The EMG signals were first preprocessed together with the EEG signals. They were demeaned, high-pass filtered at 0.5 Hz, cleared from visually detected EEG/EMG artifacts, and broadband pass filtered from 5 Hz to 200 Hz. Notch filters at 50 Hz and its harmonics were applied to reduce interference from environmental noise. After rectification, EMG signals were segmented into 3-s epochs (−1 s to 2 s) and aligned with the movement onset as defined in EEG signal preprocessing. EMG signals were smoothed by applying a 5th order Butterworth low pass filter at 5 Hz to the rectified signals (30). For each subject, the EMG signals were averaged across epochs. Based on the mean EMG signal of each subject, we calculated the EMG slope as the slope between the 25% and 75% percentiles of the normalized EMG amplitudes close to the movement onset (as shown in Fig. 1E). EMG signals were z-score normalized before computing the EMG slope to diminish individual variability of the EMG amplitudes.
