Sparse bayesian ionospheric clutter stap method based on local continuous frequency domain dimension reduction

Abstract

The sparse Bayesian ionospheric clutter STAP method based on local continuous frequency domain dimension reduction relates to the technical field of radar signal processing, and aims at the problem of poor ionospheric clutter suppression effect of the existing ground wave radar. The application is based on local continuous frequency domain dimension reduction and sparse Bayesian reconstruction technology, and is used for effectively suppressing ionospheric clutter. By constructing a high-frequency ground wave radar signal model, using dimension reduction processing of local frequency domain, combining a sparse Bayesian algorithm, efficient estimation of the ionospheric clutter space-time spectrum is realized, and the covariance matrix of the clutter is reconstructed, and finally the ionospheric clutter suppression is realized by using the STAP method. The suppression effect of the technical scheme of the application is remarkable, and the signal-to-noise ratio (SCNR) of the application is improved by 3.09 dB compared with the SBL method, and the clutter suppression performance is significantly enhanced.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of radar signal processing, in particular to a sparse Bayesian ionospheric clutter STAP method based on local continuous frequency domain dimension reduction. BACKGROUND

[0002] Ground wave radar is widely used in ocean monitoring, weather detection and other fields due to its unique propagation path and excellent coverage capability. However, ground wave radar is often disturbed by ionospheric clutter in practical application, resulting in the decline of target detection performance. The ionospheric clutter has complex angle and Doppler characteristics, and its suppression is difficult.

[0003] In 2010, Wang used space-time adaptive processing (STAP) method and dimension reduction STAP method to suppress ionospheric clutter by estimating the clutter covariance matrix and designing filter, but in the case of complex and changeable ionospheric clutter, the traditional method often has the problem of poor ionospheric clutter suppression effect of ground wave radar. SUMMARY

[0004] The purpose of the present application is to solve the problem of poor ionospheric clutter suppression effect of ground wave radar in the prior art, and to provide a sparse Bayesian ionospheric clutter STAP method based on local continuous frequency domain dimension reduction.

[0005] The technical scheme adopted by the present application to solve the above technical problem is:

[0006] The sparse Bayesian ionospheric clutter STAP method based on local continuous frequency domain dimension reduction comprises the following steps:

[0007] Step 1: Obtain local frequency range, local spatial frequency range, time domain window function and space domain window function to construct local continuous frequency domain feature matrix and local continuous spatial frequency domain feature matrix;

[0008] Step 2: Perform eigenvalue decomposition on the local continuous frequency domain feature matrix and the local continuous spatial frequency domain feature matrix respectively to obtain time domain dimension reduction matrix and space domain dimension reduction matrix;

[0009] Step 3: Obtain an attenuation threshold to construct a local range time domain observation matrix and a space domain observation matrix, and use the time domain dimension reduction matrix and the space domain dimension reduction matrix to reduce the dimension of the local range time domain observation matrix and the space domain observation matrix, and finally obtain the Kronecker product of the reduced local range time domain observation matrix and the space domain observation matrix, that is, the space-time observation matrix;

[0010] Step 4: Obtain radar echo data and pre-process the radar echo data, the pre-processing including matching filtering and array calibration processing;

[0011] Step five: based on the pre-processed radar echo data, all distance units of the radar echo data are obtained, and the time domain dimension reduction matrix and the space domain dimension reduction matrix are used to perform dimension reduction processing on all distance units of the radar echo data;

[0012] Step six: the Doppler frequency of the expected target is obtained , so as to construct the time domain steering vector of the expected target , and the time domain dimension reduction matrix is used to perform dimension reduction processing on the time domain steering vector of the expected target , to obtain the dimension-reduced time domain steering vector;

[0013] Step seven: the azimuth angle of the expected target is obtained , so as to construct the space domain steering vector of the expected target , and the space domain dimension reduction matrix is used to perform dimension reduction processing on the space domain steering vector of the expected target , to obtain the dimension-reduced space domain steering vector;

[0014] Step eight: the Kronecker product of the dimension-reduced time domain steering vector and the dimension-reduced space domain steering vector, i.e. the dimension-reduced space-time steering vector is obtained;

[0015] Step nine: in all distance units of the dimension-reduced radar echo data, the Ith distance unit is randomly selected as a detection unit, and the I-2th and I+2th distance units are selected as training samples;

[0016] Step ten: the space-time observation matrix is used, and the sparse Bayesian learning method is used to perform space-time estimation on the two training samples, to obtain the clutter space-time spectrum;

[0017] Step eleven: the clutter space-time spectrum is used to reconstruct the covariance matrix of the clutter, and the reconstructed covariance matrix of the clutter and the dimension-reduced space-time steering vector are used to construct the filter weight;

[0018] Step twelve: the filter weight is used to perform filtering processing on the detection unit;

[0019] Step thirteen: steps nine to twelve are repeated until all dimension-reduced echo data in the distance unit are processed, to obtain the echo data after removing the ionospheric clutter.

