An anti-interference control method applied to radar

Abstract

The present invention relates to the field of radar control technology, and in particular to an anti-interference control method applied to radar, which aims to solve the problem that the existing radar anti-interference technology has insufficient anti-interference ability in a complex electromagnetic environment. The method synchronously collects multi-dimensional input data of the radar system, uses a dual-channel deep network to extract interference features from the multi-dimensional data stream, and outputs interference fingerprint vectors to a dynamic feature library in real time. Based on the interference feature priority list, collaborative processing is performed, including spatial filtering, frequency domain notching, polarization projection, and residual extraction and sparse reconstruction of target signals. According to the real-time evaluation results, the processing parameters are dynamically adjusted. Data diversion processing is performed on suppression and deception interference, and the final target trajectory information is output through result fusion. The present invention also implements parallel processing on an FPGA chip to improve the processing speed. Overall, the present invention improves the anti-interference ability, stability and real-time performance of the radar system in a complex electromagnetic environment.

Description

Technical Field

[0001] The present invention relates to the field of radar control technology, and in particular to an anti-interference control method applied to radar. Background Art

[0002] In the field of radar anti-interference control, existing signal recognition algorithms are difficult to balance the real-time performance and recognition accuracy of interference suppression in a complex electromagnetic environment with dynamic time-varying. Existing technologies based on spectrum analysis, spatial filtering and adaptive beamforming methods have significant performance degradation when the interference signal and the target signal overlap in the frequency domain (such as DRFM deception interference), have similar spatial pointing (such as main lobe suppression interference) or have similar polarization characteristics (such as cross-polarization interference). When the interference intensity exceeds the dynamic range of the radar (typically >30dB suppression interference), conventional parameter estimation methods are prone to threshold effects, resulting in weak target signals being submerged. Existing anti-interference algorithms are not adaptable enough to complex interference, especially in scenarios where suppression interference and deception interference coexist. There is a lack of effective joint processing mechanism, which results in the inability of radar systems to be accurately controlled in complex electromagnetic confrontations, resulting in distortion of target information extraction, interruption of tracking trajectories, and generation of false targets. Summary of the invention

[0003] In view of the shortcomings existing in the prior art, the present invention provides an anti-interference control method applied to radar, which solves the problem that the control system of the prior art spectrum analysis and spatial domain filtering cannot realize the coordinated control of online identification of interference characteristics and fidelity extraction of target signals under multi-dimensional coupled interference of time, frequency and space, resulting in insufficient stability of anti-interference control and failure of dynamic compensation of target tracking error in a dynamic electromagnetic environment.

[0004] The present invention provides an anti-interference control method applied to a radar, comprising:

[0005] Step S101, synchronously collect multi-dimensional input data of the radar system, align and normalize the spatial guidance matrix, time-frequency distribution spectrum, polarization scattering parameters and dynamic environment perception data in time and space, and generate a standardized multi-dimensional data stream, wherein the dynamic environment perception data includes the azimuth of the interference source, the spectrum scanning result and the multipath propagation characteristics; Step S102, extract interference features from the multi-dimensional data stream through a dual-channel deep network, wherein the dual-channel deep network includes a 3D convolutional network of the first channel and a graph neural network of the second channel, outputs the interference fingerprint vector to the dynamic feature library in real time and updates the interference feature priority list, wherein the dynamic feature library refreshes the interference feature priority list with a period of 5 ms, and the priority is based on A sorting list is dynamically generated according to the power intensity and occurrence frequency of the interference signal; in step S103, collaborative processing is performed based on the interference feature priority list: spatial filtering, frequency notching and polarization projection are performed on the interference suppression data stream in sequence, and residual signal extraction and sparse reconstruction are performed in the target fidelity data stream; in step S104, spatial filtering matrix parameters, sparse dictionary configuration and data processing window length are dynamically adjusted according to the real-time evaluation results of the interference suppression ratio, target reconstruction error and tracking confidence; in step S105, data diversion processing is performed on the suppression interference and deceptive interference, and the final target trajectory information is output through parallel processing of frequency domain energy detection and multi-target association analysis, and weighted fusion of the results.

[0006] The advantages and beneficial effects of the present invention are:

[0007] The anti-interference control method applied to radar described in the present invention improves the anti-interference capability, stability and reliability, real-time and processing efficiency, adaptability and flexibility of the radar system through technical means such as multi-dimensional data fusion, dual-channel deep network feature extraction, collaborative processing mechanism, specific interference processing and result fusion, dynamic parameter adjustment mechanism, closed-loop control mechanism, FPGA chip parallel processing and rapid response capability, thereby providing strong support for the application of radar systems in complex electromagnetic environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] Figure 1 The present invention is a flow chart of an anti-interference control method applied to radar. DETAILED DESCRIPTION

[0009] An embodiment of the present invention is further described below with reference to the accompanying drawings.

[0010] See also Figure 1 The present invention provides an anti-interference control method applied to a radar, comprising:

[0011] Step S101, synchronously collecting multi-dimensional input data of the radar system, performing spatiotemporal alignment and normalization processing on the spatial steering matrix, the time-frequency distribution spectrum, the polarization scattering parameters and the dynamic environment perception data, and generating a standardized multi-dimensional data stream, wherein the dynamic environment perception data includes the azimuth of the interference source, the spectrum scanning result and the multipath propagation characteristics;

[0012] Spatial steering matrix: generated by the phase response data of each element of the radar antenna array, reflecting the directional characteristics of the signal in space.

[0013] Time-frequency distribution spectrum: generated by short-time Fourier transform (STFT), it captures the time-varying frequency characteristics of the signal and is used to analyze the modulation characteristics of the interference signal (such as frequency hopping and frequency sweeping).

[0014] Polarization scattering parameters: including the measurement values ​​of four polarization channels: HH (horizontal transmission-horizontal reception), HV (horizontal transmission-vertical reception), VH (vertical transmission-horizontal reception), and VV (vertical transmission-vertical reception), which are used to distinguish the polarization difference between the target and the interference.

[0015] Dynamic environment perception data: covers the azimuth of the interference source (obtained through wave direction estimation), spectrum scanning results (real-time spectrum energy distribution), and multipath propagation characteristics (such as delay spread and Doppler spread), which are used to evaluate the dynamic changes of the electromagnetic environment.

[0016] Spatiotemporal alignment and normalization:

[0017] Alignment mechanism: Use timestamp synchronization technology to ensure the temporal consistency of spatial, time-frequency, polarization and environmental perception data; unify spatial pointing parameters (such as azimuth and elevation) through coordinate transformation.

[0018] Normalization method: normalize the amplitude of multidimensional data (such as normalizing the power to the range of 0~1) and standardize the dimensions (such as mapping the spectrum energy to a fixed number of frequency points) to eliminate dimensional differences and form a standardized data stream. Step S102, extract interference features from the multidimensional data stream through a dual-channel deep network, wherein the dual-channel deep network includes a 3D convolutional network of the first channel and a graph neural network of the second channel, the first channel is a 3D convolutional network, containing a 7×7×3 convolution kernel; the second channel is a graph neural network, based on the graph structure constructed by the interference source node and the environment node, outputs the interference fingerprint vector to the dynamic feature library in real time and updates the interference feature priority list, the dynamic feature library refreshes the interference feature priority list with a period of 5ms, and the priority is dynamically generated in a sorted list based on the power intensity and frequency of occurrence of the interference signal;

[0019] Dual-channel deep network interference feature extraction:

[0020] Dual-channel network division of labor:

[0021] First channel (3D convolutional network):

[0022] Input data: joint time-frequency-space data block (three-dimensional structure, including time, frequency, and space dimensions).

[0023] Function: Extract the modulation pattern characteristics of the interference signal, for example:

[0024] DRFM phase jump characteristics: Identify the phase mutation pattern of digital radio frequency memory (DRFM) deceptive interference.

[0025] Spatial pointing deviation angle: detects the pointing difference between the interference signal and the radar main lobe.

[0026] Polarization mismatch parameter: Quantifies the energy difference between the interference and target polarization channels.

[0027] Second channel (graph neural network):

[0028] Input data: graph structure constructed by interference source nodes (azimuth, spectrum characteristics) and environment nodes (multipath propagation paths).

[0029] Function: Model the propagation effects of interference signals in complex electromagnetic environments (such as multipath diffusion and reflection attenuation) and predict the dynamic propagation path of interference.

[0030] Dynamic feature library update mechanism:

[0031] Interference fingerprint vector: The feature vector output by the dual-channel network that represents the unique identity of the interference (such as modulation type, spatial position, and polarization state).

[0032] Priority list: dynamically sorted according to the threat level of the interference signal (such as power intensity, frequency of occurrence), and refreshed every 5ms. Step S103, perform collaborative processing based on the interference feature priority list: perform spatial domain filtering, frequency domain notching and polarization projection on the interference suppression data stream in sequence, and perform residual signal extraction and sparse reconstruction in the target fidelity data stream;

[0033] Co-processing and signal separation:

[0034] Interference suppression data stream processing:

[0035] Spatial filtering: Generates adaptive beamforming weight vectors based on the interference feature library, forms a null in the interference direction (the depth is proportional to the interference power), and suppresses main lobe / side lobe interference.

