Also an interesting observation. Denoiser was stable with 20nF for Ccomp and 22R for Rcomp. Dienoiser didn't like the R at all, I had to short it. Also I couldn't get it to perform until I bumped C to 40nF. At 30nF for example it behaved somewhat like a denoiser in performance. I let it around 10 minutes to stabilize but same. When I went to 40nF it offered the performance shown in the previous post.
So this is very important for anyone considering trying a dienoiser. It seems the comp network depends heavily on the layout and parts. The compensation network should be tweaked in your specific application, and you should have some tools to check it. Multimeter only might not be enough. The LNA didn't show oscillation or weird behavior at 30nF, just that the performance wasn't there.
Here's for example how the output of the dienoiser looked with 20nF:
And with 30nF:
That's denoiser type performance, not dienoiser.
I have also found a broken to220 lm317 and decided to compare the die size to the d2pak one I broke yesterday. to220 is ST, d2pak is On.
So this is very important for anyone considering trying a dienoiser. It seems the comp network depends heavily on the layout and parts. The compensation network should be tweaked in your specific application, and you should have some tools to check it. Multimeter only might not be enough. The LNA didn't show oscillation or weird behavior at 30nF, just that the performance wasn't there.
Here's for example how the output of the dienoiser looked with 20nF:
And with 30nF:
That's denoiser type performance, not dienoiser.
I have also found a broken to220 lm317 and decided to compare the die size to the d2pak one I broke yesterday. to220 is ST, d2pak is On.
Last edited:
A most curious and interesting update!
I decided to get back to LT1085 as I wasn't sure of the comp network while testing it.
So 20nF with no R didn't work, 30nF with no R didn't work. 30nF with 22R seems to somewhat do something.
Now comes the interesting part. Bumping to 40nF + 22R we see a very interesting result:
This is lm317 + dienoiser type performance out of the denoiser + lt1085! -127dB total! How could this be?
I double checked and the PNP is shorted, I just measured with dmm. You can also tell it's the denoiser circuit by the characteristic hump at 27 something kHz.
Can someone please test this? Seems too good to be true. Also allow for 10 minutes for the whole thing to reach max performance. Seems true for lm317 as well.
Also am interested, is there any gain or downside in using lt1085 + denoiser and lm337 + denoiser for dual supplies? Would it be any benefit or downside to this?
I decided to get back to LT1085 as I wasn't sure of the comp network while testing it.
So 20nF with no R didn't work, 30nF with no R didn't work. 30nF with 22R seems to somewhat do something.
Now comes the interesting part. Bumping to 40nF + 22R we see a very interesting result:
This is lm317 + dienoiser type performance out of the denoiser + lt1085! -127dB total! How could this be?
I double checked and the PNP is shorted, I just measured with dmm. You can also tell it's the denoiser circuit by the characteristic hump at 27 something kHz.
Can someone please test this? Seems too good to be true. Also allow for 10 minutes for the whole thing to reach max performance. Seems true for lm317 as well.
Also am interested, is there any gain or downside in using lt1085 + denoiser and lm337 + denoiser for dual supplies? Would it be any benefit or downside to this?
Last edited:
I use jbau's LT1085 + LM337 circuit to feed my opamp based preamp. (Phase and Z graphs are here). Nothing seems alarming.
If you ask about imbalanced noise performance for each rail, I think some experienced people say more than my listening experience. It is also going to be interesting to measure LT1033+denoiser to see if it is going to match LT1083+denoiser.
I turned off some stuff around the desk, unplugged devices etc. The spectrum seems clean now.
I watched the live ARTA display for some time without averaging, and when I get the -70dB results the live display is kind of jumpy. But pushing connectors and cables seems to get to a state where it keeps a steady -60dB. I think at this level there's a lot of stuff that matters. Some signal may get attenuated into connectors. Hence I think the reality is more around -60dB in my measurements. With 60dB from the LNA that's -120dB -4.14dB that the input ripple is at, that's around 116dB total attenuation from LT1085+denoiser.
I really wish someone with gear would try to properly measure LT1085+denoiser.
I might try to measure again the audio output noise of the Magni, and file down a lt1085 and replace the lm317, and see if there's any change in the noise.
