Narrowband-Enhanced Method for Improving Frequency Recognition in SSVEP-BCIs

Steady-state visual evoked potential (SSVEP)-based brain-computer interfaces (BCI) provide a non-invasive and effective means for communication and control, which fundamentally rely on the feature of frequency information. However, filter banks in conventional spatial filter classification methods do not effectively utilize narrowband information. This study proposed a narrowband-enhanced filter bank canonical correlation analysis (NE-FBCCA) to integrate narrowband signal processing with a broadband filter bank analysis. By employing adaptive signal decomposition via multivariate fast iterative filtering (MvFIF), the specific component corresponding to the stimulus frequency can be strengthened separately. To validate the efficacy of this method, we conducted a performance evaluation using public SSVEP datasets. The results demonstrate a notable enhancement of reconstructed EEG signals in the signal-to-noise ratio (SNR) of stimulus frequency responses. Furthermore, there are significant improvements observed in classification accuracy and ITRs when compared to standard canonical correlation analysis (CCA) and filter bank CCA (FBCCA) approaches. This study provides a narrowband signal processing strategy for SSVEP responses and shows its potential to improve the performance of SSVEP-based BCI systems.