A method for mmwave sensing signal enhancement under non-coherent interference

Abstract

The application provides a millimeter wave sensing signal enhancement method under non-coherent interference, relates to the technical field of wireless sensing, and comprises the following steps: constructing a multi-dimensional correlation characteristic and a system representation model based on high-dimensional signal sparsity; performing transformation domain multi-dimensional signal interference suppression and information extraction processing based on a low-rank structure; performing decomposition optimization iteration processing based on a least square idea; and constructing a region segmentation optimization design extended dimension target detection strategy. The method provided by the application can construct an interference decoupling framework of a high-dimensional signal model, expand a sensing range, and provide an extensible technical route for noise reduction and feature extraction of a high-dimensional sensing signal.

Description

Technical Field

[0001] This invention relates to the field of wireless sensing technology, and more particularly to a method for enhancing millimeter-wave sensing signals under incoherent interference. Background Technology

[0002] To address incoherent interference, Professor Alexander Yarovoy's team at Delft University of Technology in the Netherlands used the matrix-pencil principle to extract and reconstruct time-domain interference. [1] An interference suppression method in the range-Doppler domain is proposed based on intermediate frequency signal interpolation and phase matching. [2] Professor Cao Siyang's team at the University of Arizona designed an adaptive noise canceller to reduce noise power. [3] The teams led by Professor Chen Zezong from Wuhan University and Professor Chen Jinli from Nanjing University of Information Science and Technology proposed an interference suppression algorithm based on mode decomposition to improve radar detection performance. [4][5] A team led by Professor Seong-Cheol Kim from Seoul National University and a team led by Professor Xu Zhihuo from the University of Chinese Academy of Sciences proposed a method for suppressing mutual interference in radar sensing systems based on wavelet denoising. [6][7] A team led by Professor Ruud JG at Eindhoven University of Technology in the Netherlands has proposed a neural network-enhanced contraction thresholding algorithm for reconstructing time-domain sensing signals. [8] Professor Franz Pernkopf's team at the Technical University of Graz, Austria, quantified radar signal denoising and interference suppression based on convolutional neural networks. 9] Professor Hong Wei's team at Southeast University explored the temporal-domain sparsity of interference to suppress its effects.

[10] The methods described above can eliminate the effects of incoherent interference in the time and frequency domains, but they do not yet consider spatial information. To address high-dimensional interference across the time, frequency, and spatial domains, Professor Rainer Martin's team at Ruhr University Bochum in Germany explored the impact of a general signal model for eliminating interference in multi-channel interference scenarios on detection.

[11] Professor Cui Guolong's team at the University of Electronic Science and Technology of China proposed a high-resolution compressed sensing iterative algorithm based on angle domain interference characteristics to suppress vehicle-mounted radar interference.

[12] Professor Chen Yan's team at the University of Science and Technology of China proposed a feature-based unsupervised adaptive suppression network to alleviate signal interference between sensing devices.

[13] While the methods described above output 3D perception information, they lack the ability to mine the dimensional structure of the data. Furthermore, machine learning-based methods require a high degree of matching between the data and the scene, and involve significant computational and storage demands.

[0003] Millimeter-wave sensing enhancement technology, by decoupling the target from various interference signals, can significantly improve the system's environmental perception and target information extraction capabilities, and has broad application prospects. However, multiple interference couplings in complex electromagnetic environments will cause sensing devices to face problems such as high-dimensional signal feature aliasing, severe scene incoherent interference, and difficulty in extracting sensing target information. Exploration of sensing enhancement in high-dimensional signal space is still in its early stages, and existing sensing technologies still face many challenges, mainly including: Accurate modeling of effective features of high-dimensional sensing targets is difficult: Existing methods can eliminate the influence of incoherent interference in the time domain, frequency domain, and time-frequency and time-space two-dimensional space. However, the matrix-based signal processing framework suffers from low noise reduction efficiency due to insufficient high-dimensional feature representation and decoupling capability, making it difficult to achieve false alarm suppression of sensing targets under high-dimensional signal aliasing, which further limits the sensing range.

[0004] High-dimensional interference signals are difficult to handle when there is insufficient prior information about the scene: most existing filtering techniques require prior information about the target as input, which is difficult to match the application of filtering out incoherent interference in actual unknown scenes; in addition, the nonlinear expression ability of machine learning methods makes them more promising in solving this type of problem, but the key is whether a network architecture "homogeneous" with millimeter-wave high-dimensional signals can be constructed, and such methods have weak interpretability, making it difficult to provide theoretical support for subsequent improvement of perception performance.

