Energy Detection And Feature Recognition Of Wireless Radio Frequency Signal

The rapid development of wireless communication technology and the rapid increase of wireless traffic have encouraged the demands to make full use of channel capacity,different kinds of wireless access technology.This directly leads to the problem that different wireless access technologies may compete for the limited spectrum resources or even illegally access the network.Therefore it is very important to detect and recognize different wireless radio frequency signals in time.This thesis focuses on the energy detection and feature recognition technology of wireless radio frequency signal,the main work and innovations are as follows:Firstly,energy detection method is studied,in view of the problem that the detection performance of the traditional energy detection method and double-threshold energy detection method is easily affected by the noise uncertainty,results in low detection accuracy.An energy detection method combined variance correction is proposed to ease the impact of background noise on the detection performance considering signal energy as well as the variance of signal energy in multiple segments.Simulation results prove that the proposed energy detection method can improve the accuracy of signal detection greatly.Secondly,signal recognition technology based on machine learning is studied,in view of the problem that the recognition performance of wireless radio frequency signal recognition method based on the instantaneous feature and high order cumulants is poor,a signal recognition method based on sequential forward selection is proposed,which combines the instantaneous feature and high order cumulants to compose a statistic feature vector with high dimension,then a feature subset with better recognition performance is selected with the help of sequential forward selection algorithm,improving signal recognition accuracy significantly.In addition,a wireless radio frequency signal recognition method based on cyclic time domain profile is also proposed,which extracts the maximum value of cyclic autocorrelation function through the cyclic frequencies at each time delay,composing one dimensional feature vector.This method reduces the computational complexity of operating two dimensional cyclic autocorrelation function significantly,and simulation results reveal that the proposed method has excellent signal recognition performance.