Dear Anette,
On 10 Jun 2010, at 09:42, Anette Giani <[log in to unmask]> wrote:
> Dear all (and in particular dear Vladimir),
>
>
>
> During your lecture on preprocessing (MEEG course 2010; (methods of spectral estimation , example 2)) you mentioned that high frequency 'stripes' may be seen during time-frequency analysis. If I understood you correctly, this might indicate that the chosen (Morlet) wavelet, is not appropriate for your data. Alternatively, one may do an Hilbert transform, use fixed windows or optimize the Morlet wavelet.
>
>
>
> I was wondering how to do this optimization in SPM? I have 10 seconds and 1 seconds trials and am specifically interested in phase/power at 6 & 40 Hz. When doing a simple time-frequency analysis in SPM (using a wavelet factor of 10), I get these high frequency stripes. Is it correct that in SPM the only way to optimize my analysis would be to use a different Morlet wavelet factor?
>
With standard wavelets you will always get the stripes at high frequencies, just if you have more cycles those fingers will be pushed higher and you will loose time resolution. To get rid of them you can use wavelets with fixed time window (there is an option to do that in the version you got from the course (which is not yet publicly released). Alternatively you can try a different method of spectral estimation e.g. multitaper.
> I am more keen in having a good frequency resolution (~1 second) than a good temporal resolution. What wavelet factor would you advise? Is there another way in SPM to optimize my analysis?
>
It's not quite clear to me whether you meant high time resolution or high frequency resolution but the more cycles in the wavelet or the longer the time window in other methods the higher the frequency resolution and the lower the time resolution. With other methods you can optimise your analysis in different ways. For instance with Hilbert and multi taper you can have decreasing frequency resolution with frequency which might be a good idea in some cases. To do that you need to specify the frequency resolution as a vector, the same length as the vector of frequencies.
>
> Another question concerns the conversion to images. If I want to convert my time-frequency data to images using SPM, I have the option to average above channels or frequencies. Is there a way to average over time instead?
>
If you mean that you want to estimate a spectrum rather than TF there is a function for that in MEEGTools called 'Fieldtrip multi taper power map'. This function uses multi taper as it doesn't make much sense to use wavelets if you don't want to have spectrum resolved in time. The function presently still requires signal processing toolbox.
Best,
Vladimir
>
>
> Thanks a lot for your help,
>
> Anette
>
>
|