[0036] Frequency domain notching: Design a band-stop filter group according to the interference spectrum characteristics, notch the interference frequency band, and reconstruct the target Doppler frequency shift signal (retain the characteristics of the moving target).

[0037] Polarization projection: Calculate the polarization difference between the interference and the target, select the direction with the maximum difference (such as the interference is horizontal polarization and the target is vertical polarization) for polarization projection to achieve signal separation.

[0038] Target Fidelity Data Stream Processing:

[0039] Residual signal extraction: Extract the residual target signal not covered by interference from the suppressed data stream.

[0040] Sparse reconstruction: A dynamic dictionary learning algorithm is used to match the scattering basis functions of the target (such as point scattering models and micro-Doppler characteristics of rotating components) to reconstruct high-fidelity target signals.

[0041] False target elimination: The reconstructed signal is temporally and spatially correlated with the historical target trajectory library, and the false trajectory is eliminated by using Kalman filter residual analysis (such as the deviation between the predicted position and the actual position). Step S104, dynamically adjust the spatial domain filter matrix parameters, sparse dictionary configuration and data processing window length according to the real-time evaluation results of the interference suppression ratio, target reconstruction error and tracking confidence;

[0042] Dynamic parameter adjustment mechanism:

[0043] Adjust trigger conditions and strategies:

[0044] Spatial filter matrix reconfiguration: When the interference suppression ratio drops by more than 15% of the historical average (indicating a change in interference direction or intensity), the beamforming weight vector is urgently optimized.

[0045] Sparse dictionary update: If the target reconstruction error exceeds the threshold for three consecutive frames (such as error > 10%), online dictionary learning is started and the scattering basis function is updated to adapt to the dynamic characteristics of the target.

[0046] Processing window dynamic adjustment:

[0047] High confidence (stable tracking): shorten the window to 10ms to improve real-time performance.

[0048] Low confidence (complex environment): Extend the window to 50ms and enhance data accumulation to improve the signal-to-noise ratio.

[0049] Closed-loop feedback mechanism:

[0050] Error covariance matrix: Calculate the covariance of the reconstructed signal and the original measurement data in real time. If the eigenvalue exceeds the threshold (indicating that the system error increases), adjust the deep network parameters in reverse (such as 3D convolution kernel size optimization, graph neural network edge weight update) to form a closed-loop control. Step S105: Perform data diversion processing on suppression interference and deceptive interference, perform parallel processing of frequency domain energy detection and multi-target association analysis, and perform weighted fusion on the results to output the final target trajectory information.

[0051] Fusion of disturbance flow and target trajectory:

[0052] Suppressive interference processing:

[0053] Frequency domain energy detection: Identify high-power interference frequency bands and perform frequency domain notching.

[0054] Spatial null widening: Expand the null width near the interference direction to suppress wide-beam interference.

[0055] Time domain pulse shielding: Time domain gating shielding of periodic interference pulses.

[0056] Deceptive interference processing:

[0057] Multi-target association analysis: Detect the inconsistency of distance and speed of false targets (such as sudden changes in speed and position). The distance-speed consistency check refers to calculating the target radial velocity through Doppler frequency shift and verifying its physical consistency with the target distance change rate.

[0058] Consistency check: By comparing historical target trajectories, false trajectories that do not conform to the laws of motion are eliminated.

[0059] Result fusion and target trajectory output:

[0060] Target trajectory probability fusion algorithm: The detection results of the suppression interference branch (focusing on frequency domain features) and the deception interference branch (focusing on time domain features) are weighted and fused to output high-confidence target trajectory information.

[0061] Specifically, the anti-interference control method applied to radar according to the present invention, step S101 includes:

[0062] The spatial steering matrix is ​​composed of phase response data of the radar antenna array;

[0063] The time-frequency distribution spectrum is generated by short-time Fourier transform;

[0064] The polarization scattering parameters include HH, HV, VH, and VV polarization channel measurement values;

[0065] The dynamic environment perception data includes interference source azimuth, spectrum scanning results and multipath propagation characteristics.

[0066] This step (S101) is the core link of data preprocessing of the radar anti-interference control method, which provides high-quality input for subsequent interference feature extraction and collaborative control through multi-source data collection and standardized processing.

[0067] Data source and generation method:

[0068] Spatial Steering Matrix:

[0069] Generation principle: Based on the phase response data of each element of the radar antenna array, the spatial steering vector is calculated through the direction of arrival (DOA) estimation algorithm (such as the MUSIC algorithm or Capon beamforming).

[0070] Technical function: It reflects the propagation direction characteristics of the signal in space and is used for accurate positioning of the interference direction in subsequent spatial filtering.

[0071] Implementation example:

[0072] Array structure: Uniform Linear Array (ULA) or Planar Array.

[0073] Phase calibration: The built-in calibration source eliminates phase errors between array elements to ensure the accuracy of the steering matrix.

[0074] Time-frequency distribution graph:

[0075] Generation method: Short-time Fourier transform (STFT) is used to intercept signal fragments through a sliding time window and perform Fourier transform to generate a time-frequency energy distribution spectrum.

[0076] Parameter settings:

[0077] Window function: Hamming window (balance main lobe width and side lobe attenuation).

[0078] Window length and overlap rate: The window length is 256 sampling points and the overlap rate is 50% to take into account both time domain resolution and spectrum continuity.

[0079] Technical function: Capture the time-varying spectrum characteristics of interference signals (such as frequency sweeping and frequency hopping) and provide time-frequency-space joint analysis data for the 3D convolutional network.

[0080] Polarization scattering parameters:

[0081] Data source: The echo signals of four polarization channels, HH (horizontal transmission-horizontal reception), HV (horizontal transmission-vertical reception), VH (vertical transmission-horizontal reception), and VV (vertical transmission-vertical reception), are measured through a fully polarized radar system.

[0082] Technical role:

[0083] Distinguishing targets from interference: Target scattering characteristics usually have a stable polarization response, while interference may exhibit abnormal parameters due to polarization mismatch (such as cross-polarization interference).

[0084] Implementation example: Calculate the asymmetry of HV and VH channels and identify the polarization modulation characteristics of artificial interference.

[0085] Dynamic environment perception data:

[0086] Interference source azimuth: The interference source angle (azimuth ± elevation) is estimated in real time through the spatial steering matrix combined with the DOA algorithm.

[0087] Spectrum scanning results: Use a wideband receiver to quickly scan the radar operating frequency band and generate a spectrum energy distribution histogram.

[0088] Multipath propagation characteristics:

[0089] Extraction method: Through channel impulse response (CIR) analysis, multipath delay spread, Doppler spread and path attenuation coefficient are obtained.

[0090] Technical role: Assess the complexity of the electromagnetic environment and provide a basis for modeling interference propagation paths for graph neural networks.

[0091] Space-time alignment mechanism:

[0092] Time Synchronization:

[0093] Hardware triggering: Use a unified clock source (such as GPS disciplined clock) to synchronize the radar subsystems (array, polarization channel, spectrum scanning module) to ensure that the data acquisition time deviation is ≤1μs.

[0094] Software compensation: Time alignment of asynchronous data (such as environmental perception data) is performed through interpolation algorithms.

[0095] Spatial Alignment:

[0096] Coordinate system 1: Map the azimuth angle of the spatial steering matrix, the frequency points of the spectrum scan, and the polarization channel data to the same spatial reference system (such as the radar local coordinate system).

[0097] Example: Convert the time delay information of multipath propagation paths into spatial distance and associate it with the azimuth of the interference source.

[0098] Normalization processing strategy:

[0099] Amplitude normalization:

[0100] Method: The amplitude of each data source is normalized to the maximum and minimum, the complex phase response of the spatial steering matrix is ​​mapped to the interval [0,1], and the polarization channel energy is normalized to unit power.

[0101] Function: Eliminate the gain differences of hardware channels and avoid convergence problems caused by inconsistent dimensions in subsequent network training.

[0102] Dimension normalization:

[0103] Spatial data: Compress the element dimensions of the steering matrix into a fixed length (such as a 16×16 matrix).

[0104] Time-frequency data: The resolution of the time-frequency diagram output by STFT is unified to 256×256 pixels, and the insufficient part is padded with zeros.

[0105] Environmental data: The spectrum scanning results are discretized into 128 frequency points, and the multipath characteristics are encoded into a 4-dimensional vector (delay, Doppler, attenuation, and number of paths).

[0106] Specifically, the anti-interference control method applied to radar according to the present invention, step S102 includes:

[0107] The first channel uses a 3D convolutional network to extract the modulation mode of the interference signal from the joint time-frequency-space data block. The modulation mode includes the phase jump feature of the digital RF memory, and the phase jump feature is extracted by the time axis gradient detection algorithm. The second channel constructs an interference-environment association model through a graph neural network to analyze the multipath diffusion effect of the electromagnetic environment on interference propagation. The dynamic feature library refreshes the interference feature priority list with a period of 5 ms.

