I just mean the smaller smd parts: passives and small signal transistors. The SOT-223 regulators are large enough for me to handle.
If you'd like something to test and be easier to solder and still be able to use LT3082 or LM317 in sot223 package I made this board that can also be diyed. Everything is tht apart from the regulator.
There's a small 5mm input cap for some input capacitance, 8mm output cap so you could use caps that would have the correct ESR/ESL. The coupling cap is 6.3mm diameter so you can use either 3.3-4.7uF for LT3082 either 220uF for LM317. There's a comp cap footprint, LT3082 dienoiser doesn't need any, and the LM317 dienoiser is ok with just 22nF and no R. Output cap for LM317 should have around 150-200mOhm ESR. The regulator is on the backside of the pcb, I also put some extra copper on the sides of the tab for some extra cooling. There's also a smd jumper on the backside to switch between denoiser and dienoiser. There's pdf for pcb traces for diy, gerber files in case anyone wants that, and photos with part layout and schematic. For LT3082 you don't install the 220R resistor, that's for LM317 only.
I will test in the near future Panasonic FR 150uF/63V. It should work for LT3082/LM337 dienoiser output cap. It's 8mm diameter, around 50mOhm ESR and is 20mm long. For LM317 dienoiser Panasonic FR 47uF/63V seems has the ESR in the correct area but I have not tried it. I successfully used Panasonic FC 100uF/63V (8mm), with ESR of 230mOhm.
edit: I made the board with a bit shorter resistors than in photo. Size is around 23mm x 29mm.
There's a small 5mm input cap for some input capacitance, 8mm output cap so you could use caps that would have the correct ESR/ESL. The coupling cap is 6.3mm diameter so you can use either 3.3-4.7uF for LT3082 either 220uF for LM317. There's a comp cap footprint, LT3082 dienoiser doesn't need any, and the LM317 dienoiser is ok with just 22nF and no R. Output cap for LM317 should have around 150-200mOhm ESR. The regulator is on the backside of the pcb, I also put some extra copper on the sides of the tab for some extra cooling. There's also a smd jumper on the backside to switch between denoiser and dienoiser. There's pdf for pcb traces for diy, gerber files in case anyone wants that, and photos with part layout and schematic. For LT3082 you don't install the 220R resistor, that's for LM317 only.
I will test in the near future Panasonic FR 150uF/63V. It should work for LT3082/LM337 dienoiser output cap. It's 8mm diameter, around 50mOhm ESR and is 20mm long. For LM317 dienoiser Panasonic FR 47uF/63V seems has the ESR in the correct area but I have not tried it. I successfully used Panasonic FC 100uF/63V (8mm), with ESR of 230mOhm.
edit: I made the board with a bit shorter resistors than in photo. Size is around 23mm x 29mm.
Attachments
Last edited:
I was curious about high current output low dropout regulators from Linear and from simulations seems like LT3083 would be a good alternative. The advertised 310mV drop is available by connecting a higher than input voltage to Vcontrol pin so not so simple to implement. But from simulations it does have a lower drop in comparison to LM317 or LT1085. It's a 3A device, so this could be used for filament supplies in tube gear for example. PSRR with dienoiser is higher for 50-100Hz ripple than LM317/LT1085 + dienoiser.
LM317 seems to need min 2.4V across it for best performance, LT1085 seems to need around 2V and LT3083 about 1.3V. That means that for 6.3V at 3A out it needs a 9VAC output on transformer secondary, maybe even less depending on rectifying diodes. The only downside seems to be the price, around 10$ for the to220-5 version.
It does seem to have the LT3082 internal setup with the current source, and seems to work similarly with dienoiser without needing a compensation network for stability.
The simulation is for 3A output for LT1085/LT3083 and 1.5A output for LM317.
LM317 seems to need min 2.4V across it for best performance, LT1085 seems to need around 2V and LT3083 about 1.3V. That means that for 6.3V at 3A out it needs a 9VAC output on transformer secondary, maybe even less depending on rectifying diodes. The only downside seems to be the price, around 10$ for the to220-5 version.
It does seem to have the LT3082 internal setup with the current source, and seems to work similarly with dienoiser without needing a compensation network for stability.
The simulation is for 3A output for LT1085/LT3083 and 1.5A output for LM317.
Attachments
@Trileru, please don't offend but I have to notice that your experiments on different devices and different approaches varies too much in a short of time without putting required real world experiments as outcome.
Since you have prepared yourself with devices like LNA, why don't you support your circuits with real world measurements? I guess other followers like myself expecting some useful outcome along with personal experiments.
Since you have prepared yourself with devices like LNA, why don't you support your circuits with real world measurements? I guess other followers like myself expecting some useful outcome along with personal experiments.
I posted measurements for real applications earlier in the thread. I have used the dienoisers add-ons in my Schiit Magni 3 headphone amp, successfully. I have also used them in a AK4396 DAC for opamps supply + Vref/VACC and clock recovery rails. In the DAC I used LT1085 instead of LM317. I don't really have more real world applications to measure.
I got LT3082 to use in the new DAC pcb that I designed. When I assemble it I'll measure it and post the measurements. The pcb designs I posted are more of a by-product of my tinkering with the de/dienoiser circuits. Others are welcomed to test them and post measurements, especially if they have better gear, to offer a better picture of the performance. Since I don't have the necessary measurement gear to fully vet them nor do I have the interest of selling any or invest more resources into them I won't go further into each design.
My posts have the intention of helping others that are interested in applying these low noise and cheap supplies into their own gear/products.
I got LT3082 to use in the new DAC pcb that I designed. When I assemble it I'll measure it and post the measurements. The pcb designs I posted are more of a by-product of my tinkering with the de/dienoiser circuits. Others are welcomed to test them and post measurements, especially if they have better gear, to offer a better picture of the performance. Since I don't have the necessary measurement gear to fully vet them nor do I have the interest of selling any or invest more resources into them I won't go further into each design.
My posts have the intention of helping others that are interested in applying these low noise and cheap supplies into their own gear/products.
Please consider my thoughts as a grain of salt. I was following the thread for couple of months but I ended up with buying LT3042 and LT3045/LT3094 based kits. Because every option besides bare denoiser that put on this thread seemed to me very experimental.
From my experience of other DIY threads, people are gathering their build experiences on a tested design to help others. But this didn't happen in this ~200 pages of thread because the reasons that I mentioned above. For instance, you came up with a good idea based on LT3082 with nice looking PCB's but you suddenly switched to other version with more amperes. At this point, this idea tends to get lost in thread because attention moves toward new one. I'm not saying don't come up with new things but at least you could try to support your new ideas with real world measurements, stability observations etc. in order encourage others.
From my experience of other DIY threads, people are gathering their build experiences on a tested design to help others. But this didn't happen in this ~200 pages of thread because the reasons that I mentioned above. For instance, you came up with a good idea based on LT3082 with nice looking PCB's but you suddenly switched to other version with more amperes. At this point, this idea tends to get lost in thread because attention moves toward new one. I'm not saying don't come up with new things but at least you could try to support your new ideas with real world measurements, stability observations etc. in order encourage others.
Wish I could pick a dienoiser version to build but it didn't happen. So, I can't compare. But to compare with bare LT1085/LM337 board, trebles are abviously more defined on LT3045/LT3094 one. I spared LT3042 boards for digital sections. My NOS PCM1792 dac project is on progress. I simulated I/V stage on Ltspice with proper feedback network yet.
This is the board I'm working on
I made the second LT3082 version as it's tht and can be made diy. Some people prefer this more than smallest size. For example even if I posted the tht design after the smd design, it has 14 downloads vs 9 downloads for the smd version. Also interestingly the LM317 version has 10 downloads at this time, one more than LT3082. I'm just trying to increase the chances of people making any of them and post some feedback/measurements. But yes, I understand how a flood of designs make it hard to concentrate on any. They'll all still be accessible in the future so at least that is a good thing. Maybe I make separate threads for the most desired?
The small smd LT3082 + dienoiser design I'll send to the fab house along with the similar LM3x7 versions when I'll send the DAC pcb files as well. So I'll test all three of those.
In reality after testing some of my designs I think there's high chances all work similarly. Not sure about the latest small ones as they are double layer, but the single layer ones should behave the same. Of course there's still chances of surprises at higher frequencies, sadly I don't have gear to look at that spectrum and see if there's anything bad happening.
The small smd LT3082 + dienoiser design I'll send to the fab house along with the similar LM3x7 versions when I'll send the DAC pcb files as well. So I'll test all three of those.
In reality after testing some of my designs I think there's high chances all work similarly. Not sure about the latest small ones as they are double layer, but the single layer ones should behave the same. Of course there's still chances of surprises at higher frequencies, sadly I don't have gear to look at that spectrum and see if there's anything bad happening.
Plus the designs are for different applications. You should make the one that suits you best. For example the tht full supply versions in post #1889 are for small power amplifiers and preamps, and the smd full supply versions from post #1934 are for small power applications where size matters, like DACs or preamps/headamps. The three pin versions can be used to retrofit existing gear or design them very close to the loads. They each have a slightly different application. The same with the add-on boards. There's single and dual versions, smd and tht. Also I tried to make most diy friendly. This way I maximize the chances of people making them and giving some feedback.
