I came up with a sliding FFT that does 256 N and ~3k samples/sec on one channel,
based on
http://www.ultimaterobolab.com/pdfs/5_RFT.pdf
Has anybody done better?
Interested in re-factor or translate to C?
(Would C be much faster?)
N seems limited to 256, as 512 data and twiddles over-fill the memory.
I'm working on a recursive peak-frequency finder at the moment.
Ray