High-dimensional information enhanced perception method based on chain decomposition

Abstract

The invention provides a high-dimensional information enhancement perception method based on chain decomposition, relates to the technical field of information enhancement, considers a single-station radar system working under a millimeter wave frequency, comprises an MIMO antenna array of a uniform rectangular array at a transmitting end and a receiving end, and comprises the following steps: establishing global correlation among different dimensions of a high-order signal; establishing a conversion relation between chain decomposition and CP decomposition, and introducing a minimum description length criterion to determine a chain decomposition rank number; correcting the ray tracking simulator based on the corner reflector; expanding a simulation scene, and generating a sensing environment and a sensing echo under a calibration physical condition; constructing a Riemannian manifold structure, and optimizing statistical distribution in combination with multiple geometric metrics; a retrieval problem is converted into a geometric mean value optimization problem, and the difference between a target and interference is maximized. According to the method, the perception echo performance is enhanced from the distance-speed-angle dimensions of the target object, so that the target object is clearer, and target object detection and parameter estimation are facilitated.

Description

Technical Field

[0001] This invention relates to the field of information enhancement technology, and more particularly to a high-dimensional information enhancement perception method based on chain decomposition. Background Technology

[0002] Millimeter-wave frequency-modulated continuous-wave radar is widely used in vehicle radar systems due to its advantages such as high detection accuracy, low implementation cost, and wide applicability. By flexibly adjusting the bandwidth and frame duration, high-precision range and velocity resolution can be obtained in both the time and frequency domains. Meanwhile, massively multi-input multiple-output (MIMO) antenna arrays provide spatial degrees of freedom for the radar in the angular domain, allowing for the creation of a virtual aperture by expanding the antenna array, thus achieving more accurate angular resolution. However, high-performance millimeter-wave radar still faces many challenges in practical applications, mainly including: To address the problem of low signal-to-noise ratio caused by the target object, windowing techniques are considered an ideal denoising filter. Typical signal window functions such as Hann, Hamming, Dolpher-Chebyshev, and Kaiser-Bessel are used, combined with certain signal transformations to reduce spectral leakage and lower noise levels. [9] In addition, some traditional noise reduction methods, including moving target indication and moving target detection,

[10] Background subtraction method

[11] Phase-shifted center-biased antenna technology

[12] These methods, such as [list of methods], focus on using Doppler characteristics for filtering and spectral analysis to suppress the influence of noise components. However, all of these signal processing methods require prior knowledge of the noise filtering band, which is difficult to obtain in advance and therefore cannot be applied to practical radar systems. Furthermore, methods based on wavelet transform, which do not require prior frequency band information, still have the problem of threshold selection, making it difficult to guarantee effectiveness in dynamically changing environments.

[0003] To address the surge in computational complexity caused by high-order data, MIMO radar applications require processing algorithms that balance high performance and accuracy with low computational complexity. Traditional noise reduction techniques are mostly based on matrix processing, failing to utilize high-order signal structures and incurring enormous computational costs. In recent years, tensor analysis methods have been applied to high-order data signal processing, with common methods including CP tensor (CANDECOMP / PARAFAC) decomposition. However, most CP decomposition algorithms rely on matrix transformation. Specifically, when the number of targets exceeds the number of antenna array elements, CP decomposition degenerates into matrix-based signal processing algorithms. These algorithms first convert tensors into matrix form and then transform the cost function of tensor decomposition into a matrix-approximate objective function. While this simplifies the optimization problem and iterative factor matrix process, it introduces the problem of tensor element permutation and cannot gain from the high-order tensor data structure. Furthermore, high-order data structures often involve high-dimensional computational costs. Even the least complex CP decomposition algorithm, alternating least squares, exhibits exponentially increasing computational cost with tensor order. Additionally, the CP-ALS algorithm is highly dependent on initial conditions, leading to local convergence. Furthermore, estimating the rank of CP is an NP-hard problem. Currently, there is no method to accurately estimate the tensor rank, and it often converges to a local minimum, making it difficult to find the optimal solution.

