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

It's the wire loop. I tried replacing with some small resistance thinking it's its resistance but didn't work. I posted about it three years ago in post #1,938.
edit: maybe 15-22nH?

Screenshot from 2021-01-06 17-17-59.jpg
 
With due respect for the impressive improvements, for which I count 6 resistors, two electrolytics, one diode and two transistors, a total of 11 components.

Maybe it has been asked before, but how does this compare to placing a LDO instead behind the LM317, like f.i. the LT3085.
You will only need the LDO, one resistor and one or optionally two ceramic caps, so just 3 or 4 components instead of 11.

When more current than 500 mA is needed, you can just parrallel more LT3085’s without adding any additional component.

Hans
 
I only added exactly three extra parts on top of the original Denoisator, one BJT and the RC filter. If you would install a status LED anyway. The RC tames the startup spike, quite well from my testing, and the BJT adds some dB of PSRR and lowers the noisefloor even more. Also seems to improve output impedance a bit. For me all three parts are worth it for what they bring on top of Denoisator.
Regarding performance, I have measured the noisefloor of both Denoisator and NoNoiser and I don't know of any other supply that comes close to them (especially NoNoiser). There may be some, but I don't know of any, if anyone can reference a similar performance other one I'd be curious to see it. I mean for the noise performance, PSRR is not stellar but still pretty decent, especially if using higher values R1/R2. Output impedance is under 10uOhms up to 1kHz. And all the extra parts are pretty cheap.
 
Maybe it has been asked before, but how does this compare to placing a LDO instead behind the LM317, like f.i. the LT3085.
You will only need the LDO, one resistor and one or optionally two ceramic caps, so just 3 or 4 components instead of 11.
It will push the PSRR down to unmeasurable levels, but it will eat up a few hundreds of mV (it would be unwise to use it at its absolute dropout floor) and the residual noise will be higher: with the suitable transistors, the x-noisers can achieve noise densities <2nV/sqrtHz, and even <1nV/sqrtHz for two-transistor types.
I didn't check, but the output impedance of the x-noisers is probably much lower too
 
O.K. I understand.
Instead of the LT3085 a LT3045 could be used instead.
It has 2nV/rtHz and is also parrallelable without adding extra components if more than 500mA is needed and is stable at all loads.

To compensate for the small 260mV LDO voltage drop, the divider on the LM317 could be slightly adjusted.

Nevertheless, as mentioned before, I’m impressed to what amount a mediocre LM317 has been improved by smart DIY circuitry.

Hans
 
O.K. I understand.
Instead of the LT3085 a LT3045 could be used instead.
It has 2nV/rtHz and is also parrallelable without adding extra components if more than 500mA is needed and is stable at all loads.

To compensate for the small 260mV LDO voltage drop, the divider on the LM317 could be slightly adjusted.

Nevertheless, as mentioned before, I’m impressed to what amount a mediocre LM317 has been improved by smart DIY circuitry.

Hans
Pity it doesn't come in a negative output version. For a +/- V power you would need a transformer with two separate output windings and two positive LT3045 regulators, and connect them to get the +/- voltage.
 
From issues about negative psu regulators versus positive......I definatly ruled out negative.
As I am concerned, when the need comes for +V 0 -V , the way to go is two identical positive regulators and dual transformer windings ( instead of center tab ).
From an engineer point of view this means one design ( instead of another troublesome one ) and easier sourcing.
IMO negative psu regulators design is a waste of time. Only for the look of a complementary schematic for stuff that is not really complementary.
A waste of time for a cosmetic result.
 
Last edited:
Do you think there are true complementary BJTs ?
Actually they are aproximately complements, designed with compromises and by means of degrading the performance of the NPN one.
Complementary pairs is more a commercial claim than a technical reality.

There are reasons why PNP and NPN are not born equal.
Check this with solid state engineers specialized in semi-conductors ( I am not ).
 
As a result of my tireless search for improvements to the originality of the magical circuit designed by Elvee, I found, after convincing myself that not much more could be done, a solution that allows us to go far beyond everything that has happened. What I present does not reduce the current supply capacity from the output of the regulator. There is also no noticeable output voltage overshoot (limited by the action of the multiplier implemented by R5, C2 and Q2), which could limit its use with delicate circuits. It can be wired with either 3 or 4 conductors. It doesn't necessarily have to be connected to the power supply of the circuit itself you are trying to improve, as the noise eliminator could have its own small power supply. The eliminator is likely to behave stable, since transistor Q1 would operate outside the output of the regulator, avoiding oscillations in the voltage of the node 5 (for example) or other specific nodes. Nothing would prevent the implementation of CFP variants in the same way, such as the Dienoiser or the most recent buffered one that I presented.

Circuit.jpg


PSRR.jpg


Output Transient.jpg


Output Transient (zoom in).jpg


IB Q1 Transient.jpg


P Q1 Transient.jpg


Parameters.jpg


Best regards
 
Last edited:
Not to take away from Diego's post but I ran some AC measurements for the Denoisator with R1=3.3K, both LM317 and LM337. And performance is phenomenal!

One thing I was curious about was testing varying Vce values for denoiser and strangely enough it doesn't seem to matter, 3V or 9V noise and PSRR is the same. This makes things simpler, you can set Vce of denoiser BJT to some half Vout and it'll be fine. This also means you can use any LED color for CCS, IR to UV if you want.
Another advantage of having a lower Vce is lower startup spike. Vce at 6-7V with the filter makes for virtually no startup spike.

startup.png


I have measured this and behaves like simulation predicts. 2.2uF filter cap worked nice for both 5Vout and 12Vout. Only the resistor had to change. 2.2uF caps are small, I will test for microphonics with a ceramic cap, forgot to do it this measurement session. The filter shouldn't take a lot of PCB space and is worth it. Without it it still spikes pretty high even with Vce at half Vout.

PSRR is extremely strong, 135dB for LM317 and 130dB for LM337. This is with LM3x7-N versions so I expect few dB lower with "regular" LM3x7. I'll test this since I made an addon board that I will also measure. Will use regular LM3x7.

Noise is not as good as NoNoiser but it's getting close to it. Noisefloor has a bow shape, I think it's due to Denoisator topology. NoNoiser is flatter. Total noise in audio spectrum is 170-180nV for both LM317 and LM337, but with MPSA06/56 pair.
I have tried BC807/BC817 but they are slightly higher noise, MMBTA06/56 (SOT-23 version of MPSA06/56) have lower noise, same PSRR. So I'll use them from now on for LM3x7 Denoisator.

Another interesting thing is there's no real difference in performance between 5Vout and 12Vout. Maybe 1dB lower PSRR for 5V vs 12V.


PSRR for 12Vout with MMBTA06

12V_MMBTA06_PSRR.png

For 5Vout with MMBTA06

5V_MMBTA06_PSRR.png


-12Vout with MMBTA56

-12V_MMBTA56_PSRR.png


-5Vout with MMBTA56

-5V_MMBTA56_PSRR.png

Upper trace is measurement at input of regulator, lower trace is on the output, with LNA. Difference between them at 105Hz is the PSRR at that frequency.
Still around 100dB of PSRR at 10kHz.

Noise measurements

12Vout with MMBTA06 and BC817

12V_MMBTA06-BC817_noise.png


The difference is not a lot but MMBTA06 goes under 1nV/sqrtHz level. That's pretty good for just Denoisator topology.

5Vout with MMBTA06

5V_MMBTA06_noise.png


-12Vout with MMBTA56

-12V_MMBTA56_noise.png


-5Vout with MMBTA56

-5V_MMBTA56_noise.png


Again, no real difference between 5Vout and 12Vout.

And some measurements for measuring gear itself:

Grounded ADC input with 1000x gain

grounded_ADC.png


Grounded LNA input

grounded_LNA.png


Tweaked the LNA circuit and got it to lower noise, similar to 4.7R resistor, but there is some LF extra stuff in there.

4.7R resistor measured with LNA

4.7R_measurement.png

So calibration should be pretty close.

I have done some stability tests and both LM317 and LM337 seem fine. I did use two 220uF/35V caps in parallel for both rails, same capacitors. Model number EDH227M035A9MAA from KEMET. I only added 0.2R in series with them for LM317, I didn't add anything for LM337, not even the inductors. Seemed to work fine just with two of these.

They look like this

output_caps.jpeg


I measured their ESR in circuit and I measured 0.3R for LM317 and 0.1R for LM337 and it makes sense since I added 0.2R in series with the LM317 ones. With these LM337 Denoisator seems stable without extra inductance/resistance.
Tried adding 0.2R and 22nH-68nH and AC looked bad, higher noisefloor, worse PSRR performance. Without any of them AC performance is real good and seems stable even under load and poking at it. Recovers quite fast.

This is gif of the LM337 output with LNA, I'm poking at denoiser BJT base with metal tweezers, three times, you can see the recovery, hope the gif works here.
I used no averaging to make things clearer, this is the noisefloor of LM337:

LM337.gif



The exact part values I used in measurements for both LM3x7 are these, for +/-5V and +/-12V.

capture-2024-11-21_22-11-1732220921.png


I might make a new NoNoiser run to compare results but it won't be too far. R1 at 3.3K has pretty good performance.
I've also added an archive with REW mdat files for measurements. You can open them in REW and check values.

edit: since it's Denoisator then online LM317 calculators should work for figuring out R2 value for needed Vout, with R1=3.3K. So all are standard values apart from LM3x7 R2 value. If you change LED color you need to adjust the 120R to keep 8mA through the denoiser BJT. 1.7-1.8Vdrop needs 120R.
 

Attachments

Last edited:
Datasheet max value for R1 is 1.2K, which is still a lot higher than the 120R-240R that's usually used. Low values help with the 10mA minimum current. That's not an issue with the denoiser since it uses about that much for itself.

LM317.png

This is from a LM317 datasheet, can't remember which make. I might test with R1=1.2K just to see performance out of curiosity.
 
Again, R1=1.2K is a datasheet value. It's not out of spec as it were.
Here's 4 gain stages, classic Denoisator with R1=220R, one with R1=1.2K and two more of the same but with CCS. The improvements are pretty clear, gain seen at ADJ pin. You approving or not doesn't change that. Pretty clear from measurements.

comparison.png
 
Last edited:
Test as you like.
I need no test to find what does the R1 value on: Output voltage accuracy and loop gain that changes: Stability, output impedance, PSRR and load regulation.
The performance means nothing. It is a matter of compromises.
Looks pretty simple to me also, the denoiser is in parallel with R2 and will create a resisitance less than R2, pulling more current and fixing "the problem."
 
CCS in schematic at post #3052.

The difference it makes at 100 Hz 1000 Hz on PSRR
R1=220 Ohm
Q1 collector load = CCS dynamic impedance = 1 Meg Ohm ( typical )
No CCS Q1 collector load = 1.8 K Ohm ( typical )

To make it simple ( but not simplistic.)
220 // 1000000 VERSUS 220 // 1800

About Zilch !

PSRR at 100 Hz 1000 Hz ......+ 1.14 dB
mchambin's issue was the low R2 value with R1=220, and I was curious what he'll say with higher values of R2 but he still doesn't approve. Which is strange. I'm not even sure what he's trying to do here, these denoisers are clearly not a datasheet application. We all know that.