Study On Blind Source Separation And Its Application In Brain Signal Analysis

Brain signals which reflect the brain function status are special bioelectricity activities. Brain signals are classified into electroencephalograph (EEG) signals and evoked potential (EP) signals. The EEG signals from scalp can be taken as electric potential distribution of the electric activities inside the brain on scalp, bioelectrical activities generated by other apparatus of human beings (including ECG, EOG, EMG and so on), and interferences brought by outside factors. These interferences outside brain are called artifacts. One of the main tasks of brain signal analysis is to eliminate the artifacts and get pure brain signals. The pure signals can be applied to clinical diagnosis and brain science research. EP signals are electric signals generated by the stimulated neural system. By the detection of EP, a number of pathological changes and damages can be diagnosed. The second task of brain signal analysis is to extract EP signals from spontaneous EEG in strong backgrounds.Blind Source Separation (BSS) technology is a new way to analyze signals, the essence of which is consistent with Independent Component Analysis (ICA). ICA is a technology for multi-channel signal processing, whose characteristics are to decompose the target signal into independent components without knowing any other priori knowledge besides the statistical independence between source signals. This paper mainly researches on the principles, criterions and optimized algorithms of ICA. The research results are used to remove artifacts and extract EP signals.Based on fast fixed-point algorithm of kurtosis (FastICA), this paper extends ICA with reference (rICA) and proposes an ICA algorithm with multi-references signals. The new algorithm reduces the computation quality compared with FastICA and doesn't need persons to judge the separated components. Moreover, it can extract more trails of signals than rICA, which can only extract single trail. This paper introduces the principles and algorithms of ICA with multi-references in detail. In addition, the paper proposes a concrete algorithm to eliminate artifacts and applies the algorithm to removing artifacts from brain signals. Simulations prove the effectiveness of the algorithm.ICA separates the multi-channel signals blindly in noise free conditions. However, the interference of additional noise is inevitable. In this situation, the performance of ICA goes down rapidly. The usual way is to remove the noise from multi-channel signals by using wavelet and take the result signals as the input of ICA. The steps are as follows: firstly, do...