I agree with Chris on this one. I'd add some things, though. Because
of transmission-line effects, the ground return currents for digital
ICs will tend to flow on the ground plane in the region very near the
supply lines, so keep the supply lines for the digital section away
from the sensitive analog section.

Another way to make the digital stuff quiet is to realize that a great
deal of the high frequency currents involved with digital are in the
I/O lines that go into, and especially go out of, chips. It helps to
pay special attention to all digital signal lines. You can soften the
edges by putting small series resistors physically very near each
output, so as to limit the current into the capacitance of the trace
and inputs it goes to. It can help a lot, too, to use differential
signalling even when you don't need to. Just as differential lines in
the sensitive analog input end help to reject noise, so differential
lines radiate less than single ended ones. Differential outputs also
make for quieter power supply lines, since one output turns on whenever
another turns off, maintaining constant supply current. Fully
differential ECL is particularly quiet on the power supplies, but
likely not worth the effort because of how power-hungry it is and how
little integration you can get on a single chip.

We do use little ferrites on power supply lines; they provide effective
high frequency isolation. 0.1uF bypasses aren't really very good at
bypassing 100MHz -- it's nice to also use some smaller values, and to
use physically small parts that have low parasitic inductance. It pays
to THINK about the layout quite a bit, and understand what you are
doing and why.

For sensitive analog circuitry that draws constant current, there's a
cute and very effective power supply cleaning circuit and ap note on
the Wenzel Associates website. The noise out of a typical linear
regulator will significantly add to the phase noise of a PLL. That
site also has some good info about low phase noise circuits. (somewhat
aside: Wenzel makes some really fine low-noise oscillators, and
Charles Wenzel is a very nice, and very interesting, guy.)

I think this all says that what Chris does also DOES work for me, and I
can affirm that much of it is in a commercial environment where we are
achieving cutting-edge performance.

Cheers,
Tom