Sparse Bayesian ionized layer clutter STAP method based on local continuous frequency domain dimensionality reduction

Abstract

The invention discloses a sparse Bayesian ionized layer clutter STAP method based on local continuous frequency domain dimensionality reduction, relates to the technical field of radar signal processing, and aims to solve the problem of poor ground wave radar ionized layer clutter suppression effect in the prior art, and the sparse Bayesian ionized layer clutter STAP method based on local continuous frequency domain dimensionality reduction and sparse Bayesian reconstruction technology is used for effectively suppressing ionized layer clutter. By constructing a high-frequency ground wave radar signal model, utilizing dimension reduction processing of a local frequency domain and combining a sparse Bayesian algorithm, efficient estimation of an ionized layer clutter space-time spectrum is achieved, a covariance matrix of clutters is reconstructed, and finally an STAP method is used for achieving ionized layer clutter suppression. According to the technical scheme, the suppression effect is remarkable, the signal-to-noise ratio (SCNR) of the method is improved by 3.09 dB compared with an SBL method, and clutter suppression performance is remarkably enhanced.

Description

Technical Field

[0001] The present invention relates to the technical field of radar signal processing, and specifically to a sparse Bayesian ionospheric clutter STAP method based on local continuous frequency domain dimension reduction. Background Art

[0002] Ground wave radar is widely used in fields such as marine monitoring and meteorological detection due to its unique propagation path and excellent coverage ability. However, ground wave radar is often interfered by ionospheric clutter in practical applications, resulting in a decline in target detection performance. Ionospheric clutter has complex angular and Doppler characteristics, and it is difficult to suppress.

[0003] In 2010, Wang used the space-time adaptive processing (STAP) method and the dimension reduction STAP method to suppress ionospheric clutter by estimating the clutter covariance matrix and designing filters. However, in the case of complex and variable ionospheric clutter, traditional methods often have the problem of poor suppression effect of ionospheric clutter in ground wave radar. Summary of the Invention

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

[0005] The technical solution adopted by the present invention to solve the above technical problems is:

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

[0007] Step 1: Obtain a local frequency range, a local spatial frequency range, a time domain window function W t and a spatial domain window function W s , and construct a local continuous frequency domain feature matrix and a local continuous spatial frequency domain feature matrix therefrom;

[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 a time domain dimension reduction matrix and a spatial domain dimension reduction matrix;

[0009] Step 3: Obtain an attenuation threshold, construct a time domain observation matrix and a spatial domain observation matrix in a local range therewith, and use the time domain dimension reduction matrix and the spatial domain dimension reduction matrix to perform dimension reduction on the time domain observation matrix and the spatial domain observation matrix in the local range. Finally, obtain the Kronecker product of the time domain observation matrix and the spatial domain observation matrix in the dimension-reduced local range, that is, the space-time observation matrix;

[0010] Step 4: Obtain radar echo data, and perform preprocessing on the radar echo data, the preprocessing including matched filtering and array calibration processing;

[0011] Step 5: Based on the preprocessed radar echo data, obtain all range cells of the radar echo data, and perform dimensionality reduction processing on all range cells of the radar echo data using the time-domain dimensionality reduction matrix and the space-domain dimensionality reduction matrix;

[0012] Step 6: Obtain the Doppler frequency f of the desired target t , and construct the time-domain steering vector b(f t ) of the desired target accordingly, and perform dimensionality reduction processing on the time-domain steering vector b(f t ) of the desired target using the time-domain dimensionality reduction matrix to obtain the dimensionality-reduced time-domain steering vector;

[0013] Step 7: Obtain the azimuth angle φ of the desired target t , and construct the space-domain steering vector a(φ t ) of the desired target accordingly, and perform dimensionality reduction processing on the space-domain steering vector a(φ t ) of the desired target using the space-domain dimensionality reduction matrix to obtain the dimensionality-reduced space-domain steering vector;

[0014] Step 8: Obtain the Kronecker product of the dimensionality-reduced time-domain steering vector and the dimensionality-reduced space-domain steering vector, that is, the dimensionality-reduced spatio-temporal steering vector

[0015] Step 9: Randomly select the I-th range cell as the detection cell among all range cells of the dimensionality-reduced radar echo data, and select the (I - 2)-th and (I + 2)-th range cells as training samples;

[0016] Step 10: Use the spatio-temporal observation matrix and perform spatio-temporal estimation on the two training samples using the sparse Bayesian learning method to obtain the clutter spatio-temporal spectrum;

[0017] Step 11: Reconstruct the covariance matrix of the clutter using the clutter spatio-temporal spectrum, and construct the filter weights using the reconstructed covariance matrix of the clutter and the dimensionality-reduced spatio-temporal steering vector ;

[0018] Step 12: Perform filtering processing on the detection cell using the filter weights;

[0019] Step 13: Repeat Steps 9 to 12 until all dimensionality-reduced echo data in the range cells are processed to obtain the echo data with ionospheric clutter removed.

[0020] Furthermore, the local continuous spatial frequency domain feature matrix is expressed as:

[0021]

[0022] where N represents the number of array elements, d represents the array element spacing, fs Denotes the sampling frequency, W s Denotes the spatial domain window function, f s1 And f s2 Denotes that the local spatial frequency ranges from f s1 To f s2 ;

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

[0024]

[0025] Where M denotes the number of signal pulses, W t Denotes the time domain window function, f t1 And f s2 Denotes that the local frequency ranges from f t1 To f s2 .

[0026] Furthermore, the time domain steering vector b(f t ) of the desired target is expressed as:

[0027]

[0028] Where f R Is the pulse repetition frequency;

[0029] The spatial domain steering vector a(φ t ) of the desired target is expressed as:

[0030]

[0031] Where λ is the signal wavelength;

[0032] The reduced-dimensional space-time steering vector Is expressed as:

[0033]

[0034] Where T s Denotes the spatial domain reduction matrix, T t Denotes the time domain reduction matrix.

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

[0036] Step 2-1: Perform eigenvalue decomposition on the local continuous frequency domain feature matrix and the local continuous spatial frequency domain feature matrix respectively, expressed as:

[0037] H s =U s S s U s H

[0038] H t =U t S t U t H

[0039] Among them, H s and H t They represent the local continuous frequency domain feature matrix and the local continuous spatial frequency domain feature matrix respectively, U s Represents the matrix composed of spatial eigenvectors, U t represents the matrix composed of time domain eigenvectors, S s represents the diagonal matrix composed of spatial eigenvalues, S t Represents a diagonal matrix composed of time domain eigenvalues;

[0040] Step 22: Let U s =[U s1 ,U s2 ],U t =[U t1 ,U t2 ], where U s1 and U s2 U s The first k s Column and remaining Nk s Column, U t1 and U t2 U t The first k t Column and the remaining Mk t Columns, thus obtaining the time domain dimensionality reduction matrix and the space domain dimensionality reduction matrix, expressed as:

[0041]

[0042] Furthermore, the local spatial observation matrix is ​​expressed as:

[0043]

[0044] in, Represents the spatial observation matrix of the local scope.

[0045] Furthermore, the local time domain observation matrix is ​​expressed as:

[0046]

[0047] in, Represents the time domain observation matrix of the local scope.

[0048] Furthermore, the space-time observation matrix is ​​expressed as:

[0049]

[0050] Among them, D s and D t respectively represent the time-domain observation matrix and the space-domain observation matrix of the reduced-dimensional local range.

[0051] Furthermore, the training sample is expressed as:

[0052]

[0053] Among them, X I represents the echo data of the i-th range bin, represents the echo data after dimensionality reduction of the i-th range bin, as a detection unit, and respectively represent the echo data after dimensionality reduction of the (i - 2)-th and (i + 2)-th range bins, and X is used as a training sample.

