D-Noizator: a magic active noise canceller to retrofit & upgrade any 317-based V.Reg.

[Elvee]

✪ A simple circuit that transforms a good, clumsy workhorse like the LM317 into a superreg so easily that you do not even need to cut a single track. ✪
✪ It must be too good to be true.... or ✪

.................................................. .................................................. .....................

It's Christmas!!!

The present SOTA in VR noise cancellation is this:



From Wenzel finesse regulator: Finesse Voltage Regulator Noise! |

Drawbacks are obvious:
a) Low current
b) Affects static and dynamic operation
c) Limited correction ability

Let us start with the basic regulator (including the additional bypass cap, since we strive for the lowest noise).
An additional source symbolizing the noise is shown explicitly.
The bypass cap is effective down to ~10Hz, but the level does not fall below 0dB.



Now, let us add the D-Noizator piggy-back:



The floor drops by more than 30dB.
The circuit works by changing the output node into an AC virtual ground.
Therefore, benefits do not stop at internal noise reduction: all the performances benefit: PSRR, output impedance.

Ultimately, the performances are limited by the discrete transistor.


What about the reality?

I made a test with ST and ONsemi samples. They behave as perfect clones, both with and without the D-Noizator.



With just the cap, the noise is 16μV, as measured by a Levell TM3B.
With the D-Noizator, it falls to 2.5μV~3μV, which seems to indicate that the circuit doesnt deliver all of the promised improvement... except that the noise floor of the TM3 is also 2.5μV~3μV.
The TM3 is not a bad instrument, but it is very cheap, and very dated.
I would need to add a modern LNA to make meaningful measurements.



It is more than promising, anyway.

Of course you cannot throw a large gain into the regulator's loop without consequences: it is now forced to operate at a gain much lower than unity, and without precautions, it oscillates like mad, which comes as no surprise.
However, compensation seems to be easily achieved, and very tolerant: a capacitor larger than 2nF was sufficient to stop oscillations.

A VLF peaking is visible in the response; it could probably be polished off by increasing the splitting of the poles, or similar measures

This post details the possible versions, and provides practical information:

D-Noizator: a magic active noise canceller to retrofit & upgrade any 317-based V.Reg.

[Lingwendil]

Interesting. Wonder how well it will work for digital stuff like DAC supplies.

[Elvee]

Originally Posted by Lingwendil

Wonder how well it will work for digital stuff like DAC supplies.

It depends on two things: is the target supply using a divider for the voltage setting, and can you find a discrete transistor better than the regulator's internal circuitry? Note that even a fixed regulator could benefit from the fix, but from a practical POV, things would be much more difficult: you would need to isolate the reference terminal, which generally happens to be GND. Probably a detail for compulsive modders, but a serious obstacle for all the others.

In principle, you will always be able to find a discrete superior to the internal circuits of a VR, but with modern VR's, this is going to be difficult, and perhaps not worth the trouble. With 317 style regulators, it was easy, because the IC has to cope with everything, DC, AC, thermal stability at the same time. With a discrete, you can separate the AC and DC paths, which eases matters considerably. However, IC designs have become so much smarter that bettering them requires a significant effort. It is certainly possible, but any gain is probably going to be marginal, and if you want a real improvement, you need to design everything from scratch, not try to improve on something that is barely improvable

[MarcelvdG]

Nice circuit, like in all threads that you start!

[jackinnj]

Hat tip to Elvee. I ran some simulations of Z out and PSRR with Multisim:

[Elvee]

I have been able to refine the measurements: I remembered that one channel of one of my lab amplifiers is equipped with a OPA27.

This brought the noise floor of my test set-up to 0.75µV.
When measuring the D-noizator, I got 1.0µV.
This means that the actual noise is in the vicinity of 0.66µV.
Since the noise with the cap connected normally is still 16µV, the improvement ratio is 27.7dB, much closer to what the sim predicted.

Of course, the test is made on a breadboard, and the transistor is just an ordinary BC337-40 from Philips.

With a good PCB and a super low noise transistor, it would be possible to scrape a few tens of nV

[Mark Johnson]

Summary: low noise voltage regulator achieved using twelve electronic components, one of which is an integrated circuit.

[DontHertzMe]

Would the same principle apply for the LM337?

[PRR]

Originally Posted by Mark Johnson

Summary: low noise voltage regulator achieved using twelve electronic components, one of which is an integrated circuit.

You forgot "magic".

[Elvee]

Originally Posted by Mark Johnson

Summary: low noise voltage regulator achieved using twelve electronic components, one of which is an integrated circuit.

Yes.
The piggyback alone (since the main use of this circuit is probably going to be the retrofitting existing regulators) requires 7 components

Originally Posted by DontHertzMe

Would the same principle apply for the LM337?

Yes, of course: use a PNP and reverse the polarity of E-caps.


Here is a slightly better tradeoff: the current consumption is increased, but the improvement factor is 38dB, and the VLF peak is mostly gone.

Going further probably makes little sense, since finding a transistor sufficiently good to capitalize on the gain is going to be difficult.