Whole brain resting-state EEG dynamic: A mixture of linear aperiodic and nonlinear resonant stochastic processes

Parameterizing electroencephalography (EEG) signals in the spectral domain reveals physiologically relevant components of neural stochastic processes, yet the linearity or nonlinearity of these components remains debated and could not solved by the current Spectral Parameter Analysis (SPA). We address this using BiSCA (BiSpectral EEG Component Analysis), a likelihood-based model unifies EEG spectrum and bispectrum analysis to identify inter-frequency harmonic relationships and distinguish signal components. Simulations demonstrate BiSCA's ability to separate nonlinearity from non-Gaussianity (e.g., linear non-Gaussian systems exhibit diffuse bicoherence, while nonlinear Gaussian systems show localized peaks). Analyzing 1,771 intracranial EEG (iEEG) channels and a large scalp EEG dataset, we uncover a clear organizational principle: the brain's aperiodic (ξ) activity is predominantly linear, whereas its resonant (ρ) oscillations, including Alpha (α) rhythms and other peaks, are the primary source of cortical quadratic nonlinearity. This finding challenges the long-held notion of widespread linearity in large-scale brain signals, as our analysis reveals that over two-thirds of EEG and iEEG channels exhibit significant nonlinear characteristics. Spatially, we uncover a striking dissociation between signal power and nonlinearity: while the occipital Alpha (α) rhythm dominates in power, the parietal Mu (μ) rhythm generates the strongest nonlinear signature. These findings highlight that nonlinearity is present across the brain, arising from resonant activity.