Remove Noise Research Of The EEG Based On Underdetermined Blind Source Separation Method

Brain-computer interface（BCI） is a new interactive system that does not reliy onexternal nerve and muscle to complete the communication with the outside world, ittransforms the biological signal into computer commands to complete thehuman-computer interaction. When patients can’t normal communication with the outsideworld because of external nerve and muscle severely damaged, then they cancommunicate with the outside world by BCI, so BCI has profound significance forrehabilitation engineering. Brain-computer interface studies has several aspects such as thesignal preprocessing, removing noise, feature extraction, feature classification andrecognition, all of these aspects will affect the performance and effect of BCI. Denoisingof electroencephalogram is an interesting problem for BCI research, we propose a kind ofunderdetermined blind source separation based on independent component analysismethod of removing EEG’s noise in this paper.Firstly, non-stationary EEG is partitioned into smooth approximation of the blockEEG in this paper, then using blind source separation method of removing noise. Blindsource separation method can be divided into two steps, mixing matrix estimation andrecover the source signals. This paper proposes a better robustness of mixing matrixestimation method based on second-order blind identification of underdeterminedmixtures（SOBIUM）. The method uses LM iterative algorithm to optimize the mixingmatrix of result of SOBIUM, so as to improve the accuracy of identifying mixed matrix.Secondly, this paper puts forward an improvement of beamforming method toseparate the source signal for beamformer of underdetermined blind source separation. Itis a multiple input multiple output filter based on minimizing the theoretical mean squaredistance of the estimated signal from the original ones. Also, this method takes theabsolute value of the autocovariances of original signal and constraints the AR coefficients,to ensure that the energy of output signal as a negative value to improve the stability ofbeamformer.Lastly, underdetermined blind source separation method is used to remove noises insmall number of channels of EEG, furthermore, simulation experiment utilizing BCI Competition III dataset and Competition IV₁dataset shows that the method can perfectlyseparate the source signals of EEG, this method also achieves ideal result in the denoisingof EEG.