Target protection type mask prior guided low rank sparse reconstruction method for fmcw radar jamming suppression

Abstract

This invention belongs to the field of radar signal processing and intelligent sensing technology, specifically relating to a target protection mask prior-guided low-rank sparse reconstruction method for FMCW radar interference suppression. The method includes the following steps: 1. Acquiring FMCW radar observation signals and preprocessing them; 2. Performing time-frequency transformation on the preprocessed FMCW radar observation signals to extract basic features; 3. Accurately locating the interference region in the time-frequency domain, forming a binary mask prior, and constructing a mandatory target protection region; 4. Constructing a low-rank sparse matrix and a target protection low-rank sparse decomposition model based on mask prior guidance; 5. Solving the target protection low-rank sparse decomposition model to obtain the decomposition results; 6. Reconstructing and post-processing the decomposition results, outputting the interference suppression results, and evaluating the effect. This method combines the intuitiveness of time-frequency detection methods with the reconstruction capability of low-rank sparse methods, reducing the probability of target echoes being misclassified into sparse terms.

Description

Technical Field

[0001] This application belongs to the field of radar signal processing and intelligent sensing technology, specifically relating to a target protection mask prior-guided low-rank sparse reconstruction method for FMCW radar interference suppression. Background Technology

[0002] FMCW radar acquires beat frequency information related to target distance and velocity by transmitting linear frequency modulated continuous waves and demodulating the echoes. With the large-scale deployment of vehicle-mounted radar, security radar, and industrial millimeter-wave radar, multiple FMCW radars are prone to mutual interference due to overlapping operating frequency bands, similar frequency modulation parameters, or partial temporal overlap. This type of interference often manifests as short bursts of high-energy signals in the time domain and as diagonal lines or localized areas of strong energy concentration in the time-frequency domain, significantly reducing the detectability of useful target echoes.

[0003] Existing technologies mainly fall into the following categories: The first category is interference detection and suppression methods based on time-frequency analysis. These methods typically first perform a short-time Fourier transform (STFT) or other time-frequency transform on the beat frequency signal to detect abnormally high-energy regions in the time-frequency graph. Then, they remove interference using methods such as threshold decision, mask zeroing, neighborhood interpolation, or local repair. Their advantages are intuitive flow and relatively simple implementation, but they are sensitive to threshold settings, mask range, and interference morphology. Existing interference suppression methods based on time-frequency binary masks typically focus on detecting and eliminating interference regions, which can weaken strong interference to some extent. The second category is interference suppression methods based on low-rank sparse decomposition or RPCA. These methods utilize the differences in structural characteristics between the useful target signal and the interference signal to decompose the observation matrix into low-rank and sparse parts, thereby achieving signal separation. However, these existing technologies still have the following technical problems:

[0004] (1) Interference suppression methods based on time-frequency domain binary masks usually focus on detecting and eliminating interference. When interference and target echoes overlap locally, the effective components of the target are easily misjudged as interference and eliminated together, resulting in target energy weakening, target structure damage and loss of weak targets.

[0005] (2) The FMCW radar interference suppression method based on traditional low-rank sparse decomposition (RPCA) is prone to misclassifying target echoes, especially weak targets, low-energy targets, edge targets or locally discontinuous targets, into sparse terms in the absence of prior constraints, resulting in target leakage, range image distortion and decreased detection performance.

[0006] (3) The time-frequency domain detection, mask construction, low-rank sparse decomposition and reconstruction are often independent of each other, lacking a clear coupling mechanism, unable to make full use of interference prior information to guide the decomposition process, and lacking a special protection mechanism for the target region.

[0007] Therefore, a new method is urgently needed to use prior information about interference in the time and frequency domain to initially constrain the abnormal region, improve the signal recovery quality through low-rank sparse reconstruction, and introduce a target protection mechanism during the reconstruction process to reduce the problems of target deletion and target leakage, thereby achieving a more robust interference suppression effect. Summary of the Invention

[0008] To address the aforementioned technical problems in the existing technology, the purpose of this invention is to reduce the problems of target deletion and target leakage, and to achieve a robust interference suppression effect. The technical solution is as follows:

[0009] A target-protected mask-guided low-rank sparse reconstruction method for FMCW radar jamming suppression includes the following steps:

[0010] Step 1: Acquire frequency modulated continuous wave (FMCW) radar observation signals and perform preprocessing;

[0011] Step 2: Perform time-frequency transformation on the preprocessed FMCW radar observation signal and extract basic features;

[0012] Step 3: Accurately locate the interference area in the time-frequency domain and form a binary mask prior. Based on the continuity characteristics of the target echo in the frequency domain, construct a mandatory target protection area.

[0013] Step 4: Construct a low-rank sparse matrix and a target-protected low-rank sparse decomposition model based on mask prior guidance;

[0014] Step 5: Solve the target protected low-rank sparse decomposition model using the inexact augmented Lagrange multiplier method to obtain the decomposition results;

[0015] Step 6: Reconstruct and post-process the decomposition results, output the interference suppression results, and evaluate the effect.

