Improving the performance of the SSVEP-based BCI system using optimized singular spectrum analysis (OSSA)

A brain-computer interface (BCI) is a system that provides a new communication path for people with disabilities to communicate with the outside world. Recently, steady-state visual evoked potential (SSVEP)-based BCI system has received more attention because it has a higher information transfer rate (ITR) and requires less user and system training. SSVEP is a periodic signal that generated by a periodic visual stimulation and its frequency spectrum contains stimulation frequency and its harmonics. The aim of the SSVEP based BCI system is to detect the stimulation frequency and thus understand the intention of the person. In recent years, various methods have been proposed for SSVEP recognition. Among these methods, the canonical correlation analysis (CCA) as a well-known powerful method and multivariate linear regression (MLR) as a state-of-the-art method can be mentioned. Although these methods find the dominant frequency in the short time EEG signal, their performance severely degraded by the presence of the noise. Therefore, the appropriate removal of the background EEG signal, that plays the role of the noise, is essential. In this paper, singular spectrum analysis (SSA) is used as a preprocessing method for separating random and periodic components of the recorded EEG signal. Also, the skewness coefficient and the Spearman correlation are used to optimize the values of SSA parameters. The experimental results show a significant improvement in performance after preprocessing by SSA using CCA and MLR methods, especially for short data lengths. Also, it has been shown that SSA with the optimal parameters (OSSA) has a higher performance than SSA with fixed parameters. The superiority of proposed method suggests it as an appropriate choice for implementation of the real-time SSVEP based BCI system.