For each sensor, artifact-free epochs were transformed into the time–frequency domain using complex demodulation55,56. This involves filtering the complex signal into frequency bands of a predetermined width and range, and calculating the power within each band across each successive temporal window. The resulting spectral power estimations per sensor were then averaged across trials and normalized by calculating the percent change in power relative to the baseline time period (− 400 to 0 ms) per frequency bin. Time–frequency windows of interest were then identified using paired-samples t-tests against baseline on each data point in the sensor-level spectrograms across all participants’ gradiometers (p < 0.05), and then correcting for multiple comparisons using nonparametric cluster-based permutation testing57,58.
