark the far-field target DOA estimates obtained in this embodiment, and the circles mark the far-field target DOA true values. It can be seen that compared with the CBF method and the SBL method, the method of this patent effectively suppresses near-field interference of about 0~20° and obtains more accurate far-field target DOA estimates.

[0043] Step 8.1: Based on the estimated values ​​of each latent variable obtained in Step 7, ... The first 181 values ​​are the ordinate. Plot the spatial range spectrum for the horizontal axis. ,like Figure 3 As shown by the solid black line in the middle.

[0044] Step 8.2, in the sequence elements Select all that meet the criteria. , The elements of the condition are rearranged into a new sequence. Where q=12 is a sequence Length, The minimum peak height is preset. This represents the pre-set number of far-field targets. At this point, q is 12, which is greater than... Therefore, take the sequence The angles corresponding to the four largest values ​​are used as the estimated DOA results, such as... Figure 3 As shown by the green star in the image.

[0045] Corresponding to the aforementioned embodiment of a sparse Bayesian target orientation estimation method based on far-field dictionary reconstruction under strong near-field interference, the present invention also provides an embodiment of a sparse Bayesian learning device based on far-field dictionary reconstruction under strong near-field interference.

[0046] See Figure 4The present invention provides a sparse Bayesian learning device based on far-field dictionary reconstruction under near-field strong interference environment, including a memory and one or more processors. The memory stores executable code, and when the processor executes the executable code, it is used to implement a sparse Bayesian target orientation estimation method based on far-field dictionary reconstruction under near-field strong interference environment as described in the above embodiment.

[0047] The embodiment of the sparse Bayesian learning device based on far-field dictionary reconstruction under near-field strong interference environment provided by this invention can be applied to any device with data processing capabilities, such as a computer. The device embodiment can be implemented in software, hardware, or a combination of both. Taking software implementation as an example, as a logical device, it is formed by the processor of any data processing device loading the corresponding computer program instructions from non-volatile memory into memory for execution. From a hardware perspective, such as... Figure 4 The diagram shown is a hardware structure diagram of any data processing-capable device, including the sparse Bayesian learning device based on far-field dictionary reconstruction under near-field strong interference environment provided by the present invention. (Except for...) Figure 4 In addition to the processor, memory, network interface, and non-volatile memory shown, any data processing device in the embodiment may also include other hardware depending on the actual function of the data processing device, which will not be described in detail here.

[0048] The specific implementation process of the functions and roles of each unit in the above device can be found in the implementation process of the corresponding steps in the above method, and will not be repeated here.

[0049] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of the present invention according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0050] This invention also provides a computer-readable storage medium storing a program that, when executed by a processor, implements a sparse Bayesian target location estimation method based on far-field dictionary reconstruction under near-field strong interference conditions as described in the above embodiments.

[0051] The computer-readable storage medium can be an internal storage unit of any data processing device described in any of the foregoing embodiments, such as a hard disk or memory. The computer-readable storage medium can also be an external storage device of any data processing device, such as a plug-in hard disk, smart media card (SMC), SD card, flash card, etc., equipped on the device. Furthermore, the computer-readable storage medium can include both internal storage units and external storage devices of any data processing device. The computer-readable storage medium is used to store the computer program and other programs and data required by the data processing device, and can also be used to temporarily store data that has been output or will be output.

[0052] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the sparse Bayesian target location estimation method based on far-field dictionary reconstruction under near-field strong interference environment.

[0053] Other embodiments of this application will readily occur to those skilled in the art upon consideration of the specification and practice of the disclosure herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein. The specification and embodiments are to be considered exemplary only, and the true scope and spirit of this application are indicated by the claims.

[0054] It should be understood that the foregoing general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this application. This application is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this application is limited only by the appended claims.

Claims

1. A sparse Bayesian target location estimation method based on far-field dictionary reconstruction under near-field strong interference environment, characterized in that, Includes the following steps: The output signal of the receiving array is acquired and its covariance matrix is ​​calculated. The covariance matrix is ​​decomposed into eigenvectors to obtain its eigenvectors. Discriminant features are extracted based on the eigenvectors. A support vector machine is introduced to classify the discriminant features into near-field or far-field. All eigenvectors identified as near-field are arranged into a near-field subspace matrix by columns, and their orthogonal projection matrix is ​​calculated simultaneously. Grid the near-field and far-field observation spaces and construct ultra-complete near-field and far-field dictionaries; The far-field dictionary is projected using an orthogonal projection matrix to obtain the projected far-field dictionary; the near-field dictionary and the projected far-field dictionary are then combined into a near-far-field hybrid dictionary by column. Based on sparse representation theory, the DOA estimation problem is transformed into a sparse signal reconstruction problem using a near-field and far-field hybrid dictionary. A corresponding sparse Bayesian learning probability model is established to obtain the estimated values ​​of each latent variable. The sparse power spectrum of the signal is obtained by inference, and the DOA estimate of the far-field target is extracted.

