A collaborative spectrum sensing method and system based on symmetric low-rank subspace clustering

By employing a cooperative spectrum sensing method based on symmetric low subspace clustering, the signal matrix is ​​reduced in dimensionality and a classifier is constructed. This solves the problem of poor spectrum sensing performance in multi-master user and low signal-to-noise ratio environments, and achieves efficient spectrum sensing in complex environments.

CN116707682BActive Publication Date: 2024-11-12CHENGDU QUANYI INTELLECTUAL PROPERTY OPERATION CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202310661061.1
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2023-06-05
Publication Date
2024-11-12
Estimated Expiration
2043-06-05

AI Technical Summary

Technical Problem

Existing technologies have poor spectrum sensing performance and high computational complexity in environments with multiple primary users and low signal-to-noise ratio, and cannot accurately detect spectrum status.

Method used

A cooperative spectrum sensing method based on symmetric low subspace clustering is adopted. By constructing a cognitive wireless system, the signal matrix is ​​processed in reduced dimension, and the K-means algorithm is used for cluster training to construct a classifier to improve sensing performance.

Benefits of technology

In low signal-to-noise ratio and complex environments, it reduces computational complexity and improves the accuracy and detection efficiency of spectrum sensing, making it suitable for multi-master user scenarios.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN116707682B_ABST
    Figure CN116707682B_ABST
Patent Text Reader

Abstract

A collaborative spectrum sensing method and system based on symmetric low-rank subspace clustering, firstly constructing a cognitive wireless system for collaborative spectrum sensing, including multiple primary users and multiple secondary users, then fusing the signal vectors collected by the secondary users into a signal matrix, then performing dimensionality reduction processing on the signal matrix to obtain several low-rank and structurally symmetric coefficient matrices, extracting the maximum eigenvalue of its covariance matrix, taking the maximum eigenvalue as the eigenvector, jointly forming a sample set and dividing it into a training set and a test set, constructing a classifier using the K-means algorithm, and finally using the constructed classifier to classify all eigenvectors of the test set to evaluate the system perception performance. The present invention constructs a wireless communication system with multiple primary users, and processes the signal matrix through a symmetric low-rank representation, so as to propose a spectrum sensing algorithm that is more suitable for complex communication environments, and can ensure perception accuracy and detection efficiency under low signal-to-noise ratio conditions.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of cognitive radio communications, and in particular to a collaborative spectrum sensing method and system for symmetric low subspace clustering. Background Art

[0002] At present, the rapid development of emerging technologies such as autonomous driving and drones has placed higher and higher performance requirements on wireless communication networks, while the scarcity of spectrum resources has imposed certain restrictions on the development of wireless communication technology. In order to solve this problem, cognitive radio technology that can dynamically allocate spectrum has been proposed. Cognitive radio technology can improve spectrum utilization and alleviate the current situation of tight spectrum resources. Spectrum sensing is the key technology of cognitive radio, which is used to collect spectrum information and detect spectrum status.

[0003] According to the number of secondary users, spectrum sensing technology can be divided into single-user spectrum sensing and multi-user collaborative spectrum sensing. In particular, multi-user collaborative spectrum sensing shares local sensing data and integrates effective data with the cooperation of multiple secondary users to finally obtain the result of global decision-making. According to the number of sensing antennas when the secondary user receives the signal in the sensing environment, spectrum sensing research can be divided into single-antenna sensing and multi-antenna sensing. Compared with single-antenna spectrum sensing, multi-antenna spectrum sensing can make full use of the spatial dimension in the sensing environment, and the increase in antenna diversity will improve the system's sensing accuracy and sensing performance to a certain extent.

