I borrowed 5V from the USB line of a data acquisition card -- it is somewhat
noisy so I put a 100uH choke and a small tantalum on the line -- it works.

You can get excellent noise performance (far exceeding the Wetzel modified
LM317 circuit) with Linear Technologies LT1763 and LT1964 ultra-low noise
regulators -- they will do 100mA on their own and can be used to drive a
pass transistor if you need more juice. Unfortunately, post regulating any
digital supply for analog circuitry burns up watts.

fwiw -- while dual bypass caps aren't a good idea for digital circuits
(because of the ringing aforementioned) -- in analog they are recommended
for certain low noise apps because of susceptibility to EMI and RFI.

Huh? The LT1763 claims 20uVRMS noise in a 100kHz bandwidth. That's
about 63nV/rtHz. The data sheets shows it MUCH worse at low
frequencies, and the LT1964 a bit worse yet. I'm not sure who Wetzel
is, but the (modified) Wenzel circuit I've used has given me noise down
in the very few nanovolts/rtHz region down to 10Hz and below, depending
on just how I implement it. Perhaps we're thinking of different
circuits for cleaning supplies. The one I'm thinking of is very useful
even with "quiet" regulators like the LT parts. Easy implementations
get you 20dB improvement in noise, and careful design and parts
specification can get you 40dB, with noise as low as a couple nV/rtHz
and a 1/f corner well below 100Hz, by using an op amp like the AD797
and low noise precision resistors in what is essentially a bridge
circuit to cancel the regulator noise.

Guess the jury is still out on the multiple bypass cap values, too.
Xilinx and TI at least both recommend -- strongly recommend -- multiple
values. Yes, you want to avoid resonances, but I've had much less
trouble with bypass caps in parallel than with chokes resonating with
with bypass caps. Now that's something to look out for! The trick
with parallel bypass caps is to insure that the Q of any parasitic
inductances is low.

Cheers,
Tom

More on capacitors of differing values in parallel can be found many
places, of course. Here's one web resource that should be of interest:

http://www.ultracad.com/articles/esrbcap.pdf

It has specific ideas beyond the low-Q thought I mentioned, when you
parallel different values.

Cheers,
Tom

Yes, you are right -- my recollection was faulty and I apologize.

I was just looking at the "clean up" portion of the Wenzel (not Wetzel)
regulator with an LM317 -- later on in the article (Finesse Voltage
Regulator Noise) they use an LM833 which is cited as a limiting factor.

Jung used an AD797 in the Super-Regulator series of articles, but I believe
that they switched to an AD817 or AD825 later on as there were some
problems. The Jung article can be found on EDN's website:
www.ednmag.com -- Just put "Jung" in the search engine. Some people
reported that the Jung super-regulator could be made to oscillate pretty
easily. I didn't have this problem.

I built the super-regulator and used it with a low noise preamp, but found
the Linear Parts to be much easier to implement, as long as I used a
tantalum bypass !

"K7ITM" wrote in message
ups.com...
Huh? The LT1763 claims 20uVRMS noise in a 100kHz bandwidth. That's
about 63nV/rtHz. The data sheets shows it MUCH worse at low
frequencies, and the LT1964 a bit worse yet. I'm not sure who Wetzel
is, but the (modified) Wenzel circuit I've used has given me noise down
in the very few nanovolts/rtHz region down to 10Hz and below, depending
on just how I implement it. Perhaps we're thinking of different
circuits for cleaning supplies. The one I'm thinking of is very useful
even with "quiet" regulators like the LT parts. Easy implementations
get you 20dB improvement in noise, and careful design and parts
specification can get you 40dB, with noise as low as a couple nV/rtHz
and a 1/f corner well below 100Hz, by using an op amp like the AD797
and low noise precision resistors in what is essentially a bridge
circuit to cancel the regulator noise.

Guess the jury is still out on the multiple bypass cap values, too.
Xilinx and TI at least both recommend -- strongly recommend -- multiple
values. Yes, you want to avoid resonances, but I've had much less
trouble with bypass caps in parallel than with chokes resonating with
with bypass caps. Now that's something to look out for! The trick
with parallel bypass caps is to insure that the Q of any parasitic
inductances is low.

Cheers,
Tom

K7ITM wrote:

values. Yes, you want to avoid resonances, but I've had much less
trouble with bypass caps in parallel than with chokes resonating with
with bypass caps. Now that's something to look out for!


You've got that right!

I'd be very careful sticking a choke, especially one without a swamping
resistance in parallel, in series with a supply line! Unless you hve a
specific frequency range you have to block an L/C filter can be big
problems.