[0016] Furthermore, in step 1, the FMCW radar observation signal is the original intermediate frequency beat frequency signal acquired under single-frame or multi-frame conditions. It is a one-dimensional complex sampled signal of a single linear modulation pulse chirp or a two-dimensional complex matrix signal composed of multiple chirps. The matrix form expression of the FMCW radar observation signal is as follows:

[0017]

[0018] in, The observation matrix to be processed. Indicates that there is A fast-time sampling point, Indicates that there is One chirp;

[0019] The preprocessing includes:

[0020] ① Perform window function weighting, multiplying chirp by a Hamming window, Hann window, or Blackman window;

[0021] ② The FMCW radar observation signal is normalized, and the expression is:

[0022]

[0023] in, It is a very small positive number. This is the normalized observation signal.

[0024] Furthermore, the FMCW radar observation signal is generated by superimposing the target echo component, background noise component, and interference component, as expressed in the following expression:

[0025]

[0026] in, For the target echo signal, This is an interference signal. It is Gaussian white noise.

[0027] Furthermore, the process of step 2 is as follows:

[0028] Step 2.1: Perform a Short Time Fourier Transform (STFT) on each chirp of the preprocessed FMCW radar observation signal to obtain the corresponding time-frequency complex matrix. ;

[0029] Step 2.2: Based on the time-frequency complex matrix Calculate the time-frequency amplitude matrix The calculation expression is:

[0030]

[0031] in, For indexing time windows, For frequency indexing.

[0032] Furthermore, the specific process of step 3 is as follows:

[0033] Step 3.1: Search for and locate the target energy ridge. The specific steps are as follows:

[0034] Step 3.1.1: Based on the physical characteristics of stationary or low-speed targets in FMCW radar appearing as horizontal high-energy ridges in the time-frequency diagram of a single snapshot frequency signal, energy accumulation is performed on the time-frequency amplitude matrix along the time axis.

[0035] Step 3.1.2: Calculate the first... The sum of the cumulative energy at each frequency index is calculated using the following expression:

[0036]

[0037] in, This represents the total energy of each frequency row. Indicates the total number of time windows;

[0038] Step 3.1.3: Determine the center frequency row of the useful target by searching for the maximum point in the energy accumulation sequence. ;

[0039] Step 3.2: Calculate the adaptive threshold based on statistical quantiles

[0040] The percentage threshold method is used to analyze the two-dimensional time-frequency amplitude spectrum. All pixel values ​​are sorted in ascending order, and a specific high quantile is selected. As the threshold for interference judgment ;

[0041] Step 3.3: Generate an initial binary mask and perform morphological thinning. The specific process is as follows:

[0042] Step 3.3.1: Determine the threshold based on the interference. Decisions are made on the time-frequency graph to generate a preliminary binary mask;

[0043] Step 3.3.2: Compare the pixel capability value with a threshold value. When the threshold is reached, the mask value is set to 0, i.e., the reserved area; when the pixel capability value... When the threshold is reached, the mask value is set to 1, which is the interference removal area;

[0044] Step 3.3.3: Perform a morphological dilation operation on the preliminary binary mask using specific structuring elements to obtain an interfering prior mask. ;

[0045] Step 3.4: Construct a mandatory target protection zone. The specific process is as follows:

[0046] Step 3.4.1: Define the guard band, using the center frequency from step 3.1.3. Centered on the target area, set symmetrical frequency protection radius values ​​to define the protection range for the target area. ;

[0047] Step 3.4.2: Perform a forced rewrite, forcibly setting the mask value at all time indices within the target area protection range to 1, thus obtaining the target protection mask. , which is represented as;

[0048]

[0049] in, The set frequency protection radius is symmetrical between the upper and lower sides;

[0050] Step 3.5: Use the protected binary mask as the prior guiding weight for low-rank sparse decomposition.

[0051] Furthermore, in step 3.3.3, the specific structural element is... A rectangular structural element.

[0052] Furthermore, in step 4, the process of constructing a low-rank sparse matrix guided by mask priors is as follows:

[0053] ①The interference prior mask and the target protection mask The fusion process is performed to construct prior constraints that guide RPCA decomposition; the target protection mask is then applied. After removing the interference support set, the corrected interference prior mask is obtained, and its expression is:

[0054]

[0055] in, Represents element-wise product;

[0056] ② Construct a weighting matrix to control the sparsity penalty intensity in different regions. The expression for the weighting matrix is ​​as follows:

[0057]

[0058] in, These are the weighting coefficients for different regions, and they satisfy... This indicates that the occurrence of sparse components is encouraged in the interference area and suppressed in the target protection area.

[0059] The process of constructing the target-protected low-rank sparse decomposition model is as follows:

[0060] a. Transform the time-frequency amplitude matrix The time-frequency amplitude matrix is ​​decomposed into low-rank terms, sparse terms, and noise residual terms. The expression is:

[0061]

[0062] in, This indicates the low-rank portion that retains the main target structure and useful background information; This represents the sparse anomaly associated with the interference. This represents noise and modeling error;

[0063] b. Construct a target-protected low-rank sparse decomposition model by using interference mask priors and target protection constraints. Its expression is:

[0064]

[0065]

[0066] in, For nuclear norm, For weighted sparsity constraints, To provide additional penalty for sparse components appearing in non-interference regions, and To adjust the parameters.

