[Eeglablist] The meaning of dB

[Arnaud Delorme]

Antoine and Tom,

As Tom mentioned, the inaccuracy on power scaling has no effect on standard
ERSP (power ratios). This is only important when retrieving single
trial time-frequency estimates or when plotting ERS (event related spectrum)
using the " 'baseline', NaN " option in timef() or newtimef(). Data and
component spectrums plotted using the spectopo function use the Matlab
pwelch() function so the scaling should also be correct. Anyway, thanks Tom
for checking that out and finding this problem. We will take your changes
into account and test the function further for power absolute scaling.

Arno

Thomas Ferree wrote:
> Antoine,
> In eeglab 4.515 at least, the timef function does not return power 
> exactly.
> To start with, the Matlab FFT function is not normalized for the 
> number of
> points.  When computing power or amplitude in uV, one needs to normalize
> by the number of data points.  We went through a series of steps to 
> remedy
> this and other related things in the timef fuction.  Attached please 
> find a
> slightly modified version of the timef function.  Search on the string 
> 'TF
> and MC' to find our changes.  Note these changes are only multiplicative
> factors, so they do not affect the ERSP which is a power ratio.  With
> this function, the power is the so-called average power, not total power
> or power spectral density (see Numerical Recipes), and is corrected for
> number of data points, windowing and negative frequencies.  We made
> these edits only for the basic FFT settings, but took care that the
> normalization for data points is correct when zero-padding, i.e., one
> should normalize by the number of non-zero data points only.  In order
> to compute the amplitude in uV, take sqrt(2*P) where P is power.  We
> tested this by overwriting an EEG.data array with sine waves of known
> amplitude (use something other than unit amplitude) and verified that
> we were getting the right answer.  You should verify that yourself too.
> Hope this helps, Tom.