Yes the Magni measurements are pretty clean. The issue is the DAC case+transformer combo. I'm not going to open it again for pictures but here's a photo I found online of the same DAC:
As you can see the transformer core is about 5cm away from the regulators, and the board has a ground plane in the analog output. My probing techniques and equipment are not proper. At those levels it's easy to pickup stuff.
You can tell by the lowered noise floor around those spikes that the denoiser is working. You could have a poorly made pcb layout that picks up some noise that is hidden in the higher noise floor of the rails. Once you clean them it could reveal some issues. All in all the noise is lower anyway.
Also it seems that in this particular DAC the case is not earthed. The earth goes to the transformer's shield but it isn't connected to the case. I need to start looking into case earthing etc
As you can see the transformer core is about 5cm away from the regulators, and the board has a ground plane in the analog output. My probing techniques and equipment are not proper. At those levels it's easy to pickup stuff.
You can tell by the lowered noise floor around those spikes that the denoiser is working. You could have a poorly made pcb layout that picks up some noise that is hidden in the higher noise floor of the rails. Once you clean them it could reveal some issues. All in all the noise is lower anyway.
Also it seems that in this particular DAC the case is not earthed. The earth goes to the transformer's shield but it isn't connected to the case. I need to start looking into case earthing etc
Last edited:
Here's the positive dienoiser again:
That highest peak at -60.8dB is actually at -120.8dB if we add the gain of the LNA. ADC is calibrated at 1.3V at 0dB, so that means that peak is at around 1.18uV. Sub 1uV you get into nV territory. At these levels it's very easy to pick up crap. The noise floor is at around -90dB in the graph, which is around -150dB considering the gain of the LNA. -150dB from 1.3V is around 40nV.
Here's the same rail just that I turned on my soldering station:
I'm not sure how many people turn off everything in the house to listen to the cleanest music.
Ideally you'd need a noise generator, a shielded case for testing the supply+denoiser, and a good LNA and probing gear. I don't have these things, I just made a rough measurement to see if they are working or not.
Also noise is only half the story here. You may lower the rail noise but your output signal may be dominated by noise from other devices, like a DAC or other chips. Another benefit of the denoisers is the low output impedance of the rails which I cannot measure and may play a role in how the device "sounds" in the end.
That highest peak at -60.8dB is actually at -120.8dB if we add the gain of the LNA. ADC is calibrated at 1.3V at 0dB, so that means that peak is at around 1.18uV. Sub 1uV you get into nV territory. At these levels it's very easy to pick up crap. The noise floor is at around -90dB in the graph, which is around -150dB considering the gain of the LNA. -150dB from 1.3V is around 40nV.
Here's the same rail just that I turned on my soldering station:
I'm not sure how many people turn off everything in the house to listen to the cleanest music.
Ideally you'd need a noise generator, a shielded case for testing the supply+denoiser, and a good LNA and probing gear. I don't have these things, I just made a rough measurement to see if they are working or not.
Also noise is only half the story here. You may lower the rail noise but your output signal may be dominated by noise from other devices, like a DAC or other chips. Another benefit of the denoisers is the low output impedance of the rails which I cannot measure and may play a role in how the device "sounds" in the end.
Last edited:
Elvee, Trileru, thanks, I note that the denoiser works without issues with lm337. I need -/+ 30 Vdc out. I'm going to work on pcb.
Damien
Damien
Good luck and tell us if you think the sound changed.
Elvee, could there be better channel separation as a consequence of lower output impedance for VACC/Vref in a DAC chip?
Elvee, could there be better channel separation as a consequence of lower output impedance for VACC/Vref in a DAC chip?
I remind you that 30V is borderline for LM317/337. You need an input voltage comfortably higher than that if you want to keep regulating under adverse conditions like low line voltage, etc. and if the regulator needs to charge output capacitors during startup, it will see the full input voltage, which may exceed the abs max ratings.
The denoisers will work, but the 220µF needs to be rated at 35V, and the 1K8 resistors must be >0.5W
If PSRR is poor, yes. This information is probably available in the DS.
Depending on the actual implementation, there might be other, indirect ways for degradation.
I think the ams1117 is basically lt1117 which is in LTSpice. I made a sim with the comparison between lm317+de/dienoiser and lt1117+de/dienoiser.
Output impedance of lt1117 also decreases and is more linear in the audio region. Not as low as lm317+dienoiser but good enough.
