Performance Comparison of Spatial Filter with Multiple BMFLCs for BCI Applications

The subjects can learn to modulate their EEG pattern to achieve multiple targets in brain-computer interface systems. The modulation can occur in both α and β bands of the EEG signal. To successfully identify these modulated EEG patterns, multiple band-limited multiple Fourier linear combiner (BMFLCs) are employed to estimate the amplitude variations in EEG. To achieve better signal to noise ratio, spatial filter is paired with BMFLC for classification with linear discriminant analysis. Various existing spatial filters are paired with BMFLC and the performance is compared to identify the best spatial filters for classification of four targets in BCI Competition 2003 data set II(a). Results show that the Tikhonov regularized common spatial filter (TRCSP) paired with BMFLC provides better accuracy in comparison with other spatial filters.