Study On Signal Processing Of BCI Based On Multivariate Empirical Mode Decomposition

In the study of brain-computer interface, it is not appropriate for the non-stationaryand multichannel brain signals to apply the conventional signal processing methods basedon some basic function, to analyse a few channels synchronously. The EMD method doesnot need the basic function and can decompose one-dimensional nonstationary nonlinearsignals adaptively, providing a new direction for brain signal processing. Aiming at themultichannel problem of brain signal in this paper, we analyse the property ofmultivariable EMD and study its application in brain signal processing in BCI.Firstly, the feature extraction method of MEMD combining interval maximum poweris presented. The multichannel signals are decomposed into multiscale featuressynchronously. The performance of the proposed method can be verified by experimentson BCI Competition III data set I and BCI Competition IV data sets3. The results meetthe level of competition winner group, which shows that this algorithm is very suitable forbrain signal feature extraction. Effectiveness and stability are relatively good.Then, Research the methods of signal assisted MEMD for feature extraction in MEG.Increase a few channels Gaussian White Noise or a pulse signal as the assisted channels ofthe original multichannel data, then decompose by MEMD, removing the IMFscorresponding to assisted channels. BCI Competition IV data sets3is used for test.Experimental results both are better than the one of standard MEMD. The signal assistedMEMD methods can effectively reduce the frequency mixing and Improve performance.Finally, by investigating the obstacles of EMD，the filer bank property of MEMD,component estimation using Multivariate IMFs, we prove the superiority of MEMDcompared with EMD. Using BCI Competition IV Data set I to experiment, CWT, EMD,EEMD, MEMD are applied for preprocessing and CSP, CSSD is for feature extraction.The resuls of experimental comparative analysis indicate the eminent applicability andeffectiveness of multivariate EMD method for EEG signal and further improveperformance of noise-assisted EMD method.