Figure 7. Method 4 applied to simulated EEG data.
Bursts of 75 Hz gamma in one dipole were locked to theta troughs in a different dipole (A). Activity in these two dipoles, along with a ‘distractor’ gamma signal at 50 Hz and correlated 1/f noise at all other dipoles, was projected to scalp EEG channels. (B) shows the dipole projections and their reconstructions from gedCFC on delay-embedded matrices. (C) shows the trough-locked average (TLA) over time for all electrodes (electrodes are sorted according to spatial location, with anterior electrodes on top and posterior electrodes on the bottom) on the left and the spatiotemporal component (STC) extracted via gedCFC on the right. (D) shows the time series of the filter kernel from electrode POz (left), the filter kernel power spectrum (middle), and the power spectrum of POz activity (right; note the absence of a pronounced peak at 75 Hz). Even with only minimal noise, the non-phase-locked nature of the gamma burst prevented the TLA from revealing any meaningful relationship (phase-locking does not affect the single-trial covariance matrices). Amplitudes are normalized to facilitate direct comparisons.