A vehicle-mounted radar anti-interference method and system based on time-frequency-space three-dimensional adaptive sampling

Abstract

The application relates to the technical field of radar signal processing, and discloses a vehicle-mounted radar anti-interference method and system based on time-frequency-space three-dimensional adaptive sampling. In view of the problems of low detection efficiency and unreasonable sampling resource allocation in a multi-interference source scene, a time-frequency-space three-dimensional joint interference feature tensor is constructed, and a three-dimensional cross-coupling effect is used to accurately identify interference; based on interference intensity and gradient information, the local sampling density is dynamically adjusted, the sampling is adaptively increased in interference dense and boundary areas, and a compression sensing algorithm is used to perform high-precision reconstruction on the non-uniformly sampled signals. The application effectively improves the interference identification accuracy in a complex electromagnetic environment, significantly reduces the system data throughput and calculation cost while guaranteeing target detection performance, and is suitable for high-performance vehicle-mounted radar systems.

Description

A method and system for anti-jamming vehicle radar based on time-frequency-space three-dimensional adaptive sampling Technical Field

[0001] This invention relates to the field of radar signal processing technology, and in particular to an anti-interference method and system for vehicle-mounted radar based on time-frequency-space three-dimensional adaptive sampling. Background Technology

[0002] With the rapid development of automotive intelligent technology, the installation rate of vehicle-mounted millimeter-wave radar, as a core environmental perception sensor, has seen explosive growth. In complex road scenarios with traffic congestion or dense traffic, the simultaneous operation of multiple radars in the same or adjacent frequency bands has become commonplace, leading to increasingly severe mutual interference problems. Radio frequency interference between radars can raise the system's noise floor and even create false targets in the range-Doppler spectrum, seriously affecting the detection probability and parameter estimation accuracy of vehicle-mounted radar for real targets, posing a significant safety hazard to autonomous driving systems.

[0003] Existing anti-jamming technologies for vehicle-mounted radar typically only process signal characteristics in a single dimension. For example, they may only detect pulse interference by utilizing amplitude abrupt changes in the time domain, or only identify anomalous spectral peaks in the frequency domain, or use spatial beamforming techniques to suppress interference in specific directions. However, in complex scenarios with multiple interference sources, interference signals often exhibit highly dynamic and coupled non-stationary characteristics in the time, frequency, and spatial domains. Traditional single-dimensional detection methods struggle to comprehensively capture interference features, leading to missed detections in dimensions where interference characteristics are insignificant, and false alarms in regions with complex signals.

[0004] Furthermore, existing anti-interference systems generally employ a uniform sampling strategy, acquiring data at a fixed sampling rate throughout the entire observation period and frequency band. This "one-size-fits-all" sampling method fails to consider the sparse distribution characteristics of interference signals in multidimensional space, resulting in the collection of a large amount of redundant data in "clean" areas with sparse interference, leading to a waste of storage and computing resources. Conversely, in "contaminated" areas with dense interference, the limited number of sampling points often fails to obtain sufficient information to accurately characterize and eliminate interference. This unreasonable allocation of resources not only increases the computational burden on the backend signal processor and limits the system's real-time response capability but also hinders further improvements in anti-interference performance. Summary of the Invention

[0005] The main objective of this invention is to provide a vehicle-mounted radar anti-interference method and system based on time-frequency-space three-dimensional adaptive sampling, so as to solve the technical problems of low detection accuracy and high computational overhead caused by unreasonable allocation of sampling resources in the existing technology under complex interference environment.

[0006] To achieve the above objectives, this invention provides a vehicle-mounted radar anti-interference method based on time-frequency-space three-dimensional adaptive sampling, the method comprising the following steps:

[0007] The vehicle-mounted radar acquires the raw echo signal, wherein the vehicle-mounted radar uses an array receiving unit to acquire a multi-dimensional signal containing time domain, frequency domain, and spatial domain information;

[0008] A three-dimensional joint interference feature characterization model is constructed, and the time-domain interference features, frequency-domain interference features, and spatial-domain interference features of the original echo signal are extracted respectively. Based on the time-domain interference features, frequency-domain interference features, and spatial-domain interference features, a three-dimensional interference feature tensor is constructed. Interference regions are identified by threshold determination of the three-dimensional interference feature tensor.

[0009] Adaptive sampling density adjustment is performed, the gradient components and comprehensive gradient magnitude of the three-dimensional interference feature tensor in the time domain, frequency domain and spatial domain are calculated, the sampling density of each three-dimensional unit is calculated based on the element values ​​of the three-dimensional interference feature tensor and the comprehensive gradient magnitude, and a non-uniform sampling mask is generated accordingly.

[0010] Interference suppression and signal reconstruction are performed. Suppression weights are applied to the sampled signal according to the interference region, and an observation matrix is ​​constructed based on the non-uniform sampling mask. The target signal is reconstructed from the suppressed signal using a compressed sensing algorithm.

[0011] The reconstructed target signal is used for target detection and parameter estimation, and the target's distance, velocity and angle information are output.

[0012] As a preferred embodiment of the present invention, the construction of the three-dimensional joint interference feature characterization model specifically includes:

[0013] The signal energy within a sliding time window is calculated in the time domain. When the signal energy exceeds a set threshold, the time-domain interference indication function is activated. Set to 1 otherwise; calculate the deviation of the power spectral density after short-time Fourier transform from the theoretical echo spectrum in the frequency domain. When the deviation exceeds a set threshold, the frequency domain interference indicator function is activated. Set to 1 otherwise set to 0; calculate the eigenvalue distribution and spatial spectrum of the spatial covariance matrix in the spatial domain, and activate the spatial interference indicator function when the spatial spectrum peak exceeds a set threshold. Set to 1 otherwise set to 0;

[0014] The three-dimensional interference feature tensor The calculation formula is:

[0015] in, , , These are the weighting coefficients for the time domain, frequency domain, and spatial domain, respectively. is the three-dimensional cross-coupling coefficient.

[0016] As a preferred embodiment of the present invention, the construction of the three-dimensional joint interference feature characterization model further includes:

[0017] For the three-dimensional interference feature tensor Perform low-rank approximate decomposition:

[0018]

[0019] in, , , These are feature factor vectors in the time domain, frequency domain, and spatial domain, respectively. This represents the outer product operation. For tensor rank.

[0020] As a preferred embodiment of the present invention, the sampling density of each three-dimensional unit is calculated using the following sampling density allocation function:

[0021]

[0022] in, As the baseline sampling density, This is the interference intensity adjustment coefficient. This is the gradient adjustment coefficient. and These are the normalization factors for the tensor elements and the comprehensive gradient magnitude, respectively. This is a 3D linkage correction item.

[0023] As a preferred technical solution of the present invention, the three-dimensional linkage correction item Defined as:

[0024] in, , , These are the linkage and coupling gain coefficients for each dimension. For indicator functions, and These are specific regions in the frequency domain and time domain, respectively.

[0025] As a preferred embodiment of the present invention, the non-uniform sampling mask is generated using the Poisson disk sampling algorithm, and the generated sampling point positions are... , The constraints are satisfied:

[0026]

[0027] in, Sampling points Local sampling density at a given location.

[0028] The aforementioned vehicle-mounted radar anti-interference method based on time-frequency-space three-dimensional adaptive sampling, wherein the weighting function for applying suppression weights to the sampled signal is:

[0029]

[0030] in, These are the coordinates of the sampling point. For the corresponding interference feature tensor value, Parameters used to control the intensity of suppression.

[0031] As a preferred embodiment of the present invention, the step of reconstructing the target signal from the suppressed signal using a compressed sensing algorithm includes:

[0032] Optimization problem:

[0033]

[0034] in, The target signal's sparse spectrum in the frequency domain is the solution. It is a sparse transformation basis. For the observation matrix corresponding to non-uniform sampling locations, This is the effective observation sample vector after interference suppression.

[0035] The present invention also provides a vehicle-mounted radar anti-jamming system based on time-frequency-space three-dimensional adaptive sampling, the system comprising:

[0036] The signal acquisition module is used to acquire the raw echo signal through the radar transmitting unit and the array receiving unit;

[0037] The time-frequency-space three-dimensional joint interference feature characterization module is used to extract interference features in the time domain, frequency domain, and spatial domain, construct a three-dimensional interference feature tensor, and identify interference regions.

[0038] The adaptive sampling density adjustment module is used to calculate the gradient and comprehensive gradient magnitude of the three-dimensional interference feature tensor, allocate sampling density according to interference intensity and gradient information, and generate sampling masks.

[0039] The interference suppression and signal reconstruction module is used to perform weighted suppression of signals within the interference area and reconstruct the target signal based on non-uniform sampling data using compressed sensing algorithms.

[0040] The target detection and parameter estimation module is used to process the reconstructed signal to output target parameters.

[0041] As a preferred embodiment of the present invention, the time-frequency-space three-dimensional joint interference feature characterization module includes a three-dimensional feature fusion subunit, which is used to calculate tensor element values ​​based on interference indications in the time domain, frequency domain, and spatial domain, as well as three-dimensional cross-coupling coefficients; the adaptive sampling density adjustment module includes a linkage adjustment subunit, which is used to adjust the sampling density of relevant dimensions in a linkage manner based on interference indication information in each dimension.

[0042] The vehicle-mounted radar anti-interference method and system based on time-frequency-space three-dimensional adaptive sampling provided by this invention have the following beneficial effects:

[0043] 1. This invention constructs a three-dimensional joint interference feature characterization model based on time, frequency, and space. Utilizing the joint distribution characteristics and cross-coupling effects of interference signals in the time, frequency, and spatial domains, it significantly improves the ability to identify dynamic interference signals in complex electromagnetic environments. Compared to single-dimensional detection methods, this invention can effectively distinguish between target echoes with indistinct features and interference signals, reducing false alarm and false negative rates while maintaining high detection accuracy.

[0044] 2. This invention proposes an adaptive sampling density adjustment mechanism based on interference intensity and its gradient information, which can intelligently allocate sampling resources according to the interference distribution in three-dimensional space. It automatically increases the sampling density in regions with high interference intensity and large gradient changes to obtain sufficient interference details, while reducing the sampling density in regions with sparse interference. Combined with Poisson disk sampling and compressed sensing reconstruction algorithms, this scheme achieves high-quality signal reconstruction while significantly reducing data throughput and computational overhead, effectively resolving the contradiction between anti-interference performance and real-time processing requirements. Attached Figure Description

[0045] Figure 1 is a flowchart illustrating the anti-interference method for vehicle-mounted radar based on time-frequency-space three-dimensional adaptive sampling provided in an embodiment of the present invention;

[0046] Figure 2 is a structural block diagram of the vehicle-mounted radar anti-jamming system based on time-frequency-space three-dimensional adaptive sampling provided in an embodiment of the present invention.

