Abstract
Mobile EEG enables the investigation of brain activity during real-world behavior but remains highly susceptible to motion artefacts, which limit signal interpretability and the applicability of advanced analytical techniques. Existing artefact removal methods face challenges in reliably separating motion-induced artefact from true neural signals, particularly in dynamic, non-stationary contexts. In this study, we propose a denoising method based on generalized eigenvalue decomposition (GED), which exploits the contrast between covariance matrices of artefactual and reference (resting-state) EEG to identify and suppress motion-related components. The method was evaluated on two ecologically valid datasets: one involving stereotyped head and body movements, and another featuring free and dynamic motion during table tennis gameplay. GED based denoising significantly outperformed state-of-the-art methods across several metrics, including signalto-error ratio (SER), artefact-to-residue ratio (ARR), and preservation of spectral power, particularly in the high-frequency range. The proposed approach did not introduce high-frequency distortions and demonstrated superior capability in suppressing large-amplitude non-stereotypical artefacts while preserving physiological EEG content. These results highlight GED’s potential as a robust and generalizable method for motion artefact suppression in mobile EEG.