Enhancing Individual Calibration Classification in SSVER-Based BCI with Exactly Periodic Component Analysis

Brain-computer interfaces (BCIs) based on steady-state visual evoked responses (SSVERs) offer high information transfer rates and signal-to-noise ratios, making them widely used in applications, such as brain-controlled typing and wheelchair navigation. However, traditional SSVER decoding using spatial filtering methods requires substantial training data, limiting practical usability due to the time-consuming calibration process. To address this issue, we propose an exactly periodic component analysis (EPCA) method, which extracts exactly periodic signals using multiple orthogonal subspace projections and constructs spatial filters via generalized eigendecomposition. EPCA eliminates the need for multiple training trials while maintaining high decoding performance. We validated the effectiveness of EPCA on four publicly available datasets and one self-collected BCI dataset. Results demonstrate that EPCA significantly outperforms conventional spatial filtering methods, recent single-trial calibration approaches, and deep learning-based decoding models. Across all five datasets, EPCA consistently achieved the highest decoding performance under single-trial calibration. Furthermore, an online spelling experiment confirmed the real-world effectiveness of the proposed method. By reducing calibration requirements without compromising performance, EPCA greatly enhances the practicality of SSVER-based BCIs for real-world applications.