Decomposition and deconvolution beam forming method based on subspace clustering
By using a decomposition-based deconvolution beamforming method based on subspace clustering, the problem of poor target detection capability of deconvolution beamforming algorithms under strong interference environments is solved, and the effect of improving array gain and target detection capability under strong interference is achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- HARBIN ENG UNIV
- Filing Date
- 2026-01-09
- Publication Date
- 2026-05-01
AI Technical Summary
Existing deconvolution beamforming algorithms have poor target detection capabilities in strong interference environments, resulting in decreased array gain and difficulty in effectively addressing the problem of target being overwhelmed by sidelobes in strong interference.
A decompositional deconvolutional beamforming method based on subspace clustering is adopted. By constructing the data covariance matrix, subspace decomposition is performed to obtain strong signal subspace and weak signal subspace. Combined with zero-travel constrained beamforming and signal-to-noise ratio fusion, the target detection capability is improved.
It improves target detection capability under strong interference environment, suppresses the impact of interference on weak targets, enhances the robustness of the algorithm and array gain, and effectively suppresses the energy of interference leaking into the weak signal subspace component.
Smart Images
Figure CN121955868A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of direction of arrival (DOA) estimation technology in passive detection, and specifically relates to a decomposition deconvolution beamforming method based on subspace clustering. Background Technology
[0002] Underwater target location (DOA) estimation is a popular research area in sonar passive detection, aiming to estimate the location of a target of interest from data containing interference and noise. However, conventional beamforming methods have wide main lobes and high sidelobe levels, making the target of interest easily submerged in the sidelobes of strong interference. Therefore, to detect targets in strong interference backgrounds and improve DOA estimation performance, researchers have studied a series of algorithms, such as minimum variance distortionless response beamformers, multiple signal classification, and sparse Bayesian learning, achieving some research results. However, in general, these algorithms still exhibit reduced output signal-to-noise ratio and the generation of spurious peaks under many non-ideal conditions, such as snapshot number limitations, array manifold errors, and non-uniform noise.
[0003] Deconvolutional beamforming has advantages such as high resolution, low computational complexity, and strong robustness, and can effectively cope with non-ideal conditions such as snapshot number limitations, array manifold errors, and non-uniform noise. However, since deconvolution is an ill-conditioned inverse problem, the array gain of deconvolutional beamforming will decrease at low signal-to-interference-plus-noise ratios, leading to a decline in target DOA estimation performance (Yang T C. Deconvolved conventional beamforming for a horizontal line array. IEEE J. Oceanic Eng, 2018; 43(1): 160−172). However, there is relatively little research on the problem of reduced target detection capability of deconvolution beamforming arrays under low signal-to-interference-plus-noise ratio (SNR). For example, Reference 1 (D. Sun, C. Ma, and TCYang. Improving the performance of a vector sensor line array by deconvolution. IEEE J. Oceanic Eng, 2020, 45(3):1063–1077.) proposes an extended RL deconvolution beamforming method, which enables the deconvolution algorithm to be applied to arrays with translational beammaps. Compared with DAMAS and NNLS, Reference 1 can achieve higher array gain under low SNR conditions. Reference 2 (Tianfeng Huang, Dajun Sun, et al. Direction of arrival estimation by Perona & Malik regularized Richardson-Lucy deconvolution. J. Acoust. Soc. Am, 2025; 158 (2): 1133–1145.) proposes a deconvolution beamforming method based on Perona & Malik regularization, which alleviates the ill-conditioned inverse problem by introducing a regularization term, thereby reducing background noise and improving array gain. Reference 3 (TC Yang. Deconvolution of decomposed conventional beamforming. J. Acoust. Soc. Am, 2020, 148(2): EL195–EL201.) proposes a deconvolution beamforming method based on the signal subspace of conventional beamforming from a subspace perspective, which improves array gain to some extent by projecting the data covariance matrix onto the signal subspace. Although the above methods improve the target detection capability under low signal-to-interference-plus-noise ratio to some extent, they cannot effectively cope with scenarios where strong interference sidelobes overwhelm the target.In summary, although existing deconvolution beamforming algorithms have a certain degree of robustness under non-ideal conditions, they still suffer from poor target detection capability due to the decrease in array gain under strong interference environments. Therefore, proposing a new passive DOA estimation method to solve the above problems is an urgent need. Summary of the Invention
[0004] The purpose of this invention is to address the problem of poor target detection capability in existing deconvolution beamforming algorithms, and to propose a decomposition deconvolution beamforming method based on subspace clustering.
[0005] The technical solution adopted by the present invention to solve the above-mentioned technical problems is: a decomposition deconvolution beamforming method based on subspace clustering, the method specifically including the following steps:
[0006] Step 1: Construct the data covariance matrix based on the received data from the uniform linear array;
[0007] Step 2: Analyze the data covariance matrix. Subspace decomposition is performed to obtain strong signal subspace, weak signal subspace components, and noise subspace;
[0008] Step 3: Obtain the weak signal subspace deconvolution azimuth spectrum and the strong signal subspace deconvolution azimuth spectrum respectively through beamforming;
[0009] Step 4: Fuse the deconvolution azimuth spectrum of the strong signal subspace and the deconvolution azimuth spectrum of the weak signal subspace to obtain the DOA estimation result.
