[FieldTrip] Question about least-square FIR filter

[Andreas Widmann]

Hi,

> I am trying to process my data through Hilbert transform. I am used to the EEGLAB package, so I tried to use the same filter recommended by EEGLAB, which is least-square FIR filter in fieldtrip. However, I got a warning, when running the function ft_specest_hilbert, that least-square FIR 'is not recommended for neural signals, only proceed if you know what you are doing’. I am curious why  least-square FIR is not appropriate for neural signals?
Short version: See
http://bugzilla.fieldtriptoolbox.org/show_bug.cgi?id=2453
Point (3) and first attachment (png).

Long version:
firls filters are also NOT recommended in EEGLAB. Where did you get this information? The EEGLAB firls filter are marked legacy/deprecated since 2012. See
http://journal.frontiersin.org/article/10.3389/fpsyg.2012.00233/full
for a more detailed description of the problem, in particular footnote 1. Filter roll-off/transition width is a function of the filter order. However, with MATLAB firls filters roll-off/transition width is defined independently from filter order. This might work if the selected parameters for both are appropriate, however, commonly it does not work resulting in various artifacts in the filter’s frequency and/or time domain responses, bad attenuation, non-unity DC gain etc. In Fieldtrip the problem was/is particularly problematic as a zero transition width (rectangular frequency response) filter is requested combined with an incorrect estimation of required filter order (actually infinite).

Hope this helps! Best,
Andreas