Statistically significant time-varying neural connectivity estimation using generalized partial directed coherence.

This paper illustrates the effectiveness of generalized partial directed coherence (gPDC) in characterizing time-varying neural connectivity by properly extrapolating its single trial asymptotic statistical results to a multi trial setting. Time-varying estimation is performed with a sliding-window procedure based on the proposal in [1], whereby a time-frequency map of the connectivity between channels is built. The technique is validated on a non-linear toy model generating simulated EEG and then applied to a publicly available real EEG dataset for benchmarking purposes.