Adaptive estimation of EEG-rhythms for event classification

Current brain computer interface (BCI) utilize electroencephalogram (EEG) rhythms associated with movement/ function to generate control signals. The amplitude of mu rhythm varies when the subject is not moving or not imagining and attenuates when the subject is moving or imagines movement. The classification of events is generally performed in frequency domain using fast Fourier transform (FFT) to compute band power. This papers aims to develop an alternative time-domain analysis by estimation of bandlimited signals through adaptive filtering. The design methodology estimates bandlimited signals through multiple Fourier series there by estimating the individual components of frequency weights through LMS algorithm. The knowledge of individual frequency components in time-domain provides useful insight into the classification process of EEG. Instead of using the band-power, this paper analyzes the usage of frequency weights to determine the optimum band for a subject. Study is conducted on 3 subjects for optimum band selection and classification.