[0067] Furthermore, the specific process of step 5 is as follows:

[0068] Step 5.1: Convert the time-frequency amplitude matrix It serves as input to a low-rank sparse decomposition model and decomposes it into low-rank terms. and sparse terms ;

[0069] Step 5.2: Using the Lagrange multiplier matrix The augmented Lagrangian function is constructed using the penalty parameter, and its expression is:

[0070]

[0071] in, Indicates the inner product. Denotes the Frobenius norm. For penalty parameters;

[0072] Step 5.3: The augmented Lagrange function is iteratively solved using the inexact augmented Lagrange multiplier method. The specific process is as follows:

[0073] The low-rank term, sparse term, and multiplier term are initialized, and the following update steps are performed during the iteration:

[0074] ① Update the lower-rank terms

[0075] With the current sparse terms fixed and Lagrange multiplier matrix Under the condition of constructing the intermediate matrix The expression is:

[0076]

[0077] in, Indicates the first The next iteration;

[0078] For the intermediate matrix Perform singular value decomposition and soft threshold shrink the singular values ​​to obtain the updated low-rank terms. ;

[0079] ② Update the sparse terms

[0080] Low-rank terms after fixed update and the current Lagrange multiplier matrix Under the condition of constructing the intermediate matrix The expression is:

[0081]

[0082] For the intermediate matrix Element-wise soft thresholding is performed to obtain the updated sparse terms. ;

[0083] ③ Perform constraint correction on the target protection mask.

[0084] The support region of the sparse terms is modified by updating the sparse terms and using a pre-constructed target protection binary mask. For positions in the target protection binary mask marked as target protection regions or reserved regions, the values ​​of the sparse terms at those positions are suppressed. For positions in the target protection binary mask marked as suspected interference regions, the corresponding components of the sparse terms are retained.

[0085] ④ Update the multiplier terms

[0086] After updating the low-rank and sparse terms, the Lagrange multiplier matrix is ​​updated. The expression for the updated Lagrange multiplier matrix is ​​as follows:

[0087]

[0088] in, This is the updated Lagrange multiplier matrix;

[0089] ⑤ Update the penalty parameters

[0090] After each iteration, the penalty parameter is adjusted. Perform incremental updates and record the updated penalty parameter as... ;

[0091] ⑥ Perform convergence determination

[0092] Calculate the reconstruction error between the sum of the current low-rank terms and sparse terms and the observation matrix. Set a preset error threshold. Compare the reconstruction error with the preset error threshold. Stop the iteration when the reconstruction error is lower than the preset error threshold or the change between two adjacent iterations is lower than the preset error threshold, and output the final low-rank term. and sparse terms .

[0093] Furthermore, in step 5.3, the initialization of the low-rank term, sparse term, and multiplier term involves setting the initial values ​​of the low-rank term and the sparse term to zero matrices, initializing the Lagrange multiplier matrix to a matrix of the same dimension as the FMCW radar observation matrix, and initializing the penalty parameter to a positive number.

[0094] Furthermore, in step 6, the reconstruction and post-processing involves: processing low-rank terms... Amplitude correction and normalization are performed, and the corrected time-frequency representation is then subjected to inverse STFT to recover the time-domain signal.

[0095] Beneficial effects: (1) An interference prior mask is first constructed in the time-frequency domain, and then reconstructed through low-rank sparse decomposition, which combines the intuitiveness of time-frequency detection methods with the reconstruction capability of low-rank sparse methods. (2) The introduction of a target protection mechanism effectively reduces the probability of target echoes being misclassified into sparse terms, and is suitable for weak targets, locally overlapping targets, and scenarios with multiple interferences. (3) By introducing the mask prior into the RPCA model, the update range of sparse terms or the strength of sparse penalty are constrained, reducing the target leakage and decomposition instability caused by unconstrained decomposition. (4) It can maintain the target main peak and target structure well while suppressing interference, and improve the accuracy of subsequent distance detection, Doppler estimation and target recognition. Attached Figure Description

[0096] Figure 1 This is a flowchart of the target protection mask prior-guided low-rank sparse reconstruction method for FMCW radar interference suppression according to the present invention.

[0097] Figure 2 This is a flowchart of the Augmented Lagrange Multiplier Optimization (IALM) algorithm of the present invention;

[0098] Figure 3 This is a time-domain waveform diagram of a single target and single interference signal according to the present invention;

[0099] Figure 4a This is a schematic diagram of the original time-frequency characteristics of the single-target, single-interference binary mask of the present invention;

[0100] Figure 4b This is a schematic diagram of the binary mask for the single-target, single-interference binary mask of the present invention;

[0101] Figure 4cThis is a schematic diagram of the single-target, single-interference binary mask after mask removal according to the present invention;

[0102] Figure 5 This is a schematic diagram of the RPCA decomposition simulation results of the present invention;

[0103] Figure 6 This is a comparison chart of simulation results between the multi-Chirp binary mask and low-rank sparse decomposition of the present invention;

[0104] Figure 7 This is a schematic diagram of the Range-Doppler single-target single-interference output result of the present invention;

[0105] Figure 8 This is a schematic diagram of the time-domain waveform of a single target with multiple interferences according to the present invention;

[0106] Figure 9 This is a comparison chart of simulation results between the single-target multi-interference binary mask and the low-rank sparse decomposition of the present invention;

[0107] Figure 10 This is a schematic diagram of the Range-Doppler single-target multi-interference output result of the present invention. Detailed Implementation

[0108] The specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0109] like Figure 1 As shown, the target-protected mask prior-guided low-rank sparse reconstruction method for FMCW radar interference suppression of the present invention mainly includes the following steps:

[0110] Step 1: Acquisition and Preprocessing of Frequency Modulated Continuous Wave (FMCW) Radar Observation Signals