[0054] Furthermore, the space-time estimation using the sparse Bayesian learning method is expressed as:

[0055] Initialize the hyperparameters

[0056] Among them, d i represents the i-th column of D, x i represents 's i-th column, N c is the number of columns of matrix D, ||·|| F represents the Frobenius norm of the matrix;

[0057] The steps of each iteration of the sparse Bayesian learning method are:

[0058]

[0059]

[0060] Among them, the matrix I is the identity matrix, Γ (t) represents the diagonal matrix, Y represents the clutter space-time spectrum, ⊙ represents the Hadamard product, and F represents the constant matrix N r is the number of rows of matrix D, Y i,j represents the element in the i-th row and j-th column of Y, Y :,j represents the data in the j-th column of Y, z (t+1) 、 and C (t+1) represent intermediate variables, and L represents the number of training samples;

[0061] When the convergence condition is reached, the iteration stops, and the optimal estimate of Y, that is, the clutter space-time spectrum, is obtained;

[0062] The convergence conditions are as follows:

[0063] (1) Reaching the maximum number of iterations

[0064] (2) ||γ (t) - γ (t-1) || 2 / ||γ (t) || 2 <δ, where δ is a positive number and δ = 0.001.

[0065] Furthermore, the detected unit y after filtering in step twelve is expressed as:

[0066]

[0067] R = DYY H D H

[0068] where R represents the covariance matrix of clutter and w represents the filter weight.

[0069] The beneficial effects of the present invention are:

[0070] This application is based on local continuous frequency domain dimension reduction and sparse Bayesian reconstruction technology for effectively suppressing ionospheric clutter. By constructing a high-frequency ground wave radar signal model, using dimension reduction processing in the local frequency domain, combined with the sparse Bayesian algorithm, it realizes the efficient estimation of the space-time spectrum of ionospheric clutter, reconstructs the covariance matrix of clutter, and finally uses the STAP method to achieve ionospheric clutter suppression. The suppression effect of the technical solution of this application is remarkable. This application has improved by 3.09 dB compared with the SBL method in terms of signal-to-clutter-plus-noise ratio (SCNR), significantly enhancing the clutter suppression performance.

[0071] Through local frequency domain dimension reduction processing, the running time of an average single range gate of this application is only 0.31% of that of the SBL method, greatly improving the real-time processing ability of the algorithm and being suitable for real-time applications in actual ground wave radar systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0072] Figure 1 is the off-line process flow chart of this application;

[0073] Figure 2 is the on-line process flow chart of this application;

[0074] Figure 3 is the output comparison chart after suppression. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0075] It should be particularly noted that, without conflict, the various embodiments disclosed in this application can be combined with each other.

[0076] Embodiment 1: The sparse Bayesian ionospheric clutter STAP method based on local continuous frequency domain dimensionality reduction described in this embodiment includes an offline step and an online step.

[0077] The offline step is as follows, and the flowchart is as Figure 1 shown:

[0078] Step 1: Construct a local continuous frequency domain feature matrix and a local continuous spatial frequency domain feature matrix according to the preset local frequency range, local spatial frequency range, and window function parameters;

[0079] The parameters of the HFSWR system are assumed as follows: The uniform linear array consists of N array elements with an element spacing of d, and the phase coherent accumulation period contains M signal pulses, where fs is the sampling frequency. Therefore, for the I-th range cell, a data vector of N×M dimensions can be obtained as follows:

[0080]

[0081] The spatio-temporal steering vector of the target echo is defined as follows:

[0082]

[0083] where are the spatial steering vector and the temporal steering vector respectively, is the Kronecker product, λ is the signal wavelength, φ t is the azimuth of the target echo, f t is the Doppler frequency of the target echo, f R is the pulse repetition frequency.

[0084] The received data can be expressed as the sum of the target signal, clutter, and noise:

[0085] x I = vec(X I ) = ξv(f t , φ t ) + c + n (3)

[0086] where ξ is the amplitude of the target, c is the ionospheric clutter, and n is the noise.

[0087] Since most of the ionospheric clutter is strongly correlated in the local angle and Doppler space, it is first necessary to determine the range of the angle-Doppler region of the selected samples and extract the corresponding features.

[0088] The observation matrix of the spatial steering matrix is:

[0089]

[0090] where N is the number of antennas, φ 0 is the starting frequency, p is the sparsity, and Δφ is the angular difference between adjacent atoms.

[0091] Let Rewrite A in the frequency-angle domain s :

[0092]

[0093] where Δf s is the frequency difference between adjacent atoms.

