Research On Independent Component Analysis And Its Application In Extraction Of Signal

Blind signal processing is an important branch of signal processing.Independent component analysis is one of the main methods to solve the blind signal processing problem.It can successfully estimate the source signals and mixing matrix from the mixing signals,only using the basic assumption that the source signals are non-Gaussian and mutual independent without knowing the parameters of channel and source signals.The research of independent component analysis can be divided into real valued independent component analysis and complex valued independent.The real valued independent component analysis can further more be divided into linear independent component analysis and non-linear independent component analysis.The linear independent component analysis can also be divided into sparse independent component analysis and complete independent component analysis.The complex valued independent component analysis mainly solves the problems that how to estimate source signals from the complex valued linear mixed signals.Besides,there are noise independent component analysis and independent component analysis with reference that are derived from the independent component analysis.This paper mainly focuses on the real valued linear independent component analysis with reference,complex valued independent component analysis and their applications.The main contents are as follows:Firstly,to improve the convergence speed of an independent component analysis algorithm with reference signals,pre-whitening method is applied to deal with the observed data to reduce the algorithm complexity.The factors that affect the speed of independent component analysis with reference signal are analyzed.The cost function of independent component analysis with reference signal is optimized with Newton method with 2 +1 order convergence after preprocessing,and the independent component analysis method with faster convergence and reference signal is derived.Secondly,in order to improve the performance of the complex independent component analysis algorithm based on the two order statistics,a complex independent component analysis method based on steepest descent method and a complex independent component analysis method based on adaptive orthogonalization are proposed respectively.The method of complex independent component analysis method based on the steepest descent method firstly uses adaptive whitening processing of the signal,and then in the premise of maintaining the separation matrix for orthogonal matrix,with pseudo covariance matrix of signal separation for the diagonal matrix as the cost function,using the Riemann space under the steepest descent of the cost function optimization method do not need to derive additional complex independent component analysis method of orthogonal matrix separation force.Complex independent component analysis method based on adaptive orthogonal firstly using the same method of signal adaptive whitening and also in the pseudo covariance matrix is a diagonal matrix separation signal as the cost function,then the separation matrix orthogonalization and cost function to optimize the process of integration,complex independent component derived adaptive orthogonal separation matrix analysis method,this method makes the separation matrix in the iteration and does not require additional processing to meet the orthogonalization.Thirdly,to improve the speed performance of the fetal electrocardiogram extraction method based on independent component analysis,this paper firstly analyzes the relationship between the cost function and source signal kurtosis,then derives the mathematical expression between signal kurtosis and cost function.Secondly,this paper fully utilizes the per-information that probability density distribution of the fetal electrocardiogram is super Gaussian to simplify the cost function.Finally,Newton method is used to optimize simplified cost function and we derive the improved fetal electrocardiogram extraction based on independent component analysis with faster convergence speed.Finally,to improve the performance of voltage flicker detection,we construct complex valued independent component model of voltage flicker signal,and utilize the complex valued independent component analysis to separate the component of flicker signal.We use the separated flicker components and estimated mixing matrix to compute the parameters of voltage flicker to realize higher performance voltage flicker detection method.