Wavelet vs. Multitaper

Nathan Weisz nathanweisz at MAC.COM
Mon May 3 21:06:24 CEST 2010 

hi roni,

> 1. How come when I use multitaper with 1 cycle I see a range starting at 5 hz, but when I use morelet wavelet with 1 cycle, the range starts at 10hz?

i'm not sure. what do you mean with "range": the lowest frequency where you get a spectral estimation? just looking at your plots the difference doesn't seem dramatic: the multitaper seems to start somewhere at 6-7, the wavelet at 5. you resolution is set to 1 Hz.

> 2. Can I conclude that when the trial is rather short multitaper will be more accurate, or more sensitive to changes?

without being an expert: with the multitaper you have more possibilities to do "fine-tuning" (i.e. spectral and temporal resolution can be specified separately). for the have a spectral smoothing of +/- foi*.4. for your wavelet it's +/- foi/1 (your width which for wavelets fixes the temporal AND spectral resolution). this means your frequency resolution in case of the wavelet is *terrible*!!

you should take a look here:
http://fieldtrip.fcdonders.nl/tutorial/timefrequencyanalysis

in general it seems that for doing "good" time-frequency analysis, your trial lengths are far too short, particularly if you are interested in lower frequency activity. you should be more generous when chopping your data into trials (i usually do +/- 2s or so, do TF-analysis and then home-in into interpretable periods). you can see that in your case you hardly have values in the baseline (particularly <20 Hz).

when you do that (i.e. longer epochs) and use sensible m-factors (~5-7; or just use the fieldtrip default), then you will definitely get a completely different picture.

i'm not religious about wavelets and multitapers. with proper use both give good and usually very similar results. as said multitapers offer more possibilities for fine-tuning, but more freedom in choice is sometimes a heavy burden :-)

> 3. How come the freq range in the multitaper changes depending on the baseline settings? It started at 5hz when cfg.baseline = [-0.2 -0.05], but started at 10hz when cfg.baseline = [-0.2 -0.1];

again: take a look at the pictures. you basically have NaN's everywhere between -.2 and -.1. longer epochs are your friend.

good luck,
nathan

>
> Thanks a lot!
> Roni
>
>
>
> On 3 May 2010 18:36, Roemer van der Meij <r.vandermeij at donders.ru.nl> wrote:
> Hi Roni,
>
> I haven't looked in detail at all the other settings you use, but you apparently use a different number of cycles for the wavelets in both methods. For the multitapers, you use 2 cycles per wavelet (which are then convolved with the set of dpss tapers), whereas you use 1 cycle for your Morlet wavelets (which are then convolved with a hanning taper (or something similar)). This will make the time-window for spectral estimation twice as long for the multitapers than for your Morlet wavelets. You could try setting cfg.width = 2 for 'wltconvol'.
>
> Hope it helps a bit,
>
> Roemer
>
>
>
>
> On 5/3/2010 5:08 PM, Roni Tibon wrote:
> Dear all,
>
> I've been trying to analyze the same data using both methods, but I'm
> getting very different results, depending on the method I'm using.
>
> Enclosed is a picture of the results of the analysis in 3 channels, using
> both methods. In the left column I performed a multitaper analysis for each
> channel, using the following script:
>
> cfg = [];
> cfg.output     = 'pow';
> cfg.channel    = 'A195';
> cfg.method     = 'mtmconvol';
> cfg.foi        = 1:1:50;
> cfg.t_ftimwin  = 2./cfg.foi;
> cfg.tapsmofrq  = 0.4 *cfg.foi;
> cfg.toi        = -0.2:0.05:1;
> cfg.pad        = 'maxperlen';
> f12a_a195 = ft_freqanalysis(cfg, data);
>
> In the right one, I performed wavelet analysis using:
>
> cfg = [];
> cfg.channel    = 'A195';
> cfg.method     = 'wltconvol';
> cfg.output     = 'pow';
> cfg.foi        = 1:1:50;
> cfg.width      =  1;
> cfg.toi          = -0.2:0.05:1;
> cfg.keeptrials = 'yes';
> f12b_a195=freqanalysis(cfg, data);
>
> As you can see, channel A106 looks about the same using both methods.
> However, for channels A176 and A195, the multitaper analysis revealed
> activation in frequencies ranging from ~15-25, which was not seen in the
> wavelet analysis.
>
> I used a different program (msi) to divide the trial in two (0-500ms and
> 500-1000ms) and compare activation in this frequency range between the 2
> halves, and indeed- there was more activation in the second half, as would
> be suggested by the multitaper.
>
> Can it be that when using wavelet analysis data is lost like this?
> Am I doing something wrong with the wavelet?
>
> Thanks,
> Roni
>
>
>
> ----------------------------------
> The aim of this list is to facilitate the discussion between users of the FieldTrip  toolbox, to share experiences and to discuss new ideas for MEG and EEG analysis. See also http://listserv.surfnet.nl/archives/fieldtrip.html and http://www.ru.nl/neuroimaging/fieldtrip.
>
>
> --
> Roemer van der Meij MSc
> Scientific Programmer&  Data-Analyst
> Donders Institute for Brain, Cognition and Behaviour
> Centre for Cognition
> P.O. Box 9104
> 6500 HE Nijmegen
> The Netherlands
> Tel: +31(0)24 3612631
> E-mail: r.vandermeij at donders.ru.nl
>
> ----------------------------------
> The aim of this list is to facilitate the discussion between users of the FieldTrip  toolbox, to share experiences and to discuss new ideas for MEG and EEG analysis. See also http://listserv.surfnet.nl/archives/fieldtrip.html and http://www.ru.nl/neuroimaging/fieldtrip.
>
> ----------------------------------
>
> The aim of this list is to facilitate the discussion between users of the FieldTrip toolbox, to share experiences and to discuss new ideas for MEG and EEG analysis.
>
> http://listserv.surfnet.nl/archives/fieldtrip.html
>
> http://www.ru.nl/fcdonders/fieldtrip/
>



--------------------------------------------
Dr. Nathan Weisz

OBOB-Lab
University of Konstanz
Department of Psychology
P.O. Box D23
78457 Konstanz
Germany

Tel: ++49 - (0)7531 - 88 45 84
Email: nathan.weisz at uni-konstanz.de
Homepage: http://www.uni-konstanz.de/obob

"Nothing shocks me. I'm a scientist." (Indiana Jones)


----------------------------------
The aim of this list is to facilitate the discussion between users of the FieldTrip  toolbox, to share experiences and to discuss new ideas for MEG and EEG analysis. See also http://listserv.surfnet.nl/archives/fieldtrip.html and http://www.ru.nl/neuroimaging/fieldtrip.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.science.ru.nl/pipermail/fieldtrip/attachments/20100503/228b9ec1/attachment-0002.html>

More information about the fieldtrip mailing list