I have made some progress since my last question, and again I am asking you for advice.

Hi, I'm still designing an unusual (in terms of signal analysis) EEG experiment, which forces me to design my own filter. I have made some progress since my last question, and again I am asking you for advice.

I prepared a document that describes what filter I chose, why this particular one, and how I made this decision. Ultimately, I would like this document to become an attachment to the scientific publication that I plan to publish after my experiment. The document is here. It does not contain the source code which is here.

I have little experience with filter design, so I'd like to ask a few questions:

  1. In general, does my approach makes sense?
  2. Does my filter looks good for the task intended?
  3. Is my approach to finding the best filter (parameter Grid Search) a good one?
  4. Is there any example of such a well-written attachment describing the filter selection that I could rely on?

I am happy to accept any suggestions on how to improve both the filter and the document. I'm worried about making some false assumptions. I really care about the highest possible quality of my (PhD) project.

I will be very grateful for any comments.

  • $\begingroup$ I recommend you to add the details of your filter directly here in the question. So people can also benifit from your question once the link is dear. And secondly not everyone wants to click on a link. $\endgroup$ Commented Feb 24, 2021 at 5:56
  • $\begingroup$ @BartekKroczek: I've read your linked question and looked at your proposed filter. I agree with the answer to the linked question that what you are trying to do won't work. If I were to try what you are up to (peak or trough at the end of a six second segment,) I'd go about it differently. I'd use a set of FIR filters with linear delay. Filter the whole recording at once with one of the filters, then remove the delay (remove "bad" samples from the start of the output.) The segment into six second lengths and do your trough/peak detection. $\endgroup$
    – JRE
    Commented Feb 24, 2021 at 8:30
  • $\begingroup$ A linear phase FIR means that the output is delayed a fixed number of samples compared to the input. For a filter of (for example) 55 the delay is 27 samples ($Delay = \frac{N-1}{2}$) Filter your data, clip the delay off the front, then you can segment your data - and the segments will be in the exact same places as if you had segmented the original signal. $\endgroup$
    – JRE
    Commented Feb 24, 2021 at 8:35
  • $\begingroup$ Another thing about linear phase FIR filters is that they won't distort the waveform. Even in your narrow band filtered signals, there will be more than one frequency present. A steep IIR filter will mess up the phase relationship between the various frequencies and possibly move your peaks and troughs. $\endgroup$
    – JRE
    Commented Feb 24, 2021 at 8:37


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.