[0010] Furthermore, each element of the uniform linear array in The elements are evenly spaced on the positive half-axis of the axis, and the distance between adjacent elements is denoted as . .
[0011] Furthermore, the uniform linear array in The data received at any given time is modeled as follows:
[0012]
[0013] in:
[0014] Indicates by A uniform linear array composed of 1000 array elements Data received in real time;
[0015] Indicates the first The incident azimuth of the target, , Indicates the number of incident targets;
[0016] Indicates the first The signal vector of each target;
[0017] Represents the noise vector;
[0018] Represents a signal vector;
[0019] express 3D array manifold matrix, , Indicates the first The array steering vector corresponding to the incident azimuth of each target;
[0020] ,
[0021] in:
[0022] The base of the natural logarithm;
[0023] Represents the imaginary unit;
[0024] Indicates the signal wavelength;
[0025] Represents transpose;
[0026] use Construct the data covariance matrix from the array receiving data of the number of snapshots. :
[0027]
[0028] in:
[0029] Represents the data covariance matrix;
[0030] yes The conjugate transpose of . .
[0031] Furthermore, the specific process of step two is as follows:
[0032] Step 2.1: Analyze the data covariance matrix. Perform eigenvalue decomposition:
[0033]
[0034] in:
[0035] For the data covariance matrix The 1 eigenvalue, ;
[0036] Eigenvalues The corresponding feature vector;
[0037] yes The conjugate transpose of;
[0038] Step 22: Obtain the results from Step 21 Sort the _th feature values in ascending order, and then select the _th ... The eigenvalues are denoted as ;
[0039] Steps two and three: Preprocess each feature value in the sorting result separately:
[0040]
[0041] in, For the sorted results, the first eigenvalues The logarithmic representation of Indicates multiplication;
[0042] Step 24: Construct the eigenvalue variance matrix :
[0043]
[0044] in:
[0045] Represents the eigenvalue variance matrix The Middle Line number Column elements;
[0046] Indicates the first in the sorting results The eigenvalue to the ... The average of the logarithmic representations of the eigenvalues;
[0047] when hour, The value is 0;
[0048] Step 25: Minimize the variance sum matrix The first line of the middle Column elements Minimum variance matrix The Middle Line number Column elements for:
[0049]
[0050] in, ;
[0051] Let the breakpoint matrix The Middle Line number Column elements equals making The value of reaches the minimum. ,Right now:
[0052]
[0053] in, ;
[0054] Step 26: Set the first breakpoint for:
[0055]
[0056] in, The breakpoint in the 3rd row of the matrix Column elements;
[0057] Set the second breakpoint for:
[0058]
[0059] in, The second row of the breakpoint matrix Column elements;
[0060] Step 27: Transfer the first eigenvalue to the second eigenvalue. The eigenvalue is taken as the first type of eigenvalue, and the eigenvalue is taken as the second type of eigenvalue. The eigenvalues up to the ... The eigenvalue is used as the second type of eigenvalue, and the eigenvalue is used as the second type of eigenvalue. The eigenvalues up to the ... These eigenvalues are used as the third type of eigenvalues;
[0061] Weak signal subspace components are obtained based on the second type of eigenvalues. The strong signal subspace component is obtained based on the third type of feature value. .
[0062] Furthermore, the weak signal subspace component is obtained based on the second type of feature value. Specifically:
[0063]
[0064] in:
[0065] Represents the weak signal subspace components;
[0066] For the sorted results, the first eigenvalues The corresponding feature vector, yes The conjugate transpose of .
[0067] Furthermore, the strong signal subspace component is obtained based on the third type of feature value. Specifically:
[0068]
[0069] in, This represents the strong signal subspace component.