[0111] Acquire the raw intermediate frequency beat frequency signal collected by the FMCW radar under single-frame or multi-frame conditions. The beat frequency signal is a one-dimensional complex sampled signal of a single linear modulation pulse chirp or a two-dimensional complex matrix signal composed of multiple chirps. Assume that each chirp contains... Each frame contains a fast-time sampling point. If there are 1 chirp, then the FMCW radar observation signal is represented as:

[0112]

[0113] in, The observation matrix to be processed contains target echo components, background noise components, and interference components;

[0114] The FMCW radar observation signal originates from target echo and interference superposition simulation data constructed using the MATLAB platform. First, the target echo signal is constructed, then multiple interference components are superimposed, and Gaussian white noise can be further superimposed to form the final FMCW radar observation signal, represented as:

[0115]

[0116] in, For the target echo signal, This is an interference signal. It is Gaussian white noise;

[0117] Preprocessing of FMCW radar observation signals mainly includes the following steps:

[0118] ① Perform window function weighting. To reduce spectral leakage, multiply the chirp by a Hamming window, Hann window, or Blackman window.

[0119] ② To facilitate the processing of unified thresholds or unified reconstruction parameters across different frames and chirps, the observed signal is normalized, as expressed by:

[0120]

[0121] in, It is a very small positive number. This is the normalized observation signal.

[0122] Step 2: Perform time-frequency transformation on the FMCW radar observation signal and extract basic features. The specific process is as follows:

[0123] To identify interference regions in the time-frequency domain, a short-time Fourier transform (STFT) is performed on each chirp of the preprocessed FMCW radar observation signal to obtain the corresponding time-frequency complex matrix. Further calculate the time-frequency amplitude matrix. Its calculation expression is:

[0124]

[0125] in, For indexing time windows, For frequency indexing.

[0126] In the time-frequency amplitude matrix In this context, the target signal typically appears as a horizontal energy ridge parallel to the time axis, while the interfering signal typically appears as a high-energy strip with a certain slope and short bursts or a localized anomalous high-energy cluster.

[0127] Step 3: Accurately locate the interference region in the time-frequency domain and form a binary mask prior. Based on the continuity characteristics of the target echo in the frequency domain, construct a mandatory target protection region. The specific sub-steps are as follows:

[0128] 3.1: The steps for searching and locating the target energy ridge are as follows:

[0129] 3.1.1: Based on the physical characteristics of stationary or low-speed targets in FMCW radar appearing as horizontal (or near-horizontal) high-energy ridges in the time-frequency diagram of a single snapshot frequency signal, energy accumulation is performed on the time-frequency amplitude matrix along the time axis.

[0130] 3.1.2: Calculate the first The sum of the cumulative energy at each frequency index is calculated using the following expression:

[0131]

[0132] in, This represents the total energy of each frequency row. Indicates the total number of time windows;

[0133] 3.1.3: Searching for energy accumulation sequences The maximum point in the value is used to determine the center frequency row where the useful target is located. ;

[0134] 3.2: Adaptive threshold calculation based on statistical quantiles. To robustly extract high-energy interference regions under complex interference backgrounds, this invention employs a percentage thresholding method to calculate the two-dimensional time-frequency amplitude spectrum. All pixel values ​​are sorted in ascending order, and a specific high quantile is selected. (In this example, the 92nd quantile is used; the typical value range is 90% to 95%) as the interference threshold. ;

[0135] This method can effectively cope with fluctuations in interference energy due to environmental changes, ensuring that the mask can cover the area with the highest energy. The proportion of interfering components;

[0136] 3.3: Generate an initial binary mask and perform morphological refinement. The specific process is as follows:

[0137] 3.3.1: Determine the threshold based on the interference. Decisions are made on the time-frequency graph to generate a preliminary binary mask;

[0138] 3.3.2: Compare the pixel capability value with a threshold; when the pixel capability value... When the threshold is reached, the mask value is set to 0, i.e., the reserved area; when the pixel capability value... When the threshold is reached, the mask value is set to 1, which is the interference removal area;

[0139] 3.3.3: To address the common spectral leakage phenomenon in radar signal processing, a specific structural element is employed ( A rectangular structural element is used to perform a morphological dilation operation on the initial binary mask to obtain an interfering prior mask. ;

[0140] This operation can appropriately expand the coverage of the high-energy interference area, ensuring that the residual energy at the edge of the interference strip is also included in the area to be reconstructed;

[0141] 3.4: Constructing a mandatory target protection zone, the specific process is as follows:

[0142] 3.4.1: Define a guard band, based on the center frequency. Centered on a central point, a frequency protection radius value is set symmetrically above and below, thereby defining the protection range of the target area. ;

[0143] 3.4.2: Perform mask forced rewriting, that is, force the mask value at all time indices within the protection range of the target area to be set to 1, and obtain the target protection mask. , which is represented as;

[0144]

[0145] in, The set frequency protection radius is symmetrical between the upper and lower sides;

[0146] 3.5: The protected binary mask is used as the prior guiding weight for low-rank sparse decomposition (RPCA). In the subsequent iteration of the inexact augmented Lagrange multiplier method (IALM), this matrix will directly act on the penalty factor of the sparse term, guiding the algorithm to strongly remove interference in the unprotected area and relax the constraints in the protected area, thereby achieving accurate signal reconstruction.

