[FieldTrip] log transform of power values

Frederic Roux f.roux at bcbl.eu
Mon Dec 3 16:58:13 CET 2012


Dear all,

I am planning to run freq_statistics on single trial data,
with the power data computed in the following way:

cfg = [];
cfg.method = 'sincos';
etc ...
planar_data = ft_meplanar(cfg,raw_data);

cfg = [];
cfg.foi = 30:150;
cfg.output = 'pow';
cfg.taper = 'dpss';
etc ...

freq_planar = ft_freqanalysis(cfg,planar_data);

freq = ft_combineplanar([],freq_planar);


When I then look at the topographies 

cfg = [];
cfg.grad = freq.grad;
cfg.layout = ft_perpare_layout(cfg);
cfg.xlim = [.1 .6];
cfg.ylim = [40 60];
cfg.parameter = 'powspctrm';

figure;
ft_topoplotTFR(cfg,freq);

I noticed that it makes a huge difference if the data are log transformed
before the plotting, ie 

freq.powspctrm = 20*log10(freq.powspctrm);

The data looks more 'clustered' when I do the log transform, whereas the 'raw' 
power values look scattered and like tiny islands.

Now I am wondering if this could help with the statistical analysis, meaning
that it would give me more power to detect spatio-temporal clusters.
(Baseline correction is not really a good idea, as there seems to be important
stuff going on in the baseline)

Any (quick) help and advice would be highly appreciated.

Fred






Frédéric Roux
Postdoctoral Researcher
www.bcbl.eu





More information about the fieldtrip mailing list