Forty-Class SSVEP-Based Brain-Computer Interface to Inter-subject Using Complex Spectrum Features

The Steady-State Visually Evoked Potential (SSVEP) is one of the most popular paradigms for Brain-Computer Interface (BCI) applications. In this study, we address two challenges in designing SSVEP-based BCI. Firstly, our BCI system must be able to discriminate among the 40 available visual stimuli. In addition to the complexity brought by the high number of classes, visual stimuli flicker at close frequencies, only 0.2 Hz apart in the range of 8 to 15.8 Hz. The second challenge we addressed was the attempt to eliminate individualized system tuning. Our SSVEP-based BCI was designed using only data from subjects other than the user, that is, with cross-subject training. In the treatment of these two challenges, we extracted features with frequency and phase information for each of the 40 visual stimuli and applied them to a Linear Discriminant Analysis. The database has data from 35 subjects, so we trained with 34 subjects and tested with the remaining ones. We applied three different time windows of 1, 2 and 3 s to segment brain data and analyze the effect on classification accuracy. Our results reached an average classification, considering 40 classes, of 28.14%, 56.85% and 71.45% for a time window of 1, 2 and 3 s, respectively.