[0047] Figure labeling: 100, Signal acquisition module; 110, Transmitting unit; 120, Array receiving unit; 200, Time-frequency-space three-dimensional joint interference feature characterization module; 210, Time domain feature extraction subunit; 220, Frequency domain feature extraction subunit; 230, Spatial domain feature extraction subunit; 240, Three-dimensional feature fusion subunit; 300, Adaptive sampling density adjustment module; 310, Gradient calculation subunit; 320, Density allocation subunit; 330, Linkage adjustment subunit; 340, Sampling point generation subunit; 400, Interference suppression and signal reconstruction module; 410, Interference suppression subunit; 420, Compressed sensing reconstruction subunit; 500, Target detection and parameter estimation module. Detailed Implementation

[0048] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0049] The present invention provides a vehicle radar anti-interference method and system based on time-frequency-space three-dimensional adaptive sampling. It addresses the problems of low interference detection efficiency and unreasonable allocation of sampling resources faced by vehicle radar in complex scenarios with multiple interference sources. By performing joint feature characterization of interference signals in three dimensions of time, frequency and space, and adaptively adjusting the sampling density based on interference intensity and its gradient information, it achieves accurate identification and efficient suppression of interference signals.

[0050] Referring to Figure 1, the vehicle-mounted radar anti-jamming method provided in this embodiment of the invention includes five main steps: acquiring the original echo signal, constructing a three-dimensional joint interference feature characterization model, adaptive sampling density adjustment, interference suppression and signal reconstruction, and target detection and parameter estimation.

[0051] During the acquisition of the original echo signal, the vehicle-mounted radar uses a frequency-modulated continuous wave (FM-CW) system for target detection. The transmitted signal generated by the radar transmitting unit can be expressed as:

[0052]

[0053] in Indicates the amplitude of the transmitted signal. Indicates the carrier frequency. Indicates the frequency modulation bandwidth. This indicates the frequency sweep period. In practical applications, the carrier frequency... Typically, the millimeter wave band between 77 GHz and 81 GHz is selected, with a frequency modulation bandwidth of [missing information]. The sweep frequency period can be set from hundreds of MHz to several GHz depending on the distance resolution requirements. The value is determined based on the maximum unambiguous distance and speed measurement requirements, with a typical range of tens of microseconds to several milliseconds.

[0054] The radar receiver employs an array antenna structure to acquire spatial dimension information. Assume the array contains... There are 10 receiving array elements, and each element is arranged in a uniform linear array or uniform area array manner. The echo signal received by each array element is represented as:

[0055]

[0056] in For the target echo component, For interference signal components, The noise component is another component. The target echo component carries the target's range, velocity, and angle information, which is the useful signal that the radar needs to extract. The interference signal component originates from the electromagnetic radiation of other vehicle-mounted radars or communication equipment. It may exhibit pulsed or continuous characteristics in the time domain, may overlap with the operating frequency band of this radar in the frequency domain, and has a specific incident direction in the spatial domain. The noise component mainly includes receiver thermal noise and external environmental noise, which can usually be modeled as additive white Gaussian noise.

[0057] In scenarios where multiple vehicle-mounted radars operate simultaneously, the time-frequency-spatial distribution of interference signals exhibits complex dynamic characteristics. Radars from adjacent vehicles may use similar operating frequencies and modulation parameters. When the transmitted signals of two radars overlap in the time domain, beat interference will occur at the receiver. This interference manifests as a sudden energy change in the time domain, may fall within the beat frequency range of the target echo in the frequency domain, and corresponds to the direction of arrival of the interference source in the spatial domain. Traditional single-dimensional interference detection methods only focus on the abnormal characteristics of the signal in one dimension, making it difficult to effectively distinguish between the target echo and the interference signal, especially when the interference signal and the target echo have similar characteristics in a single dimension, easily leading to missed detections.

[0058] The time-frequency-space three-dimensional joint interference feature characterization model described in this embodiment of the invention comprehensively utilizes the distribution characteristics of interference signals in three dimensions for interference identification. Time-domain feature extraction is achieved through short-time energy analysis. For the received signal... In time Centered on, with a length of Calculate the signal energy within the sliding time window:

[0059]

[0060] Time-domain analysis window length The selection of the window requires a trade-off between time resolution and energy estimation stability. A window that is too short will increase the variance of the energy estimation, while a window that is too long may blur the temporal boundaries of the interference pulse. In a specific embodiment of the present invention, It can be set to the frequency sweep period. One-fifth to one-third.

[0061] Define a time-domain disturbance indication function based on short-time energy calculation results:

[0062]

[0063] in To obtain the average energy over the reference time period, the average energy of the interference-free region at the beginning of the signal segment can be used, or it can be obtained online using a recursive moving average method. Threshold coefficient The sensitivity of time-domain interference detection is determined by its magnitude. A higher value will increase the detection threshold and thus reduce the probability of false alarms, but may lead to missed detections of weak interference; a smaller value... The value is the opposite. In automotive applications, considering the large range of variation in interference signal strength, It can be adaptively adjusted based on the signal-to-noise ratio estimation results, with a typical value range of 1.5 to 4.