[0108] This step (S102) realizes dynamic extraction and modeling of interference features through a dual-channel deep network architecture. Its innovation lies in combining the time-frequency and spatial domain signal characteristics with the correlation characteristics of the electromagnetic environment to build a closed-loop updated interference feature library.

[0109] First channel: 3D convolutional network time-frequency-space feature extraction:

[0110] Input data structure:

[0111] Time-frequency-space joint data block: A three-dimensional data cube formed by splicing the standardized multi-dimensional data stream (time-frequency distribution spectrum, spatial domain steering matrix, polarization parameters) generated in step S101 according to a time window (such as 20ms), and the dimensions include:

[0112] Time axis: signal segment time series (such as 10 frames of continuous data).

[0113] Frequency axis: The number of frequency points generated by short-time Fourier transform (e.g. 256 points).

[0114] Spatial axis: the number of radar antenna elements (such as 16 elements) or the result of spatial grid division.

[0115] Network structure and functions:

[0116] 3D convolution layer design:

[0117] First layer: Large-size convolution kernels (such as 7×7×3) extract coarse-grained features in time, frequency and space to capture the global modulation rules of the interference signal (such as scanning period and spatial pointing trend).

[0118] Second layer: Small-sized convolution kernels (such as 3×3×1) focus on local features and identify details such as DRFM phase jumps (sudden changes in the time axis) and polarization mismatches (energy differences in polarization channels).

[0119] Feature output:

[0120] DRFM phase jump feature: quantify the phase mutation amplitude and interval period through time axis gradient detection.

[0121] Spatial pointing deviation angle: Calculate the angle between the interference signal direction and the radar main lobe direction (error < 0.5°).

[0122] Polarization mismatch parameters: output HH-VV channel energy ratio, HV-VH asymmetry and other indicators.

[0123] Implementation example:

[0124] For DRFM deception interference, the 3D convolutional network can identify the periodic phase jumps caused by its repeated forwarding (e.g., once every 5ms);

[0125] The interference of the main lobe is suppressed, and the spatial separation between the main lobe and the target is detected through the spatial pointing deviation angle, which provides a basis for the spatial nulling in step S103.

[0126] Second channel: Graph neural network environment association modeling:

[0127] Graph structure construction:

[0128] Node definition:

[0129] Interference source node: contains attributes such as azimuth, spectrum energy, and polarization state.

[0130] Environmental nodes: including parameters such as multipath propagation path (delay, attenuation), terrain reflectivity, and meteorological conditions (such as rainfall attenuation).

[0131] Edge definition:

[0132] Propagation relationship edge: The connection weight from the interference source node to the environment node, indicating the multipath reflection intensity (such as -20dB attenuation).

[0133] Environmental coupling edge: the interaction between environmental nodes (such as multipath superposition caused by building reflection).

[0134] Network training and inference:

[0135] Analysis of multipath diffusion effect:

[0136] Input: interference source azimuth, spectrum characteristics and environmental perception data.

[0137] Output: Predict the multipath propagation path of the interference signal (such as 3 reflection paths) and its delay-Doppler spread parameters.

[0138] Dynamic weight update:

[0139] According to real-time environmental changes (such as the emergence of new reflectors), the graph edge weights are adaptively adjusted and the interference propagation model is updated.

[0140] Technical relevance:

[0141] The output result is used for frequency domain notch design (suppressing multipath interference frequency band) in step S103 and dynamic window adjustment (extending the processing window in complex environment) in step S104.

[0142] Dynamic feature library and priority mechanism:

[0143] Interference fingerprint vector generation:

[0144] Data fusion: Encode the dual-channel output features (modulation mode of 3D convolutional network + multipath diffusion parameters of graph network) into a 128-dimensional feature vector as a unique identifier of interference.

[0145] Priority list update logic:

[0146] Sort by:

[0147] Threat level: Interference signal power (>30dBm is high priority).

[0148] Occurrence frequency: the number of times interference is repeated per unit time (e.g. 10 times / second raises to high priority).

[0149] Timeliness: New interference will have a higher initial priority, and will be downgraded if it continues to go undetected.

[0150] Refresh mechanism:

[0151] The feature library is sorted and updated once every 5 ms to ensure that high-priority interferences enter the collaborative processing flow of step S103 first.

[0152] Implementation example:

[0153] When high-power suppression interference (priority TOP1) is detected, the spatial null depth of step S103 is increased (e.g., -40dB suppression);

[0154] For low-frequency deception interference (lower priority), default parameters are used to save computing resources.

[0155] Technical advantages:

[0156] Dual-channel synergy advantages:

[0157] Complementarity: 3D convolutional networks are good at extracting intrinsic signal features (such as modulation patterns), while graph neural networks are good at modeling external environmental influences (such as multipath effects). The combination of the two improves the robustness of interference identification.

[0158] Real-time performance: Dual-channel parallel processing (3D convolution calculation and graph reasoning are executed asynchronously) meets the requirement of updating the feature library in a 5ms cycle.

[0159] Closed-loop feedback design:

[0160] The error covariance matrix detection result of step S104 can reversely trigger the optimization of the 3D convolution kernel size (such as increasing the convolution kernel to capture long-period interference), forming an adaptive learning mechanism.

[0161] Hardware compatibility:

[0162] 3D convolution calculations can be accelerated in parallel through FPGA (e.g. 16 PE units process data blocks in parallel);

[0163] The edge weight update of graph neural network adopts incremental calculation to reduce hardware resource consumption.

[0164] Application scenario examples:

[0165] Dense multipath environment (such as complex urban environment):

[0166] The graph neural network analyzes the reflection path of the building and collaborates with the 3D convolutional network to identify the DRFM jump characteristics of deceptive interference. The dynamic feature library marks the multipath interference as a high priority, triggering the joint processing of frequency domain spectrum line reconstruction and polarization projection in step S103.

[0167] Composite interference scenario (suppression + deception coexistence):

[0168] The 3D convolutional network separates the spatial pointing deviation angle (>5°) of the suppressed interference and the polarization mismatch parameter (HV / VH ratio>3dB) of the deceptive interference. The dynamic feature library assigns independent priorities to the two types of interference. Step S105 implements data diversion processing.

[0169] Through the collaborative design of 3D convolutional networks and graph neural networks, the joint modeling of the intrinsic characteristics of interference signals and the external propagation environment is achieved.

[0170] Specifically, the anti-interference control method applied to radar described in the present invention, the interference suppression data stream processing in step S103 includes: spatial domain processing, generating a beamforming weight vector according to an interference feature library, and implementing null-pointing control on the original data, wherein the null-pointing depth is proportional to the power of the interference signal; frequency domain processing, generating a band-stop filter group based on the interference spectrum characteristics, reconstructing the spectral lines of overlapping frequency band signals, and retaining the target Doppler frequency shift characteristics; polarization processing, calculating the polarization difference between the interference and the target, and selecting the polarization projection operator in the direction of the maximum difference to perform signal separation.

[0171] This step (S103) achieves a balance between interference suppression and target fidelity through a multi-dimensional collaborative processing mechanism in the spatial domain, frequency domain, and polarization.

[0172] Airspace Processing: Adaptive Null Control:

[0173] Beamforming weight vector generation:

[0174] Input basis: based on the interference fingerprint vector (including interference azimuth and power level) in the dynamic feature library in step S102.

[0175] Algorithm selection: The minimum variance distortionless response (MVDR) algorithm is used to calculate the optimal weight vector by inverting the covariance matrix. The formula logic is:

[0176] Goal: Create a null in the direction of the jammer while keeping the main lobe pointed toward the target.

[0177] Constraints: The null depth is positively correlated with the interference power (e.g., for every 10dB increase in interference power, the null depth increases by 6dB).

[0178] Implementation example:

[0179] For high power suppression interference (such as 40dBm), a deep null (-40dB) is generated to suppress interference while avoiding main lobe distortion.

[0180] For low-power interference (such as 20dBm), a shallow null (-15dB) is used to reduce signal distortion.

[0181] Dynamic Pointing Control:

[0182] Real-time update mechanism: When the interference azimuth changes by more than 0.5° (detected by the feature library in step S102), the weight vector is recalculated.

[0183] Anti-disturbance design: For fast-moving interference (such as airborne interference sources), a sliding average filter is used to smooth the azimuth estimate to avoid zero-sink jitter.

[0184] Frequency domain processing: spectral line reconstruction and feature preservation:

[0185] Bandstop filter bank design:

[0186] Parameter configuration:

[0187] Stopband position: set according to the interference spectrum characteristics (such as center frequency, bandwidth) extracted in step S102.

[0188] Stopband width: Dynamically expands to 1.2 times the interference bandwidth to cover frequency diffusion (such as spectrum broadening caused by multipath).

[0189] Transition band optimization: Kaiser window is used to design the filter to ensure that the passband (target Doppler shift area) ripple is ≤0.1dB.

