Research On Algorithm And Application Of Blind Source Signal Separation

Blind Source/Signal Separation(BSS)is the process of recovering components of source signals from the observed signals based on statistics characteristics of source signals.In BSS,source signals and transmission channel parameters are unknown.Because of stable theoretical basis and broad application prospect,BSS is one of the research hotspot in signal processing.Currently,BSS has been developed and applied in many areas,especially in biomedicine,electronic countermeasures,speech enhancement,remote sensing,seismic exploration,communication system,geophysics,econometrics,mechanics of machinery etc.BSS demands little prior information of mixed source signal and transmission channel.It can finish signal estimation and recovery at receiver end only,which is of incomparable advanced to other signal processing technologies.With applications of passive radar system,FH-signal system and adjacent satellite interference,this dissertation investigates BSS algorithm in theory and practical engineering to improve spectrum efficiency of communication system and signal detection performance.The main work includes four parts: 1)separation of weak signal with strong jamming in passive radar system,2)separation of blind source signal in orthogonal frequency hopping system,3)separation of undetermined blind source signal in non-orthogonal frequency hopping system and 4)separation of blind source signal with adjacent satellite interference.Detailed researches are as below:For blind source separation of weak signal with strong jamming,an interference elimination method in passive radar system is proposed.Under strong jamming condition,each observed signal's amplitude is quite different and preprocessing is needed to make sure succeeding signal separation can be completed.Interference signal estimation and reestablishment are included in preprocessing and the accuracy affects BSS performance directly.Based on estimation and reestablishment,this dissertation establishes BSS model with strong jamming.Weak target mixed signals are got after interference cancellation,and KM-FastICA algorithm is proposed for weak target mixed signals separation.After interference canceallation,the residual signals' intensity also affects separation performance.The impact of residual signals' intensity on mixed signal separation is analyzed according to information theory and then BSS method is proposed with condition that strong jamming is non-cooperative signals.The proposed interference elimination method cannot be applied directly.Under this condition,this dissertation proposes “separate first eliminate second” method,which has better performance than the condition that strong jamming is cooperative signals.This method relaxes the requirement of prior information and has broader application prospect.For separation of blind source signal in orthogonal(orthogonality means inner product of every two mixed source signals is zero)frequency hopping system,this dissertation proposes a separation algorithm based on Density Clustering algorithm(DC-algorithm),which contains two parts: 1)applying Short-Time Fourier Transform(STFT)with sparsity of FH-signal to get Time-frequency domain information of the sample signals,with which cost function pair(?,?)and decision coordinate system can be built.Then cost function pair(?,?)is used to perform density clustering on time-frequency domain value of sample signals to find cluster center.From which,mixed source signal number can be got,because number of mixed source signals equals to that of cluster centers.2)classifying sample signals with cluster centers and then recovers signals with inverse transform of STFT to finish BSS.For separation of undetermined blind source signal in non-orthogonal(non-orthogonality means inner product of every two mixed source signals is not zero)system,this dissertation proposes Matching Optimization Blind Separation algorithm(MOBS).Signals are classified into two parts: 1)Sample signals without collision.For this kind of signals,DC-algorithm is used to finish BSS.2)Sample signals with collision.Sample points for this kind of signal are sum of multiple signals,which means sparsity is no longer satisfied,so cluster method cannot be used.MOBS is proposed for this kind of signal.Based on signal sampling feature,target optimization cost function is proposed and optimal value of cost function is got with steepest descent method,convergence,complexity and impact of noise on algorithm are further analyzed.For blind source signal separation with adjacent satellite interference,this dissertation proposes a BSS method based on particle swarm optimization.This algorithm contains three steps: Firstly,time-frequency domain information of sample signals is got by computing STFT of each sample point.Secondly,K-means clustering algorithm is used to preprocess sample signals,thus get better separation performance and reduce computing complexity.Thirdly,a new iteration parameter is defined based on characteristics of adjacent satellite interference,and BSS with particle swarm optimization algorithm is finished.In particle swarm optimization algorithm,global optimal position Gb is replaced by cluster center set C.This dissertation discusses convergence and robustness of this algorithm and a series of simulation experiments are performed to prove the validity of the proposed algorithm.