In the field, the useful signal drowns under movement, blinks and mains. ASR, ICA templates and online ORICA, and the risk of over-cleaning.
In the lab, the subject is still and the environment controlled. In the field, the useful EEG signal drowns under movement artifacts, blinks, muscle activity and mains noise. No simple filter separates these sources, because they overlap the bands of interest.
Three complementary tools
ASR, or Artifact Subspace Reconstruction, learns a reference statistic on clean segments, then reconstructs online the portions where the signal leaves that subspace. ICA template projection removes stereotyped artifact components, such as blinks, identified once. ORICA updates an ICA decomposition continuously, to track artifacts that evolve during the session.
These three tools act on different time scales and complement rather than replace one another.
Ordering and calibration
The real challenge is not picking a tool but ordering and calibrating them. Too aggressive, ASR erases portions of real signal and impoverishes the measurement. Too permissive, it lets through noise that contaminates downstream analysis. The threshold depends on the quality of the calibration segment, which must be genuinely clean.
The takeaway: a poorly calibrated artifact rejection does not clean the signal, it manufactures another one.