[0190] Spectral line reconstruction technology:

[0191] Target signal recovery:

[0192] For overlapping frequency band signals (the target and interference frequency domains overlap), the residual signal after interference suppression is used to perform inverse Fourier transform to reconstruct the time domain waveform.

[0193] Preserve Doppler characteristics: Compensate for the group delay introduced by filtering through phase interpolation to ensure that the target velocity information is not distorted.

[0194] Implementation example:

[0195] After notching the interference frequency band (e.g., 9.8-10.2 GHz), the complete spectrum is retained for the target Doppler frequency shift region (e.g., ±500 Hz).

[0196] Polarization processing: Difference maximization projection:

[0197] Polarization difference calculation:

[0198] Quantification method:

[0199] The polarization covariance matrix of the target and the interference is calculated, and the eigenvalue difference ratio is extracted (e.g., the target HV / VH ratio = 1.2, the interference HV / VH ratio = 3.5).

[0200] Difference threshold: Set the polarization projection to start when the difference is greater than 3dB, otherwise skip it to save resources.

[0201] Dynamic selection logic:

[0202] If the interference is horizontal polarization (H), select vertical polarization (V) direction projection;

[0203] If the interference is circularly polarized, choose an elliptical polarization axis projection to maximize the difference.

[0204] Signal separation implementation:

[0205] Projection operator generation: A polarization filter is constructed based on the direction of maximum difference, and the original signal is projected to suppress interference components.

[0206] Fidelity design: The integrity of the polarization scattering matrix of the projected signal is preserved to ensure that subsequent target identification (such as aircraft model classification) is not affected.

[0207] Multi-dimensional collaborative processing mechanism:

[0208] Parallel processing flow:

[0209] Data splitting: The original data is simultaneously input into the spatial domain, frequency domain, and polarization processing modules, and the output results are weighted and fused to generate suppressed signals.

[0210] Weight distribution: Dynamically adjusted according to interference type:

[0211] Suppression interference: airspace weight 70%, frequency domain 20%, polarization 10%;

[0212] Deception interference: polarization weight 50%, spatial domain 30%, frequency domain 20%.

[0213] Dynamic resource scheduling:

[0214] For high priority interference (such as TOP1 interference marked in step S102), spatial nulling and polarization projection resources are preferentially allocated;

[0215] For minor interferences, only frequency domain notching is performed to reduce the computational load.

[0216] Technical effects and implementation relevance:

[0217] Improved anti-interference performance:

[0218] The joint suppression of spatial-frequency-polarization domain improves the signal-to-interference ratio (SIR) by ≥25dB ​​(measured data), which is better than the traditional method of a single dimension (such as spatial filtering alone improves by 10-15dB).

[0219] The Doppler frequency shift retention rate is >95%, ensuring the accuracy of moving target tracking.

[0220] Connection with upstream and downstream steps:

[0221] Input dependency: The interference signature library provided in step S102 drives the dynamic configuration of processing parameters;

[0222] Output connection: The suppressed data stream is input into the target fidelity branch of step S103 for residual reconstruction, and at the same time provides a reference signal for the evaluation of step S104.

[0223] Hardware compatibility:

[0224] Spatial beamforming weight vector calculation is realized through FPGA parallelization, with a latency of <1ms;

[0225] The polarization projection operator uses a lookup table (LUT) to store commonly used polarization modes to reduce the amount of real-time calculations.

[0226] Implementation Notes:

[0227] Parameter calibration:

[0228] The airspace null depth needs to be calibrated according to the radar dynamic range to avoid excessive suppression leading to target signal loss;

[0229] The polarization difference threshold needs to be dynamically adjusted according to external factors such as ambient humidity and antenna cover loss.

[0230] Exception handling:

[0231] When the interference and target polarizations are completely consistent (difference < 1 dB), the multi-target correlation analysis in step S105 is triggered to replace the polarization projection.

[0232] Through the spatial-frequency-polarization multi-dimensional coordinated suppression mechanism, the technical bottleneck of traditional methods that lose sight of one thing while focusing on another in complex interference scenarios is solved. Its core innovation lies in:

[0233] Dynamic weight allocation: adaptively adjust the resource ratio of processing modules according to interference type and priority;

[0234] Feature-fidelity design: maximally retains the target Doppler and polarization characteristics while suppressing interference;

[0235] Specifically, in the anti-interference control method applied to radar described in the present invention, the target fidelity data stream processing in step S103 includes: using a dynamic dictionary learning algorithm to match the target scattering basis function, the basis function includes a point scattering model, a sliding scattering model and a micro-Doppler characteristic of a rotating component; the reconstructed signal is temporally and spatially associated with the historical tracking data in the target trajectory library, and false targets are eliminated through Kalman filter residual analysis.

[0236] This step (target fidelity data stream processing in S103) achieves high-fidelity reconstruction of the target signal and elimination of false trajectories after interference suppression through dynamic dictionary learning and spatiotemporal correlation verification technology.

[0237] Dynamic dictionary learning algorithm:

[0238] Base function library construction and matching mechanism:

[0239] Basis function type:

[0240] Point scattering model: Applicable to stationary or uniformly moving targets (such as ground vehicles), modeled as scattering centers at fixed positions.

[0241] Sliding scattering model: describes the scattering characteristics of the target's translational motion (such as an airplane flying level), and characterizes the continuous displacement through time-varying position parameters.

[0242] Micro-Doppler signature of rotating parts: Captures the periodic motion of target parts (e.g. helicopter rotors, low-altitude aircraft propellers) and models it as a frequency modulated signal (micro-Doppler sidebands).

[0243] Dynamic update mechanism:

[0244] Dictionary initialization: preload typical target basis functions (such as 10 types of aerial targets and 5 types of ground targets).

[0245] Online Learning: Based on real-time radar echoes, basis function weights are optimized through sparse coding to add environment-specific scattering patterns (such as unique micro-motion features of newly emerging targets).

[0246] Signal reconstruction process:

[0247] Sparse representation: Decompose the residual signal after interference suppression into a linear combination of basis functions, and select the 3-5 basis functions with the strongest energy for reconstruction.

[0248] Fidelity Design:

[0249] Preserve the phase continuity of the target scattering matrix to avoid distortion of the reconstructed signal;

[0250] Time-frequency analysis and calibration are used for micro-Doppler characteristics to ensure the integrity of frequency modulation characteristics.

[0251] Implementation example:

[0252] Scenario 1: For helicopter targets, match the basis functions of rotating components and reconstruct the micro-Doppler spectrum corresponding to the rotor speed (e.g. 20 Hz);

[0253] Scenario 2: For high-speed moving targets, the sliding scattering model is matched first to extract its acceleration parameters (such as 50m / s²).

[0254] Spatiotemporal correlation and false target elimination:

[0255] Space-time correlation logic:

[0256] Data mapping: Align the timestamp and spatial coordinates (range-azimuth) of the reconstructed signal with the historical data in the target trajectory library:

[0257] Time window: associate the tracking data of the last 5 frames (50ms);

[0258] Spatial tolerance: distance tolerance ±50m, angle tolerance ±1° (adjusted according to radar resolution).

[0259] Association rules:

[0260] Continuous association: The current signal must match the historical target trajectory continuously within 3 frames, otherwise it will be marked as a suspicious target;

[0261] Motion consistency: speed change rate <10m / s³, acceleration <50m / s² (exceeding this will trigger anomaly detection).

[0262] Kalman filter residual analysis:

[0263] Prediction-correction mechanism:

[0264] State prediction: predict the current target position and speed based on historical target trajectories;

[0265] Residual calculation: the deviation between the actual measured position and the predicted position (e.g. lateral deviation > 10m);

[0266] False target determination:

[0267] Single frame judgment: If the residual exceeds 3 times the measurement noise variance (e.g. σ²=5m²), it is marked as a potential false target;

[0268] Continuous judgment: If the residuals of two consecutive frames are abnormal and there is no micro-Doppler feature support, the trajectory will be eliminated.

[0269] Implementation example:

[0270] Real target: high-speed moving target flying at a constant speed, residual fluctuation range ±3m, micro-Doppler characteristics matching the engine speed;

[0271] False target: DRFM deception interference generates false tracks, the residual suddenly jumps to 15m, and there is no micro-Doppler support, which triggers elimination.

[0272] Technical synergy and advantages:

[0273] Cooperation with the interference suppression branch:

[0274] Input dependency: The target fidelity processing uses the residual signal after interference suppression in step S103 as input to ensure data purity;

[0275] Feedback mechanism: Reconstruction error (such as basis function matching failure rate>30%) reversely triggers the sparse dictionary update in step S104.

[0276] Technical advantages:

[0277] High-fidelity reconstruction: Dynamic dictionary learning improves the reconstruction accuracy of weak targets (SNR>6dB) by ≥40%;

[0278] Anti-spoofing interference: Spatiotemporal correlation and micro-Doppler verification can identify more than 99% of false trajectories without physical motion characteristics;

[0279] Resource optimization: Basis functions are dynamically loaded according to priority (such as giving priority to matching typical target models in highly dynamic and complex environments) to reduce computational redundancy.

[0280] Implementation Notes:

[0281] Parameter calibration:

[0282] The basis function library needs to be calibrated according to the radar wavelength (e.g., X-band radar needs to optimize micro-Doppler resolution);

[0283] The Kalman filter process noise parameters need to be dynamically adjusted with the target maneuverability.

[0284] Exception handling:

[0285] When the matching degree between the reconstructed signal and all basis functions is less than 60%, the manual review process is triggered to avoid missed alarms;

[0286] For sudden high-maneuver targets (such as evasive actions), temporarily relax the motion consistency threshold (such as increasing the acceleration threshold to 100m / s²).

[0287] Hardware Adaptation:

[0288] Dynamic dictionary learning achieves parallel matching through FPGA (e.g. 4 logic units synchronously process different basis function categories);

[0289] The spatiotemporal correlation algorithm adopts a pipeline design with a delay of less than 2ms, meeting real-time requirements.

[0290] This step solves the problem of target signal distortion and false trajectory residue after interference suppression in traditional methods through the dual mechanism of dynamic dictionary learning and spatiotemporal correlation verification. Its core innovation lies in:

[0291] Adaptive basis function matching: compatible with the static, translational and micro-motion characteristics of complex targets, improving reconstruction accuracy;

[0292] Multi-dimensional verification: Combine time-space correlation, motion logic and physical characteristics (micro-Doppler) to eliminate false targets;

[0293] Closed-loop optimization: Reconstruction error is fed back to the upstream step to drive dynamic adjustment of system parameters.

[0294] Those skilled in the art can adjust the basis function library and association rules according to the radar application scenarios (such as early warning radar focusing on micro-motion features and fire control radar enhancing motion consistency verification) to achieve optimal anti-interference performance.

[0295] Specifically, the anti-interference control method applied to radar according to the present invention, step S104 includes:

[0296] When it is detected that the interference suppression ratio drops by more than 15% compared with the historical average, the emergency reconfiguration of the spatial domain filter matrix is ​​triggered; if the target reconstruction error exceeds the preset threshold for three consecutive frames, the online update process of the sparse dictionary is started; the data processing window length is dynamically adjusted according to the tracking confidence level: the window is shortened to 10ms when the confidence is high, and extended to 50ms when the confidence is low.

[0297] This step (S104) realizes closed-loop feedback adjustment of processing parameters by dynamically monitoring the interference suppression effect and target reconstruction accuracy, thereby ensuring the adaptability and robustness of the radar system in a complex electromagnetic environment.

[0298] Emergency reconfiguration of the spatial filter matrix:

[0299] Trigger conditions and detection logic:

[0300] Interference Suppression Ratio (JSR) definition:

[0301] Calculation formula: JSR = (power after interference suppression / original interference power) × 100%, which represents the interference suppression effect.

[0302] Historical mean calculation: Take the JSR sliding average of the most recent 50 frames (500ms) as the benchmark.

[0303] Falling threshold judgment: When the real-time JSR is 15% lower than the historical average (for example, the historical average is 80%, and the current value is ≤ 68%), reconfiguration is triggered.

[0304] Reconfiguration implementation process:

[0305] Covariance matrix update: recalculate the spatial covariance matrix based on the latest interference features (output of step S102) to improve the interference direction estimation accuracy.

[0306] Weight vector optimization: Use robust adaptive beamforming algorithms (such as diagonal loading) to reduce signal distortion while suppressing interference.

[0307] Dynamic adjustment of null depth: Set the suppression strength according to the current interference power (e.g. 30dBm → null depth -25dB) and type (suppression / deception).

[0308] Implementation example:

[0309] Scenario 1: The external interference source releases sweep frequency suppression interference, causing the JSR to drop sharply to 65%, triggering the spatial filter matrix reconfiguration, and the null depth deepened to -30dB;

[0310] Scenario 2: Multipath interference causes JSR fluctuations, and sliding average filtering is used to eliminate false triggering of instantaneous disturbances.

[0311] Sparse dictionary online update process:

[0312] Error monitoring and update trigger:

[0313] Preset threshold setting: dynamically set according to radar resolution and target type (e.g. high-speed moving target error threshold is ±5m, low-altitude aircraft ±2m).

[0314] Continuous exceeding of the threshold: The target reconstruction error (deviation between the measured position and the predicted position) exceeds the threshold for 3 consecutive frames (30ms) (e.g. high-speed moving target>5m).

[0315] Dynamic dictionary learning mechanism:

[0316] Basis function expansion: Add target scattering characteristics with current significant errors (such as the micro-Doppler characteristics of new stealth coatings) to the basis function library.

[0317] Weight optimization: The sparse coding algorithm is used to adjust the basis function combination weights to prioritize matching of high-frequency target patterns.

[0318] Implementation example:

[0319] For low-altitude aircraft that suddenly appear (reconstruction error > 3m for 3 consecutive frames), high-resolution micro-motion basis functions are added to reduce the error to within 1m.

[0320] Connection with step S103:

[0321] The updated dictionary is used in real time to reconstruct the residual signal of the target fidelity data stream, forming a closed-loop optimization link.

[0322] Dynamic adjustment of data processing window:

[0323] Tracking confidence grading criteria:

[0324] High confidence (level 1):

[0325] Residual fluctuation < measurement noise variance (e.g. σ²=4m²);

[0326] The historical target trajectory is continuous for 10 frames without interruption;

[0327] The matching degree between micro-Doppler characteristics and target type is >90%.

[0328] Low confidence (level 3):

[0329] Residual fluctuation > 3σ²;

[0330] The target trajectory is intermittent or there are multiple candidate targets.

[0331] Window length adjustment strategy:

[0332] High confidence (10ms window):

[0333] Advantages: Improve target update rate (100Hz) to adapt to high-speed targets;

[0334] Data volume: A single window contains 1 frame of data, relying on Kalman filter prediction to compensate for short-term data loss.

[0335] Low confidence (50ms window):

[0336] Advantages: Accumulating 5 frames of data improves the signal-to-noise ratio, suitable for weak targets or complex environments;

[0337] Data processing: Sliding window splicing technology is used to avoid truncation of target information.

[0338] Dynamic switching mechanism:

[0339] When the confidence level changes from high to medium (e.g., the residual error of two consecutive frames is greater than 2σ²), the window is gradually extended to 30ms;

[0340] When the confidence level returns to high, the 10ms window takes effect immediately.

[0341] Technical advantages and synergy:

[0342] Closed-loop feedback mechanism:

[0343] The front end (S102 interference feature library) and the back end (S104 parameter adjustment) form a two-way interaction:

[0344] The JSR drop triggers the update of the spatial matrix, which is also fed back to S102 to optimize the feature extraction network weights;

[0345] The sparse dictionary update results are synchronized to the S103 target fidelity module to improve the reconstruction accuracy.

[0346] Dynamic adaptability:

[0347] Design differentiated response strategies for suppression jamming (JSR sensitive) and deception jamming (reconstruction error sensitive);

[0348] The window length is associated with the confidence level, balancing real-time performance and detection reliability.

[0349] Resource optimization:

[0350] Shorten the window to reduce the computational load at high confidence levels (FPGA resource usage is reduced by 40%).

[0351] Dictionary updates are triggered only when necessary to avoid hardware overheating caused by over-learning.

[0352] Implementation Notes:

[0353] Parameter calibration:

[0354] The length of the JSR historical mean window needs to be adjusted according to the type of interference: a short window (such as 30 frames) is used for suppression interference, and a long window (100 frames) is used for deception interference;

[0355] The target error threshold needs to change dynamically with the radar operating mode (search / tracking).

[0356] Exception handling:

[0357] For JSR drops caused by instantaneous interference (such as lightning pulses), a 200ms delay confirmation mechanism is added to avoid false triggering;

[0358] When the dictionary update fails (such as insufficient hardware resources), the backup base function library is enabled to maintain basic functions.

[0359] Hardware Adaptation:

[0360] Spatial matrix reconfiguration is performed through FPGA parallel computing with a latency of <0.5ms;

[0361] Dynamic window adjustment uses a double buffering mechanism to ensure seamless switching.

[0362] Through JSR monitoring, error feedback and confidence-driven dynamic adjustment, a closed-loop optimization system for radar anti-interference control is constructed. Its core innovations are:

[0363] Multi-index collaborative decision-making: adaptively adjust parameters based on interference suppression effect, target accuracy and environmental complexity;

[0364] Real-time hierarchical guarantee: The real-time requirements of different scenarios are matched through window length grading;

[0365] Resource performance balance: trigger high-load operations only when necessary to improve system efficiency.

[0366] Those skilled in the art can adjust the threshold parameters (such as setting the JSR drop threshold to 10%-20%) according to the specific radar model (such as airborne radar giving priority to real-time performance, and ground-based radar focusing on accuracy) to achieve the best engineering adaptation.

[0367] Specifically, in the anti-interference control method for radar according to the present invention, in step S105:

[0368] The suppression interference processing includes the cascade operation of frequency domain energy detection, spatial domain null widening and time domain pulse shielding; the deception interference processing includes the iterative process of starting multi-target correlation analysis, range-Doppler consistency check and false trajectory removal; the result fusion adopts the target trajectory probability fusion algorithm to weightedly fuse the spatial-frequency domain joint features of the suppression interference branch and the spatial-time domain correlation features of the deception interference branch, wherein:

[0369] The spatial-frequency domain joint feature is generated by comprehensively evaluating the spatial null suppression effect and the interference spectrum energy distribution;

[0370] The spatial-temporal correlation features are extracted based on the consistency of multi-target spatial distribution and the temporal continuity analysis of trajectories.

[0371] This step (S105) designs a diversion processing and fusion mechanism based on the differentiated characteristics of suppression interference and deceptive interference to achieve precise anti-interference control in complex interference scenarios. The following is a detailed analysis from three aspects: interference type characteristics, processing strategy and fusion logic to ensure that those skilled in the art have a clear understanding of the technical implementation path:

[0372] Suppressive interference handling strategy:

[0373] Frequency domain energy detection:

[0374] Detection logic:

[0375] Based on the interference spectrum characteristics output in step S102, locate the frequency band where the energy is significantly higher than the background noise (e.g., power density>30dBm / MHz);

[0376] Mark the detected frequency band and calculate its bandwidth, center frequency and power slope (to determine whether it is swept frequency interference).

[0377] Implementation example:

[0378] Generate frequency domain notch template for narrowband high-power interference (bandwidth 5MHz, center frequency 10GHz);

[0379] For wideband interference (bandwidth 200MHz), the spectrum sensing tracking algorithm is activated to dynamically adjust the notch range.

[0380] Airspace zero-sag widening:

[0381] Widening mechanism:

[0382] Based on the null generated in step S103, the coverage range is expanded to both sides of the interference azimuth (eg, ±3°) to suppress interference beam jitter or multipath diffusion effects;

[0383] The null width is positively correlated with the interference power (e.g., for every 10 dB increase in interference power, the width expands by 1°).

[0384] Anti-Distortion Design:

[0385] Adaptive sidelobe weighting technology is used to maintain the mainlobe gain fluctuation less than 0.5dB when widening the null.

[0386] Time domain pulse masking:

[0387] Gating strategy:

[0388] Identify periodic interference pulses (e.g., repetition period 1ms, pulse width 50μs), and generate a shielding window in the time domain (pulse width + protection interval 60μs);

[0389] The signal inside the masking window is set to zero, and the signal outside the window is retained for target detection.

[0390] Dynamic Adjustment:

[0391] When the interference pulse period changes by more than 10%, the window parameter reconfiguration (such as the period tracking algorithm) is triggered.

[0392] Deceptive interference handling strategy:

[0393] Multi-target association analysis:

[0394] Association logic:

[0395] Construct target attribute matrix: including distance, speed, azimuth, RCS (radar cross section) and micro-Doppler characteristics;

[0396] Calculate the attribute similarity between the newly detected target and the target in the target trajectory library (if the Euclidean distance is less than the threshold, then they are associated).

[0397] False target identification:

[0398] If multiple similar targets appear at a certain azimuth at the same time (such as three targets with similar distances but large speed differences), they are marked as deception interference candidates.

[0399] Range-Doppler consistency check:

[0400] Verification rules:

[0401] Physical constraints: target acceleration < 100m / s² (if it exceeds this, it will be a false trajectory);

[0402] Consistency condition: The deviation between the distance change rate and the Doppler velocity is less than 10% (for example, if the measured velocity is 100m / s, the distance change rate should be ±100m / s±10%).

[0403] Iterative culling:

[0404] In the first round, targets with abnormal acceleration are eliminated;

[0405] In the second round, targets with inconsistent range and Doppler are eliminated;

[0406] The remaining targets enter the historical target trajectory comparison.

[0407] False track removal:

[0408] Verification of space-time continuity:

[0409] The real target must satisfy the smooth motion trajectory of at least 3 consecutive frames;

[0410] The false trajectories generated by deception interference usually have position jumps (such as sudden changes in the distance between frames > 50m).

[0411] Example:

[0412] The range-drag trajectory generated by DRFM interference has a mismatch between the Doppler velocity and the range change rate, which triggers its rejection.

[0413] Target trajectory probability fusion algorithm:

[0414] Feature weighting mechanism:

[0415] Suppress interference branch weights:

[0416] Depends on the confidence of frequency domain features (such as the improvement of signal-to-noise ratio after notching), with a weight of 60%-70%;

[0417] Deceptive interference branch weight:

[0418] Depends on the time domain trajectory continuity score, with a weight of 30%-40%.

[0419] Fusion Logic:

[0420] Probability calculation:

[0421] For each candidate target trajectory, calculate its frequency domain confidence (based on the energy proportion of the target signal outside the notch frequency band) and time domain confidence (based on the motion continuity score);

[0422] Weighted fusion formula: total confidence = 0.7×frequency domain confidence + 0.3×time domain confidence.

[0423] Target trajectory output:

[0424] The target tracks with total confidence > 80% are marked as true targets;

[0425] Target trajectories with a confidence level of 50%-80% enter the review queue;

[0426] Target tracks with confidence less than 50% are directly eliminated.

[0427] Dynamic weight adjustment:

[0428] When the environment complexity increases (e.g. the number of multiple targets > 20), the time domain weight is increased to 50% to enhance motion logic verification;

[0429] For high-frequency interference scenarios (such as Ku-band radar), the frequency domain weight is increased to 80%.

[0430] Technical advantages and implementation relevance:

[0431] Improved anti-interference performance:

[0432] The suppression rate of suppressive interference is ≥90% (measured data), and the rejection rate of deceptive interference is ≥95%;

[0433] The fusion algorithm makes the target trajectory output false positive rate less than 0.1%, which is better than the traditional single-branch processing method (false positive rate is about 5%).

[0434] Connection with upstream steps:

[0435] The interference suppression process depends on the spatial-frequency domain suppression result of step S103;

[0436] The deception interference processing uses the micro-Doppler characteristics of the target fidelity branch in step S103 for auxiliary verification.

[0437] Hardware resource optimization:

[0438] Frequency domain notching and time domain shielding are implemented through FPGA pipeline, with processing delay less than 2ms;

[0439] Multi-target association analysis uses distributed computing (such as 4-core parallelism) to meet real-time requirements.

[0440] Implementation Notes:

[0441] Parameter calibration:

[0442] The null width of the jammer needs to be calibrated according to the radar beam width (e.g. 3° beam width corresponds to a null of ±2°);

[0443] The acceleration threshold in the deception jamming check needs to be adapted to the target type (e.g. 200m / s² for missiles and 50m / s² for low-altitude aircraft).

[0444] Exception handling:

[0445] For sudden and intensive deception interference (such as generating 100 false targets per second), the degradation mode is activated: only the trajectory of high-priority targets (confidence> 70%) is verified;

[0446] When the fusion algorithm fails, it switches to independent branch output mode (suppression and deception branches report independently).

[0447] This step improves the radar's target recognition capability in complex interference scenarios by differentially processing suppression and deception interference and combining multi-dimensional feature fusion. Its core innovations are:

[0448] Interference characteristic driven: Design a dedicated processing chain to suppress the frequency domain concentration of interference and the time domain contradiction of deceptive interference;

[0449] Dynamic weight fusion: adaptively adjust feature weights according to environmental complexity and interference type to balance detection probability and false alarm rate;

[0450] Hardware co-design: Accelerate key modules through FPGA to ensure real-time execution of complex algorithms.

[0451] Those skilled in the art can adjust the parameter weights according to the radar operating mode (such as the search mode focuses on frequency domain processing and the tracking mode strengthens time domain verification) to achieve optimal anti-interference performance.

[0452] Specifically, in the anti-interference control method applied to radar according to the present invention, the dynamic adjustment further includes:

[0453] Calculate the error covariance matrix between the target reconstructed signal and the radar measurement data in real time;

[0454] When the eigenvalue of the covariance matrix exceeds the dynamic threshold, the parameter adjustment instruction is triggered and fed back to the deep network of step S102. The adjustment instruction includes 3D convolution kernel size optimization, graph neural network edge weight update and dynamic feature library query priority reset.

[0455] This step (dynamic adjustment mechanism) realizes adaptive optimization of the radar anti-interference system through real-time monitoring of the error covariance matrix and a closed-loop feedback mechanism.

[0456] Error covariance matrix calculation and monitoring:

[0457] Data sources and alignment:

[0458] Target reconstruction signal: the output from the target fidelity data stream in step S103, including parameters such as target position, velocity and micro-Doppler characteristics.

[0459] Radar measurement data: real-time measurement results of the original echo signal after preprocessing (such as pulse compression and moving target detection).

[0460] Spatiotemporal alignment: By unifying the timestamp (accuracy ≤ 1μs) and spatial coordinate conversion, ensure that the two data are compared in the same reference frame.

[0461] Covariance matrix generation:

[0462] Error vector construction: Calculate the deviation between the target reconstructed signal and the measured data in each dimension (distance, speed, azimuth) to form an error vector (such as ΔR=2m, Δv=1m / s, Δθ=0.3°).

[0463] Matrix calculation logic:

[0464] The covariance matrix reflects the correlation between the error dimensions (e.g., range-velocity coupling error);

[0465] The matrix eigenvalues ​​characterize the overall error level of the system (the maximum eigenvalue > threshold indicates the risk of error out of control).

[0466] Dynamic threshold setting:

[0467] Determination of benchmark value: set the initial threshold (e.g., characteristic value threshold = 5) according to the radar performance index (e.g., ranging accuracy ±1m);

[0468] Adaptive Adjustment:

[0469] When the environment complexity increases (e.g. the number of interference sources > 5), the threshold is dynamically relaxed to 8;

[0470] The threshold is tightened to 3 for high confidence tracking.

[0471] Parameter adjustment instruction feedback mechanism:

[0472] Trigger conditions and instruction generation:

[0473] Eigenvalue out-of-limit judgment: When the maximum eigenvalue of the covariance matrix exceeds the threshold for two consecutive frames, the adjustment instruction is triggered.

[0474] Instruction content:

[0475] 3D convolution kernel size optimization: dynamically adjust the kernel size according to the error distribution characteristics (e.g., expand from 3×3×3 to 5×5×3 to capture long-period interference features);

[0476] Graph neural network edge weight update: strengthen the connection weight of high error associated nodes (such as multipath environment node weight +20%);

[0477] Dynamic feature library query priority reset: raise the query priority of the current high error interference feature (such as DRFM phase jump) to TOP1.

[0478] Feedback implementation process:

[0479] 3D Convolutional Network Optimization:

[0480] Kernel size adjustment: Expand the time dimension kernel length based on the error frequency domain distribution to enhance the ability to capture time-varying interference;

[0481] Example: For frequency sweep interference, increase the time dimension from 3 to 5 to improve the detection accuracy of the frequency sweep period.

[0482] Graph neural network edge weight update:

[0483] Multipath effect compensation: Increase the edge weight from interference source nodes to nodes in strong reflection environments (such as buildings) to improve the accuracy of multipath interference prediction;

[0484] Example: In an urban environment, the building reflection path weight is increased from 0.3 to 0.7.

[0485] Dynamic signature database priority reset:

[0486] Threat-driven ranking: re-rank interference features according to error contribution (e.g., the priority of phase jump feature is raised from 5th to 1st);

[0487] Example: Detecting a spike in errors from a DRFM interference, immediately set its signature to the highest priority.

[0488] Closed-loop feedback effect verification:

[0489] Iterative optimization: After adjusting the parameters, the covariance matrix is ​​recalculated. If the eigenvalue falls back to within the threshold, the current configuration is maintained; otherwise, a secondary adjustment is triggered (such as further increasing the convolution kernel size).

[0490] Performance improvement indicators: Actual measurements show that the closed-loop mechanism can increase the system error convergence speed by ≥50% and reduce the target tracking interruption rate by 60% in complex scenarios.

[0491] Technical synergy and implementation linkage:

[0492] Connection with step S102:

[0493] Input dependency: The priority reset of the dynamic feature library directly affects the dual-channel feature extraction process in step S102, ensuring that high-threat interference is processed first;

[0494] Output optimization: The adjusted 3D convolution kernel and graph neural network parameters improve the accuracy of subsequent interference feature extraction.

[0495] Linkage with step S104:

[0496] Error data sharing: The covariance matrix eigenvalues ​​are used as auxiliary decision basis for sparse dictionary update and window adjustment in step S104;

[0497] Example: When the eigenvalue exceeds the limit and the target reconstruction error exceeds the limit, the convolution kernel optimization is triggered first instead of the dictionary update.

[0498] Hardware resource adaptation:

[0499] FPGA implementation: Covariance matrix calculation is completed through the hardware acceleration unit, with a latency of <0.2ms;

[0500] Parallel adjustment: 3D convolution kernel optimization and graph network weight update are assigned to independent logic units to avoid processing conflicts.

[0501] Implementation Notes:

[0502] Parameter calibration:

[0503] The initial threshold needs to be calibrated according to the radar model (e.g. the threshold for phased array radar is set to 4, and for mechanical scanning radar is set to 6);

[0504] The update range of edge weights needs to be limited to ±30% to prevent network oscillation.

[0505] Exception handling:

[0506] For continuous over-limit errors (e.g., 5 consecutive frames of feature values ​​> threshold), start the system self-check mode to eliminate hardware faults;

[0507] When the adjustment command fails to execute, it rolls back to the last stable configuration and triggers an alarm.

[0508] Real-time guarantee:

[0509] Feedback command generation cycle ≤ 1ms, ensuring closed-loop response speed;

[0510] Use incremental parameter updates (such as weight fine-tuning ±5%) to reduce computational load.

[0511] This dynamic adjustment mechanism builds a closed-loop adaptive optimization system for the radar anti-interference system through real-time monitoring of the error covariance matrix and parameter feedback. Its core innovations are:

[0512] Multi-dimensional error correlation analysis: reveal the inherent correlation of errors through the covariance matrix and guide precise parameter adjustment;

[0513] Cross-module collaborative optimization: feedback instructions link feature extraction, interference suppression and target fidelity modules to achieve global performance improvement;

[0514] Hardware acceleration guarantee: The key computing links are implemented through FPGA to meet the real-time requirements of complex algorithms.

[0515] Those skilled in the art can adjust parameter thresholds and update strategies according to specific application scenarios (such as complex electromagnetic environment countermeasures that require enhanced graph network updates, and low-altitude surveillance scenarios that focus on convolution kernel time domain optimization) to achieve optimal engineering adaptation.

[0516] Specifically, in the anti-interference control method applied to radar described in the present invention, steps S101 to S105 implement the following parallel processing on the FPGA chip: hardware acceleration of spatial filter matrix calculation and beamforming weight vector update; pipeline processing of time-frequency distribution spectrum generation and spectral line reconstruction operations; dedicated logic unit allocation for polarization projection operator calculation and dynamic dictionary learning.

[0517] The anti-interference control method of the present invention realizes the efficient execution of a multi-dimensional anti-interference algorithm through the parallel architecture design of an FPGA chip.

[0518] Hardware acceleration of spatial filter matrix calculation and beamforming weight vector update:

[0519] Parallel computing architecture design:

[0520] Covariance matrix calculation:

[0521] The spatial data received by the radar array is divided into multiple sub-arrays (e.g., 4×4 sub-arrays) and assigned to the parallel processing units (PEs) of the FPGA to synchronously calculate the local covariance matrix;

[0522] The sub-array results are aggregated through a tree accumulator structure to generate a global covariance matrix, reducing the latency to 1 / 5 of that of traditional DSP solutions.

[0523] Weight vector optimization:

[0524] Using hardware implementation of the gradient descent algorithm, each PE unit independently calculates one dimension of the weight vector, and the convergence speed is improved through parallel iterative updates.

[0525] Dynamic update mechanism:

[0526] Real-time guarantee:

[0527] The weight vector update period is ≤1ms, supporting real-time tracking of high-speed mobile interference sources;

[0528] Interrupt response mechanism: When a sudden change in interference direction is detected (rate of change > 10° / s), the weight vector recalculation is triggered immediately.

[0529] Resource reuse strategy:

[0530] Idle PE units are automatically assigned to covariance calculation tasks of adjacent channels to improve hardware utilization.

[0531] Technical advantages: Compared with traditional serial processing, the computational efficiency of spatial filtering is increased by 8 times, meeting the millisecond-level response requirements in complex interference scenarios.

[0532] Pipeline processing of time-frequency distribution spectrum generation and spectrum line reconstruction:

[0533] Pipeline stage division:

[0534] Stage 1 (time-frequency analysis):

[0535] The input data is framed (256 sampling points per frame) and a time-frequency diagram is generated through multiple parallel STFT units (16 parallel channels, each channel processes 16 points);

[0536] The window function (Hamming window) and FFT calculation are integrated into the same logic unit to reduce data handling overhead.

[0537] Stage 2 (spectral line reconstruction):

[0538] Perform inverse FFT calculation on the notched spectrum to restore the time domain signal;

[0539] The interpolation filter is used to compensate for the group delay and retain the target Doppler shift characteristics.

[0540] Data Flow Management:

[0541] Double Buffer Design:

[0542] When the current frame is performing STFT calculation, the next frame data is synchronously stored in the cache to eliminate processing gaps;

[0543] The reconstructed time domain signal is seamlessly connected with subsequent modules (such as polarization processing) through a ring buffer.

[0544] Dynamic resolution adjustment:

[0545] Dynamically select the number of FFT points based on the target speed (512 points for high-speed targets and 256 points for low-speed targets), balancing resolution and real-time performance.

[0546] Technical advantages: The time-frequency processing throughput reaches 10^6 sampling points per second, supporting full-band real-time analysis of broadband radar signals.

