BCIwiki (bciwiki.com) — Researchers led by Kunming University of Science and Technology have proposed a unified theoretical framework for brain-computer interfaces (BCIs) that uses information theory and statistical decision theory to define when user intent can be decoded and what bounds limit system performance, reporting in Frontiers in Human Neuroscience on September 16, 2026 (DOI: 10.3389/fnhum.2026.1964902). The authors are Yunfa Fu, Liu Yan, Xiaogang Chen, Jiahui Pan, Fan Wang and Tianwen Li.
The paper argues that while BCI signal acquisition and decoding algorithms have advanced, the field's theoretical foundations remain fragmented, with limited characterization of decodability, performance bounds and dynamic constraints. The authors formalize a BCI as a closed-loop stochastic process with five stages: intention generation, neural encoding, signal observation, statistical decoding and feedback regulation. On that basis they build a hierarchical framework linking scientific hypotheses, neuroscientific principles, fundamental theorems and fundamental laws.
The theorems address the conditions for intention decodability, the upper bound of observable information, the lower bound of optimal decoding error and the convergence of closed-loop learning. The laws describe constraints imposed by signal-to-noise ratio, low-dimensional neural representations, irreversible loss of observational information, class-distribution separability and non-stationary co-adaptation between user and system.
As a theoretical case study, the authors parameterize the theorems and laws for a four-class steady-state visual evoked potential BCI (SSVEP-BCI) to show how the framework applies to a concrete paradigm, and translate it into design guidelines. They describe the paper as, to their knowledge, the first attempt to systematically set out a BCI theoretical framework from a unified perspective. The abstract reports no new experimental data.