I have had a brilliant experience using mne so far, so thank you for this invaluable software and support. This is more a data interpretation question. The current dataset I am working with has two features that I am finding it tricky to understand, and they may relate. I’d really appreciate any advice, insight of others into these:
No alpha peak in the data, both before and after processing as shown by a psd visualisation (image attached)
A really high amount of noise in the data before processing. I am curious to know what likely causes would be, and if this data is usable. (images of data before and after filtering)
Thanks for your response! Sorry I should have added more detail to my original post. The last image I attached is after filtering between 1 and 45 hz. It seems to get rid of the noise, but I guess I am hesitant that with such high noise to begin with that the filtering would end up distorting the signals. Can line noise show up in the data even after notch filtering for 60hz and harmonics? I experimented with just applying a notch filter (without 1-40hz filtering) and there was still a lot of noise in the data.
Line noise is typically restricted to that specific frequency, so if you look at e.g. frequencies from 1–40Hz, this should be OK. Or do you need frequencies greater than that (i.e., above 60Hz)?
I recommend that you look at the PSD after filtering. This should also explain why notch filtering didn’t really work (I suspect there will still be a large peak at 60Hz).
Thats really great to know, thanks @cbrnr! I don’t need frequencies above 45 Hz, so very happy to cut out everything above this. How do we know in this case that is its line noise as oppose to other EMI interfering with the data, such as a telephone plugged in for example?
I will experiment with looking at the psd after notch filtering to see what you mean! Really helpful, thanks
If you don’t need those higher frequencies, use a low-pass filter instead of a notch. You can achieve much better attenuation with a suitably wide transition band as opposed to a very narrow notch filter.
A sharp peak at 50/60Hz is power line interference, that’s how electrical power is transmitted over wires that are all around us. This can include a power brick (with a phone plugged it), although these devices can generate interferences at different frequencies as well. But given your PSD, that’s power line interference for sure.
Ah that is really useful to know about the low-pass filter being preferable for notch.
In previous datasets I’ve worked with I’ll have the 60Hz power line noise peak in my PSD, but will not have this level of noise showing up in the raw data plots. This is what made me worried about there being some other broadband noise present in the data across all frequencies, without a PSD peak. So essentially, line noise plus other environmental noise. In this case, do you think then the explanation is more intense/larger power line noise than in previous datasets?
The other reason I thought there might be other noise, is the lack of alpha peak - as its resting state data, the explanation that came to mind was the brain signal being lost or distorted due to other noise
Let’s look at the PSD of the lowpass-filtered data before trying to interpret too much into it. The spectrum does look too flat and not 1/f-like at all, but that’s maybe because it’s hard to see in the PSD with the large 60Hz spike.
As you already mentioned, this doesn’t look like nice EEG at all. The time series plot you showed looks OK-ish though (hard to say), so yes, I don’t know either what might have caused high power across all frequencies. It might still be usable (as in there’s information in the signal), but this depends on what you want to do.
its confusing when the time-series looks okay whilst at the same time something is off about the data. The analyses include mostly: connectivity analyses, signal diversity (lempel ziv complexity), and absolute broadband power. As in comparisons are made between two conditions for each subject. Interestingly for this subject, there is no alpha peak in both conditions despite the conditions being about 2 weeks apart.
Maybe you can find out more about how the data was recorded? Are there more datasets from this study? Do they look similar? Which EEG amp was used? Etc.
Luckily I have lots of information about the recording - there wasn’t anything plugged in other than the EEG amplifier (which was an ant-neuro hub). Also interestingly, when recording the data (on eego software), apparently the signal looked completely fine, they didn’t see this level of noise in it. I wasn’t sure what to make of that, for example if its a scaling difference or something else between eego and mne+eeglab. Other factors were that ECG and breathing were being gathered at the same time, and ECG was very noisy; and two ppts were measured simultaneously (i.e. two 64 channel caps with one 128 channel amplifier)
OK, so it’s not related to MNE-Python then. I have no idea what the problem could be – maybe someone else has seen such signals before? Again, if it’s just one dataset I’d not investigate too much, only if all of your datasets look like this – do they?