[0070] Furthermore, the specific process of step three is as follows:
[0071] Step 31, in Beamforming is performed on the strong signal subspace components to obtain the strong signal subspace azimuth spectrum. ;
[0072]
[0073] in: The scanning angle on the azimuth scanning grid The corresponding beamforming weight vector, for The conjugate transpose of . , Indicates angle The corresponding array steering vector, ;
[0074] Step 3.2: Set the maximum number of iterations to... ;
[0075] Step 3: Initialize the strong signal subspace source distribution function and initialize the number of iterations. ;
[0076] Steps three and four: Calculate the first... The strong signal subspace source distribution function of the next iteration :
[0077]
[0078] in:
[0079] Indicates the first The strong signal subspace source distribution function of the next iteration;
[0080] Indicates the azimuth of the beamforming deconvolution estimation;
[0081] Indicates beamforming deconvolution in A dictionary of point spread functions in terms of direction;
[0082] Step 35: Determine whether it has been obtained. ;
[0083] If obtained Then let the strong signal subspace deconvolve azimuth spectrum And continue with step three six;
[0084] If not received Then let Return to steps three and four;
[0085] Step 36: Deconvolve the azimuth spectrum of the strong signal subspace. The angles corresponding to each local spectral peak are taken as the null angles for forming the null constraint beam in the weak signal subspace, and the first... The angles corresponding to each local spectral peak are denoted as . , , Represents the azimuth spectrum of deconvolution in the strong signal subspace The total number of local spectral peaks;
[0086] Calculate the weight vector of the null-constrained beam and point spread function dictionary :
[0087]
[0088]
[0089] in:
[0090] The scanning angle on the azimuth scanning grid The corresponding null-containment beamforming weight vector, yes The conjugate transpose of;
[0091] This indicates that null-constrained beamforming deconvolution is in A dictionary of point spread functions in terms of direction;
[0092] express The conjugate transpose of;
[0093] The azimuth spectrum represents the beamforming deconvolution estimation. The corresponding array steering vector;
[0094]
[0095] in: express The conjugate transpose of;
[0096] for The orthogonal projection matrix of the array manifold of the interference:
[0097]
[0098] in: express An array manifold matrix of interference, express The conjugate transpose of , where the superscript -1 denotes the inverse of the matrix. For unit array;
[0099]
[0100] in: Indicates angle The corresponding array steering vector, ;
[0101] Step 37, in Null-constrained beamforming is performed on the weak signal subspace components to obtain the weak signal subspace azimuth spectrum. :
[0102]
[0103] Step 38: Calculate intermediate variables , Indicates the scanning orientation The normalization function, Initialize the weak signal subspace source distribution function ;
[0104] Set the maximum number of iterations to and initialize the number of iterations. ;
[0105] Step 39: Calculate the weak signal subspace source distribution function :
[0106]
[0107] in:
[0108] Indicates the first The weak signal subspace source distribution function obtained in the next iteration;
[0109] This represents the normalized point spread function. ;
[0110] It is the normalized weak signal subspace orientation spectrum. ;
[0111] Step 30: Determine whether it has been obtained. ;
[0112] If obtained Then let the weak signal subspace deconvolve azimuth spectrum , , This indicates the azimuth estimation based on beamforming deconvolution. The normalization function;
[0113] If not received Then let Return to step 39.
[0114] Furthermore, the specific process of step four is as follows:
[0115]
[0116] in:
[0117] The azimuth spectrum after fusion;
[0118] The signal-to-noise ratio in each direction of the strong signal subspace component;
[0119] The signal-to-noise ratio in each direction of the weak signal subspace component;
[0120] Obtain the azimuth spectrum The local spectral peaks are used to estimate the DOA of each target, and the angle corresponding to each local spectral peak is used as the DOA estimation result of each target.
[0121] Furthermore, the signal-to-noise ratio in each direction of the strong signal subspace component Signal-to-noise ratio in each direction of the weak signal subspace component They are respectively:
[0122]
[0123]
[0124] in, express background value, express Background value;
[0125] The method for calculating the background value is as follows:
[0126] Step 1: Record the number of scanning angles on the azimuth grid as... , will the The azimuth corresponding to each scanning angle is denoted as . Then, the strong signal subspace deconvolution azimuth spectrum is represented as a length of... vector The weak signal subspace deconvolution azimuth spectrum is represented as a spectrum of length . vector ;
[0127] Step 2: Initialize the forward filtering results: , For have:
[0128]
[0129]
[0130] in:
[0131] These are the filter coefficients. Indicates the azimuth corresponding to the first scanning angle. The corresponding value in the deconvolution azimuth spectrum of the strong signal subspace. Indicates the first The azimuth corresponding to each scanning angle The corresponding value in the deconvolution azimuth spectrum of the strong signal subspace;
[0132] from Start traversing to End, and the forward filtering result is obtained. and ;
[0133] Step 3: Initialize the backward filtering results: , For have:
[0134]
[0135]
[0136] from Start traversing to The process ends, and the back-filtering result is obtained. and ;
[0137] Then the first strong signal subspace Each scanning angle corresponds to the azimuth. Background values and weak signal subspace Each scanning angle corresponds to the azimuth. The background values are respectively , .
[0138] The beneficial effects of this invention are:
[0139] To mitigate the impact of strong interference on weak targets, this invention proposes a subspace self-clustering method. It adaptively decomposes the covariance matrix into strong signal and weak signal subspaces based on the magnitude of the eigenvalues, improving target detection capability and addressing the array gain reduction problem of traditional deconvolution beamforming algorithms in strong interference scenarios. The proposed DOA estimation method introduces null-constrained deconvolution beamforming, which effectively suppresses energy leakage of interference into the weak signal subspace components under non-ideal conditions. Furthermore, the fusion processing enhances the algorithm's robustness, further improving target detection capability. Attached Figure Description
[0140] Figure 1 This is a flowchart of a decomposition deconvolution beamforming method based on subspace clustering according to the present invention;
[0141] Figure 2 A comparison diagram of the deconvolution azimuth spectrum of the strong signal subspace and the deconvolution azimuth spectrum of the weak signal subspace;
[0142] Figure 3 The results of the orientation spectrum comparison between the method of this invention and five other methods are shown.
