[Eeglablist] newcrossf coherence matrix

[Arnaud Delorme]

On Aug 5, 2009, at 9:21 AM, Sacks, Annette E wrote:

> I've been doing a channel cross-coherence analysis in EEGlab v6.01b  
> using newcrossf() with a sampling freq of 250Hz, epoch start/end  
> [-400 1896], and frequency range [0 100], which outputs a 102 x 200  
> coherence matrix.  This is the call I used:
>
> >> [coh,mcoh,timesout,freqsout,cohboot,cohangles] =  
> pop_newcrossf(EEG, 1, 1, 2, [-400 1896], 0, 'freqs', [0 100],  
> 'alpha', .05);
>
> Which gives this output:
>
> Adjust min freq. to 0.98 Hz to match FFT output frequencies Adjust max
> freq. to 99.61 Hz to match FFT output frequencies Generating 200 time
> points (-143.6 to 1639.6 ms) The window size used is 128 samples (512
> ms) wide. Estimating 102 linear-spaced frequencies from 1.0 Hz to  
> 99.6 Hz.
>
> My question is: how do I calculate exactly how the time and  
> frequency bins are being split up in the 102x200 (freq x time) coh  
> matrix ?
> ie, we don't know the exact beginning and ending time/frequency that  
> is being collected in each entry of the coh matrix.


The freqsout and timesout matrices contain the actual time latencies  
and frequencies. When you use FFT, even if you specify actual  
frequencies 'freqs', [1 2 3 4 5 6 7.... ] (which is possible), the FFT  
decomposition is run based on your window size and the function then  
searches for frequencies closest to the one you have asked for. When  
you use wavelets, you get the decomposition at the frequency you have  
actually asked for.

To obtain round frequencies for FFT, just choose a window size  
multiple of the sampling rate. For instance 250, 125, etc...

Hope this helps,

Arno


>
> Thanks,
> Annette Sacks
> <ATT00001.txt>

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://sccn.ucsd.edu/pipermail/eeglablist/attachments/20090806/4196511b/attachment.html>