A collaborative spectrum sensing method based on subspace clustering

Abstract

The present invention proposes a cooperative spectrum sensing method based on subspace clustering, which relates to the technical field of cognitive radio communication. Firstly, a cognitive wireless system for cooperative spectrum sensing is established, the cognitive wireless system includes a primary user and multiple secondary users, then a signal collected by any secondary user is fused to obtain a signal matrix, and then the signal matrix is ​​processed by low-rank subspace clustering to obtain an adjacency matrix of the association matrix, and then the maximum eigenvalue of the adjacency matrix is ​​used as a feature sample set, which is divided into a training set and a test set, and the training set is trained by a k-means algorithm to obtain an expression of a classifier for spectrum sensing, and finally the test set is used to detect the spectrum sensing performance of the classifier to obtain the detection result of the primary user state, and the spectrum sensing performance of the classifier is judged according to the detection result, so as to ensure the performance of spectrum sensing under a low signal-to-noise ratio and reduce the computational complexity.

Description

Technical Field

[0001] The present invention relates to the technical field of cognitive radio communication, and in particular to a cooperative spectrum sensing method based on subspace clustering. Background Art

[0002] Wireless communication technology is a communication method that uses the characteristics of electromagnetic wave signals propagating in free space to exchange information. Various wireless devices require certain spectrum resources and continuous energy supply to transmit data. With the rapid development of wireless communication technology, the demand for spectrum is increasing, resulting in tight spectrum resources, but the utilization rate of spectrum resources is very low. The introduction of cognitive radio has improved the utilization rate of spectrum resources to a certain extent. Spectrum sensing is a key link in cognitive radio, which is used to collect spectrum information and detect spectrum occupancy status.

[0003] Currently, spectrum sensing technology can be divided into single-user spectrum sensing technology and multiple-user collaborative spectrum sensing technology according to the number of sub-users participating in spectrum sensing. Single-user spectrum sensing technology is easily interfered by factors such as shadow effect, channel fading and terminal anonymity; multiple-user collaborative spectrum sensing technology, with the cooperation of multiple sub-users, shares local sensing data, integrates effective data, and finally obtains the result of global decision-making. Compared with single-user spectrum sensing technology, multiple-user collaborative spectrum sensing technology can overcome problems such as path attenuation, shadow and hidden terminals. However, with the increase of sub-users, the channel overhead and computational complexity also increase. Therefore, it is of great significance to study the balance of channel overhead while maximizing the performance of multiple-user collaborative sensing algorithms.

[0004] In recent years, spectrum sensing problems can be converted into binary classification problems. Clustering algorithms based on machine learning are a hot topic in research on solving binary classification problems. Common clustering algorithms based on machine learning include k-means clustering or Gaussian mixture models. k-means clustering or Gaussian mixture models are used as unsupervised learning frameworks, with dominant features, maximum and minimum eigenvalues ​​or geometric features as feature vectors for classifier training.

[0005] The prior art discloses a collaborative spectrum sensing method, which first extracts signal features based on information geometry, then uses information geometry theory to map the signal vector onto a manifold for calculation, and obtains geometric features as training samples for a clustering algorithm. The training samples are further used to train a differential evolution clustering algorithm to obtain a classifier for spectrum sensing. However, this scheme only processes sample signals, and does not consider the correlation information between sample signals, that is, it ignores the relationship between sample signals. The performance of spectrum sensing cannot be guaranteed under low signal-to-noise ratio conditions, and the signal vector is mapped onto a manifold for calculation, which has high computational complexity. Summary of the invention

[0006] In order to solve the problems of poor spectrum sensing performance and high computational complexity of current collaborative spectrum sensing methods, the present invention proposes a collaborative spectrum sensing method based on subspace clustering, which takes into account the relationship between sample signals, ensures the performance of spectrum sensing under low signal-to-noise ratio, and reduces the computational complexity.