[0005] High levels of incoherent interference can lead to missed detection of high-dimensional weakly perceived targets: Current research mainly focuses on constant false alarm rate (CFAR) detection. Common cell-averaged CFAR detection methods include range-velocity and range-angle detection. These methods use only two-dimensional range and velocity information and ignore important spatial information, which can result in a loss of detection performance.

[0006] Reference documents: [1] Wang J, Ding M, Yarovoy A. Matrix-pencil approach-basedinterference mitigation for FMCW radar systems[J]. IEEE Transactions onMicrowave Theory and Techniques, 2021, 69(11): 5099-5115. [2] Neemat S, Krasnov O, Yarovoy A. An interference mitigation technique for FMCW radar using beat-frequencies interpolation in the STFTdomain[J]. IEEE Transactions on Microwave Theory and Techniques, 2018, 67(3):1207-1220. [3] Jin F, Cao S. Automotive radar interference mitigation using adaptive noise canceller[J]. IEEE Transactions on Vehicular Technology, 2019,68(4): 3747-3754. [4] Chen Z, Xie F, Zhao C, et al. Radio frequency interferencemitigation in high-frequency surface wave radar based on CEMD[J]. IEEEGeoscience and Remote Sensing Letters, 2017, 14(5): 764-768. [5] Li Jiaqiang, Liu Haobo, Wang Xingyu, Yao Changhua, Chen Jinli. Anti-jamming method for vehicle-mounted radar based on successive variational mode decomposition and wavelet threshold [J / OL]. Radar Science and Technology, 1-13 [2025-03-20]. http: / / kns.cnki.net / kcms / detail / 34.1264.TN.20250108.1711.007.html. [6] Lee S, Lee JY, Kim S C. Mutual interference suppression using wavelet denoising in automotive FMCW radar systems[J]. IEEE Transactions on Intelligent Transportation Systems, 2019, 22(2): 887-897. [7] Xu Z, Yuan M. An interference mitigation technique for automotivemillimeter wave radars in the tunable Q-factor wavelet transform domain[J].IEEE Transactions on Microwave Theory and Techniques, 2021, 69(12): 5270-5283. [8] Overdevest J, Koppelaar A G C, Youn J, et al. Neurally-augmenteddeep unfolding for automotive radar interference mitigation[J]. IEEETransactions on Radar Systems, 2024, 2: 712-724. [9] Rock J, Roth W, Toth M, et al. Resource-efficient deep neuralnetworks for automotive radar interference mitigation[J]. IEEE Journal ofSelected Topics in Signal Processing, 2021, 15(4): 927-940.

[10] Wang Y, Huang Y, Wen C, et al. Mutual interference mitigationfor automotive FMCW radar with time and frequency domain decomposition[J].IEEE Transactions on Microwave Theory and Techniques, 2023, 71(11): 5028-5044.

[11] Chen S, Klemp M, Taghia J, et al. Improved target detection through DNN-based multi-channel interference mitigation in automotive radar[J]. IEEE Transactions on Radar Systems, 2023, 1: 75-89.

[12] Zhang H, Wei S, Wang M, et al. FUAS-Net: Feature-orientedunsupervised network for FMCW radar interference suppression[J]. IEEETransactions on Microwave Theory and Techniques, 2023, 72(4): 2602-2619.

[13] Yang S, Shang X, Zhang D, et al. IMIA: Interference mitigation via iterative approaches for automotive radar[J]. IEEE Transactions on RadarSystems, 2023, 1: 753-766. Summary of the Invention

[0007] As mentioned above, environmental noise such as rain, snow, and building reflections in the sensing scene will generate a large number of incoherent interference signals. These signals randomly superimpose with the target signal in the time, frequency, and spatial dimensions, resulting in a high degree of overlap between the sensing target and interference features in these dimensions. This makes it difficult to effectively separate the sensing target from the interference, further affecting the technical problem of extracting sensing target information. Therefore, this invention provides a method for enhancing millimeter-wave sensing signals under incoherent interference. This invention preserves the cross-dimensional signal correlation in time, space, and frequency based on the orthogonal Tucker tensor processing framework, improving the reliability of sensing in complex scenes. First, a multi-dimensional transform domain sensing representation model of sparse signals is constructed. Second, the low-rank structure of the signal is adaptively extracted for incoherent interference filtering. Finally, a region segmentation optimization detection method is designed based on the signal after interference filtering, comprehensively enhancing the performance of the sensing signal by integrating multi-dimensional information.

