Band pass filter: edge effect

Discussions
1

Hi,

I am not interested in delta frequency band for calculating PSD but interested in theta PSD (4-7 Hz) and above.

Is it OK to filter raw data from 3-30 Hz (just 1 bellow 4 Hz) or there can be edge effect affecting theta PSD?
I am removing slow frequencies to exclude noise.

Thanks

2

Hi Niko,

That would depend on the length of your data. When selecting the bandpass from 3-30 Hz in Brainstorm you will see that the transient effects of the filtering are 3.6 seconds for the full transient. The frequency content from 4-7 Hz will not be affected by the filtering in terms of amplitude.

When calculating PSD you could also set the option to already group it in the different frequency bands with the edit option. In that case you could discard the results for the other frequency bands.

Kind regards,
Steven

3

Thank you Beumer

Recording is 3 minutes long, continuous recording.
Is it ok, let us say to apply filter at 3-100Hz and then to have a look at edge frequency bands (e.g. 4-7Hz, 65-95Hz). Would they be affected?

Do you recommend to exclude some portions of recording from PSD calculation following the band pass filtering?
Also, for PSD calculation should be done on continuous data, right?

Thanks

4

I am not an expert in PSD calculations, but yes PSD calculations should be done on continuous data. Due to filter transients it would be better to cut off the first few seconds and the last few.

A good overview on PSD and filtering might be the following tutorial, especially the advanced section: https://neuroimage.usc.edu/brainstorm/Tutorials/ArtifactsFilter

Have a nice weekend!

5

Thank you Beumer!

6

If your measure of interest is the power in the frequency band 4-7Hz, and not ERPs: do not filter your recordings before. The noise at 3Hz or below in the original recordings would not affect the power estimated at 4Hz, the FFT used in the PSD estimation is a frequency filter.

Is it ok, let us say to apply filter at 3-100Hz and then to have a look at edge frequency bands (e.g. 4-7Hz, 65-95Hz). Would they be affected?

Look at the frequency response of the filter you designed in the band-pass filter process: it would show you the frequencies that are affected by the filter.
As @SBeumer mentioned, you should exclude the edges of the recordings from the computation of the PSD if you filtered your recordings, if possible of the duration of the entire filter. Another reason for not filtering your recordings before computing the power.

In general, minimize as much as possible the preprocessing of the recordings, use only the filters/ICA/SSP cleaning you really need for computing the measures you are interested in.

7

Thank you Francois.

Actually, I filtered the EEG before ICA to get less noisy ICs.