[0143] Figure 4 The curves showing the array gain versus signal-to-noise ratio for the method of this invention and the conventional beamforming deconvolution method are shown.
[0144] Figure 5 The curves showing the array gain versus interference-to-noise ratio for the method of this invention and the conventional beamforming deconvolution method are shown. Detailed Implementation
[0145] Specific implementation method one: Combining Figure 1 This embodiment describes a method for decomposition deconvolution beamforming based on subspace clustering, which specifically includes the following steps:
[0146] Step 1: Construct the data covariance matrix based on the received data from the uniform linear array;
[0147] Specifically, each element of a uniform linear array in The elements are evenly spaced on the positive half-axis of the axis, and the distance between adjacent elements is denoted as . ;
[0148] Taking due east as the positive x-axis and due north as the positive y-axis, the target incident angle is defined as the angle between the two axes in the north-northeast direction. Assume that... If there are independent far-field narrowband signals, then the uniform linear array in The data received at any given time is modeled as follows:
[0149]
[0150] in, Indicates by A uniform linear array composed of 1000 array elements Data received in real time Indicates the first The incident azimuth of the target, , Indicates the number of incident targets. Indicates the first The signal vector of each target , Indicates the first The array steering vector corresponding to the incident azimuth of each target. The base of the natural logarithm. Represents the imaginary unit. Indicates the signal wavelength, superscript Represents transpose. Represents the noise vector. Represents a signal vector. express 3D array manifold matrix, ;
[0151] use Construct the data covariance matrix from the array receiving data of the number of snapshots. :
[0152]
[0153] in, Represents the data covariance matrix. yes The conjugate transpose of .
[0154] It should be noted that: to prevent the sample covariance matrix from differing too much from the actual covariance matrix due to an insufficient number of snapshots, thus causing errors in subsequent eigenvalue decomposition, a specific number of snapshots is set. Usually not less than That is, the number of quick shots It is usually not less than the number of array elements.
[0155] Step 2: Analyze the data covariance matrix using clustering methods. Subspace decomposition is performed to obtain strong signal subspace, weak signal subspace components, and noise subspace;
[0156] Specifically:
[0157] Step 2.1: Analyze the data covariance matrix. Perform eigenvalue decomposition:
[0158]
[0159] in, For the data covariance matrix The 1 eigenvalue, Eigenvalues The corresponding feature vector, yes The conjugate transpose of . ;
[0160] Step 22: Obtain the results from Step 21 Sort the _th feature values in ascending order, and then select the _th ... The eigenvalues are denoted as ;
[0161] Steps 2 and 3: Preprocess the eigenvalues:
[0162]
[0163] in, For the first eigenvalues The logarithmic representation of Indicates multiplication;
[0164] Since the eigenvalues are arranged in ascending order, the logarithmic representation is... They are also arranged in ascending order;
[0165] Step 24: Construct the eigenvalue variance matrix :
[0166]
[0167] in, Represents the eigenvalue variance matrix The Middle Line number Column elements, Indicates the first in the sorting results The eigenvalue to the ... The average of the logarithmic representations of the eigenvalues;
[0168] when hour, The value of is 0, which means the eigenvalue variance matrix is... It is an upper triangular matrix. The elements on the main diagonal and below it are all set to 0, which have no actual meaning and are not involved in subsequent calculations;
[0169] Step 25: Initialize the minimum variance and matrix For a 3-line Given a matrix of columns, find the minimum variance matrix. The Middle Line number Column elements ; This indicates that the first few in the sorted results will be selected. The eigenvalues are optimally divided into When there are 10 categories, the minimum sum of the internal variances of each category that can be achieved is... This means that when the first [item] in the sorting result is [the first item], [the result] is [the first item]. The case where each feature value is classified into only one class;
[0170] The criterion for classifying categories is to make those within the same category as similar as possible, and those between different categories as different as possible. Minimizing the sum of variances within classes is equivalent to maximizing the sum of variances between classes; the goal is to find the breakpoint that minimizes the sum of variances within classes.
[0171] Minimum variance matrix The Middle Line number Column elements for:
[0172]
[0173] in, ;
[0174] Let the breakpoint matrix The Middle Line number Column elements equals making The value of reaches the minimum. ,Right now:
[0175]
[0176] in, ;
[0177] Record the top of the sorted results The eigenvalues are optimally divided into The location of the last breakpoint when considering each category; and it should be noted that when At that time, due to The eigenvalues cannot be divided. Class, so at this point in the minimum variance and matrix The corresponding elements in the matrix have no practical meaning. Furthermore, the breakpoint matrix... The first line in the text has no practical meaning because it will be the first line of the text. No breakpoints need to be set when the data is optimally divided into 1 class, and if At that time, matrix elements in They also have no practical meaning, and these elements that have no practical meaning can all be set to inf.
