Eeg Signal Dsp-based Blind Separation

For the extremely weak, brain source signals, an important issue is how to use noised EEG/MEG data to detect reliably, enhance signal and position them. It is effective that EEG/MEG data's noise reduction and interference elimination using Independent Component Analysis (ICA) methods, and the method reveals important development prospects in depicting human cognizance and the processing of nervous system.Firstly, we introduced the basic principles of independent component analysis. The basic theory of blind signals separation (BSS) algorithms, and the statistical signals processing includes higher order cumulants and information theory is described. Some classical blind signals separation algorithms' principles are compared. The criteria of algorithm's performance are mentioned.Secondly, based on information theory, combining fixed-point algorithm and extended-ICA, a fast BSS algorithm is proposed aimed at linear-instantaneous mixed model. This algorithm is effective to separate EEG signals with impulsive noise. Eigenvalue decomposition method is used to whiten signals. Minimized mutual information is choosed for ICA contract function that measure the mutual independence of the output signal. The Newton iterative method is applied to obtain algorithm's expression. By the t-distribution model with family of super-Gaussian distribution and the model with super-Gaussian distribution, we can separate the mixture of sub-Gaussian and super-Gaussian source components. Through simulation, we illustrate the efficacy of this approach with noiseless signals and the robustness with low signal noise ratio (SNR) signals.With programmable and powerful calculation ability, digital signal processor(DSP) make it possible to let BSS apply to medical instrument from laboratory simulation. Because of the DSP's limit digit, cutting-tail error occurs during hardware's realization. Using Matlab simulate core function's approximate calculation and cutting-tail error's impact on algorithms, when the algorithms work in hardware. Simulation experiments showed that under the approximate calculation and finite length cutting-tail situation, separation algorithms remain valid. The effective digit length is confirmed. The blindsignals separation algorithms are accomplished by using the TMS320C54X C language, and it is the foundation for hardware realization.