[0147] Step 4: Construct a low-rank sparse matrix and a target-protected low-rank sparse decomposition model based on mask prior guidance;

[0148] The process of constructing a low-rank sparse matrix based on mask prior guidance is as follows:

[0149] ① Mask the interference prior and target protection mask The fusion is performed to construct prior constraints that guide RPCA decomposition; preferably, the target protection mask is used. After removing the interference support set, the corrected interference prior mask is obtained, and its expression is:

[0150]

[0151] in, This represents element-wise multiplication; the significance of the above processing is that even if some target regions are falsely detected as interference in the initial interference detection, they can be removed from the interference mask after the target protection operation, thereby avoiding the subsequent RPCA from preferentially including them in the sparse terms.

[0152] ② Construct a weighting matrix to control the sparsity penalty intensity in different regions. The expression for the weighting matrix is ​​as follows:

[0153]

[0154] in, These are the weighting coefficients for different regions, and they satisfy... This indicates that the occurrence of sparse components is encouraged in the interference area and suppressed in the target protection area.

[0155] The process of constructing the target-protected low-rank sparse decomposition model is as follows:

[0156] Time-frequency amplitude matrix The decomposition consists of low-rank terms, sparse terms, and noisy residual terms, expressed as:

[0157]

[0158] in, The low-rank portion retains the main target structure and useful background information. This represents the sparse anomaly associated with the interference. This represents noise and modeling error;

[0159] The traditional RPCA optimization objective is usually:

[0160]

[0161]

[0162] in, The nuclear norm of matrix L is the sum of all its singular values, used to constrain low-rank properties. Representation matrix of Norm, which is the sum of the absolute values ​​of all elements, is used to constrain sparsity; These are the sparse constraint weight parameters;

[0163] Based on this, the present invention introduces interference mask prior and target protection constraints to construct a target-protected low-rank sparse decomposition model, the expression of which is:

[0164]

[0165]

[0166] in, For nuclear norm, For weighted sparsity constraints, To provide additional penalty for sparse components appearing in non-interference regions, and To adjust the parameters.

[0167] Step 5: Solve the target protected low-rank sparse decomposition model using the inexact augmented Lagrange multiplier method to obtain the decomposition results. The specific process is as follows:

[0168] 5.1: The time-frequency amplitude matrix It serves as input to a low-rank sparse decomposition model and decomposes it into low-rank terms. and sparse terms Among them, the low-rank term Used to represent the main structure of the target and background after interference suppression, sparse term Interference components used to represent local anomalies and high-energy bursts;

[0169] 5.2: To facilitate the solution, the standard RPCA basic form is adopted as the solution framework. To solve the above constrained optimization problem, a Lagrange multiplier matrix is ​​introduced. And the penalty parameter, construct the augmented Lagrangian function, the expression of which is:

[0170]

[0171] in, Indicates the inner product. Denotes the Frobenius norm. For penalty parameters;

[0172] 5.3: The augmented Lagrange function is solved iteratively using the inexact augmented Lagrange multiplier method, such as... Figure 2 As shown, the specific process is as follows:

[0173] The low-rank term, sparse term, and multiplier term are initialized. Specifically, the initial values ​​of the low-rank term and sparse term are set to zero matrices, the Lagrange multiplier matrix is ​​initialized to a matrix with the same dimension as the observation matrix, the penalty parameter is initialized to a positive number, and the following update steps are performed during the iteration:

[0174] ① Update the lower-rank terms

[0175] With the current sparse terms fixed and Lagrange multiplier matrix Under the condition of constructing the intermediate matrix The expression is:

[0176]

[0177] in, Indicates the first The next iteration;

[0178] For the intermediate matrix Perform singular value decomposition and soft threshold shrink the singular values ​​to obtain the updated low-rank terms. The essence of this process is to preserve the main structural components and suppress local abnormal structures caused by interference through singular value thresholding.

[0179] ② Update the sparse terms

[0180] Low-rank terms after fixed update and the current Lagrange multiplier matrix Under the condition of constructing the intermediate matrix The expression is:

[0181]

[0182] For the intermediate matrix Element-wise soft thresholding is performed to obtain the updated sparse terms. The purpose of this process is to extract interference components that are local anomalies, high-energy bursts and lack an overall continuous structure, so that the sparse terms correspond as closely as possible to the interference region.

[0183] ③ Perform constraint correction on the target protection mask.

[0184] By updating the sparse terms and using a pre-constructed target protection binary mask, the supporting region of the sparse terms is modified. The positions marked as target protection regions or reserved regions in the target protection binary mask are suppressed, and the sparse terms are allowed to retain their corresponding components at the positions marked as suspected interference regions in the target protection binary mask. In this way, the low-rank sparse decomposition results not only satisfy the low-rank and sparse characteristics of the matrix structure, but also remain consistent with the target protection prior in the time-frequency domain, thereby reducing the probability of the target being misclassified into the sparse terms.

