EEG signal classification using time-varying autoregressive models and common spatial patterns

The performance of EEG signal classification methods based on Common Spatial Patterns (CSP) depends on the operational frequency bands of the events to be discriminated. This problem has been recently addressed by using a sub-band decomposition of the EEG signals through filter banks. Even though this approach has proven effective, the performance still depends on the number of filters that are stacked and the criteria used to determine their cutoff frequencies. Therefore, we propose an alternative approach based on an eigenstructure decomposition of the signals' time-varying autoregressive (TVAR) models. The eigen-based decomposition of the TVAR representation allows for subject-specific estimation of the principal time-varying frequencies, then such principal eigencomponents can be used in the traditional CSP-based classification. A series of simulations show that the proposed classification scheme can achieve high classification rates under realistic conditions, such as low signal-to-noise ratio (SNR), a reduced number of training experiments, and a reduced number of sensors used in the measurements.