[0064] Frequency domain feature extraction employs the short-time Fourier transform method. For the received signal... After windowing, a Fourier transform is performed to obtain the time-frequency distribution:

[0065]

[0066] in For the analysis window function, Hamming window, Hanning window, or Gaussian window can be selected. Power spectral density of each frequency unit. This reflects the energy distribution of the signal at that frequency. To distinguish between power spectrum anomalies caused by interference and the normal power spectrum distribution of the target echo, the frequency domain interference characteristic is defined as the deviation of the power spectrum from the theoretical echo spectrum:

[0067]

[0068] Theoretical echo spectrum It can be pre-calculated based on the transmitted signal parameters and typical target scenarios, or it can be obtained statistically from measured data under interference-free conditions. The frequency domain interference indication function is defined as:

[0069]

[0070] in This represents the standard deviation of the power spectrum deviation, which can be estimated globally in the frequency domain. Threshold coefficient. The typical value range is 2 to 5, and the specific value can be adjusted according to the type and intensity distribution characteristics of the interference. For narrowband interference, it manifests as an abnormal concentration of energy at local frequency points in the frequency domain. A significant peak will appear at the interference frequency; for broadband interference, its energy distribution in the frequency domain is relatively flat, but the overall energy level is higher than that of the normal echo spectrum, and a comprehensive judgment needs to be made by combining time domain and spatial domain characteristics.

[0071] Spatial feature extraction utilizes the spatial correlation characteristics of the array-received data. An array snapshot vector is defined. ,in Indicates the first The array element in the first The received signal value at each sampling time. By analyzing... The spatial covariance matrix is ​​constructed by statistically averaging the snapshot vectors.

[0072]

[0073] superscript This represents the conjugate transpose operation. The eigenvalue decomposition of the covariance matrix reveals the spatial structure characteristics of the received signal:

[0074]

[0075] in For the first 1 eigenvalue, This represents the corresponding eigenvector. When interference signals are present, the eigenvalues ​​corresponding to the direction of the interference source will be significantly larger than the eigenvalues ​​in the noise subspace. This invention defines spatial interference characteristics as an index of the non-uniformity of eigenvalue distribution:

[0076]

[0077] in and These are the maximum and minimum eigenvalues, respectively. A larger value indicates a more pronounced directional energy distribution in the received signal, potentially corresponding to the presence of an interference source or target. To further distinguish between interference sources and targets, the MUSIC algorithm is used for spatial spectrum estimation to obtain spectral peak values ​​at various angles. The airspace interference indication function is defined as follows:

[0078]

[0079] in The average value of the spatial spectrum. This is the spatial threshold coefficient.

[0080] Three-dimensional joint feature fusion is one of the core technical aspects of this invention. The time domain is discretized into... Each time unit is discretized in the frequency domain as follows: Each frequency unit is spatially discretized into... Each angular unit is used to construct a three-dimensional disturbance feature tensor. The calculation of tensor elements comprehensively considers the interference indicators of each dimension and the cross-coupling effects between dimensions:

[0081]

[0082] in , , These are the weighting coefficients for each dimension, reflecting the importance of different dimensions of interference features in the comprehensive judgment; The three-dimensional cross-coupling coefficient is used to characterize the synergistic enhancement effect when the interference signal exhibits anomalous characteristics in three dimensions simultaneously.

[0083] The introduction of a three-dimensional cross-coupling term is the key difference between this invention and traditional single-dimensional or two-dimensional joint detection methods. Some interference signals may not have sufficiently obvious features in a single dimension; for example, their temporal energy may only be slightly higher than the background, their frequency distribution may be similar to the target echo, or their spatial orientation may be close to the target. In single-dimensional detection, these signals may be missed. However, when interference signals exhibit a regular distribution pattern in the three-dimensional joint space, the cross-coupling term can effectively amplify these joint anomaly features and improve detection sensitivity. Weighting coefficients , , and coupling coefficient The specific values ​​can be optimized according to the statistical characteristics of the actual interference scenario. In one embodiment of the present invention, they can be set to 0.25, 0.25, 0.25 and 0.5 respectively.

[0084] To further improve the robustness and computational efficiency of interference feature representation, this invention employs tensor decomposition to perform a low-rank approximation of the three-dimensional interference feature tensor:

[0085]

[0086] in , , These are feature factor vectors in the time domain, frequency domain, and spatial domain, respectively, with symbols... This represents the outer product operation. Tensor rank. The choice of determines the balance between the accuracy and computational complexity of the decomposition; a larger ... The value can more accurately approximate the original tensor, but the computational cost also increases accordingly. In real-time automotive applications, It can be set to a smaller integer between 3 and 10.

[0087] Interference regions are identified by applying threshold values ​​to tensor elements.

[0088]

[0089] in The threshold for determining interference. (Falling into the set) The three-dimensional time-frequency-space unit was determined to be affected by interference and needs to be suppressed in subsequent processing. Threshold The selection of the detection probability and the false alarm probability needs to be considered comprehensively. The constant false alarm detection principle can be adopted to adaptively determine the detection probability based on the statistical distribution of the tensor elements.

[0090] Adaptive sampling density adjustment based on interference intensity gradient is another core technical aspect of this invention. Traditional uniform sampling strategies allocate sampling resources evenly across time-frequency-space dimensions, failing to differentiate based on the actual distribution of interference. This results in sampling redundancy in sparse interference regions and insufficient sampling in dense interference regions. The adaptive sampling density adjustment mechanism proposed in this invention intelligently allocates sampling resources based on the interference distribution information provided by the three-dimensional interference feature tensor, achieving synergistic optimization of sampling efficiency and anti-interference performance.

[0091] The calculation of the interference intensity gradient field is fundamental to sampling density adjustment. To eliminate the influence of inconsistencies in physical dimensions across the time, frequency, and spatial domains on gradient calculation, this invention first establishes a normalized index coordinate system. ,in Gradient components are calculated along the time, frequency, and spatial domains on the three-dimensional disturbance feature tensor. For nodes inside the tensor, central difference is used for calculation; for boundary nodes, forward or backward difference is used for calculation.

[0092] The temporal gradient is:

[0093] The frequency domain gradient is:

[0094] The spatial gradient is:

[0095] The overall gradient magnitude is defined as:

[0096]

[0097] The overall gradient magnitude reflects the degree of drastic change in the disturbance intensity along each direction in the normalized feature space.

[0098] This invention employs a unified sampling density allocation model based on multi-dimensional feature fusion, integrating interference intensity, gradient information, and a three-dimensional linkage mechanism into a single mathematical expression, thus avoiding sampling conflicts caused by multiple parallel logic sets. The sampling density allocation function is:

[0099]

[0100] in As the baseline sampling density, This is the interference intensity adjustment coefficient. This is the gradient adjustment coefficient. and These are the normalization factors for the tensor elements and the gradient magnitude, respectively. The physical implication of this design is that regions with high interference intensity require denser sampling to obtain sufficient information for interference characteristic analysis and suppression; boundary regions with large interference intensity gradients require dense sampling to accurately determine the boundary between the interference region and the non-interference region, avoiding accidental damage to the target signal during interference suppression. The three-dimensional linkage correction term, used to characterize cross-dimensional collaborative sampling enhancement, is defined as follows:

[0101]

[0102] in This is an indicator function, representing the response across the entire frequency band or a specific frequency band region when an interference pulse is detected in the time domain. Increase the linkage density; These are the linkage and coupling gain coefficients for each dimension.

[0103] The sampling point locations are determined using a Poisson disk sampling algorithm based on normalized distance. Within the three-dimensional time-frequency-spatial domain, the generation of sampling points must satisfy the following constraints:

[0104]

[0105] in This is the normalized index coordinate vector. Indicates the first The three-dimensional coordinates of each sampling point are defined. This constraint ensures that the minimum distance between adjacent sampling points is adapted to the local sampling density, with denser sampling point distribution in high sampling density areas and sparser distribution in low sampling density areas. The Poisson disk sampling algorithm can generate a sampling point distribution with blue noise characteristics, avoiding aliasing artifacts of regular grid sampling while ensuring uniform coverage of sampling points.

[0106] In terms of hardware implementation, considering that automotive radar hardware typically uses ADCs (Analog-to-Digital Converters) with a fixed sampling rate, this embodiment adopts an architecture of "high-frequency oversampling + digital sparse decimation". The radar front-end ADC performs full acquisition at the highest fixed rate satisfying the Nyquist sampling theorem. After the data enters the FPGA (Field-Programmable Gate Array), real-time data decimation is performed according to the sampling mask generated by the Poisson disk algorithm described above. The retained non-uniform sampling points are processed by the subsequent DSP, while the discarded data points no longer occupy transmission bandwidth and storage resources. This strategy effectively reduces the data throughput and power consumption of the back-end signal processing without increasing the complexity of the ADC design.

[0107] During the interference suppression and signal reconstruction stage, weighted suppression processing is applied to the sampled signals within the identified interference area. The suppressed signal is represented as follows:

[0108]

[0109] The suppression weight function is defined as follows:

[0110]

[0111] The weighting function is an exponentially decaying function with respect to the values ​​of the elements in the disturbance feature tensor. Parameters Control the intensity of inhibition, smaller A higher value corresponds to a stronger suppression effect, but may cause signal distortion; a larger value... The suppression effect is weaker at certain values, and some interference may remain. The value of can be determined based on the statistical characteristics of the interference intensity distribution, with a typical value being 0.5 to 1 times the mean of the tensor elements in the interference region.

[0112] Because the sampling points generated by the adaptive sampling strategy are non-uniformly distributed, traditional uniform sampling signal processing methods cannot be directly applied for subsequent processing. This invention uses compressed sensing theory for signal reconstruction. It is particularly important to note that the echo signal of vehicle-mounted radar is typically a full-duration continuous wave in the time domain, lacking sparsity. Directly minimizing the L1 norm in the time domain will lead to reconstruction failure. Therefore, this invention constructs an optimization problem in the sparse transform domain (i.e., the frequency domain or beat frequency domain).

[0113] Define sparse transformation basis Let be the Discrete Fourier Transform (DFT) matrix (or, in a specific scenario, the Discrete Cosine Transform (DCT) basis). Let the sparse coefficient vector of the clean signal to be reconstructed in the frequency domain be . Then the time-domain signal .

