Re: [NBThelp] artifacts in spectrogram and DFA

[Simon-Shlomo P. <poi...@gm...>]

Dear Matt,

I didn't explore this in detail, but I believe the sawtooth power spectrum
is caused by a very strong low frequency component (and perhaps an
inadequate PSD algorithm). Did you try to use the pwelch function (with
default settings)?

If the sawtooth power spectrum disappears after high-pass filtering (e.g.
using nbt_filter_firHp), and the signal looks normal, then it's probably ok
to proceed.

If you plan to use DFA as describe here (
http://www.nbtwiki.net/doku.php?id=tutorial:detrended_fluctuation_analysis_dfa#.Uz7JU8eF9Yw);
i.e., to measure long-range temporal correlations in the amplitude
modulation of different frequency bands, you need to band-pass filter your
signal and, therefore, you do not need to do a high-pass. However, if you
do a high-pass it is also not a problem, because this high-pass should not
influence the modulation of higher frequencies (to be sure you can check
the frequency response of the filter).

If you plan to use DFA on the raw signal trace, then you need to be very
careful, because a high-pass filtering could induce correlations.

Best wishes,
Simon

--
Simon-Shlomo Poil, Dr.

Mobile number: +41 (0)76 399 5809
LinkedIn profile: http://ch.linkedin.com/in/simonshlomopoil/

The Neurophysiological Biomarker Toolbox (NBT): http://www.nbtwiki.net

My latest publications:
Integrative EEG biomarkers predict progression to Alzheimer's disease at
the MCI stage: http://l.nbtwiki.net/19jMuy8

The Amsterdam Resting-State Questionnaire reveals multiple phenotypes of
resting-state cognition: http://l.nbtwiki.net/17yblK7



2014-04-03 16:57 GMT+02:00 Matt Euler <mat...@ps...>:

> Dear NBT list,
>
> First, thanks for the toolbox. It looks like it is going to be a great
> resource for the field!
>
> I'm hoping to use NBT to conduct DFA on some .bdf and ANT EEprobe .cnt
> files, but I have a question related to artifacts I'm noticing in my EEGlab
> spectral plots. The files are continuous data from a resting period,
> sampled at 1024 Hz, and about 200,000 samples in length. The spectra always
> have a regular, saw-tooth or spiking appearance in all the channels for
> both file types (even after converting the .bdf files to .cnt in ASA-lab
> software).
>
> I've noticed that downsampling the data and removing the channel mean
> appears to improve the look of the spectra, but I'm having trouble
> identifying the actual underlying issue. I'm especially concerned that the
> saw-tooth artifacts might be due to large voltage changes in my time domain
> data, potentially due to skin potentials or some other artifact, but these
> are only evident when I plot the data in matlab rather than using the
> EEGlab channel data scroll function or the NBT visualization tools.
>
> I know that similar artifacts have been discussed in regards to biosemi
> data on the EEGlab list before:
> http://sccn.ucsd.edu/pipermail/eeglablist/2008/002229.html, and
> http://darb.ketyov.com/professional/raw-spectrumZoomed.jpg, but the
> proposed fix to cast the raw data first to double and subtract the channel
> mean hasn't solved the issue in my raw .bdfs, and it's unclear to me
> whether that accounts for the same problem in the native .cnt files. Also,
> since the ultimate goal is to conduct DFA, I'm very reluctant to simply use
> a low high-pass filter or mean detrend every channel, without guidance
> regarding the implications of that for DFA.
>
> Any advice about how to correct the issue in prep for DFA would be greatly
> appreciated!
>
> Thanks,
> Matt
>
> Matthew J. Euler, Ph.D.
> Department of Psychology
> University of Utah
>
>
> ------------------------------------------------------------------------------
> _______________________________________________
> The NBT help mailing list.
>
> Visit our webpage:
> http://www.nbtwiki.net
>
> LinkedIn: http://l.nbtwiki.net/nbtlinkedin
> Facebook: http://l.nbtwiki.net/nbtfacebook
>
> Do you need help with NBT?, see
> http://l.nbtwiki.net/HRIMOF
>
> Get involved, see
> http://l.nbtwiki.net/HRf5Ns
>
> To unsubcribe write a mail to nbt...@li... 'unsubcribe' in the subject line (without ' ')
>
>
>