Don't put series resistance with the output. Put it after the output cap to ground.
Elvee mentioned the denoiser was stable with 20mOhm caps, in earlier posts. Adding resistance in series with the output of the reg ruins the ultra low output impedance that the regs have after adding the denoiser.
I see the lm337 denoiser has no series resistance for the comp cap. You could try adding 20R or so and see what happens.
Also seems like there's differences between lm3x7 regs, from manufacturer to manufacturer, and between packages. A to220 lm337 made by ST worked fine without series R in comp network, while a lm337t made by OnSemi had issues without the R. You may have to tweak the comp network on the denoising circuit at fault.
Also the output voltage varies over temperature, there's a sim in earlier postings, made by Elvee. Doesn't drift too much but 20mV drift is within the denoiser specs I think.
The issues might be internal to the reg. I only checked with dmm but without Rcomp the lm337 would sit at a lower output voltage and output looked stable. It might have had stuff that the dmm didn't pick up. I wouldn't stray from the typical part values, and even if Elvee tested with your lower currents, your reg manufacturer/package might add extra details that make for your results.
Also I had 0.1uF ceramics on the output of the reg that I took out before adding the denoiser. If your 0.1uF ceramics are enough distance away from the reg, you could leave them as the tracks might add the required extra impedance to keep the regulator stable. Calculate for your track/wire length. Don't forget that the 0.1uF ceramics are in parallel with the output cap. Maybe measure the ESR that the regulator sees by measuring at the regulator output.
Elvee mentioned the denoiser was stable with 20mOhm caps, in earlier posts. Adding resistance in series with the output of the reg ruins the ultra low output impedance that the regs have after adding the denoiser.
I see the lm337 denoiser has no series resistance for the comp cap. You could try adding 20R or so and see what happens.
Also seems like there's differences between lm3x7 regs, from manufacturer to manufacturer, and between packages. A to220 lm337 made by ST worked fine without series R in comp network, while a lm337t made by OnSemi had issues without the R. You may have to tweak the comp network on the denoising circuit at fault.
Also the output voltage varies over temperature, there's a sim in earlier postings, made by Elvee. Doesn't drift too much but 20mV drift is within the denoiser specs I think.
The issues might be internal to the reg. I only checked with dmm but without Rcomp the lm337 would sit at a lower output voltage and output looked stable. It might have had stuff that the dmm didn't pick up. I wouldn't stray from the typical part values, and even if Elvee tested with your lower currents, your reg manufacturer/package might add extra details that make for your results.
Also I had 0.1uF ceramics on the output of the reg that I took out before adding the denoiser. If your 0.1uF ceramics are enough distance away from the reg, you could leave them as the tracks might add the required extra impedance to keep the regulator stable. Calculate for your track/wire length. Don't forget that the 0.1uF ceramics are in parallel with the output cap. Maybe measure the ESR that the regulator sees by measuring at the regulator output.
These measurements and findings about stability are very interesting.
Stability issues are those of any regulator that uses high loop gain to obtain extremely low Zout.
No wonder why this design is sensible to the output cap, the compensation network and the pass device headroom.
I see all of this in simulations using a TL431 and a CCS, instead of an LM317.
About 100nF ceramic caps.
These are bypass capacitors, at the ICs on the board powered by the regulated PSU.
They are not directly at the regulator, there is wires and tracks inductance in between.
You get good result inserting a diode, I suspect you would get the same with a low value resistor or a couple of inches wires.
I think there is some isolation needed to have unconditional stability whatever decoupling that are on the powered board. Typically 100nF caps at every IC Vcc pin and one 10uF for all the circuitry.
Stability issues are those of any regulator that uses high loop gain to obtain extremely low Zout.
No wonder why this design is sensible to the output cap, the compensation network and the pass device headroom.
I see all of this in simulations using a TL431 and a CCS, instead of an LM317.
About 100nF ceramic caps.