[0094] When p→∞ and Δf s →0, the continuous spatial frequency domain is:

[0095]

[0096] Take the local spatial frequency from f s1 to f s2 , and use the window function W s . The local continuous spatial frequency domain feature matrix can be expressed as:

[0097]

[0098] Since

[0099]

[0100] Similarly, take the local frequency from f t1 to f s2 , and use the window function W t . The local continuous frequency domain feature matrix can be expressed as:

[0101]

[0102] j is the imaginary unit in mathematics;

[0103] Step 2: Perform eigenvalue decomposition on the local continuous frequency domain feature matrix and the local continuous spatial frequency domain feature matrix respectively to construct a time-domain dimensionality reduction matrix and a spatial-domain dimensionality reduction matrix;

[0104] Perform eigenvalue decomposition on H s and H t :

[0105] H s = U s S s V s H , H t = U t S t V tH (10)

[0106] Let U s =[U s1 ,U s2 ,U t =[U t1 ,U t2 , where U s1 and U s2 are the first k s columns and the remaining N - k s columns of U s respectively, and U t1 and U t2 are the first k t columns and the remaining M - k t columns of U t respectively. Obtain the spatial domain dimensionality reduction matrix T s , and the time domain dimensionality reduction matrix T t :

[0107]

[0108] Step 3: Construct the time domain observation matrix and the spatial domain observation matrix for the local range according to the preset attenuation threshold, and perform dimensionality reduction on the time domain observation matrix and the spatial domain observation matrix for the local range.

[0109] Obtain the spatial domain observation matrix for the local range and the time domain observation matrix for the local range:

[0110]

[0111] where

[0112] Perform dimensionality reduction on the observation matrix:

[0113]

[0114]

[0115] Then the spatio - temporal observation matrix is:

[0116]

[0117] The online steps are as shown in the flowchart Figure 2 as follows:

[0118] Step 1: Obtain the radar echo data, perform matched filtering and array calibration on the radar echo data; then use the time domain dimensionality reduction matrix and the spatial domain dimensionality reduction matrix to perform dimensionality reduction processing on the processed radar echo data to obtain the dimensionality - reduced data;

[0119] Perform dimensionality reduction on the radar echo data:

[0120]

[0121] When the test data of the radar echo data is the I-th range bin, select the (I - 2)-th and (I + 2)-th range bins as the training data:

[0122]

[0123] Step 2: Use the spatio-temporal observation matrix and perform spatio-temporal estimation on the dimension-reduced training data using the sparse Bayesian learning method to obtain the clutter spatio-temporal spectrum Y;

[0124] The optimal estimation for each iteration of the sparse Bayesian learning method is:

[0125] C (t+1) = DΓ (t) D H + σ 2(t) I (19)

[0126]

[0127] where ⊙ represents the Hadamard product,

[0128] The algorithm of sparse Bayesian is an iterative process, initialize the hyperparameters

[0129] Convergence condition for iteration:

[0130] (1) Reach the set maximum number of iterations maxiter.

[0131] (2) ||γ (t) - γ (t-1) || 2 / ||γ (t) || 2 < δ, where δ is a very small positive number, usually set to 0.001.

[0132] Step 3: Reconstruct the covariance matrix of the clutter using the clutter spatio-temporal spectrum; construct the filter weights using the covariance matrix of the clutter and the spatio-temporal steering vector of the desired target; perform filtering on the dimension-reduced test data using the filter weights to remove the ionospheric clutter.

[0133] Specific method:

[0134] Reconstruct the covariance matrix of the clutter using the clutter spatio-temporal spectrum:

[0135] R = DYY H D H (25)

[0136] Solve for the filter weights:

[0137] w = R -1 v (26)

[0138] Filter the dimension-reduced test data to obtain the final output:

[0139]

[0140] Example:

[0141] Adopt the ground wave radar data of an N-element uniform linear array with a phase coherent accumulation period containing M signal pulses. The element spacing is d, and the radar operating frequency is fs. Set the relevant parameters by injecting simulated targets into the ionospheric clutter range.

