Hi everyone,
Let's assume that we have an N x T resting-state EEG signal where N and T represent the number of channels and samples, respectively, and the signal is simply the sum of independent clean EEG and noises. How can we compute the SNR of this EEG signal?
For example, if we compute the power of the signal (at each channel separately), which is the sum of power of clean EEG and noises, but in order to compute SNR, we should be able to compute either the power of only clean EEG or that of only noises. If the EEG and noises exist for the whole period of recording simultaneously, then is there any way to compute the SNR in this scenario?
Or practically can we often find time interval(s), which contain either only clean EEG or only noises? How do people usually compute the SNR or quantify the quality of their EEG recordings?
Thanks in advance!
Hi SangKim,
In principle finding the exact SNR for your recordings requires you to fully separate the "clean" signal, from all the noise sources. In Simulation studies you can quite easily define a SNR if you have a known clean EEG. Perfectly clean data or only noise data for a measured EEG is in my opinion not possible.
In MEG you can do some estimation with for example the empty room recordings which gives you sensor noise, but not physiological noise.
How it is usually computed is a good question which I do not have an answer though, maybe someone else can chime in.
Kind regards,
Steven
Unless you are working with simulations, you never know in resting-state EEG what is "interesting" brain signal, physiological noise, or instrumental noise. Therefore you can't compute the SNR of real EEG signals.
If you Google for "estimate SNR EEG", you could find some definitions of the SNR which compare the power during the peak of an ERP vs the power of the baseline. If this makes sense to you, you could follow this literature.