Spectrum Sensing Research Based On Machine Learning And Information Geometry

Spectrum is a non-renewable natural resource.With the continuous development of modern wireless communication technology,the number of wireless devices has also increased rapidly,resulting in an increasingly tight spectrum of available spectrum resources.The licensed spectrum is basically used in daily life.However,the licensed spectrum will be in an idle state when it is not used for a certain period of time,which greatly reduces the utilization rate of the spectrum.Therefore,how to improve the utilization of spectrum under limited spectrum resources has become an urgent problem to be solved,and Cognitive Radio(CR)technology is considered to be an effective means to solve this problem.As a key step of CR technology,spectrum sensing can accurately detect the availability of licensed spectrum in a complex wireless environment in a flexible and intelligent manner.If the spectrum is in an idle state,the secondary user can dynamically access and use the spectrum reasonably,so as to achieve the purpose of improving spectrum utilization.In this paper,we will study methods that can effectively improve the performance of spectrum sensing by combining information geometry and machine learning.The main work is as follows:Firstly,the background and research significance of spectrum sensing are briefly introduced,and some current research status is discussed.Then,the methods and principles of single-user spectrum sensing technology and multi-user cooperative spectrum sensing technology are analyzed.At the same time,the advantages and disadvantages of these two technologies are summarized.In order to solve the problem that the traditional spectrum sensing method based on random matrix eigenvalues has poor sensing performance under low signal-to-noise ratio,a spectrum sensing scheme based on KL divergence and k-medoids clustering algorithm is proposed.KL divergence,as a measure of information geometry,contains more information about the geometry of the covariance matrix on the statistical manifold.Therefore,in this scheme,the KL divergence is used as the signal feature,and then the signal feature is trained on the k-medoids clustering algorithm to obtain the corresponding classifier.Finally,the classifier is used to determine whether the primary user signal exists.Simulation experiments show that the scheme can effectively avoid the calculation of the threshold value and improve the performance of spectrum sensing under low signal-to-noise ratio.In order to solve the problem of loss of covariance matrix information in traditional spectrum sensing methods based on random matrix eigenvalues,a spectrum sensing scheme based on Bhattacharyya mean and BDK clustering algorithm is proposed.In this scheme,the Bhattacharyya mean in the information geometry is used to fuse multiple signal covariance matrices to improve the estimation accuracy of the covariance matrix.At the same time,in order to cluster the points on the statistical manifold,a BDK clustering method is proposed in this scheme.The Bhattacharyya mean obtained after fusion is used to train the BDK clustering algorithm and obtain the corresponding classifier.Finally,the test set is input into the classifier to get the final decision result.According to the simulation results,it is verified that the scheme can reduce the influence of noise on the performance of spectrum sensing,and can fully utilize the information of the covariance matrix.