[0008] The technical means employed in this invention are as follows: A method for enhancing millimeter-wave sensing signals under incoherent interference includes the following steps: S1. Construct a multi-dimensional correlation characteristic and system representation model based on the sparsity of high-dimensional signals; S2. Multi-dimensional signal interference suppression and information extraction processing in the transform domain based on low-rank structure; S3. Decomposition and optimization iterative processing based on the least squares idea; S4. Construct an extended-dimensional target detection strategy based on region segmentation optimization design.

[0009] Further, S1 includes the following steps: S11. Establish approximate correlations between dimensions; S12, characterizes the coupling features of sparse signals and global correlation.

[0010] Furthermore, in S11, a single-station multiple-input multiple-output sensing system operating in the millimeter-wave band is considered, the system including: a uniform linear array of Each transmitting antenna element and Each receiving antenna element is in tandem; time-division multiplexing technology is used to achieve the inclusion of... A virtual array of elements, increasing the spatial degrees of freedom of the signal; To provide high-resolution target parameter estimation, a frequency-modulated continuous wave chirp signal is used as the transmitted signal; therefore, the same chirp signal is transmitted in the antenna element of each transmitter. , is represented as:

[0011] in, Indicates the index of the chip signal. , The starting frequency and initial phase of the chirp signal are given. The slope of the chirp signal. These represent the signal transmission bandwidth and period, respectively. Assuming ideal signal transmission conditions without incoherent interference, the scenario contains... If the first effective perception target is 1, then the second effective perception target is 1. The ideal sensed echo signal in each virtual antenna element is: ; in, Indicates by the first The sensing echo scaling factor caused by backscattering from the sensing target; Indicates the first The virtual antenna element received the first... The chip signal comes from the first chip. The round-trip delay of the reflected echo from a single sensing target is expressed as: ; in, , , They represent the first The distance, Doppler velocity, and azimuth angle of the target relative to the sensing system; After processing by the downconverter mixer, the first... The intermediate frequency signal in each virtual antenna element is represented as: .

[0012] Furthermore, the spacing between the antenna elements at the receiving end is half the operating wavelength, denoted as... The spacing between the antenna elements at the transmitting end is .

[0013] Furthermore, in S12, a generalized Tucker tensor decomposition model is introduced to represent the high-dimensional sensing signal, expressed as: ; in, , as well as Let represent the kernel tensor, factor matrix, and Tucker rank, respectively; Let Tucker's rank be kernel tensor For a diagonal tensor, its diagonal elements correspond to the first... Scaling factor for the backscattering of the target object's echo; for any tensor and arbitrary matrices Its modulus-p product can be expressed as ,and The elements in can be represented as ; The intermediate frequency signal enters the analog-to-digital converter (ADC) for discrete sampling, and the ADC output signal is modeled as a tensor: ; in, , Indicates the sampling interval; , and ;in Wavelength representation; orthogonal matrix , and These represent the tensor decomposition factor matrices for the distance, velocity, and azimuth angle domains, respectively.

[0014] Further, S2 includes the following steps: S21. Project the target energy onto the primary eigenvalue; , and Based on the signal modulus along the corresponding dimension Expanded, this is obtained by taking the average of the samples, and can be expressed as: ; ; ; right , and Performing EVD operations yields a Western matrix composed of the corresponding eigenvectors. , , and the , and The corresponding eigenvalues , and ; S22. Establish physical mapping and extract low-rank structure to suppress interference; the resulting three-dimensional low-rank filter is expressed as: = ; in, The truncated tensor can be expressed as follows: ; in, Tensor rank; kernel noise reduction processor The elements in are represented as: = .

[0015] Further, S3 includes the following steps: S31. Construct the optimization problem and set initial values; if the rank of the signal tensor is greater than 1, then it is necessary to decompose the matrix. , ,as well as Iterative updates are performed based on least squares regression; the decomposition matrix is ​​iteratively optimized by constructing the following optimization problem: ; st ; S32. Calculate the factor matrix based on alternating iterations; for each iteration, fix the other two decomposition matrices, and solve the objective matrix by minimizing the error: for each decomposition matrix... First, calculate the intermediate tensor variables. : ; Next to Perform R-truncated singular value decomposition, expressed as: = ; Among them, the first R left singular variables Construct the updated decomposition matrix , This represents a diagonal matrix, where the diagonal elements are the singular values ​​after decomposition. Composition; each iteration updates the three-dimensional decomposition matrix sequentially until the iteration termination condition is met: ; S33. Update the decomposition factors to improve decomposition accuracy; after satisfying the iteration termination condition and completing the matrix update, calculate... = The updated expression is: .

