Selection of an Optimal Feature Space for Separating Mental Tasks based on the EMD Algorithm

Designing brain-computer interface (BCI) systems is one of the concerns of people today. These systems operate by brain signals and so far, much research has been done in this regard. The most conventional systems are based on mental task signals. In the design of BCI systems based on mental activity, selecting a feature space with higher resolution and less processing time is important. In this study, Anderson mental task signals, a known and available database in such systems, were used. According to the nonlinear and non-stationary properties of electroencephalogram (EEG) signals, this study tried to review and analyze new empirical mode decomposition (EMD) algorithms, as well as conventional and successful methods such as autoregressive (AR) spectrum and entropy, for discrimination of mental task signals. EMD algorithm is compatible with nonlinear and non-stationary properties of EEG signals. Therefore, using an EMD algorithm along with the concept of entropy for modeling complexity values and AR spectrum, as a significant function in the frequency domain would provide great discrimination. Application of EMD algorithm and its parallel schemes (EMD entropy) would result in a feature vector with less dimensions requiring less than 2 seconds to extract features. Thus, such combination would require a maximum of 0.1 seconds to separate 10-second signals which can be beneficial in real-time BCI systems.