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

Managed to etch the die/denoiser version and found some hardware and parts. Need to search for resistor values and see if it's working. Also found a 100uF cap for output that measured around 0.210 ohm for ESR. Sadly I'll only be able to check the board with a multimeter. I'm thinking of testing at 1A and use a series 1ohm resistor and test the sense lines this way. See if I get the set voltage at the sense lines connection. I'll use 10cm of wire or so.

Board is a bit finicky around the bjt traces, and I'd remove the BCE copper text before etching, complicates soldering in that spot. Also I swapped BC327 around in the last revision and E and C are the other way around, as in the text. So better to ignore the copper text for BCE.
H95N1Ds.png

UIwD5j5.png

WE87mKb.png
 
Last edited:
I managed to test it and it seems to be working, at least as a LM317 reg. I don't have the tools to measure the performance. I just realized the sense lines are not for setting the output voltage at the point of load.
I used BC550/BC560 and I injected DC in the R from the CRC filter. So I didn't test the rectifying bridge but hopefully I got it right. I tested at 5V/500mA load and for 30 minutes the output voltage seemed stable.
I didn't notice any voltage oscillation at startup. I remember some people noticed some voltage swings for some seconds. I'll test more tonight, check voltages at different points in the denoiser circuit.
Any way I can inject some 1kHz AC and check the output? Would something like the LTSpice schematics work, inject into the ADJ pin? Multimeter would pick it up on the output.
edit: I just realized I stacked two 100nF caps instead of 10nF for compensation cap. Might have something to do with it.
2nd edit: yeah couldn't wait, I replaced the sandwiched 2x100nF with 2x10nF caps and now the denoiser seems to be working. At least I get the swings at startup. For a 4.97V nominal output highest I had 5.28V but happens only on the first swing. Subsequent swings decrease in amplitude to nominal output voltage. Cycled it about 10 times and that's as high as it went. So it seems to be safe for 5.5V max recommended voltage/ 6V absolute maximum circuits. It seems anyway, ymmv. Use this at your own risk as always.

fcEjkjq.png

GLP3baS.png
 
Last edited:
Someone mentioned that at 15V it went to 16-16.1V max, so that seems to be consistent with my findings. Seems it goes up about x1.06-x1.07 output voltage. So for 25V output that would be around 26.75V on the first swing. I'm curious about this, I'll test it tonight.
 
I think I found the culprit. Whoever designed the LM317 power supply (acting as a DAC pre-regulator) was following the latest diy-trend of that time (about 2005) and used ground power planes. These apparently act as antennas that pickup small noise artefacts. I made a quick&dirty new LM317 regulator (without any ground planes) and the spikes were gone. Here is the new graph.
Planes are dangerous: they give a (mostly) false sense of security, but in fact you control nothing at all.
A sensibly thought out perfboard layout can be vastly superior. Planes are good for electrostatic shielding (when used properly)
 
Planes are dangerous: they give a (mostly) false sense of security, but in fact you control nothing at all.
A sensibly thought out perfboard layout can be vastly superior. Planes are good for electrostatic shielding (when used properly)
Hmmmm....that is an interesting statement. I've always been led to believe that ground planes were a MUST in circuit board design. Please explain why they "control nothing at all".
 
How critical is the 220uF value from the adj pin? My Magni 3 seems to be using a LM317/LM337 combo (TO-263/D2PAK smd) for power supply. There also seems to be a small ceramic cap on the ADJ pin, no large electro. Seems like some 100nF job. Also has large 1000uF on the output (paralleled with I guess 100nF 0805).
If I'd want to add a denoiser the space there is really at a premium. The smd add-on would fit if I used tantalum caps. I could parallel 2x100uF but I'd go lower if possible. With smd tantalum I probably could make it work.
I'd also need to replace the output caps with something around 100uF but that isn't the issue.
I tried the simulation with 100uF for C3 but didn't notice much change.
Also are there ESR requirements for C3? I see the LTSpice sim has 0.6R added as ESR for that cap.
 
Last edited:
Planes are dangerous: they give a (mostly) false sense of security, but in fact you control nothing at all.
A sensibly thought out perfboard layout can be vastly superior. Planes are good for electrostatic shielding (when used properly)

Completely agreed with this: planes, at least in audio level frequencies, are not necessary. An exception might be very low level signal circuits, and even those that are unbalanced. The problem is digital circuits, which do radiate RFI through the ground, and digital ground should be separated from analog ground.

One interesting thing Jan Didden said on the Super Regulator thread that he abandoned the ground plane he had used on his first design of the pcb for Jung's regulator, as not providing any benefits.

What still is very effective is separating grounds inside a regulator from those coming from the main input filter.
 
Is this valid for 4 layers boards as well? Where the 2 internal layers are ground and no other traces on them?

