That's interesting! Checking the optocoupler's datasheet I see that Vdrop of that LED is 1.3V at 1.6mA. That's a bit low for this application. I can only use an IR LED in the CCS which is some 1.1-1.2Vdrop. Vref needs to be around 1.1V+Vbe at minimum. Or else it doesn't work as good. So minimum some 1.7Vdrop across Vref, makes it red I suppose, but I'll also try with green and blue, see what happens.
sorry, off topic question: what circuit design software do you use? the 3d renders look great, i use altium but the 3d renders are pretty bad
Last edited:
Please post in English.
sorry, off topic question: what circuit design software do you use? the 3d renders look great, i use altium but the 3d renders are pretty bad
It's a Kicad addon from the official repositories, named pcb2blender. It exports the board in certain format.
But it also has a Blender addon for importing them in Blender. Does add a few nice things, solder on pads, has predefined colors and textures for various materials.
Found it helps me notice some issues with the board, having correct size for all parts. Especially for boards this small.
There's also lighting which is a full topic in its own right, I just add a few lights above and rotate the board until I find some interesting view but I usually don't bother with it.
At least from simulation seems I could adjust current through the denoiser BJT for lower temperature variation.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)
Left side R4 CCS current setting resistor is 120R and right side R12 is 100R, AC performance doesn't change much, if at all.
Ran the simulation for 25C and 75C.
Figured out why sim was showing good tempco, I was using simple LED models. Using real LED models from LTSpice changes things.
Left side without compensation looks way worse (and about what I measured in reality, over 1V variation 30C-75C), and compensation arrangement seems to work for right side circuit.
I tested it in practice, made a test board and baked it a few times, managed to get the variation down to ~600mV but I think I can get it spot on.
AC doesn't seem affected
(in sim only atm) I lowered the current through Vref, about 1.5mA same as CCS LED. But doubled the current through the CCS.
LED models I used for sim are a white LED for Vref, has the right Vdrop, and CCS LED is still a red one.
Left side without compensation looks way worse (and about what I measured in reality, over 1V variation 30C-75C), and compensation arrangement seems to work for right side circuit.
I tested it in practice, made a test board and baked it a few times, managed to get the variation down to ~600mV but I think I can get it spot on.
AC doesn't seem affected
(in sim only atm) I lowered the current through Vref, about 1.5mA same as CCS LED. But doubled the current through the CCS.
LED models I used for sim are a white LED for Vref, has the right Vdrop, and CCS LED is still a red one.
Doubled the current through the CCS, about 19mA total, 6.5mA through the denoiser BJT and 12.5mA through the extra resistor. This makes for little temperature variation, 20-30mV 30C-75C. And some of the semiconductors should be thermally coupled. Or wait a bit so they equalize.
The diode I previously mentioned is not required but helps with measuring the potentimeter in circuit while testing.
It's not ideal but it works. By comparison with LM317 version I do have extra bits that I can poke at, with this circuit. I'll try different LED combos, maybe add another in series for Vref. Hope I can minimize this extra wasted current, although 12-13 mA wasted is not a lot.
edit: IR + red LEDs combo works better. Red + blue was really bad for compensation.
The diode I previously mentioned is not required but helps with measuring the potentimeter in circuit while testing.
It's not ideal but it works. By comparison with LM317 version I do have extra bits that I can poke at, with this circuit. I'll try different LED combos, maybe add another in series for Vref. Hope I can minimize this extra wasted current, although 12-13 mA wasted is not a lot.
edit: IR + red LEDs combo works better. Red + blue was really bad for compensation.
I think I found a way to reduce that extra current from compensation resistor:
I added D6 which seriously eats into the overall negative temperature coefficient of the circuit, allowing for way less current in R9 to get it pretty spot on between 30C-75C. This way only about 2-3mA are required while 7-8mA go in Q3. Which is acceptable for me.
D4 speeds up startup so it stays, worth the part cost for what it does. Its negative tempco turns Vout more positive so it helps a bit this way, takes some of the load off R9.
C5 is required for AC but don't go over that value or you'll have an overshoot on startup. 0.47uF seems like a good value.
In my case R9 was ~5Kohm and I could probably go even higher, I suspect a standard value resistor would work just fine. I tried with BC3x7 and MPSAx6 pairs and the compensation mechanism seems somewhat BJT agnostic. A pot would allow you to get it spot on.
Best LED combo is red in CCS with white as Vref which has to have a minimum Vdrop.
