Wavelet analysis of P300 EEG signals for brain-computer interface applications

Abstract

In this work the use of Time-Frequency techniques in the analysis of electroencephalographic (EEG) signals for brain-computer interface (BCI) applications is investigated. The continuous wavelet transform (CWT) is used to automatically select robust features in an offline P300 BCI system. Examination of the entire Time Scale domain exhibits additional features to the P300, including the N1 component of the event-related potential (ERP) and event-related desynchronization (ERD) at the end of target epochs. Information corresponding to the shape of the Time-Scale function near extrema points is also proposed as a relevant feature parameter, and shown to improve feature selection and robustness. Comparative results indicate that the Market wavelet is more suitable than the Mexican Hat wavelet in extracting ERP information, contrary to hypotheses in recent literature. Receiver operating characteristic curves are used as a more comprehensive measure of BCI system performance. This work shows that summing the Time-Scale coefficients in regions around extrema can be used to achieve 100% correct target character detection by extracting features from seven averaged singe-trials, using information from only one EEG channel.