An interesting point about the last variant that I presented (where the noise amplification stage is powered from the input of the regulator), is that it keeps the improvement parameters acceptably constant, even between very different output voltages. The same is not possible when the noise amplification stage is powered from the regulator output, in which case certain optimizations are necessary at each voltage.
This advantage favors application in adjustable power supplies (within certain limits).
Low current version:
Green: R2 = 12K6; Red: R2 = 1177,9 ohms
Last edited:
Tested the higher gain dienoiser version:
It's a bit more finicky, Vce of Q2 had to be about 8.5V when the high performance really kicked in. I wouldn't try to build one without a LNA, you need to confirm the performance when you build it.
I had to short the 0.2R resistor, so cap's normal ESR of 0.1R worked fine (two caps with ESR of 0.2R in parallel).
Didn't try more than 22nF for comp cap, extra BJT was soldered to its ground connection and was too complicated to replace it.
R9 should be 150-220R. Tried with 560R and didn't work no matter Vce value. As in noisefloor was really high, and poor PSRR performance. CCS resistor decreased to 100R for a bit more current. Also used R1=4.7K. I threw everything at it.
But when you get it working you get a bit under 150dB of PSRR:
And the lowest noise that I've had a LM317 at:
86nV but removing my LNA's noise it works out at 76nV, for the regulator only. Equivalent noise for a 17.6R resistor. This had to be the highest PSRR/lowest noise any LM317 ever had! But didn't test much for stability, just some poke test and seemed fine but don't think it was very stable. Ideally I'd increase comp cap to 33nF and maybe tweak output RLC network with individual parts.
Attached REW measurement file.
It's a bit more finicky, Vce of Q2 had to be about 8.5V when the high performance really kicked in. I wouldn't try to build one without a LNA, you need to confirm the performance when you build it.
I had to short the 0.2R resistor, so cap's normal ESR of 0.1R worked fine (two caps with ESR of 0.2R in parallel).
Didn't try more than 22nF for comp cap, extra BJT was soldered to its ground connection and was too complicated to replace it.
R9 should be 150-220R. Tried with 560R and didn't work no matter Vce value. As in noisefloor was really high, and poor PSRR performance. CCS resistor decreased to 100R for a bit more current. Also used R1=4.7K. I threw everything at it.
But when you get it working you get a bit under 150dB of PSRR:
And the lowest noise that I've had a LM317 at:
86nV but removing my LNA's noise it works out at 76nV, for the regulator only. Equivalent noise for a 17.6R resistor. This had to be the highest PSRR/lowest noise any LM317 ever had! But didn't test much for stability, just some poke test and seemed fine but don't think it was very stable. Ideally I'd increase comp cap to 33nF and maybe tweak output RLC network with individual parts.
Attached REW measurement file.
Thanks! Yes, it's an LTSpice theme. You can get it from here:
https://draculatheme.com/ltspice
It's just the color values in the LTspiceXVII.ini file.
https://draculatheme.com/ltspice
It's just the color values in the LTspiceXVII.ini file.
Does it work well with a step-down input?
Usually when the regulator is not stable, it generates oscillations at the input when powered by a step-down switch.
That higher gain version should be considered very experimental. Didn't test for stability much, and I used a battery supply for the regulator.
I might do some stress testing later on, on this higher gain version. I do have a step down converter as a bench power supply. But I need to figure out a type of load which stresses the regulator on stability. So far I have an electronic load which has very poor AC performance, under 1Hz with some 4 bit resolution or something. Maybe a small audio amp?
I might do some stress testing later on, on this higher gain version. I do have a step down converter as a bench power supply. But I need to figure out a type of load which stresses the regulator on stability. So far I have an electronic load which has very poor AC performance, under 1Hz with some 4 bit resolution or something. Maybe a small audio amp?
I played with the LM337 CCS Denoisator setup and I found that 47nF+3.9R comp works fine with any ~0.2R ESR output cap.
I was able to have it stable with any low ESR output cap that it has issues with, just by correcting their ESR, with small resistance in series.
I've tested with the addon board, which I figured is the worst case scenario, and copied same values to the SMD dual board, and LM337 behaved just the same (although higher PSRR as it's the -N version on the PCB, and denoiser is integrated on the PCB).
With these values I can't get it to oscillate, it doesn't even seem close to. Always quickly recovers even under load. Low noise/high PSRR.
Vce of denoiser BJT was always in the -7V to -8.5V range (for -12Vout). SMD supply has MPSAx6 pair, addon board has BC8x7.
So LM317 output caps should work fine for LM337 as well, as long as they're in the 0.2R ESR range. One example that worked fine is Panasonic FC 100uF/63V.
I have also tested with a small audio amp, powered by the LM337 Denoisator with wires the other way around, seemed stable with a few test signals, while poking at the denoiser innards. Quickly recovered as well.
Not sure if I can gauge the output impedance with this chip amp, I should maybe have an audio signal directly driving a transistor connected to the supply?
So far I got this shot, 8R resistor load for amplifier, one channel grounded, 12Vin supply, 0.77V RMS for a 70Hz sine into the 8R resistor I got a doubled frequency on the power rail, 140Hz.
The amp draws 50mA idle and it was drawing 150mA with the 70Hz signal.
The 140Hz peak is at -133.8dBV which is 204nV RMS, and at 100mA that means an output impedance of 2uOhm. But again, not sure this type of load is adequate to gauge output impedance. Don't know why the frequency is doubled. They seem like harmonics and 70Hz is there as well just very low.
I also shorted the 33R protection resistor. Seemed like it was affecting stability and worked nicer without it. I abused the BJT bases under load by poking at them and didn't have a BJT fail, without that resistor. I've done random startups/shutdowns under load and again, seemed fine.
The Denoisator is AC coupled so if denoiser BJT fails there isn't much danger, you lose the AC performance.
For the CCS version the LED is not just a power rail status LED, it's more of a Denoisator status LED. If the denoiser is badly oscillating it will flicker, even with (relatively) steady Vout. Or can even go out completely with nominal Vout. That's always a sign of something wrong with the denoiser.
I was able to have it stable with any low ESR output cap that it has issues with, just by correcting their ESR, with small resistance in series.
I've tested with the addon board, which I figured is the worst case scenario, and copied same values to the SMD dual board, and LM337 behaved just the same (although higher PSRR as it's the -N version on the PCB, and denoiser is integrated on the PCB).
With these values I can't get it to oscillate, it doesn't even seem close to. Always quickly recovers even under load. Low noise/high PSRR.
Vce of denoiser BJT was always in the -7V to -8.5V range (for -12Vout). SMD supply has MPSAx6 pair, addon board has BC8x7.
So LM317 output caps should work fine for LM337 as well, as long as they're in the 0.2R ESR range. One example that worked fine is Panasonic FC 100uF/63V.
I have also tested with a small audio amp, powered by the LM337 Denoisator with wires the other way around, seemed stable with a few test signals, while poking at the denoiser innards. Quickly recovered as well.
Not sure if I can gauge the output impedance with this chip amp, I should maybe have an audio signal directly driving a transistor connected to the supply?
So far I got this shot, 8R resistor load for amplifier, one channel grounded, 12Vin supply, 0.77V RMS for a 70Hz sine into the 8R resistor I got a doubled frequency on the power rail, 140Hz.
The amp draws 50mA idle and it was drawing 150mA with the 70Hz signal.
The 140Hz peak is at -133.8dBV which is 204nV RMS, and at 100mA that means an output impedance of 2uOhm. But again, not sure this type of load is adequate to gauge output impedance. Don't know why the frequency is doubled. They seem like harmonics and 70Hz is there as well just very low.
I also shorted the 33R protection resistor. Seemed like it was affecting stability and worked nicer without it. I abused the BJT bases under load by poking at them and didn't have a BJT fail, without that resistor. I've done random startups/shutdowns under load and again, seemed fine.
The Denoisator is AC coupled so if denoiser BJT fails there isn't much danger, you lose the AC performance.
For the CCS version the LED is not just a power rail status LED, it's more of a Denoisator status LED. If the denoiser is badly oscillating it will flicker, even with (relatively) steady Vout. Or can even go out completely with nominal Vout. That's always a sign of something wrong with the denoiser.
Last edited:
Maybe this is a better output impedance measurement setup?
Made a simple MOSFET amplifier, DMM showed around 4.9V RMS across R1 (150R) and some 33mA RMS flowing through it. Tried a few frequencies in the audio spectrum.
Measured on the output of the power supply board, right where denoiser is sensing.
115Hz sits at -121dBV
That is 890nV, and at 33mA it means an output impedance of 27uOhms. It's similar up to 3kHz.
At the higher end of the spectrum it goes higher, at 12kHz it's 143uOhms and 20kHz it's 354uOhms.
Not sure if this is the proper way to do it or I missed something.
Made a simple MOSFET amplifier, DMM showed around 4.9V RMS across R1 (150R) and some 33mA RMS flowing through it. Tried a few frequencies in the audio spectrum.
Measured on the output of the power supply board, right where denoiser is sensing.
115Hz sits at -121dBV
That is 890nV, and at 33mA it means an output impedance of 27uOhms. It's similar up to 3kHz.
At the higher end of the spectrum it goes higher, at 12kHz it's 143uOhms and 20kHz it's 354uOhms.
Not sure if this is the proper way to do it or I missed something.
Last edited:
Besides R11, should R8 adjusted for different output regulated voltages? Is there a list for that?
In my case I would be setting for 15v.
What voltage should I adjust to in the joint node of C5 and R8?
In my case I would be setting for 15v.
What voltage should I adjust to in the joint node of C5 and R8?
You need to adjust R9 to keep ~3mA (not precise) through the LED and R8 to have about Vout/2 at that node. Not critical, any standard value. That voltage depends on BJT you use, try a few values until you get anywhere 7-10V.
Measured the output impedance of ST LM337 THT board with addon, came out a bit better at around 19uOhm in lower part of spectrum and 62uOhm at 20kHz.
I attached the REW measurement files, each measurement is named with fundamental test frequency. Had 4.9-5V RMS across 150R resistor in schematic from two posts up, so 33mA.
Then I tested the LM317-N dienoiser output impedance and that thing is a beast!
Has 1.3.uOhm in lower part of spectrum and goes to 11uOhm at 20kHz.
I attached the REW measurement files, each measurement is named with fundamental test frequency. Had 4.9-5V RMS across 150R resistor in schematic from two posts up, so 33mA.
Then I tested the LM317-N dienoiser output impedance and that thing is a beast!
Has 1.3.uOhm in lower part of spectrum and goes to 11uOhm at 20kHz.
LM317-N on the SMD dual board came out similar to LM337-N. But I'd hold off on the dienoiser measurement, might have screwed it up I'll have to confirm it again.
And ST LM317T with addon board, similar to the TI LM317-N in the extra harmonics that are present.
My speculation is that there's some different interaction between it and simple amplifier. When powered with reversed negative Vout those extra harmonics aren't present.
CCS Denoisator addon board had 33nF comp cap. Uses BC8x7 pair. Dual SMD with LM3x7-N uses MPSAx6 pair.
My speculation is that there's some different interaction between it and simple amplifier. When powered with reversed negative Vout those extra harmonics aren't present.
CCS Denoisator addon board had 33nF comp cap. Uses BC8x7 pair. Dual SMD with LM3x7-N uses MPSAx6 pair.
I don't have a project ATM but a typical preamp or phono stage retrofit to a Maida style with 3oov @ 60 to 80ma. would probably be excellent!
