Hi, list,
I've improved on my previous attempt to create a hilbert transform.
The hilbert_by_fft patch actually had a problem with its operation,
similar to errors caused by circular convolution. The improved patch
eliminates this problem by reformatting the transform as a causal
filter (with latency). Hilbert2~.pd goes down as low as 10 Hz, with
good accuracy. The latency is 8128 samples, which is about as low as
it can be for the frequency response (unless we use a rectangular
window instead of hanning window). I have very good confidence that
this is the correct implementation. Note: uses z~
outlets of hilbert2~ from left are: amplitude, phase, real part, and
imaginary part (the hilbert transform'ed real part)
Chuck