Research On Spectrum Sensing Algorithm Based On Random Matrix And Gaussian Mixture Model

Currently,as the number of wireless access devices continues to increase,spectrum resources are becoming more and more tense.However,many licensed bands are idle most of the time and are not used by authorized users.Cognitive radio,as an important technology to improve spectrum utilization,has attracted more and more researchers' attention.Spectrum sensing is one of the key technologies of cognitive radio.The goal of spectrum sensing is to detect the frequency band and the status of the primary user by continuously detecting the target frequency band,and detect the spectrum holes.In order to accurately calculate the spectrum holes in real time,spectru m sensing has become a research hotspot in the field of wireless communication.This paper mainly studies the application of random matrix theory and Gauss mixture model in spectrum sensing.The research of spectrum sensing technology mainly includes two aspects: One is single-user detection,the other is multi-user cooperative detection.Firstly,several existing spectrum sensing algorithms are summarized.Several algorithms based on transmitter sensing are introduced,such as energy detection algorithm,m atched filter detection algorithm and cyclostationary feature detection algorithm.Some shortcomings of single-user spectrum sensing technology lead to the centralized cooperative spectrum detection technology in multi-user cooperative spectrum sensing,focusing on the more mature data fusion method centralized hard fusion method.This includes the “and” criteria for this approach,the “or” criteria,and the “K-rank” criteria.Then the application of random matrix theory in spectrum sensing is briefly introduced.Finally,in order to improve the traditional spectrum sensing model,a spectrum sensing model based on machine learning is introduced,and the process of spectrum sensing based on machine learning is also analyzed.Firstly,the application of random matrix theory in spectrum sensing is studied.Because the computational complexity of the spectrum-aware algorithm based on random matrix theory is relatively high,a spectral sensing algorithm(MSE)that combines the difference between the largest eigenvalue and the average eigenvalue and a spectrum sensing algorithm(ED-MSE)combined with the energy detection algorithm is proposed.In high Signal-to-noise ratio,energy detection algorithm is used to detect,which reduces the amount of calculation;in low Signal-to-noise ratio,MSE algorithm is used to detect.The detection performance of ED-MSE algorithm is better than that of the other two algorithms.Then the spectrum sensing algorithm based on machine learning theory i s studied.Because the derivation of decision threshold of spectrum sensing algorithm based on random matrix theory is complex and inaccurate,and in order to improve the spectrum sensing performance in attenuated channels,a multi-antenna cooperative spectrum sensing method based on wavelet transform and Gauss hybrid model is proposed.This method combines multi-user multi-antenna technology to improve diversity and reduce the influence of path attenuation and shadows on spectrum sensing performance.In th e aspect of signal feature calculation,in order to reduce the influence of noise on the calculation of signal characteristics,wavelet transform is used to denoise the signal.Then,a spectrum sensing classifier is obtained by training the Gauss mixture c lustering algorithm in machine learning,which avoids the complex decision threshold deduction.Finally,the trained classifier is used for spectrum sensing.In the experimental simulation part,the performance of Rayleigh attenuation channel and Rician attenuation channel is simulated,and the performance of several different features is compared with that of K-means clustering algorithm.The results show that the multi-antenna cooperative spectrum sensing method based on wavelet transform and Gauss mixture model with MSE features has better spectrum sensing performance.