Event Abstract Back to Event High-frequency SSVEP responses parametrized by multichannel matching pursuit Piotr Durka1*, Rafal Kus1, Jaroslaw Zygierewicz1, Piotr Milanowski1 and Gary Garcia2 1 University of Warsaw, Poland 2 Philips Research Europe, Netherlands High-frequency stimuli (above 30Hz) are crucial for implementing steady-state-visual-evoked-potential (SSVEP) based brain-computer interfaces (BCIs). Indeed SSVEPs induced by high-frequency stimuli significantly diminish user's fatigue and risk of photosensitive epileptic seizures. However, SSVEPs in the high-frequency range are up to two orders of magnitude weaker than that in the low frequency range. This poses a considerable challenge to the sensitivity of applied signal processing methods. This problem has been addressed by using an optimal linear combination of the electroencephalogram (EEG) channels to maximize the contribution of SSVEP components while minimizing the background EEG and noise [1]. In this abstract we present a novel approach based on multichannel matching pursuit [2], which provides a joint multivariate decomposition of all the EEG channels. In this study we scan all the EEG channels for matching Gabor oscillations, with potentially varying amplitudes and phases. A topographic representation of the response, taken as the amplitude of Gabor atoms fitted between 38.5 and 39.5 Hz, is depicted in Figure 1 (c). It corresponds to the distribution of response's energy across electrodes. In Figure 1 (a), the time-frequency energy density, computed for electrode Oz is represented. This high-resolution estimate, apart from the clear frequency-following response, reveals also two alpha bands with different desynchronization patterns in relation to the stimulus onset. Figure 1 (b) confirms the statistical significance of the observed changes, displaying only the area in the stimulation period where the event-related desynchronization or synchronization in relation to the pre-stimulus period was statistically significant with the false discovery rate set at 5% [3].These preliminary results indicate the potential of high-resolution adaptive time-frequency estimates in the SSVEP research targeted at enhancing BCI operation. While the average reports are primarily of exploratory and research value, we are working towards single-trial detection schemes. Complete software for multivariate matching pursuit can be downloaded via http://bci.fuw.edu.pl/wiki/Matching_pursuit. Acknowledgement:Part of the work presented here has been sponsored by EU-project BRAIN (ICT-2007-224156)