[Eeglablist] triangular / sawtooth / zig-zag pattern in spectral data

Tue May 13 12:32:22 PDT 2014

Hi Matt,

> Andreas, can you say more about why Biosemi recordings with these breaks should not be filtered, or whether approaches such as the one I described above might be acceptable? Is there a better way to correct the breaks? Thanks so much for any input.

The DC potential may change substantially already during short breaks in the recording. The paused recording now contains a DC offset (a step in the signal). Any filtering across those DC offsets/steps will necessarily result in major filter artifacts. Sharp/long filters as commonly used for high-pass filtering may easily contaminate several tens of seconds of data with filter artifacts. Demeaning the complete recording doesn’t work either as the data segments will still show (possibly large) DC offsets due to the step(s) in DC potential.

Demeaning or filtering should be done per recorded segment. The filter implementation should automatically check for boundary markers. The BioSemi recording software marks discontinuities in the recording. You should either import with a routine also importing the discontinuity/boundaries markers (e.g., https://raw.githubusercontent.com/widmann/erptools/master/pop_bdfread.m; I assume also BioSig) or do not pause recordings but rather always save separate files.

Best,
Andreas

Am 13.05.2014 um 20:01 schrieb Matt Euler <matt.euler at psych.utah.edu>:

> Hi All,
> 
> I've been encountering this same issue in my Biosemi data. I am planning to analyze the continuous data files using detrended fluctuation analysis, and have been worried about these non-zero shifts, and the saw-tooth artifacts. My approach has been to high-pass filter using the firws function at 0.5 Hz to correct the saw-tooth appearance of the spectra. This seems to be working for my purposes, since the next step for DFA is narrow-band filtering (e.g. 8-13 Hz), and the high-pass appears to remove the saw-tooth artifact in the bandpass of interest. 
> 
> Andreas, can you say more about why Biosemi recordings with these breaks should not be filtered, or whether approaches such as the one I described above might be acceptable? Is there a better way to correct the breaks? Thanks so much for any input.
> 
> Best,
> 
> Matt
> 
> 
> Matthew J. Euler, Ph.D.
> Assistant Professor 
> Department of Psychology
> University of Utah
> Salt Lake City, UT 84112
> 
> -----Original Message-----
> From: Andreas Widmann [mailto:widmann at uni-leipzig.de] 
> Sent: Tuesday, May 13, 2014 3:01 AM
> To: Bachman, Peter
> Cc: eeglablist at sccn.ucsd.edu
> Subject: Re: [Eeglablist] triangular / sawtooth / zig-zag pattern in spectral data
> 
> Hi Peter,
> 
>> I think sometimes demeaning based on the full continuous data might not work if there's an abrupt shift in offset somewhere in the data - the abrupt shift means the data either side of the shift isn't centred on zero as it should be after subtraction of the mean.
> How do you import your BioSemi data? The bdfimport plugin does not import discontinuity/boundary triggers or CMS out of range triggers (https://sccn.ucsd.edu/bugzilla/show_bug.cgi?id=1373). In case you have breaks and/or CMS out of range epochs in your recordings the effect well described by Matt will occur. Demeaning will not help. BioSemi recordings including breaks and imported with the bdfimport plugin must not be filtered.
> 
> Best,
> Andreas
> 
> Am 12.05.2014 um 20:31 schrieb Matt Craddock <matt.craddock at uni-leipzig.de>:
> 
>> On 12/05/2014 18:53, Bachman, Peter wrote:
>>> Hi everyone,
>>> 
>>> 
>>> When I plot spectral data from continuous EEG, the resulting 
>>> Power-by-Frequency graph looks very strange.  Specifically, starting 
>>> at about 5 Hz and continuing on to higher frequencies, the plots for 
>>> each channel have a repeating triangular, or sawtooth-like, or zig-zag shape.
>>> 
>>> 
>>> Here is a Dropbox link to an image of representative data:
>>> https://www.dropbox.com/s/lddxgzlti9m20vj/FFT_triangle_pattern.JPG
>>> 
>>> 
>>> It looks like some kind of wacky filter has been applied, but the 
>>> only filter in the processing pipeline is a very conventional 
>>> 0.16-100 Hz bandpass.
>>> 
>>> 
>>> The problem isn't a feature of the raw data itself, because opening 
>>> the same continuous data in different EEG analysis programs and 
>>> applying a FFT produces very normal looking spectra.  This happens 
>>> across different versions of EEGLAB, Matlab, and Windows.
>>> 
>>> 
>>> The continuous data were recorded using a Biosemi system and saved in 
>>> .bdf format.
>>> 
>>> 
>>> Has anyone run into this before?  I'd be very grateful for any suggestions.
>>> 
>>> 
>>> Thanks!
>>> 
>>> Pete
>>> 
>>> 
>> 
>> Hi Pete,
>> 
>> This is typical with Biosemi. Here's an old post from Bradley Voytek about this very issue:
>> 
>> http://sccn.ucsd.edu/pipermail/eeglablist/2008/002229.html
>> 
>> It's probably some sort of rounding error somewhere down the line. He suggests removing the mean from each channel during import. I tried that and it didn't solve the issue; I think sometimes demeaning based on the full continuous data might not work if there's an abrupt shift in offset somewhere in the data - the abrupt shift means the data either side of the shift isn't centred on zero as it should be after subtraction of the mean. Anyway, if you're planning on epoching the data, baseline correction will also solve the problem, in my experience.
>> 
>> Cheers,
>> Matt
>> 
>> --
>> Dr. Matt Craddock
>> Research Fellow
>> Institute of Psychological Sciences
>> University of Leeds
>> 
>> _______________________________________________
>> Eeglablist page: http://sccn.ucsd.edu/eeglab/eeglabmail.html
>> To unsubscribe, send an empty email to 
>> eeglablist-unsubscribe at sccn.ucsd.edu
>> For digest mode, send an email with the subject "set digest mime" to eeglablist-request at sccn.ucsd.edu
> 
> 