BCIwiki (bciwiki.com) — A 64-target code-modulated visual evoked potential (c-VEP) brain-computer interface (BCI) using 5-color relative-color stimuli reached a mean classification accuracy of 93.72% ± 3.31% across 10 participants with a hybrid TDCA-MLP decoder, researchers at the Tianjin Information Sensing and Intelligent Control Key Laboratory, Tianjin University of Technology and Education, reported in Journal of Engineering Research and Reports on September 26, 2026. The authors are Xinting Yu and Yan Bian.
Conventional c-VEP systems typically flash high-contrast black-and-white stimuli, which can cause visual fatigue, while time-shift encoding based on a single pseudorandom sequence limits the number of targets that can be recognized, the paper said. The team applied a grouped modulation strategy to a quinary relative-color design using red, green, blue, yellow and grey, which the authors describe as a first, aiming to improve multi-target recognition and reduce calibration requirements.
The decoder uses task-discriminant component analysis (TDCA) for spatial filtering and dimensionality reduction, then a lightweight multilayer perceptron (MLP) to model nonlinear features for 64-target classification. The experiment ran from April to June 2026 at the laboratory.
Mean accuracies were 85.72% ± 8.52% for CCA, 87.56% ± 7.00% for FBCCA, 85.78% ± 8.55% for TRCA, 91.27% ± 5.06% for TDCA and 93.72% ± 3.31% for TDCA-MLP. A Friedman test found a significant overall difference among the 5 algorithms (P < .001, Kendall's W = 0.601). Holm-corrected two-sided Wilcoxon signed-rank tests showed TDCA-MLP significantly outperformed CCA and FBCCA (adjusted P = .020) and TRCA and TDCA (adjusted P = .031). The authors concluded the system achieves high multi-target performance with reduced calibration and a favorable subjective visual experience.