These are bypass capacitors, at the ICs on the board powered by the regulated PSU.
They are not directly at the regulator, there is wires and tracks inductance in between.
You get good result inserting a diode, I suspect you would get the same with a low value resistor or a couple of inches wires.
I think there is some isolation needed to have unconditional stability whatever decoupling that are on the powered board. Typically 100nF caps at every IC Vcc pin and one 10uF for all the circuitry.
Hi Trileru,
Compliments for your persistance and careful measurements.
Can you please elaborate on the current draw in your measurements? Both LM317 and LM337 require a minimum current (various threads mention at least about 50mA) to get stable results.
Compliments for your persistance and careful measurements.
Can you please elaborate on the current draw in your measurements? Both LM317 and LM337 require a minimum current (various threads mention at least about 50mA) to get stable results.
Trileru :
The series R at the output was for testing only , yet some designs have a resistor
and a cap near an opamp to isolate it from the rest.
I've been on this thread from the beginning an have read it all over the past year.
Elvee has said that the series R is bad with the LM337 , no need for it there and I
have no problem with the 337's . The LM317 is al lot more tollerant , but it is just
this one that has this problem. The 317's are ST and the 337's are ON's all in
TO220. I had a picture here about a year ago .
I know the output varies with temp and load , but the 3 others don't show this
behavior like the problem one . 20 mV is not much but it is 20 mV more than the
others (with the same load and temperature ) , they're on the same piece of
aluminium.
Elvee said it was ok to stray from typical part values , and it works for the others.
LM's work fine with even less than 3 mA . I usually use 560 to 1k as LM bias R over the 1,25V ref/adj or powering digital circuits. But to be closer to Elvee 's design I chose 365 ohm .
Instead of using kelvin connections , I use thick and short copper , but this means that ceramics near the opamps seem as close to the regulator as if they were there, minus some very low inductance.
It is confusing , Elvee insists (and proves) that ceramics parallel with a electrolitic is bad , but manufacturers (and Elektor) show it nearly all the time . Yet the Denoiser hates it so I can't use them , even further from the regulator.
The series R at the output was for testing only , yet some designs have a resistor
and a cap near an opamp to isolate it from the rest.
I've been on this thread from the beginning an have read it all over the past year.
Elvee has said that the series R is bad with the LM337 , no need for it there and I
have no problem with the 337's . The LM317 is al lot more tollerant , but it is just
this one that has this problem. The 317's are ST and the 337's are ON's all in
TO220. I had a picture here about a year ago .
I know the output varies with temp and load , but the 3 others don't show this
behavior like the problem one . 20 mV is not much but it is 20 mV more than the
others (with the same load and temperature ) , they're on the same piece of
aluminium.
Elvee said it was ok to stray from typical part values , and it works for the others.
LM's work fine with even less than 3 mA . I usually use 560 to 1k as LM bias R over the 1,25V ref/adj or powering digital circuits. But to be closer to Elvee 's design I chose 365 ohm .
Instead of using kelvin connections , I use thick and short copper , but this means that ceramics near the opamps seem as close to the regulator as if they were there, minus some very low inductance.
It is confusing , Elvee insists (and proves) that ceramics parallel with a electrolitic is bad , but manufacturers (and Elektor) show it nearly all the time . Yet the Denoiser hates it so I can't use them , even further from the regulator.
The under and overvoltage voltage and DC detector use the same supply as the amp , but I isolate them from it with diodes , so the Denoiser doesn't "see" the ceramics and I can use them as I like .
LM's alone have no problem with ceramics and I use them plentyful and without problems with digital circuits.
You can use the third detector, but with C5=10nF.
I don't think you have oscillations though.
Remember that the regulators need a minimum load to regulate properly; that's especially true for the 337.
Regarding the excess current drain, you probably have a leaky cap.
A leaky 220µ could also cause slight voltage variations.
