Research On Robust Cooperative Spectrum Sensing Based On Information Geometry

With the rapid development of the 5G communication field,the explosive growth of wireless communication equipment makes the limited spectrum resources increasingly scarce.Therefore,it has far-reaching research significance to study how to solve the problems of tight spectrum resources and low utilization of spectrum.At present,many spectrum sensing schemes have been developed in cognitive radio technology to search for spectrum holes,thereby achieving the purpose of improving spectrum utilization.However,some of the spectrum sensing schemes still have some problems,such as: the calculation of the decision threshold is complex and inefficient,and the influence of individual abnormal user interference in the sensing environment is not considered.In order to solve the problems of insufficient performance and low robustness of some existing spectrum sensing schemes,this thesis proposes a scheme based on random theory to extract eigenvalues and two data fusion schemes based on information geometry theory,and applies machine learning to spectrum sensing.The specific content of this article is as follows:In order to improve the sensing performance of the algorithm in low signal-tonoise ratio scenarios and solve the problems that traditional algorithms relying on prior knowledge and calculate fixed thresholds,a spectrum sensing algorithm based on random matrix theory and mean shift clustering is proposed.In the single-antenna sensing scenario,the algorithm uses random matrix theory to construct a novel information feature vector.Different from distance-based clustering,this algorithm proposes a new spectrum sensing scheme based on the idea of density clustering.The idea of the overall scheme is as follows: features extraction of the signal matrix sensed by the secondary user to construct a two-dimensional information feature vector;cluster the feature vector based on the mean shift algorithm to construct a spectrum sensing classifier.It is used to determine whether there are spectrum holes in the sensing environment.Taking into account the interference of individual abnormal users to the global decision in the multi-user collaboration scenario,a spectrum sensing scheme based on Riemannian distance fusion and improved genetic algorithm is proposed.In the scenario of multi-antenna cooperative sensing,the solution is based on Information Geometry Theory(IGT)and uses geometric tools to fuse data on statistical manifolds,aiming to eliminate the interference of individual abnormal users.In order to realize the clustering process on the manifold,this scheme improves genetic algorithm based on Riemannian distance and symmetric Kullback-Leibler divergence distance The overall process of the scheme: map the local information covariance matrix collected by the secondary user to the statistical manifold,and perform data fusion based on the Riemannian distance to eliminate the interference of individual abnormal secondary user.Secondly,an improved genetic algorithm is used to train a spectrum decision classifier on the statistical manifold using the fusion samples to detect whether there is a Primary User signal in the perceptual environment.In order to further improve the spectrum sensing performance and the efficiency of data fusion,a scheme based on geodesic projection fusion and symmetric KullbackLeibler divergence distance-based particle swarm algorithm is proposed.Considering in the multi-antenna scenario,this scheme is based on the framework of information geometry theory and proposes a geodesic projection data fusion scheme to eliminate the interference of individual abnormal users.In order to realize the clustering process on Riemannian manifolds,a particle swarm optimization algorithm based on symmetric KL divergence distance is proposed.The overall process of the scheme:map the sampling covariance matrix perceived by each secondary user to the statistical manifold,perform data fusion processing based on the geodesic projection method to eliminate interference.and then the fused samples are directly clustered on the manifold by the improved particle swarm algorithm to obtain a spectrum sensing classifier.Finally,the classifier can be used to determine whether the secondary user is allowed to be connected to the authorized frequency band.It is worth noting that this algorithm has better data fusion efficiency than the previous paragraph,and the training of the classifier is easier to achieve convergence.