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

Thanks Elvee.
The "Finesse" circuit works OK (and it can be adjusted)
Sure, it works, but not only can you adjust it, you need to, to suit the peculiarities of the target regulator: you cannot use the same setup for 50mA or 1A.

In addition, you improve one parameter at the expense of another.

My proposition is "one size fits all", and it will remain identical for any 317 used in a "regular" manner: it will of course not work with circuits that already have been augmented in a way or another, by including an opamp in the loop for example.
It also improves all the three key parameters at once: you don't need to compromise.
Finally, it is also minimally invasive and purely additive: no need to remove something or cut tracks.

OMG. A LT3042 with an external transistor to carry the dissipation is
> 40 dB better than any LM317, right from the start.
Circuit is in the data sheet.
You probably missed the title and the point of this thread: it is not about developing some kind of new revolutionary superreg, but about upgrading existing equipments using 317/337's, which certainly number in the millions, if not more.

All you have to do is lift the cover, stick a post-stamp-sized PCB somewhere using a double-sided adhesive or a dab of silicone, solder 3 connections and you are done. The worst you might have to do is to lift one side of a capacitor.
No rework or butchering of the existing PCB, and all the DC voltages remain the same to the mV.
As a result, and for a negligible cost, you get a ~30dB improvement on all three key parameters: 90dB of cumulated improvement.
How sweeter could it be?


Elvee, nice idea! When I read the posts I thought, could you use another '317 to do the de-noise duty instead of the extra NPN? After all, from the outside the 317 is just a transistor with 1.2V Vbe on steroids ;-)
I love the idea, although it may be a bit circular: I need to think about it, for exactly three months + one week. It will then be perfectly ripe...
 
Does it need only BC337?
I made this circuit with BC547C and LM338 - and don't have good result. It works as usual, without extra PSRR. :(
I tried different R4, R5, R6 and different C2, C3, C4 (it is stable with C4 1 nF).
What voltage have to be on the Q1 collector? (I have about 3V). Thank you.
 
Last edited:
Does it need only BC337?
No, certainly not: I have tested it with various types, including a BC547 (don't remember the selection letter), and all worked, but the 337 gave a marginally lower noise
I made this circuit with BC547C and LM338 - and don't have good result. It works as usual, without extra PSRR. :(
I would look for dry solders or similar problems. The circuit in itself is very simple and straightforward, and if all the E-caps are OK and correctly connected, it should deliver the goods
I tried different R4, R5, R6 and different C2, C3, C4 (it is stable with C4 1 nF).
What voltage have to be on the Q1 collector? (I have about 3V). Thank you.
3V is OK: it should be around 1/4 of the output voltage, but as long as the transistor remains in it linear domain, it ought to work.
Try disconnecting C3, and observe the waveform at the collector of Q1 at the max sensitivity of the O-scope: you should see a lot of noise.

Note that I only tested it on a 317, but the 338 should work exactly the same.
There has to be some silly material error somewhere.

There is little benefit in making C4 as small as possible: in fact, you can use 10nF or 22nF, it will make the circuit more tolerant (but won't solve the issue of non-functionality)
 
I haven't done photos (I forget my phone at home, but my lab is at work). It is about 7x20 mm, with three 30-40 mm wires.
I tested it mostly at 12 V output, so 1/4 is 3 V - it is ok. Now, I have about 0.6 mV noise with or without de-noiser.
There is a DC/DC pre-reg (XL4016) before the LM338 - can it be a reason of high noise?
Thank you for help, will look for a silly mistake.
I think, I have to test this cirquit with linear power supply first (because dc/dc radiates hum).
 
Last edited:
I have now remeasured the output noise in good conditions: it is <0.3µV in a 10Hz to 10kHz bandwidth;
nothing particularly exceptional, but it all depends on the auxiliary transistor, an ordinary BC337 in this case.
The figure could be improved by using a really high performance, low noise transistor
 
Indeed, and thanks again for that.

As I took care to keep them in a safe place, separate from my main stock, I had some difficulty locating them.

In the end I found most of them.

The KSC1845 was disappointing, at 0.6µV
The KSC3503 is already better than the BC337 (which itself isn't bad at all): 0.25µV
The 2SC6043 manages 0.2µV and the ZTX851 0.16µV, but for those two, the actual value is certainly better: most of the noise appears to be mains-related, fundamental and harmonics.
The hum was already present with the BC337, but since it was not really dominant, it didn't bother me too much.
With better transistors, it becomes the main contributor.