2. The sparse Bayesian target location estimation method based on far-field dictionary reconstruction under near-field strong interference environment as described in claim 1, characterized in that, The eigenvalue decomposition of the covariance matrix includes: in, Let the covariance matrix be u1,…,u i …,u M (i=1,…,M) represent the eigenvectors of the covariance matrix, σ1,…, σ i, …, σ M (i=1,…,M) represents the eigenvalues ​​corresponding to each eigenvector.

3. The sparse Bayesian target location estimation method based on far-field dictionary reconstruction under near-field strong interference environment as described in claim 1, characterized in that, The extraction of discriminative features based on feature vectors is as follows: Near-field and far-field dictionary matching degree, phase consistency measure, phase curvature measure and beamwidth measure are extracted from the feature vector. The four features are concatenated into a four-dimensional discriminative feature vector, which is used as the input feature of the support vector machine. The support vector machine classifies samples of different categories into near-field or far-field by finding the maximum margin hyperplane.

4. The sparse Bayesian target location estimation method based on far-field dictionary reconstruction under near-field strong interference environment as described in claim 3, characterized in that, The pre-training process of the support vector machine includes: Calculate the representation of each feature using the simulated array element received data as training data: Near-field and far-field dictionary matching degree: , in, The angles selected for the far-field steering vector and the near-field steering vector. The distance chosen for the near-field guidance vector. Select the range of angles for the far-field steering vector and the near-field steering vector. The range of distances selected for the near-field guidance vector. For far-field steering vector, This is the near-field steering vector. , , where is the element index and η is the element spacing. Indicates the signal wavelength. c The speed of sound in water, f For processing frequency; Phase consistency measure: in, For feature vectors The m-th value The phase information, where real() and imag() represent taking the real and imaginary parts of the complex number, This represents the parameters of the linear model after least-squares fitting of the phase information. ; Phase curvature measurement: in, The phase curvature of the eigenvector is specifically calculated as follows: ; Beamwidth measurement: Among them, B w For spatial spectrum The beamwidth is the difference in angles between the left and right sides of the maximum peak power in the spatial spectrum when the power drops to half of the peak power. The selection step size is 1°; The extracted features are fused. , Update using the objective function: ; in, It is the first i The labels of each training sample are set, with the near-field label being -1 and the far-field label being 1. Indicates the first i One training sample; For parameters to be optimized, It is the vector 2 norm.

5. The sparse Bayesian target location estimation method based on far-field dictionary reconstruction under near-field strong interference environment as described in claim 1, characterized in that, The step of constructing a near-field subspace matrix from all the feature vectors identified as near-field along their columns, and simultaneously calculating its orthogonal projection matrix, includes: The eigenvectors identified as near-field are arranged in columns to form a near-field subspace matrix U. nf where n1,…,n k Indexes for classification as near-field feature vectors: ; Calculate the orthogonal projection matrix P of the near-field subspace matrix. nf ,in I M express The identity matrix, Represents the near-field subspace matrix U nf Conjugate transpose: .

6. The sparse Bayesian target location estimation method based on far-field dictionary reconstruction under near-field strong interference environment as described in claim 1, characterized in that, The construction of the overcomplete near-field dictionary and far-field dictionary includes: Gridded far-field angle observation space, constructing an overcomplete far-field dictionary matrix A F ; to observe the space from the far-field angle within the range The inner part is divided into uniformly spaced 1° intervals to obtain an angular grid point set. K f K is the total number of far-field grid points, and K f >>N f Based on the far-field angle grid point set Construct the far-field dictionary matrix A F : ; Grid the near-field angle and range observation space, and construct an overcomplete near-field dictionary matrix A. N ; Near-field angle observation space within range Inner The angles are evenly divided to obtain the angle grid point set. ,in This represents the total number of grid points in the near-field angle, and The near-field distance observation space is uniformly divided within the range [200m, 1000m] at 100m intervals to obtain a distance grid point set. ,in The total number of grid points in the near field distance, and Based on the near-field angle grid point set and near-field distance point set Construct the near-field dictionary matrix A N Its dimensions are : .