Note that if the main output Ecap has some ESR, it will be sufficient to damp a resonance caused by ceramic or film caps.
A real load will also have various parallel damping mechanisms.
These aren't present when you test the regulators in isolation.
I don't think you have oscillations though.
Remember that the regulators need a minimum load to regulate properly; that's especially true for the 337.
Regarding the excess current drain, you probably have a leaky cap.
A leaky 220µ could also cause slight voltage variations.
Note that if the main output Ecap has some ESR, it will be sufficient to damp a resonance caused by ceramic or film caps.
A real load will also have various parallel damping mechanisms.
These aren't present when you test the regulators in isolation.
I have not tested the current draw, but I remember the Magni was 2W and something while checking with a watt meter at the 230VAC plug. For 34V (+-17V) that's around 59mA so should be thereabouts. I might add two 800ohm Rload on each rail just to be sure. But Magni had lm3x7 originally so I guess they at least took that into account.
For the LNA I had some issues as I got a 100mA resetable fuse
and whole thing was limited. I bypassed it and noticed it draws around 120-130mA so again, should be fine. I think I added some 1-1.5K Rload on each reg on the LNA anyway. I remember there being around 10mA min draw for a lm3x7 reg to be ok.
Yet my LM337 denoiser doesn't work without some R in comp network. Elvee said the LM337 went unstable past 82R in that spot, and he generally recommended not using anything there. Doesn't mean that using any R there would not work. Aparently I needed to use some for the thing to work. Might go out again if I increase it more. Maybe there's a sweet spot.
0.1uF caps seem to be regarded like a lucky charm. You just don't feel good if they are not there. They are used by some manufacturers to cover most conditions, but it doesn't mean they are always necessary. Especially on LM3x7 regs. Datasheet doesn't mention them. I don't know what Elektor did, but I'd be more trusting of the manufacturer's datasheet than Elektor.
It has been stated ad nauseam and I'm going to repeat it: it's not the Denoiser that does not "like" ceramics, it's the LM3x7 regs that don't do well with their low ESR, and it's not only ceramics, it's any cap that has a very low ESR. Denoiser has no stricter requirements for output cap, than lm3x7 originally have anyway. This is not about the denoiser.
My Magni used 0.1uF caps directly on the output of the reg, which bypassed 30mOhm 1000uF output caps. Some complained about this on different forums and Schiit said they got good measurements so they used it. Yet they are not required considering the lm3x7 datasheet. People do regard them as lucky charms sadly. Some even have their preferred brand/color/size for these.
I've updated the double add-on designs to have only denoiser and Rcomp for lm337 as well. RF shield version has two footprints in parallel for 220uF cap, so you can use two 100uF in parallel or maybe bump the value to 440uF. Higher voltage higher capacitance tantalums in that footprint are harder to find.
I think the tht cap double denoiser add-on could be diy-ed.
RF shield part number BMI-S-203-F is only the frame. The metal cover for it has part number BMI-S-203 without the F.
I still need to fix the complete dual rail power supplies, tht and smd versions to include Rcomp for lm337. I didn't need it for to220 lm337 made by ST but ymmv.
Updated Kicad projects attached.
Until Elvee or anyone else confirms the Nonoiser works with LM337 I wouldn't make any Nonoiser board version that has denoiser for LM337.
One very nice thing about the denoiser is that it allows to use a single output transformer with half wave rectifying and you can get away with not much input capacitance. For dual rails! I think the denoiser circuit costs less than the extra needed input capacitance.
edit: forgot about warning! these boards have not been tested yet. you make them at your own risk!
I think the tht cap double denoiser add-on could be diy-ed.
RF shield part number BMI-S-203-F is only the frame. The metal cover for it has part number BMI-S-203 without the F.
I still need to fix the complete dual rail power supplies, tht and smd versions to include Rcomp for lm337. I didn't need it for to220 lm337 made by ST but ymmv.
