Seems to me that 256 or 512 is reasonable. Then decide how many
kHz/rev you want, and that tells you the step size. If you want
10kHz/rev, 512 steps would give you just under 20Hz/step, which seems
reasonable to me. I wouldn't want it any coarser than that, and I'd
actually prefer finer. Processors are very cheap; you can easily
process the quadrature step info up to dozens of revolutions per
second. If you budget ten instruction cycles to process each step, and
you're using a slow processor at 1usec/instruction cycle, you can
process 100,000 steps a second, if the processor has nothing else to
do. You can process 10,000 steps a second, or 200kHz/second, with just
ten percent of the processor's time. A nice "plus" is to accelerate
the tuning when the steps come fast, so you might bump the tuning up to
higher Hz/step as the steps come faster. Expect to spend some time on
the algorithm to get a smooth "feel" to it, though. The accelerated
tuning when the knob is turned more quickly is a feature commonly found
in test instruments. I haven't ever had a problem with the HP/Agilent
encoders with respect to the output looking like "noise." The outputs
are very clean digital signals with no "bounce" if you're turning in
one direction.

Cheers,
Tom

PS...Tim, did you find an encoder yet? Drop me an email if not...