Common spatial spectral pattern for motor imagery tasks in small channel configuration

Targeting to the deficient recording condition with 2-3 channels, a novel method, multiple time-delayed common spatial spectral pattern (CSSP), was proposed in this paper. In this approach, 2-10 time-delayed Electroencephalogram (EEG) signals for each channel were used to expand the number of EEG channels and then common spatial pattern (CSP) was used to extract the features. Mutual information criterion was applied to choose the optimal number of time-delayed signals. The proposed method was tested by the IIb datasets of 9 subjects in BCI Competition 2008 and datasets of 13 subjects in our experiments. The average Kappa coefficients for the two datasets are 0.6 and 0.34, respectively. This method automatically weights the importance of frequency in 8-30 Hz and enhances the classification accuracy in the case of deficient recording consequently.