[0016] Further, S4 includes the following steps: S41. Dynamically segment the space between the training unit and the protection unit of the unit under test; based on the signal after noise filtering. In the CFAR detection with mean rejection, the power of the first M and last N guard units is removed when estimating the background clutter power, and the average value of the remaining training units is used as the interference background estimate; a detection mask is defined. The detection mask is initialized, that is... ; for Each unit under test in First, define its segmented protection unit area. Represent it as Secondly, calculate the regional power of the segmented protection unit area. Represent it as ; Simultaneously define the segmentation training unit. Represent it as Secondly, calculate the noise benchmark power in the training unit region. Represent it as ; S42. Optimize the detection area and update the detection threshold; based on the regional power of the segmented protection unit area. and the noise benchmark power of the training unit area The redundant interference power in the isolated segmentation process is calculated, and an adaptive threshold detection power threshold is calculated, which is expressed as: ; in, and This is the scaling factor; Compare the unit to be tested with the detection threshold, and update the detection mask accordingly. ; Detection mask =1, which means index ( At the distance, Doppler angle, and spatial angular position corresponding to ), there exists a reflecting target.

[0017] Compared with the prior art, the present invention has the following advantages: This invention proposes a method for enhancing millimeter-wave sensing signals under incoherent interference. First, it overcomes the denoising bottleneck caused by the high dimensionality and unstructured nature of millimeter-wave sensing data by leveraging the multi-dimensional correlation characteristics of sparse signals represented by orthogonal Tucker representations. Second, to achieve optimal denoising performance, transform domain signal denoising is performed based on a low-rank structure, and sensing performance is improved iteratively by minimizing the second-order error. Finally, considering the structure of the denoised sensing signal, an extended-dimensional target detection algorithm is designed based on region segmentation optimization to further reduce the impact of incoherent interference and improve robust detection performance. The proposed method can construct an interference decoupling framework for high-dimensional signal models, expand the sensing range, and provide a scalable technical route for denoising and feature extraction of high-dimensional sensing signals. Attached Figure Description

[0018] 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.

[0019] Figure 1 This is a schematic diagram of the overall process of the present invention.

[0020] Figure 2 This is a schematic diagram of the high-dimensional data transform domain signal enhancement processing under incoherent interference according to the present invention.

[0021] Figure 3 This is a diagram showing the SNR gain effect of the present invention for enhancing incoherent interference.

[0022] Figure 4 This diagram illustrates the improved detection rate achieved by the present invention in its incoherent interference enhancement processing. Detailed Implementation

[0023] 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.

[0024] 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.

[0025] like Figure 1 As shown, this invention provides a method for enhancing millimeter-wave sensing signals under incoherent interference. It can avoid the destruction of global correlation caused by matrix processing when facing high-dimensional data structures, and does not require prior scene information. It has stronger interference decoupling capabilities. At the same time, it designs a target information extraction strategy based on the decoupled data structure, which effectively improves the sensing signal-to-noise ratio and the accuracy of weak sensing target information extraction.

[0026] Includes the following steps: S1. Construct a multi-dimensional correlation characteristic and system representation model based on the sparsity of high-dimensional signals; S11. Establish approximate correlations between dimensions; Consider a single-station multiple-input multiple-output sensing system operating in the millimeter-wave band, in which... Each transmitting antenna element and A uniform linear array is composed of several co-located receiving antenna elements. Time-division multiplexing technology is used to achieve [the desired effect]. A virtual array of elements is constructed, increasing the spatial degrees of freedom of the signal. The spacing between the antenna elements at the receiving end is half the operating wavelength, denoted as . The spacing between the antenna elements at the transmitting end is To provide high-resolution target parameter estimation, a frequency-modulated continuous wave chirp signal is used as the transmitted signal. Therefore, the same chirp signal is transmitted in each antenna element at the transmitting end. It can be represented as: ; in, Indicates the index of the chip signal. , The starting frequency and initial phase of the chirp signal are given. The slope of the chirp signal. These represent the signal transmission bandwidth and period, respectively.

[0027] Assuming ideal signal transmission conditions without incoherent interference, the scenario contains... If the first effective perception target is 1, then the second effective perception target is 1. The ideal sensed echo signal in each virtual antenna element is ; in, For the first The sensing echo scaling factor caused by backscattering from the sensing target For the first The virtual antenna element received the first... The chip signal comes from the first chip. The round-trip delay of the reflected echo from a perceived target can be expressed as: ; in, , , They represent the first The distance, Doppler velocity, and azimuth angle of the target relative to the sensing system.

