Research On CCA Based ICA Algorithm And Its Application In EEG Signal Processing

Independent component analysis(ICA)refers to the process in which,under the condition that the source signals and transmission channel characteristics are unknown,the source signals are recovered and the transmission channel is estimated only by observed mixed signals and the mutually statistical independence among the source signal.At present,the technique of ICA have a widely applications in wireless communication,biomedical,image processing and speech signal processing.This paper focus on the independent component analysis algorithm via the canonical correlation analysis(CCA)approach and its applications in electroencephalogram(EEG)signal processing.The main work of this thesis are shown as follows: 1.The basic theory and method of the ICA problem are systematically discussed.The basic model and the basic assumptions in ICA problem are introduced,and the two indeterminacies inherent in ICA are analyzed.This paper summarizes several classical objective functions and optimization algorithms in independent component analysis.2.The CCA technique of ICA is introduced.The adaptive ICA algorithm and pre-whitening adaptive ICA algorithm based on CCA are both introduced.Pre-whitening of the observed data is a useful pre-processing method in ICA.The correlation between the data can be eliminated by pre-whitening processing,thus the workload of the independent component analysis can be reduced.Simulation results show that the separation effect of the CCA based on pre-whitening adaptive ICA algorithms is better than the non-pre-whitening algorithms.3.The objective function is the key to deduce the ICA algorithms,which determines the algorithm statistical performance.Based on the analysis of classic CCA contrast function,this thesis proposed a new contrast function.Theory analysis shows that the new contrast function has unique global minimal point,which is just the maximal point of classical canonical correlation analysis criterion.Then,based on the new contrast function,a fast and efficient adaptive algorithm using quasi-Newton iteration is proposed.Simulation results show that the new algorithm possess has faster convergence speed than the existing algorithms and the better separation accuracy.4.The proposed adaptive quasi-Newton algorithm based on CCA is applied to the clinical EEG signals for separation and artifact removal.The simulation results show that the proposed adaptive quasi-Newton algorithm based on CCA can complete the EEG signals separation and can successfully remove electrooculogram(EOG)artifacts in EEG signals.Compared with the existing algorithms,the algorithm converges faster and more stable.