Hi
Yes, they should match very close, I suspect you have forgotten to remove the DC component. If you do
f=abs(fft(t)).^2;
f(1)=[]
and then correlate things should work fine...
hth
Christian
On 23 Feb 2013, at 01:40, SURIL GOHEL <[log in to unmask]> wrote:
> Dear Experts,
>
> I have a question about melodic output time series and the power spectrum associated with it.
>
> I am trying to run a group ICA on 30 subjects resting state data with 460 time points each.
>
> In the previous melodic version,
>
> the plot of ICA time series had 460 time points and an power spectrum associated with it.
> I could access the time series by looking at stats/t*txt and power spectrum by looking at stats/f*txt.
>
>
> In the latest version of melodic.
> the plot of ICA time series is a concated version of time series across subjects with 13800 (460 * 30 ) time points.
> and there is also a power spectrum associated with it.
>
> The question is ,
> when i load the ICA time series in to matlab and calculate the power spectrum (abs(fft(time-series))).^2,
> there are between the power spectrum obtained from fsl melodic and the one i obtained from matlab (correlation =0.5)
> shoouldn't it match very closely?
>
> Also, if i want to show the time series associated with a component at group level and related power spectrum,
> which melodic output should i use ? the current version or earlier version?
>
> Thanks for any help on the subject.
>
>
>
>
>
>
|