[0020] Further, the local continuous spatial frequency domain feature matrix is represented as:

[0021]

[0022] Wherein, N represents the number of array elements, d represents the array element spacing, represents the sampling frequency, denotes a spatial domain window function, and denotes a local spatial frequency from to ;

[0023] The local continuous frequency domain feature matrix is denoted as:

[0024]

[0025] wherein M denotes the number of signal pulses, denotes a time domain window function, and denotes a local frequency from to .

[0026] Further, the time domain steering vector of the desired target is denoted as:

[0027]

[0028] wherein is a pulse repetition frequency;

[0029] The spatial domain steering vector of the desired target is denoted as:

[0030]

[0031] wherein is a signal wavelength;

[0032] The reduced dimension space-time steering vector is denoted as:

[0033]

[0034] wherein denotes a spatial domain reduction matrix, denotes a time domain reduction matrix.

[0035] Further, the specific steps of the step two include:

[0036] Step two one: performing eigenvalue decomposition on the local continuous frequency domain feature matrix and the local continuous spatial frequency domain feature matrix respectively, denoted as:

[0037]

[0038]

[0039] wherein and respectively represent the local continuous frequency domain feature matrix and the local continuous spatial frequency domain feature matrix, represents the matrix composed of spatial domain feature vectors, represents the matrix composed of time domain feature vectors, represents the diagonal matrix composed of spatial domain feature values, represents the diagonal matrix composed of time domain feature values.

[0040] Step two: let and respectively represent the first columns and the remaining columns of , and respectively represent the first columns and the remaining columns of , so as to obtain the time domain dimension reduction matrix and the spatial domain dimension reduction matrix, denoted as:

[0041]

[0042] .

[0043] Further, the local range spatial domain observation matrix is denoted as:

[0044]

[0045] wherein, represents the local range spatial domain observation matrix.

[0046] Further, the local range time domain observation matrix is denoted as:

[0047]

[0048] wherein, represents the local range time domain observation matrix.

[0049] Further, the space-time observation matrix is denoted as:

[0050]

[0051]

[0052]

[0053] wherein, and respectively represent the local range time domain observation matrix and the local range spatial domain observation matrix after dimension reduction.

[0054] ​​Further, the training sample is represented as:

[0055]

[0056]

[0057] wherein, represents echo data of the Ith range cell, represents dimension-reduced echo data of the Ith range cell as a detection unit, and respectively represent dimension-reduced echo data of the I-2th and I+2th range cells, as the training sample.

[0058] Further, the space-time estimation using the sparse Bayesian learning method is represented as:

[0059] Initialize hyperparameters , ,

[0060] wherein, represents the i-th column of D, represents the i-th column of , is the column number of the matrix , represents the Frobenius norm of a matrix;

[0061] The steps of each iteration of the sparse Bayesian learning method are:

[0062]

[0063]

[0064]

[0065]

[0066]

[0067]

[0068]

[0069] wherein, the matrix I is an identity matrix, represents a diagonal matrix, represents a clutter space-time spectrum, represents a Hadamard product, represents a constant matrix , wherein, n is the number of rows of matrix D, represents the element of Y at the i-th row and j-th column, represents the data of Y at the j-th column, , and represents an intermediate variable, represents the number of training samples;

[0070] When the convergence condition is reached, the iteration stops, and Y obtains the optimal estimate, i.e., the clutter space-time spectrum;

[0071] The convergence condition is:

[0072] (1) the maximum number of iterations is reached

[0073] (2) , is a positive number, = 0.001.

[0074] Further, the detection unit after the filtering processing in step twelve is represented as:

[0075]

[0076]

[0077]

[0078] wherein, represents the covariance matrix of the clutter, represents the filter weight.

[0079] The application has the following beneficial effects:

[0080] The application is based on local continuous frequency domain dimension reduction and sparse Bayesian reconstruction technology, and is used for effectively suppressing ionospheric clutter. By constructing a high-frequency ground wave radar signal model, using dimension reduction processing in the local frequency domain, combining the sparse Bayesian algorithm, efficient estimation of the ionospheric clutter space-time spectrum is realized, the covariance matrix of the clutter is reconstructed, and finally the STAP method is used to realize ionospheric clutter suppression. The suppression effect of the technical solution of the application is remarkable, and the signal-to-noise ratio (SCNR) of the application is improved by 3.09 dB compared with the SBL method, and the clutter suppression performance is significantly enhanced.

[0081] The application realizes dimension reduction processing in the local frequency domain, and the average running time of a single distance gate is only 0.31% of the SBL method, which greatly improves the real-time processing capability of the algorithm and is suitable for real-time application in actual ground wave radar systems. BRIEF DESCRIPTION OF DRAWINGS

[0082] Figure 1 is an offline process flowchart of the application;​

[0083] Figure 2 For the online process flowchart of the present application;

[0084] Figure 3 For the inhibition of the output comparison chart. DETAILED DESCRIPTION

[0085] It should be particularly noted that the various embodiments disclosed in the present application can be combined with each other without conflict.

[0086] Embodiment one: the sparse Bayesian ionospheric clutter STAP method based on local continuous frequency domain dimension reduction in the present embodiment includes offline steps and online steps.

[0087] The offline steps are as follows, and the flowchart is shown in Figure 1 .

[0088] Step one, according to the preset local frequency range, local spatial frequency range and window function parameter, construct the local continuous frequency domain feature matrix and the local continuous spatial frequency domain feature matrix;

[0089] The HFSWR system parameters are assumed as follows: the uniform linear array is composed of N array elements, the array element spacing is d, the phase coherent accumulation period contains M signal pulses, and fs is the sampling frequency. Therefore, for the Ith range unit, the data vector of dimension can be obtained as follows:

[0090] (1)

[0091] The space-time steering vector of the target echo is defined as follows:

[0092] (2)

[0093] Wherein, , are the spatial steering vector and the time steering vector respectively, is the Kronecker product, λ is the signal wavelength, is the azimuth of the target echo, is the Doppler frequency of the target echo, is the pulse repetition frequency.

[0094] The received data can be represented as the sum of the target signal, clutter and noise:

[0095] (3)

[0096] Wherein is the amplitude of the target, c is the ionospheric clutter, and n is the noise.​

[0097] Since most ionospheric scintillation is strongly correlated in local angle and Doppler space, the range of angle-Doppler region of selected samples needs to be determined first, and the corresponding features are extracted.

[0098] The observation matrix of the spatial domain steering matrix is:

[0099] (4)

[0100] where N is the number of antennas, is the starting frequency, p is the sparsity, is the angle difference between adjacent atoms.

[0101] Let , rewrite the frequency angle as :

[0102] (5)

[0103] where, is the frequency difference between adjacent atoms.

[0104] When , the continuous spatial frequency domain is:

[0105] (6)

[0106] Take the local spatial frequency from to , use the window function , and the local continuous spatial frequency domain feature matrix can be expressed as:

[0107] (7)

[0108] Since :

[0109] (8)

[0110] Similarly, take the local frequency from to , use the window function , and the local continuous frequency domain feature matrix can be expressed as:

[0111] (9)

[0112] j is the imaginary unit in mathematics;

[0113] Step two, perform eigenvalue decomposition on the local continuous frequency domain feature matrix and the local continuous spatial frequency domain feature matrix respectively, and construct the time domain dimension reduction matrix and the spatial domain dimension reduction matrix.

[0114] characteristic value decomposition is performed on the matrix , characteristic value decomposition is performed on the matrix

[0115] (10)

[0116] Let , where and are the first columns and the remaining columns of , respectively, and are the first columns and the remaining columns of , respectively, obtain the spatial dimension reduction matrix , and the time domain dimension reduction matrix :

[0117] (11)

[0118] Step three, according to the preset attenuation threshold, construct the local range time domain observation matrix and the local range spatial observation matrix, and reduce the dimension of the local range time domain observation matrix and the local range spatial observation matrix.

[0119] Obtain the local range spatial observation matrix and the local range time domain observation matrix:

[0120] (12)

[0121] (13)

[0122] wherein .

[0123] Reduce the dimension of the observation matrix:

[0124] (14)

[0125] (15)

[0126] Then the space-time observation matrix is:

[0127] (16)

[0128] In the step, the flow chart is as shown in Figure 2 :

[0129] Step one, obtain the radar echo data, and after matching filtering and array calibration of the radar echo data; using the time domain dimension reduction matrix and the spatial dimension reduction matrix to reduce the dimension of the processed radar echo data, and obtain the reduced data;

[0130] Dimension reduction of radar echo data:

[0131] (17)

[0132] When the test data of radar echo data is the Ith range cell, the I-2th and I+2th range cells are selected as training data:

[0133] (18)

[0134] Step two, using the space-time observation matrix, the training data after dimension reduction is used for space-time estimation by sparse Bayesian learning method, and the clutter space-time spectrum Y is obtained;

[0135] The optimal estimation of each iteration of the sparse Bayesian learning method is:

[0136] (19)

[0137] (20)

[0138] (21)

[0139] (22)

[0140] (23)

[0141] (24)

[0142] wherein, represents Hadamard product, .

[0143] The algorithm of sparse Bayesian is an iterative process, and the hyperparameters are initialized , , .

[0144] Convergence condition of iteration:

[0145] (1) The maximum number of iterations maxiter is reached.

[0146] (2) , is a very small positive number, usually set to 0.001.

[0147] Step three, reconstruct the covariance matrix of clutter by clutter space-time spectrum; construct the filter weight by the covariance matrix of clutter and the space-time steering vector of expected target; filter the reduced dimension test data using the filter weight to remove ionospheric clutter.

[0148] Specific method:

[0149] Reconstruct the covariance matrix of clutter by clutter space-time spectrum:

[0150] (25)

[0151] Solve the filter weight:

[0152] (26)

[0153] Filter the reduced dimension test data to obtain the final output:

[0154] (27)

[0155] Embodiment:

[0156] Adopt Uniform linear array, the ground wave radar data of phase coherent accumulation period containing Signal pulses, the array element spacing is d, and the radar operating frequency is fs. By injecting a simulated target into the ionospheric clutter range, relevant parameters are set.

[0157] Offline steps:

[0158] Step one: according to the preset local frequency range, local spatial frequency range, window function type parameter, construct local continuous frequency domain feature matrix and local continuous spatial frequency domain feature matrix;

[0159] Take the local spatial frequency from To , use the rectangular window function , the local continuous spatial frequency domain feature matrix can be expressed as:

[0160] (28)

[0161] Similarly, take the local frequency from To , use the Hanning window function , the local continuous frequency domain feature matrix can be expressed as:

[0162] (29)

[0163] Step two, eigenvalue decomposition is performed on the local continuous frequency domain feature matrix and the local continuous spatial frequency domain feature matrix, time domain dimension reduction matrix and spatial domain dimension reduction matrix are constructed, and the final space-time observation matrix is obtained;

[0164] Eigenvalue decomposition is performed on , Eigenvalue decomposition is performed on

[0165] (30)

[0166] Let , wherein and are the first columns and the remaining columns of , respectively, and are the first columns and the remaining columns of , respectively, the spatial domain dimension reduction matrix , and the time domain dimension reduction matrix are obtained:

[0167] (31)

[0168] Step three, according to the preset attenuation threshold, the local range time domain observation matrix and the local range spatial domain observation matrix are constructed, and the local range time domain observation matrix and the local range spatial domain observation matrix are dimension reduced.

[0169] Taking the local spatial frequency from to , using the rectangular window function , the local range spatial domain observation matrix can be expressed as:

[0170] (32)

[0171] Taking the local frequency from to , using the Hanning window function , the local range time domain observation matrix can be expressed as:

[0172] (33)

[0173] Dimension reduction is performed on the observation matrix:

[0174] (34)

[0175] (35)

[0176] Then the space-time observation matrix is:

[0177] (36)

[0178] The online steps are:

[0179] Step one, obtain the radar echo data, and after matching filtering and array calibration of the radar echo data; use the time domain dimension reduction matrix and the space domain dimension reduction matrix to perform dimension reduction processing on the radar echo data, to obtain the dimension-reduced data;

[0180] Dimension reduction is performed on the radar echo data:

[0181] (37)

[0182] When the test data is the first distance unit, the first and the first distance units are selected as training data:

[0183] (38)

[0184] Step two, use the space-time observation matrix to perform space-time estimation on the dimension-reduced training data using the sparse Bayesian learning method, to obtain the clutter space-time spectrum;

[0185] The optimal estimate of the sparse Bayesian learning method in each iteration is:

[0186] (39)

[0187] (40)

[0188] (41)

[0189] (42)

[0190] (43)

[0191] (44)

[0192] wherein, represents the Hadamard product, .