[0004] Clustering algorithms in machine learning can be used to solve the classification problem of whether there is a primary user signal in the frequency band. Spectrum sensing research based on clustering methods can effectively improve the accuracy of spectrum sensing and avoid the derivation of complex thresholds. However, in the current spectrum sensing research based on clustering algorithms, the sum of signal energy, the eigenvalues ​​of the signal matrix, or the geometric distance between signals are generally used as sample features, and then clustering algorithms are used to classify spectrum signals into two categories: the presence or absence of primary users. For example, a joint spectrum sensing method based on a clustering algorithm is disclosed in the prior art. First, a sampling matrix is ​​input; after processing, a covariance matrix is ​​obtained; then, the sum of the maximum eigenvalue and the minimum eigenvalue of the covariance matrix and the difference between them are taken, and then, the sampling matrix of the input signal to be tested and the sampling matrix in the presence and absence of the primary user signal are processed to obtain their corresponding difference vectors and sum vectors; the center point of each case is determined; the points in the set of the presence and absence of the primary user signal are clustered by the K-means++ clustering algorithm to obtain the cluster centers respectively; finally, the center point of the sampling matrix of the signal to be tested is compared with the cluster center point, so as to determine whether the center point to be tested contains the primary user information.

[0005] However, on the one hand, the solution in the prior art has high computational complexity and cannot obtain accurate data under low signal-to-noise ratio conditions; on the other hand, for multi-antenna collaborative spectrum sensing research, the spectrum signal received by the secondary user is usually high-dimensional and contains redundant information, and the features of the directly extracted perception signal cannot accurately represent the signal, resulting in a decrease in the perception performance of the cognitive wireless system. In addition, the rapid development of wireless communications has made communication networks increasingly dense, and the appearance of multiple primary user signals in communication networks has gradually become a normal phenomenon, especially when the multi-antenna system adopts spatial multiplexing technology, or when multiple independent primary users access the same frequency channel, multiple primary user signals usually appear in the communication network. In general, an increase in the number of primary users can easily lead to an increase in interference signals, thereby increasing the difficulty for secondary users to correctly access idle frequency bands. Therefore, eliminating various interferences in complex environments to improve detection accuracy and reducing the computational complexity of spectrum sensing are of great significance in improving spectrum sensing utilization, thereby solving the problem of poor spectrum sensing performance in multiple primary users and low signal-to-noise ratio environments. Summary of the invention

[0006] In order to reduce the computational complexity of spectrum sensing and solve the problem of poor spectrum sensing performance in an environment with multiple primary users and low signal-to-noise ratio, the present invention proposes a collaborative spectrum sensing method and system based on symmetric low subspace clustering, which reduces the dimension of sample data through symmetric low-rank representation and can improve the system's perception performance and detection accuracy in the case of multiple primary users and low signal-to-noise ratio.

[0007] According to a first aspect of the present invention, a collaborative spectrum sensing method based on symmetric low subspace clustering is provided, comprising:

[0008] S1. Construct a cognitive wireless system for collaborative spectrum sensing, the cognitive wireless system comprising a plurality of primary users and a plurality of secondary users, the primary users communicating with the secondary users via an access frequency band, each secondary user being equipped with a plurality of sensing antennas for collecting a signal vector, the signal vector being composed of a noise signal generated during the communication process or any primary user signal superimposed with a noise signal;

[0009] S2. Fusing the signal vectors collected by any user through multiple sensing antennas into a signal matrix;

[0010] S3. Performing dimensionality reduction processing on the signal matrix to obtain several symmetric coefficient matrices with low rank and symmetric structure;

[0011] S4. extracting the covariance matrix of each of the symmetric coefficient matrices, and then extracting the maximum eigenvalue of the covariance matrix, taking the maximum eigenvalue as the eigenvector, and jointly forming a sample set, and dividing the sample set into a training set and a test set;

[0012] S5. Perform clustering training on the feature vectors in the training set using the K-means algorithm, and output the cluster centroid; construct a classifier using the cluster centroid;

[0013] S6. Use the classifier to classify all feature vectors of the test set, compare the classification results with the actual results of the test set, and evaluate the system perception performance.

[0014] Preferably, in step S1, the number of primary users is M, the number of secondary users is N, the number of sensing antennas of each secondary user is A, the total number of signal sampling points collected by the secondary users in the sampling time slot is d, and the signal collected by each sensing antenna is regarded as a signal vector. Let n = 1, 2, 3, ..., N, and let sampling point t = 1, 2, 3, ..., d, then the sampled signal expression of the nth secondary user is:

[0015]

[0016] Among them, H1 means that the primary user signal exists, and the frequency band is unavailable; H0 means that the primary user signal does not exist, and the frequency band is available. na represents the main user signal, and the signal is an amplitude modulation signal, η na is a product with a mean of 0 and a covariance of Gaussian white noise signal, h m,na (t)~CN(0,d -α ) is the channel gain from the mth primary user to the nth secondary user, m = 1, 2, 3…M, a represents the sensing antenna channel selected during signal transmission, and the channel gain depends on the distance d between the primary user and the secondary user -α , where α is the path loss exponent.

