How to identify the user specific stimulation frequencies for SSVEP-based BCI

The aim of this paper is to compare some approaches used for detecting user specific stimulation frequencies in terms of the number of detected Steady State Visually Evoked Potentials (SSVEPs). We performed our analysis with 6 subjects with respect to three factors: channels, time windows and harmonics/subharmonics used for power spectrum calculations. Results: 1. The best results across original channels and their combinations were obtained for O2 referenced to pz and for the average of O2 and O1 referenced to Pz. 2. The length of time window (2, 5, or 10 seconds) did not have an influence on the average results. 3. The best results across the harmonics/subharmonics used for power spectrum correction were obtained when the first harmonic was added to the original power spectrum. 4. While with the best combination of all three factors (the average of O2 and O1 referenced to Pz; time window of 10 seconds; fundamental frequency + first harmonic in the power spectrum), we were able to detect 41% out of all 85 stimulation frequencies used in the survey, only 13% was detected with the worst combination (the O1 referenced to O2; time window of 2 seconds; fundamental frequency + first subharmonic in the power spectrum).