[0547] Dedicated logic unit allocation for polarization projection operator calculation and dynamic dictionary learning:

[0548] Dedicated logic unit design:

[0549] Polarization projection operator module:

[0550] The hardware implements the eigendecomposition of the polarization covariance matrix and outputs the projection vector in the direction of maximum difference.

[0551] Supports parallel processing of 4 groups of polarization channel (HH / HV / VH / VV) data with a latency of <0.1ms.

[0552] Dynamic dictionary learning module:

[0553] Assign independent lookup tables (LUTs) to each type of basis function (point scattering, sliding scattering, micro-Doppler);

[0554] The basis function weights are optimized online through a sparse coding hardware accelerator, with an update cycle of ≤5ms.

[0555] Resource isolation and priority scheduling:

[0556] Static resource division:

[0557] The polarization module exclusively occupies 20% of DSP Slice resources, and the dictionary learning module occupies 15% of BRAM resources;

[0558] The remaining resources are dynamically allocated to other tasks (such as spatial filtering).

[0559] Task Priority:

[0560] Polarization projection is a high-priority task that can preempt dictionary learning resources to ensure real-time performance;

[0561] Dictionary learning automatically triggers batch updates when system load is low.

[0562] Technical advantages: The resource competition rate of polarization processing and dictionary learning is reduced by 90%, ensuring deterministic latency under multi-task concurrency.

[0563] Collaboration and system-level optimization:

[0564] Data interaction mechanism:

[0565] Cross-module data bus:

[0566] The spatial filtering output is transmitted to the polarization processing module via the high-speed AXI bus;

[0567] The time-frequency reconstruction data is shared to the dictionary learning module through distributed RAM.

[0568] Synchronous signal design:

[0569] The global clock is divided to generate multi-phase synchronization pulses to coordinate the data handover between each stage of the pipeline.

[0570] Power and area optimization:

[0571] Dynamic Power Management:

[0572] Inactive processing units automatically enter low-power modes (e.g. clock gating);

[0573] Peak power consumption is reduced by 40% compared with traditional solutions.

[0574] Logical unit multiplexing:

[0575] Polarization projection and dictionary learning share some multiplier resources to reduce hardware overhead.

[0576] Implementation effect and scene adaptation:

[0577] Performance indicators:

[0578] Real-time: The whole process processing delay is ≤8ms, meeting the real-time control requirements of the radar;

[0579] Resource usage: A single FPGA (such as the Xilinx Ultrascale+ series) can achieve 8-channel processing and support multi-target tracking.

[0580] Scenario adaptation strategy:

[0581] Airborne radar: Improve the resource ratio of airspace filtering and polarization modules to adapt to high-speed dynamic environments;

[0582] Ground-based early warning radar: Expand the number of time-frequency processing pipeline levels and enhance broadband signal processing capabilities.

[0583] The present invention uses the parallel architecture design of FPGA chips to map core algorithms such as spatial filtering, time-frequency analysis, polarization processing, and dictionary learning into hardware executable logic units, thus achieving real-time and efficient operation of multi-dimensional anti-interference control algorithms. The core technology is:

[0584] Heterogeneous computing division of labor: divide parallel computing, pipeline processing and dedicated logic units according to algorithm characteristics to maximize hardware efficiency;

[0585] Dynamic resource allocation: Balance real-time performance, accuracy and resource overhead through priority scheduling and task preemption mechanism;

[0586] System-level collaborative optimization: Cross-module data bus and synchronization mechanism ensure low latency across the entire link.

[0587] Technical personnel in this field can adjust the resource allocation ratio (such as 15%-25% of the polarization module DSP Slice) according to the number of channels, processing accuracy requirements and FPGA model of the specific radar system to adapt to different application scenarios.

Claims

1. An anti-interference control method applied to radar, characterized in that: The steps include: Step S101, synchronously collect multi-dimensional input data of the radar system, align and normalize the spatial domain steering matrix, time-frequency distribution spectrum, polarization scattering parameters and dynamic environment perception data in time and space, and generate a standardized multi-dimensional data stream, wherein the dynamic environment perception data includes the interference source azimuth, spectrum scanning results and multipath propagation characteristics; Step S102, extract interference features from the multi-dimensional data stream through a dual-channel deep network, wherein the dual-channel deep network includes a 3D convolutional network of a first channel and a graph neural network of a second channel, outputs interference fingerprint vectors to a dynamic feature library in real time and updates the interference feature priority list, wherein the dynamic feature library refreshes the interference feature priority list with a period of 5 ms, and the priority dynamically generates a sorted list according to the power intensity and occurrence frequency of the interference signal; Step S103, perform collaborative processing based on the interference feature priority list: perform spatial domain filtering, frequency domain notching and polarization projection on the interference suppression data stream in sequence, and perform residual signal extraction and sparse reconstruction in the target fidelity data stream; Step S104, dynamically adjust the spatial filter matrix parameters, sparse dictionary configuration and data processing window length according to the real-time evaluation results of the interference suppression ratio, target reconstruction error and tracking confidence; Step S105, implement data diversion processing on the suppression interference and deceptive interference, through the parallel processing of frequency domain energy detection and multi-target association analysis, and perform weighted fusion on the results to output the final target trajectory information.

2. The anti-interference control method for radar according to claim 1, characterized in that: In step S101: The spatial steering matrix is ​​composed of phase response data of the radar antenna array; the time-frequency distribution spectrum is generated by short-time Fourier transform; the polarization scattering parameters include HH, HV, VH, and VV polarization channel measurement values; the dynamic environment perception data includes the azimuth of the interference source, spectrum scanning results, and multipath propagation characteristics.

3. The anti-interference control method for radar according to claim 1, characterized in that: The step S102 includes: The first channel uses a 3D convolutional network to extract the modulation mode of the interference signal from the joint time-frequency-space data block. The modulation mode includes the phase jump feature of the digital RF memory, and the phase jump feature is extracted by the time axis gradient detection algorithm. The second channel constructs an interference-environment association model through a graph neural network to analyze the multipath diffusion effect of the electromagnetic environment on interference propagation. The dynamic feature library refreshes the interference feature priority list with a period of 5 ms.

4. The anti-interference control method for radar according to claim 1, characterized in that: The interference suppression data stream processing in step S103 includes: spatial domain processing, generating a beamforming weight vector according to the interference feature library, implementing null-pointing control on the original data, and the null-pointing depth is proportional to the interference signal power; frequency domain processing, generating a band-stop filter group based on the interference spectrum characteristics, reconstructing the spectral lines of the overlapping frequency band signals, and retaining the target Doppler frequency shift characteristics; polarization processing, calculating the polarization difference between the interference and the target, and selecting the polarization projection operator in the direction of the maximum difference to perform signal separation.

5. The anti-interference control method for radar according to claim 1, characterized in that: The target fidelity data stream processing in step S103 includes: using a dynamic dictionary learning algorithm to match the target scattering basis function, the basis function includes a point scattering model, a sliding scattering model and a rotating component micro-Doppler feature; performing spatiotemporal correlation between the reconstructed signal and the historical tracking data in the target trajectory library, and eliminating false targets through Kalman filter residual analysis.

6. The anti-interference control method for radar according to claim 1, characterized in that: The step S104 includes: When it is detected that the interference suppression ratio drops by more than 15% compared with the historical average, the emergency reconfiguration of the spatial domain filter matrix is ​​triggered; if the target reconstruction error exceeds the preset threshold for three consecutive frames, the online update process of the sparse dictionary is started; the data processing window length is dynamically adjusted according to the tracking confidence level: the window is shortened to 10ms when the confidence is high, and extended to 50ms when the confidence is low.

7. The anti-interference control method for radar according to claim 1, characterized in that: In the step S105: The processing of suppression interference includes performing cascade operations of frequency domain energy detection, spatial domain null widening and time domain pulse shielding; the processing of deception interference includes starting the iterative process of multi-target association analysis, distance and speed consistency verification and false trajectory elimination; the result fusion adopts the target trajectory probability fusion algorithm to weightedly fuse the frequency domain characteristics of the suppression interference branch and the time domain characteristics of the deception interference branch.

8. The anti-interference control method for radar according to claim 1, characterized in that: The dynamic adjustment also includes: Calculate the error covariance matrix between the target reconstructed signal and the radar measurement data in real time; When the eigenvalue of the covariance matrix exceeds the dynamic threshold, a parameter adjustment instruction is triggered and fed back to the deep network of step S102. The adjustment instruction includes 3D convolution kernel size optimization, graph neural network edge weight update, and dynamic feature library query priority reset.

9. The anti-interference control method for radar according to claim 1, characterized in that: The steps S101 to S105 implement the following parallel processing on the FPGA chip: Hardware acceleration of spatial filter matrix calculation and beamforming weight vector update; pipeline processing of time-frequency distribution spectrum generation and spectral line reconstruction operations; dedicated logic unit allocation for polarization projection operator calculation and dynamic dictionary learning; The hardware acceleration of spatial domain filter matrix calculation is realized through parallel processing units on the FPGA chip, and the covariance matrix calculation adopts 16 parallel processing units to process sub-array data in parallel.

Images