[0017] According to the above technical means, a wireless communication system with multiple primary users can be completely constructed to propose a spectrum sensing method that is more suitable for complex communication environments.

[0018] Preferably, in step S2, the signal vector collected by the nth secondary user through the ath sensing antenna is represented as x na =[x na (1),x na (2),…,x na (d)] T , then the matrix form of the signal vector collected by n secondary users through all antennas is:

[0019] Among them, X n ∈R d×A .

[0020] Preferably, in step S3, the signal matrix is ​​subjected to dimensionality reduction processing by means of a symmetric low-rank representation mathematical model, and the mathematical model for solving the symmetric coefficient matrix with the lowest rank and symmetric structure is:

[0021]

[0022]

[0023] Among them, Z n ∈R d×A is the signal matrix X n ∈R d×A The symmetric and low-rank coefficient representation, E n is the error term, l 1,2 The norm makes the error term infinitely close to 0, and λ>0 is used to balance the error term E n and low-rank structure Z n , Y1 and Y2 are Lagrange multipliers; μ is the penalty parameter, and μ>0.

[0024] According to the above technical means, the signal vector matrix can be reduced in dimensionality to reduce the computational complexity of subsequent steps.

[0025] Preferably, the symmetric low-rank representation model is solved using an alternating direction multiplier iteration method to obtain a symmetric coefficient matrix Z n .

[0026] According to the above technical means, the alternating direction multiplier iteration method is used for solving, and the complex model in the above steps can be decomposed into multiple simple small problems, thereby reducing the computational complexity and efficiently solving the symmetric coefficient matrix.

[0027] Preferably, in step S4, the maximum eigenvalue of the symmetric coefficient matrix is ​​extracted as an eigenvector, and the eigenvector is expressed as V = [v1, v2, ..., v L ]∈R N×L , where v L =[v1,v2,…,v N ] T is the N coefficient matrix Z n The maximum eigenvalue set of the covariance matrix of , L represents the size of the sample set, and the sample set V is divided into a training set and test set

[0028] Preferably, in step S5, the specific steps of the training include:

[0029] S51. Randomly divide the feature vectors in the training set into 2 M classes, let the set of feature vectors belonging to class k be C k,k=1,2,…,2 M ; Let the cluster centroid of class k be c k , define c k is the mean of all feature vectors in this class, where c1 is fixed to the mean of the noise feature vector;

[0030] S52. Calculate the feature vector in the training set With each cluster centroid c k The Euclidean distance is Divide the feature vectors v in the training set into corresponding classes, calculate the mean of all feature vectors in each class after update, and update the corresponding cluster centroid c k , until all cluster centroids no longer change, the training is completed and the corresponding classifier is output.

[0031] According to the above technical means, the K-means clustering algorithm is used to classify the samples by judging the Euclidean distance of each centroid and updating it after each classification, and finally a classifier is obtained to classify new samples.

[0032] Preferably, in step S6, the expression of the classifier is:

[0033] Among them, c k and They represent the cluster centroid and the test set feature vector respectively, and β is a parameter; the feature vectors in the test set are input into the classifier in turn. In each classification, if the classification result of the classifier is greater than β, H1 is output, indicating that the feature vector is classified into the class where the primary user exists, the primary user signal exists in the frequency band, and the secondary user cannot access the frequency band. If the left-hand side of the classifier expression is less than β, that is, the expression is not established, H0 is output, indicating that the feature vector is classified into the class where the primary user does not exist, the primary user signal does not exist in the frequency band, and the secondary user can access the frequency band.

[0034] According to the above technical means, the feature vector in the test set is input into the classifier trained in the previous step to obtain the classification result of the test set, and the test set data is used to evaluate the classifier performance in the subsequent steps.

[0035] Preferably, in step S6, the classification results of the test set are compared with the actual results, and the test probability and the false alarm probability are calculated. The test probability indicates: when the primary user exists, the probability that the classifier determines that the primary user exists; the false alarm probability indicates: when the primary user does not exist, the probability that the classifier determines that the primary user exists; under the same false alarm probability, the higher the test probability, the better the classifier performance.

[0036] According to the above technical means, the detection probability and false alarm probability are calculated, so that when the classifier adjusts the parameters, it can provide an intuitive performance standard under the premise of ensuring the confidence level, thereby objectively evaluating the classifier performance.

[0037] According to another aspect of the present invention, a collaborative spectrum sensing system based on symmetric low-rank subspace clustering is provided, comprising:

[0038] A system building module, used to build a cognitive wireless system for collaborative spectrum sensing, the cognitive wireless system includes multiple primary users and multiple secondary users, the primary users communicate with the secondary users through access frequency bands, each secondary user is equipped with multiple sensing antennas for collecting signal vectors, the signal vectors are composed of noise signals generated during the communication process or any primary user signal superimposed on the noise signal;

[0039] A matrix synthesis module is used to fuse the signal vectors collected by any user through multiple sensing antennas into a signal matrix;

[0040] A matrix dimension reduction module, used for performing dimension reduction processing on the signal matrix to obtain a plurality of symmetric coefficient matrices with low rank and symmetric structure;

[0041] A sample partitioning module, used for extracting the covariance matrix of each of the symmetric coefficient matrices, then extracting the maximum eigenvalue of the covariance matrix, taking the maximum eigenvalue as an eigenvector, and jointly forming a sample set, and dividing the sample set into a training set and a test set;

[0042] A clustering training module, used to call the K-means algorithm to perform clustering training on the feature vectors in the training set, output the cluster centroid; and construct a classifier through the cluster centroid;

[0043] The performance evaluation module is used to classify all feature vectors of the test set using the classifier, compare the classification results with the actual results of the test set, and evaluate the system perception performance.

[0044] Compared with the prior art, the present invention proposes a spectrum sensing method that is more suitable for complex communication environments by constructing a wireless communication system with multiple main users, and performs dimensionality reduction processing on the signal matrix to filter out redundant information and reduce computational complexity. Therefore, the present invention can ensure perception accuracy and detection efficiency under low signal-to-noise ratio conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 A schematic flow chart showing the steps of a collaborative spectrum sensing method based on symmetric low-rank subspace clustering proposed in an embodiment of the present invention.

[0046] Figure 2A schematic diagram showing the overall process of a collaborative spectrum sensing method based on symmetric low-rank subspace clustering proposed in an embodiment of the present invention.

[0047] Figure 3 FIG. 4 is a schematic diagram of a cognitive wireless system according to an embodiment of the present invention.

[0048] Figure 4 A performance diagram showing different spectrum sensing methods proposed in the embodiments of the present invention.

[0049] Figure 5 A schematic diagram of a collaborative spectrum sensing system based on symmetric low-rank subspace clustering proposed in an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0050] The embodiments of the present invention provide a spectrum sensing method and device, which can effectively improve detection performance.

[0051] The following will combine the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention and are only used for illustrative purposes and cannot be understood as limitations on this patent. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention. The drawings are only used for illustrative purposes and cannot be understood as limitations on this patent; in order to better illustrate this embodiment, some parts of the drawings will be omitted, enlarged or reduced, and do not represent the actual size. The description of the directions of parts such as "up" and "down" is not a limitation on this patent; for those skilled in the art, it is understandable that some well-known content descriptions in the drawings may be omitted; the description of the positional relationship in the drawings is only for illustrative purposes and cannot be understood as a limitation on this patent.

[0052] The technical solution of the present invention is further described below in conjunction with the accompanying drawings and embodiments.

[0053] Example 1

[0054] This embodiment provides a collaborative spectrum sensing method based on symmetric low-rank subspace clustering, see Figure 1 and 2 ,include:

[0055] S1. Building a cognitive wireless system for collaborative spectrum sensing, refer to Figure 3 The cognitive wireless system includes multiple primary users and multiple secondary users. The primary users communicate with the secondary users through an access frequency band. Each secondary user is equipped with multiple sensing antennas for collecting signal vectors. The signal vector is composed of a noise signal generated during the communication process or any primary user signal superimposed on a noise signal.

[0056] In cognitive wireless systems, frequency bands are divided into authorized frequency bands and unlicensed frequency bands. Primary users can only use authorized frequency bands for communication, while secondary users can use authorized frequency bands and unlicensed frequency bands. The priority of primary users in accessing authorized frequency bands is higher than that of secondary users.

[0057] In step S1, assume that in a cognitive wireless system, the number of primary users is M, the number of secondary users is N, the number of sensing antennas of each secondary user is A, the total number of signal sampling points collected by the secondary user in the sampling time slot is d, and the signal collected by each sensing antenna is regarded as a signal vector. Assume n = 1, 2, 3, ..., N, and the sampling point t = 1, 2, 3, ..., d, then the sampled signal expression of the nth secondary user is

[0058] H1:x na (t) = h m,na (t)s na (t)+η na (t)

[0059] H0:x na (t) = η na (t)

[0060] Among them, H1 means that the primary user signal exists, and the frequency band is unavailable, and H0 means that the primary user signal does not exist, and the frequency band is available; na represents the main user signal, and the signal is an amplitude modulation signal. The amplitude modulation signal is a signal modulation method commonly used in wireless communication. Amplitude modulation makes the amplitude of the carrier signal change with the signal, but the frequency remains unchanged; η na is a product with a mean of 0 and a covariance of Gaussian white noise signal, h m,na (t)~CN(0,d -α ) is the channel gain from the mth primary user to the nth secondary user, m = 1, 2, 3…M, a represents the sensing antenna channel selected during signal transmission, the channel gain is the transmission characteristic of the channel itself, has nothing to do with input and output, and the channel gain depends on the distance d between the primary user and the secondary user -α , where α is the path loss exponent.

[0061] S2. Fusing the signal vectors collected by any user through multiple sensing antennas into a signal matrix;

[0062] In step S2, let the signal vector collected by the nth secondary user through the ath sensing antenna be represented as x na =[x na (1),x na (2),…,x na (d)] T, then the matrix form of the signal vector collected by n secondary users through all antennas is:

[0063] Among them, X n ∈R d×A .

[0064] S3. Perform dimensionality reduction processing on the signal matrix to obtain several symmetric coefficient matrices with low rank and symmetric structure.

[0065] In wireless communication systems, the spectrum signals received by secondary users usually contain redundant information and are difficult to solve with high dimensions. Directly extracted signal feature information cannot accurately represent the signal, so a symmetric low-rank representation is used to process the signal matrix. The low-rank representation mathematical model is used to solve the clustering problem of high-dimensional data. It can find the lowest rank representation of the data and capture the global structure of the data. The symmetric constraint enables points in the same subspace to be linearly represented by other points in the space, ensuring the consistency of the weights of each pair of data points and avoiding symmetric processing of the coefficient matrix, thereby improving the accuracy of clustering and reducing computational complexity.

[0066] In step S3, the signal matrix is ​​subjected to dimensionality reduction processing by using a symmetric low-rank representation mathematical model, and the specific steps include:

[0067] S31. The lowest rank and structurally symmetrical coefficient matrix of the signal matrix is ​​constructed through a symmetric low-rank representation model. The algorithm model can be expressed as:

[0068]

[0069] xT n =X n Z n +E n ,Z n =Z n T

[0070] Among them, Z n ∈R d×A is the signal matrix X n ∈R d×A Symmetric and low-rank coefficient representation; E n is the error term, l 1,2 The norm can make the error term infinitely close to 0, and λ>0 is used to balance the error term E n and low-rank structure Z n .

[0071] S32. In order to facilitate the separation of the above objective functions and simplify the subsequent solution steps, it is necessary to introduce an auxiliary variable J based on the objective function in S31. n :

[0072]

[0073]

[0074] S33. The augmented Lagrangian function of the mathematical model of the function in S32 can be expressed as:

[0075]

[0076] Among them, Y1 and Y2 are Lagrange multipliers; μ is the penalty parameter, and μ>0.

[0077] The optimization problem in S34.S33 can be further combined into:

[0078]

[0079] S35. The alternating direction multiplier iteration method is an algorithm for solving convex optimization problems with separable structures. The model in S34 is solved using the alternating direction multiplier iteration method. The specific process includes: setting the maximum number of iterations to i, and using the alternating direction multiplier method for calculation in each iteration. When the number of iterations reaches i, the calculation result is output, that is, the low-rank and structurally symmetric coefficient matrix of the output signal.

[0080] S4. extract the covariance matrix of each of the symmetric coefficient matrices, then extract the maximum eigenvalue of the covariance matrix, use the maximum eigenvalue as the eigenvector, together form a sample set, and divide the sample set into a training set and a test set.

[0081] In step S4, the maximum eigenvalue of the symmetric coefficient matrix obtained in S3 is extracted as an eigenvector, and the eigenvector is expressed as V = [v1, v2, ..., v L ]∈R N×L , where v L =[v1,v2,…,v N ] T is the N coefficient matrix Z n The maximum eigenvalue set of the covariance matrix of , L represents the size of the sample set, and the sample set V is divided into a training set and test set

[0082] S5. Perform clustering training on the feature vectors in the training set obtained in step S4 using the K-means algorithm, and output the cluster centroid; and construct a classifier using the cluster centroid.

[0083] In step S5, the specific steps of using the K-means algorithm to train the classifier include:

[0084] S51. Randomly divide the feature vectors in the training set into 2 M classes, let the set of feature vectors belonging to class k be C k ,k=1,2,…,2 M ; Let the cluster centroid of class k be c k , define c k is the mean of all feature vectors in this class, where c1 is fixed to the mean of the noise feature vector.

[0085] S52. Calculate the feature vector in the training set With each cluster centroid c k The Euclidean distance is The feature vector in the training set Divide them into corresponding classes, calculate the mean of all eigenvectors in each class after update, and update the corresponding cluster centroid c k , until all cluster centroids no longer change, the training is completed and the corresponding classifier is output.

[0086] The expression of the classifier is:

[0087] Among them, c k and They represent the cluster centroid and test set feature vector respectively, and β is a parameter.

[0088] S6. Use the classifier to classify all feature vectors of the test set, compare the classification results with the actual results of the test set, and evaluate the system perception performance.

[0089] The specific steps of step S6 include:

[0090] S61. Input the feature vectors in the test set into the classifier one by one. In each classification, if the classifier The classification result is greater than β, and H1 is output, which means that the feature vector is classified into the class where the primary user exists, the primary user signal exists in the frequency band, and the secondary user cannot access the frequency band. If the left-hand side of the classifier expression is less than β, that is, the expression does not hold, H0 is output, which means that the feature vector is classified into the class where the primary user does not exist, the primary user signal does not exist in the frequency band, and the secondary user can access the frequency band.

[0091] S62. Compare the classification result obtained in S61 with the actual result of the test set, and calculate the test probability and false alarm probability. The test probability is the ratio of the classifier judging the existence of the primary user when the primary user exists; the false alarm probability is the probability of the classifier judging the existence of the primary user when the primary user does not exist. Under the same false alarm probability, the higher the test probability, the better the classifier performance.

[0092] It can be concluded from this embodiment that under the same conditions, different spectrum sensing algorithms are trained and the spectrum sensing performance is compared. Figure 4 It can be seen that when the signal-to-noise ratio is -20db, the detection probability of the spectrum sensing algorithm proposed in the present invention is higher when the false alarm probability is the same, which means that the spectrum sensing algorithm proposed in the present invention has better performance than other algorithms and reduces the complexity of the algorithm while maintaining the perception accuracy.

[0093] Example 2

[0094] This embodiment provides a cooperative spectrum sensing system based on symmetric low-rank subspace clustering. Figure 5 ,include:

[0095] A system building module, used to build a cognitive wireless system for collaborative spectrum sensing, the cognitive wireless system includes multiple primary users and multiple secondary users, the primary users communicate with the secondary users through access frequency bands, each secondary user is equipped with multiple sensing antennas for collecting signal vectors, the signal vectors are composed of noise signals generated during the communication process or any primary user signal superimposed on the noise signal;

[0096] A matrix synthesis module is used to fuse the signal vectors collected by any user through multiple sensing antennas into a signal matrix;

[0097] A matrix dimension reduction module, used for performing dimension reduction processing on the signal matrix to obtain a plurality of symmetric coefficient matrices with low rank and symmetric structure;

[0098] A sample partitioning module, used for extracting the covariance matrix of each of the symmetric coefficient matrices, then extracting the maximum eigenvalue of the covariance matrix, taking the maximum eigenvalue as an eigenvector, and jointly forming a sample set, and dividing the sample set into a training set and a test set;

[0099] A clustering training module, used to call the K-means algorithm to perform clustering training on the feature vectors in the training set, output the cluster centroid; and construct a classifier through the cluster centroid;

[0100] The performance evaluation module is used to classify all feature vectors of the test set using the classifier, compare the classification results with the actual results of the test set, and evaluate the system perception performance.

[0101] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the embodiments of the present invention. For those skilled in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all the embodiments here. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the claims of the present invention.

Claims

1. A collaborative spectrum sensing method based on symmetric low-rank subspace clustering, characterized in that: The following steps are involved: S1. Construct a cognitive wireless system for collaborative spectrum sensing, the cognitive wireless system comprising a plurality of primary users and a plurality of secondary users, the primary users communicating with the secondary users via an access frequency band, each secondary user being equipped with a plurality of sensing antennas for collecting a signal vector, the signal vector being composed of a noise signal generated during the communication process or any primary user signal superimposed with a noise signal; S2. Fusing the signal vectors collected by any user through multiple sensing antennas into a signal matrix; S3. Performing dimensionality reduction processing on the signal matrix to obtain several low-rank and structurally symmetric coefficient matrices; wherein, in step S3, the signal matrix is ​​subjected to dimensionality reduction processing by a symmetric low-rank representation mathematical model, and the mathematical model for solving the lowest-rank and structurally symmetric symmetric coefficient matrix is: Among them, Z n ∈R d×A is the signal matrix X n ∈R d×A The symmetric and low-rank coefficient representation, E n is the error term, l 1,2 The norm makes the error term infinitely close to 0, and λ>0 is used to balance the error term E n and low-rank structure Z n , Y1 and Y2 are Lagrange multipliers; μ is the penalty parameter, and μ> 0 ;as well as, The symmetric low-rank representation mathematical model is solved using the alternating direction multiplier iteration method to obtain the symmetric coefficient matrix Z n ; S4. extracting the covariance matrix of each of the symmetric coefficient matrices, and then extracting the maximum eigenvalue of the covariance matrix, taking the maximum eigenvalue as the eigenvector, and jointly forming a sample set, and dividing the sample set into a training set and a test set; S5. Perform clustering training on the feature vectors in the training set using the K-means algorithm, and output the cluster centroid; construct a classifier using the cluster centroid; S6. Use the classifier to classify all feature vectors of the test set, compare the classification results with the actual results of the test set, and evaluate the system perception performance.

2. The collaborative spectrum sensing method based on symmetric low-rank subspace clustering according to claim 1 is characterized in that: In step S1, let the number of primary users be M, the number of secondary users be N, the number of sensing antennas of each secondary user be A, the total number of signal sampling points collected by the secondary user in the sampling time slot be d, and the signal collected by each sensing antenna is regarded as a signal vector. Let n = 1, 2, 3, ..., N, let sampling point t = 1, 2, 3, ..., d, then the sampled signal expression of the nth secondary user is H1:x na (t)=h m,na (t)s na (t)+η na (t) H0:x na (t)=η na (t) Among them, H1 means that the primary user signal exists, and the frequency band is unavailable; H0 means that the primary user signal does not exist, and the frequency band is available. na represents the main user signal, and the signal is an amplitude modulation signal, η na is a product with a mean of 0 and a covariance of Gaussian white noise signal, h m,na (t)~CN(0,d -α ) is the channel gain from the mth primary user to the nth secondary user, m = 1, 2, 3…M, a represents the sensing antenna channel selected during signal transmission, and the channel gain depends on the distance d between the primary user and the secondary user -α , where α is the path loss exponent.

3. The collaborative spectrum sensing method based on symmetric low-rank subspace clustering according to claim 2 is characterized in that: In step S2, let the signal vector collected by the nth secondary user through the ath sensing antenna be represented as x na =[x na (1),x na (2),…,x na (d)] T , then the matrix form of the signal vector collected by n secondary users through all antennas is: Among them, X n ∈R d×A .

4. The collaborative spectrum sensing method based on symmetric low-rank subspace clustering according to claim 1, characterized in that: In step S4, the maximum eigenvalue of the symmetric coefficient matrix is ​​extracted as an eigenvector, and the eigenvector is represented by V = [v1, v2, ..., v L ]∈R N×L , where v L =[v1,v2,…,v N ] T is the N coefficient matrix Z n The maximum eigenvalue set of the covariance matrix of , L represents the size of the sample set, and the sample set V is divided into a training set and test set 5. The collaborative spectrum sensing method based on symmetric low-rank subspace clustering according to claim 4 is characterized in that: In step S5, the specific steps of the training include: S51. Randomly divide the feature vectors in the training set into 2 M classes, let the set of feature vectors belonging to class k be C k ,k=1,2,…,2 M ; Let the cluster centroid of class k be c k , define c k is the mean of all feature vectors in this class, where c1 is fixed to the mean of the noise feature vector; S52. Calculate the feature vector in the training set With each cluster centroid c k The Euclidean distance is The feature vector in the training set Divide them into corresponding classes, calculate the mean of all eigenvectors in each class after update, and update the corresponding cluster centroid c k , until all cluster centroids no longer change, the training is completed and the corresponding classifier is output.

6. The collaborative spectrum sensing method based on symmetric low-rank subspace clustering according to claim 5, characterized in that: In step S6, the expression of the classifier is: Among them, c k and They represent the cluster centroid and the test set feature vector respectively, and β is a parameter; The feature vectors in the test set are input into the classifier in sequence. In each classification, if the classification result of the classifier is greater than β, H1 is output, indicating that the feature vector is classified into the class where the primary user exists, the primary user signal exists in the frequency band, and the secondary user cannot access the frequency band. If the classification result of the classifier is less than β, H0 is output, indicating that the feature vector is classified into the class where the primary user does not exist, the primary user signal does not exist in the frequency band, and the secondary user can access the frequency band.

7. The collaborative spectrum sensing method based on symmetric low-rank subspace clustering according to claim 6, characterized in that: In step S6, the classification result of the test set is compared with the actual result, and the test probability and false alarm probability are calculated. The test probability represents: when the primary user exists, the classifier determines the probability that the primary user exists; The false alarm probability indicates the probability that the classifier judges that the primary user exists when the primary user does not exist. Under the same false alarm probability, the higher the detection probability, the better the classifier performance.

8. A collaborative spectrum sensing system based on symmetric low-rank subspace clustering, characterized in that: include: A system construction module, used to construct a cognitive wireless system, wherein the cognitive wireless system includes multiple primary users and multiple secondary users, wherein the primary users communicate with the secondary users through an access frequency band, and each secondary user is equipped with multiple sensing antennas for collecting signal vectors, wherein the signal vectors are composed of a noise signal generated during the communication process or a signal of any primary user superimposed on a noise signal; A matrix synthesis module is used to fuse the signal vectors collected by any user through multiple sensing antennas into a signal matrix; The matrix dimension reduction module is used to perform dimension reduction processing on the signal matrix to obtain several symmetric coefficient matrices with low rank and symmetric structure; wherein, the signal matrix is ​​reduced in dimension by using a symmetric low rank representation mathematical model, and the mathematical model for solving the symmetric coefficient matrix with the lowest rank and symmetric structure is: Among them, Z n ∈R d×A is the signal matrix X n ∈R d×A The symmetric and low-rank coefficient representation, E n is the error term, l 1,2 The norm makes the error term infinitely close to 0, and λ>0 is used to balance the error term E n and low-rank structure Z n , Y1 and Y2 are Lagrange multipliers; μ is the penalty parameter, and μ>0; and, The symmetric low-rank representation mathematical model is solved using the alternating direction multiplier iteration method to obtain the symmetric coefficient matrix Z n ; A sample partitioning module, used for extracting the covariance matrix of each of the symmetric coefficient matrices, then extracting the maximum eigenvalue of the covariance matrix, taking the maximum eigenvalue as an eigenvector, and jointly forming a sample set, and dividing the sample set into a training set and a test set; A clustering training module, used to call the K-means algorithm to perform clustering training on the feature vectors in the training set, output the cluster centroid; and construct a classifier through the cluster centroid; The performance evaluation module is used to classify all feature vectors of the test set using the classifier, compare the classification results with the actual results of the test set, and evaluate the system perception performance.