[0004] In terms of detection, target detection is often considered a binary hypothesis statistical testing problem. Traditional CFAR detection mainly achieves target detection by statistically modeling the clutter background, that is, by using single-point data from training units around the detection unit to estimate the interference level and comparing it with single-point data from the target unit. To overcome the problems of using single-point data, this invention adopts matrix CFAR detection, that is, modeling echo data as a positive definite covariance matrix of Emmetts. This set of matrices constitutes a Riemannian matrix manifold with a certain spatial structure, thereby establishing the connection between the geometric structure of the Riemannian manifold and target detection, transforming the statistical testing problem into a distance difference measurement problem on the manifold, providing a new technical approach for researching target detection problems and methods. Utilizing the high-dimensional transformation space characteristics of echo data, the echo data is modeled as a geometric mean tensor for estimating clutter power, and detection is based on information geometry manifold spatial transformation, aiming to reduce the impact of coherent interference and improve the robustness of detection.

[0005] In response to coherent interference, Professor Wen Yinghong's team at Beijing Jiaotong University studied the spectral spread characteristics of ghost targets in the design of sensing systems. [1] Professor Feng Zhiyong's team at Beijing University of Posts and Telecommunications proposed a coherent interference suppression method that combines frequency hopping technology and binary phase modulation. [2] Professor Graham M. Brooker of the University of Sydney, Australia, studied the statistical properties of coherent interference signals in the time, space, and frequency dimensions.[3] Professor Zheng Le's team at Beijing Institute of Technology studied the ghost detection problem in a multipath environment based on the binary composite hypothesis. [4] Professor André Bourdoux's team at KU Leuven in Belgium has proposed a low-complexity method for multipath target recognition and suppression of ghosting. [5] A team led by Professor Francesco Soldovieri of the Italian National Research Council has proposed a deep learning method based on convolutional neural networks and microwave tomography to achieve ghosting multipath filtering. [6] Professor Seongwook Lee's team at Korea Aerospace University designed a classifier based on deep neural networks to identify ghost images in the detection results. [7] A team led by Professor Félix Pérez-Martínez at the Polytechnic University of Madrid, Spain, has proposed a time-reversal-based method to eliminate ghosting in inverse synthetic aperture radar (ISAR) imaging. [8] Professor Yu Kegen's team at China University of Mining and Technology explored a non-line-of-sight multipath elimination method based on Taylor series. [9] Professor Marko Beko's team at the University of Lisbon mitigated the impact of non-line-of-sight multipath by solving the wide-area trust domain subproblem.

[10] Professor Cui Guolong's team at the University of Electronic Science and Technology of China explored a multipath ghosting recognition method based on multi-frame cumulative difference.

[11] Professor Xu Hongtao's team at Fudan University proposed a data-driven, end-to-end neural network-based method for eliminating radar ghost targets.

[12] .

[0006] In summary, low-complexity adaptive signal interference suppression and detection enhancement technologies are needed to ensure the effectiveness and feasibility of signal processing, so as to serve the high-precision and high-dimensional target object information perception of the Internet of Vehicles.

[0007] Reference documents: [1]Wang M, Wen Y, Wang Q, et al. Demonstration on range and dopplerspread of ghost target in automotive LFMCW radar[J]. IEEE Transactions on Instrumentation and Measurement , 2024, 73: 1-11. [2]Wang Y, Zhang Q, Wei Z, et al. Performance analysis of uncoordinated interference mitigation for automotive radar[J]. IEEETransactions on Vehicular Technology , 2023, 72(4): 4222-4235. [3]Brooker G M. Mutual interference of millimeter-wave radar systems[J]. IEEE Transactions on Electromagnetic Compatibility , 2007, 49(1): 170-181. [4]Zheng L, Long J, Lops M, et al. Detection of ghost targets forautomotive radar in the presence of multipath[J]. IEEE Transactions on Signal Processing , 2024, 72: 2204-2220. [5]Feng R, De Greef E, Rykunov M, et al. Multipath ghost recognitionand joint target tracking with wall estimation for indoor MIMO radar[J]. IEEE Transactions on Radar Systems, 2024, 2: 154-164. [6]Esposito G, Catapano I, Ludeno G, et al. A deep learning strategyfor multipath ghosts filtering via microwave tomography[J]. IEEE Transactions on Geoscience and Remote Sensing , 2023, 62: 1-14. [7]Jeong T, Lee S. Ghost target suppression using deep neural networkin radar-based indoor environment map[J]. IEEE Sensors Journal , 2022, 22(14):14378-14386. [8]de Arriba-Ruiz I, Pérez-Martínez F, Muñoz-Ferreras J M. Time-reversal-based multipath mitigation technique for ISAR images[J]. IEEE Transactions on geoscience and remote sensing , 2012, 51(5): 3119-3138. [9]Yu K, Wen K, Li Y, et al. A novel NLOS mitigation algorithm for UWB localization in harsh indoor environments[J]. IEEE Transactions on Vehicular Technology , 2018, 68(1): 686-699.

[10] Tomic S, Beko M. A robust NLOS bias mitigation technique for RSS-TOA-based target localization[J]. IEEE Signal Processing Letters , 2018, 26(1):64-68.

[11] Luo H, Zhu Z, Jiang M, et al. An effective multipath ghostrecognition method for sparse MIMO radar[J]. IEEE Transactions on Geoscience and Remote Sensing , 2023, 61: 1-11.

[12] Liu R, Song X, Qian J, et al. A data-driven method for indoor radar ghost recognition with environmental map[J]. IEEE Transactions on Radar Systems , 2024, 2: 910-923. Summary of the Invention In response to the technical problems mentioned in the background section, this invention provides a high-dimensional information-enhanced sensing method based on chain decomposition. By fusing information geometry and tensor processing, this invention offers a new paradigm for high-performance radar detection, thereby enhancing the separability of perceived targets and coherent interference.

[0008] The technical means employed in this invention are as follows: A high-dimensional information enhancement sensing method based on chain decomposition, considering a monostation radar system operating at millimeter-wave frequencies, including a MIMO antenna array with uniform rectangular arrays at both the transmitting and receiving ends, includes the following steps: S1. Establish the global correlation between different dimensions of high-order signals; obtain the four-dimensional noise reduction signal to be processed in the MIMO radar system. The four-dimensional noise reduction signal is obtained by mixing and discrete sampling the echo of the FMCWchirp signal through the virtual array unit of the radar receiver, wherein the index These correspond to the horizontal dimension, vertical dimension, sampling point dimension of the chirp signal, and chirp signal sequence dimension of the virtual array unit, respectively; S2. Establish the conversion relationship between chain decomposition and CP decomposition, and introduce the minimum description length criterion to determine the rank of chain decomposition; S3, Corner reflector-based ray tracing simulator; S4. Expand the simulation scene to generate the sensing environment and sensing echo under the calibrated physical conditions; S5. Construct a Riemannian manifold structure and optimize the statistical distribution by combining multiple geometric metrics; S6. Transform the retrieval problem into a geometric mean optimization problem to maximize the difference between the target and the interference.

[0009] Furthermore, in S1, the MIMO radar system employs time-division multiplexing MIMO technology, with a uniform rectangular array at the transmitting end and a uniform rectangular array at the receiving end; the number of virtual array elements is U×V, where... , , , These represent the number of antennas in the horizontal and vertical directions of the receiver array, respectively. , These represent the number of antennas in the horizontal and vertical directions of the transmitter array, respectively.

[0010] Furthermore, the horizontal and vertical spacing of the receiving array elements is... , Indicates the signal wavelength; the horizontal spacing of the transmitting array elements is... Vertical interval is , .

[0011] Further, S2 includes: S21, Expand the matrix; Soon Mapping to a matrix The There are elements, among which ; S22, Contraction operation; for order tensor and order tensor ,satisfy By performing a contraction operation, we obtain order tensor; S23, For a order tensor Decompose CP; S24, For a order tensor Its TT decomposition is expressed as: ; in, It is a third-order kernel tensor. and For matrix factors, It is denoted as TT rank.

[0012] Further, S5 includes the following steps: S51, Tensor for radar echo data Based on the relevant characteristics of the sensed echo data stored in each signal dimension unit, a sample HPD covariance tensor with modulus-n expansion is constructed; for each signal dimension unit, an HPD tensor is generated through multidimensional covariance estimation. ;in, To represent the signal dimension, the element is defined as: ; S52. To reduce computational complexity, the HPD tensor... Tensor chain decomposition is performed again for dimensionality reduction, preserving principal components to reduce redundant information between detection unit data. ; in For kernel tensors; S53. Project the original data onto a low-dimensional manifold subspace to enhance the sparsity of target features and the difference between the target and the interference on the Riemannian manifold; the manifold mapping process preserves the local geometric structure, so that the target and the interference are still discriminative after dimensionality reduction.