I watched the live ARTA display for some time without averaging, and when I get the -70dB results the live display is kind of jumpy. But pushing connectors and cables seems to get to a state where it keeps a steady -60dB. I think at this level there's a lot of stuff that matters. Some signal may get attenuated into connectors. Hence I think the reality is more around -60dB in my measurements. With 60dB from the LNA that's -120dB -4.14dB that the input ripple is at, that's around 116dB total attenuation from LT1085+denoiser.
I really wish someone with gear would try to properly measure LT1085+denoiser.
I might try to measure again the audio output noise of the Magni, and file down a lt1085 and replace the lm317, and see if there's any change in the noise.
Removed the cable I used and I directly soldered a coax to the LNA input on the pcb. I turned off everything apart from computer and it seems I can consistently get the same results. Noise floor dropped a bit as well. I can constantly see the 100Hz just under 71dB, as in -70.something.
I put an album with a few captures here, they are almost identical. All captures are averages of 100 samples for each capture.
Imgur: The magic of the Internet
This is a example:
So almost -127dB for LT1085+denoiser. I also have a LT1084, will try it tomorrow. LT1085 is not as cheap as LM317 but 28.5V/3A is nice. If LT1084 works the same that is a 5A device.
Too bad we don't have a negative equivalent to LT108x.
I put an album with a few captures here, they are almost identical. All captures are averages of 100 samples for each capture.
Imgur: The magic of the Internet
This is a example:
So almost -127dB for LT1085+denoiser. I also have a LT1084, will try it tomorrow. LT1085 is not as cheap as LM317 but 28.5V/3A is nice. If LT1084 works the same that is a 5A device.
Too bad we don't have a negative equivalent to LT108x.
Last edited:
I may have elucidated the mystery. Indeed the LT1085+denoiser offers the performance, just that it's at low frequency. I used Simetrix which has both lm317 and lt1085 models and these are the results:
lm317+dienoiser:
And lt1085+denoiser:
So LT1085+denoiser offers the performance just that at low frequency. Seems like lm317+denoiser is still the better option all around considering everything. Hope I made the simulation correct, but it kind of explains my findings.
And a comparison of lm317+denoiser/lm317+dienoiser and lt1085+dienoiser(if it would be stable):
lm317+dienoiser:
And lt1085+denoiser:
So LT1085+denoiser offers the performance just that at low frequency. Seems like lm317+denoiser is still the better option all around considering everything. Hope I made the simulation correct, but it kind of explains my findings.
And a comparison of lm317+denoiser/lm317+dienoiser and lt1085+dienoiser(if it would be stable):
Last edited:
Decided to try the dienoiser option on the Magni. Positive worked, I think I used 57nF Ccomp. 40nF didn't work, and I used a 47nF+10nF in parallel. Works great, stable. I tested it for some time, I listened to it while watching the positive rail output. I also made some sweeps, and it seems that only the tested frequency goes up as seen directly on the regulator output. No issues, great performance.
This time I directly soldered microphone shielded wire (softer than coax) into the LNA and on the output of the rails. Noise floor dropped a bit from doing this. And also denoiser measurements make more sense now.
Denoiser +lm317 offers total 114dB psrr at 50Hz
Denoiser +lm337 offers total 111.4dB psrr at 50Hz
Dienoiser +lm317 offers total 128.95dB psrr at 50hz
I couldn't get lm337 dienoiser to work like the last time I tried it on the LNA pcb. I tried all combinations of 5-20-47R with each 10nF to 100nF with 10nF increments.
Denoiser + lm337 works with 20nF/23.5R.
I noticed more ripple drop after installing the dienoiser, measured on the output of the headamp. So even if it's only the positive dienoiser + negative denoiser I still benefited off the dienoiser.
Now for some graphs. This is the positive denoiser on the output cap (input ripple into lm317 is at -9.26dB) fed into 60dB LNA then LNA-ADC:
So total around -114dB for lm317+denoiser.
This is the dienoiser on positive output cap fed into 60dB LNA then LNA-ADC:
As you can see the noise floor is lower as well. This has a psrr of almost 129dB at 50Hz.
And this is the negative denoiser (negative input ripple at regulator is at -7.65dB):
So this comes out at around -115dB psrr at 50Hz.
Now a comparison of before and after I went from positive denoiser to positive dienoiser. There's two channels on two levels of gain (lo/hi).
Left channel with both rails equiped with denoisers.
Low gain:
And after dienoiser on lm317:
High gain denoiser (still left channel):
High gain left channel with lm317 + dienoiser:
Noise floor seems the same, a bit of higher psrr at 50Hz.
Now right channel low gain with denoiser:
And right channel low gain after dienoiser:
This channel output shows around -5dB better after upgrading to dienoiser for lm317.
Right channel high gain with lm317+denoiser:
And right channel high gain after upgrading to lm317+dienoiser;
On high gain we notice around -5dB improvement at 50Hz. Left channel seems to have less noise than right channel. It's a clear improvement.
And a couple of measurements. Used my phone as signal generator into Magni, Magni output to 60dB LNA, LNA to ADC. 0dB signal output from LNA to ADC (around 1.3Vrms into ADC, around 13mVrms (low gain) from Magni output into LNA:
And Magni output directly into ADC, 0dB is 1.3Vrms, no LNA, phone as signal generator:
And output on high gain (I recalibrated 0dB from the Magni pot):
Now the low/high outputs look the same, with and without 60dB LNA.
A few gotchas while playing with the dienoiser. It might look stable on dmm but performance might suffer. I managed to get a steady output, but ADC was overloaded from noise. Without a LNA I couldn't have been sure the dienoisers are working fine. Tweaking R/Ccomp I could see differences on the ADC capture only, not on dmm. If you want to do it I think there are very slim chances of working with the values I used. You should tweak the dienoiser comp network while checking the psrr. Dmm might fool you into thinking the dienoiser is stable.
Sorry for the many screenshots but maybe someone finds the info useful. I'll listen to it for some time to hear for differences/issues etc. So far it sounds absolutely great!
This time I directly soldered microphone shielded wire (softer than coax) into the LNA and on the output of the rails. Noise floor dropped a bit from doing this. And also denoiser measurements make more sense now.
Denoiser +lm317 offers total 114dB psrr at 50Hz
Denoiser +lm337 offers total 111.4dB psrr at 50Hz
Dienoiser +lm317 offers total 128.95dB psrr at 50hz
I couldn't get lm337 dienoiser to work like the last time I tried it on the LNA pcb. I tried all combinations of 5-20-47R with each 10nF to 100nF with 10nF increments.
Denoiser + lm337 works with 20nF/23.5R.
I noticed more ripple drop after installing the dienoiser, measured on the output of the headamp. So even if it's only the positive dienoiser + negative denoiser I still benefited off the dienoiser.
Now for some graphs. This is the positive denoiser on the output cap (input ripple into lm317 is at -9.26dB) fed into 60dB LNA then LNA-ADC:
So total around -114dB for lm317+denoiser.
This is the dienoiser on positive output cap fed into 60dB LNA then LNA-ADC:
As you can see the noise floor is lower as well. This has a psrr of almost 129dB at 50Hz.
And this is the negative denoiser (negative input ripple at regulator is at -7.65dB):
So this comes out at around -115dB psrr at 50Hz.
Now a comparison of before and after I went from positive denoiser to positive dienoiser. There's two channels on two levels of gain (lo/hi).
Left channel with both rails equiped with denoisers.
Low gain:
And after dienoiser on lm317:
High gain denoiser (still left channel):
High gain left channel with lm317 + dienoiser:
Noise floor seems the same, a bit of higher psrr at 50Hz.
Now right channel low gain with denoiser:
And right channel low gain after dienoiser:
This channel output shows around -5dB better after upgrading to dienoiser for lm317.
Right channel high gain with lm317+denoiser:
And right channel high gain after upgrading to lm317+dienoiser;
On high gain we notice around -5dB improvement at 50Hz. Left channel seems to have less noise than right channel. It's a clear improvement.