So for 5V it certainly is a good proposition for upgrading gear that uses 1117. 3.3V doesn't work as the output voltage rises to 3.8V. Maybe a series diode on the output? What would be the drawbacks of the series diode to the performance of the de/dienoiser?
edit: it seems there's a lt1117-2.85V, which simulated with dienoiser makes for a 3.3V output. That could be interesting as well.
also for the 1117-5.0V the originally 180R resistor from dienoiser circuit should be around 1K-1.1K for best results. The Diego table of values for fixed regulators (LM78xx) does not apply for best results with the 1117-5.0V. There's an extra 1.5dB to be had if using 1k instead of the recommended 1.5k from the Diego table for fixed regulators for 5V. And 1.3K for the 2.85V version.
Output impedance of lt1117 also decreases and is more linear in the audio region. Not as low as lm317+dienoiser but good enough.
So for 5V it certainly is a good proposition for upgrading gear that uses 1117. 3.3V doesn't work as the output voltage rises to 3.8V. Maybe a series diode on the output? What would be the drawbacks of the series diode to the performance of the de/dienoiser?
edit: it seems there's a lt1117-2.85V, which simulated with dienoiser makes for a 3.3V output. That could be interesting as well.
also for the 1117-5.0V the originally 180R resistor from dienoiser circuit should be around 1K-1.1K for best results. The Diego table of values for fixed regulators (LM78xx) does not apply for best results with the 1117-5.0V. There's an extra 1.5dB to be had if using 1k instead of the recommended 1.5k from the Diego table for fixed regulators for 5V. And 1.3K for the 2.85V version.
Last edited:
For sot-223 regs like the 1117 there also is the lm317 in that package, which would have higher psrr and lower output impedance than the variable output 1117 version. But the 1117 version deals with low esr on the output, and also has a more linear output impedance in the audio range. It's actually smaller at 20kHz for variable 1117+dienoiser than lm317+dienoiser. The variable output version of 1117 can also be found in LTSpice.
Yes, this is pecisely the reason I do not use a LM317 for a 32V power supply.
I switched to using a LT341 and a pass transistor with it's emitter at the 32V output side.
It works like wonder, low drop, using a CCS to feed the base from the collector side, a CCS made of a two bjt ring.
This is a very low part design with very low cost and excellent performance, perfectly compatible with the denoiser.
I was playing with the denoiser circuit by itself to see how the compensation value affects the response. Interesting to note that there's a significant difference between 20nF and 40nF. Since I only could get the lm317+dienoiser to work with minimum 40nF I have the inclination to use lm78xx + dienoiser where possible. I have it stable with 20nF but in reality I didn't try lower. 10nF would be a good compromise.
Last edited:
Also another trick to try to get lm337+dienoiser working would be to install a variable resistor instead of R6, and go from denoiser to where it starts going crazy, and see if you could muster more performance out of it, compared to just lm337+denoiser. You might get extra even if not full dienoiser performance.
The following graph shows what I mean (red trace is denoiser and going up is dienoiser territory):
This could be something worth experimenting with. Maybe the LM317 table of values for R6 value does not apply to LM337. It may have a different sweet spot that I didn't hit using the LM317 resistance values.
The following graph shows what I mean (red trace is denoiser and going up is dienoiser territory):
This could be something worth experimenting with. Maybe the LM317 table of values for R6 value does not apply to LM337. It may have a different sweet spot that I didn't hit using the LM317 resistance values.
Last edited:
Elvee I'm trying to understand my measurements. I found a LM337 spice model.
Is the general noise floor that I measure on the output of the reg a consequence of the self-noise reduction value, equivalent to the adj pin noise attenuation in the LTSpice simulation? And PSRR would deal only with the 50/100Hz higher ripple? Because that would make more sense. As in LM337+denoiser would have an relatively equivalent PSRR to the LM7912+dienoiser at 50/100Hz that I also see in the measurements and in the simulation:
And it would also explain the higher reduction in the general noise floor that I observe in measurements with LM7912 + dienoiser vs LM337+denoiser:
Either way practically it's still better to use the LM7912+dienoiser for the lowered noise floor even if the ripple is similar. That would imply that user terranigma's worries were justified and it wasn't bad measurement technique but the ripple is not that reduced as I'd hoped it would be.
