Re: [ATM] Foucault images

["Stephen Koehler" <s.c.koehler@gmail.com>]

Mike,

> Depending on software, a power-of-two-length FFT should be fastest
> (1024, 2048), but I'm guessing you know that already.

The FFT in R works well, as long as the array size is "highly
composite", meaning that it has lots of small factors.  There's a
function I make use of (nextn) that picks the next higher
highly-composite number above a given number.  In this case, 1000*1.2
is already highly composite, so the FFTs were actually performed in
1200x1200 arrays.  This technique allows me to use any combination of
pupil size and padding, and still get the FFTs to run fast without
changing my preferred padding too much.  This is much more convenient
than being force to work with the closest power of 2, which would be
very slow if I was forced to jump all the way from 1024 to 2048.

--
Steve Koehler
_______________________________________________
ATM mailing list http://www.atmlist.net/