Updated Kicad projects attached.
Until Elvee or anyone else confirms the Nonoiser works with LM337 I wouldn't make any Nonoiser board version that has denoiser for LM337.
One very nice thing about the denoiser is that it allows to use a single output transformer with half wave rectifying and you can get away with not much input capacitance. For dual rails! I think the denoiser circuit costs less than the extra needed input capacitance.
edit: forgot about warning! these boards have not been tested yet. you make them at your own risk!
Last edited:
Thanks Elvee.
You could be right about a cap leaking . It's been on all day , I just measured it , it's now an extra 1 mA more while the other 3 are the same as before.
If it keeps escalating I will find out which one . Never had a Ecap fail on me.
I use regular Ecaps , never low ESR for series regulator , but I remember you posting graphs of how bad it was putting Ecap and ceramics together. Even worse using film caps , I think.
You could be right about a cap leaking . It's been on all day , I just measured it , it's now an extra 1 mA more while the other 3 are the same as before.
If it keeps escalating I will find out which one . Never had a Ecap fail on me.
I use regular Ecaps , never low ESR for series regulator , but I remember you posting graphs of how bad it was putting Ecap and ceramics together. Even worse using film caps , I think.
Yet the trouble I had last year with the LM337 + DN , were caused only by the ceramic on the output. Shorting the transistor of the Denoiser , making it a regular circuit WITH 1uF ceramic , no problem. So it is the Denoiser that makes the LM's unstable with it.
I've never had any trouble with LM3x7's with lots of ceramics ( every IC has it's own 10 or 100nF or even more). SMD ceramics are cheap and fun to solder , so I use them a lot , even when not needed . 🙂 So yes maybe they are my lucky charm.😉
You're not using the circuit as intended. The Denoiser does not add any extra requirements on the datasheet circuit.
The last statement should be read as: "I understand that some people are pushing the limits of LM3x7, some are testing these limits, some are plain lucky. The Denoiser is stated to be used on datasheet regular circuits. Going out of these bounds does not guarantee it works. If your LM3x7 circuits work with ceramics on the output it does not mean that the Denoiser increases the requirements vs datasheet. It means it just makes the LM3x7 margins more apparent. Hence why the Denoiser is stated many times by Elvee to be used on regular datasheet applications."
You having a working LM3x7 circuit with ceramics on the output, and having a not working Denoiser circuit on the same regulator does not mean that the Denoiser induces extra requirements, it means you were using the LM3x7 circuit in ways that is not intended in the datasheet, and the Denoiser showed you why.
The last statement should be read as: "I understand that some people are pushing the limits of LM3x7, some are testing these limits, some are plain lucky. The Denoiser is stated to be used on datasheet regular circuits. Going out of these bounds does not guarantee it works. If your LM3x7 circuits work with ceramics on the output it does not mean that the Denoiser increases the requirements vs datasheet. It means it just makes the LM3x7 margins more apparent. Hence why the Denoiser is stated many times by Elvee to be used on regular datasheet applications."
You having a working LM3x7 circuit with ceramics on the output, and having a not working Denoiser circuit on the same regulator does not mean that the Denoiser induces extra requirements, it means you were using the LM3x7 circuit in ways that is not intended in the datasheet, and the Denoiser showed you why.
I've updated the complete dual supply, tht version at the moment. This can also be diy-ed as it's one side only. I've used only the denoiser circuit, and has the Rcomp for LM337 as well.
This has not been tested, you make it at your own risk. There's gerbers and pdf for diy in the attached archive, along with the kicad project.
edit: I won't update the smd version as there's not much to gain size wise. On the basis that the dienoiser had issues with the lm337 and there's no working example for a lm337+nonoiser I hereby invalidate all previous projects with dienoiser/nonoiser versions for lm337. Only denoiser versions are to be used if you are expecting good results. There's high chances they'll work on any of the updated designs I posted today.
This has not been tested, you make it at your own risk. There's gerbers and pdf for diy in the attached archive, along with the kicad project.
edit: I won't update the smd version as there's not much to gain size wise. On the basis that the dienoiser had issues with the lm337 and there's no working example for a lm337+nonoiser I hereby invalidate all previous projects with dienoiser/nonoiser versions for lm337. Only denoiser versions are to be used if you are expecting good results. There's high chances they'll work on any of the updated designs I posted today.
Last edited:
I just measured the ripple on both reg inputs and it's around 0.72Vpp for the negative and 0.65Vpp for the positive. I used LTSpice to work out the current, I put 24VAC for the supply, with around 1.25ohm resistance of the secondary winding of the transformer, 2200uF input cap after the rectifying diode, and to get those ripples it means that the negative draws around 89mA and the positive around 77mA. And it makes sense since the relay coil is on the negative supply. The relay is a NEC UB2-12NU for which the coil operates at 12V and 140mW which means around 12mA, and it all makes sense. So no need for extra load resistor, it's past 50mA on both rails.
I had some time to listen to it, I'm using a pair of AKG K712Pro and it seems that I can perceive more details, especially on highs. It seems like I'm able to focus on tiny details. I don't know if I'm imagining it or not but for sure I get this feeling.
I wish someone else would do the denoiser mod to a headphone amp (which has a similar supply) and tell us if they perceive the same thing.
Made a DIY version for a full wave voltage doubler so you can have +- supplies from a single secondary transformer. It's what I used on the LNA and what's used in the Magni. It's very convenient. The drawback is that the ripple is 50Hz instead of 100Hz and needs double the input capacitance, but using the denoiser you can get away with not much input capacitance, depending on your needs. Do calculate the output ripple taking the denoiser into account and see if it fits your application. I also added a jumper so you could wire in an on/off switch.
edit: also warning, this board is not tested, you make it at your own risk.
edit: also warning, this board is not tested, you make it at your own risk.
Last edited:
We have discussed about using half wave rectifiers here but that was for separate pcbs. This way, we can use wall wart AC-AC adapters which makes casing work is easier and avoids transformer interference.
But I still wonder why LM337 didn't behave like LM317. Did you just gave it up about negative dienoiser? Will you build a negative nonoiser? It would be also interesting to compare LT1033 if you have around in place of LM337.
I made Elvee's oscillation sniffer , not sure what to expect .
Even with the power off , I measure around 0,6 mV dc , with power on , me not touching anything, gets down to around or less than 1mVdc but 20mV ac. Good thing is , it's like this for all 4 Denoisers , so I guess no oscillation . Not much change with load or no load .
No shielding and when I touch leads , voltages jumps up.
I tried a LM317 with no Denoiser and no AdjCap with 5 V on its output , only a couple of mV dc more , ac the same.
Good that I made the sniffer to check out other regulators and opamps.
Even with the power off , I measure around 0,6 mV dc , with power on , me not touching anything, gets down to around or less than 1mVdc but 20mV ac. Good thing is , it's like this for all 4 Denoisers , so I guess no oscillation . Not much change with load or no load .
No shielding and when I touch leads , voltages jumps up.
I tried a LM317 with no Denoiser and no AdjCap with 5 V on its output , only a couple of mV dc more , ac the same.
Good that I made the sniffer to check out other regulators and opamps.
I didn't have time to tinker further with the dienoiser. I couldn't make it work in the dual supply for the LNA, and I tried two different to220 lm337, both behaved the same. And I searched this thread and found no posts about anyone making a lm337 dienoiser, or nonoiser. Since the nonoiser pushes the regulator even further I didn't even consider trying it. I personally think it doesn't work, or it's very touchy. I hope someone proves me wrong, but I don't have time for that. I think I've released some designs for nonoiser add-ons and also full supplies, if anyone wants to diy and test they are welcome.
