Dual-Frequency SSVEP-based BCI for Reducing Eye Fatigue and Improving Classification Rate

Abstract

The steady-state visual-evoked potential (SSVEP)-based brain-computer interface (BCI) has been widely investigated because of its high signal-to-noise ratio (SNR), and little requirement for training. However, the stimulus for evoking SSVEP causes high visual fatigue and has a risk of epileptic seizure. Furthermore, stimulation frequency is limited and the SSVEP amplitude diminishes when a monitor is used as a stimulator. In this thesis, a dual-frequency SSVEP is examined to resolve the aforementioned issues. Employing dual-frequency SSVEPs, two novel SSVEP-based BCIs are introduced to decrease eye fatigue and use harmonic frequencies with increased performance. First, the spectral characteristics of dual-frequency SSVEPs are investigated and frequency recognition methods for dual-frequency SSVEPs are suggested. Three methods based on power spectral density analysis (PSDA) and two methods based on canonical correlation analysis (CCA) were tested. The proposed CCA with a novel reference signal exhibited the best BCI performance, and the use of harmonic components improved the classification rate of the dual-frequency SSVEP. Second, the dual-frequency SSVEP response to an amplitude-modulated stimulus (AM-SSVEP) was explored to verify its performance with reduced eye fatigue. An amplitude-modulated stimulus was generated using the product of two sine waves at a carrier frequency (fc) and a modulating frequency (fm). The carrier frequency was higher than 40 Hz to reduce eye fatigue, and the modulating frequency ranged around the α-band (9–12 Hz) to utilize low-frequency harmonic information. The feasibility of AM-SSVEP with high BCI performance and low eye fatigue was confirmed through offline and online experiments. Using an optimized combination of the harmonic frequencies, the online experiments demonstrated that the accuracy of the AM-SSVEP was 97%, equivalent to that of the low-frequency SSVEP. Furthermore, subject evaluation indicated that an AM stimulus caused lower eye fatigue and less perception of flickering than a low-frequency stimulus, in a manner similar to a high-frequency stimulus. Third, a novel dual-frequency SSVEP-based hybrid SSVEP-P300 speller is introduced to overcome the frequency limitations and improve the performance. The hybrid speller consists of nine panels flickering at different frequencies. Each panel contains four different characters that appear in a random sequence. The flickering panel and the periodically updating character evoke the dual-frequency SSVEP, and the oddball stimulus of the target character evokes the P300. Ten subjects participated in offline and online experiments, in which accuracy and information transfer rate (ITR) were compared with those of conventional SSVEP and P300 spellers. The offline analysis revealed that the proposed speller elicited dual-frequency SSVEP. Moreover, the dual-frequency SSVEP significantly improved the SSVEP classification rate and ITR with a monitor in online experiments by 4 % accuracy and 18.8 bpm ITR. In conclusion, the proposed dual-frequency SSVEP-based BCIs reduce eye fatigue and improve SSVEP classification rate. The results indicate that this study provides a promising approach to make SSVEP-based BCIs more reliable and efficient for practical use.