Spatial Attention and Subspace Calibration for Personalized SSVEP Measurement

초록

Reliable acquisition and calibration of electroencephalography (EEG) signals are essential for accurate and consistent brain-computer interface (BCI) measurement systems. However, conventional steady-state visual evoked potential (SSVEP) frameworks often overlook variability in the EEG measurement chain, including sensor gain, electrode impedance, synchronization jitter, and calibration cost. These limitations degrade measurement stability and hinder the deployment of EEG as a dependable instrumentation modality. This article presents spatial attention-augmented TRCA (SA-TRCA), a calibration-oriented framework designed to improve the reliability and robustness of SSVEP-based EEG acquisition and analysis. The method integrates: 1) a spatial attention module that estimates channelwise relevance as sensor gain factors to compensate for electrode variability and 2) an adaptive subspace calibration (ASC) strategy that determines the effective number of task-relevant components using profile likelihood, thereby reducing dependence on fixed assumptions and improving calibration efficiency. Comprehensive validation on two public SSVEP datasets (Benchmark and BETA) demonstrates that SA-TRCA consistently enhances measurement reliability and robustness across different signal lengths, calibration trials, and channel counts. Overall, SA-TRCA provides a unified calibration framework that balances accuracy, efficiency, and robustness, advancing EEG measurement methodology for reliable and scalable SSVEP-based instrumentation.

키워드