[0028] After processing by the downconverter mixer, the first... The intermediate frequency signal in a virtual antenna element can be represented as: .

[0029] S12, characterizes the coupling features of sparse signals and global correlation.

[0030] At this point, if the high-dimensional data is reshaped into a two-dimensional matrix, performing signal processing based on matrix operations will prevent subsequent algorithms from fully utilizing the multidimensional linear structure of the high-order data, leading to a decrease in algorithm performance. In high-dimensional signal processing, tensors have a natural high-dimensional data structure and are also a higher-order generalization of matrix eigenvalue decomposition and singular value decomposition. Therefore, a "time-frequency-space" high-dimensional sparse sensing space is constructed. Based on this, a generalized Tucker tensor decomposition model is first introduced to represent the high-dimensional sensing signal, expressed as: ; in, , as well as Let represent the kernel tensor, factor matrix, and Tucker rank, respectively. In this invention, let the Tucker rank be . kernel tensor For a diagonal tensor, its diagonal elements correspond to the first... The scaling factor for the backscattering of the echo from a target object. For any tensor and arbitrary matrices Its modulus-p product can be expressed as ,and The elements in can be represented as Since the perceived signal dimensions in this invention are time, frequency, and space, the signal processing is three-dimensional, i.e. .

[0031] The intermediate frequency signal enters the analog-to-digital converter for discrete sampling, and the output signal of the ADC is modeled as a tensor: ; in, , Indicates the sampling interval. (Note: The last part is a typo and can be left as is.) , and Used to simplify expressions, where Represents wavelength. Orthogonal matrix. , and These represent the tensor decomposition factor matrices for the distance, velocity, and azimuth angle domains, respectively.

[0032] Specifically, the vector in the factor matrix corresponding to the l-th objective and They can be represented as follows: ; ; ; Using tensors Represents the above combined noise term, where Let represent the elements in the noise tensor. Let the noise-free signal tensor be denoted as . The received signal can then be modeled as: .

[0033] S2. Multi-dimensional signal interference suppression and information extraction processing in the transform domain based on low-rank structure; S21. Project the target energy onto the main eigenvalue; , and It is possible to perform modal-based analysis on the signal along the corresponding dimension. Expanded, this is obtained by taking the average of the samples, and can be expressed as: ; ; ; Further , and Performing EVD operations yields a Western matrix composed of the corresponding eigenvectors. , , and the , and The corresponding eigenvalues , and .

[0034] S22. Establish physical mapping and extract low-rank structures to suppress interference. Finally, the obtained three-dimensional low-rank filter can be expressed as: = ; Here, To truncate the tensor, it can be expressed as follows: ; in, Represents the tensor rank. Kernel noise reduction processor. The elements in can be expressed by the following formula: = .

[0035] S3. Decomposition and optimization iterative processing based on the least squares idea; S31. Construct the optimization problem and set initial values; the above process can obtain ideal results based on the rank-1 assumption. If the rank of the signal tensor is greater than 1, then it is necessary to perform decomposition on the matrix. , ,as well as Iterative updates based on least squares regression improve decomposition accuracy.

[0036] Therefore, the decomposition matrix is ​​first iteratively optimized by constructing the following optimization problem: ; st ; For the optimization problem described above, the matrix factors calculated in the orthogonal Tucker are used as the initial iteration values, and optimization is performed based on these values.

[0037] S32. Calculate the factor matrix based on alternating iterations; for each iteration, fix the other two decomposition matrices, and solve the objective matrix by minimizing the error: for each decomposition matrix... First, calculate the intermediate tensor variables. : ; Next to Perform R-truncated singular value decomposition, and express it as: = ; Among them, the first R left singular variables Construct the updated decomposition matrix , It is a diagonal matrix, and the diagonal elements are composed of the singular values ​​after decomposition. Composition. Each iteration updates the three-dimensional decomposition matrix sequentially until the iteration termination condition is met: .

[0038] S33. Update the decomposition factors to improve decomposition accuracy; after satisfying the iteration termination condition and completing the matrix update, calculate... = The updated expression is: ; S4. Construct an extended-dimensional target detection strategy based on region segmentation optimization design; S41. Dynamically segment the space between the training unit and the protection unit of the unit under test; based on the signal after noise filtering. In the average-removal CFAR detection method, the power of the first M and last N guard units is removed when estimating the background clutter power, and the average value of the remaining training units is used as the interference background estimate. A detection mask is defined. The detection mask is initialized, that is... .

[0039] for Each unit under test in First, define its segmented protection unit area. Represent it as Secondly, calculate the regional power of the segmented protection unit area. Represent it as .

[0040] Simultaneously define the segmentation training unit. Represent it as Secondly, calculate the noise benchmark power in the training unit region. Represent it as .

[0041] S42. Optimize the detection area and update the detection threshold; based on the regional power of the segmented protection unit area. and the noise benchmark power of the training unit area The redundant interference power in the isolated segmentation quyu is calculated, and the adaptive threshold detection power threshold is calculated, which is expressed as: ; in, and This is the scaling factor.

[0042] Finally, the unit to be tested is compared with the detection threshold, and the detection mask is updated accordingly. ; Detection mask =1, which means index ( At the distance, Doppler angle, and spatial angular position corresponding to ), there exists a reflecting target.

[0043] As one embodiment of the present invention, firstly, for antenna configurations with 3 and 5 elements in the uniform linear arrays of the transmitting and receiving antennas respectively, the center frequency of the system is 77 GHz. In time-division multiplexing mode, the uniform linear arrays of the transmitting and receiving ends will generate a 1 A virtual uniform linear array of 15, with the element spacing in the array being... In the sensed echo signal received by each virtual array element, the echo signal reflected by the point scattering target is sampled along the fast time dimension (range dimension) and the slow time dimension (velocity dimension), respectively. The number of sampling points in the fast time dimension is 256 and the number of sampling points in the slow time dimension is 128, thereby obtaining the received signal tensor data. Figure 2This diagram illustrates the high-dimensional data transform domain signal enhancement processing under incoherent interference, as described in this invention. First, characteristic values ​​that significantly reflect the echo power of the target object are extracted from the transform domain. Since the prior input of traditional tensor decomposition is related to the interference power and is an unknown parameter in practical applications, an information content criterion is used to replace this input. Simultaneously, low-rank structural features of the transform domain of the perceived target signal are adaptively extracted. Next, equivalence conditions between the orthogonal Tucker tensor decomposition method and the optimal Tucker decomposition are established by quantifying the interference power. A physical mapping relationship is also established between the low-rank filtering interference cancellation method and the orthogonal Tucker tensor decomposition. To improve anti-interference performance under non-optimal conditions, an improved interference cancellation strategy based on the least squares approach is proposed. An optimization problem is constructed to reveal the impact mechanism of the alternating least squares iteration process on improving sensing performance. Redundancy is further utilized to reduce the interference level under non-ideal conditions. Based on the above processing, this invention then implements an extended-dimensional optimization detection method based on region segmentation in the original signal detection space. Green units represent the units to be tested, blue units represent protection units, and gray units represent training units. If the radar resolution is high enough that a potential target might occupy multiple range-velocity cells, more than one protection cell needs to be placed around the cell under test. In 3D processing, a reference window is applied to the 3D data matrix, and both the protection cell and the cell under test are expanded into row-column-page reference windows, corresponding to the range-velocity-angle dimensions, respectively. In the power spectrum of the output sensing signal based on the transform domain anti-interference strategy, the power of the sensing target cell in any dimension is higher than the power of the non-sensing target cell in any dimension, which will present a spatial cross structure centered on the sensing target. Therefore, this invention divides the window under test into regions and isolates the redundant interference power in the cross structure for adaptive threshold power calculation. This optimized design can eliminate the influence of aliasing interference signal cells, aiming to further improve the detection rate of 3D sensing targets.

[0044] Example 1 Please see Figure 3 This case study compares the interference suppression performance of target based on orthogonal Tucker decomposition and least-squares iterative optimization algorithms under different transmitter SNR conditions. The main factor causing the difference in target echo power is the different distances between the target and the radar system. The SNR range of the target's transmission point shown is -22 dB to -16 dB, with the number of targets... and half the dimension in each dimension As a pre-set rank The subplot in the figure highlights the SNR gain between the receiver SNR results of the noise reduction algorithm proposed in this chapter and the receiver without interference suppression algorithm. Because electromagnetic waves propagate in free space, signal path loss occurs. This loss is a large-scale fading that is a function of distance, and it occurs when mobile devices travel long distances. This type of fading is usually independent of frequency, thus causing a loss in the output SNR. As can be seen from the figure, under different transmitter SNR conditions, both noise reduction algorithms perform better when the rank is low. This indicates that reducing the rank within a certain range can effectively suppress the output noise signal, resulting in a higher output signal SNR. When the signal SNR is greater than 18 dB, the SNR gain tends to level off, better reflecting the algorithm's effect on improving SNR gain. At this point, the average SNR gain of the target based on orthogonal Tucker decomposition and the algorithm after iterative optimization are 10.89 dB and 11.58 dB, respectively, when the rank equals the number of targets; the latter provides 0.69 dB more SNR gain than the former.

[0045] Example 2 Please see Figure 4 This case study compares the detection performance of interference suppression based on orthogonal Tucker decomposition and least-squares iterative optimization under different transmitter SNR conditions, using both traditional CFAR detection and region segmentation detection strategies. Detection rates were compared across all simulated SNR conditions from -23 dB to -14 dB. When the region segmentation-based detector is applied to the tensor interference-suppressed signal, it consistently maintains a higher detection probability compared to the traditional CFAR method, especially showing a significant advantage in challenging low SNR scenarios. The target information extraction effect gradually improves from orthogonal Tucker decomposition to least-squares iterative optimization. High-performance detection up to 21% is achieved compared to traditional CFAR, as region detection utilizes subtle target features that are mistakenly considered noise loss in traditional CFAR operations. Even with two-dimensional denoising of two targets, the region segmentation detector can improve the detection rate at different SNRs, thereby improving perception performance. Furthermore, as the SNR decreases, the detection rate of region segmentation-based detection and transmission... 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.

[0046] 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 method for enhancing millimeter-wave sensing signals under incoherent interference, characterized in that, Includes the following steps: S1. Construct a multi-dimensional correlation characteristic and system characterization model based on the sparsity of high-dimensional signals: After sampling the intermediate frequency echo signal of the millimeter-wave MIMO sensing system, construct a three-dimensional sensing tensor with distance dimension, velocity dimension and orientation dimension. Introduce a generalized orthogonal Tucker tensor decomposition model to characterize the high-dimensional sensing signal and establish the coupling characteristics and global correlation between each dimension. S2. Transform domain multidimensional signal interference suppression and information extraction processing based on low-rank structure: The three-dimensional sensing tensor is expanded along each dimension by modulus-p, and the covariance matrix of each dimension is constructed and eigenvalue decomposition is performed to obtain the unitary matrix of the corresponding dimension; the physical mapping relationship between orthogonal Tucker decomposition and three-dimensional low-rank filtering is established, and the low-rank structure of the signal is extracted by truncating the kernel tensor to filter out non-coherent interference components. S3. Decomposition and optimization iterative processing based on the least squares idea: construct a tensor decomposition optimization problem with the goal of minimizing the decomposition error, and use an alternating iterative method to update the factor matrix and kernel tensor of each dimension in turn until the iteration termination condition is met, thereby improving the tensor decomposition accuracy and noise reduction effect. S4. Construct an extended-dimensional target detection strategy based on region segmentation optimization design: Perform dynamic region segmentation on the denoised 3D signal, divide the surrounding protection unit space and training unit space for each unit to be tested, and adaptively calculate the detection threshold by combining the power of the protection unit region and the noise power of the training unit to complete the target detection in the 3D space.

2. The method for enhancing millimeter-wave sensing signals under incoherent interference according to claim 1, characterized in that, S1 includes the following steps: S11. Establish approximate correlations between dimensions; S12, characterizes the coupling features of sparse signals and global correlation.

3. The method for enhancing millimeter-wave sensing signals under incoherent interference according to claim 2, characterized in that, In S11, consider a single-station multiple-input multiple-output sensing system operating in the millimeter-wave band, the system including: a uniform linear array of... Each transmitting antenna element and Each receiving antenna element is in tandem; time-division multiplexing technology is used to achieve the inclusion of... A virtual array of elements, increasing the spatial degrees of freedom of the signal; To provide high-resolution target parameter estimation, a frequency-modulated continuous wave chirp signal is used as the transmitted signal; therefore, the same chirp signal is transmitted in the antenna element of each transmitter. , is represented as: in, Indicates the index of the chip signal. , The starting frequency and initial phase of the chirp signal are given. The slope of the chirp signal. These represent the signal transmission bandwidth and period, respectively. Assuming ideal signal transmission conditions without incoherent interference, the scenario contains... If the first effective perception target is 1, then the second effective perception target is 1. The ideal sensed echo signal in each virtual antenna element is: ； in, Indicates by the first The sensing echo scaling factor caused by backscattering from the sensing target; Indicates the first The virtual antenna element received the first... The chip signal comes from the first chip. The round-trip delay of the reflected echo from a single sensing target is expressed as: ； in, , , They represent the first The distance, Doppler velocity, and azimuth angle of the target relative to the sensing system; After processing by the downconverter mixer, the first... The intermediate frequency signal in each virtual antenna element is represented as: 。 4. The method for enhancing millimeter-wave sensing signals under incoherent interference according to claim 3, characterized in that, The spacing between the antenna elements at the receiving end is half the operating wavelength, denoted as... The spacing between the antenna elements at the transmitting end is .

5. The method for enhancing millimeter-wave sensing signals under incoherent interference according to claim 2, characterized in that, In step S12, a generalized Tucker tensor decomposition model is introduced to represent the high-dimensional sensing signal, expressed as: ； in, , as well as Let represent the kernel tensor, factor matrix, and Tucker rank, respectively; Let Tucker's rank be kernel tensor For a diagonal tensor, its diagonal elements correspond to the first... Scaling factor for the backscattering of the target object's echo; for any tensor and arbitrary matrices Its modulus-p product can be expressed as ,and The elements in can be represented as ; The intermediate frequency signal enters the analog-to-digital converter (ADC) for discrete sampling, and the ADC output signal is modeled as a tensor: ； in, , Indicates the sampling interval; , and ;in Wavelength representation; orthogonal matrix , and These represent the tensor decomposition factor matrices for the distance, velocity, and azimuth angle domains, respectively.

6. The method for enhancing millimeter-wave sensing signals under incoherent interference according to claim 1, characterized in that, S2 includes the following steps: S21. Project the target energy onto the primary eigenvalue; , and Based on the signal modulus along the corresponding dimension Expanded, this is obtained by taking the average of the samples, and can be expressed as: ； ； ； right , and Performing EVD operations yields a Western matrix composed of the corresponding eigenvectors. , , and the , and The corresponding eigenvalues , and ; S22. Establish physical mapping and extract low-rank structure to suppress interference; the resulting three-dimensional low-rank filter is expressed as: = ； in, The truncated tensor can be expressed as follows: ； in, Tensor rank; kernel noise reduction processor The elements in are represented as: = 。 7. The method for enhancing millimeter-wave sensing signals under incoherent interference according to claim 1, characterized in that, S3 includes the following steps: S31. Construct the optimization problem and set initial values; if the rank of the signal tensor is greater than 1, then it is necessary to decompose the matrix. , ,as well as Iterative updates are performed based on least squares regression; the decomposition matrix is ​​iteratively optimized by constructing the following optimization problem: ； s.t. ; S32. Calculate the factor matrix based on alternating iterations; for each iteration, fix the other two decomposition matrices, and solve the objective matrix by minimizing the error: for each decomposition matrix... First, calculate the intermediate tensor variables. : ； Next to Perform R-truncated singular value decomposition, expressed as: = ； Among them, the first R left singular variables Construct the updated decomposition matrix , This represents a diagonal matrix, where the diagonal elements are the singular values ​​after decomposition. Composition; each iteration updates the three-dimensional decomposition matrix sequentially until the iteration termination condition is met: ; S33. Update the decomposition factors to improve decomposition accuracy; after satisfying the iteration termination condition and completing the matrix update, calculate... = The updated expression is: 。 8. The method for enhancing millimeter-wave sensing signals under incoherent interference according to claim 1, characterized in that, S4 includes the following steps: S41. Dynamically segment the space between the training unit and the protection unit of the unit under test; based on the signal after noise filtering. In the CFAR detection with mean rejection, the power of the first M and last N guard units is removed when estimating the background clutter power, and the average value of the remaining training units is used as the interference background estimate; a detection mask is defined. The detection mask is initialized, that is... ; for Each unit under test in First, define its segmented protection unit area. Represent it as Secondly, calculate the regional power of the segmented protection unit area. Represent it as ; Simultaneously define the segmentation training unit. Represent it as Secondly, calculate the noise benchmark power in the training unit region. Represent it as ; S42. Optimize the detection area and update the detection threshold; based on the regional power of the segmented protection unit area. and the noise benchmark power of the training unit area The redundant interference power in the isolated segmented region is calculated, and an adaptive threshold detection power threshold is applied, which is expressed as: ； in, and This is the scaling factor; Compare the unit to be tested with the detection threshold, and update the detection mask accordingly. ； Detection mask =1, which means index ( At the distance, Doppler angle, and spatial angular position corresponding to ), there exists a reflecting target.

Images