[FieldTrip] Question pertaining to frequency resolution for time-frequency analysis

[Eelke Spaak]

Dear Chiel,

I think you understood frequency resolution very well. The only
"trick" you can apply is some form of padding (most typically
zero-padding using cfg.pad in ft_freqanalysis). That will increase the
number of frequencies you get out (e.g. cfg.pad = 1 [seconds] will
allow you to get 1 Hz frequency spacing), but note that this is
"artificial", in a sense, and the trade-off between time and frequency
resolution is a fundamental property of nature (and Fourier analysis).
However padding is still commonly applied, and can be very useful
nonetheless.

Best,
Eelke


On Wed, 12 May 2021 at 12:24, Wijs, C. <c.wijs at student.rug.nl> wrote:
>
> Dear all,
>
> For a uni research project I am trying to perform a TFR analysis on segmented data. This was not the main aim of my analysis (which are ERP amplitudes), but me and my supervisor decided that I would have an (exploratory) look at it. The issue I am facing is that the trials I have defined (and preprocessed/cleaned) are rather short, ranging from -0.1s prior to a cue, to 0.6s after a cue, which leaves me with little wiggle room.
>
> I've decided to follow the method used in section "Time-frequency analysis I" from the tutorial at : https://www.fieldtriptoolbox.org/tutorial/timefrequencyanalysis/#time-frequency-analysis-i . At the end of this section a "warning" is given that one should think critically about the used frequency resolution. When using a 2Hz frequency resolution, corresponding to a 500ms time window for the taper,  one cannot, according to the message, interpret any odd-numbered frequencies. I do get values for the odd numbered frequencies, but these are meaningless/incorrect/invalid I assume?
>
> What I would like to do is have a 1Hz frequency resolution, such that I get (meaningful) values for each integer frequency (within a range of course). If I understand the type of analysis correctly, this would correspond to a 1s time window for the taper, making the first and last 500ms of my trial not be included in the output of the analysis. The problem of course, is that my trials last only 700 ms, from 100ms prior till 600ms post cue, so I would be left with nothing.
>
> Am I just out of luck, and will I have to do the analysis with a minimal 2Hz frequency resolution? (which already leaves me with quite a short period of TFR analysis, as this cuts about 250ms from either end of the trial).
>
> Thank you very much for your time and kind regards,
>
> Chiel Wijs
> _______________________________________________
> fieldtrip mailing list
> https://mailman.science.ru.nl/mailman/listinfo/fieldtrip
> https://urldefense.com/v3/__https://doi.org/10.1371/journal.pcbi.1002202__;!!HJOPV4FYYWzcc1jazlU!rJ_Qm31kRT1OmvvGkgIacMHNLW0o7y2qHuh2yP1pZKQWVyB6Gej1Ar78cuQ_WAUHVNc$