[0178] Step 26: Set the first breakpoint for:
[0179]
[0180] in, The breakpoint in the 3rd row of the matrix Column elements;
[0181] Set the second breakpoint for:
[0182]
[0183] in, The second row of the breakpoint matrix Column elements;
[0184] This invention utilizes breakpoints and breakpoints The feature values in the sorting results are divided into three categories. The first category includes the first feature value in the sorting results up to the second. The second type of feature value includes the first feature value in the sorting result. The eigenvalues up to the ... The third type of feature value includes the first feature value in the sorting result. The eigenvalues up to the ... One eigenvalue;
[0185] Step 27: Obtain the noise subspace components based on the first type of feature values:
[0186]
[0187] in, For the sorted results, the first eigenvalues The corresponding feature vector, yes The conjugate transpose of . Represents the noise subspace components;
[0188] Weak signal subspace components are obtained based on the second type of eigenvalues:
[0189]
[0190] in, Represents the weak signal subspace components;
[0191] The strong signal subspace components are obtained based on the third type of eigenvalues:
[0192]
[0193] in, This represents the strong signal subspace component.
[0194] Step 3: Obtain the deconvolution azimuth spectrum of the weak signal subspace and the deconvolution azimuth spectrum of the strong signal subspace through beamforming; wherein, the strong signal subspace uses the conventional beamforming deconvolution method, and the weak signal subspace uses the null-constrained beamforming deconvolution method, and the null azimuth is the interference azimuth obtained in the strong signal subspace.
[0195] Specifically:
[0196] Step 31, in Beamforming is performed on the strong signal subspace components to obtain the strong signal subspace azimuth spectrum. :
[0197]
[0198] in, The scanning angle on the azimuth scanning grid The corresponding beamforming weight vector, for The conjugate transpose of . , Indicates angle The corresponding array steering vector, ;
[0199] For azimuth scanning grids, the scanning angle range needs to be divided into grids according to the actual direction finding accuracy requirements. Generally, the quantization accuracy of the grid is required to be less than 1 / 5 of the measurement accuracy requirement.
[0200] The strong signal subspace azimuth spectrum is subjected to beamforming deconvolution using the RL algorithm, i.e., steps 32 to 35 are executed.
[0201] Step 3.2: Use the RL algorithm to perform beamforming deconvolution on the strong signal subspace azimuth spectrum, setting the maximum number of iterations to [value missing]. (In this invention, the value is taken as 500 times);
[0202] Step 3: Initialize the strong signal subspace source distribution function and initialize the number of iterations. ;
[0203] Steps three and four: Calculate the first... The strong signal subspace source distribution function of the next iteration :
[0204]
[0205] in, Indicates the first The source distribution function of the strong signal subspace in the next iteration. Indicates beamforming deconvolution in A dictionary of point spread functions in terms of direction. Indicates the azimuth of the beamforming deconvolution estimation;
[0206]
[0207] in, Represents the Sink function;
[0208] Step 35: Determine whether it has been obtained. ;
[0209] If obtained Then let the strong signal subspace deconvolve azimuth spectrum And continue with step three six;
[0210] If not received Then let Return to steps three and four;
[0211] Step 36: Deconvolve the azimuth spectrum of the strong signal subspace. The angles corresponding to each local spectral peak are taken as the null angles for forming the null constraint beam in the weak signal subspace, and the first... The angles corresponding to each local spectral peak are denoted as . , , Represents the azimuth spectrum of deconvolution in the strong signal subspace The total number of local spectral peaks;
[0212] Calculate the weight vector of the null-constrained beam and point spread function dictionary :
[0213]
[0214]
[0215] in, The scanning angle on the azimuth scanning grid The corresponding null-containment beamforming weight vector, express The conjugate transpose of . for The orthogonal projection matrix of the array manifold of the interference. yes The conjugate transpose of . This represents the array steering vector corresponding to the azimuth spectrum estimated by beamforming deconvolution. , express The conjugate transpose of . This indicates that null-constrained beamforming deconvolution is in A dictionary of point spread functions in terms of direction;
[0216] The orthogonal projection matrix of the array manifold of the interference is:
[0217]
[0218] in, express An array manifold matrix of interference, express The conjugate transpose of , where the superscript -1 denotes the inverse of the matrix. For unit array;
[0219]
[0220] in, Indicates angle The corresponding array steering vector, ;
[0221] Step 37, in Null-constrained beamforming is performed on the weak signal subspace components to obtain the weak signal subspace azimuth spectrum. :
[0222]
[0223] The extended RL algorithm is used to perform null-constrained beamforming deconvolution on the weak signal subspace azimuth spectrum, i.e., steps 38 to 30 are executed.
[0224] Step 38: Calculate intermediate variables , Indicates the scanning orientation The normalization function, Initialize the weak signal subspace source distribution function ;
[0225] Set the maximum number of iterations to (In this invention, the number of iterations is set to 500), and the number of iterations is initialized. ;
[0226] Step 39: Calculate the weak signal subspace source distribution function :
[0227]
[0228] in, Indicates the first The weak signal subspace source distribution function obtained in the next iteration This represents the normalized point spread function. , It is the normalized weak signal subspace orientation spectrum. ;
[0229] Step 30: Determine whether it has been obtained. ;
[0230] If obtained Then let the weak signal subspace deconvolve azimuth spectrum , , This indicates the azimuth estimation based on beamforming deconvolution. The normalization function;
[0231] If not received Then let Return to step 39.
[0232] Step 4: Fuse the deconvolution azimuth spectrum of the strong signal subspace and the deconvolution azimuth spectrum of the weak signal subspace to obtain the DOA estimation result;
[0233] Specifically:
[0234]
[0235] in, This is the fused azimuth spectrum. The signal-to-noise ratio (SNR) of each direction in the strong signal subspace component. The signal-to-noise ratio in each direction of the weak signal subspace component;
[0236] Obtain the azimuth spectrum The local spectral peaks are used to estimate the DOA of each target, and the angle corresponding to each local spectral peak is used as the DOA estimation result of each target.
[0237] The following is about and The calculation process is explained below:
[0238] Step 1 and They are respectively:
[0239]
[0240]
[0241] in, express background value, express Background value;
[0242] Step 2: Record the number of scanning angles on the azimuth grid as... The first scanning angle corresponds to the azimuth. , No. Each scanning angle corresponds to the azimuth. , will the Each scanning angle corresponds to an azimuth. Then, the strong signal subspace deconvolution azimuth spectrum is represented as a length of... vector The weak signal subspace deconvolution azimuth spectrum is represented as a spectrum of length . vector ;
[0243] Initialize the forward filtering results: , For have:
[0244]
[0245]
[0246] in, This is the filter coefficient (usually set to 0.1 to obtain a relatively stable background value). Indicates the azimuth corresponding to the first scanning angle. The corresponding value in the deconvolution azimuth spectrum of the strong signal subspace. Indicates the first Each scanning angle corresponds to the azimuth. The corresponding value in the deconvolution azimuth spectrum of the strong signal subspace;
[0247] from Start traversing to End, and the forward filtering result is obtained. and ;
[0248] Initialized backward filtering results: , For have:
[0249]
[0250]
[0251] from Start traversing to The process ends, and the back-filtering result is obtained. and ;
[0252] Then the first strong signal subspace Each scanning angle corresponds to the azimuth. Background values and weak signal subspace Each scanning angle corresponds to the azimuth. The background values are respectively , ,Will and Substituting the fused azimuth spectrum formula from step four, we can obtain the fused azimuth spectrum formed by each scanning angle.
[0253] In summary, the spectral clustering-based subspace decomposition method proposed in this invention does not require estimating the number of signal sources. It adaptively decomposes the covariance matrix into strong signal and weak signal subspaces based on the magnitude of eigenvalues, solving the array gain reduction problem of traditional deconvolution beamforming algorithms in strong interference scenarios and improving target detection capability. Furthermore, the DOA estimation method proposed in this invention introduces zero-concave-constrained deconvolution beamforming and a signal-to-noise ratio-based fusion algorithm. Under non-ideal conditions, it can effectively suppress the energy leaked into the weak signal subspace component by interference, effectively suppressing the impact of strong interference on weak targets and improving the robustness and target detection capability of the algorithm. This method can be applied to direction-of-arrival estimation in passive detection.
[0254] Simulation Experiment
[0255] Simulation conditions: The array used in the experiment is a linear array with 18 equally spaced half-wavelength uniformly distributed elements, with a spacing of 5m between adjacent elements and a sound speed in water of 1500m / s. Assume the presence of one target and one strong interference, with a signal-to-noise ratio of -5dB, an interference-to-noise ratio of 30dB, a target incident angle of 60°, and a strong interference incident angle of 120°. A non-ideal environment is introduced, considering that the element spacing error follows a Gaussian distribution with a mean of 0 and a standard deviation of 0.01 times the wavelength, and that the non-uniform noise power of each element follows a uniform distribution in the range [1,10]. The number of snapshots is set to 20, and the number of Monte Carlo simulations is set to 300.
[0256] Figure 2 The deconvolutional azimuth spectra of the strong signal subspace and the weak signal subspace are normalized according to the background. The black dashed lines represent the true incident azimuth of the target. In the strong signal subspace, error-free DOA estimation can be achieved despite interference, with an output signal-to-noise ratio of approximately 52 dB. In the weak signal subspace, the sine value of the DOA estimation for the target has a very small error of only 0.01, and its output signal-to-noise ratio is approximately 17 dB.
[0257] The deconvolutional azimuth spectra of the strong signal subspace and the deconvolutional azimuth spectra of the weak signal subspace are fused to output the final DOA estimation result. Figure 3 The results show a comparison of the azimuth spectrum between the algorithm of this invention and five other algorithms. In terms of target detection capability, the algorithm of this invention exhibits the highest output signal-to-noise ratio compared to the other algorithms, and produces almost no spurious peaks. Regarding DOA estimation performance, the local spectral peaks of the algorithm of this invention correspond to the azimuth closest to the true target azimuth, and the DOA estimation error is the smallest. This demonstrates that the algorithm of this invention possesses excellent target detection capability and DOA estimation performance under strong interference environments, and remains robust even under non-ideal conditions, verifying the feasibility of the proposed algorithm.