Hmmmm....that is an interesting statement. I've always been led to believe that ground planes were a MUST in circuit board design. Please explain why they "control nothing at all".
Some reminders: in our universe, nothing can be equipotential under dynamic conditions, not even a superconducting plane.
This applies to PCB planes.

Two main rules apply to the distribution of AC currents sent through planes:

One is the most direct/lowest impedance path (equivalent, because the most direct path has the least inductance)

The other is the mutual attraction of the forward and return current paths to minimize the enclosed area.

For example, if on a board you have a dynamic load and a main bypass cap connected through two planes, but physically distant, the currents will follow a straight line from the load to cap, but they will tend to superpose one another for much of the path.
In fact, the two planes will boil down to two tracks and the rest of the copper will play no role.
If other signals share or are coupled to this path, they will be subjected to Xtalk, just as with simple tracks.

If a plane is not subjected to a varying current and is used as a reference or an electrostatic shield, then it will appear ~equipotential
 
Would it matter for a LM317+denoiser/nonoiser + groundplane if the supply is delivering the current through wires to another board? If the compensation cap is close to the bjt, how could a groundplane create problems locally on the lm317 board? For the whole board there's only one current draw on the wire that goes out.

edit: so hypothetically if designing a lm317+denoiser and groundplane, one would do best to keep the denoiser circuit out of the high current path between the point of connection of the output voltage wire and first cap after rectifying bridge?
 
Last edited:
Lowering the 220µ will bring the input and output poles closer together, thus increasing the VLF peaking.
With 220µ, it is just tolerable, but 330µ or 470µ would be even better.
The input cap could be reduced, but the LF correction would be diminished
 
I tried to measure the noise using the phono input of my Marantz 4230. Datasheet has a stated gain of 40dB but I measured around 48.8dB using a tone generator from my phone, for 100Hz (to rule out gain variation between different frequencies). Since best chances to see anything is the rectifying bridge ripple I wanted to test at 100Hz.
The lm317 was loaded at 500mA for 5V output. I used a 1uF cap on the output to the amp since the phono preamp has an input impedance of around 40K so should let 100Hz without attenuation. I used one of those cheap red pcb oscilloscopes that are rated for audio frequency only. With the 220uF adj cap removed from the circuit I measured around 80mV rms 100Hz ripple on the output of the phono preamp. With a gain of around 48.8dB that means around 290uV 100Hz ripple on the output of plain LM317 setup (no adj cap). Input ripple to lm317 measured at around 484mV rms. That means that the lm317 has around 64.4dB rejection which is pretty consistent with the datasheet.
I next installed a 22uF cap between adj and ground and measured 24mV rms on the phono output. That translates to 87uV 100Hz ripple on the output of lm317, and around 74.5dB which is a bit better than datasheet but thereabouts.
I added the 220uF cap from adj to denoiser, shorted R3 and measured almost nothing on the output of phono. There seems to be a smaller 100Hz ripple under the 50Hz ripple that I couldn't minimize. I had no case/shield for the lm317 board.

jIqolFX.png


There seems to be like around 5-6mVpp? Hard to tell, and at this level I really question the measurements. Phono amp might not be able to properly pick up such a low level signal and the oscilloscope might not be able to accurately measure it. It should translate to around 7.6uV 100Hz ripple output with denoiser. Which is around 96dB.
I then removed the short across R3 for dienoiser version and got nothing visible on the oscilloscope. Whatever is there is 110dB+ and no way of seeing it.
5jg6Pf9.png


40dB LNA doesn't do much for analyzing de/dienoiser noise. Maybe with a better oscilloscope that can go 1mV/div and lower, or a decent ADC.
So apparently my board is working, no telling if there's any higher level oscillations, as I simply don't have the gear to see anything relevant.
One important issue that I discovered, while using the 22uF on adj to ground, was that initially I had a higher 100Hz ripple after adding the cap. That left me scratching my head for some time as I couldn't figure out what the problem was. Then it hit me, the ground trace that I used, is the one that is separate and comes from the first cap from CRC filter. I switched the cap ground to the other side of the board and then got lower ripple. So this is one problem that the board has, and will rectify it in the future. I drew that trace for those damned 100nF caps that people like to use on everything. I don't use it but it bit me in the *** while testing as I used it for the adj cap (was closest to adj pin). So that trace is not to be used for bypass caps. Will inject ripple.
XDfhOsl.png


All these measurements should be taken with a grain of salt as the phono amp might have varying gains at different input levels. It was more to assure me that the circuit works.
I deliberately shorted the resistor from the input CRC as I wanted a higher level ripple into lm317 so I had a chance to measure anything. That resistor drops the ripple by almost 10dB at the voltage/current I used.

Measuring a Nonoiser is out of the question with the gear that I have now.
 
Last edited: