Cooperative Spectrum Sensing Based On Statistical Manifolds And Machine Learning

Cognitive radio(CR)technology is a practical technology to solve the low utilization rate of spectrum resources,and spectrum sensing technology is the most critical link in the cognitive radio network.At present,there are many kinds of spectrum sensing algorithms proposed by domestic and foreign researchers,but most of them have great performance attenuation in the low signal-noise ratio(SNR)environment,and some of them can not even undertake the task of spectrum sensing in this environment.Therefore,to improve the spectrum sensing properties of the low SNR environment,the concept of the statistical manifold in information geometry is used to transform the spectrum perception problem into a geometric problem on the statistical manifold,and then the related algorithms of genetic algorithm and deep learning(DL)are applied to cooperative spectrum sensing(CSS).The specific contents of this thesis are as follows:To improve the sensing performance in a low SNR environment,a genetic simulated annealing algorithm based on quadratic covariance matrix and information geometry is proposed for cooperative spectrum sensing in a single antenna system.Firstly,the quadratic covariance matrix of cooperative sub-users(SUs)is used as the feature matrix for feature extraction.Secondly,based on information geometry,the eigenmatrix is mapped to the statistical manifold to avoid information loss.On this basis,the genetic simulated annealing algorithm is used to obtain the classifier on the statistical manifold,and a new mutation operator is used to improve the mutation process to speed up the convergence of the algorithm.Finally,a classifier is used to realize fast spectrum sensing.In the simulation analysis,the proposed method has better spectrum sensing performance and faster convergence speed than other popular methods under a low SNR environment.To solve the problem that excessive sensing signal data harms the computational overhead and computational pressure of fusion center(FC)in the multi-antenna system,and improve the robustness of the whole system,a convolution neural network algorithm based on quadratic covariance matrix and information geometry is proposed for cooperative spectrum sensing.A deep learning correlation network is used to improve the spectrum sensing performance of the whole algorithm.Firstly,each SU equipped with multiple antennas performs matrix fusion operation on the collected signal matrix through the concept of Riemannian matrix in information geometry to achieve data dimension reduction,and then sends the data to the fusion center through reporting channel.Secondly,the fusion center collects the received signal matrix to form the training data set.The convolutional neural network is trained and obtains a classifier model.Finally,the trained classifier can realize fast spectrum sensing.In the simulation analysis,this method still has better spectrum sensing performance under very low SNR environment,and has obvious performance advantages over other algorithms.