[0258] With a fixed signal-to-noise ratio of 0dB, Figure 4 The diagram illustrates the array gain curve of the algorithm of this invention as a function of interference-to-noise ratio (INR). The array gain of conventional beamforming deconvolution algorithms decreases as the target SNR decreases, while the array gain of the algorithm of this invention remains almost within the theoretical range. At an input SNR of -10dB, the array gain only decreases by about 0.9dB. With a fixed INR of 0dB, Figure 5The diagram illustrates the array gain curve of the algorithm of this invention as a function of the interference-to-noise ratio (IRR). The array gain of conventional beamforming deconvolution algorithms decreases with increasing IRR intensity because strong background interference overwhelms the target, degrading the performance of deconvolution beamforming. In this case, the dominant factor affecting the target array gain is the strong interference sidelobes. The algorithm of this invention effectively eliminates the influence of strong interference on the target through subspace clustering. Compared to conventional beamforming deconvolution algorithms, the method of this invention can achieve higher array gain and is more suitable for target DOA estimation scenarios under strong interference.
[0259] 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 decomposition deconvolution beamforming method based on subspace clustering, characterized in that, The method specifically includes the following steps: Step 1: Construct the data covariance matrix based on the received data from the uniform linear array; Step 2: Analyze the data covariance matrix. Subspace decomposition is performed to obtain strong signal subspace, weak signal subspace components, and noise subspace; Step 3: Obtain the weak signal subspace deconvolution azimuth spectrum and the strong signal subspace deconvolution azimuth spectrum respectively through beamforming; Step 4: Fuse the deconvolution azimuth spectrum of the strong signal subspace and the deconvolution azimuth spectrum of the weak signal subspace to obtain the DOA estimation result.
2. The decomposition deconvolution beamforming method based on subspace clustering according to claim 1, characterized in that, The elements of the uniform linear array are in The elements are evenly spaced on the positive half-axis of the axis, and the distance between adjacent elements is denoted as . .
3. The decomposition deconvolution beamforming method based on subspace clustering according to claim 2, characterized in that, The uniform linear array The data received at any given time is modeled as follows: in: Indicates by A uniform linear array composed of 1000 array elements Data received in real time; Indicates the first The incident azimuth of the target, , Indicates the number of incident targets; Indicates the first The signal vector of each target; Represents the noise vector; Represents a signal vector; express 3D array manifold matrix, , Indicates the first The array steering vector corresponding to the incident azimuth of each target; in: The base of the natural logarithm; Represents the imaginary unit; Indicates the signal wavelength; Represents transpose; use Construct the data covariance matrix from the array receiving data of the number of snapshots. : in: Represents the data covariance matrix; yes The conjugate transpose of . .
4. The decomposition deconvolution beamforming method based on subspace clustering according to claim 3, characterized in that, The specific process of step two is as follows: Step 2.1: Analyze the data covariance matrix. Perform eigenvalue decomposition: in: For the data covariance matrix The 1 eigenvalue, ; Eigenvalues The corresponding feature vector; yes The conjugate transpose of; Step 22: Obtain the results from Step 21 Sort the _th feature values in ascending order, and then select the _th ... The eigenvalues are denoted as ; Steps two and three: Preprocess each feature value in the sorting result separately: in, For the sorted results, the first eigenvalues The logarithmic representation of Indicates multiplication; Step 24: Construct the eigenvalue variance matrix : in: Represents the eigenvalue variance matrix The Middle Line number Column elements; Indicates the first in the sorting results The eigenvalue to the ... The average of the logarithmic representations of the eigenvalues; when hour, The value is 0; Step 25: Minimize the variance sum matrix The first line of the middle Column elements Minimum variance matrix The Middle Line number Column elements for: in, ; Let the breakpoint matrix The Middle Line number Column elements equals making The value of reaches the minimum. ,Right now: in, ; Step 26: Set the first breakpoint for: in, The breakpoint in the 3rd row of the matrix Column elements; Set the second breakpoint for: in, The second row of the breakpoint matrix Column elements; Step 27: Transfer the first eigenvalue to the second eigenvalue. The eigenvalue is taken as the first type of eigenvalue, and the eigenvalue is taken as the second type of eigenvalue. The eigenvalues up to the ... The eigenvalue is used as the second type of eigenvalue, and the eigenvalue is used as the second type of eigenvalue. The eigenvalues up to the ... These eigenvalues are used as the third type of eigenvalues; Weak signal subspace components are obtained based on the second type of eigenvalues. The strong signal subspace component is obtained based on the third type of feature value. .
5. The decomposition deconvolution beamforming method based on subspace clustering according to claim 4, characterized in that, The weak signal subspace component is obtained based on the second type of feature value. Specifically: in: Represents the weak signal subspace components; For the sorted results, the first eigenvalues The corresponding feature vector, yes The conjugate transpose of .
