Thank you very much, I will be talking to the device vendor.
DIUC that EDF does not specify the units? I.e., they might be writing the values in V or mV or uV?
It does depend. You have to look at your data to figure out this kind of threshold, not just put a magic number in. It depends on the EEG amplifier you are using, if you are using dry, wet, or gel-based caps, the environment, and the preprocessing you applied to the raw data.
The point is, you have to look at your data after preprocessing, figure out what are the artifacts/issues you want to capture with flat and reject and select a threshold accordingly.
There is even automatic methods to compute the rejection threshold that have been developed. See for instance autoreject: autoreject — autoreject 0.3.1 documentation
THIS. SO. MUCH.
Visually inspect the data after each (pre)rocessing step and adjust your procedure / pipeline as needed. Otherwise you’ll never know what was actually going on.
what preprocessing do you recommend?
I believe @mscheltienne meant you have to inspect the data to determine sensible rejection thresholds. This would ideally involve visual inspection. You need to get a “feel” for the data.