Research On Cooperative Spectrum Sensing Technology Based On Unsupervised Learning

With the rapid development of broadband mobile communication technologies 5G and B5G and the massive access of intelligent terminal equipment,coupled with the complex network characteristics of high heterogeneity and high dynamics,the current spectrum working environment is becoming more complicated,and the contradiction between the supply and demand of the spectrum is intensifying.Cognitive radio,as one of the important technologies to solve the contradiction between spectrum supply and demand,its core technology,spectrum sensing,faces the problems of low sensing efficiency,poor sensing accuracy,and complicated sensing devices in the current network environment.Applying machine learning to cooperative spectrum sensing has become an effective solution to the problems of low sensing efficiency and poor accuracy in traditional spectrum sensing algorithms.Therefore,this paper proposes a new unsupervised learning-based cooperative spectrum sensing scheme that combines principal component analysis(PCA)with improved K-means clustering(K-Means++)or Gaussian mixture model(GMM)to improve the efficiency and accuracy of spectrum sensing in the complex dynamic cognitive networks.In the sensing schemes proposed in this paper,the energy information perceived by cognitive users is first divided into different levels of energy vectors through the data fusion center,and the energy vectors form a feature matrix.Secondly,the PCA algorithm is used to convert the feature matrix into a low-dimensional feature matrix to reduce the amount of spectral data and training complexity.Then,the low-dimensional feature matrix is used to train the unsupervised learning K-Means++/GMM cluster.Finally,the trained K-Means++/GMM cluster can be used to divide the required spectrum into idle spectrum(accessed by cognitive users)and busy spectrum(not accessed by cognitive users)In order to verify the performance of the proposed cooperative spectrum sensing scheme and to select the best cooperative spectrum sensing scheme,this paper proposes four schemes of K-Means++,GMM,PCA-K-Means++and PCA-GMM in two scales of cognitive networks.Simulation analysis and comparison were carried out,and the scheme with the optimum sensing performance was selected.The simulation results show that in the schemes of PCA-K-Means++and PCA-GMM,the dimension of the energy feature matrix(input data)is reduced,and the training time is significantly reduced while the sensing accuracy is not affected.The PCA-GMM training time is an order of magnitude less than PCA-K-Means++.In addition,for the power of the primary user(PU),when the power of PU is 200mW,the detection accuracy of PCA-K-means++ and PCA-GMM is very high,and close to 1.Therefore,among the several proposed cooperative spectrum sensing schemes,when the PU's power is 200 mW and the smaller number of cognitive users,the cooperative spectrum sensing scheme using the PCA-GMM performs is the best.