An Optimized Encoding BCI Framework: Implementing Massive Command With Minimal Calibration

As a critical metric for brain-computer interfaces (BCIs), the number of commands directly defines the control capacity for practical applications. However, existing BCIs often suffer from limited command sets and prohibitive calibration costs. To address these problems, this study presents a functional optimization-based encoding framework to generate massive commands with high discriminability while minimizing calibration burden. Specifically, a functional optimization theory enhances command distinguishability by optimizing the encoding function, while a few-shot training strategy leverages symbol reusability to reduce calibration data. Additionally, a symbol-joint decoding approach improves recognition accuracy. Using this framework, we developed an online BCI system with an unprecedented 1,008 commands-establishing a dual state-of-the-art (SOTA) in both command scale and calibration efficiency for large-scale BCIs (> 100 commands). Comparative analysis shows that the functional optimization strategy improved accuracy by 13.94% and the information transfer rate (ITR) by 26.12% over the widely adopted baseline. Remarkably, with only 72 seconds of calibration data, the system achieved an average accuracy of $86.60~\pm ~13.35$ % and an average ITR of $122.74~\pm ~24.64$ bits/min across 15 subjects, peaking at 100%. The framework features high flexibility in command encoding and robust cross-paradigm compatibility, significantly enhancing BCI performance and practicality.