Got curious about HV regulators like 21st Century Maida, since it uses a LT3080. I'm not sure if I correctly recreated it in LTSpice and also if I correctly adapted the dienoiser but it seems to be working, also without comp network.
The PSRR is already pretty good for the normal regulator, and dienoiser adds on top of that. What I consider meaningful gains are a serious lowering of the self-noise (lower noisefloor). There's also the crazy low output impedance as well but I don't know if that matters that much at the hundreds of mA level.
I increased the 1.8K resistor to 33K, there's the extra 10mA draw from the dienoiser circuit. That resistor has to be high power type as it dissipates few watts. Both dienoiser circuit caps need to be HV rated. But apart from that looks like the normal transistors could be used?
The value of the output voltage setting resistor changes a bit.
I attached the sim file. I want to also give a serious warning to not play with high voltage! It's always a big risk even for experienced people! I'd also not try to adapt an existing regulator pcb. If this really works and is needed people should wait for products that have it built in, and have been also tested. I won't attempt any practical testing for this.
First photo is PSRR, second is self-noise reduction, third is output impedance and fourth is a zoom-in for the dienoiser version output impedance.
The PSRR is already pretty good for the normal regulator, and dienoiser adds on top of that. What I consider meaningful gains are a serious lowering of the self-noise (lower noisefloor). There's also the crazy low output impedance as well but I don't know if that matters that much at the hundreds of mA level.
I increased the 1.8K resistor to 33K, there's the extra 10mA draw from the dienoiser circuit. That resistor has to be high power type as it dissipates few watts. Both dienoiser circuit caps need to be HV rated. But apart from that looks like the normal transistors could be used?
The value of the output voltage setting resistor changes a bit.
I attached the sim file. I want to also give a serious warning to not play with high voltage! It's always a big risk even for experienced people! I'd also not try to adapt an existing regulator pcb. If this really works and is needed people should wait for products that have it built in, and have been also tested. I won't attempt any practical testing for this.
First photo is PSRR, second is self-noise reduction, third is output impedance and fourth is a zoom-in for the dienoiser version output impedance.
Attachments
Found some time to test the LM317N TO220 version (from TI). I measured it and LM317 from ST. Both are marked LM317T, but as far as I understand the one from TI is the N version while the one from ST is the normal one. The normal version from TI is marked LM317K. The price of LM317K from TI is around 60cents the same as the LM317T from ST. The LM317T (N version) from ST costs about 1.6$, so more than double the price.
Measurements for PSRR show a very small performance gain for the LM317 N version. I don't think it's worth the price difference. There might be larger performance gains in output impedance or higher frequencies PSRR but I cannot test for that.
Attached measurements in order: normal configuration without Cadj, with Cadj, denoiser and last dienoiser. The yellow trace is LM317 from ST and red trace is LM317N from TI.
Last picture is a measurement for the SLNR denoiser made by Diego. I used a 860K resistor I think, output voltage was around 9V. I'm not sure what the issue was but the performance was bad compared to even Cadj version. It was working as in all measurements I used the 60dB LNA so if there was anything bad it would have overloaded the ADC. I made it on perfboard and it was pretty simple as layout, don't know what the issue is. I used a BC550C.
Measurements for PSRR show a very small performance gain for the LM317 N version. I don't think it's worth the price difference. There might be larger performance gains in output impedance or higher frequencies PSRR but I cannot test for that.
Attached measurements in order: normal configuration without Cadj, with Cadj, denoiser and last dienoiser. The yellow trace is LM317 from ST and red trace is LM317N from TI.
Last picture is a measurement for the SLNR denoiser made by Diego. I used a 860K resistor I think, output voltage was around 9V. I'm not sure what the issue was but the performance was bad compared to even Cadj version. It was working as in all measurements I used the 60dB LNA so if there was anything bad it would have overloaded the ADC. I made it on perfboard and it was pretty simple as layout, don't know what the issue is. I used a BC550C.
Attachments
Ah sorry forgot to mention, it's the same tht board I used for previous tht lm317/lt1085 measurements, it's an early design from post #1402
I use a 14VAC rms transformer, has a secondary resistance of around 1.25ohms. Cap values after rectifier are 470uF + 220uF (edit: I didn't use high values so I'll have some ripple in the end. Upping the value here could bury the output ripple into the noisefloor with dienoiser). LM317 resistors are 200R/1.6K for around 11.25Vout. Load is 150R resistor. Comp network is made out of two 10nF in parallel, no R. 100uF output cap with 0.2ohm ESR. I used no inductor or resistor on the input CLC/CRC, I shorted that footprint. For Cadj measurements I used a 220uF cap. The load resistor is soldered on the output of the pcb. I use a thin coax cable that is soldered on the load pins and on the LNA pcb on each side.
Using these values in LTSpice sim we see an input ripple of 1Vpp, output ripple of 530nV for dienoiser. My ADC has 0dB point set at 3672mVpp (1.298Vrms).
Take the measurement for dienoiser at around -80dB at 100Hz on the ARTA graph, adding the 60dB from LNA means -140dB from 1.298V which is 0.13uVrms. Checking the LTSpice sim the input ripple is 1Vpp (0.353Vrms), which means around -11.3dB from my 0dB on ADC. So the dienoiser attenuation at 100Hz measures at around -140dB from which we subtract the input ripple of -11.3dB which works out at around -128.7dB for the measured LM317+dienoiser attenuation. From simulation input ripple is at 0.353Vrms and output ripple at 529nV and the difference is 125.5dB. So measurement and simulation are within 3dB or so of each-other. I guess there's other factors at this level that I didn't account for in the simulation, the two caps after rectifier ESR, track resistances etc. I set the 0dB based on the datasheet of my motherboard's audio chip (alc892), but the integration on the board might differ slightly.
It seems that Ti's spice model for LM317 is pretty accurate and closer to real LM317 regulators than the RH117K model.
edit: from my testing so far it seems LM317 + dienosier works fine with around 22nF for comp cap and no R (with 100uF output cap with around 0.2Ohm ESR). This "should" work on most implementations. For LM337 + dienoiser it seems that 47nF+3.9R works fine (as long as the output cap requirements are met, at least 10uF, around 50mOhm max total ESR and around 30nH).
I use a 14VAC rms transformer, has a secondary resistance of around 1.25ohms. Cap values after rectifier are 470uF + 220uF (edit: I didn't use high values so I'll have some ripple in the end. Upping the value here could bury the output ripple into the noisefloor with dienoiser). LM317 resistors are 200R/1.6K for around 11.25Vout. Load is 150R resistor. Comp network is made out of two 10nF in parallel, no R. 100uF output cap with 0.2ohm ESR. I used no inductor or resistor on the input CLC/CRC, I shorted that footprint. For Cadj measurements I used a 220uF cap. The load resistor is soldered on the output of the pcb. I use a thin coax cable that is soldered on the load pins and on the LNA pcb on each side.
Using these values in LTSpice sim we see an input ripple of 1Vpp, output ripple of 530nV for dienoiser. My ADC has 0dB point set at 3672mVpp (1.298Vrms).
Take the measurement for dienoiser at around -80dB at 100Hz on the ARTA graph, adding the 60dB from LNA means -140dB from 1.298V which is 0.13uVrms. Checking the LTSpice sim the input ripple is 1Vpp (0.353Vrms), which means around -11.3dB from my 0dB on ADC. So the dienoiser attenuation at 100Hz measures at around -140dB from which we subtract the input ripple of -11.3dB which works out at around -128.7dB for the measured LM317+dienoiser attenuation. From simulation input ripple is at 0.353Vrms and output ripple at 529nV and the difference is 125.5dB. So measurement and simulation are within 3dB or so of each-other. I guess there's other factors at this level that I didn't account for in the simulation, the two caps after rectifier ESR, track resistances etc. I set the 0dB based on the datasheet of my motherboard's audio chip (alc892), but the integration on the board might differ slightly.
It seems that Ti's spice model for LM317 is pretty accurate and closer to real LM317 regulators than the RH117K model.
edit: from my testing so far it seems LM317 + dienosier works fine with around 22nF for comp cap and no R (with 100uF output cap with around 0.2Ohm ESR). This "should" work on most implementations. For LM337 + dienoiser it seems that 47nF+3.9R works fine (as long as the output cap requirements are met, at least 10uF, around 50mOhm max total ESR and around 30nH).
Last edited:
Using a voltage doubler one could use the LT3083+dienoiser with the advertised 0.31Vdrop. The input capacitors will have to be higher value but for a 3A device with 0.31Vdrop might worth it depending on case. Care should be taken with taking the secondary transformer resistance into account but maybe this is interesting to someone. The only downside is that the regulator is on the expensive side.
I attached the sim file.
edit: unless both rails have to be equally loaded?
I attached the sim file.
edit: unless both rails have to be equally loaded?
Last edited: