[time-nuts] Linear voltage regulator hints...

[ed breya]

I'd recommend going with what Bob Stewart mentioned, using separate 
gates as buffers, operated from a better-grade reference, to shift 
from the noisier and driftier logic supplies, into the more critical 
circuits. It's simple, and can be powered from a modest reference circuit.

If the logic circuits themselves need better supply noise and tempco 
performance, don't use any kind of three-terminal regulators - use a 
good opamp driving a pass transistor. Use a reference IC that has a 
buried zener for lowest noise - this eliminates all the low voltage 
references and three-terminal etc regulators that use band-gap 
references. The down side is that the good kind of reference ICs will 
need a higher (like 10V and up) operating voltage than may be 
available, so that complicates it.

For a system using a conventional PC-style supply, with +5V and +12V 
available, an LM399, for example, could run from the +12V, along with 
the opamp circuitry, while the pass transistor could feed from the 
+5V, dropping to the +3.3V or whatever low logic supply is needed. 
For modest current requirement, use only an NPN pass transistor in 
emitter-follower mode. For higher currents, add another NPN 
emitter-follower in front of it for more drive - its collector can be 
supplied from the +12V via some limiting R, to ensure enough 
overhead. The opamp and associated network resistors, of course, 
should have performance commensurate with the reference, and 
sufficient for the application. Since there's also plenty of digital 
and PS noise around, a lot of bypassing in the right spots should help a lot.

Ed