An EEG-Based Brain-Computer Interface Using Spectral Correlation Function

Brain-Computer Interface (BCI) is a promising technique because of its wide variety of applications, from treating cognition in humans to person authentication. Brain signals can be transmitted straight to a prosthetic device from the BCI system, without the need for nerve or muscle activity. For accurately identifying the transmitted signals at the prosthetic device, considering the nature of the Electroencephalography (EEG) signal, and extracting the most informative features are effective keys. In this paper, we studied the cyclostationarity of the Slow Cortical Potential (SCP) EEG signals for BCI applications, following our previous studies. Cyclostationary analysis reveals the hidden periodicity in the signal and provides a second-order statistical description in the frequency domain. We used the FFT Accumulation Method (FAM), an effective computational algorithm, to extract the features of the Spectral Correlation Function (SCF). The features are classified using SVM RBF, SVM polynomial, and K-Nearest Neighbor classifiers, and they are considered with different pre-processing. Our research indicates that the SCP EEG signal has cyclostationary properties and this idea is applied to the BCIs as well. The classification accuracy on the BCI Competition 2003 dataset Ia’s increased considerably, by spotting the intrinsic correlation between just two EEG signals.