[0142] Offline steps:

[0143] Step 1: Construct a local continuous frequency domain feature matrix and a local continuous spatial frequency domain feature matrix according to the preset local frequency range, local spatial frequency range, and window function type parameters;

[0144] Take the local spatial frequency from f s1 = -0.0625 to f s2 = 0.0625, and use the rectangular window function W s . The local continuous spatial frequency domain feature matrix can be expressed as:

[0145]

[0147] Similarly, take the local frequency from f t1 = -0.00495 to f t2 = 0.00495, and use the Hanning window function W t . The local continuous frequency domain feature matrix can be expressed as:

[0148]

[0149] Step 2: Perform eigenvalue decomposition on the local continuous frequency domain feature matrix and the local continuous spatial frequency domain feature matrix, construct a time domain dimension reduction matrix and a spatial domain dimension reduction matrix, and obtain the final spatio-temporal observation matrix;

[0150] Perform eigenvalue decomposition on H s , H t :

[0151] H s = U s S s V sH , H t = U t S t V t H (30)

[0152] Let U s = [U s1 , U s2 , U t = [U t1 , U t2 , where U s1 and U s2 are the first k s = 6 columns and the remaining N - k s columns of U s respectively, and U t1 and U t2 are the first k t = 5 columns and the remaining M - k t columns of U t respectively. Obtain the spatial domain dimensionality reduction matrix T s , and the time domain dimensionality reduction matrix T t :

[0153]

[0154] Step 3: According to the preset attenuation threshold, construct the time domain observation matrix and the spatial domain observation matrix of the local range, and perform dimensionality reduction on the time domain observation matrix and the spatial domain observation matrix of the local range.

[0155] Take the local spatial frequency from to and use the rectangular window function W s . The spatial domain observation matrix of the local range can be expressed as:

[0156]

[0157] Take the local frequency from to and use the Hanning window function W t . The time domain observation matrix of the local range can be expressed as:

[0158]

[0159] Perform dimensionality reduction on the observation matrix:

[0160]

[0161] Then the spatio-temporal observation matrix is:

[0162]

[0163] The online steps are as follows:

[0164] Step 1: Obtain radar echo data, perform matched filtering and array calibration on the radar echo data, and then use a time-domain dimensionality reduction matrix and a spatial-domain dimensionality reduction matrix to perform dimensionality reduction processing on the radar echo data to obtain the dimensionality-reduced data;

[0165] Perform dimensionality reduction on the radar echo data:

[0166]

[0167] When the test data is the I = 281st range cell, select the I - 2nd and I + 2nd range cells as training data:

[0168]

[0169] Step 2: Use the spatio-temporal observation matrix to perform spatio-temporal estimation on the dimensionality-reduced training data using the sparse Bayesian learning method to obtain the clutter spatio-temporal spectrum;

[0170] The optimal estimate for each iteration of the sparse Bayesian learning method is:

[0171] C (t+1) = DΓ (t) D H + σ 2(t) I (39)

[0172]

[0173]

[0174] where ⊙ represents the Hadamard product (Hadamard product),

[0175] The algorithm of sparse Bayesian is an iterative process. Initialize the hyperparameters

[0176] Convergence condition for iteration:

[0177] (1) Reach the set maximum number of iterations of 100.

[0178] (2) ||γ (t) - γ (t-1) || 2 / ||γ (t) || 2 <δ, where δ is a very small positive number, set to 0.001.

[0179] Step 3: Reconstruct the clutter covariance matrix using the clutter spatio-temporal spectrum; construct the filter weights using the clutter covariance matrix and the spatio-temporal steering vector of the expected target; filter the dimension-reduced test data using the filter to remove the ionospheric clutter.

[0180] Specific method:

[0181] Reconstruct the clutter covariance matrix using the clutter spatio-temporal spectrum:

[0182] R = DYY H D H (45)

[0183] Solve for the filter weights:

[0184] w = R -1 v (46)

[0185] Filter the dimension-reduced test data to obtain the final output:

[0186]

[0187] Finally, compare this application with the SBL method. Figure 3 Figure and Table 1 respectively give the comparison diagram of the outputs after suppression by different algorithms and the comparison diagram of the SCNR performance of the outputs after suppression by the algorithms. Table 2 gives the comparison diagram of the average running time of a single range gate for different algorithms. The results show that the suppression performance of this application is better than that of the SBL method, and it can significantly reduce the actual running time.

