A High-speed 120-target SSVEP-BCI Employing Dual-Frequency and Phase Modulation with Minimal Calibration

Large instruction-set brain-computer interfaces (BCIs) allow users to issue many commands through a single interface, greatly expanding their application scope. Increasing the number of targets, however, raises encoding complexity and intensifies the trade-off between calibration time and decoding performance. We introduce a 120-target steady-state visual evoked potential (SSVEP)-BCI that pairs dual-frequency phase modulation (DFPM) with a lightweight global multi stimulus canonical correlation analysis-based spatiotemporal filtering (gmsCCA-st) method. DFPM encodes the 120 targets with only 23 low-frequency carriers by simultaneously flickering two frequency-phase tags in a checkerboard pattern, thereby mitigating the "frequency scarcity" problem and eliciting pronounced harmonic and intermodulation responses. Instead of training a separate filter for each target, gmsCCA-st learns a set of shared spatiotemporal filters from all targets. With just one calibration trial per target in the offline experiment, the system achieved a peak information transfer rate (ITR) of 326.49±55.13 bits/min. During online cue-guided spelling, the system attained 94.69±5.99% accuracy and 251.47±25.47 bits/min ITR; in free-spelling mode, accuracy was 91.72±6.89% at 176.64±23.75 bits/min. These findings demonstrate the feasibility of a high-performance 120 target SSVEP-BCI after only three minutes of calibration, overcoming the compromise among instruction-set size, calibration burden, and performance. This study therefore offers a practical pathway toward high-performance, minimal-calibration large instruction-set BCIs.