> When going back from the frequency domain to signal, I have to divide the 
> real part of the result by the length of the signal. Is it the convention 
> that I do that, or should the library?

Looks like every library does its own thing here - ie scale, not scale, offer 
options to scale - for now, fftr does no scaling (since scaling might be needed 
in either direction and sometimes by the square root, if done in both 
directions) - this is probably a reasonable starting point though maybe 
confusing for ... python and matlab users. hm.. see also 
<https://stackoverflow.com/questions/62881003/fftw-backwards-transform-is-multiplied-by-n>

> It seems that depending on the length of the signal, the ifft result is 
> flipped sign wise.

Bug, should be fixed now

Reply via email to