Vref >= Vf D5+ Vf D6 + Vf D4 + Vbe Q1 which works out to about 2.4V. So I went with white. I used a plain 5mm white LED. Measured 2.6Vdrop across it in circuit. Tried IR in CCS but didn't work so good with D6, I think tempco was flipped, heavily positive overall, I had no mechanism to compensate it like that. But I might have had a Schottky for D6. Red LED is good anyway.
I have only tested it for 12Vout but 5V-30Vout seems possible from simulation. 5V even seems to have the higher PSRR.
Didn't yet test AC that's another headache for another day but simulation doesn't predict much loss with the compensation mechanism. With and without the compensation arrangement:
So in the end I did need the 1N4148 diode but a total of two extra parts for compensation are totally worth it for me.
I added D6 which seriously eats into the overall negative temperature coefficient of the circuit, allowing for way less current in R9 to get it pretty spot on between 30C-75C. This way only about 2-3mA are required while 7-8mA go in Q3. Which is acceptable for me.
D4 speeds up startup so it stays, worth the part cost for what it does. Its negative tempco turns Vout more positive so it helps a bit this way, takes some of the load off R9.
C5 is required for AC but don't go over that value or you'll have an overshoot on startup. 0.47uF seems like a good value.
In my case R9 was ~5Kohm and I could probably go even higher, I suspect a standard value resistor would work just fine. I tried with BC3x7 and MPSAx6 pairs and the compensation mechanism seems somewhat BJT agnostic. A pot would allow you to get it spot on.
Best LED combo is red in CCS with white as Vref which has to have a minimum Vdrop.
Vref >= Vf D5+ Vf D6 + Vf D4 + Vbe Q1 which works out to about 2.4V. So I went with white. I used a plain 5mm white LED. Measured 2.6Vdrop across it in circuit. Tried IR in CCS but didn't work so good with D6, I think tempco was flipped, heavily positive overall, I had no mechanism to compensate it like that. But I might have had a Schottky for D6. Red LED is good anyway.
I have only tested it for 12Vout but 5V-30Vout seems possible from simulation. 5V even seems to have the higher PSRR.
Didn't yet test AC that's another headache for another day but simulation doesn't predict much loss with the compensation mechanism. With and without the compensation arrangement:
So in the end I did need the 1N4148 diode but a total of two extra parts for compensation are totally worth it for me.
As user mchambin pointed out in the Denoisator thread, voltage regulation is not great so AC coupling the denoiser seems like a better idea with less headaches.
I think D6 is temperature compensation for Q1 Vbe. Current through D4/D7 Zeners can be tweaked from R1 for some tempco adjustment although it looks pretty good in simulation.
This version only needs one pot to trim Vout and that's it. Apart from it the other parts can be standard values.
I think D6 is temperature compensation for Q1 Vbe. Current through D4/D7 Zeners can be tweaked from R1 for some tempco adjustment although it looks pretty good in simulation.
This version only needs one pot to trim Vout and that's it. Apart from it the other parts can be standard values.
Made a test PCB and assembled the negative version for this circuit:
Seems stable with and without load, even poking at its innards. Tempco is under 300mV for 50C span (with values in photo). Maybe I can get it tighter, I'll try with LEDs as well.
Seems stable with and without load, even poking at its innards. Tempco is under 300mV for 50C span (with values in photo). Maybe I can get it tighter, I'll try with LEDs as well.
Managed to get it to negative 30mV (edit: actually positive as it's a negative voltage circuit) across 50C, at 12.5Vout, with three 1N4148 diodes in series.
A simplified schematic since the denoiser doesn't do anything to tempco
Should be fine to 15Vout where it should be positive 50-80mV (edit: more negative as it's negative voltage). For higher Vout some other higher voltage Zeners might be required to keep the tempco variation low. For lower Vout I should be able to drop from the diodes one by one as I go down in voltage but 5Vout is not possible with these Zener values. LEDs as Vref might be a bigger tempco headache but I'll try it. They're a bit lower noise.
If you look at the datasheet for these Zener diodes you'll see the 6.2V version has positive tempco, 0.4-3.7mV/k at 5mA.
https://www.onsemi.com/download/data-sheet/pdf/bzx84c2v4lt1-d.pdf
Higher voltage ones, like 12V for example, have (positive) 6-10mV/k tempco. That might help for 20-30Vout for regulator but should be tested. I only have the 6.2V ones.