6. The decomposition deconvolution beamforming method based on subspace clustering according to claim 5, characterized in that, The strong signal subspace component is obtained based on the third type of feature value. Specifically: in, This represents the strong signal subspace component.
7. The decomposition deconvolution beamforming method based on subspace clustering according to claim 6, characterized in that, The specific process of step three is as follows: Step 31, in Beamforming is performed on the strong signal subspace components to obtain the strong signal subspace azimuth spectrum. ; in: The scanning angle on the azimuth scanning grid The corresponding beamforming weight vector, for The conjugate transpose of . , Indicates angle The corresponding array steering vector, ; Step 3.2: Set the maximum number of iterations to... ; Step 3: Initialize the strong signal subspace source distribution function and initialize the number of iterations. ; Steps three and four: Calculate the first... The strong signal subspace source distribution function of the next iteration : in: Indicates the first The strong signal subspace source distribution function of the next iteration; Indicates the azimuth of the beamforming deconvolution estimation; Indicates beamforming deconvolution in A dictionary of point spread functions in terms of direction; Step 35: Determine whether it has been obtained. ; If obtained Then let the strong signal subspace deconvolve azimuth spectrum And continue with step three six; If not received Then let Return to steps three and four; Step 36: Deconvolve the azimuth spectrum of the strong signal subspace. The angles corresponding to each local spectral peak are taken as the null angles for forming the null constraint beam in the weak signal subspace, and the first... The angles corresponding to each local spectral peak are denoted as . , , Represents the azimuth spectrum of deconvolution in the strong signal subspace The total number of local spectral peaks; Calculate the weight vector of the null-constrained beam and point spread function dictionary : in: The scanning angle on the azimuth scanning grid The corresponding null-containment beamforming weight vector, yes The conjugate transpose of; This indicates that null-constrained beamforming deconvolution is in A dictionary of point spread functions in terms of direction; express The conjugate transpose of; The azimuth spectrum represents the beamforming deconvolution estimation. The corresponding array steering vector; in: express The conjugate transpose of; for The orthogonal projection matrix of the array manifold of the interference: in: express An array manifold matrix of interference, express The conjugate transpose of , where the superscript -1 denotes the inverse of the matrix. For unit array; in: Indicates angle The corresponding array steering vector, ; Step 37, in Null-constrained beamforming is performed on the weak signal subspace components to obtain the weak signal subspace azimuth spectrum. : Step 38: Calculate the normalized weak signal subspace orientation spectrum , Indicates the scanning orientation The normalization function, Initialize the weak signal subspace source distribution function ; Set the maximum number of iterations to and initialize the number of iterations. ; Step 39: Calculate the weak signal subspace source distribution function : in: Indicates the first The weak signal subspace source distribution function obtained in the next iteration; This represents the normalized point spread function. ; Step 30: Determine whether it has been obtained. ; If obtained Then let the weak signal subspace deconvolve azimuth spectrum , , This indicates the azimuth estimation based on beamforming deconvolution. The normalization function; If not received Then let Return to step 39.
8. The decomposition deconvolution beamforming method based on subspace clustering according to claim 7, characterized in that, The beamforming deconvolution is Dictionary of point spread functions in directions for: in, This represents the Sink function.
9. The decomposition deconvolution beamforming method based on subspace clustering according to claim 8, characterized in that, The specific process of step four is as follows: in: The azimuth spectrum after fusion; The signal-to-noise ratio in each direction of the strong signal subspace component; The signal-to-noise ratio in each direction of the weak signal subspace component; Obtain the azimuth spectrum The local spectral peaks are used to estimate the DOA of each target, and the angle corresponding to each local spectral peak is used as the DOA estimation result of each target.
10. The decomposition deconvolution beamforming method based on subspace clustering according to claim 9, characterized in that, Signal-to-noise ratio in each direction of the strong signal subspace component Signal-to-noise ratio in each direction of the weak signal subspace component They are respectively: in, express background value, express Background value; The method for calculating the background value is as follows: Step 1: Record the number of scanning angles on the azimuth grid as... , will the The azimuth corresponding to each scanning angle is denoted as . Then, the strong signal subspace deconvolution azimuth spectrum is represented as a length of... vector The weak signal subspace deconvolution azimuth spectrum is represented as a spectrum of length . vector ; Step 2: Initialize the forward filtering results: , For have: in: These are the filter coefficients. Indicates the azimuth corresponding to the first scanning angle. The corresponding value in the deconvolution azimuth spectrum of the strong signal subspace. Indicates the first The azimuth corresponding to each scanning angle The corresponding value in the deconvolution azimuth spectrum of the strong signal subspace; from Start traversing to End, and the forward filtering result is obtained. and ; Step 3: Initialize the backward filtering results: , For have: from Start traversing to The process ends, and the back-filtering result is obtained. and ; Then the first strong signal subspace Each scanning angle corresponds to the azimuth. Background values and weak signal subspace Each scanning angle corresponds to the azimuth. The background values are respectively , .