[FieldTrip] [ext] ft_freqanalysis time and freq settings

[Mathis Kaiser]

Hi Martin,

it's correct that if you want 1Hz resolution, you need to increase the
length of your cfg.t_ftimwin to 1s, at the expense of decreasing your
temporal resolution. Another way would be to increase padding of your
data, at the expense of decreasing total power.
What is often done, and makes the time-freq analysis similar to a
wavelet analysis, is using a frequency-dependent time window (e.g. 3-7
cycles of a given frequency, depending on the desired temporal resolution).
Your cfg. toi does not depend on the time window, it just specifies on
which time points the chosen window will be centered.
Most of this is explained in some detail in the tutorial (sections
Background and TFR analysis I and II).

Hope this helps,
Mathis

On 03.07.19 10:52, Martin Rosenfelder wrote:
> Dear Fieldtrip community,
> 
> I am analyzing a two-class motor imagery task using power spectra as features for training and testing in a machine-learning setting. 
> Reading through the tutorial on "Time-frequency analysis using Hanning window, multitapers and wavelets" I found out about how to set the time and frequency parameters (cfg.t_ftimwin, cfg.toi and cfg.foi). However I did not figure out how to determine a good way of setting all these parameters. 
> 
> My current configuration loooks like this:
> 
> % Motor Imagery condition
>         cfg = [];
>         cfg.output = 'pow';
>         cfg.method = 'mtmconvol';
>         cfg.taper = 'hanning';
>         cfg.foi = 7:30
>         cfg.t_ftimwin = ones(length(cfg.foi),1).*0.5;
>         cfg.channel = {'E8' 'E9' 'E17' 'E43' 'E44' 'E45' 'E51' 'E52' 'E53' 'E58' 'E59' 'E60' 'E65' 'E66' 'E79' 'E80' 'E81' 'E131' 'E132' 'E143' 'E144' 'E155' 'E164' 'E182' 'E183' 'E184' 'E185' 'E186' 'E195' 'E196' 'E197' 'E198'} 
>         cfg.toi = 1.0:0.25:4.0;
>         cfg.trials = find(data_ref.trialinfo(:,4) == 1);
>         cfg.keeptrials = 'yes';
>         cfg.pad = 'nextpow2';
>         tfrswim = ft_freqanalysis(cfg,data_ref);
>         
> % Rest condition
>         cfg = [];
>         cfg.output = 'pow';
>         cfg.method = 'mtmconvol';
>         cfg.taper = 'hanning';
>         cfg.foi = 7:30
>         cfg.t_ftimwin = ones(length(cfg.foi),1).*0.5;
>         cfg.channel = {'E8' 'E9' 'E17' 'E43' 'E44' 'E45' 'E51' 'E52' 'E53' 'E58' 'E59' 'E60' 'E65' 'E66' 'E79' 'E80' 'E81' 'E131' 'E132' 'E143' 'E144' 'E155' 'E164' 'E182' 'E183' 'E184' 'E185' 'E186' 'E195' 'E196' 'E197' 'E198'} 
>         cfg.toi = 1.0:0.25:4.0;
>         cfg.trials = find(data_ref.trialinfo(:,4) == 2);
>         cfg.keeptrials = 'yes';
>         cfg.pad = 'nextpow2';       
>         tfrrest = ft_freqanalysis(cfg,data_ref);
> 
> So far I selected a time window of 0.5 seconds, which results in a power calculation at 2Hz, 4Hz, 6Hz etc. Presumably, this does not fit the selected frequency bands between 7 and 30 Hz. 
> 
> My main concerns are:
> Would another time window in cfg.t_ftimwin (e.g. 1 second) fit better here? 
> In which relation is cfg.toi to the setting in cfg.t_ftimwin and what steps would match the 1 second in cfg.t_ftimwin?
> 
> I hope that the description was adequately understandable and would like to thank you in advance for any kind of comment.
> 
> Kind regards,
> Martin 
> 
_______________________________________________
fieldtrip mailing list
https://mailman.science.ru.nl/mailman/listinfo/fieldtrip
https://doi.org/10.1371/journal.pcbi.1002202