And a couple of measurements. Used my phone as signal generator into Magni, Magni output to 60dB LNA, LNA to ADC. 0dB signal output from LNA to ADC (around 1.3Vrms into ADC, around 13mVrms (low gain) from Magni output into LNA:
And Magni output directly into ADC, 0dB is 1.3Vrms, no LNA, phone as signal generator:
And output on high gain (I recalibrated 0dB from the Magni pot):
Now the low/high outputs look the same, with and without 60dB LNA.
A few gotchas while playing with the dienoiser. It might look stable on dmm but performance might suffer. I managed to get a steady output, but ADC was overloaded from noise. Without a LNA I couldn't have been sure the dienoisers are working fine. Tweaking R/Ccomp I could see differences on the ADC capture only, not on dmm. If you want to do it I think there are very slim chances of working with the values I used. You should tweak the dienoiser comp network while checking the psrr. Dmm might fool you into thinking the dienoiser is stable.
Sorry for the many screenshots but maybe someone finds the info useful. I'll listen to it for some time to hear for differences/issues etc. So far it sounds absolutely great!
Last edited:
I remembered that my DAC has LM7812/7912 supplies for the analog output stage. I've tried the denoiser and dienoiser with it and it works, and LM7912 dienoiser works fine with 20nF comp cap. The only thing I noticed is that the DAC has the transformer close to the board, and the denoiser picks up the mains frequency easily, and I can tweak for best performance by positioning the sense gnd/vcc lines. So it's best to check the output of the denoiser while installing it.
Both 7x12 regulators worked with just 20nF with dienoiser. Looking back I probably should have left the 7x12 instead of the lm3x7 I switched to. I couldn't check for output impedance but from a noise point of view they seemed thereabouts.
So if you can find your lm79xx in your needed voltage flavor you might be better off by using them rather than lm3x7 combo.
I did have some issues as I discovered the sense lines pickup crap while checking the lm3x7 circuit as I couldn't understand the bad ripple, but once I tweaked it the noise looked ok. But I got some screenshots and you can see the rest of the noise floor dropping similarly for lm7x12 and lm3x7. I'll post the graphs tomorrow.
edit: this was for low currents, 50mA or less.
Both 7x12 regulators worked with just 20nF with dienoiser. Looking back I probably should have left the 7x12 instead of the lm3x7 I switched to. I couldn't check for output impedance but from a noise point of view they seemed thereabouts.
So if you can find your lm79xx in your needed voltage flavor you might be better off by using them rather than lm3x7 combo.
I did have some issues as I discovered the sense lines pickup crap while checking the lm3x7 circuit as I couldn't understand the bad ripple, but once I tweaked it the noise looked ok. But I got some screenshots and you can see the rest of the noise floor dropping similarly for lm7x12 and lm3x7. I'll post the graphs tomorrow.
edit: this was for low currents, 50mA or less.
Last edited:
Tested ams1117-5V and there's improvements. I tested on the diy supply and I guess you could have more improvement if you used an add-on board to sense directly at load.
This is a graph showing the improvement. Top line is normal output, middle blue line is ams1117+dienoiser and bottom line is the 60dB LNA input grounded at sense point, to give an idea of the lowest range it can pick up. There's some visual artifacts from overlaying screenshots.
Around 35dB lower noise floor. I can't easily clear the mains pick up as I'm testing on the desk with no case. I didn't test denoiser, I tested dienoiser directly.
Getting back to the lm7812/lm7912 here's the comparison. In the end I used dienoiser lm317 for positive and denoiser lm337 for negative, but I might switch to dienoiser lm7912 at some point.
The following graph shows the positive regulators. Top red line is the stock lm7812, the teal line is the lm7812+dienoiser and pink line is lm317+dienoiser. I first tested lm7x12 and figured later that I can adjust the mains pickup by moving the sense wires around. So lm7812 +dienoiser has some mains crap, but you can see the noise floor. Bottom line is the 60dB LNA sense input line, grounded at sense point.
As you can see the lm317 and lm7812 have similar noise floors after dienoiser. 7812 seems a bit better, and worked with 20nF Ccomp. lm317 needed 40nF to be stable.
The following graph is the lm337/lm2912 comparison. Top line is the lm7912 output measured at the output opamp pin. Teal line is the lm7912 + dienoiser (again screenshot made before I figured I could adjust the mains pick up). Pink line is lm337+denoiser. Bottom blue line is the LNA grounded at input sense point.