[0188] Table 1 Comparison diagram of the SCNR performance of the output after suppression of this application

[0189]

[0190] Table 2 Comparison diagram of the actual running time of this application

[0191]

[0192] This application realizes the efficient suppression and real-time processing of ionospheric clutter through effective dimension reduction processing and sparse reconstruction technology, significantly improving the target detection performance and computational efficiency of the ground wave radar system, and having broad application prospects.

[0193] It should be noted that the specific implementation manners are only explanations and illustrations of the technical solutions of the present invention, and the scope of the right protection cannot be limited thereby. Any changes that are merely partial based on the claims and the description of the present invention should still fall within the protection scope of the present invention.

Claims

1. The sparse Bayesian ionospheric clutter STAP method based on local continuous frequency domain dimensionality reduction is characterized by The method comprises the following steps: Step 1: Obtain the local frequency range, local spatial frequency range, and time domain window function W t And the spatial window function W s , thereby constructing a local continuous frequency domain feature matrix and a local continuous spatial frequency domain feature matrix; 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 a time domain dimensionality reduction matrix and a spatial domain dimensionality reduction matrix; Step 3: Obtain the attenuation threshold to construct the local time domain measurement matrix and the spatial domain measurement matrix, and use the time domain dimension reduction matrix and the spatial domain dimension reduction matrix to reduce the dimension of the local time domain measurement matrix and the spatial domain measurement matrix. Finally, obtain the Kronecker product of the local time domain measurement matrix and the spatial domain measurement matrix after dimension reduction, that is, the spatial-temporal measurement matrix. Step 4: Acquire radar echo data and preprocess the radar echo data, wherein the preprocessing includes matched filtering and array calibration processing; Step 5: Based on the preprocessed radar echo data, all distance units of the radar echo data are obtained, and all distance units of the radar echo data are reduced in dimension using the time domain dimensionality reduction matrix and the space domain dimensionality reduction matrix; Step 6: Get the Doppler frequency f of the desired target t , and construct the time-domain steering vector b(f t ), and use the time domain dimension reduction matrix to calculate the time domain steering vector b(f t ) performs dimensionality reduction processing to obtain a time domain steering vector after dimensionality reduction; Step 7: Get the azimuth angle φ of the desired target t , in order to construct the desired target's spatial guidance vector a(φ t ), and use the spatial dimension reduction matrix to calculate the spatial guidance vector a(φ t ) performs dimensionality reduction processing to obtain a spatial guidance vector after dimensionality reduction; Step 8: Obtain the Kronecker product of the reduced-dimensional time-domain steering vector and the reduced-dimensional space-domain steering vector, i.e., the reduced-dimensional space-time steering vector Step 9: Among all the distance units of the radar echo data after dimensionality reduction, randomly select the Ith distance unit as the detection unit, and select the I-2th and I+2th distance units as training samples; Step 10: Use the space-time observation matrix and the sparse Bayesian learning method to perform space-time estimation on the two training samples to obtain the clutter space-time spectrum; Step 11: Reconstruct the covariance matrix of the clutter using the clutter space-time spectrum, and use the reconstructed clutter covariance matrix and the reduced-dimensional space-time steering vector Construct filter weights; Step 12: Filter the detection unit using the filter weights; Step 13: Repeat steps 9 to 12 until all the echo data after dimension reduction in the range unit are processed, and the echo data with ionospheric clutter removed is obtained.

2. The sparse Bayesian ionospheric clutter STAP method based on local continuous frequency domain dimensionality reduction according to claim 1 is characterized in that The local continuous spatial frequency domain feature matrix is ​​expressed as: Where N is the number of array elements, d is the array element spacing, and f s represents the sampling frequency, W s represents the spatial window function, f s1 and f s2 Indicates that the local spatial frequency changes from f s1 to f s2 ; The local continuous frequency domain feature matrix is ​​expressed as: Where M represents the number of signal pulses, W t represents the time domain window function, f t1 and f s2 Indicates that the local frequency changes from f t1 to f s2 .

3. The sparse Bayesian ionospheric clutter STAP method based on local continuous frequency domain dimensionality reduction according to claim 2 is characterized in that The time domain steering vector b(f t ) is expressed as: Among them, f R is the pulse repetition frequency; The spatial guidance vector a(φ t ) is expressed as: Where, λ is the signal wavelength; The space-time steering vector after dimension reduction It is expressed as: Among them, T s represents the spatial dimension reduction matrix, T t Represents the time domain dimension reduction matrix.

4. The sparse Bayesian ionospheric clutter STAP method based on local continuous frequency domain dimensionality reduction according to claim 3 is characterized in that The specific steps of step 2 include: Step 21: Perform eigenvalue decomposition on the local continuous frequency domain feature matrix and the local continuous spatial frequency domain feature matrix respectively, expressed as: H s =U s WITH s IN s H H t =U t WITH t IN t H Among them, H s and H t They represent the local continuous frequency domain feature matrix and the local continuous spatial frequency domain feature matrix respectively, U s Represents the matrix composed of spatial eigenvectors, U t represents the matrix composed of time domain eigenvectors, S s represents the diagonal matrix composed of spatial eigenvalues, S t Represents a diagonal matrix composed of time domain eigenvalues; Step 22: Let U s =[U s1 ,U s2 ],U t =[U t1 ,U t2 ], where U s1 and U s2 U s The first k s Column and remaining Nk s Column, U t1 and U t2 U t The first k t Column and the remaining Mk t Columns, thus obtaining the time domain dimensionality reduction matrix and the space domain dimensionality reduction matrix, expressed as:

5. The sparse Bayesian ionospheric clutter STAP method based on local continuous frequency domain dimensionality reduction according to claim 4 is characterized in that The spatial domain observation matrix of the local range is expressed as: in, Represents the spatial observation matrix of the local scope.

6. The sparse Bayesian ionospheric clutter STAP method based on local continuous frequency domain dimensionality reduction according to claim 5 is characterized in that The local range time domain observation matrix is ​​expressed as: in, Represents the time domain observation matrix of the local scope.

7. The sparse Bayesian ionospheric clutter STAP method based on local continuous frequency domain dimensionality reduction according to claim 6 is characterized in that The space-time observation matrix is ​​expressed as: Among them, D s and D t They respectively represent the time domain measurement matrix and spatial domain measurement matrix of the local range after dimensionality reduction.

8. The sparse Bayesian ionospheric clutter STAP method based on local continuous frequency domain dimensionality reduction according to claim 7 is characterized in that The training sample is expressed as: Among them, X I Represents the echo data of the I-th range unit, Represents the echo data after dimension reduction of the I-th distance unit, as the detection unit, and They represent the echo data of the I-2th and I+2th distance units after dimensionality reduction, and X is used as a training sample.

9. The sparse Bayesian ionospheric clutter STAP method based on local continuous frequency domain dimensionality reduction according to claim 8 is characterized in that The space-time estimation using sparse Bayesian learning method is expressed as: Initializing Hyperparameters Among them, d i represents the i-th column of D, x i express The i-th column, N c is the number of columns of matrix D, ||·|| F It means to find the Frobenius norm of the matrix; The steps of each iteration of the sparse Bayesian learning method are: Among them, matrix I is the unit matrix, Γ (t) represents a diagonal matrix, Y represents the clutter space-time spectrum, ⊙ represents the Hadamard product, and F represents the constant matrix N r is the number of rows in matrix D, Y i,j represents the element in the i-th row and j-th column of Y, :,j represents the data in the jth column of Y, z (t+1) , and C (t+1) represents the intermediate variable, L represents the number of training samples; When the convergence condition is reached, the iteration stops and Y is optimally estimated, i.e., the clutter space-time spectrum; The convergence condition is: (1) Reaching the maximum number of iterations (2)||c (t) -c (t-1) ||2 / ||c (t) ||2<δ, δ is a positive number, δ=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 y after filtering in step 12 is expressed as: R=DYY H D H Among them, R represents the covariance matrix of clutter, and w represents the filter weight.

Images