Research On Spectrum Sensing For Cognitive Radio Based On Machine Learning And Compressed Sensing

Spectrum sensing is one of the important technologies of cognitive radio(CR),which helps to alleviate the spectrum shortage in wireless communication.In recent years,more and more blind detection algorithms based on covariance matrix were applied to spectrum sensing.The performances of blind spectrum sensing are suceptible because the detection thresholds of blind spectrum sensing are approximative.At the same time,the tracking performances of the spectrum reconstruction algorithms are weak due to the high similarity of the sensing matrices column vector in the wideband compress spectrum sensing.The research on the setting of the asymptotic threshold for blind spectrum sensing by machine learning algorithms and the optimization of the compressed sensing matrix are carried out.The content and innovation of the full text are as follows:First,an improved blind spectrum sensing scheme is established by the covariance matrix Cholesky decomposition and radial basis function(RBF)-support vector machine(SVM)decision classication at low signal-to-noise ratios(SNRs).In the RBF-SVM training process,the ratio of the maximum-to-minimum eigenvalues of a covariance matrix obtained by the Cholesky decomposition are used to construct the eigenvectors,which contains the statistical information of each antenna on the SU.The spectrum decision classifier is generated in the training process,which replaces the asymptotic threshold of traditional blind spectrum sensing as a non-asymptotic threshold.Under strong background noises,the proposed scheme improves the recognition rate of primary users(PUs)than that of the current blind spectrum sensing and thus it can effectively realize the cooperative spectrum sensing with multi-antenna and multi-user.Second,the choice of parameters has a significant impact on the performance of SVM classification model.The Genetic Algorithm(GA)is used to optimize the parameters of mixed kernel SVM.The accuracy of the spectrum classification is selected as the fitness function valuse of GA algorithm and when it reaches the optimal value,the corresponding parameters of mixed kernel SVM are selected in training process.Compared with the scheme without optimization parameters,the missed alarm probability(Pm)and false alarm probability(Pf)of mixed kernel SVM scheme with GA optimization reduce by about 0.28 and 0.29,respectively,which improves the detection performance of the mixed kernel SVM in the blind spectrum detection.Third,in the wideband compressed spectrum sensing of CR,for the redundancy problem between the observation matrix and the sparse variation matrix,the correlation between the column vectors of the sensing matrix are reduced by the elements contraction calculation of the Grammatrix.Combining the sensing matrix optimization and sequential compressed sensing process,the spectrum sensing method of sequential adaptive compressed sensing based on the optimization of sensing matrix is proposed.The simulation results show that the proposed method has lower mean square error(MSE)than the existing sequential adaptive compressed spectrum sensing algorithm,and the detection probability is effectively improved.The optimization sensing matrix sparsity sdaptive matching pursuit(SAMP)scheme outperforms the SAMP method in detection probability about 0.03 at the ratio of 0.3.In this thesis,a non-asymptotic threshold method for blind spectrum detection by machine learning algorithm with optimized parameters and an adaptive compressed spectrum sensing scheme that optimizes the sensing matrix are studied.The research shows that both the proposed method and the optimization scheme improve the spectrum detection probability and the construction of eigenvector in the former significantly reduces the algorithm complexity,which are beneficial to improve spectrum sensing performance and reduces communication overhead in 5G communication.