[Eeglablist] .datspec freq bins and epoch size

[Mikołaj Magnuski]

Unfortunatelly, that's the basic time/frequency trade-off - if you want to
resolve frequencies that are closer than your frequency resolution
you need a longer singnal sample (longer analysis window).
At least this is the case for standard methods like FFT.

There are also so-called 'parametric' methods that are said to overcome
some of the limitations of the classical methods (see for example MUSIC
algorithm), but are less general (require some prior knowledge about the
signal). I have never used any of these methods however, so I cannot
tell you more than that they exist :)

And then - it depends on what you are trying to examine in the signal.
Maybe extending analysis window and/or using padding will be satisfactory.




Pozdrawiam,
Mikołaj Magnuski


2013/9/13 Steven Pillen <stevendpillen at gmail.com>

> That is, without interpolation  (which probably isn't ideal for the kind
> of things I'm trying to see)?
>
>
> On Fri, Sep 13, 2013 at 11:02 AM, Steven Pillen <stevendpillen at gmail.com>wrote:
>
>> So there's no way to get smaller frequency windows with an epoch size
>> that small?
>>
>>
>> On Wed, Sep 11, 2013 at 5:56 PM, Mikołaj Magnuski <
>> imponderabilion at gmail.com> wrote:
>>
>>>
>>> Hi Steven,
>>>
>>> Frequency resolution depends on window size, so what you're seeing is
>>> normal.
>>> However, for visualisation purposes you can interpolate the spectrum
>>> (which is implemented by zero-padding your data to some desired length) -
>>> EEGlab functions allow to specify the degree of this smoothing.
>>> For example you can control the zero-padding in spectopo function with
>>> 'nfft' (specific length to zero-pad the data to).
>>> By specifying 'nfft' to be two times the number of samples in your
>>> analysis window you increase the frequency resolution twofold (although you
>>> have to remember that this is not a genuine increase in frequence
>>> resolution but an interpolation).
>>>
>>> Also, see the help messages to pop_spectopo and spectopo functions.
>>>
>>> For time-frequency you can use the 'nfreqs' key in newtimef.m to control
>>> the frequency resolution
>>>  12 wrz 2013 01:27, "Steven Pillen" <stevendpillen at gmail.com>
>>> napisał(a):
>>>
>>>>  Hello, EEGLABlist.
>>>>
>>>> When we ran power spectrum measures through the study functions on
>>>> files whose epochs were about 0.2 seconds long, it looked like the power
>>>> spectra chart generated had very few points of reference.  Looking at the
>>>> .datspec files using the function
>>>> *
>>>> *
>>>> *m = importdata('design1_2003_regular.datspec')*
>>>> *
>>>> *
>>>> in m.freqs, it looked like there was one bin for every 5 hz, which
>>>> matched what the charts looked like.
>>>>
>>>> We ran the same data again, this time cut into 1 second epochs, and the
>>>> power spectra looked like they had many more reference points.  Likewise,
>>>> there appeared to be a bin in m.freqs for every 1 hz. as opposed to the
>>>> previous 5.
>>>>
>>>> Is there any way to increase the number of bins with a smaller sized
>>>> epoch so that the power spectra is more robust?
>>>>
>>>> _______________________________________________
>>>> Eeglablist page: http://sccn.ucsd.edu/eeglab/eeglabmail.html
>>>> To unsubscribe, send an empty email to
>>>> eeglablist-unsubscribe at sccn.ucsd.edu
>>>> For digest mode, send an email with the subject "set digest mime" to
>>>> eeglablist-request at sccn.ucsd.edu
>>>>
>>>
>>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://sccn.ucsd.edu/pipermail/eeglablist/attachments/20130914/7c0a7446/attachment.html>