[0185] ④ Update the multiplier terms

[0186] After updating the low-rank and sparse terms, update the Lagrange multiplier matrix. The expression for the updated Lagrange multiplier matrix is ​​as follows:

[0187]

[0188] in, The updated Lagrange multiplier matrix is ​​then updated through multiplier terms to improve the constraint conditions in subsequent iterations. The level of satisfaction gradually increases;

[0189] ⑤ Update the penalty parameters

[0190] To improve the efficiency of iterative convergence, the penalty parameter is adjusted after each iteration. The penalty parameter is incrementally updated to gradually strengthen its constraint on the decomposition error term. The updated penalty parameter is denoted as... ;

[0191] ⑥ Perform convergence determination

[0192] After each iteration, the reconstruction error between the sum of the current low-rank term and sparse term and the observation matrix is ​​calculated. A preset error threshold is set, and the reconstruction error is compared with the preset error threshold. When the reconstruction error is lower than the preset error threshold or the change between two adjacent iterations is lower than the preset error threshold, the iteration stops, and the final low-rank term is output. and sparse terms ;

[0193] Through the above solution process, the main structural components in the time-frequency amplitude matrix can be separated into low-rank terms, and the interference components of local anomalies can be separated into sparse terms. Furthermore, by combining a target-protected binary mask, target misclassification and target leakage problems can be further reduced, thus providing a reliable decomposition basis for subsequent background consistency filling and time-domain signal reconstruction.

[0194] Step Six: Reconstruct and post-process the decomposition results, output the interference suppression results, and evaluate the effect.

[0195] After solving, the low-rank term is obtained. sparse terms and optional residuals Among them, the low-rank terms Considered as the effective echo reconstruction result after interference suppression, the sparse term It is considered the main interference component. To further improve the output signal quality, the low-rank term is... Amplitude correction and normalization are performed, and the time-domain signal is recovered by inverse STFT on the corrected time-frequency representation. If there are still discontinuous regions, smoothing is performed. If the background is excessively weakened after low-rank decomposition, a background noise reconstruction strategy can be combined to recover part of the noise floor in order to avoid excessive sparsity of the output signal.

[0196] The reconstructed signal can be used in subsequent radar detection processes: range FFT, Doppler FFT, range-Doppler imaging, etc.

[0197] Finally, the interference suppression results are output and the effectiveness is evaluated. The processing results of this invention can be output and displayed in the following forms:

[0198] (1) Comparison of the original time-domain waveform, the disturbed time-domain waveform, and the reconstructed time-domain waveform;

[0199] (2) Comparison of the original time-frequency diagram, the interference time-frequency diagram, the binary mask diagram, and the reconstructed result diagram;

[0200] (3) Comparison of the original range spectrum, the disturbed range spectrum, and the suppressed range spectrum;

[0201] (4) Visual comparison of sparse terms and low-rank terms;

[0202] (5) Comparison of processing results of multiple frames or multiple chirp.

[0203] The present invention will be further illustrated below with a typical FMCW radar simulation example, but the present invention is not limited to this example.

[0204] Example 1: Target-protected mask prior-guided RPCA interference suppression under single-target, single-interference conditions

[0205] This example uses a 77GHz millimeter-wave radar, and some parameters are shown below:

[0206]

[0207] The FMCW radar is configured to transmit a linear frequency modulated signal with a sampling frequency of [frequency value missing]. The number of sampling points per chirp is Each frame contains One chirp;

[0208] Constructing a single target echo (with fixed range and position, and stable echo amplitude) is expressed as follows:

[0209]

[0210] in, Represents fast-time variables. Indicates the chirp sequence number. The beat frequency, For Doppler frequency shift, For Chirp cycles;

[0211] An interference signal is constructed, wherein the interference signal is a frequency sweep interference with a slope inconsistent with the frequency modulation slope of this radar, and its expression is:

[0212]

[0213] in, Indicates the amplitude of the interference signal; For the interference slope, This is the radar's transmission slope. This is for additional frequency offset.

[0214] The target signal and the interference signal are superimposed to form the interfered observation signal, such as Figure 3 As shown, the interfered observation signal exhibits obvious sudden high-energy oscillations in the time domain waveform. This high-energy envelope covers part of the original Chirp signal, causing the echo characteristics of the useful target to be obscured.

[0215] like Figure 4a As shown, a short-time Fourier transform (STFT) is performed on the observed signal to obtain the time-frequency diagram;

[0216] A preliminary binary mask is generated based on the local high-energy oblique line region in the time-frequency graph, such as... Figure 4b As shown, a morphological dilation operation is performed on the generated preliminary binary mask. By performing percentage threshold judgment and morphological dilation processing on the time-frequency map, the mask accurately covers the oblique interference area.

[0217] The original signal is subjected to distance FFT to locate the main peak of the target, and several units are extended on both sides of the main peak to form a target protection zone;

[0218] like Figure 4c As shown, the target protection area is removed from the interference mask to obtain the corrected interference prior;

[0219] A weighted RPCA model is constructed based on the corrected interference prior.

[0220] The IALM algorithm is used to solve the problem, yielding the low-rank and sparse terms. The simulation results of the low-rank sparse decomposition are as follows: Figure 5 As shown, the observation matrix was successfully decomposed into low-rank terms (mainly stationary target echoes) and sparse terms (captured high-energy sweep frequency interference) using the IALM algorithm.

[0221] The low-rank term is output as the effective signal after interference suppression, such as... Figure 6 As shown, the mask continuity and decomposition stability under multi-Chirp conditions are demonstrated. Experiments show that the mask prior can adapt to the changes in the interference positions between different Chirps, and the low-rank term can always stably extract the target features across Chirps.