Also output impedance seems lower on the LM7912+dienoiser compared to LM337+denoiser (still not as low as LM3x7):
Is the general noise floor that I measure on the output of the reg a consequence of the self-noise reduction value, equivalent to the adj pin noise attenuation in the LTSpice simulation? And PSRR would deal only with the 50/100Hz higher ripple? Because that would make more sense. As in LM337+denoiser would have an relatively equivalent PSRR to the LM7912+dienoiser at 50/100Hz that I also see in the measurements and in the simulation:
And it would also explain the higher reduction in the general noise floor that I observe in measurements with LM7912 + dienoiser vs LM337+denoiser:
Either way practically it's still better to use the LM7912+dienoiser for the lowered noise floor even if the ripple is similar. That would imply that user terranigma's worries were justified and it wasn't bad measurement technique but the ripple is not that reduced as I'd hoped it would be.
Also output impedance seems lower on the LM7912+dienoiser compared to LM337+denoiser (still not as low as LM3x7):
Last edited:
All models have to be taken with a pinch of salt, especially if you look at some characteristics, like noise, HF response, etc.
Unless it is a native LTspice model, the noise will be computed from the theoretical values of semiconductor junctions, resistors, etc., and will neglect any excess noise
Broadly speaking, the die-, de-, no-noisers are agnostic: their "business" side is connected to Vout and GND (or +sense and -sense), and they will attempt to cancel any non-zero AC voltage there, using their gain (loaded).
The AC can be the result of an imperfect PSRR, internal noise or voltage caused by the product of an AC current and a non-zero output impedance.
However, if you dig deeper, the correction path through the adj pin may not be able to correct all the different sources of perturbation in a fair and uniform manner, which can explain some apparent oddities.
With a plain-vanilla LM317 reg augmented by a standard denoiser, the uniform reduction rule applies almost perfectly: you see typically a 32dB improvement for the PSRRR, internal noise and output impedance, but the situation can be different for other configurations, where some errors are not correctable
We used the same circuit. If you are referring to the board I don't know I didn't make that one.
But ideally you'd use an add-on that you install close to the load.
Thank you for the info!
Managed to get LT1085 to work with dienoiser. I used a 2K pot to go from denoiser to best result. Works with 50nF/5R comp network and the trick was to use a low esr output cap. Best results were at 560R and 1.8K (I'd stick with 560R) for the dienoiser resistor value. Between it didn't look that good. Noise performance of LT1085 with dienoiser seems similar to LM317 with dienoiser.
Now I'm curious about LM337, maybe it works with dienoiser. Will report back the results.
Now I'm curious about LM337, maybe it works with dienoiser. Will report back the results.
Last edited:
Finally, LM337 works with dienoiser as well! The issue seems to have been the higher ESR output cap used for LM317. Apparently LT1085/LM337 doesn't like it.
LM337 works without R just with 40nF Ccomp. Dienoiser resistor value around 390-560R seems ok.
Here's a graph with some measurements. Yellow is LM337 denoiser, green is LM337 dienoiser, red is LT1085 dienoiser and bottom trace is the LNA input grounded at sense point.
LM337 works without R just with 40nF Ccomp. Dienoiser resistor value around 390-560R seems ok.
Here's a graph with some measurements. Yellow is LM337 denoiser, green is LM337 dienoiser, red is LT1085 dienoiser and bottom trace is the LNA input grounded at sense point.
Seems to have the same amplitude regardless of attenuation factor. Regardless of regulator as well. Might be related either to LNA, either to environment. Actually when I ground the LNA it's gone. But still I kinda could tell it's there just that it's attenuated when LNA is grounded. I don't know, but seems to be there on most denoisers I played with. I also use denoiser for LNA rails.
edit: I'll try with the laptop and turn off everything else, see if I still get it. It's strange as it's on denoiser trace as well, which should be stable.
edit: I'll try with the laptop and turn off everything else, see if I still get it. It's strange as it's on denoiser trace as well, which should be stable.
Last edited:
Seems to be LNA related, or environment pickup.
I have the LNA turned on, output connected to ADC, input connected to a diy lm3x7 denoiser power supply that is turned off, laying on the desk. This is what the ADC sees:
If I ground the LNA input cable at sense point it goes away (bottom trace). So I don't think that is related to the DUT.
I have the LNA turned on, output connected to ADC, input connected to a diy lm3x7 denoiser power supply that is turned off, laying on the desk. This is what the ADC sees:
If I ground the LNA input cable at sense point it goes away (bottom trace). So I don't think that is related to the DUT.
Last edited: