Non-cooperative signal detection method and device based on random tensor theory

Abstract

The invention relates to the technical field of wireless communication, and discloses a non-cooperative signal detection method and device based on a random tensor theory, a medium and equipment. According to the method, signals are collected and preprocessed through multiple antennas, high-dimensional tensor data are constructed, a tensor sample covariance matrix is calculated, eigenvalue distribution is analyzed, eigenvalue statistics are calculated, a detection threshold value is determined, and whether target signals exist or not is judged through comparison of the eigenvalue statistics and the threshold value. Updating the signal data in real time by adopting a dynamic window method, and repeating the step from tensorization processing to signal judgment to realize real-time monitoring and tracking of the target signal; the preprocessing is used for optimizing the data quality, and the high-dimensional tensor data is obtained by performing tensor processing on the preprocessed signal; priori information of a main user signal is not needed, the detection performance is excellent under the conditions of low signal-to-noise ratio, uncertain noise and the like, signals can be accurately recognized, the number change can be indicated, and the method is suitable for scenes of cognitive radio networks, radar detection and the like.

Description

Technical Field

[0001] This invention relates to the field of wireless communication technology and discloses a method and apparatus for detecting non-cooperative signals based on random tensor theory. Background Technology

[0002] With the rapid development of 6G and the Internet of Things (IoT) technologies, the number of communication devices has increased dramatically, leading to a growing demand for channel transmission rates and increasingly strained and scarce spectrum resources. To fully utilize limited spectrum resources, the concept of Cognitive Radio (CR) has emerged and gained widespread development. In this technology, unlicensed users or secondary users (SUs) need to monitor channel occupancy in real time, improving spectrum resource utilization through dynamic spectrum access. Secondary users must be aware of current spectrum activity in real time to determine optimal communication parameters and avoid interfering with primary users (PUs). Therefore, spectrum awareness is crucial in CR communication.

[0003] To achieve accurate perception of channel spectrum activity, numerous spectrum sensing methods have been proposed and studied, which can be broadly classified into three categories: Energy detection (ED) methods do not require prior information about the primary user signal and have low computational complexity, but their detection performance is poor at low signal-to-noise ratios (SNR) and it is difficult to determine an appropriate threshold.

[0004] Matched Filter Detection (MFD) method: Compared with the ED algorithm, it has a lower SNR tolerance, but it requires complete prior information about the primary user signal, such as physical implementation and structure. In practice, due to implementation issues, the detection performance often degrades to varying degrees.

[0005] Feature detection (FD) methods: These methods detect the primary user signal by designing appropriate feature extraction techniques. Common methods include cyclostationary feature (CF) methods, wavelet transform (WT) methods, machine learning (ML) methods, and deep learning (DL) methods. Traditional CF and WT methods have low tolerance for signal-to-noise ratio (SNR) and require complete or partial prior information about the primary user signal. ML and DL methods typically do not require prior information, but they perform poorly at low SNR and have poor interpretability, resulting in low reliability and usability in practical implementations.

[0006] In summary, existing spectrum sensing methods have shortcomings in scenarios with low signal-to-noise ratios and lack of prior information, and there is an urgent need for a new signal detection method to meet practical application requirements. Summary of the Invention

[0007] The purpose of this invention is to provide a non-cooperative signal detection method, apparatus, medium, and device based on random tensor theory, to address the poor detection performance of existing spectrum sensing methods under conditions of low signal-to-noise ratio, uncertain noise, or lack of prior information from the primary user. By constructing a high-dimensional tensor sample covariance matrix from the received multi-antenna sample signals, and using random tensor theory to analyze the eigenvalue distribution and set an appropriate detection threshold, rapid and accurate detection and localization of potential primary user signals can be achieved, improving the efficiency and real-time performance of spectrum sensing. High-dimensional tensor mapping enhances signal feature recognition, and adaptive threshold detection is achieved by combining statistical theory, ultimately realizing efficient and real-time detection and tracking of target signals.

