Adaptive Bands on EEG Signals Extracted with Empirical Mode Decomposition

In this work, it has been analyzed the performance of the Empirical Mode Decomposition (EMD) of electroencephalographic (EEG) signals for classification of different mental tasks. The EMD is a suitable method for processing nonstationary and nonlinear signals, as the EEG. Moreover this method is adaptive to the signal and it is able to decompose the analyzed signal into a set of adaptive frequency bands. This is quite different from the traditional approach, where EEG signals, using a spectral analysis technique, are divided in several frequency bands with fixed bandwidth, namely, delta (<4 Hz), theta (4-7 Hz), alpha (8-13 Hz), beta (14-30 Hz) and gamma (>30 Hz). This paper presents an analysis of these adaptive bands. In order to contrast results, the proposed and the traditional methods were applied in a database of EEG signals acquired in 6 subjects performing 5 mental tasks in a Brain-Computer Interface experiment. It was found that using adaptive frequency bands of EMD technique the results were improved on an average of 12% respect to fixed bands. Also, it was observed an adaptation process trial over trial in the spectral content of adaptive frequency bands. Additionally, frequency content of adaptive bands shows some relationship with the performed mental task.