7. The sparse Bayesian target location estimation method based on far-field dictionary reconstruction under near-field strong interference environment as described in claim 1, characterized in that, The method of using a near-field and far-field hybrid dictionary to transform the DOA estimation problem into a sparse signal reconstruction problem is as follows: ; in, Represents a near-field and far-field mixed dictionary, where X represents... A signal matrix of dimension n, where n represents The additive white Gaussian noise matrix of dimension X, since X contains only K non-zero row vectors, X is a sparse matrix, and K f This represents the total number of far-field grid points. This represents the total number of grid points in the near-field angle. This represents the total number of grid points in the near-field distance. To establish a sparse Bayesian probability model, first, a complex Gaussian distribution assumption is made for each column vector of the array output signal Y. Then, the likelihood function of Y is expressed as: ; in, Represents the i-th column vector of matrices Y and X. The noise precision is expressed as the reciprocal of the noise variance. Perform a gamma prior distribution, i.e.: , These are the parameters of the gamma distribution; Next, a hierarchical sparse prior is constructed for the signal matrix X. In the first layer of priors, a set of latent variables is introduced. 3D auxiliary latent variable matrix , K×1 matrix W and K×1 signal mean vector Perform a linear transformation on each column of X. ,in Let the i-th column vector of matrix Z be represented. Let X represent a complex Gaussian vector with zero mean. Then the linear Gaussian prior model is: X follows the following complex Gaussian distribution: ; in, , This indicates the operation of generating a diagonal matrix; Let be a scalar, representing the reciprocal of the incident target signal power in the direction of the i-th grid point; using a linear Gaussian prior model allows the autocorrelation component of signal X to be included in the matrix. In this matrix, the cross-correlation components are contained in matrix W; In the second layer of priors, the hyperparameters are respectively... Make a prior hypothesis, assume Each column and All follow a zero-mean complex Gaussian distribution. Each element is independently and identically distributed, and all follow a gamma distribution, i.e.: in, Let W represent the i-th column vector. express The i-th element, , For the parameters of the gamma distribution, For accuracy parameters, This represents the matrix inversion operation; In the third layer prior, regarding hyperparameters For each element, make a gamma prior hypothesis: ; in, Parameters for the gamma distribution; set initial parameter values. ,remember This is called the set of latent variables; The latent variables are calculated using the variational Bayesian inference method. Approximate posterior distribution ,get: That is, the first in X i column vector The posterior distribution is a complex Gaussian distribution with a mean of and variance They are respectively: , ;in, , Let Y and Z represent the i-th column vectors, respectively. This represents the expectation operation. This represents the matrix conjugate transpose operation; That is, the j-th column vector in W The posterior distribution is a complex Gaussian distribution with a mean of and variance They are respectively: in Represents the first j Line 1 i Column elements, express The j One element; That is, the first in Z i Column vector Z i The posterior distribution is a complex Gaussian distribution with a mean of and variance They are respectively: ,Right now The posterior distribution is a complex Gaussian distribution with a mean of and variance They are respectively: ,Right now The Middle i element The posterior distribution of is a gamma distribution, with parameters g and h as follows: ,in For vector 2 norm operations; ,Right now The posterior distribution of is a gamma distribution, and the parameters of this distribution are... c,d for: ,in Represents the trace of a matrix; ,Right now The Middle j element The posterior distribution of is a gamma distribution, and the parameters a and b of this distribution are: W j Let W be the j-th column vector. Represents absolute value; according to The update formula, after setting initial values ​​for the hidden variables, iterates and updates each variable until the convergence condition is met, then stops the iteration; initial values ​​for the hidden variables... , , ,in The variable representing the r-th iteration. express A 1-dimensional matrix of all ones. This represents the F-norm of the matrix.

8. A sparse Bayesian target location estimation method based on far-field dictionary reconstruction under near-field strong interference environment as described in claim 7, characterized in that, By inferring the sparse power spectrum of the signal, the estimated DOA of the far-field target is extracted, including: Based on the obtained estimates of each latent variable, The former The value is the y-coordinate. Plot the spatial range spectrum for the horizontal axis. ; In sequence elements Select all that meet the criteria. , The elements of the condition are rearranged into a new sequence. , where q is a sequence Length, For the preset minimum peak height, if Then take the sequence The largest The angle corresponding to each value is used as the estimated DOA result; if Then take As the estimated DOA result.

9. A sparse Bayesian learning device based on far-field dictionary reconstruction under near-field strong interference environment, comprising a memory and one or more processors, wherein the memory stores executable code, characterized in that, When the processor executes the executable code, it implements a sparse Bayesian target orientation estimation method based on far-field dictionary reconstruction under near-field strong interference environment as described in any one of claims 1-8.

10. A computer-readable storage medium having a program stored thereon, characterized in that, When the program is executed by the processor, it implements a sparse Bayesian target orientation estimation method based on far-field dictionary reconstruction under near-field strong interference environment as described in any one of claims 1-8.

Images