[0008] To achieve the above objectives, the present invention provides the following technical solution: According to one aspect of the present invention, a method for detecting non-cooperative signals based on random tensor theory is provided, comprising: Acquire multi-channel unknown signals and preprocess the signals to optimize data quality; Tensor quantization is performed on the preprocessed signal to map the low-dimensional signal into high-dimensional tensor data; Calculate the covariance matrix of the high-dimensional tensor data; Analyze the eigenvalue distribution of the covariance matrix and calculate the eigenvalue statistics to extract signal features; The detection threshold is determined based on statistical theory, and the presence of a target signal is determined by comparing the feature value statistics with the threshold. The signal data is updated in real time using a dynamic window method, and the above steps of tensor quantization processing to signal judgment are repeated to achieve real-time monitoring and tracking of the target signal.

[0009] According to one embodiment of the present invention, the preprocessing includes noise reduction processing and normalization processing, wherein the noise reduction processing employs moving average filtering, and the normalization processing converts the signal to the [0,1] or [-1,1] interval to eliminate dimensional differences.

[0010] According to one embodiment of the present invention, the tensor quantization process specifically involves: dividing the multi-channel signal at each time step into k sub-vectors of equal length, where k ≥ 2, and mapping the sub-vectors to high-dimensional random vectors through a k-order tensor product operation. And the sub-vector length satisfies M=n×k, where M represents the multi-channel signal dimension and n represents the sub-vector length.

[0011] According to one embodiment of the present invention, the covariance matrix is ​​calculated in the following manner: The covariance matrix is ​​constructed using the high-dimensional vectors in each column of the high-dimensional tensor data matrix through a rank-1 outer product operation. The specific calculation formula is: ф ,in, This indicates the order of a high-dimensional random variable. This is the weighting coefficient, which is usually set to 1. Let be the mapped high-dimensional random vector; ф represents the covariance matrix, which is a high-dimensional random vector at N sampling times. and The sum of the products of transposes; According to one embodiment of the present invention, the eigenvalue statistic is the linear eigenvalue statistic LES, defined as: ,in, Let i be the i-th eigenvalue of the covariance matrix. Here is a preset test function, where n represents the length of the subvector and k is the number of subvectors; According to one embodiment of the present invention, when determining the detection threshold based on statistical theory, the formula for calculating the standardized statistic is as follows: ; in, The theoretical expectation of the LES statistic is derived based on the theory of random tensors. The theoretical standard deviation of the LES statistic; According to one embodiment of the present invention, the size of the dynamic window is adaptively adjusted according to the signal change rate, the size of the dynamic window ranges from 50 to 500 sampling times, and the sliding step ranges from 1 to 10 sampling times.

[0012] On the other hand, the present invention also provides a non-cooperative signal detection device based on random tensor theory to implement the method, comprising the following steps: The signal acquisition module is used to receive unknown wireless signals through multiple channels and acquire raw signal data; The preprocessing module is used to perform noise reduction and standardization on the raw signal data and output an optimized signal matrix; The tensor quantization module is used to perform segmentation and tensor product operations on the optimized signal matrix, mapping low-dimensional signals into high-dimensional tensor data. The covariance calculation module is used to calculate the tensor sample covariance matrix based on high-dimensional tensor data. The feature analysis module is used to analyze the eigenvalue distribution of the covariance matrix and calculate linear eigenvalue statistics. The threshold detection module is used to standardize the eigenvalue statistics according to the central limit theorem, determine the threshold based on the preset confidence level, and determine whether the target signal exists. The real-time tracking module is used to dynamically update signal data through a sliding window, control the tensor quantization module to repeatedly execute operations from the threshold detection module, and realize the real-time monitoring and tracking of the target signal.

[0013] On the other hand, the present invention also provides a computer storage medium storing instructions that, when executed, implement the non-cooperative signal detection method based on random tensor theory.

[0014] On the other hand, the present invention also provides a computing device, including a processor and a communication interface coupled to the processor; the processor is used to run computer programs or instructions to implement the non-cooperative signal detection method based on random tensor theory.

[0015] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. No prior information required: The method of this invention is entirely data-driven and does not require any prior information about the master user signal, making it suitable for non-cooperative scenarios.

[0016] 2. Can indicate changes in signal quantity: It can not only detect the main user signal, but also indicate changes in the number of signals.

[0017] 3. Excellent performance at low signal-to-noise ratio: By using high-dimensional tensor mapping and tensor product operations, the feature recognition of target signals is enhanced at low signal-to-noise ratios, improving detection sensitivity by 10% to 20% compared to traditional methods (such as random matrix theory); 4. Sensitive and robust detection: By using tensor product to map low-dimensional signals to high-dimensional space, the sensitivity to abnormal signals is enhanced, making it more sensitive to the detection of main user signals and more robust to complex noise environments. This improves detection accuracy while effectively reducing false detection rate.

[0018] 5. Real-time performance and adaptability: The dynamic window mechanism can adjust parameters according to signal changes, balancing detection accuracy and real-time performance, and is suitable for diverse scenarios such as sudden signals and continuous signals.

[0019] 6. Applicable to multiple scenarios: It can be applied to the detection and identification of unknown or "non-cooperative" radio signals in cognitive radio networks, and can also be extended to scenarios involving large-scale data analysis and signal detection, such as radar detection, IoT spectrum management, and network security anomaly detection. Attached Figure Description

[0020] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this invention, illustrate exemplary embodiments of the invention and are used to explain the invention, but do not constitute an undue limitation of the invention. In the drawings: Figure 1 This is a complete flowchart of the non-cooperative signal detection method based on random tensor theory of the present invention, from signal acquisition to real-time tracking; Figure 2 This is a schematic diagram of a non-cooperative signal detection system based on random tensor theory according to the present invention; Figure 3 This is a trend diagram of the LES curve under different signal-to-noise ratios for a non-cooperative signal detection method based on random tensor theory according to the present invention. Figure 4 This is a comparison chart of the detection performance of the random tensor theory-based non-cooperative signal detection method RTT and the random tensor theory-based RMT method, which are proposed in this invention. Detailed Implementation

[0021] To facilitate a clear description of the technical solutions in the embodiments of the present invention, the terms "first" and "second" are used to distinguish identical or similar items with essentially the same function and effect. For example, the first threshold and the second threshold are merely used to distinguish different thresholds and do not limit their order. Those skilled in the art will understand that the terms "first" and "second" do not limit the quantity or execution order, and that the terms "first" and "second" are not necessarily different.

[0022] It should be noted that in this invention, the terms "exemplary" or "for example" are used to indicate examples, illustrations, or descriptions. Any embodiment or design described as "exemplary" or "for example" in this invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of terms such as "exemplary" or "for example" is intended to present the relevant concepts in a concrete manner.

[0023] In this invention, "at least one" refers to one or more, and "more than one" refers to two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, or B alone, where A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one" or similar expressions refer to any combination of these items, including any combination of singular or plural items. For example, "at least one of a, b, or c" can represent: a, b, c, a combination of a and b, a combination of a and c, a combination of b and c, or a, b, and c, where a, b, and c can be single or multiple.

[0024] like Figure 1 As shown, the complete process of a non-cooperative signal detection method based on random tensor theory of the present invention, from signal acquisition to real-time tracking, includes the following steps: Step S11: Signal acquisition and preprocessing to obtain the preprocessed data matrix; The signal acquisition specifically involves: receiving unknown wireless signals from multiple antennas (i.e., multiple channels) in a cognitive radio network to obtain a set of sampled signal data, forming a spatiotemporal data matrix; each antenna receives signals from N consecutive sampling times to form an M×N spatiotemporal data matrix; The preprocessing includes noise reduction and standardization.

[0025] The noise reduction process specifically involves using a moving average filter to eliminate high-frequency noise from the obtained spatiotemporal data matrix. The standardization process involves standardizing the denoised data, specifically by standardizing the signal to the range of [0,1] or [-1,1] to eliminate the dimensional differences between different channels, converting it to a uniform scale to eliminate dimensional differences, and ensuring data consistency.

[0026] The standardization process employs Z-score standardization, which transforms the signal data into a distribution with a mean of 0 and a standard deviation of 1, eliminating dimensional differences while preserving the statistical distribution characteristics of the data. This applies to the signal value of the m-th channel at the t-th sampling time in a multi-channel signal. The standardized value is: ,in, Let be the mean of the m-th channel signal, and the range of the mean is the sampled data within the current dynamic window; Let be the standard deviation of the m-th channel signal. Z-score normalization ensures that the preprocessed signal data conforms to a normal distribution, which aligns well with subsequent statistical analysis logic. This reduces the interference of non-normally distributed data on LES statistic calculations and improves the accuracy of threshold determination. In low signal-to-noise ratio (e.g., -20dB scenarios), Z-score normalization can be further amplified by high-dimensional tensor mapping after normalization, avoiding the loss of weak signal features. Z-score can be calculated in real-time based on the local mean within the current window, without relying on the extreme values ​​of the entire data set, making it particularly suitable for real-time preprocessing in bursty signal scenarios.

[0027] The standardization process employs L2 normalization. For a multi-channel signal vector at a given sampling time t... Where M is the number of channels. The normalized vector is: .

[0028] in, The L2 normalization method removes amplitude interference and focuses on directional features, enabling a clearer distinction between the "cooperative direction of the signal" and the "random direction of noise" in a high-dimensional tensor space, thus improving feature recognition at low SNR. Even if the noise amplitude changes randomly (such as sudden impulse noise), the direction of the normalized vector can still retain the relative relationship of the signal, avoiding misjudgment caused by sudden changes in noise amplitude.

[0029] Step S12: Construct a high-dimensional tensor data matrix; The high-dimensional tensor data is specifically obtained by tensorizing the preprocessed data matrix to achieve high-dimensional expansion of the preprocessed data matrix. That is, step S12 amplifies the feature differences between the target signal and noise through high-dimensional mapping.

[0030] Step S12 includes the following sub-steps: S12A. Divide the M-dimensional antenna received signal at each of the N consecutive sampling times into k sub-vectors of length n; the length n and the number k of the sub-vectors satisfy n×k=M; M represents the multi-channel signal dimension, n represents the sub-vector length, and k represents the number of sub-vectors; S12B: Mapping subvectors to high-dimensional random vectors through a k-order tensor product operation; specifically: performing a tensor product operation on k standardized subvectors to form a vector of dimension n. k High-dimensional random vectors; S12C arranges high-dimensional random vectors from N consecutive sampling times into a high-dimensional tensor data matrix.

[0031] Step S13: Construct the tensor sample covariance matrix through the outer product operation; The high-dimensional vectors in each column of the high-dimensional tensor data matrix The covariance matrix is ​​constructed by an outer product operation with rank 1; The rank-1 outer product operation specifically involves: for a single sampling time... High-dimensional random vectors Calculate its conjugate transpose with itself. In wireless communication scenarios, signals are often in complex form, so the conjugate transpose is used to ensure the Hermitian property of the covariance matrix. The outer product matrix, with a rank of 1, is the basic building block of the covariance matrix. The outer product result not only contains the second-order moment information of the signal, but also indirectly preserves the spatiotemporal coupling characteristics of the original multi-channel signal through high-dimensional mapping. It is the key to distinguishing weak signals from noise and avoids the problem of losing key features in traditional low-order outer products.

[0032] The specific calculation formula is as follows: ; in, This is the weighting coefficient, which is usually set to 1. This is the high-dimensional vector after mapping in step S12. Let the covariance matrix of the nth sub-vector of the k-th high-dimensional vector be denoted by , and let N be high-dimensional random vectors at sampling times. and The sum of the transpose products; the covariance matrix Preserving the higher-order statistical properties of the signal, it can better characterize the essence of the signal than traditional matrices, so as to further extract signal features; the final covariance matrix, through the superposition of N low-rank units, can not only preserve the characteristics of the high-dimensional signal at each time, but also highlight the "high-order correlation characteristics of the signal" through the superposition effect, which is significantly different from the random characteristics of noise.

[0033] Step S14: Eigenvalue statistics and analysis to obtain standardized statistics; Based on the obtained tensor sample covariance matrix, the eigenvalue distribution is calculated to identify the presence of abnormal signals (i.e., master user PU signals); the actual eigenvalue distribution is calculated according to theoretical spectrum laws (such as the MP (Marchenko-Pastur) law and the circular law). For example, the Marchenko-Pastur (MP) law can be used to calculate the theoretical distribution range of eigenvalues ​​when noise dominates, i.e., the noise eigenvalue interval. For a matrix of dimension pxN, N is the number of sampling times, and the theoretical lower and lower bounds of the noise eigenvalues ​​are as follows: , in This represents the noise variance. The eigenvalues ​​are compared with those obtained from the actual solution. and It can filter out signal feature values ​​that are outside the noise range.

[0034] The Circular Law is applicable to scenarios where noise is not independent and identically distributed or where the signal distribution is more complex after high-dimensional tensor mapping. It can describe the theoretical distribution shape of noise eigenvalues, such as a circular region on the complex plane, and can be further adapted to scenarios where noise is uncertain.

[0035] The standardized statistic is the Linear Eigenvalue Statistic (LES). By linearly combining the eigenvalues ​​of the covariance matrix, higher-order statistical properties of the signal are extracted to distinguish target signals from noise under low signal-to-noise ratio conditions. When there is no target signal, the high-dimensional vector of noise is randomly distributed, and the eigenvalues ​​of the outer product matrix are scattered and have small amplitudes. When a target signal is present, the higher-order correlation characteristics of the signal make the high-dimensional vector exhibit regularity, and the eigenvalues ​​of the outer product matrix show significant peaks. This difference can be further amplified by the LES statistic, ultimately supporting the accuracy of threshold detection. The Linear Eigenvalue Statistic (LES) is used to comprehensively statistically analyze the eigenvalues ​​of the covariance matrix. The LES statistic is defined as follows: ; Where nk is the dimension of the high-dimensional tensor data, n represents the length of the sub-vector, and k is the number of sub-vectors; Let i be the i-th eigenvalue of the covariance matrix. The preset test function is used to enhance the feature weights corresponding to the target signal and improve the detection sensitivity; preferably, the preset test function is a logarithmic function or a quadratic function.

[0036] Step S15: Determine the threshold and detect the signal, including the following sub-steps: S15A: Standardizing the LES statistic according to the Central Limit Theorem (CLT) yields the standardized statistic: , in, and These are the expected value and standard deviation of the LES statistic, respectively. S15B: Determine the detection threshold for the LES statistic based on the set confidence level; Preferably, the confidence level is greater than or equal to 85%, and specifically in this embodiment, the confidence level is selected as 95% or 99%. S15C: When the standardized statistic exceeds the set detection threshold, a PU signal is determined to exist; otherwise, no signal is determined. In order to achieve real-time and accurate signal detection, a signal detection threshold is obtained based on statistical theory.

[0037] PU signals typically refer to Primary User (PU) signals. In cognitive radio networks, spectrum resources are allocated to PUs, while Secondary Users (PSUs) need to use the spectrum without interfering with PUs. The presence or absence of a PU signal determines whether a PSU can use a particular frequency band. The time-domain characteristics of PU signals need to be adapted to high-dimensional tensor feature extraction logic. In non-cooperative scenarios, the start time and data content of PU signals in the time domain are random. However, core features such as modulation patterns and fundamental periods are relatively fixed. Therefore, they can still be detected without prior information. Fixed features can be captured through high-dimensional tensors, filtering out random noise interference. The energy of PU signals is concentrated in the licensed carrier frequency and sideband range in the frequency domain, rather than being randomly distributed across a wide frequency band like noise. This energy concentration causes significant peaks in the eigenvalues ​​of the tensor covariance matrix, making it clearly distinguishable from the smooth eigenvalue distribution of noise.

[0038] Step S16: Real-time tracking and updating; In practical applications, a sliding window method is used to achieve real-time monitoring and tracking of PU signals: a dynamic window mechanism is adopted, and the window size (50~500 sampling times) and sliding step size (1~10 sampling times) are adaptively adjusted according to the signal change rate. The data window slides continuously in the time domain, and steps S12~S15 are repeated after each window update. Based on the real-time calculated trend of LES statistics, the system enables real-time tracking and rapid response to the appearance, persistence, or disappearance of the target signal (PU signal). Specifically, the dynamic window adaptive adjustment adjusts the window size according to the rate of change of the LES statistics: when the rate of change of LES > a threshold (e.g., 0.1), the window shrinks to 50-200 sampling times to improve real-time performance; when the rate of change of LES ≤ the threshold, the window expands to 200-500 sampling times to improve detection accuracy. The sliding step size is adjusted proportionally to the window size at a ratio of 1:50; for example, when the window size is 50, the step size is 1, and when the window size is 500, the step size is 10.

[0039] like Figure 2 The diagram shows a non-cooperative signal detection device based on random tensor theory, used to implement the method described above. The device is characterized by comprising: The signal acquisition module is used to receive unknown wireless signals through multiple channels and acquire raw signal data; The preprocessing module is used to perform noise reduction and standardization on the raw signal data and output an optimized signal matrix; The tensor quantization module is used to perform segmentation and tensor product operations on the optimized signal matrix, mapping low-dimensional signals into high-dimensional tensor data. The covariance calculation module is used to calculate the tensor sample covariance matrix based on high-dimensional tensor data. The feature analysis module is used to analyze the eigenvalue distribution of the covariance matrix and calculate linear eigenvalue statistics. The threshold detection module is used to standardize the eigenvalue statistics according to the central limit theorem, determine the threshold based on the preset confidence level, and determine whether the target signal exists. The real-time tracking module is used to dynamically update signal data through a sliding window, control the tensor quantization module to repeatedly execute operations from the threshold detection module, and realize the real-time monitoring and tracking of the target signal.

[0040] This invention introduces random tensor theory (RTT) into non-cooperative signal detection, replacing the traditional random matrix theory (RMT). It is an upgrade of the order of the signal representation dimension, rather than a simple dimensional expansion.

[0041] Existing RMT methods can only characterize the second-order statistical properties of signals (such as the covariance matrix), failing to capture higher-order correlation information of multi-channel signals, such as spatiotemporal coupling features. The method proposed in this invention uses a k-order tensor product mapping, specifically dividing the low-dimensional signal into k sub-vectors satisfying M=n×k, and then mapping them to an nᵏ-dimensional high-dimensional vector through tensor product, directly preserving the higher-order statistical properties of the signal. This processing is not dimensional superposition, but rather amplifies the essential difference between signal and noise through the tensor structure. Noise is more likely to exhibit randomness in a high-dimensional tensor space, while the correlation characteristics of the signal are more prominent.

[0042] While tensor theory is a mathematical tool, its application to non-cooperative signal detection, especially in scenarios with no prior information and low SNR, has not yet been explored. The proposed method essentially addresses the core contradiction of traditional low-order matrix representations' inability to distinguish weak signals from noise through high-order structural representations. This requires a creative connection between mathematical theory and the engineering needs of signal detection. Tensor quantization provides a high-dimensional data foundation for tensor covariance, LES normalization provides an adaptive basis for threshold detection, and dynamic windows offer engineering feasibility for real-time tracking. This combination represents an innovative solution designed for the entire process of non-cooperative signal detection, requiring a deep understanding of the entire chain of signal detection, including data processing, feature extraction, decision-making, and real-time tracking.

[0043] This invention discloses a non-cooperative signal detection method based on random tensor theory. In practice, its performance advantages under different signal-to-noise ratio (SNR) conditions are verified through Monte Carlo experiments. Before the Monte Carlo experiments, data is generated through simulation using the following data model: ; in, Let represent the signal data received by the i-th antenna at time t, where p is the number of analog primary user (PU) signals. For factor loading, Main user factor, For noise; , The elements in the middle follow a standard normal distribution, that is... , ~N(0,1), This indicates the control noise level; different signal-to-noise ratio levels (-20dB, -10dB, -5dB, 5dB, 10dB, 20dB) and the number of primary user signals (0, 1, 2) are set to generate data respectively; The Monte Carlo experiment procedure includes the following steps: Step 21, Preprocessing: The generated data is preprocessed by dividing it into sub-vectors, performing tensor product mapping to a high-dimensional space, and generating a tensor data matrix: each received signal vector is divided into k sub-vectors of length n and standardized. Step 22, Tensorization: Perform tensor product operation to generate high-dimensional vectors. For example, divide the 28-dimensional signal at each time step into two 14-dimensional sub-vectors, and map them to a 14×14=196-dimensional high-dimensional vector through second-order tensor product. Step 23, covariance matrix calculation: Using the high-dimensional vector obtained in step 22, calculate the tensor sample covariance matrix. : in, This is the weighting coefficient, which is usually set to 1. It is a high-dimensional random vector after mapping. Let the covariance matrix be a high-dimensional random vector at N sampling times. and The sum of the products of transposes.

[0044] Step 24, Eigenvalue Statistics: Calculate the linear eigenvalue statistic (LES) based on the tensor covariance matrix to obtain the LES-t detection curve, as shown below. Figure 3 As shown, observe the changing trend of the LES curve under different signal-to-noise ratios; from Figure 3 As can be seen, when the channel is idle, the LES curve remains stable at a low level. When the primary user signal appears, the LES curve rises significantly, and the magnitude of the rise gradually increases with the increase of the signal-to-noise ratio. When a new signal appears, the curve shows a rise of different magnitudes. It is worth noting that even at extremely low signal-to-noise ratios (such as -20dB), the method proposed in this invention can still effectively identify the primary user signal, verifying the robustness of the method to noise in complex electromagnetic environments.

[0045] It can still effectively detect the primary user signal even at an SNR of -20dB (extremely low signal-to-noise ratio). Figure 3As shown, even at -20dB, the LES curve still rises significantly after the main user signal appears, improving the detection sensitivity by 10% to 20% compared to the traditional RMT method. Figure 4 The results show that the LES curve of RTT changes much more significantly than that of RMT, making it easier to identify signals when they appear, thus solving the industry problem of weak signal detection in complex electromagnetic environments.

[0046] The method of the present invention is further verified based on simulation data from a real communication environment. Specifically, this includes the following steps: Step 31, Scene Setup: In the simulation environment, the primary user signal adopts amplitude modulation (AM) mode with a carrier frequency of 702kHz; there are 28 receiving antennas for secondary users, each user is sampled 1000 times, and the sampling rate is 4MHz; the signal-to-noise ratio is set to -20dB; the PU signal is set to appear from the 501st sampling time.

[0047] Step 32, Data Window and Tensor Processing: A 28×200 data window is used to sequentially extract data from the simulation dataset in a sliding manner; the data within the window is mapped to a high-dimensional tensor space by tensor product to obtain a high-dimensional data vector.

[0048] Step 33, Tensor covariance matrix and eigenvalue calculation: The tensor covariance matrix of each data window is calculated, and then the LES index is calculated and the LES-t curve is plotted in real time. To facilitate comparison of the detection effects of Random Matrix Theory (RMT) and the Random Tensor Theory (RTT) method proposed in this patent, the curve is standardized, as shown in the attached figure. Figure 3 As shown.

[0049] from Figure 4 As can be seen, the LES curve remains relatively stable before the 500th sampling time, indicating the absence of a primary user signal. At the 501st sampling time, the LES curve increases rapidly, accurately indicating the presence of the PU signal. At the same time, it can be seen that the LES curve based on RTT changes more significantly than that based on RMT, demonstrating the advantages of the RTT method proposed in this patent in terms of sensitivity to primary user signals and robustness to noise.

[0050] This invention is not only applicable to spectrum sensing in core cognitive radio networks, but can also be extended to areas such as weak target signal detection, IoT spectrum management (signal identification of massive devices), and network security anomaly detection (abnormal traffic signal location). This scenario extension is not a simple reuse, but is based on the core advantages of no prior knowledge, low SNR robustness, and real-time tracking, reflecting the versatility and value extension of the technical solution.

[0051] Furthermore, an exemplary embodiment of the present invention may also provide a computer-readable storage medium storing a computer program. This computer-readable storage medium stores a computer program that, when executed by a processor, causes the processor to perform a non-cooperative signal detection method based on random tensor theory according to an exemplary embodiment of the present invention. This computer-readable recording medium is any data storage device capable of storing data read by a computer system. Examples of computer-readable recording media include: read-only memory, random access memory, read-only optical disk, magnetic tape, floppy disk, optical data storage device, and carrier waves (such as data transmission via the Internet through wired or wireless transmission paths).

[0052] Furthermore, an exemplary embodiment of the present invention may also provide a computing device. The computing device includes a processor and a memory. The memory stores a computer program. The computer program is executed by the processor, causing the processor to execute a computer program for a non-cooperative signal detection method based on random tensor theory according to an exemplary embodiment of the present invention.

[0053] Although the invention has been described herein in conjunction with various embodiments, those skilled in the art will understand and implement other variations of the disclosed embodiments by reviewing the accompanying drawings, disclosure, and other materials. In this specification, the word "comprising" does not exclude other components or steps, and "a" or "an" does not exclude multiple components. A single processor or other unit can implement several functions listed in the specification. While certain measures are described in different embodiments, this does not mean that these measures cannot be combined to produce good results.

[0054] Although the invention has been described in conjunction with specific features and embodiments, it is obvious that various modifications and combinations can be made therein without departing from the spirit and scope of the invention. Accordingly, this specification and drawings are merely illustrative of the invention and are considered to cover any and all modifications, variations, combinations, or equivalents within the scope of the invention. Clearly, those skilled in the art can make various alterations and modifications to the invention without departing from its spirit and scope. Thus, if such modifications and modifications fall within the scope of the invention and its equivalents, the invention is also intended to include such modifications and modifications.

Claims

1. A non-cooperative signal detection method based on random tensor theory, characterized in that, Includes the following steps: Acquire multi-channel unknown signals and preprocess the signals to optimize data quality; Tensor quantization is performed on the preprocessed signal to map the low-dimensional signal into high-dimensional tensor data; Calculate the covariance matrix of the high-dimensional tensor data; Analyze the eigenvalue distribution of the covariance matrix and calculate the eigenvalue statistics to extract signal features; The detection threshold is determined based on statistical theory, and the presence of a target signal is determined by comparing the feature value statistics with the threshold. The signal data is updated in real time using a dynamic window method, and the above steps of tensor quantization processing to signal judgment are repeated to achieve real-time monitoring and tracking of the target signal.

2. The non-cooperative signal detection method based on random tensor theory according to claim 1, characterized in that, The preprocessing includes noise reduction and normalization. The noise reduction uses moving average filtering, and the normalization transforms the signal to the [0,1] or [-1,1] interval to eliminate dimensional differences.

3. The non-cooperative signal detection method based on random tensor theory according to claim 1, characterized in that, The tensor quantization process specifically involves dividing the multi-channel signal at each time step into k sub-vectors of equal length, where k ≥ 2, and mapping these sub-vectors to high-dimensional random vectors through a k-order tensor product operation. And the sub-vector length satisfies M=n×k, where M represents the multi-channel signal dimension and n represents the sub-vector length.

4. The non-cooperative signal detection method based on random tensor theory according to claim 1, characterized in that, The covariance matrix is ​​calculated in the following manner: The covariance matrix is ​​constructed using the high-dimensional vectors in each column of the high-dimensional tensor data matrix through a rank-1 outer product operation. The specific calculation formula is as follows: ,in, This is the weighting coefficient, which is usually set to 1. It is a high-dimensional random vector after mapping; Let the covariance matrix be a high-dimensional random vector at N sampling times. and The sum of the products of transposes.

5. The non-cooperative signal detection method based on random tensor theory according to claim 1, characterized in that, The eigenvalue statistic is the linear eigenvalue statistic LES, defined as follows: ,in, Let i be the i-th eigenvalue of the covariance matrix. Here is the preset test function, where n represents the length of the subvector and k is the number of subvectors.

6. The non-cooperative signal detection method based on random tensor theory according to claim 5, characterized in that, When determining the detection threshold based on statistical theory, the formula for calculating the standardized statistic is as follows: ,in, The theoretical expectation of the LES statistic is derived based on the theory of random tensors. Let be the theoretical standard deviation of the LES statistic.

7. The non-cooperative signal detection method based on random tensor theory according to claim 1, characterized in that, The size of the dynamic window is adaptively adjusted according to the rate of signal change. The size of the dynamic window ranges from 50 to 500 sampling times, and the sliding step ranges from 1 to 10 sampling times.

8. A non-cooperative signal detection device based on random tensor theory, used to implement the method described in claims 1 to 7, characterized in that, include: The signal acquisition module is used to receive unknown wireless signals through multiple channels and acquire raw signal data; The preprocessing module is used to perform noise reduction and standardization on the raw signal data and output an optimized signal matrix; The tensor quantization module is used to perform segmentation and tensor product operations on the optimized signal matrix, mapping low-dimensional signals into high-dimensional tensor data. The covariance calculation module is used to calculate the tensor sample covariance matrix based on high-dimensional tensor data. The feature analysis module is used to analyze the eigenvalue distribution of the covariance matrix and calculate linear eigenvalue statistics. The threshold detection module is used to standardize the eigenvalue statistics according to the central limit theorem, determine the threshold based on the preset confidence level, and determine whether the target signal exists. The real-time tracking module is used to dynamically update signal data through a sliding window, control the tensor quantization module to repeatedly execute operations from the threshold detection module, and realize the real-time monitoring and tracking of the target signal.

9. A computer storage medium, characterized in that, The computer storage medium stores instructions that, when executed, implement the non-cooperative signal detection method based on random tensor theory as described in any one of claims 1 to 7.

10. A computing device, characterized in that, It includes a processor and a communication interface coupled to the processor; the processor is used to run computer programs or instructions to implement the non-cooperative signal detection method based on random tensor theory as described in any one of claims 1 to 7.

Images