Adaptive modelling of eeg signals to produce accurate time-frequency decompositions for use in BCI

Abstract

Motor imagery and actual movement are both tasks that bring forth a noticeable change in the subject's EEG mu rhythm known as Even-Related Desynchronisation (ERD). They appear as magnitude decreases of the frequencies included in the said band and can be tracked and measured for the automatic real-time detection and classification of the events. It has been proven that an important percent of the changes happen within a narrow frequency band called a reactive band, providing thus the means to significantly improve the efficiency of the interpretation of such events by concentrating the decisive information. The algorithm presented in this paper automatically identifies a subject's specific reactive band by detecting the highest decrease in power. The decision is made after the recorded EEG signal is modeled with a Band-Limited Multiple Fourier Linear Combiner (BMFLC). The method adaptively estimates the amplitude of each frequency component in the given band of interest and produces a precise time-frequency map that can be afterwards used for increasingly accurate classification and BCI applications.