As you can see it appears there's more to be had with 7912+dienoiser, plus it's stable with just 20nF Ccomp.
The DAC output noise barely changed because AVCC is powered by a ams1117-5V that is powered from the lm317+dienoiser. The ams1117 has the same output noise regardless of what I use on its input. I might try and make a board for the ams1117 as well, just that it's really cramped in there.
I've used this diy tht add-on board:
http://
I attached the Kicad project. This can be easily diyable. Also this design is tested and working, the same as the smd diy add-on board.
To add the denoiser to lm78xx/lm79xx you just need to lift the ground pin and put 100R between it and ground. The same point where you join the gnd leg to the 100R you also feed ADJ from add-on board. If you retrofit lm3x7 circuits with lm78xx/79xx you just need to remove R1 and replace R2 with 100R, the rest should be the same. The output voltage will be 0.3V higher I think, on the fixed voltage regs.
This is a graph showing the improvement. Top line is normal output, middle blue line is ams1117+dienoiser and bottom line is the 60dB LNA input grounded at sense point, to give an idea of the lowest range it can pick up. There's some visual artifacts from overlaying screenshots.
Around 35dB lower noise floor. I can't easily clear the mains pick up as I'm testing on the desk with no case. I didn't test denoiser, I tested dienoiser directly.
Getting back to the lm7812/lm7912 here's the comparison. In the end I used dienoiser lm317 for positive and denoiser lm337 for negative, but I might switch to dienoiser lm7912 at some point.
The following graph shows the positive regulators. Top red line is the stock lm7812, the teal line is the lm7812+dienoiser and pink line is lm317+dienoiser. I first tested lm7x12 and figured later that I can adjust the mains pickup by moving the sense wires around. So lm7812 +dienoiser has some mains crap, but you can see the noise floor. Bottom line is the 60dB LNA sense input line, grounded at sense point.
As you can see the lm317 and lm7812 have similar noise floors after dienoiser. 7812 seems a bit better, and worked with 20nF Ccomp. lm317 needed 40nF to be stable.
The following graph is the lm337/lm2912 comparison. Top line is the lm7912 output measured at the output opamp pin. Teal line is the lm7912 + dienoiser (again screenshot made before I figured I could adjust the mains pick up). Pink line is lm337+denoiser. Bottom blue line is the LNA grounded at input sense point.
As you can see it appears there's more to be had with 7912+dienoiser, plus it's stable with just 20nF Ccomp.
The DAC output noise barely changed because AVCC is powered by a ams1117-5V that is powered from the lm317+dienoiser. The ams1117 has the same output noise regardless of what I use on its input. I might try and make a board for the ams1117 as well, just that it's really cramped in there.
I've used this diy tht add-on board:
http://
I attached the Kicad project. This can be easily diyable. Also this design is tested and working, the same as the smd diy add-on board.
To add the denoiser to lm78xx/lm79xx you just need to lift the ground pin and put 100R between it and ground. The same point where you join the gnd leg to the 100R you also feed ADJ from add-on board. If you retrofit lm3x7 circuits with lm78xx/79xx you just need to remove R1 and replace R2 with 100R, the rest should be the same. The output voltage will be 0.3V higher I think, on the fixed voltage regs.
Last edited:
For the fixed regulators I see in the simulation that the 220uF should be reversed. Is that so? I don't remember if I read anything about this in earlier posts. I see it put both ways in different schematics.
The ams1117-5.0 has been tested with it configured for lm3x7 application. I don't know if it would take away from the performance as the delta is pretty small, about 0.3V
edit: I just realized that I made the 7x12 tests with it configured for lm3x7. Also you need to mind the value for the originally 180R resistor. Fixed regulators have another table of recommended values, you can find that in earlier posts. But it's different from the lm3x7 dienoiser values.
The ams1117-5.0 has been tested with it configured for lm3x7 application. I don't know if it would take away from the performance as the delta is pretty small, about 0.3V
edit: I just realized that I made the 7x12 tests with it configured for lm3x7. Also you need to mind the value for the originally 180R resistor. Fixed regulators have another table of recommended values, you can find that in earlier posts. But it's different from the lm3x7 dienoiser values.