A simplified schematic since the denoiser doesn't do anything to tempco
Should be fine to 15Vout where it should be positive 50-80mV (edit: more negative as it's negative voltage). For higher Vout some other higher voltage Zeners might be required to keep the tempco variation low. For lower Vout I should be able to drop from the diodes one by one as I go down in voltage but 5Vout is not possible with these Zener values. LEDs as Vref might be a bigger tempco headache but I'll try it. They're a bit lower noise.
If you look at the datasheet for these Zener diodes you'll see the 6.2V version has positive tempco, 0.4-3.7mV/k at 5mA.
https://www.onsemi.com/download/data-sheet/pdf/bzx84c2v4lt1-d.pdf
Higher voltage ones, like 12V for example, have (positive) 6-10mV/k tempco. That might help for 20-30Vout for regulator but should be tested. I only have the 6.2V ones.
Last edited:
I found a better temperature compensation setup but I don't really like it:
It's clear the positive tempco of the Zener doesn't help, it should have negative tempco so I can eliminate the other diodes. The second reversed Zener is not needed here. So probably a lower voltage Zener from same series could work better, they exhibit negative tempco.
But what worked same as good was just a regular 5mm red LED, it's within 40mV drift between 25C-75C.
This works very good, has even a bit lower noise. No extra compensation diodes.
The circuit that I actually measured is this (-12Vout):
PSRR at 105Hz is a bit over 130dB:
And noisefloor:
This contains my LNA noise as well, which is this:
So subtracting it from the measurements leaves the power supply at a bit under 80nV total noise in audio spectrum.
This is the negative voltage output version, I'll test the positive version as well.
It's clear the positive tempco of the Zener doesn't help, it should have negative tempco so I can eliminate the other diodes. The second reversed Zener is not needed here. So probably a lower voltage Zener from same series could work better, they exhibit negative tempco.
But what worked same as good was just a regular 5mm red LED, it's within 40mV drift between 25C-75C.
This works very good, has even a bit lower noise. No extra compensation diodes.
The circuit that I actually measured is this (-12Vout):
PSRR at 105Hz is a bit over 130dB:
And noisefloor:
This contains my LNA noise as well, which is this:
So subtracting it from the measurements leaves the power supply at a bit under 80nV total noise in audio spectrum.
This is the negative voltage output version, I'll test the positive version as well.
Tested the positive version, tempco is same, 45mV drift 25C-75C with just a red LED.
I have measured with these values/parts:
PSRR for positive output version came out pretty strong, 139dB:
Noise was a touch higher compared to negative output version, 116.5nV in audio spectrum:
Pretty happy with the results!
I have measured with these values/parts:
PSRR for positive output version came out pretty strong, 139dB:
Noise was a touch higher compared to negative output version, 116.5nV in audio spectrum:
Pretty happy with the results!
Nice work Trileru!
I’d like to try it, has there been any thinking of a group buy of boards and maybe also comps?
I’d like to try it, has there been any thinking of a group buy of boards and maybe also comps?
Thank you!
I designed and just finished assembling a test THT PCB for remote sensing.
Has AC input with a voltage doubler so you can use a single secondary transformer to get +/- Vout. Hope the 50Hz ripple will be low enough to make it make sense.
I'm using the SMD denoisers I used for LM3x7 testing, just had to add a 10K resistor on each. But THT ones should work just the same. I'm just tired of drilling.
If performance checks out I'll release the project files with gerbers and all. Hope to do it in few days.
edit: there will also be a single Vout version I just won't test it.
This is the general idea I'm starting with (including test load, some 100mA RMS) but I'll probably tweak some values. I will also try the higher gain version but my hopes are low. 10nF compensation is too much for lower gain (not stable), but seems fine with 1nF, even without compensation (for lower gain version).
I designed and just finished assembling a test THT PCB for remote sensing.
Has AC input with a voltage doubler so you can use a single secondary transformer to get +/- Vout. Hope the 50Hz ripple will be low enough to make it make sense.
I'm using the SMD denoisers I used for LM3x7 testing, just had to add a 10K resistor on each. But THT ones should work just the same. I'm just tired of drilling.
If performance checks out I'll release the project files with gerbers and all. Hope to do it in few days.
edit: there will also be a single Vout version I just won't test it.
This is the general idea I'm starting with (including test load, some 100mA RMS) but I'll probably tweak some values. I will also try the higher gain version but my hopes are low. 10nF compensation is too much for lower gain (not stable), but seems fine with 1nF, even without compensation (for lower gain version).
Last edited:
Hi,
If you presented more data in addition to the diagram, everything would be perfect.
