David,
On the other hand I've had the suspicion that the actual subband sample rate was too small by a factor of 2.

I suspected this because the subbands are apparently divided into approximate 10KHz bands which implies that the subband sample frequency  should be about 20KHz (2*9765.62 = 19531.24 Hz).

Al

> David Woolley wrote:

In article <41EB17FB.1080505@sbcglobal.net>,
Alfred A. Aburto Jr. <aburto@sbcglobal.net> wrote:

How is a "short FFT" defined?

Short enough that it has sufficient time resolution to perform gaussian
(or pulse) analysis.  If one ignores very high chirp rates, most FFTs have a length of 128K, which corresponds to about 13 seconds, or between
about 0.5 and 1 beam transit times, and therefore unusable for detecting
gaussians etc.

A quarter the number are 64K, one sixteenth 32K etc., one 64th 16K.  Only
about 16K (I haven't checked the actual figure, which depends on the angle rate) is there enough time resoultion to compute gaussians.

My argument is that the ratios should be two to one, not four to one.  One
doesn't need to actually start doing that until one reaches short FFTs, but
one would need many more of the first FFT length for which this applied
than at present (actually more than if one had simply used a ratio of
two to one all the way, as typical transit times are typically twice the
time for the long FFTs).

Are the bins shaded? The effective "FFT length" would be a bit different if the bins were "shaded" ...

Do you mean do they window the FFT.  It looks like they might not.  I'm not
sure of the impact of that.  Arguably they ought to be overlapping FFTs in
time.  (Arguably they ought to also do this in frequency, but the additional
processing might be quite significant.)  However, I'd rather any discussion
of such be moved to a new thread.