[0013] Furthermore, in S53, any point on the Riemannian manifold corresponds to an HPD tensor, and the shortest distance connecting two points is the geodesic distance; in the Riemannian manifold The similarity between any two points P and Q is quantified geometrically using any one of the following: (a) Riemann distance: ; in, Represents the matrix logarithm. Denotes the Frobenius norm; (b) Logarithmic Euclidean distance: ; (c) For two covariance tensors P and Q, the Kullback-Leibler divergence between them is derived as follows: ; Using Riemann distance, log-Euclidean distance, and the Kullback-Leibler separation metric as geometric measures, an optimization algorithm is used to calculate the geometric mean of the dimension-reduced tensor based on these geometric distances. This makes it the "best reference point". G Maximizing the geometric distance to the interfering samples can be expressed as: ; in, and These represent the sets of target and interference elements, respectively. This represents one of the aforementioned geometric measures.

[0014] Furthermore, in step S6, the retrieval problem is transformed into a geometric mean optimization problem, maximizing the difference between the target and the interference, given a test unit. The detection statistic is defined as: ; in, Interference obtained based on different distance metrics geometric mean , This represents the mean distance calculated using different distance metrics; if If the adaptive threshold is exceeded, the target is determined to exist.

[0015] Compared with the prior art, the present invention has the following advantages: The perception enhancement algorithm of this invention can effectively suppress background noise levels and enhance the perception echo performance from the dimensions of distance, velocity, and angle of the target object, making the target object clearer and facilitating target object detection and parameter estimation. By applying echo enhancement technology, noise power is significantly reduced and the signal-to-noise ratio of the target object is significantly improved.

[0016] Although data models can be modeled using CP tensor models, the computational cost increases with the tensor order due to the high complexity of the classic CP-ALS tensor decomposition method. It exhibits exponential growth. Meanwhile, the CP-ALS algorithm is highly dependent on initial conditions; estimating the rank of CP is an NP-hard problem, often converging to local minima and making it difficult to find the optimal solution. To address these issues, Tucker decomposition is considered an alternative to CP decomposition, which can reduce the computational cost of CPD to [missing information]. However, when the tensor order... At that time, the computational cost of Tucker decomposition remained high. In contrast, the algorithmic complexity of TT decomposition was O(n log n). It has lower computational cost when the tensor order is large, making it more suitable for real-time data processing in practical applications of millimeter-wave radar systems.

[0017] Due to the diversity of road environments and the complexity of the motion states of objects in a scene, millimeter-wave sensing systems face severe challenges in perceiving and understanding the scene. A data generation method based on ray-tracing simulation and real-data correction can exemplify the propagation of radio waves in the form of rays in the modeled 3D scene, bridging the gap between the feasibility of sensing signal acquisition and actual needs, and providing a reliable verification basis for subsequent detection and processing. Manifold geometry can amplify the inherent differences between targets and interference, and measured data shows an improvement in detection probability (compared to the traditional CFAR method). Optimization based on geometric mean suppresses noise and coherent interference, reducing the false alarm rate at low signal-to-noise ratios. The method is applicable to various radar systems (such as FMCW and pulse radar) and supports extension to higher-dimensional signals.

[0018] The effective improvement in the level and efficiency of information perception mentioned above is of great significance to the design and research of intelligent transportation vehicle networking systems, and provides important theoretical reference and key technical support for the development of intelligent and information-based transportation. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a schematic diagram of the overall process of the present invention.

[0021] Figure 2 This is a schematic diagram (I) showing a comparison of the speed-distance dimension before and after processing according to the present invention.

[0022] Figure 3 This is Schematic II, showing a comparison of the speed-distance dimension before and after processing in this invention.

[0023] Figure 4This invention utilizes simulation to generate data. Detailed Implementation

[0024] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0025] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0026] like Figures 1-4 As shown, this invention provides a high-dimensional information enhancement sensing method based on chain decomposition, considering a monostation radar system operating at millimeter-wave frequencies, including a MIMO antenna array with uniform rectangular arrays at both the transmitting and receiving ends, comprising the following steps: S1. Establish the global correlation between different dimensions of high-order signals; obtain the four-dimensional noise reduction signal to be processed in the MIMO radar system. The four-dimensional noise reduction signal is obtained by mixing and discrete sampling the echo of the FMCWchirp signal through the virtual array unit of the radar receiver, wherein the index These correspond to the horizontal dimension, vertical dimension, sampling point dimension of the chirp signal, and sequence dimension of the chirp signal, respectively, of the virtual array unit.

