Complex CNN incorporating Hilbert transform for steady-state visual evoked potential BCI

Brain computer interface (BCI) is a system that converts brain activity information into commands, allowing people to communicate with the outside world without the need for physical activity. Steady-state visual evoked potential (SSVEP) which is a kind of event related potentials (ERPs) is used for BCI. SSVEP shows higher performance and has been improved in various ways to to improve the information transfer rate and expand the number of commands, such as phase-modulated SSVEP-BCI. To further improve the performance of SSVEP-BCI, the frequency and phase of SSVEP needs be detected in a short time.In this study, we apply the Hilbert transform to the complex-valued convolutional neural network (CVCNN) for phase-modulated SSVEP-BCI. The Hilbert transform and accompanying analytic signal enable us to determine the instantaneous frequency and phase. The proposed CVCNN incorporating the Hilbert transform not only detects instantaneous frequency and phase information from electroencephalogram (EEG), but also learns convolutoinal filter automatically to detect SSVEP frequency, phase, and their harmonics components. As the result of evaluation experiments on open SSVEP datasets, the proposed method showed higher accuracy than conventional methods.