Novel frequency-based approach for detection of steady-state visual evoked potentials for realization of practical brain computer interfaces

Various algorithms for recognizing Steady-State Visual Evoked Potentials have been proposed over the last decade for realizing Brain-Computer Interfaces. However, frequency-domain techniques aside from classical FFT have been generally neglected. While close to perfect accuracies have been reported in SSVEP-based BCI studies, achieving high accuracy in a realistic scenario is still challenging. Here several frequency-domain algorithms were evaluated for SSVEP detection for the first time, and a new algorithm based on spectral averaging on resampled signal (SAoRS) was proposed, when a single EEG channel and high-frequency flickers were considered to improve user experience. Spectral Envelope (SE) and Maximum Entropy (ME) methods outperformed Burg, MUSIC, and Welch for processing window lengths of 0.5-2 s. The newly developed SAoRS algorithm significantly outperformed SE and the benchmark CCA algorithms for 0.5 s processing window. The results suggest that Spectral Envelop and SAoRS algorithms can provide high accuracies in SSVEP BCI systems.