[0027] Consider a monostationary radar system operating at millimeter-wave frequencies, consisting of a MIMO antenna array composed of uniform rectangular arrays (URAs) at both the transmitter and receiver ends. The system employs the most common practical Time Division Multiplexing (TDM) MIMO technique, utilizing phase differences to construct a virtual antenna array. The system antenna array is a two-dimensional matrix, with the number of transmitter antennas being... The number of antennas at the receiving end is The array elements at the receiving end are spaced at half a wavelength in both the horizontal and vertical directions. ,Right now The horizontal and vertical spacing of the transmitting array elements are respectively and Therefore, the number of virtual array elements is ,in , .

[0028] The FMCW radar transmits a chirp signal, i.e., a linear frequency modulated continuous wave signal. The system's range and velocity resolution are adjusted by controlling the system bandwidth and signal duration. Assuming the array elements transmit the same chirp signal, where the th... Chirp signal for: (1) in , and These represent the starting frequency, initial phase, and chirp signal slope, respectively. and These represent the system bandwidth and chirp duration, respectively. Assume the received signal is from... The reflection of a point target consists of n points, then the nth point target... The signal received by each virtual array element is (2) in For the first The backscattering coefficients of a point target. After mixing, the intermediate frequency signal is obtained as follows: (3) Where, constant No. The complex amplitude of the intermediate frequency signal of a point target. Indicates the first Noise terms in the intermediate frequency signal of each virtual array element For the first The virtual array element received the first The chirp signal passes through the first The time delay of the reflection from a point target is... (4) in, , , and They represent the first The range, velocity, azimuth, and elevation angles of a point target. For intermediate frequency signals. Discrete sampling is performed, with a sampling interval of . , No. The first chirp signal Each sampling time point can be represented as Substituting the above equation into equation (3) yields the discrete sampled intermediate frequency signal: (5) Thus, the initial four-dimensional signal to be denoised has been obtained. To unify the dimensional index, the signal is re-expressed as (6) Among them, index Corresponding to , Indicates the noise term. Representing vectors The One element, and The guiding vector of the URA can be represented as: (7) (8) Here, the first The horizontal spatial frequency of a point target can be expressed as: The pitch spatial frequency can be expressed as Similarly, and Represented as (9) (10) in, .

[0029] Furthermore, S2, establish the conversion relationship between chain decomposition and CP decomposition, and introduce the minimum description length criterion to determine the rank of chain decomposition; As can be seen from equation (6), the received signal is affected by four dimensions: range, velocity, azimuth, and elevation. Traditional matrix representations are insufficient to achieve a concise and compact expression; therefore, a fourth-order tensor is introduced as a mathematical tool to describe the received signal. order tensor , its first The elements are The tensor operations involved are summarized as follows: S21, mode- Expansion operation. This operation yields a matrix. Soon Mapping to a matrix The There are elements, among which .

[0030] S22, Contraction operation. For order tensor and order tensor ,satisfy This can be obtained through the shrinking operation. order tensor, i.e. (11) S23, CP decomposition. For a order tensor Its CP decomposition can be expressed as:

[0031]

[0032] (12) In the formula This represents the outer product operation. For weight vectors, order tensor The The diagonal elements are , For the first dimensional matrix factor, The minimum value is denoted as the rank of CP.

[0033] Chain decomposition. For a... order tensor Its TT decomposition can be expressed as (13) In the formula, It is a third-order kernel tensor. and For matrix factors, It is denoted as TT rank.

[0034] Based on equation (6) and the tensor decomposition model described above, it can be seen that the intermediate frequency signal of MIMO radar... It has a high-order data structure, and the signal to be denoised can be represented as a CP tensor, i.e. (14) In the formula, Tensor and The Each element corresponds to a specific term in equation (6), i.e. , The echo signal after frequency mixing can be modeled as a CP tensor with a sparse structure; therefore, by enhancing... The sparsity of the structure can reduce the impact of noise.

[0035] Although echo data While CP models are modeled as tensors, current tensor decomposition algorithms for CP models suffer from several problems. For example, they can degenerate into matrix operations under certain conditions, making it difficult to extract gains from higher-order data structures; the increasing data dimensionality leads to a surge in computational complexity; and estimating the rank of a CP model is an NP-hard problem, making it difficult to obtain an accurate rank. These issues hinder the application of CP decomposition algorithms to real-world radar systems. Therefore, the TT tensor decomposition algorithm is introduced to make it suitable for practical radar signal processing applications.

