CN111934797B - Collaborative spectrum sensing method based on covariance eigenvalue and mean shift clustering - Google Patents

Abstract

The invention discloses a collaborative spectrum sensing method based on covariance eigenvalues and mean shift clustering. And finally, obtaining a classifier by using the training mean shift clustering. The invention not only can well avoid the derivation of the traditional threshold value, but also avoids the condition that the two-dimensional signal characteristic vector can be constructed only by preprocessing the covariance matrix. Meanwhile, the classifier is obtained through a training mean shift clustering algorithm, so that the setting of the number of classes can be avoided, and the classes hidden by the training set can be better shown.

Description

Collaborative spectrum sensing method based on covariance eigenvalue and mean shift clustering

Technical Field

The invention relates to the field of cognitive radio, in particular to a collaborative spectrum sensing method based on covariance eigenvalues and mean shift clustering.

Background

In recent years, the number of various wireless devices and intelligent mobile terminals is rapidly increasing, people have increasingly increased demand for wireless spectrum, and the scale of wireless communication networks is continuously expanding, which undoubtedly makes spectrum resources increasingly tense. Cognitive radio technology aims to alleviate the current spectrum shortage problem. The main idea of cognitive radio technology is to enable radio communication devices to discover free spectrum and to make reasonable use of spectrum resources. The spectrum sensing technology is one of the important technologies of cognitive radio, and is also the basis of other applications such as spectrum sharing and spectrum management. However, in an actual radio environment, the spectrum sensing technology is affected by shadows, attenuation and the like, so that the signal-to-noise ratio of sensed signals is small, and the performance of spectrum sensing is reduced.

The traditional spectrum sensing technology comprises energy detection, matched filter detection and cyclic characteristic detection spectrum sensing methods. In recent years, random matrix theory has been proposed and gradually applied to spectrum sensing methods. The literature provides a spectrum sensing algorithm based on covariance matrix decomposition, which decomposes and processes a signal through a covariance matrix of the signal, and then performs decision through derivation of a threshold. The literature provides a spectrum sensing based on a random matrix theory, the algorithm still adopts a threshold judgment method, and the problems of inaccurate threshold derivation and complex calculation are definitely existed. The spectrum sensing technology is used for detecting whether a master user exists, so that spectrum sensing can be regarded as a two-classification problem, machine learning can well process the two-classification problem, and therefore the spectrum sensing method based on machine learning gradually becomes a hotspot of research of people. There is a proposed spectrum sensing algorithm based on K-means clustering, which uses the energy of signals as signal features, and then classifies and tests the signals through the K-means clustering algorithm. The method uses a random matrix principle, takes characteristic values of a signal matrix as signal characteristics, and classifies and tests the signals through a clustering algorithm.

Disclosure of Invention

The invention provides a collaborative spectrum sensing method based on covariance eigenvalue and mean shift clustering, which avoids the problem that two-dimensional signal eigenvector can be constructed only by preprocessing a covariance matrix.

In order to solve the technical problems, the technical scheme of the invention is as follows:

a collaborative spectrum sensing method based on covariance eigenvalue and mean shift clustering comprises the following steps:

s1: acquiring a plurality of received signal matrixes X under different time periods, wherein the elements of the received signal matrixes X are received signals of different secondary users;

s2: calculating an eigenvalue of a covariance matrix of the received signal matrix X;

s3: forming a two-dimensional signal feature vector by using the maximum feature value and the minimum feature value obtained by calculation of S2;

s4: two-dimensional signal feature vectors obtained by calculation at different time periods form a training set, and the training set is used as the input of a training mean shift clustering algorithm;

s5: after training is finished, obtaining a classifier for judging whether an authorized channel is available;

s6: judging whether a master user exists by using the classifier obtained in the S5, if the master user signal exists, indicating that the frequency band is occupied, and not counting the frequency band; if the master user signal does not exist, the frequency band is idle, and the frequency band can be accessed.

The method provides a collaborative spectrum sensing method (A collaborative spectral sensing method based on covariance eigenvalue and mean shift clustering, CEMSC). First, a plurality of secondary users transmit perceived data to a fusion center. Then, the fusion center integrates and processes the received data, calculates a covariance matrix, and obtains a maximum eigenvalue and a minimum eigenvalue of the covariance matrix. Secondly, a two-dimensional signal feature vector is formed by using the maximum feature value and the minimum feature value. And finally, training mean shift clustering through a large number of signal feature vectors to obtain a relevant classifier, and judging whether the master user signal exists or not by using the classifier.

Preferably, in step S1, each secondary user has N sampling points.

Preferably, in step S1, the received signal of the ith secondary user is represented as:

in the formula, H₀Indicates a band idle state, H₁Indicating the occupied state of the frequency band, x_i(n) denotes a received signal of an nth sampling point of an ith sub-user, 0<n<N,w_i(n) represents a mean value of 0 and a variance of σ²White gaussian noise signal, s_iAnd (n) represents a primary user signal.

Preferably, in step S1, the received signal matrix X is specifically:

in the formula, x_iRepresenting the ith secondary userThe sample vector of (2).

Preferably, in step S2, the covariance matrix of the received signal matrix X is calculated in the following manner:

R＝E[XX^T]

wherein R represents a covariance matrix of a received signal matrix X, X^TDenotes the transposition of X, E [, ]]Indicating the desire.

Preferably, in step S3, the maximum eigenvalue and the minimum eigenvalue calculated in step S2 form a two-dimensional signal eigenvector, specifically:

the eigenvalues of the covariance matrix R are arranged from large to small:

λ₁≥λ₂≥…≥λ_i≥…≥λ_M

the above formula is divided into two cases:

using maximum eigenvalues lambda₁And minimum eigenvalue λ_MForming a two-dimensional signal feature vector:

T＝[λ₁,λ_M]

in the formula, T is a two-dimensional signal feature vector.

Preferably, the training of the mean shift clustering algorithm in step S4 specifically includes:

(a) each data point in the training set represents a two-dimensional signal characteristic vector obtained in a time period, and a data point is randomly selected in the training set to serve as a central point;

(b) finding out all data points of the circular area with the radius h away from the central point, and recording the data points as a set S_hThese points are considered to belong to cluster k;

(c) the calculation starts from the center point to the set S_hAdding the vectors of each element to obtain an offset vector;

(d) the center point moves along the direction of the offset vector, and the moving distance is the mode of the offset vector;

(e) repeating the steps (b), (c) and (d),until the size of the offset vector meets the set threshold requirement, the central point is psi_k；

(f) Repeating steps (a), (b), (c), (d), (e) until all data points are categorized;

(g) and according to each class, for the access frequency of each data point, taking the class with the highest access frequency as the class to which the current point set belongs.

In the K-Means algorithm, the final clustering effect is influenced by the initial clustering center, the K-Means + + algorithm provides a basis for selecting a better initial clustering center, but the clustering category number K still needs to be established in the algorithm in advance, and for a data set with the category number unknown in advance, the K-Means and the K-Means + + are difficult to accurately solve, so that some improved algorithms are provided for processing the condition that the clustering number K is unknown. The Mean Shift algorithm, also known as the Mean Shift algorithm, is a clustering algorithm based on clustering centers like the K-Means algorithm, except that the Mean Shift algorithm does not need to make the number K of categories in advance. Mean-shift clustering is a sliding-window based algorithm that attempts to find dense regions of data points. This is a centroid based algorithm, which means that its goal is to locate the center point of each group/class by updating the candidate points for the center point to the mean of the points within the sliding window. These candidate windows are then filtered in a post-processing stage to eliminate approximate duplicates, forming a final set of center points and their corresponding groups.

Preferably, in step (c), the offset vector is calculated by:

in the formula, T_dIs S_hData points within the circular region, g being S_hNumber of data points in the circular region, T₀Is the center point.

Preferably, in step (d), the center point moves along the direction of the offset vector, specifically:

in the formula, M_h ^tFor the offset vector found in the t state,

is the central point in the t state,

is the central point in the t +1 state.

Preferably, in step S5, the classifier obtained after the training is represented as:

wherein the parameter gamma is used to control the false alarm probability P_f，Ψ₁Indicates after training at H₁Class center of state, Ψ₂Indicates after training at H₀Class center of state, if data point T_dIf the above formula is satisfied, the main user signal exists, the frequency band is occupied, and the frequency band cannot be counted; if the data point T_dIf the formula does not satisfy the above formula, the master user signal does not exist, the frequency band is idle, and the frequency band can be accessed.

Compared with the prior art, the technical scheme of the invention has the beneficial effects that:

the invention well avoids the derivation of the traditional threshold value and also avoids the problem that the two-dimensional signal characteristic vector can be constructed only by preprocessing the covariance matrix. Meanwhile, the classifier is obtained through the training mean shift clustering algorithm, so that the setting of the number of classes can be avoided, and the classes hidden by the training set can be better shown.

Drawings

FIG. 1 is a schematic flow chart of the method of the present invention.

Fig. 2 is a schematic diagram of a cognitive radio network model.

Fig. 3 is a ROC graph of a simulation experiment performed under the conditions that the number M of secondary users is 8, the number N of sampling points is 5000, and the SNR is-18 dB.

FIG. 4 is a ROC graph of a simulation experiment performed under the conditions that the number M of secondary users is 14, the number N of sampling points is 5000, and the SNR (signal to noise ratio) is-20 dB

Detailed Description

The drawings are for illustrative purposes only and are not to be construed as limiting the patent;

for the purpose of better illustrating the embodiments, certain features of the drawings may be omitted, enlarged or reduced, and do not represent the size of an actual product;

it will be understood by those skilled in the art that certain well-known structures in the drawings and descriptions thereof may be omitted.

The technical solution of the present invention is further described below with reference to the accompanying drawings and examples.

Example 1

The embodiment provides a collaborative spectrum sensing method based on covariance eigenvalue and mean shift clustering, as shown in fig. 1, including the following steps:

s1: acquiring a plurality of received signal matrixes X under different time periods, wherein the elements of the received signal matrixes X are received signals of different secondary users;

s2: calculating an eigenvalue of a covariance matrix of a received signal matrix X;

s3: forming a two-dimensional signal feature vector by using the maximum feature value and the minimum feature value obtained by calculation of S2;

s4: two-dimensional signal feature vectors obtained by calculation at different time periods form a training set, and the training set is used as input of a training mean shift clustering algorithm;

s5: after training is finished, obtaining a classifier for judging whether an authorized channel is available;

s6: judging whether a master user exists or not by using the classifier obtained in the S5, if a master user signal exists, indicating that the frequency band is occupied, and not counting the frequency band; if the master user signal does not exist, the frequency band is idle, and the frequency band can be accessed.

The main idea of the cooperative spectrum sensing algorithm based on covariance matrix decomposition is to detect by utilizing the difference of correlation between a main user (PU) signal and a Gaussian white signal. After the PU signal is subjected to shadowing effect and multipath attenuation, the perceived signal still has correlation, and whether the PU exists can be judged through the correlation. As shown in fig. 2, assuming that M (i ═ 1,2, …, M) Secondary Users (SUs) and one PU are shared in the cognitive radio network, the ith SU is affected by shadowing effect and multipath fading, and therefore, whether the PU exists cannot be accurately detected.

In step S1, N sampling points are provided for each secondary user.

In step S1, the received signal of the ith secondary user is represented as:

in the formula, H₀Indicating a band idle state, H₁Indicating the occupied state of the frequency band, x_i(n) denotes a received signal of an nth sampling point of an ith sub-user, 0<n<N,w_i(n) represents a mean value of 0 and a variance of σ²White gaussian noise signal, s_iAnd (n) represents a primary user signal.

In step S1, the received signal matrix X is specifically:

in the formula, x_iRepresenting the sample vector of the ith secondary user.

In step S2, the covariance matrix of the received signal matrix X is calculated in the following manner:

R＝E[XX^T]

wherein R represents a covariance matrix of a received signal matrix X, X^TDenotes the transposition of X, E [, ]]Indicating the desire.

In step S3, the maximum eigenvalue and the minimum eigenvalue calculated in step S2 form a two-dimensional signal eigenvector, which specifically includes:

the eigenvalues of the covariance matrix R are arranged from large to small:

λ₁≥λ₂≥…≥λ_i≥…≥λ_M

the above formula is divided into two cases:

using maximum eigenvalues lambda₁And minimum eigenvalue λ_MForming a two-dimensional signal feature vector:

T＝[λ₁,λ_M]

in the formula, T is a two-dimensional signal feature vector.

The training of the mean shift clustering algorithm in the step S4 specifically includes:

(a) each data point in the training set represents a two-dimensional signal characteristic vector obtained in a time period, and a data point is randomly selected from the training set to serve as a central point;

(b) finding out all data points of the circular area with the radius h away from the central point, and recording the data points as a set S_hThese points are considered to belong to cluster k;

(c) the calculation starts from the center point to the set S_hAdding the vectors of each element to obtain an offset vector;

(d) the center point moves along the direction of the offset vector, and the moving distance is the mode of the offset vector;

(e) repeating the steps (b), (c) and (d) until the size of the offset vector meets the set threshold requirement, and the central point at the moment is psi_k；

(f) Repeating steps (a), (b), (c), (d), and (e) until all data points are categorized;

(g) and according to each class, for the access frequency of each data point, taking the class with the highest access frequency as the class to which the current point set belongs.

In the step (c), the calculation method of the offset vector comprises the following steps:

in the formula, T_dIs S_hData points within the circular region, g being S_hNumber of data points, T, in the circular region₀Is the center point.

In step (d), the center point moves along the direction of the offset vector, specifically:

in the formula, M_h ^tFor the offset vector found in the t state,

is the central point in the t-state,

is the central point in the t +1 state.

In step S5, the classifier obtained after the training is represented as:

wherein the parameter gamma is used to control the false alarm probability P_f，Ψ₁Indicates after training at H₁Class center of state, Ψ₂Indicates after training at H₀Class center of state, if data point T_dIf the above formula is satisfied, the main user signal exists, the frequency band is occupied, and the frequency band cannot be counted; if the data point T_dIf the formula does not satisfy the above formula, the master user signal does not exist, the frequency band is idle, and the frequency band can be accessed.

In the specific implementation process, the provided algorithm is subjected to experiment and comparative analysis in a Matlab environment, in order to ensure the accuracy and reliability of the experiment result, the simulated main user signal adopted in the experiment is an AM signal, and the noise of the AM signal is ideal Gaussian white noise. The experiment extracted 2000 signal features, of which 1000 were used for training and 1000 for testing.

In simulation experiments, the CEMSC method proposed herein is compared with some popular spectrum sensing algorithms in the prior art. For example, a spectrum sensing method based on a Ratio of Maximum eigenvalue to minimum eigenvalue (MME) and a K-means clustering algorithm, a spectrum sensing method based on a Difference between Maximum eigenvalue and minimum eigenvalue (DMM) and a K-means clustering algorithm, a spectrum sensing method based on a Ratio of Maximum eigenvalue to trace (RMET) and a spectrum sensing method based on a K-means clustering algorithm.

FIG. 3 is a ROC graph showing simulation experiments performed under conditions where the number of sub-users M is 8, the number of sampling points N is 5000, and the SNR is-18 dB, where P is_dIndicates the probability of detection, P_fRepresenting the false alarm probability. As is clear from fig. 3, the CEMSC method proposed herein has better perceptual performance under the same conditions.

Fig. 4 shows an ROC graph of a simulation experiment performed under the conditions that the number of times of users M is 14, the number of sampling points N is 5000, and the SNR is-20 dB. As is clear from fig. 4, the CEMSC method proposed herein has better perceptual performance compared to the mentioned perceptual algorithm under the same conditions.

As is clear from fig. 3 and 4, the CEMSC method proposed herein can better improve the spectrum sensing performance. At low signal-to-noise ratios, the CEMSC method proposed herein still maintains good perceptual performance.

The same or similar reference numerals correspond to the same or similar parts;

the terms describing positional relationships in the drawings are for illustrative purposes only and are not to be construed as limiting the patent;

it should be understood that the above-described 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. Other variations and modifications will be apparent to persons skilled in the art in light of the above description. And are neither required nor exhaustive of all embodiments. Any modification, equivalent replacement, and improvement made within the spirit and principle of the present invention should be included in the protection scope of the claims of the present invention.

Claims (9)

1. A collaborative spectrum sensing method based on covariance eigenvalues and mean shift clustering is characterized by comprising the following steps:

s1: acquiring a plurality of received signal matrixes X under different time periods, wherein elements of the received signal matrixes X are received signals of different secondary users;

s2: calculating an eigenvalue of a covariance matrix of the received signal matrix X;

s3: forming a two-dimensional signal feature vector by using the maximum feature value and the minimum feature value obtained by calculation of S2;

s4: two-dimensional signal feature vectors obtained by calculation at different time periods form a training set, and the training set is used as the input of a training mean shift clustering algorithm;

s5: after training is finished, obtaining a classifier for judging whether an authorized channel is available;

s6: judging whether a master user exists or not by using the classifier obtained in the S5, if a master user signal exists, indicating that the frequency band is occupied, and not counting the frequency band; if the master user signal does not exist, the frequency band is idle, and the frequency band can be accessed;

the training of the mean shift clustering algorithm in the step S4 specifically includes:

(a) each data point in the training set represents a two-dimensional signal characteristic vector obtained in a time period, and a data point is randomly selected from the training set to serve as a central point;

(b) finding out all data points of the circular area with the radius h away from the central point, and recording the data points as a set S_hThese points are considered to belong to cluster k;

(c) the calculation starts from the center point to the set S_hAdding the vectors of each element to obtain an offset vector;

(d) the center point moves along the direction of the offset vector, and the moving distance is the mode of the offset vector;

(e) repeating the steps (b), (c) and (d) until the size of the offset vector meets the set threshold requirement, and the central point is psi_k；

(f) Repeating steps (a), (b), (c), (d), (e) until all data points are categorized;

(g) and according to each class, for the access frequency of each data point, taking the class with the highest access frequency as the class to which the current point set belongs.

2. The method for cooperative spectrum sensing based on covariance eigenvalue and mean shift clustering of claim 1, wherein in step S1, each secondary user has N sampling points.

3. The cooperative spectrum sensing method based on covariance eigenvalue and mean shift clustering of claim 2, wherein in step S1, the received signal of the ith secondary user is represented as:

in the formula, H₀Indicating a band idle state, H₁Indicating the occupied state of the frequency band, x_i(N) represents the received signal of the nth sampling point of the ith secondary user, 0 < N < N, w_i(n) represents a mean value of 0 and a variance of σ²White gaussian noise signal, s_iAnd (n) represents a primary user signal.

4. The cooperative spectrum sensing method based on covariance eigenvalue and mean shift clustering as claimed in claim 3, wherein in step S1, the received signal matrix X specifically is:

in the formula, x_iA sample vector representing the ith secondary user.

5. The method for cooperative spectrum sensing based on covariance eigenvalue and mean shift clustering as claimed in claim 4, wherein in step S2, the covariance matrix of the received signal matrix X is calculated by:

R＝E[XX^T]

wherein R represents a covariance matrix of a received signal matrix X, X^TDenotes the transposition of X, E [ ]]Indicating the desire.

6. The collaborative spectrum sensing method based on covariance eigenvalue and mean shift clustering as claimed in claim 5, wherein the maximum eigenvalue and minimum eigenvalue calculated in step S2 form a two-dimensional signal eigenvector in step S3, specifically:

the eigenvalues of the covariance matrix R are arranged from large to small:

λ₁≥λ₂≥…≥λ_i≥…≥λ_M

the above formula is divided into two cases:

using maximum eigenvalues lambda₁And minimum eigenvalue λ_MForming a two-dimensional signal feature vector:

T＝[λ₁,λ_M]

in the formula, T is a two-dimensional signal feature vector.

7. The collaborative spectrum sensing method based on covariance eigenvalue and mean shift clustering according to claim 1, wherein in step (c), the offset vector is calculated by:

in the formula, T_dIs S_hData points within the circular region, g being S_hNumber of data points, T, in the circular region₀Is the center point.

8. The collaborative spectrum sensing method based on covariance eigenvalue and mean shift clustering as claimed in claim 7, wherein in step (d), the center point moves along the direction of the offset vector, specifically:

in the formula, M_h ^tFor the offset vector found in the t state,

is the central point in the t state,

is the central point in the t +1 state.

9. The collaborative spectrum sensing method based on covariance eigenvalue and mean shift clustering according to any one of claims 7 to 8, wherein in step S5, the classifier obtained after training is expressed as:

wherein the parameter gamma is used to control the false alarm probability P_f，Ψ₁Indicates after training at H₁Class center of state, Ψ₂Indicates after training at H₀Class center of state, if data point T_dIf the above formula is satisfied, the main user signal exists, the frequency band is occupied, and the frequency band cannot be counted; if the data point T_dIf the formula does not satisfy the above formula, the master user signal does not exist, the frequency band is idle, and the frequency band can be accessed.

Images