Calibration-Efficient Dual-Frequency SSVEP-BCI for Head-Mounted AR-Based UAV Control

Steady-state visual evoked potential (SSVEP)-based brain-computer interfaces (BCIs) integrated with head-mounted augmented reality (AR) enable wearable, intuitive interaction, yet two fundamental issues limit practical adoption: (i) the role of different dual-frequency stimulation paradigms in head-mounted optical see-through displays is not systematically understood, and (ii) calibration-efficient decoding under multi-target settings remains challenging. This study presents a 16-target AR-SSVEP-BCI on HoloLens 2 and conducts a controlled comparison between a conventional single-frequency paradigm and three dual-frequency binocular paradigms derived from joint frequency-phase modulation. To address calibration burden, we propose a calibration-efficient encoding-decoding framework that leverages a row-column encoding strategy and a row-column decoding strategy with a task-related component analysis (TRCA)-based ensemble spatial filtering scheme, enabling reuse of shared frequency-phase components across targets. In offline evaluations, the best-performing right-and-left field dual-frequency and phase modulation paradigm achieved an average information transfer rate of 99.79 ± 18.96 bits/min with only five calibration blocks. Building on this paradigm, we developed an online AR-SSVEP-BCI with a training-free dynamic stopping strategy and a control-state detection module for asynchronous decision making. In online tasks, under the optimal world-referenced mode, the system reached 89.32 ± 8.43% accuracy in a 16-target Random Cue Task and 95.13 ± 4.39% accuracy in an 8-command Unmanned Aerial Vehicle (UAV) Control Task, demonstrating robust, calibration-efficient multi-command control. These findings provide practical guidance for designing calibration-efficient, wearable AR-SSVEP BCIs that can support accessible assistive control-an important step toward real-world applications.