Enhance Detection of SSVEPs through a Sinusoidal-Referenced Task-Related Component Analysis Method

The brain-computer interface (BCI) technology is deemed a pivotal technology in future wireless communication systems, e.g. 6G, for its capability to connect a brain and a machine. The device, paradigm, and algorithm are three most important aspects of a practical BCI. Among them, the detection algorithm has a decisive impact on the efficiency and robustness of the system. Great potential of artificial intelligence (AI) for decoding brains signals is also revealed. In this paper, we propose a new detection algorithm for the steady-state visual-evoked potential (SSVEP) based BCI, which is a typical noninvasive BCI paradigm and achieves by far the highest information transfer rate (ITR) among various noninvasive systems. The new algorithm is termed sinusoidal-referenced task-related component analysis (srTRCA). It resembles conventional algorithms like TRCA in the way that it is also based on spatial filtering and template matching. However, compared with conventional algorithms like TRCA, srTRCA makes better use of the prior knowledge of the SSVEP signal being sinusoidal. By introducing a new item which characterizes the correlation between the task-related component and sinusoidal reference to its objective function, srTRCA is expected to achieve an enhanced detection performance, especially in the situation where training is insufficient. The performance of srTRCA is tested on a benchmark SSVEP dataset which includes 35 subjects. Three algorithms are taken as baselines with them being canonical correlation analysis (CCA), TRCA, and similarity-constrained TRCA (scTRCA). Results show that srTRCA achieves a fair performance enhancement compared with three baselines. The validity of the proposed srTRCA algorithm is proved.