[0036] Data processing of tensors based on TT decomposition essentially imposes low-rank constraints on the multilinear structure of the CP model, giving it a potential advantage in modeling low-quality data structures. Specifically, by processing tensors... Perform mode- Expand to obtain a matrix ,right Singular value decomposition is performed, and the few larger eigenvalues ​​obtained reflect... The low-rank nature of the sparsity. Therefore, the sparsity between different dimensions is characterized by constructing the following optimization problem, namely... (15) This invention employs the TT-SVD algorithm for TT decomposition, by introducing auxiliary variables. conduct Truncation of SVD operation ( -SVD, here First, regarding Perform mode- Expand and initialize Then proceed -SVD operation, i.e.

[0037] (16) in, and Represent matrices respectively and The Column and number OK, Represents the noise tensor mode- Expanded and Let each represent a unitary matrix composed of eigenvectors. Given a diagonal matrix, the first few eigenvalues ​​are sorted in descending order. The elements form its diagonal elements. Furthermore, in the... In the next iteration, an auxiliary variable is introduced. Used to update variables ,Right now (17) right conduct -SVD operation, i.e. Finally, the TT decomposition factor was obtained. (18) Because the traditional TT-SVD algorithm requires a pre-set precision as an input parameter to determine... The rank of the -SVD budget requires prior knowledge of the noise variance, which contradicts practical applications of noise reduction. Therefore, the rank cannot be determined by pre-setting the accuracy. This invention introduces the MDL criterion to adaptively estimate the TT rank. This criterion is widely used to determine the number of target signals and does not require any prior information or assumptions. First, by adjusting the variables in each iteration... Normalization preprocessing is performed to obtain Furthermore, its covariance matrix is ​​obtained. At this time, the MDL model is (19) For the iteration , , for The eigenvalues ​​are then used to obtain the rank of TT.

[0038] Based on the estimated low rank Substituting equation (19) into equation (18) and combining it with equation (13), we can finally obtain the observed signal. The noise-reduced echo signal is obtained by restoring the signal, and is represented as follows: .

[0039] S3, Corner reflector-based ray tracing simulator; S4. Expand the simulation scenario to generate the sensing environment and sensing echoes under calibrated physical conditions. In the actual test scenario, a trihedral corner reflector is used to correct the test scenario. A trihedral corner reflector with a constant radar cross section (RCS) is selected. Using the constructed radar hardware test platform, experiments are conducted in a relatively open area. Under the condition of a fixed incident angle, radar echo signals from two corner reflectors at different distances are acquired, and the relative echo power of the received signals from the corner reflectors under different test conditions is calculated through signal processing. Here, the target echo power model is based on the classic Friis formula for radar ranging, expressed as follows: (20) in, , and Let $\mathbf$ represent the distance between the target and the radar system, the radar system's transmit power, the radar system's transmitting antenna gain, the radar system's receiving antenna gain, the wavelength, and the RCS of a typical structure, respectively. Transforming the above equation, we can obtain the expression for RCS, namely: (twenty one) Because of the parameter terms in the above formula and The system parameters are independent of the target object information, determined solely by the performance of the measurement system, and remain constant throughout the measurement process. Therefore, the system parameters are defined using the above formula. The following expression: (twenty two) Finally, we can obtain the following expression: (twenty three) The above formula indicates that during multiple measurements while keeping the equipment parameters constant, due to the system parameters... The RCS of the target object is a constant value, and its calculation depends only on the received power. and the distance between the target and the radar system Both of the above variables can be obtained from radar echo signals; therefore, only the system parameters need to be determined. Then the calculated value of the target object's RCS can be obtained.

[0040] Based on the above-mentioned experimental system verification and calibration, ray tracing simulation was used to configure and simulate the same scene and RCS objects. A multi-antenna radar was used to acquire data based on the set scene, and a virtual array was implemented using time-division multiplexing technology. Based on the virtual array's metadata, angle domain information and range-angle spectrum within the radar's field of view were obtained. The simulator was reverse-engineered by comparing the simulation and experimental data before and after processing. After successful verification, the extended scene was changed, and dynamic echo data for the extended scene was generated.

[0041] S5. Construct a Riemannian manifold structure and optimize the statistical distribution using multiple geometric metrics; firstly, for the radar echo data tensor... (in For each signal dimension (e.g., distance, velocity, angle), a modulo-n expanded sample HPD covariance tensor is constructed based on the relevant characteristics of the sensed echo data stored in each signal dimension unit. For each signal dimension unit, an HPD tensor is generated through multidimensional covariance estimation. Its elements are defined as: (twenty four) This tensor satisfies Hermitian positive definiteness and constitutes a higher-order Riemannian manifold with a certain spatial structure. The manifold dimension of the points on the signal is determined by the signal structure.

[0042] To reduce computational complexity, the HPD tensor... Tensor chain decomposition is performed again for dimensionality reduction, preserving principal components to reduce redundant information between detection unit data. This can be represented as... (25) in The kernel tensor is used. After decomposition, the original data is projected onto a low-dimensional manifold subspace, enhancing the sparsity of target features and the difference between the target and interference on the Riemannian manifold. The manifold mapping process preserves the local geometric structure, ensuring that the target and interference remain discriminative after dimensionality reduction.

[0043] Any point on a Riemannian manifold corresponds to an HPD tensor, and the shortest distance between two points is the geodesic distance. Above, the similarity between any two points P and Q (corresponding to the covariance tensors of the target and the disturbance) is quantified by the following geometric measure: Riemann distance: (26) in, For the matrix logarithm, It is the Frobenius norm.

[0044] Logarithmic Euclidean distance: (27) For two covariance tensors P and Q, the Kullback-Leibler divergence between them is derived as follows: (28) Of the three metrics mentioned above, the Riemann distance is a fundamental method based on manifold geodesics, the log-Euclidean distance approximates the geodesic distance and is characterized by its computational efficiency, and the Kullback-Leibler divergence excels in measuring the difference in probability distributions. Therefore, based on the dimensionality-reduced low-dimensional manifold, we utilize the Riemann distance, log-Euclidean distance, and the information divergence metric Kullback-Leibler separation as geometric metrics, and employ an optimization algorithm to calculate the geometric mean of the dimensionality-reduced tensor under these geometric distances. This makes it the "best reference point". G Maximizing the geometric distance to the interfering samples can be expressed as: (29) in, and They represent sets of target and interference units, respectively. This is one of the aforementioned geometric metrics. The optimization problem is essentially finding an "optimal discriminant point" by optimizing the perturbation geometric mean. Position, maximizing the target unit and The degree of aggregation, while minimizing interference units and The degree of aggregation enhances the separability of the target from the interference.

[0045] In the Riemannian manifold Above, objective function Riemann gradient Defined as: (30) in, This is the gradient of the distance function on the manifold. Therefore, the gradient is calculated as follows, depending on the different distance metrics: (a) Riemann distance: (31) in, For the matrix logarithm, It is the Frobenius norm.

[0046] (b) Logarithmic Euclidean distance: (32) (c) For two covariance tensors P and Q, the Kullback-Leibler divergence between them is derived as follows: (33) Update using Riemann gradient ascent algorithm It can be represented as: (34) in, For Riemann index mapping, Let be the learning rate. Since direct calculation of the matrix square root and inverse needs to be avoided in practical computation, eigenvalue decomposition is usually used. Therefore, we can obtain Furthermore: (35) For a positive definite tensor manifold, the closed-form solution of the exponential mapping is: (36) S6. Transform the retrieval problem into a geometric mean optimization problem, maximizing the difference between the target and the interference. This transforms the traditional hypothesis testing problem into a statistical geometry problem in a manifold space, given a test unit. The detection statistic is defined as: (37) in, Interference obtained based on different distance metrics geometric mean , This represents the mean distance calculated using different distance metrics. If... If the threshold is exceeded, the target is considered to exist. The threshold can be dynamically adjusted based on the distribution characteristics on the manifold to adapt to environmental changes.

[0047] This invention provides a new paradigm for high-performance radar detection by integrating information geometry and tensor processing, thereby enhancing the separability of perceived targets and coherent interference.

[0048] The sequence numbers of the above embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments. In the above embodiments of the present invention, the descriptions of each embodiment have their own emphasis; parts not described in detail in a certain embodiment can be referred to in the relevant descriptions of other embodiments. It should be understood that the disclosed technical content in the several embodiments provided in this application can be implemented in other ways.

[0049] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A high-dimensional information enhancement perception method based on chain decomposition, characterized in that, Considering a monostatic radar system operating at millimeter-wave frequencies, including a MIMO antenna array with uniform rectangular arrays at both the transmitting and receiving ends, the following steps are included: S1. Establish the global correlation between different dimensions of high-order signals; obtain the four-dimensional noise reduction signal to be processed in the MIMO radar system. The four-dimensional noise reduction signal is obtained by mixing and discrete sampling the echo of the FMCWchirp signal through the virtual array unit of the radar receiver, wherein the index These correspond to the horizontal dimension, vertical dimension, sampling point dimension of the chirp signal, and chirp signal sequence dimension of the virtual array unit, respectively; S2. Establish the conversion relationship between chain decomposition and CP decomposition, and introduce the minimum description length criterion to determine the rank of chain decomposition; S3, Corner reflector-based ray tracing simulator; S4. Expand the simulation scene to generate the sensing environment and sensing echo under the calibrated physical conditions; S5. Construct a Riemannian manifold structure and optimize the statistical distribution by combining multiple geometric metrics; S6. Transform the retrieval problem into a geometric mean optimization problem to maximize the difference between the target and the interference.

2. The high-dimensional information enhancement perception method based on chain decomposition according to claim 1, characterized in that, In step S1, the MIMO radar system employs time-division multiplexing MIMO technology, with a uniform rectangular array at the transmitting end and a uniform rectangular array at the receiving end; the number of virtual array elements is U×V, where... , , , These represent the number of antennas in the horizontal and vertical directions of the receiver array, respectively. , These represent the number of antennas in the horizontal and vertical directions of the transmitter array, respectively.

3. The high-dimensional information enhancement perception method based on chain decomposition according to claim 2, characterized in that, The horizontal and vertical spacing of the receiving array elements is... , Indicates the signal wavelength; the horizontal spacing of the transmitting array elements is... Vertical interval is , .

4. The high-dimensional information enhancement perception method based on chain decomposition according to claim 1, characterized in that, S2 includes: S21, Expand the matrix; Soon Mapping to a matrix The There are elements, among which ; S22, Contraction operation; for order tensor and order tensor ,satisfy By performing a contraction operation, we obtain order tensor; S23, For a order tensor Decompose CP; S24, For a order tensor Its TT decomposition is expressed as: ； in, It is a third-order kernel tensor. and For matrix factors, It is denoted as TT rank.

5. The high-dimensional information enhancement perception method based on chain decomposition according to claim 1, characterized in that, S5 includes the following steps: S51, Tensor for radar echo data Based on the relevant characteristics of the sensed echo data stored in each signal dimension unit, a sample HPD covariance tensor with modulus-n expansion is constructed; for each signal dimension unit, an HPD tensor is generated through multidimensional covariance estimation. ;in, To represent the signal dimension, the element is defined as: ； S52. To reduce computational complexity, the HPD tensor... Tensor chain decomposition is performed again for dimensionality reduction, preserving principal components to reduce redundant information between detection unit data. ； in For kernel tensors; S53. Project the original data onto a low-dimensional manifold subspace to enhance the sparsity of target features and the difference between the target and the interference on the Riemannian manifold; the manifold mapping process preserves the local geometric structure, so that the target and the interference are still discriminative after dimensionality reduction.

6. The high-dimensional information enhancement perception method based on chain decomposition according to claim 1, characterized in that, In S53, any point on the Riemannian manifold corresponds to an HPD tensor, and the shortest distance connecting two points is the geodesic distance; in the Riemannian manifold The similarity between any two points P and Q is quantified geometrically using any one of the following: (a) Riemann distance: ； in, Represents the matrix logarithm. Denotes the Frobenius norm; (b) Logarithmic Euclidean distance: ； (c) For two covariance tensors P and Q, the Kullback-Leibler divergence between them is derived as follows: ； Using Riemann distance, log-Euclidean distance, and the Kullback-Leibler separation metric as geometric measures, an optimization algorithm is used to calculate the geometric mean of the dimension-reduced tensor based on these geometric distances. This makes it the "best reference point". G Maximizing the geometric distance to the interfering samples can be expressed as: ； in, and These represent the sets of target and interference elements, respectively. This represents one of the aforementioned geometric measures.

7. The high-dimensional information enhancement perception method based on chain decomposition according to claim 1, characterized in that, In step S6, the retrieval problem is transformed into a geometric mean optimization problem, maximizing the difference between the target and the interference, given a test unit. The detection statistic is defined as: ； in, Interference obtained based on different distance metrics geometric mean , This represents the mean distance calculated using different distance metrics; if If the adaptive threshold is exceeded, the target is determined to exist.

Images