Research On Underdetermined Blind Separation For Frequency Hopping Signals Based On Time-Frequency Analysis

Frequency hopping communication is one of the most common way of spread spectrum communication.It has been widely used in the fields of military and civil communication with advantages such as low intercept probability,strong anti-jamming performance,adaptive networking ability,high confidentiality.Because of the strong non-stationary of hopping signal,so signal processing methods strong at non-stationary signals is needed.This thesis proposes an underdetermined blind source separation algorithm based on the time-frequency analysis.The main research in this thesis includes:Firstly,a brief description of the basic theory of the frequency hopping communication system is present.Several common time-frequency methods which are suitable with the frequency hopping signal characters are studied.The MATLAB simulation results have been discussed and compared.Based on the theory of blind source separation,the mixed model of frequency hopping signal underdetermined blind source separation condition is determined based on the theory of blind source separation.Common algorithms for mixing matrix estimation and signal source separation are present.Secondly,due to the heavy noise in the complex electromagnetic environment,an adaptive time-frequency supporting threshold setting method is proposed to eliminate the low energy noise inference.The combination of Davies-Bouldin Index rule with K-means cluster algorithms is used to get the time-frequency points and the following single source points are tested to improve the results precision.The time frequency ratio matrix is constructed by the time frequency single source point which is used in the estimation of the column vectors of the mixing matrix leading to the underdetermined blind source separation mixing matrix.Simulation results show that the improved algorithm has better estimation performance for the underdetermined blind source separation mixed matrix and the results improvements are significant.Finally,due to the difference carrier frequency of hopping signals,the short-time Fourier transform is applied in the time-frequency analysis.In the different periods of short-time Fourier transform window function,the hopping time and location of hopping frequency determination are determined by the judgments of the difference of the number of frequencies clustering.In combination with the time-frequency ratio matrix and the incidence Angle estimation algorithm,the sequential uncertainty of the blind source separation algorithm is improved,which leads to the realization of the estimation of time-varying mixing matrix.According to the improved subspace projection algorithm,the separation of multi-hop signal in time-frequency domain and further elimination of discrete noise interference can achieve better separation performance.The simulation results show that the improved algorithm has feasibility in the separation of multi-hop frequency signals.