Undetermined Blind Source Separation Of Radar Signals Based On Sparse Component Analysis

Blind source separation is an effective method to solve signal separation in the field of signal processing.It has been widely used in many research fields such as medical image,biomedicine,neural network,communication signal processing.In practice,the underdetermined blind source separation(UBSS)situation in which the number of observed signals is less than the number of source signals is more often encountered.Blind source separation only uses the received observation signal to extract and separate the source signals.According to the magnitude relationship between the observed signal and the number of source signals,it is defined as positive,underdetermined and overdetermined.Therefore,it is important to study the underdetermined blind source separation of radar signals.UBSS technology can separate complex mixed signals into multiple single-signal components.It has been widely used in electronic surveillance and radar applications.The main method to solve the problem of underdetermined blind source separation is "two steps" based on sparse component analysis theory.The existing radar reconnaissance algorithm recovers the source signal performance is not high.In view of the above problems,this paper proposes a new two-stage underdetermined blind source separation algorithm based on tensor decomposition and compressed sensing(CS).The main contents are as follows:(1)This paper introduces the model and basic theory of blind source separation,and introduces the clustering algorithms commonly used in hybrid matrix estimation: K-means clustering,fuzzy c-means clustering and density clustering,and expounds two main methods of blind source separation.Independent component analysis and sparse component analysis;(2)For the radar signal model,analyze its time-frequency domain characteristics and related characteristics,and analyze the accuracy error and application range of traditional eigenvalue decomposition,fuzzy C-means clustering and tensor decomposition method to estimate the hybrid matrix.In order to reduce the estimation error of hybrid matrix,an improved spectral clustering algorithm is proposed.In this improved method,the tensor decomposition algorithm is introduced to process the feature vector obtained in the process of spectral clustering algorithm to estimate the mixing matrix,and the estimation corresponds to the maximum.The eigenvector of the eigenvalues is used as the estimated mixing matrix,which greatly reduces the estimation error of the mixing matrix;(3)In the recovery process of source signal,a hierarchical coupled dictionary training method based on K-means singular value decomposition(K-SVD)is proposed.The dictionary training method obtains a priori training signal by pre-separation and uses the idea of hierarchical coupling to train effectively.Under the condition of unknown prior information,the two-stage method is used to realize the separation of radar signals.In the first stage,an algorithm based on spectral clustering and tensor decomposition is proposed to estimate the mixing matrix.In this method,the estimation corresponds to the maximum eigenvalue.The eigenvectors are used as the estimated mixing matrix.In the second stage,a unified model of UBSS and CS is first established to apply the reconstruction algorithm in the CS domain to UBSS.On this basis,the layered coupled dictionary training method based on k-means singular value decomposition recovers the source signal.The simulation results show that the source signal recovery performance of this method is better than the traditional method,and the recovery effect is better under the actual situation of unknown prior information.