enable method='gust' for sosfiltfilt

[mike-lawrence]

Hopefully a pretty straightforward FR: at present scipy.signal.filtfilt() has a method argument by which one can choose the gust method of padding, but scipy.signal.sosfiltfilt() doesn't, meaning one has to choose between an implementation of filtfilt that has the gust method but is numerically unstable for higher order filter designs, or a numerically stable filter with no gust method.

[larsoner]

In principle this does sound like a useful addition. I am not familiar with the gust method (only seen that it exists in the code) but if someone can take a look at some point and see if it's doable, that would be great. @mike-lawrence if you have already looked at the code and/or know the method well enough to know it's doable (and have time to work on it), feel free to give it a shot!

[rkube]

This addition would be useful for me too.

[CrepeGoat]

I'm also interested in this addition; atm I'm making do with a loop like the one below, but it feels a little hacky (a formal version would also almost certainly be faster):

sos_coeffs = scipy.signal.butter(..., output=sos)
for b, a in zip(*np.split(sos_coeffs, [3], axis=-1)):
    array = scipy.signal.filtfilt(b, a, array, axis=-1, method="gust")
return array