I am really happy with the denoiser as it is. It pushed the 50hz ripple into the noise floor. I don't really want or need more.
The only way that you could succesfully use the dienoiser or nonoiser would be to build lm317 supplies using two transformer windings. That way you could have positive and negative with dienoiser or nonoiser.
I decided to directly measure the power supply from the Magni. Previous measurements were of the audio outputs.
The setup was real touchy as I had to solder a cable inside the headamp so I didn't put the cover. To rule out the errors of the crappy oscilloscope I calculated the attenuation relatively between the 60dB setting and 20dB setting. The gain settings of the LNA should be really close to spec. I used 60dB setting for the output of the regs, directly on the output cap, and I used the 20dB setting on the input of the reg directly on the regulator. Negative input shows around -7.40dB and with 20dB from the LNA it works out to -27.40dB. Negative output shows -70.14dB and with 60dB from the LNA it works out to -130.14dB. So the difference is about 102.74dB. Positive worked out to around 106dB I think, but I didn't take screenshots for that.
This is for negative input on 20dB gain:
And this is for negative output on 60dB gain:
So around 100dB I'd call this a successful application of the denoiser.
The setup was real touchy as I had to solder a cable inside the headamp so I didn't put the cover. To rule out the errors of the crappy oscilloscope I calculated the attenuation relatively between the 60dB setting and 20dB setting. The gain settings of the LNA should be really close to spec. I used 60dB setting for the output of the regs, directly on the output cap, and I used the 20dB setting on the input of the reg directly on the regulator. Negative input shows around -7.40dB and with 20dB from the LNA it works out to -27.40dB. Negative output shows -70.14dB and with 60dB from the LNA it works out to -130.14dB. So the difference is about 102.74dB. Positive worked out to around 106dB I think, but I didn't take screenshots for that.
This is for negative input on 20dB gain:
And this is for negative output on 60dB gain:
So around 100dB I'd call this a successful application of the denoiser.
Ah crap, I checked the specs for the input of the LNA and it's 130mVpp max for 20dB gain. Ripple was higher on the input of the reg, so it clipped for sure. It has 1.3Vpp/460mVrms output limit. That's roughly -9dB from 1.3Vrms max input of the ADC and measurement seems thereabouts. I have to take that measurement again. Since the ripple is high enough I think I could get away with directly connecting to the input of the ADC.
edit:
But since I measured the input ripple with that crappy oscilloscope which seemed to somewhat agree with the dmm on the output DC, that was around 0.72Vpp for the negative input. That means 0.255Vrms. 60dB measurement showed -70.14dB. Adding the 60dB means 130.14dB. Out of the 1.3Vrms 0dB for the ADC, means around 0.4uVrms. The difference between 0.255Vrms and 0.4uVrms is around 116dB! Is that possible for only the denoiser?
Positive rail was around -77dB on 60dB gain, that means -137dB. 0.65Vpp (roughly) ripple is 0.230Vrms. -137dB from 1.3Vrms is 0.18uV. Difference between 0.230Vrms and 0.18uV is almost 122dB! This seems way more than the denoiser is specified at. Maybe I've done something wrong.
edit:
But since I measured the input ripple with that crappy oscilloscope which seemed to somewhat agree with the dmm on the output DC, that was around 0.72Vpp for the negative input. That means 0.255Vrms. 60dB measurement showed -70.14dB. Adding the 60dB means 130.14dB. Out of the 1.3Vrms 0dB for the ADC, means around 0.4uVrms. The difference between 0.255Vrms and 0.4uVrms is around 116dB! Is that possible for only the denoiser?
Positive rail was around -77dB on 60dB gain, that means -137dB. 0.65Vpp (roughly) ripple is 0.230Vrms. -137dB from 1.3Vrms is 0.18uV. Difference between 0.230Vrms and 0.18uV is almost 122dB! This seems way more than the denoiser is specified at. Maybe I've done something wrong.
Last edited: