# [Eeglablist] FFT vs. PSD

Arnaud Delorme arno at ucsd.edu
Fri Oct 24 13:35:49 PDT 2014

```Hi Sebastian,

We will look into it. I think when you use single trials, the computation has to be done with 'logtrials', 'on'.

Arno

On Oct 24, 2014, at 12:23 PM, Sebastian Grissmann <sebastian.grissmann at lead.uni-tuebingen.de> wrote:

> Dear Arno,
>
> I just realized that I was getting the same results ('logtrials' 'on' or 'off') when I was also saving the single-trial measures for single-trials statistics.
>
> Could this be a bug?
>
> Since I was always saving the single-trial measures, I´m now wondering if I can assume that I already got the log data (even without 'logtrials' 'on')?... Or should I rather recompute my results?
>
> Best,
> Sebastian
>
> Quoting Sebastian Grissmann <sebastian.grissmann at lead.uni-tuebingen.de>:
>
>> Dear Arno,
>>
>> No, I was using 'specmode', 'fft', 'nfft', 1024 (to get a finer grained picture) 'logtrials', 'off' (the default)
>>
>> Thanks for the info! I was assuming (based on the plots) that the log is always computed.
>>
>> Best,
>> Sebastian
>>
>> Quoting Arnaud Delorme <arno at ucsd.edu>:
>>
>>> Dear Sebastian,
>>>
>>> Did you try 'specmode', 'fft', 'logtrials', 'on'?
>>> In one case (logtrials off), the function computes the power for each trials, average power across trials, then compute the log for display.
>>> In the second case (logtrials on), the function computes the log power for each trial then averages these trials.
>>> So it is a matter of when you take the log.
>>>
>>> It is a matter of taste. Log transformation tend to reduce the noise, so maybe the second option is better.
>>>
>>> Arno
>>>
>>> On Oct 2, 2014, at 10:44 AM, Makoto Miyakoshi <mmiyakoshi at ucsd.edu> wrote:
>>>
>>>> Dear Sebastian,
>>>>
>>>> It seems you are asking 'which one is correct' or 'what the correct method', but there is no such a thing as 'the only solution' here. If you change the computational method, the result also changes. As long as your result can be replicated by other researchers by using the methods and parameters you describe in the paper, and also as long as your reviewer says no to your methods and parameters (or don't say anything about them), you are fine. You can, and should, choose your methods and parameters and you owe full responsibility.
>>>>
>>>> To conclude, you can choose whichever methods/results you like. Just be confident; instead of saying 'which one should I use', which is so nice of you, but once you learn both are valid you should say 'what's wrong with using this'. I know how you feel since my degree is psychology too.
>>>>
>>>> Makoto
>>>>
>>>> On Thu, Oct 2, 2014 at 12:37 AM, Sebastian Grissmann <sebastian.grissmann at lead.uni-tuebingen.de> wrote:
>>>> Hi there,
>>>>
>>>> I´m a PhD student (who studied Psychology) and currently trying to analyze my first EEG dataset. I first started to compute my channel spectra ( pop_precomp() ) via a FFT (using the default spectopo parameters: ‘specmode’, ‘fft’, ‘logtrials’, ‘off’), but when I was later looking at specfreqs (returned from std_specplot) I found that my frequency bins were quite broad (>1Hz). Since I need a higher frequency resolution for my analysis I tried PSD instead of FFT to compute my spectra (spectopo parameters: ‘specmode’, ‘psd’, ‘logtrials’, ‘off’). Now I had a very good spectral resolution, BUT the spectra looked quite different. For example, the alpha peaks were gone (or strongly diminished) in the PSD spectra and the statistics also returned very different results. I later found out that I can also increase the resolution of the FFT via zero-padding (spectopo parameters: ‘specmode’, ‘fft’, ‘nfft’, 1024, ‘logtrials’, ‘off’), but the spectra still look quite different.
>>>>
>>>> Here is the link to some figures. The p-value for the statistics was always 0.05.
>>>>
>>>>
>>>> I´m using EEGLAB v13.2.1; Sample rate = 250Hz; epoch length = 700ms; trials = 59-306; Bandpassfilter = 1-30Hz
>>>>
>>>> Can anyone help me?… PLEASE…
>>>>
>>>> Best,
>>>> Sebastian
>>>>
>>>>
>>>> Sebastian Grissmann (Mag. rer. nat)
>>>> Neuroengineer / PhD student
>>>>
>>>> University of Tübingen
>>>> Europastrasse 6
>>>> 72072 Tübingen
>>>> Germany
>>>> Phone  +49 7071 29-73604
>>>> _______________________________________________
>>>> 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
>>>>
>>>>
>>>>
>>>> --
>>>> Makoto Miyakoshi
>>>> Swartz Center for Computational Neuroscience
>>>> Institute for Neural Computation, University of California San Diego
>>>> _______________________________________________
>>>> 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
>>>
>>>
>>
>>
>>
>> Sebastian Grissmann (Mag. rer. nat)
>> Neuroengineer / PhD student
>>
>> University of Tübingen
>> Europastrasse 6
>> 72072 Tübingen
>> Germany
>> Phone  +49 7071 29-73604
>> _______________________________________________
>> 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
>
>
>
> Sebastian Grissmann (Mag. rer. nat)
> Neuroengineer / PhD student
>