[0193] The algorithm of sparse Bayesian is an iterative process, and the hyperparameters , , are initialized.

[0194] The convergence condition of iteration is:

[0195] (1) The maximum number of iterations set is 100.

[0196] (2) , is a very small positive number, set to 0.001.

[0197] Step three, reconstruct the covariance matrix of the clutter using the clutter space-time spectrum; construct the filter weight using the covariance matrix of the clutter and the space-time steering vector of the expected target; filter the reduced test data using the filter to remove the ionospheric clutter.

[0198] Specific method:

[0199] Reconstruct the covariance matrix of the clutter using the clutter space-time spectrum:

[0200] (45)

[0201] Solve the filter weight:

[0202] (46)

[0203] Filter the reduced test data to obtain the final output:

[0204] (47)

[0205] Finally, compare the present application with the SBL method. Figure 3 Table 1 and Table 2 respectively show the comparison chart of the output after suppression of different algorithms and the comparison chart of the SCNR performance of the output after suppression of different algorithms. Table 2 shows the comparison chart of the average single distance gate running time of different algorithms. The results show that the suppression performance of the present application is better than that of the SBL method, and the actual running time can be greatly reduced.

[0206] Table 1 Comparison chart of SCNR performance of output after suppression of the present application

[0207]

[0208] Table 2 Comparison chart of actual running time of the present application

[0209]

[0210] The present application realizes efficient suppression and real-time processing of ionospheric clutter through effective dimension reduction processing and sparse reconstruction technology, significantly improves the target detection performance and computing efficiency of the ground wave radar system, and has wide application prospects.

[0211] It should be noted that the embodiments are only intended to explain and illustrate the technical solutions of the present application, and cannot be used to limit the protection scope. Any slight change made according to the claims and the specification of the present application shall still fall within the protection scope of the present application.

Claims

1. A sparse Bayesian STAP method for ionospheric clutter based on local continuous frequency domain dimensionality reduction, characterized in that... The method includes the following steps: Step 1: Obtain the local frequency range, local spatial frequency range, and time-domain window function. and spatial window function In this way, a local continuous frequency domain feature matrix and a local continuous spatial frequency domain feature matrix are constructed; Step 2: Perform eigenvalue decomposition on the local continuous frequency domain feature matrix and the local continuous spatial frequency domain feature matrix respectively to obtain the time-domain dimensionality reduction matrix and the spatial-domain dimensionality reduction matrix; Step 3: Obtain the attenuation threshold to construct the local temporal and spatial observation matrices. Then, use the temporal and spatial dimensionality reduction matrices to reduce the dimensionality of the local temporal and spatial observation matrices. Finally, obtain the Kronecker product of the dimensionality-reduced local temporal and spatial observation matrices, which is the spatiotemporal observation matrix. Step 4: Acquire radar echo data and preprocess the radar echo data, including matched filtering and array calibration processing; Step 5: Based on the preprocessed radar echo data, obtain all range cells of the radar echo data, and use time-domain dimensionality reduction matrix and spatial-domain dimensionality reduction matrix to perform dimensionality reduction processing on all range cells of the radar echo data; Step Six: Obtain the Doppler frequency of the desired target This is used to construct the temporal steering vector of the desired target. And use the time-domain dimensionality reduction matrix to the time-domain steering vector of the desired target. Dimensionality reduction is performed to obtain the dimensionality-reduced time-domain steering vector; Step 7: Obtain the azimuth of the desired target. This is used to construct the spatial guidance vector of the desired target. And use a spatial dimension reduction matrix to adjust the spatial steering vector of the desired target. Dimensionality reduction is performed to obtain the dimensionality-reduced spatial steering vector; Step 8: Obtain the Kronecker product of the dimension-reduced temporal steering vector and the dimension-reduced spatial steering vector, i.e., the dimension-reduced spatiotemporal steering vector. ; Step 9: In all the range cells of the reduced-dimensional radar echo data, randomly select the I-th range cell as the detection cell, and select the I-2 and I+2 range cells as training samples; Step 10: Using the spatiotemporal observation matrix and the sparse Bayesian learning method, perform spatiotemporal estimation on the two training samples to obtain the clutter spatiotemporal spectrum; Step 11: Reconstruct the covariance matrix of the clutter using its spatiotemporal spectrum, and then use the reconstructed covariance matrix and the dimension-reduced spatiotemporal steering vector... Construct the filter weights; Step 12: Filter the detection unit using filter weights; Step 13: Repeat steps 9 to 12 until all the dimension-reduced echo data in the distance cell has been processed to obtain echo data with ionospheric clutter removed.

2. The sparse Bayesian ionospheric clutter STAP method based on local continuous frequency domain dimensionality reduction according to claim 1, characterized in that... The local continuous spatial frequency domain feature matrix is ​​represented as follows: Where N represents the number of array elements, and d represents the spacing between array elements. Indicates the sampling frequency. Represents the spatial window function. and Indicates local spatial frequency from arrive ; The local continuous frequency domain feature matrix is ​​represented as follows: Where M represents the number of signal pulses, Represents the time-domain window function. and Indicates local frequency from arrive .

3. The sparse Bayesian ionospheric clutter STAP method based on local continuous frequency domain dimensionality reduction according to claim 2, characterized in that... The temporal steering vector of the desired target Represented as: in, The pulse repetition frequency; The spatial steering vector of the desired target Represented as: in, The wavelength of the signal; The reduced spacetime steering vector Represented as: in, Represents a spatial dimension reduction matrix. This represents the time-domain dimensionality reduction matrix.

4. The sparse Bayesian ionospheric clutter STAP method based on local continuous frequency domain dimensionality reduction according to claim 3, characterized in that... The specific steps of step two include: Step 21: Perform eigenvalue decomposition on the local continuous frequency domain characteristic matrix and the local continuous spatial frequency domain characteristic matrix, respectively, as follows: in, and These represent the characteristic matrix in the local continuous frequency domain and the characteristic matrix in the local continuous spatial frequency domain, respectively. This represents a matrix composed of spatial eigenvectors. This represents a matrix composed of time-domain eigenvectors. This represents a diagonal matrix composed of spatial eigenvalues. This represents a diagonal matrix composed of time-domain eigenvalues; Step 22: Let ,in and They are respectively The former Columns and Remaining List, and They are respectively The former Columns and Remaining Columns are used to obtain the time-domain and spatial-domain dimensionality-reduced matrices, which are expressed as: 。 5. The sparse Bayesian ionospheric clutter STAP method based on local continuous frequency domain dimensionality reduction according to claim 4, characterized in that... The spatial observation matrix of the local area is represented as: in, This represents the spatial observation matrix for a local area.

6. The sparse Bayesian ionospheric clutter STAP method based on local continuous frequency domain dimensionality reduction according to claim 5, characterized in that... The time-domain observation matrix of the local area is represented as: in, This represents the time-domain observation matrix within a local area.

7. The sparse Bayesian ionospheric clutter STAP method based on local continuous frequency domain dimensionality reduction according to claim 6, characterized in that... The spatiotemporal observation matrix is ​​represented as follows: in, and These represent the time-domain observation matrix and the spatial-domain observation matrix of the local area after dimensionality reduction, respectively.

8. The sparse Bayesian ionospheric clutter STAP method based on local continuous frequency domain dimensionality reduction according to claim 7, characterized in that... The training samples are represented as follows: in, This represents the echo data of the I-th range cell. This represents the echo data after dimensionality reduction of the I-th distance unit, used as the detection unit. and These represent the echo data after dimensionality reduction for the (I-2)th and (I+2)th distance cells, respectively. As training samples.

9. The sparse Bayesian ionospheric clutter STAP method based on local continuous frequency domain dimensionality reduction according to claim 8, characterized in that... The spatiotemporal estimation using the sparse Bayesian learning method is expressed as follows: Initialize hyperparameters , , in, Represents the i-th column of D. express The i-th column, For matrix The number of columns, This indicates the calculation of the Frobenius norm of a matrix; The steps for each iteration of the sparse Bayesian learning method are as follows: Where matrix I is the identity matrix. Indicates a diagonal matrix. Represents the space-time spectrum of clutter. Representing Hadamaji, Represents a constant matrix , Let D be the row number of matrix D. Let Y represent the element in the i-th row and j-th column. This represents the data in the j-th column of Y. , and Indicates intermediate variables. Indicates the number of training samples; When the convergence condition is met, the iteration stops, and Y is obtained as the optimal estimate, i.e., the clutter space-time spectrum; The convergence condition is: (1) Reaching the maximum number of iterations (2) , It is a positive number. =0.

001.

10. The sparse Bayesian ionospheric clutter STAP method based on local continuous frequency domain dimensionality reduction according to claim 9, characterized in that... The detection unit after filtering in step twelve Represented as: in, The covariance matrix of clutter is represented. This represents the filter weights.

Images