[0222] Range-Doppler analysis was performed on the output results to compare the target peak value, residual interference, and background level before and after processing. Figure 7As shown, the Range-Doppler analysis diagrams are shown before (left) and after (right). The diagrams show that the noise floor rise caused by the interference after processing is effectively suppressed. The target peak is clearly presented at a distance of 50m and a speed of 10m / s, and the main peak structure of the target is intact, which verifies the effectiveness of the target protection mechanism.

[0223] Example 2: Target-protected mask prior-guided RPCA interference suppression under multi-source composite interference conditions

[0224] Based on a single-target signal, multiple interferences are superimposed simultaneously, with the remaining procedures the same as in Example 1, such as... Figure 8 As shown, the time-domain waveform is affected by multi-source interference, exhibiting multi-segment, high-frequency energy distortion; as... Figure 9 As shown, this invention demonstrates its advantages in handling multi-region interference. Even if the interference is distributed in a complex time-frequency domain, the generated binary mask can still guide the RPCA algorithm through prior weights to extract multiple sparse interference components from the complex background while retaining the low-rank target component.

[0225] like Figure 10 As shown, the Range-Doppler results are analyzed before (left) and after (right) interference signal suppression. Before the interference signal is suppressed, multi-source interference will cause severe linear or block artifacts on the entire spectrum, and weak targets are easily submerged. After reconstruction by the method of this invention, the background noise level is significantly reduced and the contrast between the target and the interference residue is greatly improved.

Claims

1. A target-protected mask-guided low-rank sparse reconstruction method for FMCW radar jamming suppression, characterized in that, Includes the following steps: Step 1: Acquire frequency modulated continuous wave (FMCW) radar observation signals and perform preprocessing; Step 2: Perform time-frequency transformation on the preprocessed FMCW radar observation signal and extract basic features; Step 3: Accurately locate the interference area in the time-frequency domain and form a binary mask prior. Based on the continuity characteristics of the target echo in the frequency domain, construct a mandatory target protection area. Step 4: Construct a low-rank sparse matrix and a target-protected low-rank sparse decomposition model based on mask prior guidance; Step 5: Solve the target protected low-rank sparse decomposition model using the inexact augmented Lagrange multiplier method to obtain the decomposition results; Step 6: Reconstruct and post-process the decomposition results, output the interference suppression results, and evaluate the effect.

2. The target protection mask prior-guided low-rank sparse reconstruction method for FMCW radar interference suppression according to claim 1, characterized in that, In step 1, the FMCW radar observation signal is the original intermediate frequency beat frequency signal acquired under single-frame or multi-frame conditions. It is a one-dimensional complex sampled signal of a single linear modulation pulse chirp or a two-dimensional complex matrix signal composed of multiple chirps. The matrix form expression of the FMCW radar observation signal is as follows: in, The observation matrix to be processed. Indicates that there is A fast-time sampling point, Indicates that there is One chirp; The preprocessing includes: ① Perform window function weighting, multiplying chirp by a Hamming window, Hann window, or Blackman window; ② The FMCW radar observation signal is normalized, and the expression is: in, It is a very small positive number. This is the normalized observation signal.

3. The target-protected mask prior-guided low-rank sparse reconstruction method for FMCW radar interference suppression according to claim 1, characterized in that, The FMCW radar observation signal is generated by superimposing the target echo component, background noise component, and interference component, and the expression is: in, For the target echo signal, This is an interference signal. It is Gaussian white noise.

4. The target-protected mask prior-guided low-rank sparse reconstruction method for FMCW radar interference suppression according to claim 1, characterized in that, The process of step 2 is as follows: Step 2.1: Perform a Short Time Fourier Transform (STFT) on each chirp of the preprocessed FMCW radar observation signal to obtain the corresponding time-frequency complex matrix. ; Step 2.2: Based on the time-frequency complex matrix Calculate the time-frequency amplitude matrix The calculation expression is: in, For indexing time windows, For frequency indexing.

5. The target-protected mask prior-guided low-rank sparse reconstruction method for FMCW radar interference suppression according to claim 1, characterized in that, The specific process of step 3 is as follows: Step 3.1: Search for and locate the target energy ridge. The specific steps are as follows: Step 3.1.1: Based on the physical characteristics of stationary or low-speed targets in FMCW radar appearing as horizontal high-energy ridges in the time-frequency diagram of a single snapshot frequency signal, energy accumulation is performed on the time-frequency amplitude matrix along the time axis. Step 3.1.2: Calculate the first... The sum of the cumulative energy at each frequency index is calculated using the following expression: in, This represents the total energy of each frequency row. Indicates the total number of time windows; Step 3.1.3: Determine the center frequency row of the useful target by searching for the maximum point in the energy accumulation sequence. ; Step 3.2: Calculate the adaptive threshold based on statistical quantiles The percentage threshold method is used to analyze the two-dimensional time-frequency amplitude spectrum. All pixel values ​​are sorted in ascending order, and a specific high quantile is selected. As the threshold for interference judgment ; Step 3.3: Generate an initial binary mask and perform morphological thinning. The specific process is as follows: Step 3.3.1: Determine the threshold based on the interference. Decisions are made on the time-frequency graph to generate a preliminary binary mask; Step 3.3.2: Compare the pixel capability value with a threshold value. When the threshold is reached, the mask value is set to 0, i.e., the reserved area; when the pixel capability value... When the threshold is reached, the mask value is set to 1, which is the interference removal area; Step 3.3.3: Perform a morphological dilation operation on the preliminary binary mask using specific structuring elements to obtain an interfering prior mask. ; Step 3.4: Construct a mandatory target protection zone. The specific process is as follows: Step 3.4.1: Define the guard band, using the center frequency from step 3.1.

3. Centered on the target area, set symmetrical frequency protection radius values ​​to define the protection range for the target area. ; Step 3.4.2: Perform a forced rewrite, forcibly setting the mask value at all time indices within the target area protection range to 1, thus obtaining the target protection mask. , which is represented as; in, The set frequency protection radius is symmetrical between the upper and lower sides; Step 3.5: Use the protected binary mask as the prior guiding weight for low-rank sparse decomposition.

6. The target protection mask prior-guided low-rank sparse reconstruction method for FMCW radar interference suppression according to claim 5, characterized in that, In step 3.3.3, the specific structural element is... A rectangular structural element.

7. The target protection mask prior-guided low-rank sparse reconstruction method for FMCW radar interference suppression according to claim 1, characterized in that, In step 4, the process of constructing a low-rank sparse matrix based on mask prior guidance is as follows: ①The interference prior mask and the target protection mask The fusion process is performed to construct prior constraints that guide RPCA decomposition; the target protection mask is then applied. After removing the interference support set, the corrected interference prior mask is obtained, and its expression is: in, Represents element-wise product; ② Construct a weighting matrix to control the sparsity penalty intensity in different regions. The expression for the weighting matrix is ​​as follows: in, These are the weighting coefficients for different regions, and they satisfy... This indicates that the occurrence of sparse components is encouraged in the interference area and suppressed in the target protection area. The process of constructing the target-protected low-rank sparse decomposition model is as follows: a. Transform the time-frequency amplitude matrix The time-frequency amplitude matrix is ​​decomposed into low-rank terms, sparse terms, and noise residual terms. The expression is: in, This indicates the low-rank portion that retains the main target structure and useful background information; This represents the sparse anomaly associated with the interference. This represents noise and modeling error; b. Construct a target-protected low-rank sparse decomposition model by using interference mask priors and target protection constraints. Its expression is: in, For nuclear norm, For weighted sparsity constraints, To provide additional penalty for sparse components appearing in non-interference regions, and To adjust the parameters.

8. The target protection mask prior-guided low-rank sparse reconstruction method for FMCW radar interference suppression according to claim 1, characterized in that, The specific process of step 5 is as follows: Step 5.1: Convert the time-frequency amplitude matrix It serves as input to a low-rank sparse decomposition model and decomposes it into low-rank terms. and sparse terms ; Step 5.2: Using the Lagrange multiplier matrix The augmented Lagrangian function is constructed using the penalty parameter, and its expression is: in, Indicates the inner product. Denotes the Frobenius norm. For penalty parameters; Step 5.3: The augmented Lagrange function is iteratively solved using the inexact augmented Lagrange multiplier method. The specific process is as follows: The low-rank term, sparse term, and multiplier term are initialized, and the following update steps are performed during the iteration: ① Update the lower-rank terms With the current sparse terms fixed and Lagrange multiplier matrix Under the condition of constructing the intermediate matrix The expression is: in, Indicates the first The next iteration; For the intermediate matrix Perform singular value decomposition and soft threshold shrink the singular values ​​to obtain the updated low-rank terms. ; ② Update the sparse terms Low-rank terms after fixed update and the current Lagrange multiplier matrix Under the condition of constructing the intermediate matrix The expression is: For the intermediate matrix Element-wise soft thresholding is performed to obtain the updated sparse terms. ; ③ Perform constraint correction on the target protection mask. The support region of the sparse terms is modified by updating the sparse terms and using a pre-constructed target protection binary mask. For positions in the target protection binary mask marked as target protection regions or reserved regions, the values ​​of the sparse terms at those positions are suppressed. For positions in the target protection binary mask marked as suspected interference regions, the corresponding components of the sparse terms are retained. ④ Update the multiplier terms After updating the low-rank and sparse terms, the Lagrange multiplier matrix is ​​updated. The expression for the updated Lagrange multiplier matrix is ​​as follows: in, This is the updated Lagrange multiplier matrix; ⑤ Update the penalty parameters After each iteration, the penalty parameter is adjusted. Perform incremental updates and record the updated penalty parameter as... ; ⑥ Perform convergence determination Calculate the reconstruction error between the sum of the current low-rank terms and sparse terms and the observation matrix. Set a preset error threshold. Compare the reconstruction error with the preset error threshold. Stop the iteration when the reconstruction error is lower than the preset error threshold or the change between two adjacent iterations is lower than the preset error threshold, and output the final low-rank term. and sparse terms .

9. A target-protected mask-guided low-rank sparse reconstruction method for FMCW radar interference suppression according to claim 8, characterized in that, In step 5.3, the initialization of the low-rank term, sparse term, and multiplier term involves setting the initial values ​​of the low-rank term and the sparse term to zero matrices, initializing the Lagrange multiplier matrix to a matrix of the same dimension as the FMCW radar observation matrix, and initializing the penalty parameter to a positive number.

10. The target-protected mask prior-guided low-rank sparse reconstruction method for FMCW radar interference suppression according to claim 1, characterized in that, In step 6, the reconstruction and post-processing are as follows: for low-rank terms Amplitude correction and normalization are performed, and the corrected time-frequency representation is then subjected to inverse STFT to recover the time-domain signal.

Images