>Another project had a 12 bit A/D converter, and clean values were critical.  I
>was getting 3 bits of noise, and filtering the crap out of it to get
semi-stable
>values (sound familiar, anyone?).  The customer had some very seasoned analog
>hands at their place, and they borrowed the unit. When it came back, it had a
>nest of wires soldered all over the boards, but the converter gave me rock
>steady readings to one bit without a filter.  (Using a good millivolt source).
>Very few times in my endevors had I seen a 12 bit value sit on a bit.  And they
>didn't use a ground plane, they just re-wired the supply and reference
networks.

This effect is what is sometimes known as "ground bounce". Typically solved as
described by being careful with where sensitive grounds get connected.

When wire wrapping a logic board or laying out a double sided board for logic,
the rule used to be to make all power lines a grid across all chips in both X
and Y directions. this was necessary in TTL because of the current consumption.
People have gone got away from this when using CMOS and other low current
devices, as the power drop considerations were reduced. It can still be a
problem however, and if you have analogue interfacing it would probably pay to
do this on your wire wrap. Also keep the analogue and digital portions separate
up to the point where your a-d interface is, as this is the point at which the
grounds should join.

It would be worth looking at any application notes you can find on laying out
PCB's around a-d converters. They will often give pointers on exactly this sort
of problem.