Discrete regulators with denoiser

The higher voltage shouldn't be an issue. I'd have to test higher current configurations. I'll try to prepare a higher current testing setup.
This is just a sim but as a general example:

View attachment 1121362


It's at 36V/20A output. Seems it could be doable. I used some random high power MOSFET model just to see how it would behave.
Higher voltage should also work but at some point there's going to be heat buildup in certain parts of the circuit (clearly apart from the pass transistor). So might need to use lower current for the denoiser circuit, or upgrade the power rating of the BJTs.

edit: also in this configuration Q1 takes a big hit at startup with a high base current surge. A protection resistor will have to be added, output capacitance would have to be increased, C3's value lowered etc. It's just a general example only.
Seeing that -140dB line makes me happy 😀
 
Tried that IRFB3607PBF MOSFET for the negative voltage output version. 36V/100A output, seems to work with two in parallel. But I'm not sure about how large the source resistors should be. Input capacitors would need a 30mOhm total ESR, or lower if the source is lower than 10mOhm.
I might have made a mistake, never tried this before. No way of making a board for that kind of output current, I wouldn't be able to test it.
Different MOSFETs might improve performance a bit.
I attached a screenshot for +/- versions.
 

Attachments

  • Screenshot_20221223_140027.png
    Screenshot_20221223_140027.png
    39.3 KB · Views: 247
  • Like
Reactions: pbilous
Figured I'd update this thread as I designed a set of PCBs for the DC coupled denoiser.

PSRR.png


PCBs have RC/LC/FC input options, the LC version PSRR should be similar to this sim (with the parts I got):

PSRR C vs LC.png


Output impedance seems pretty good:

output_impedance.png


But having this on the PCB output is a bit of a challenge. I used 4mm wide traces, with 1 oz copper at around 4cm long traces their resistance comes out at ~5mOhm which is two orders of magnitude higher than the regulator output impedance (at 20kHz). So I cleared the solder mask and will try to solder 2mm diameter solid copper wire across the whole length of the traces. The wire should have around 200uOhm which is considerably lower.


PCB_dual.png


PCB_single.png


Ordered the various SMD parts and I'll do a measurement session along with the LM3x7 boards from the Denoisator thread. Will update with measurements once I'm done.
 
Tried it with the LM3x7 board (same 4mm wide output traces), I had to scrape off the solder mask as I didn't remove it for that design. Bit of a pain but worked out fine.
I used 6mm^2 section wire, I think it's between AWG 9 and 10, some 2.76mm diameter. The cable datasheet showed 3.08 Ohm per km which is a bit higher than pure copper (2.80 Ohm/km), so 1.1x multiplier compared to pure copper. So 4cm length should work out to about 123 uOhm.
I do have a thicker 10mm^2 section wire (unused piece in photos) which should work out at 74uOhm for 4cm. But not sure it fits. I'll have to also hammer it into a rectangular shape, maybe it fits.

edit: should I enlarge the output pads holes? they're 2mm atm
 

Attachments

  • 2.png
    2.png
    362.1 KB · Views: 110
  • 1.png
    1.png
    1.5 MB · Views: 1,599
Last edited:
  • Like
Reactions: stv
The solder wick should be simpler to work with but how much does it lower the resistance? Seems like a nice simple way to add some current carrying capability to smaller traces. But the extra solder doesn't do that much by comparison.

Be careful: the deformation will cause strain-hardening and increase the resistivity unless you anneal it afterwards, thus it might be counter productive
Interesting, I did not know that!
 
Yeah the 10mm^2 copper rod can be hammered to fit the 4mm wide trace. But just. Solder seems to wick underneath it on the trace, from first attempt experience, so using it should work out at ~80uOhm resistance, at 4cm length.
But I'm not sure it's worth the trouble since there should be low output impedance at PCB output, that's where I'm sensing with denoiser.
For DC 4mm wide tracks should already be enough. MOSFETs are some 8A rated (continuous) so they should be able to do 5A but limiting factor is heat in this case. Maybe 5A at 0.2Vdrop but input trace is just 1.5mm wide. I'd think 2A output should be fine for some 1W of dissipated heat? Output traces will work as heatsinks, and I also added some extra holes across them to help with air flow.
I'd keep solder mask off the +/-V output traces
PCB.png
 
Also for some reason tempco simulation is very good compared to DC-coupled LM317 Denoisator. Tried with various transistors and results are similar. Expected it to be worse, but I'd have to test it in reality, I do have a temperature controlled oven.
 

Attachments

  • tempco.png
    tempco.png
    74.7 KB · Views: 77
I was trying to figure out if I can improve cooling even more and I remembered I have some small heatsink from my 3D printer stepper motor drivers. It does barely fit on top of the MOSFETs

v1_1.png

v1_2.png

But then I noticed their application on the driver PCB and figured I could do the same.

open_graph_image.jpg
HEAT-BLUE-1.jpg


They moved the IC on the bottom of the PCB and used vias to transfer the heat on top of the PCB, where it goes into the heatsink. The heatsink has good surface for its small size.
The driver IC normally dissipates around 2.6W as per datasheet, but also has a high operating temperature rating, 125C.

TMC2130.png


Trying the same trick on my PCB design results in this:

v2_1.png

v2_2.png


I had to increase the board width by 3mm to have a more comfortable fit, so the board is 50mm x 53mm this way.
The heatsinks and the exposed output traces should allow for maybe 2W dissipated heat, comfortably. Hopefully.
 
I'm trying to understand why the tempco of this latest incarnation of circuit looks way better than the AC-coupled denoiser version. And apart from other interactions, the fact that I modified the startup circuit seems to help.
Looking at the normal regulator without the denoising circuit, the new startup circuit seems to make for lower Vout variation due to temperature changes.
1.png


D3 can affect Vout while D1 doesn't do anything after startup.
It also isolates the two nets, and helps with PSRR
2.png


And this matters even more if we use the two Zeners back to back as Vref
3.png
4.png


There is a cost though, at least for the standard application there's one extra resistor that's being added to the circuit, and there's around 1mA flowing through that resistor chain, at all times. But for me it's totally worth the cost just for the better temperature variation performance. Let alone the boost in PSRR.
But even with the new startup circuit, there's still a lot of difference between the DC coupled and AC coupled denoising circuit. My best guess atm is that Vref tempco works against denoiser BJT tempco and they sort of cancel out.
5.png


edit:
For example in the DC-coupled versions using a blue LED as Vref has tighter Vout across temperature than using the two Zeners back to back which makes me think that LED tempco works against BJT tempco. And if it works like this in practice, that would make the blue LED the superior Vref for the DC-coupled circuit.

6.png

And there doesn't seem to be any difference between PSRR performance in both LED or Zener back to back cases. If anything the Zeners need a bit more current.
7.png


The simulations are for both 25C and 75C, so 50C span between same colored traces.
 
Last edited:
I have successfully used the LED of a peculiar optocoupler (6N139) as a Vbe compensation:
This opto uses a visible LED, unlike regular ones, and works nicely to compensate the Vbe (that's reality: sim might be different)
 
  • Thank You
Reactions: Trileru