[0114] Constructing the observation matrix It consists of a sampling mask matrix. It is composed of sparse transformation basis, that is, the effective observation process is modeled as Signal reconstruction can be formulated as a basis pursuit denoising (BPDN) optimization problem as follows: ;

[0115] in: The target signal is represented by its sparse frequency domain spectrum. The L1 norm is used to constrain the radar target by utilizing the sparsity of the radar target in the frequency domain (point targets each occupy a small number of frequency points); This is the valid observation sample vector retained after adaptive sampling and interference removal; This is the row selection matrix (i.e., the non-uniform sampling operator) corresponding to the reserved sample positions.

[0116] This optimization problem can be solved using the Orthogonal Matching Pursuit (OMP) algorithm or the Fast Iterative Shrinking Threshold Algorithm (FISTA). The solution yields the sparse coefficients. That is, directly corresponding to the target's range-velocity spectrum information, or through inverse transformation. The time-domain signal is recovered. Compared with traditional methods, this process can still recover the complete target spectrum with high accuracy using the remaining non-uniform samples even after removing interference samples (considered as missing data), thus achieving true anti-interference signal reconstruction.

[0117] The reconstructed clean signal is fed into the target detection and parameter estimation stage for conventional frequency-modulated continuous wave radar signal processing. First, a range-dimensional Fast Fourier Transform (FFT) is performed on the beat signals within each sweep cycle to obtain the spectrum resolved by each range cell. Then, a Doppler-dimensional FFT is performed on each range cell along the slow time dimension to obtain a range-velocity two-dimensional spectrum. A constant false alarm rate (CFAR) detection algorithm is used on the two-dimensional spectrum to extract targets and estimate the range and velocity parameters of each target. For systems using array antennas, the phase difference of the received signals from each array element can be further utilized to estimate the angle of arrival and obtain the target's angular information.

[0118] Referring to Figure 2, this embodiment of the invention also provides a vehicle-mounted radar anti-jamming system based on time-frequency-space three-dimensional adaptive sampling. The system includes a signal acquisition module 100, a time-frequency-space three-dimensional joint interference feature characterization module 200, an adaptive sampling density adjustment module 300, an interference suppression and signal reconstruction module 400, and a target detection and parameter estimation module 500.

[0119] The signal acquisition module 100 includes a frequency-modulated continuous wave radar transmitting unit 110 and an array receiving unit 120. The transmitting unit 110 generates a linear frequency-modulated continuous wave signal and radiates it into space through a transmitting antenna. The array receiving unit 120 includes... Each receiving antenna element has an independent receiving channel, capable of synchronously acquiring echo signals and performing analog-to-digital conversion. The analog signals are converted to digital signals and then sent to subsequent processing modules. Good phase and gain consistency must be maintained between the receiving channels to ensure the accuracy of spatial feature extraction. In one specific embodiment, It can be set to an integer between 4 and 16, and the element spacing can be set to half a wavelength to meet the grating lobe-free condition.

[0120] The time-frequency-space three-dimensional joint interference feature characterization module 200 includes a time-domain feature extraction subunit 210, a frequency-domain feature extraction subunit 220, a spatial-domain feature extraction subunit 230, and a three-dimensional feature fusion subunit 240. The time-domain feature extraction subunit 210 performs short-time energy analysis on the received signals from each channel, calculates the signal energy sequence using a sliding window, compares it with a reference energy, and outputs a time-domain interference indication sequence. The frequency domain feature extraction subunit 220 performs a short-time Fourier transform on the received signal to obtain the time-frequency distribution, calculates the power spectral density and compares it with the theoretical echo spectrum to obtain the power spectral deviation sequence, and then outputs the frequency domain interference indication sequence. The spatial feature extraction subunit 230 constructs a spatial covariance matrix using array snapshot data, performs eigenvalue decomposition to obtain eigenvalue distribution information, and uses the MUSIC algorithm for spatial spectrum estimation to output a spatial interference indication sequence. The three-dimensional feature fusion subunit 240 receives the interference indication sequence output by the above three subunits and constructs a three-dimensional interference feature tensor according to the tensor element calculation formula. Tensor decomposition is performed to achieve a low-rank approximation, and a threshold is used to determine the set of output interference regions. .

[0121] The adaptive sampling density adjustment module 300 includes a gradient calculation subunit 310, a density allocation subunit 320, a linkage adjustment subunit 330, and a sampling point generation subunit 340. The gradient calculation subunit 310 receives the three-dimensional interference feature tensor. Calculate the gradient components in the time domain, frequency domain, and spatial domain respectively. , , And synthesize a comprehensive gradient magnitude field Output. The density allocation sub-unit 320 calculates the target sampling density of each 3D unit according to the sampling density allocation function based on the values ​​of the interference feature tensor elements and the comprehensive gradient magnitude. The linkage adjustment subunit 330 executes the inter-dimensional coupling adjustment rules, and adjusts the sampling density of related dimensions in a linkage manner according to the interference indication information of each dimension, so as to achieve the coordinated optimization of the three-dimensional sampling density. The sampling point generation subunit 340 generates and outputs the sampling point position sequence that satisfies the density constraint based on the finally determined sampling density distribution using the Poisson disk sampling algorithm.

[0122] The interference suppression and signal reconstruction module 400 includes an interference suppression subunit 410 and a compressed sensing reconstruction subunit 420. The interference suppression subunit 410 receives a set of interference regions. Information and sampled signal data are used to apply a suppression weighting function to the sampling points that fall into the interference area. Output the weighted and suppressed signal The compressed sensing reconstruction subunit 420 receives adaptive sampling location information and suppressed signal data to construct an observation matrix. Solve the sparse reconstruction optimization problem and output the clean reconstructed signal. .

[0123] The target detection and parameter estimation module 500 receives the reconstructed signal, performs two-dimensional fast Fourier transform processing in the range dimension and Doppler dimension, performs constant false alarm detection on the range-velocity two-dimensional spectrum to extract the target, estimates the range and velocity parameters of each detected target, and estimates the angle parameters of the target by combining the array signal processing results, and finally outputs a target list containing range, velocity and angle information.

[0124] There is a close synergy between the two core technical components of this invention. The time-frequency-space three-dimensional joint interference feature characterization module 200 provides accurate interference distribution information to the adaptive sampling density adjustment module 300, and the three-dimensional feature tensor... This not only identifies the location of interference areas but also quantifies the interference intensity level at each location, enabling the adaptive sampling density adjustment module 300 to intelligently allocate sampling resources accordingly. The adaptive sampling strategy, in turn, enhances the detection efficiency of interference feature representation. Increasing the sampling density in densely interfering areas yields richer interference feature information, improving the resolution and representation accuracy of the three-dimensional feature tensor; gradient-guided dense sampling in boundary regions ensures accurate delineation of the boundary between the interference region and the target region.

[0125] During system operation, the time-frequency-spatial three-dimensional joint interference feature characterization module 200 and the adaptive sampling density adjustment module 300 form a closed-loop optimization mechanism. The time-frequency-spatial three-dimensional joint interference feature characterization module 200 outputs the interference distribution estimate of the current frame, and the adaptive sampling density adjustment module 300 optimizes the sampling strategy for the next frame accordingly. The optimized sampling data is then fed back to the time-frequency-spatial three-dimensional joint interference feature characterization module 200 to update the interference estimate. This iterative and collaborative mechanism enables the system to dynamically adapt to complex and ever-changing interference environments, maintaining stable anti-interference performance even in scenarios with rapidly changing interference characteristics.

[0126] The application of this invention in practical vehicle-mounted scenarios allows for parameter configuration based on different interference types and intensities. For scenarios dominated by co-channel interference, the weighting coefficient of the frequency domain features can be appropriately increased. Coupling coefficients related to the frequency domain; for scenarios dominated by impulse interference, the weighting coefficients of time-domain features can be increased. The coupling coefficient is related to the time domain; for complex scenarios with multiple interference sources coexisting, the weights of each dimension can be balanced and the three-dimensional cross-coupling coefficient can be appropriately increased. Reference density for sampling density adjustment and adjustment coefficient , The settings can be adjusted based on the computing power and real-time requirements of the onboard processor.

[0127] When implementing this invention on different hardware platforms, each module can be implemented in different ways. On a high-performance computing platform, the temporal feature extraction subunit 210, the frequency domain feature extraction subunit 220, and the spatial domain feature extraction subunit 230 can be executed simultaneously using a parallel processing architecture to fully utilize computing resources and shorten processing latency. On a resource-constrained embedded platform, a pipelined processing architecture can be used to execute each processing step sequentially, and the tensor decomposition algorithm in the 3D feature fusion subunit 240 can be appropriately simplified to reduce computational complexity. The selection of the reconstruction algorithm in the compressed sensing reconstruction subunit 420 can also be optimized according to platform characteristics. When computing resources are sufficient, a convex optimization algorithm with higher reconstruction accuracy can be used, and when real-time requirements are strict, a greedy algorithm with faster convergence speed can be used.

[0128] It should be noted that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A vehicle-mounted radar anti-interference method based on time-frequency-space three-dimensional adaptive sampling, characterized in that, Includes the following steps: The process involves acquiring the raw echo signal from an onboard radar, which uses an array receiving unit to acquire multi-dimensional signals containing time-domain, frequency-domain, and spatial-domain information; constructing a three-dimensional joint interference feature characterization model, extracting time-domain, frequency-domain, and spatial-domain interference features from the raw echo signal, and constructing a three-dimensional interference feature tensor based on these features; identifying interference regions by thresholding the three-dimensional interference feature tensor; performing adaptive sampling density adjustment, calculating the gradient components and comprehensive gradient magnitude of the three-dimensional interference feature tensor in the time, frequency, and spatial domains, calculating the sampling density of each three-dimensional unit based on the element values ​​and comprehensive gradient magnitude of the three-dimensional interference feature tensor, and generating a non-uniform sampling mask accordingly; Interference suppression and signal reconstruction are performed. Suppression weights are applied to the sampled signal according to the interference region, and an observation matrix is ​​constructed based on the non-uniform sampling mask. The target signal is reconstructed from the suppressed signal using a compressed sensing algorithm. The reconstructed target signal is used for target detection and parameter estimation, and the target's distance, velocity, and angle information are output. The construction of the three-dimensional joint interference feature characterization model specifically includes: calculating the signal energy within a sliding time window in the time domain, and when the signal energy exceeds a set threshold, applying a time-domain interference indicator function. Set to 1 otherwise; calculate the deviation of the power spectral density after short-time Fourier transform from the theoretical echo spectrum in the frequency domain. When the deviation exceeds a set threshold, the frequency domain interference indicator function is activated. Set to 1 otherwise set to 0; calculate the eigenvalue distribution and spatial spectrum of the spatial covariance matrix in the spatial domain, and activate the spatial interference indicator function when the spatial spectrum peak exceeds a set threshold. Set to 1 otherwise set to 0; the three-dimensional interference feature tensor The calculation formula is: ;in, 、 、 These are the weighting coefficients for the time domain, frequency domain, and spatial domain, respectively. For three-dimensional cross-coupling coefficients, For a normalized indexed coordinate system, where The sampling density of each three-dimensional element is calculated using the following sampling density allocation function: ;in, As the baseline sampling density, This is the interference intensity adjustment coefficient. This is the gradient adjustment coefficient. and These are the normalization factors for the tensor elements and the comprehensive gradient magnitude, respectively. This is a 3D linkage correction item. To synthesize the gradient magnitude; the three-dimensional linkage correction term Defined as: ;in, 、 、 These are the linkage and coupling gain coefficients for each dimension. For indicator functions, and These are specific regions in the frequency domain and time domain, respectively.

2. The vehicle-mounted radar anti-interference method based on time-frequency-space three-dimensional adaptive sampling according to claim 1, characterized in that, The construction of the three-dimensional joint interference feature characterization model also includes: processing the three-dimensional interference feature tensor. Perform low-rank approximate decomposition: ;in, 、 、 These are feature factor vectors in the time domain, frequency domain, and spatial domain, respectively. This represents the outer product operation. For tensor rank.

3. The vehicle-mounted radar anti-interference method based on time-frequency-space three-dimensional adaptive sampling according to claim 1, characterized in that, The non-uniform sampling mask is generated using the Poisson disk sampling algorithm, and the generated sampling point positions are... 、 The constraints are satisfied: ;in, Sampling points Local sampling density at a given location.

4. The vehicle-mounted radar anti-interference method based on time-frequency-space three-dimensional adaptive sampling according to claim 1, characterized in that, The weighting function for applying suppression weights to the sampled signal is: ;in, These are the coordinates of the sampling point. For the corresponding interference feature tensor value, Parameters used to control the intensity of suppression.

5. The vehicle-mounted radar anti-interference method based on time-frequency-space three-dimensional adaptive sampling according to claim 1, characterized in that, The process of reconstructing the target signal from the suppressed signal using compressed sensing algorithm includes: constructing an optimization problem: ;in, The target signal's sparse spectrum in the frequency domain is the solution. It is a sparse transformation basis. For the observation matrix corresponding to non-uniform sampling locations, This is the effective observation sample vector after interference suppression. It is an L1 norm.

6. A vehicle-mounted radar anti-interference system based on time-frequency-space three-dimensional adaptive sampling, characterized in that, include: The signal acquisition module is used to acquire the raw echo signal through the radar transmitting unit and the array receiving unit; the time-frequency-space three-dimensional joint interference feature characterization module is used to extract time-domain, frequency-domain and spatial-domain interference features, construct a three-dimensional interference feature tensor and identify the interference area; The adaptive sampling density adjustment module is used to calculate the gradient and comprehensive gradient magnitude of the three-dimensional interference feature tensor, allocate sampling density according to interference intensity and gradient information, and generate sampling masks; the interference suppression and signal reconstruction module is used to perform weighted suppression on signals in the interference area and reconstruct the target signal based on non-uniform sampling data using compressed sensing algorithm. The target detection and parameter estimation module is used to process the reconstructed signal to output target parameters; The construction of a three-dimensional joint interference feature characterization model specifically includes: calculating the signal energy within a sliding time window in the time domain, and when the signal energy exceeds a set threshold, applying a time-domain interference indicator function. Set to 1 otherwise; calculate the deviation of the power spectral density after short-time Fourier transform from the theoretical echo spectrum in the frequency domain. When the deviation exceeds a set threshold, the frequency domain interference indicator function is activated. Set to 1 otherwise set to 0; calculate the eigenvalue distribution and spatial spectrum of the spatial covariance matrix in the spatial domain, and activate the spatial interference indicator function when the spatial spectrum peak exceeds a set threshold. Set to 1 otherwise set to 0; the three-dimensional interference feature tensor The calculation formula is: ;in, 、 、 These are the weighting coefficients for the time domain, frequency domain, and spatial domain, respectively. For three-dimensional cross-coupling coefficients, For a normalized indexed coordinate system, where The sampling density of each three-dimensional element is calculated using the following sampling density allocation function: ;in, As the baseline sampling density, This is the interference intensity adjustment coefficient. This is the gradient adjustment coefficient. and These are the normalization factors for the tensor elements and the comprehensive gradient magnitude, respectively. This is a 3D linkage correction item. To synthesize the gradient magnitude; the three-dimensional linkage correction term Defined as: ;in, 、 、 These are the linkage and coupling gain coefficients for each dimension. For indicator functions, and These are specific regions in the frequency domain and time domain, respectively.

Images