[0007] In order to achieve the above technical effects, the technical solution of the present invention is as follows:

[0008] A collaborative spectrum sensing method based on subspace clustering, characterized by comprising the following steps:

[0009] S1. Establishing a cognitive wireless system for collaborative spectrum sensing, the cognitive wireless system comprising a primary user and a plurality of secondary users;

[0010] S2. Fusing the signals collected by any secondary user to generate a signal matrix;

[0011] S3. construct an association matrix of the signal matrix, process the association matrix using low-rank subspace clustering, and obtain an adjacency matrix of the association matrix;

[0012] S4. Obtain the maximum eigenvalue of the adjacency matrix of S3 to form a feature sample set, and divide the feature sample set into a training set and a test set;

[0013] S5. Use the k-means algorithm to train the training set to obtain cluster centers, and then output the cluster centers to obtain the expression of the classifier for spectrum sensing;

[0014] S6. Use the test set to detect the spectrum sensing performance of the classifier, obtain the detection result of the primary user state, and judge the spectrum sensing performance of the classifier according to the detection result.

[0015] In the technical scheme, a cognitive wireless system for collaborative spectrum sensing is first established, the cognitive wireless system includes a primary user and multiple secondary users, and then the signal collected by any secondary user is fused to obtain a signal matrix, and then the signal matrix is ​​processed by low-rank subspace clustering to obtain an adjacency matrix of the association matrix, so that the association matrix is ​​balanced and the performance of spectrum sensing can be guaranteed under low signal-to-noise ratio conditions. Furthermore, the maximum eigenvalue of the adjacency matrix is ​​used as a feature sample set, which is divided into a training set and a test set, and the training set is trained using a k-means algorithm to obtain an expression of a classifier for spectrum sensing. Finally, the test set is used to detect the spectrum sensing performance of the classifier. The present invention takes into account the relationship between sample signals, guarantees the performance of spectrum sensing under low signal-to-noise ratio, and reduces computational complexity.

[0016] Preferably, in step S1, the number of secondary users is m, the number of collection points of each secondary user is n, and a number of antennas for collecting signals are provided, and the signals are correlated.

[0017] Preferably, within the time window for collecting signals, the signal collected by each antenna of the secondary user is a high-dimensional signal.

[0018] Preferably, in step S2, assuming i=1, 2, 3, ..., m, the composition expression of the signal collected by the i-th secondary user using the a-th antenna is:

[0019]

[0020] Where s(t) represents the transmission signal of the primary user; w ia (t) has a mean of 0 and a covariance of σ 2 Gaussian white noise; H0 means that the transmission signal of the primary user does not exist and the channel is available; H1 means that the transmission signal of the primary user exists and the channel is unavailable;

[0021] Assume that the vector expression of the signal collected by the i-th user using the a-th antenna is:

[0022] x ia (t) = [x ia (1),x ia (2)...x ia (n)] T (2)

[0023] Fusion of the S21 signal vector into the signal matrix X i The expression is:

[0024]

[0025] Preferably, in step S3, the correlation matrix of the signal matrix is ​​constructed as a minimum rank correlation matrix, and the minimum rank correlation matrix is ​​used as the objective function. The objective function expression is:

[0026]

[0027] The constraints are:

[0028] X i =X i C i (5)

[0029] Among them, C i Represents the signal matrix X i The specific process of processing the association matrix using low-rank subspace clustering is:

[0030] S31. Using the calculated nuclear norm to replace the rank of the matrix in the objective function;

[0031] S32. The objective function is transformed by using the alternating direction multiplier iteration method, the number of iterations is set to n, and it is determined whether the number of iterations in the iteration process is greater than or equal to n. If so, the lowest rank correlation matrix of the signal matrix is ​​output, and step S33 is executed; otherwise, the iteration is continued until the number of iterations meets n;

[0032] S33. Based on the lowest rank association matrix of S32, obtain the adjacency matrix of the lowest rank association matrix.

[0033] Preferably, in step S4, the training set is represented as Training set The expression is:

[0034]

[0035] Among them, T l represents the eigenvector, the eigenvector T l The expression is:

[0036] T l =[t1,t2,...t m ] T (7)

[0037] Among them, l represents the number of feature vectors of training samples.

[0038] Preferably, in step S5, the specific process of training the training set is:

[0039] S51. The training set The eigenvector T in l Divided into K categories, let Θ k , k=1,2,...K represents the feature vector T belonging to class k l Set, class Θ k Assume the corresponding center of mass ψ k ;

[0040] S52. Define the expression of the center of mass ψ1 of Θ1 as:

[0041] ψ1=E[T l |H0] (8)

[0042] Define the centroid of other clusters k=2,3,...K as ψ k All training feature vectors T l The mean, centroid ψ k The specific expression is:

[0043]

[0044] S53. Calculate the eigenvector T l The distances to ψ1 and ψ2 are calculated as follows:

[0045]

[0046] S54. According to the minimum distance to ψ1 and ψ2, the feature vector T l Divide into the corresponding class Θ k,k＝1,2 ;

[0047] S55. Calculate all eigenvectors T in Θ2 l , and update ψ2 through formula S52 until ψ2 no longer changes, and output the classifier for spectrum sensing.

[0048] Preferably, the expression of the classifier is:

[0049]

[0050] Where ξ is a metric parameter; when equation (11) is satisfied, it means that the primary user does not exist, otherwise it means that the primary user exists.

[0051] Preferably, in step S6, the detection result of the primary user state includes a detection probability of correctly detecting the primary user state and a false alarm probability of incorrectly detecting the primary user state.

[0052] Preferably, the spectrum sensing performance of the classifier is judged according to the detection probability under the same false alarm probability. If the detection probability is compared under the same false alarm probability, the spectrum sensing performance of the classifier is good if the detection probability is high; otherwise, the spectrum sensing performance of the classifier is poor.

[0053] Compared with the prior art, the technical solution of the present invention has the following beneficial effects:

[0054] The present invention proposes a cooperative spectrum sensing method based on subspace clustering. In the technical scheme, a cognitive wireless system for cooperative spectrum sensing is first established, the cognitive wireless system includes a primary user and multiple secondary users, and then the signal collected by any secondary user is fused to obtain a signal matrix, and then the signal matrix is ​​processed by low-rank subspace clustering to obtain an adjacency matrix of the association matrix, so that the association matrix is ​​balanced and the performance of spectrum sensing can be guaranteed under low signal-to-noise ratio conditions. Further, the maximum eigenvalue of the adjacency matrix is ​​used as a feature sample set, which is divided into a training set and a test set, and the training set is trained using a k-means algorithm to obtain an expression of a classifier for spectrum sensing. Finally, the test set is used to detect the spectrum sensing performance of the classifier. The present invention takes into account the relationship between sample signals, guarantees the performance of spectrum sensing under low signal-to-noise ratio, and reduces the computational complexity. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 A schematic diagram showing a flow chart of a cooperative spectrum sensing method based on subspace clustering proposed in an embodiment of the present invention;

[0056] Figure 2 A schematic diagram of spectrum sensing based on subspace clustering proposed in an embodiment of the present invention is shown;

[0057] Figure 3 A comparison diagram of spectrum sensing performance under different spectrum sensing algorithms proposed in an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0058] The drawings are for illustrative purposes only and should not be construed as limiting the present patent;

[0059] In order to better illustrate the present embodiment, some parts of the drawings may be omitted, enlarged or reduced, and do not represent actual sizes. The description of the directions of parts such as "upper" and "lower" does not limit the present patent;

[0060] It is understandable to those skilled in the art that some well-known contents may be omitted in the drawings;

[0061] The positional relationships described in the drawings are only for illustrative purposes and should not be construed as limiting the present patent.

[0062] The technical solution of the present invention is further described below in conjunction with the accompanying drawings and embodiments.

[0063] Example 1

[0064] See also Figure 1 and Figure 2 , a cooperative spectrum sensing method based on subspace clustering, comprising the following steps:

[0065] S1. Establishing a cognitive wireless system for collaborative spectrum sensing, the cognitive wireless system comprising a primary user and a plurality of secondary users;

[0066] In step S1, the number of secondary users is set to m=3, the collection points of each secondary user are 1000, and 4 antennas are provided for collecting signals, and the signals are correlated; within the time window for collecting signals, the signal collected by each antenna of the secondary user is a high-dimensional signal.

[0067] S2. Fusing the signals collected by any secondary user to generate a signal matrix;

[0068] S3. construct an association matrix of the signal matrix, process the association matrix using low-rank subspace clustering, and obtain an adjacency matrix of the association matrix;

[0069] S4. Obtain the maximum eigenvalue of the adjacency matrix of S3 to form a feature sample set, and divide the feature sample set into a training set and a test set;

[0070] S5. Use the k-means algorithm to train the training set to obtain cluster centers, and then output the cluster centers to obtain the expression of the classifier for spectrum sensing;

[0071] S6. Use the test set to detect the spectrum sensing performance of the classifier, obtain the detection result of the primary user state, and judge the spectrum sensing performance of the classifier according to the detection result.

[0072] In this embodiment, in this embodiment, a cognitive wireless system for collaborative spectrum sensing is first established, the cognitive wireless system includes a primary user and multiple secondary users, and then the signal collected by any secondary user is fused to obtain a signal matrix, and then the signal matrix is ​​processed by low-rank subspace clustering to obtain an adjacency matrix of the association matrix, so that the association matrix is ​​balanced and the performance of spectrum sensing can be guaranteed under low signal-to-noise ratio conditions. Further, the maximum eigenvalue of the adjacency matrix is ​​used as a feature sample set, which is divided into a training set and a test set, and the training set is trained using a k-means algorithm to obtain an expression of a classifier for spectrum sensing. Finally, the test set is used to detect the spectrum sensing performance of the classifier. The present invention takes into account the relationship between sample signals, ensures the performance of spectrum sensing under low signal-to-noise ratio, and reduces computational complexity.

[0073] Example 2

[0074] See also Figure 1 In step S2, let i = 1, 2, 3 ... m, the composition expression of the signal collected by the i-th secondary user using the a-th antenna is:

[0075]

[0076] Where s(t) represents the transmission signal of the primary user; w ia (t) has a mean of 0 and a covariance of σ 2 Gaussian white noise; H0 means the primary user's transmission signal does not exist and the channel is available; H1 means the primary user's transmission signal exists and the channel is unavailable; Assume that the primary user's transmission signal is an AM modulated signal, also known as an amplitude modulation modulated signal, which is a modulation method used in electronic communications and is most commonly used for radio carrier transmission of information. In amplitude modulation, the amplitude or signal strength of the carrier changes in proportion to the wave transmitted;

[0077] Assume that the vector expression of the signal collected by the i-th user using the a-th antenna is:

[0078] x ia(t) = [x ia (1),x ia (2)...x ia (n)] T (2)

[0079] Fusion of the S21 signal vector into the signal matrix X i The expression is:

[0080]

[0081] In step S3, the correlation matrix of the signal matrix is ​​constructed as the lowest rank correlation matrix, and the lowest rank correlation matrix is ​​used as the objective function. The objective function expression is:

[0082]

[0083] The constraints are:

[0084] X i =X i C i (5)

[0085] Among them, C i Represents the signal matrix X i The specific process of processing the association matrix using low-rank subspace clustering is:

[0086] S31. Using the calculated nuclear norm to replace the rank of the matrix in the objective function;

[0087] In step S31, the nuclear norm is the sum of the singular values ​​of the matrix, which is used to constrain the low rank of the matrix. For sparse data, the matrix is ​​low rank and contains a lot of redundant information, which can be used to restore data and extract features. Since it is difficult to calculate the rank of the association matrix, using the nuclear norm instead of the matrix rank can also obtain a good clustering effect. Therefore, formula (4) can be converted into:

[0088]

[0089] S32. The objective function is transformed by using the alternating direction multiplier iteration method, and the number of iterations is set to 500. It is determined whether the number of iterations in the iterative process is greater than or equal to 500. If so, the lowest rank correlation matrix of the signal matrix is ​​output, and step S33 is executed; otherwise, the iteration is continued until the number of iterations meets 500;

[0090] In step S32, the alternating direction multiplier iteration method is an efficient algorithm for solving convex optimization problems with separable structures, and can be used to solve the correlation matrix C i , according to the algorithm of alternating direction multiplier iteration method, formula (5.1) is transformed into:

[0091]

[0092] The constraints are:

[0093] C i =Z i (5.3)

[0094] Iteratively solve the augmented Lagrangian function of equation (5.2) i , C i and Y i .

[0095] S33. Based on the lowest rank association matrix of S32, obtain the adjacency matrix of the lowest rank association matrix.

[0096] In step S33, the expression of the adjacency matrix is:

[0097] U i =(|C i |+|C i | T ) / 2 (5.4)

[0098] The k-means algorithm is applied on the adjacency matrix. If the transmission signal of the primary user does not exist, the channel is judged to be available; otherwise, the transmission signal of the primary user exists and the channel is unavailable.

[0099] In step S4, let the training set be represented as Training set The expression is:

[0100]

[0101] Among them, T l represents the eigenvector, the eigenvector T l The expression is:

[0102] T l =[t1,t2,...t m ] T (7)

[0103] Among them, l represents the number of feature vectors of training samples.

[0104] Example 3

[0105] See also Figure 1 In step S5, the specific process of training the training set is:

[0106] S51. The training set The eigenvector T in l Divided into K categories, let Θ k, k=1,2,...K represents the feature vector T belonging to class k l Set, class Θ k Assume the corresponding center of mass ψ k ;

[0107] S52. Define the expression of the center of mass ψ1 of Θ1 as:

[0108] ψ1=E[T l |H0] (8)

[0109] Define the centroid of other clusters k=2,3,...K as ψ k All training feature vectors T l The mean, centroid ψ k The specific expression is:

[0110]

[0111] S53. Calculate the eigenvector T l The distances to ψ1 and ψ2 are calculated as follows:

[0112]

[0113] S54. According to the minimum distance to ψ1 and ψ2, the feature vector T l Divide into the corresponding class Θ k,k＝1,2 ;

[0114] S55. Calculate all eigenvectors T in Θ2 l , and update ψ2 through formula S52 until ψ2 no longer changes, and output the classifier for spectrum sensing.

[0115] The expression of the classifier is:

[0116]

[0117] Where ξ is a metric parameter; when equation (11) is satisfied, it means that the primary user does not exist, otherwise it means that the primary user exists.

[0118] In step S6, the detection result of the primary user state includes the detection probability of correctly detecting the primary user state and the false alarm probability of incorrectly detecting the primary user state; the spectrum sensing performance of the classifier is judged according to the detection probability under the same false alarm probability. If the detection probabilities are compared under the same false alarm probability, the spectrum sensing performance of the classifier is good if the detection probability is high; otherwise, the spectrum sensing performance of the classifier is poor.

[0119] This embodiment results in Figure 3 , which is a comparison chart of spectrum sensing performance under different spectrum sensing algorithms. Figure 3It can be seen from the figure that when the signal-to-noise ratio SNR=-18dB and the same false alarm probability, the spectrum sensing method proposed in this embodiment is compared with other spectrum sensing algorithms. It can be clearly seen that the spectrum sensing performance of this embodiment is greatly improved, which shows that this embodiment takes into account the relationship between sample signals, has good spectrum sensing performance at low signal-to-noise ratio, and reduces calculation complexity.

[0120] Obviously, the above embodiments of the present invention are only 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 subspace clustering, characterized in that: The following steps are involved: S1. Establishing a cognitive wireless system for collaborative spectrum sensing, the cognitive wireless system comprising a primary user and a plurality of secondary users; S2. Fusing the signals collected by any secondary user to generate a signal matrix; S3. construct an association matrix of the signal matrix, process the association matrix using low-rank subspace clustering, and obtain an adjacency matrix of the association matrix; S4. Obtain the maximum eigenvalue of the adjacency matrix of S3 to form a feature sample set, and divide the feature sample set into a training set and a test set; S5. Use the k-means algorithm to train the training set to obtain cluster centers, and then output the cluster centers to obtain the expression of the classifier for spectrum sensing; S6. Use the test set to detect the spectrum sensing performance of the classifier, obtain the detection result of the primary user state, and judge the spectrum sensing performance of the classifier based on the detection result; Assume that the number of secondary users is m, the collection point of each secondary user is n, and there are several antennas for collecting signals, and the signals are correlated; In the time window of signal collection, the signal collected by each antenna of the secondary user is a high-dimensional signal; In step S2, let i = 1, 2, 3, ..., m, and the composition expression of the signal collected by the i-th secondary user using the a-th antenna is: Where s(t) represents the transmission signal of the primary user; w ia (t) has a mean of 0 and a covariance of σ 2 Gaussian white noise; H0 means that the transmission signal of the primary user does not exist and the channel is available; H1 means that the transmission signal of the primary user exists and the channel is unavailable; Assume that the vector expression of the signal collected by the i-th user using the a-th antenna is: x ia (t)=[x ia (1),x ia (2)...x ia (n)] T (2) Fusion of the S21 signal vector into the signal matrix X i The expression is: In step S3, the correlation matrix of the signal matrix is ​​constructed as the lowest rank correlation matrix, and the lowest rank correlation matrix is ​​used as the objective function. The objective function expression is: The constraints are: X i =X i C i (5) Among them, C i Represents the signal matrix X i The specific process of processing the association matrix using low-rank subspace clustering is: S31. Using the calculated nuclear norm to replace the rank of the matrix in the objective function; S32. The objective function is transformed by using the alternating direction multiplier iteration method, the number of iterations is set to n, and it is determined whether the number of iterations in the iteration process is greater than or equal to n. If so, the lowest rank correlation matrix of the signal matrix is ​​output, and step S33 is executed; otherwise, the iteration is continued until the number of iterations meets n; S33. Based on the lowest rank association matrix of S32, obtain an adjacency matrix of the lowest rank association matrix; In step S4, let the training set be represented as Training set The expression is: Among them, T l represents the eigenvector, the eigenvector T l The expression is: T l =[t1,t2,...t m ] T (7) Among them, l represents the number of training sample feature vectors; In step S5, the specific process of training the training set is: S51. The training set The eigenvector T in l Divided into K categories, let Θ k , k=1,2,...K represents the feature vector T belonging to class k l Set, class Θ k Assume the corresponding center of mass ψ k ; S52. Define the expression of the center of mass ψ1 of Θ1 as: ψ1=E[T l |H0] (8) Define the centroid of other clusters k=2,3,...K as ψ k All training feature vectors T l The mean, centroid ψ k The specific expression is: S53. Calculate the eigenvector T l The distances to ψ1 and ψ2 are calculated as follows: S54. According to the minimum distance to ψ1 and ψ2, the feature vector T l Divide into the corresponding class Θ k,k＝1,2 ; S55. Calculate all eigenvectors T in Θ2 l , and update ψ2 through formula S52 until ψ2 no longer changes, and output the expression of the classifier for spectrum sensing.

2. The collaborative spectrum sensing method based on subspace clustering according to claim 1, characterized in that: The expression of the classifier is: Where ξ is a metric parameter; when equation (11) is satisfied, it means that the primary user does not exist, otherwise it means that the primary user exists.

3. The collaborative spectrum sensing method based on subspace clustering according to any one of claims 1-2, characterized in that: In step S6, the detection result of the primary user state includes a detection probability of correctly detecting the primary user state and a false alarm probability of incorrectly detecting the primary user state.

4. The collaborative spectrum sensing method based on subspace clustering according to claim 3 is characterized in that: The spectrum sensing performance of the classifier is judged according to the detection probability under the same false alarm probability. If the detection probability is compared under the same false alarm probability, the higher the detection probability, the better the spectrum sensing performance of the classifier; otherwise, the spectrum sensing performance of the classifier is poor.

Images