Discrimination of steady-state visual evoked potentials' frequency and phase using multiset canonical correlation analysis

Abstract

Responses to a periodic visual stimuli called steady-state visual evoked potentials (SSVEPs) play important roles in applications of brain--computer interfaces (BCIs) as well as understanding of visual information processing mechanism. The measurement of the responses to the periodic stimuli from electroencephalography (EEG) involves accurately estimating its frequency and phase. This paper introduces a novel approach to recognition of the frequency and the phase of SSVEPs that can achieve high performance in BCI applications. The proposed method exploits multiset canonical correlation analysis (Mset CCA) to enhance SSVEP components in EEG recordings. An experiment of the simultaneous recognition of the frequency and phase was conducted to compare performance of the proposed method with the previous work. The proposed method achieved the ITR of 126.74 bits/min when analyzing a very short time window of 500 ms in the EEG signal, which was significantly higher than the state-of-the-art.