Back to the main page.
Bug 225 - what's the effect of slow drifts in my data on my spectral decomposition?
| Status |
CLOSED FIXED |
| Reported |
2010-11-18 09:59:00 +0100 |
| Modified |
2011-07-06 16:38:12 +0200 |
| Product: |
FieldTrip |
| Component: |
core |
| Version: |
unspecified |
| Hardware: |
PC |
| Operating System: |
Mac OS |
| Importance: |
P1 normal |
| Assigned to: |
Jan-Mathijs Schoffelen |
| URL: |
|
| Tags: |
|
| Depends on: |
|
| Blocks: |
|
| See also: |
|
Jan-Mathijs Schoffelen - 2010-11-18 09:59:07 +0100
A fundamental feature of spectral analysis (when using tapers to control for spectral leakage) is that the basis functions are not strictly orthogonal. In particular the low frequencies (DC + ) present in the data can as a consequence leak to the whole spectral range. As this may be an unwanted outcome of the spectral analysis, it is generally best to mean subtract/detrend/highpassfilter the time domain data prior to the spectral decomposition.
After a discussion with Robert we decided not to let the user be responsible for this, but build it into fieldtrip. More specifically, at the level of the low-level specest functions we will implement this as
'detrend' 0/1/2, which will fit and subtract a 0/1/2 order polynome to the time series. The higher level calling function will set the default (did we agree to let ft_freqanalysis put it to 2?).
Additionally, we will write some documentation in the form of a FAQ, similar to one already existing FAQ related to the TFR and DC.