[FieldTrip] zero-padding VS mirror-padding (ft_freqanalysis)
tzvetan.popov at uni-konstanz.de
Sat Sep 19 09:40:28 CEST 2015
You could also use data padding, i.e. epoch the data as longer segments perform freqanalysis and use ft_selectdata to re-epoch to the desired length.
> Am 18.09.2015 um 17:35 schrieb Maris Skujevskis <icelandhouse at gmail.com>:
> Dear Fieldtrip community,
> I am currently doing frequency analysis on an EEG dataset.
> My question is a general one about the dis/advantages of zero- VS mirror-padding a finite time series prior to frequency analysis (ft_freqanalysis).
> The way ft_freqanalysis is implemented suggests that zero padding is always the best option, e.g., ft_freqanalysis does not support mirror padding.
> However, it seems to me that mirror padding is also a good and a valid way to address the 'missing values' issue at the temporal edges of a TFR.
> Here is my (intuitive) reasoning:
> The disadvantage of zero padding is that it creates a discontinuity at the edges of the data segment, thus introducing additional frequency content and distorting the power estimates. Mirror padding might overestimate the power of the frequencies present, but, to its advantage, it preserves the frequency content of the actual data. Short summary: both ways of padding have their strengths and weaknesses, there is no clear winner.
> It would be good to hear what other researchers think about the advantages/disadvantages of the two ways of padding.
> Is zero-padding always a better way of padding than mirror- (or any other type of) padding?
> Does the answer depend on some further factors?
> Best wishes,
> fieldtrip mailing list
> fieldtrip at donders.ru.nl
More information about the fieldtrip