Introduction: Brain--computer interfaces (BCIs) based on steady-state visual evoked potentials (SSVEPs) are widely used alternative communication modalities due to their high information transfer rate and systematic responses in occipital cortical areas. In particular, single-flicker modalities have been of great interest in recent years. These have mainly been studied using many recording channels in the occipital area. To minimise the number of recording electrodes, we analysed the clustering of SSVEPs under different stimulation conditions (gaze directions) using a single channel.
Methods: Using a publicly available dataset, EEG signals were recorded from 7 subjects exposed to a central flickering visual stimulus (15 Hz) surrounded by four static targets in the cardinal directions. Participants focused their gaze sequentially on these targets while their cortical responses were recorded. We discretized the energy of the SSVEP, analysed it according to its principal components, and quantified the resulting clustering using the Calinski--Harabasz (CH) index.
Results and Discussion: The energy patterns showed specific characteristics that allowed them to be grouped according to the different stimulation conditions. Principal component analysis revealed that the first three components explained, on average, 93% of the total variance of the data. Quantification of separability using the CH index showed that up to four different stimuli were effectively grouped by a single channel, with an average CH index of 296.36 for the best channel, indicating excellent separability between states. Pairwise comparisons showed that the N--S and N--W directions were the most discriminable, with an average CH above 200 in all cases. A between-subjects analysis revealed that this clustering efficiency was mainly located in a specific region of the occipital cortex (O1, O2, and Oz), where CH indices were consistently higher.
Conclusions: This approach demonstrates the feasibility of significantly reducing the number of recording channels in the implementation of a single-flicker SSVEP BCI.
