Good afternoon. I'm dealing with similar issues. I have an article coming soon on this topic: "Analysis of electrocardiographic signals using a new method based on the principles of wavelet processing and Gauss-Hermit functions"

If you are interested in it, I will send it for review.

You can email me dabalakin@yandex

The filtering process is summarized as follows:

output(t)=signal(t)+LFnoise(t)+HFnoise(t);

LFnoiseAvrg(t)=MovingAverage(output(t)) LFpass(output(t));

HFoutput(t)=output(t)-LFnoiseAvrg(t) not eqv LFpass(output(t));

Filtered_output(t)=LFpass(HFoutput(t)).

I think that these papers will be of some use to you ... with my best wishes for you

Digital Filter Performance Based on Squared Error

https://ieeexplore.ieee.org/document/9626820

Conference Paper Digital Filter Performance Based on Squared Error

Convergence Rate For Low-Pass Infinite Impulse Response Digital Filter

Conference Paper Convergence Rate For Low-Pass Infinite Impulse Response Digital Filter

If one uses a sensor that is hardwired to capture within a certain bandwidth say 10/20 to 100/500Hz, the signal is now restricted to that band. So, if you find a power change that is in the bands under 10/20 is likely to be noise unless one can provide clear evidence for it to be what you recorded. This is especially true in multiple articles where the argument is given that there is an increase in activity in the lower bands but no clear explanation is provided to explain where that may arise from in the data.

one can assume the interactions between motoneurones be it homonymous, heteronymous or recurrent mechanisms.

Without that information to suggest that the results seen are biologically meaningful is poor science.