Development of an SSVEP-based BCI spelling system adopting a QWERTY-style LED keyboard

► We proposed an SSVEP mental speller that can spell one character per each target selection. ► A QWERTY style LED keyboard was used to implement the proposed BCI speller. ► Human SSVEPs with a frequency resolution of 0.1Hz can be classified with high accuracy. ► The average LPM of 9.39 is one of the best results reported in previous BCI literature. In this study, we introduce a new mental spelling system based on steady-state visual evoked potential (SSVEP), adopting a QWERTY style layout keyboard with 30 LEDs flickering with different frequencies. The proposed electroencephalography (EEG)-based mental spelling system allows the users to spell one target character per each target selection, without the need for multiple step selections adopted by conventional SSVEP-based mental spelling systems. Through preliminary offline experiments and online experiments, we confirmed that human SSVEPs elicited by visual flickering stimuli with a frequency resolution of 0.1Hz could be classified with classification accuracy high enough to be used for a practical brain–computer interface (BCI) system. During the preliminary offline experiments performed with five participants, we optimized various factors influencing the performance of the mental spelling system, such as distances between adjacent keys, light source arrangements, stimulating frequencies, recording electrodes, and visual angles. Additional online experiments were conducted with six participants to verify the feasibility of the optimized mental spelling system. The results of the online experiments were an average typing speed of 9.39 letters per minute (LPM) with an average success rate of 87.58%, corresponding to an average information transfer rate of 40.72 bits per minute, demonstrating the high performance of the developed mental spelling system. Indeed, the average typing speed of 9.39 LPM attained in this study was one of the best LPM results among those reported in previous BCI literatures.