Periodicity Transforms For Multichannel and Multiclass Detection of Visual Evoked Potentials

A repetitive visual stimulus induces a brain response known as the Steady State Visual Evoked Potential (SSVEP) whose frequency matches that of the stimulus. Reliable SSVEP-based Brain-Computer-Interfacing (BCI) is premised in part on the ability to efficiently detect and classify the true underlying frequencies in real time. We pose the problem of detecting different frequencies corresponding to different stimuli as a composite multi-hypothesis test, where measurements from multiple electrodes are assumed to admit a sparse representation in a Ramanujan Periodicity Transform (RPT) dictionary. We develop an RPT detector based on a generalized likelihood ratio test of the underlying periodicity that accounts for the spatial correlation between the electrodes. Unlike the existing supervised methods which are highly data-dependent, the RPT detector only uses data to estimate the per-subject spatial correlation. The RPT detector is shown to yield promising results comparable to state-of-the-art methods such as standard CCA and IT CCA based on experiments with real data. Its ability to yield high accuracy with short epochs holds potential to advance real-time BCI technology.