The test-jig is breadboard-based, and this makes it susceptible to stray fields present in the lab.


I have also tested other devices: the mythical 2N4401, but they are disappointing: the NS manages 0.25µV and the Motorola 0.3µV, short of audiophile expectations.
A generic S8050 purchased from China did a little better: 0.22µV
 
D1 has no active role under normal circumstances: it just protects the fragile and all-important health of Q1's base-emitter junction for situations like power up or down, output shorts and similar mishaps.

It is absolutely non-critical.

C5 is equally non-critical: in fact, it is not part of the denoizator, it should already be present before you install it.
Note that I didn't draw it, but an input cap should also be present, and is much more important than C5; refer to the 317 datasheet for guidance, or simply leave the existing capacitor in place.

If you use the denoizator as a complete, autonomous regulator built from scratch rather than as an add-on, you can use whatever value that suits you, however very large values combined with very low ESR's can upset the stability of any 317/337 based regulator, including the modded one.

BTW, I already said it somewhere in the thread, but it is probably preferable to mention it again: C4 can and should be larger than 3.3nF (10nF~22nF), because the HF benefits of using a minimal cap are very small, but the risks involved with component variations, wiring etc. far outweigh this small benefit.

Q1 and it associated components will ~double the quiescent current consumption (it depends on the exact output voltage).

Note that I am currently developing and testing an autonomous superreg based on the denoizator principles.

It would not be easy and practical as an upgrade for existing regulators, but it offers ~20dB improvement compared to the denoizator (which is already 30dB better than an optimized 317 reg), and has a quiescent current similar to an unmodified 317, thus if your goal is to build a new, standalone regulator, I recommend you wait a week or two, until I finalize and test my design properly.

If you simply want to upgrade an existing regulator, the denoizator is still the way to go.
 
BTW, the denoizator is supposed to be used on a regular 317 regulator; the circuits you showed are already augmented, which in principle makes them incompatible with the denoizator: you are not supposed to add layer after layer, you should opt for the best solution alone.

The electronic filter will normally do no harm, but I am less sure about the circuit in post #85.

Of course, you can always dream that the goodness of each improvement will linearly add up, but in engineering that's not the way things usually behave
 
Here is the NoNoiser....

It is a standalone regulator derived from the Denoizator, and it is based on the same principles, but the performances are pushed to the maximum.

attachment.php


An additional transistor is used as a follower, to increase the loop gain and to ease the trade-offs regarding the coupling capacitors, amongst other things.

Its quiescent current consumption is ~the same as a native 317 regulator.

The performance improvement is purely AC, and the DC accuracy is even somewhat sacrificed to this end, but it remains acceptable, and for audio-oriented circuits AC performance is paramount.
The DC stability just needs to be sufficient, which it is for 99% of the cases.
The situation would be different for metrology-oriented applications, where both aspects are important.


The Nonoiser is a super regulator: compared to the archetypal Jung superreg, it brings improvement on all key aspects, but of course there is no great merit in that, as Jung's regulator was designed ~30years ago, when the technology was less advanced.
More interestingly, it also beats all modern alternatives: this page includes a survey of the main superegs available:
Super Regulator – diyAudio Store

How does it compare?

Let us begin with the PSRR, which is the weakest aspect of the NoNoiser: the measured value @100Hz is 125dB, 1 to 2dB better than the best regulator of the survey (Salas reg, fig. 5).
If a ripple compensation resistor is added (R7), it is possible to gain an additional 5dB.
I didn't attempt to refine it very much, but if both the phase and the magnitude of the error signal are tweaked, up to ~20dB should be attainable.
The sim predicts a much better PSRR: 140dB, probably because the 117 model does not include Early-type effects.

attachment.php


In the sim, any resistance value for R7 degrades this figure.

The output impedance @1kHz is 15µΩ, more than 5 times better than the best reg of the survey (Jung AD797, fig. 8).
The sim predicts a slightly better value, and it is probably possible to improve the reality too, because the Kelvin configuration I adopted is not optimal: I placed the drive and sense terminals side by side, but this results in a small mutual inductance between the two circuits, sufficient to degrade the impedance.

attachment.php


I should either have completely separated the two circuits, or have alternated/intertwined them to cancel the mutual inductance


With a ZTX851 as Q1, the total noise in a 10Hz to 10kHz bandwidth is 0.09µV.
At 1kHz, the noise density is 0.8nV/√Hz, more than 3 times better than the best reg of the survey (SuperTeddy, fig. 7).
If Q1 is a BC337, the total noise becomes 0.16µV, and with a Chinese S8050 0.13µV.

Test conditions:
Vout=12V, Iout=500mA, Vdropout (average): 4.5V, voltage stimulus for PSRR: 5Vpp, current stimulus for Zout: 100mA pp.
The 317 of the prototype is made by ST.

The compensation scheme I adopted is different from the Denoizator: it involves an inductor in the emitter of the main transistor.
The advantage compared to a parallel capacitor is a greatly increased open-loop bandwidth for a comparable stability.
The required value can be as small as 50nH (with a ZTX851), but the S8050 needs more than 500nH.
These values could vary, according to the layout, the ESR of the caps, etc.

Unlike the Denoizator, this regulator is not suitable as a retrofit: even if you are ready to carry out the mods, it can only work properly and deliver its performances with Kelvin connections: the input, output and sense terminals need to be completed independent: any common ground will ruin it and bring it to the level of an ordinary regulator.

The performances vary with the output current, but the variation is modest: the PSRR remains almost unchanged, the output impedance is stable too, except at very low currents where it increases a bit, and the noise is proportional to the output current, but the proportionality is not 1:1, it is much weaker.
This means that in practice it is usable from ~0 to 1500mA, just like an ordinary 317 reg.
 

Attachments

  • Nonoiser0.png
    Nonoiser0.png
    28.6 KB · Views: 7,872
  • NoNoiser.asc
    3.3 KB · Views: 306
  • Nonoiser4.png
    Nonoiser4.png
    90.2 KB · Views: 4,645
  • Nonoiser3.jpg
    Nonoiser3.jpg
    398.1 KB · Views: 626
  • Nonoiser2.jpg
    Nonoiser2.jpg
    303.5 KB · Views: 715
  • Nonoiser1.jpg
    Nonoiser1.jpg
    368.4 KB · Views: 2,148
  • Nonoiser5.png
    Nonoiser5.png
    80.1 KB · Views: 4,057
  • Like
Reactions: 1 user
I said earlier that the DC stability of the NoNoiser had been somewhat sacrificed, to favor AC performances.

The main issue is temperature stability, namely the thermal tracking between D1 and Q2.

Other aspects of the stability are quite good, even though R1 is ~10x the recommended datasheet value: thanks to the emitter follower, the stability is in fact better than with just a 240 ohm resistor.

Here is a plot of the output voltage vs. temperature: it is not perfect obviously, but it remains within acceptable standards for an audio-oriented regulator.

attachment.php


One more thing: by adding a bootstrap circuit to extract the maximum gain of the low-noise correction amplifier, it is possible to reach even lower noise levels.

This plot shows the noise reduction effect of the unmodified amplifier on the 317's internal noise: it is around 49dB for the audio range:

attachment.php


With the bootstrap circuit, it increases to 57dB:

attachment.php


This means that it should bring the noise density to ~0.4nV/√Hz, closer to the 0.25nV/√Hz intrinsic limit of the ZTX851 (the collector current might need to be increased too).
I didn't implement it, because IMHO 0.09µV of total noise is already good enough, and the added complication is not worth the trouble, but if others find it necessary, they can always use it.

If it is implemented, it would be good idea to reduce the 10 ohm base protection resistor to an even lower value, because at this level it will begin to have an impact
 

Attachments

  • Nonoiser6.png
    Nonoiser6.png
    73.1 KB · Views: 3,920
  • Nonoiser7.png
    Nonoiser7.png
    77.5 KB · Views: 3,904
  • Nonoiser8.png
    Nonoiser8.png
    82.7 KB · Views: 3,907
Last edited:
It depends very much on your situation: do you have bothersome noise/ripple in the output, and what it is the frequency range of the disturbance?

The advantage of a passive filter like yours is that it does not introduce noise of itself in its useful range; however, the reduction capacity for LF noise is more limited, unless you resort to really extreme values.

The denoisator also has the advantage of not degrading the internal resistance, in fact it improves it. Is it important for you?

You can combine the denoisator with a passive filter too: you will get some of the advantages of both.

If your main problem is just ripple rejection (100Hz/120Hz), the denoisator is going to perform better than your present passive filter.