Last edited:
Yes, the GND pin sits practically at GND potential, and the output of the dienoiser has a few volts DC average
Correct me if I'm wrong but denoiser/nonoiser doesn't work well with lm337. So could we simply use 2 x LM317 + denoiser to make a symetrical regulated with low noise power simply ?
I made (quick) tests on the 337, and it seems to work without particular problem:
D-Noizator: a magic active noise canceller to retrofit & upgrade any 317-based V.Reg.
I tested Fairchild and ST samples
The denoiser works without issues with lm337. Dienoiser is what is not stable. And nonoiser we don't have it confirmed yet on the 337.
I added a dienoiser to the ams1117-5.0 that is feeding AVCC/Vref of the AK4396 DAC. I had to make a small pcb. I got lucky as I found a 220uF/6.3V small cap, and also 10uF/10V 0805 caps, so I could make something small that would fit. Seems to be working fine. It's in dienoiser configuration. 20nF Ccomp. Noise floor seems to have dropped about 38dB. Output sits at 5.28V pretty solid. At startup it goes to 5.33V max. I tested a few times starting up. And is within the 5.5V recommended maximum operating conditions.
Some pictures of the board and comparison:
Orange is stock ams1117 and teal is with dienoiser. Seems I have some pickup issues, I used a long 10cm wire to tap into 5V. I left it since I installed the dienoiser as I have no way of soldering anything in there once all is put together. There's the other dienoiser board that sits on top of this one. Kinda messy.
Since I took out the pcb I decided to switch to LM7912 + dienoiser for better performance on the negtive rail. LM7912 is pin compatible with LM337. LM7812 is more involving to adapt instead of LM317. Rotating LM317 allowed me to install a larger heatsink.
And a before/after of the positive and negative rails, now with LM317+dienoiser instead of LM7812:
And negative rail with LM7912+dienoiser instead of just LM7912:
The DAC+headamp sound absolutely great!
I attached the Kicad project for the board I used. This one is also tested and working. This is to be used for fixed regulators at 5V. I don't think you can find 220uF caps this small at higher than 6.3V rating. You can also use it for variable regulators but you need to rotate the 220uF cap. Also mind the originally 180R value. For 5V fixed regulators it's recommended to be 1.5K. The board is also diyable. No holes.
Some pictures of the board and comparison:
Orange is stock ams1117 and teal is with dienoiser. Seems I have some pickup issues, I used a long 10cm wire to tap into 5V. I left it since I installed the dienoiser as I have no way of soldering anything in there once all is put together. There's the other dienoiser board that sits on top of this one. Kinda messy.
Since I took out the pcb I decided to switch to LM7912 + dienoiser for better performance on the negtive rail. LM7912 is pin compatible with LM337. LM7812 is more involving to adapt instead of LM317. Rotating LM317 allowed me to install a larger heatsink.
And a before/after of the positive and negative rails, now with LM317+dienoiser instead of LM7812:
And negative rail with LM7912+dienoiser instead of just LM7912:
The DAC+headamp sound absolutely great!
I attached the Kicad project for the board I used. This one is also tested and working. This is to be used for fixed regulators at 5V. I don't think you can find 220uF caps this small at higher than 6.3V rating. You can also use it for variable regulators but you need to rotate the 220uF cap. Also mind the originally 180R value. For 5V fixed regulators it's recommended to be 1.5K. The board is also diyable. No holes.
Come to think of it, I think one of the advantages of using the fixed regulators approach is you can use a low voltage rating 220uF cap, for any output voltage. The only constraint remains on the 22uF cap voltage rating. I used LTSpice with LM7815+dienoiser and I see at most around 1V difference between the two sides of the 220uF cap.
I don't know if the denoiser/dienoiser offers the same low output impedance to the fixed regs as it does to the LM317.
Are there any reliable LM78xx LTSpice models?
I don't know if the denoiser/dienoiser offers the same low output impedance to the fixed regs as it does to the LM317.
Are there any reliable LM78xx LTSpice models?
Have you ever get a flat noise floor with your recent measurements? Those spiky graphs look concerning to my eyes. Makes me think that failure on own build will be highly possible if you can't measure.