Enhancing unsupervised canonical correlation analysis-based frequency detection of SSVEPs by incorporating background EEG

Steady-state visual evoked potential (SSVEP)-based brain-computer interfaces (BCIs) have potential to provide a fast communication channel between human brain and external devices. In SSVEP-based BCIs, Canonical Correlation Analysis (CCA) has been widely used to detect frequency-coded SSVEPs due to its high efficiency and robustness. However, the detectability of SSVEPs differs among frequencies due to a power-law distribution of the power spectra of spontaneous electroencephalogram (EEG) signals. This study proposed a new method based on the fact that changes of canonical correlation coefficients for SSVEPs and background EEG signals follow the same trend along frequency. The proposed method defined a normalized canonical correlation coefficient, the ratio of the canonical correlation coefficient for SSVEPs to the mean of the canonical correlation coefficients for background EEG signals, to enhance the frequency detection of SSVEPs. An SSVEP dataset from 13 subjects was used for comparing classification performance between the proposed method and the standard CCA method. Classification accuracy and simulated information transfer rates (ITR) suggest that, in an unsupervised way, the proposed method could considerably improve the frequency detection accuracy of SSVEPs with little computational effort.