Finesse was created by Charles Wenzel, a few years before Elvee modified and improved Finesse to create his DeNoiser. Mr. Wenzel has rearranged his website a few times since then, the most recent web path appears to be (here) .
diyAudio was aware of Finesse more than thirteen years ago, here is (a thread) about it from March 2011.
diyAudio was aware of Finesse more than thirteen years ago, here is (a thread) about it from March 2011.
The "clean up circuit" I see in the Finesse at the above link has nothing to do with the Elvee’s denoisers.
Apart of being "add on"s to a LM317 regulator, they are entirely different noise reducing technics.
Apart of being "add on"s to a LM317 regulator, they are entirely different noise reducing technics.
They are different, yet related: both use an external amplifier to improve the performance of an existing regulator. However, the finesse relies on a series resistance to operate which either/or degrades the output impedance and the effectiveness of the correction.
The denoiser is globally more effective, but it requires a surgical intervention into the existing regulator
The denoiser is globally more effective, but it requires a surgical intervention into the existing regulator
I remember reading something on his webpage, quite a few years back, but it was general considerations for LM3x7 supplies if I remember right.
Anyway, made first measurements with transformer powering the board, resistive load and sensing on output connector (at the moment). Still have to assemble the dynamic load PCB.
So far it looks good, ~130dB of PSRR for both positive and negative rails. And under 150nV total noise for both, with 150R load, which comes at around 110-120mA total for each rail (15Vout).
Positive has some extra LF stuff going on but might be either the MOSFET I used or some other thing, I'll see if I can get it better, but highest signal is at -140dBV which is pretty fine for a voltage doubler.
Positive PSRR:
Positive noisefloor:
Negative PSRR:
Negative noisefloor:
And schematic with part values I used:
Works fine with ~100mOhm ESR for output caps on both rails. 22nF compensation is stable (tested for stability with electronic load).
Might still tweak a few values but looks good as it is. Addon boards already have the final length wires that can be seen in my previous post photo, so I'm not expecting much change in performance but I guess I'll find out.
I still want to try a higher Vf LED for Vref, blue or white. I'll update with final measurements once I get to test with dynamic load, also get output impedance.
edit: forgot to mention, I added a small 47uF in parallel with 2200uF input cap, on each rail, and output caps were actually 330uF for negative rail and 470uF for positive, just what I found in the cap box for ~100mOhm ESR. Will try a few more just to make sure.
Also adjusting Vout is really sensitive. Needed ~107K (R18/R19 in schematic) for 15Vout and the extra 7K makes for 1V. I think it's because of the low Vf of the LED. So it would be better to add a series resistor to take the bulk of the value and use a lower value multi turn pot. I struggled to get it at 15V with a 200K single turn pot.
Will also try to lower Vref current, I used 2.2K which was better for BJT pass transistor but I'll test with 3.3K and 4.7K see what happens.
Anyway, made first measurements with transformer powering the board, resistive load and sensing on output connector (at the moment). Still have to assemble the dynamic load PCB.
So far it looks good, ~130dB of PSRR for both positive and negative rails. And under 150nV total noise for both, with 150R load, which comes at around 110-120mA total for each rail (15Vout).
Positive has some extra LF stuff going on but might be either the MOSFET I used or some other thing, I'll see if I can get it better, but highest signal is at -140dBV which is pretty fine for a voltage doubler.
Positive PSRR:
Positive noisefloor:
Negative PSRR:
Negative noisefloor:
And schematic with part values I used:
Works fine with ~100mOhm ESR for output caps on both rails. 22nF compensation is stable (tested for stability with electronic load).
Might still tweak a few values but looks good as it is. Addon boards already have the final length wires that can be seen in my previous post photo, so I'm not expecting much change in performance but I guess I'll find out.
I still want to try a higher Vf LED for Vref, blue or white. I'll update with final measurements once I get to test with dynamic load, also get output impedance.
edit: forgot to mention, I added a small 47uF in parallel with 2200uF input cap, on each rail, and output caps were actually 330uF for negative rail and 470uF for positive, just what I found in the cap box for ~100mOhm ESR. Will try a few more just to make sure.
Also adjusting Vout is really sensitive. Needed ~107K (R18/R19 in schematic) for 15Vout and the extra 7K makes for 1V. I think it's because of the low Vf of the LED. So it would be better to add a series resistor to take the bulk of the value and use a lower value multi turn pot. I struggled to get it at 15V with a 200K single turn pot.
Will also try to lower Vref current, I used 2.2K which was better for BJT pass transistor but I'll test with 3.3K and 4.7K see what happens.
Last edited:
Managed to measure the positive rail at load. Seems that sense wire length doesn't matter that much. I tried some 20cm and worked fine. I tested a thin coax that I recovered from an old wifi router, it connected the router PCB to antenna connector. Worked just a bit better than enameled wire, but with shield floating. If I connected it to ground on either end it would be a bit worse overall, noise and PSRR. Not sure if it's worth doing it. Plain wire is fine.
Main Vout cables had to be twisted together or else worse results. Would it be better to have all three twisted together or have a twisted pair for each rail?
Load PCB I designed is not so good, whenever my phone would output signal the noisefloor would shoot up, but once I muted it/grounded the input, noisefloor was low.
Noise came out under 200nV with load PCB grounded signal input. There was 7.5VDC across the 75R so 100mA static load.
Tried to test output impedance but DMM isn't reliable for AC across the load resistor, and my ADC is limited to 1V RMS input so I had to do it at lower AC current, some 8mA AC across 75R resistor. Output impedance worked out at about 100uOhms at 100-200Hz and 1mOhm at 20kHz, but I don't really have the gear or knowledge to properly do it.
Noise is pretty close to onboard sensing so I suppose output impedance should be good as well. No issues with stability.
All in all I'd clearly use remote sensing but ideally you sense in a spot where V/GND are relatively close, keep addon sense wires twisted right to where you are sensing and make sure there's not much distance between V and GND sense points.
Main Vout cables had to be twisted together or else worse results. Would it be better to have all three twisted together or have a twisted pair for each rail?
Load PCB I designed is not so good, whenever my phone would output signal the noisefloor would shoot up, but once I muted it/grounded the input, noisefloor was low.
Noise came out under 200nV with load PCB grounded signal input. There was 7.5VDC across the 75R so 100mA static load.
Tried to test output impedance but DMM isn't reliable for AC across the load resistor, and my ADC is limited to 1V RMS input so I had to do it at lower AC current, some 8mA AC across 75R resistor. Output impedance worked out at about 100uOhms at 100-200Hz and 1mOhm at 20kHz, but I don't really have the gear or knowledge to properly do it.
Noise is pretty close to onboard sensing so I suppose output impedance should be good as well. No issues with stability.
All in all I'd clearly use remote sensing but ideally you sense in a spot where V/GND are relatively close, keep addon sense wires twisted right to where you are sensing and make sure there's not much distance between V and GND sense points.
Here's some of yesterday's measurements for the positive rail:
I used V RMS scale for this graph. Upper trace is ADC directly sensing across the 75R load resistor with some AC signal into the remote PCB load. Makes for 11mA RMS current from supply.
Yellow trace is LNA sensing on the power rail at the load. The noisefloor is higher because I messed up the grounding on the remote load, but ignoring it we see the 125Hz peak at 1.5uV. For 11mA it means an impedance of 136uOhms.
Green bottom trace is like yellow trace, LNA sensing the power rail but with signal input grounded on load PCB.
Switching to dBV scale for that trace:
We see that 50Hz peak is even lower than last measurement I made with sensing on power supply output connector. Still 100mA load with 7.5VDC across the 75R resistor, so ripple on the input of the regulator should be the same. So we get a bit better PSRR at remote load, and noisefloor is low (apart from the extra picked up crap without shielding/case).
So I think performance is there at remote load.
Here's some more output impedance measurements for other frequencies:
Same as before upper trace ADC sensing across the 75R resistor, lower trace LNA sensing the power rail. Still 11mA and lower trace peak is at 1.35uV so impedance is 122uOhms at 1kHz.
Here's 18kHz:
This works out at 1mOhm.
edit:
This is the schematic for the remote load I used:
I used V RMS scale for this graph. Upper trace is ADC directly sensing across the 75R load resistor with some AC signal into the remote PCB load. Makes for 11mA RMS current from supply.
Yellow trace is LNA sensing on the power rail at the load. The noisefloor is higher because I messed up the grounding on the remote load, but ignoring it we see the 125Hz peak at 1.5uV. For 11mA it means an impedance of 136uOhms.
Green bottom trace is like yellow trace, LNA sensing the power rail but with signal input grounded on load PCB.
Switching to dBV scale for that trace:
We see that 50Hz peak is even lower than last measurement I made with sensing on power supply output connector. Still 100mA load with 7.5VDC across the 75R resistor, so ripple on the input of the regulator should be the same. So we get a bit better PSRR at remote load, and noisefloor is low (apart from the extra picked up crap without shielding/case).
So I think performance is there at remote load.
Here's some more output impedance measurements for other frequencies:
Same as before upper trace ADC sensing across the 75R resistor, lower trace LNA sensing the power rail. Still 11mA and lower trace peak is at 1.35uV so impedance is 122uOhms at 1kHz.
Here's 18kHz:
This works out at 1mOhm.
edit:
This is the schematic for the remote load I used:
Last edited:
Tried a blue LED (2.8Vf) for Vref and works fine, even better for voltage regulation. I increased its resistor from 2.2K to 3.9K.
I measured with denoisers sensing on supply board output connector, 150R load on each rail output, transformer on the input.
Upper trace is on the input of the regulator, bottom traces on the output. PSRR is ~128dB for positive and ~130dB for negative. Must be the MOSFET or some other thing on the positive, but they're pretty close anyway.
Noise is pretty close as well, 125nV for negative and 145nV for positive. I'm going by the 22Hz-22kHz UNW value, that seems to limit it to audio spectrum only. I do have bypass active in REW settings but "Input RMS" value varies (wildly sometimes) with lower than 20Hz activity. The UNW value seems steady whenever audio spectrum is steady.
From now on I'll use blue LEDs where possible, seems a better choice. At least for 12Vout and higher.
Speaking of which, +/-5Vout should also be possible, with red/yellow/orange LEDs at most, you could also use IR.
Schematic example values for +/-5Vout. I might measure this setup as well just for completeness sake.
edit: Curious about 365nm UV LEDs. Should have higher Vf. I might get some at some point.
I measured with denoisers sensing on supply board output connector, 150R load on each rail output, transformer on the input.
Upper trace is on the input of the regulator, bottom traces on the output. PSRR is ~128dB for positive and ~130dB for negative. Must be the MOSFET or some other thing on the positive, but they're pretty close anyway.
Noise is pretty close as well, 125nV for negative and 145nV for positive. I'm going by the 22Hz-22kHz UNW value, that seems to limit it to audio spectrum only. I do have bypass active in REW settings but "Input RMS" value varies (wildly sometimes) with lower than 20Hz activity. The UNW value seems steady whenever audio spectrum is steady.
From now on I'll use blue LEDs where possible, seems a better choice. At least for 12Vout and higher.
Speaking of which, +/-5Vout should also be possible, with red/yellow/orange LEDs at most, you could also use IR.
Schematic example values for +/-5Vout. I might measure this setup as well just for completeness sake.
edit: Curious about 365nm UV LEDs. Should have higher Vf. I might get some at some point.
Last edited:
Made a set of PCBs for latest version of the circuit.
There's AC input and DC input versions. Each has a single or dual version. The AC input ones have local and remote sensing versions while the DC input ones have only remote sensing versions. The addon denoiser boards are separate and have positive and negative versions, the negative is mirrored design.
The DC input versions don't have local sensing as I was looking for smallest size, you can adapt AC input to DC input by omitting rectifying diodes and shorting two of them.
AC input dual versions can work with single secondary voltage doubler or dual secondary transformer. For voltage doubler just short the negative side AC input to middle GND in connector.
For 200mAout for each rail either 2200uF for voltage doubler either 1000uF for dual secondaries should be enough.
All are DIY-able, traces on bottom side with no topside passes.
Addon positive board:
Addon negative board:
AC input dual remote sensing:
AC input dual local sensing:
AC input single remote sensing:
AC input single local sensing:
DC input single remote sensing:
DC input dual remote sensing:
I have attached the Kicad project files, also contains gerbers zip file and DIY and schematic PDFs.
If you have questions about part values ask here. These haven't been tested so you make them at your own risk. I did test the AC input dual version with remote sensing.
All remote sensing boards have the DC sensing voltage divider resistors on board. They are also on the addon boards so you choose to populate one or the other. So the remote sensing supply boards can also be used as a regular discrete regulator, but you'll have worse AC performance (~80dB of PSRR at 50Hz)
This PSRR measurement did have a 22uF cap across R2.
Trimpots have two regular resistors in parallel on top so it's either pots either resistors.
The blue LED is close to the BJT on all boards, so you can have them touch or maybe add a bit of thermal paste between them. Don't know how critical this is but should help with tempco.
BJT footprint pinout is for MPSA06/MPSA56 pair. Pass transistor footprint should work for BJTs such as MJE15032/MJE15033 but I have not tested performance with BJT as pass transistor.
There's AC input and DC input versions. Each has a single or dual version. The AC input ones have local and remote sensing versions while the DC input ones have only remote sensing versions. The addon denoiser boards are separate and have positive and negative versions, the negative is mirrored design.
The DC input versions don't have local sensing as I was looking for smallest size, you can adapt AC input to DC input by omitting rectifying diodes and shorting two of them.
AC input dual versions can work with single secondary voltage doubler or dual secondary transformer. For voltage doubler just short the negative side AC input to middle GND in connector.
For 200mAout for each rail either 2200uF for voltage doubler either 1000uF for dual secondaries should be enough.
All are DIY-able, traces on bottom side with no topside passes.
Addon positive board:
Addon negative board:
AC input dual remote sensing:
AC input dual local sensing:
AC input single remote sensing:
AC input single local sensing:
DC input single remote sensing:
DC input dual remote sensing:
I have attached the Kicad project files, also contains gerbers zip file and DIY and schematic PDFs.
If you have questions about part values ask here. These haven't been tested so you make them at your own risk. I did test the AC input dual version with remote sensing.
All remote sensing boards have the DC sensing voltage divider resistors on board. They are also on the addon boards so you choose to populate one or the other. So the remote sensing supply boards can also be used as a regular discrete regulator, but you'll have worse AC performance (~80dB of PSRR at 50Hz)
This PSRR measurement did have a 22uF cap across R2.
Trimpots have two regular resistors in parallel on top so it's either pots either resistors.
The blue LED is close to the BJT on all boards, so you can have them touch or maybe add a bit of thermal paste between them. Don't know how critical this is but should help with tempco.
BJT footprint pinout is for MPSA06/MPSA56 pair. Pass transistor footprint should work for BJTs such as MJE15032/MJE15033 but I have not tested performance with BJT as pass transistor.
Last edited:
I think there's something wrong with the downloads counter 
I don't think even a 10th of that number ever clicked on this thread.
I don't think even a 10th of that number ever clicked on this thread.
Where would I see this extra noise of blue LEDs? As mentioned in last measurement post I got this for both rails, noise measurement, and is kind of the same thing as with red LED.
I think the denoiser acts on the LED noise as well since it would be on the regulator output.
But higher Vf does seem to help with regulation. I'm curious in trying an UV LED.
Two LEDs in series would also most likely throw tempco off
I think the denoiser acts on the LED noise as well since it would be on the regulator output.
But higher Vf does seem to help with regulation. I'm curious in trying an UV LED.
Two LEDs in series would also most likely throw tempco off
Last post was noise with blue LED and these are with red LED from few days ago:
Looks like the same thing, same values, for either red or blue LED.
Looks like the same thing, same values, for either red or blue LED.
If the LED noise is an issue, a parallel capacitor of 470 or even 1000µ is a possible solution.
For such a low voltage, the size can remain manageable. I think that my discrete regulator used one
For such a low voltage, the size can remain manageable. I think that my discrete regulator used one
Unless I did something wrong there's no difference in regulator output noise between using a red LED or blue LED as Vref.
There may be a difference in the noisefloor without a denoiser, just the discrete regulator by itself, for example in this measurement:
This was with a blue LED, maybe using a red LED would make for a lower noisefloor, but you have bigger problems than noisefloor without a denoiser. Namely the high ripple content. Using a red LED instead of a blue one won't do you much practical good. Any lowering in noisefloor is irrelevant in practical terms.
But using the denoiser the noisefloor is as low as it goes even with a blue LED. The total noise in audio spectrum, with the attenuated 50Hz ripple included, is 125-150nV, total. As seen from measurements I posted there's no real difference between using red LED and blue LED as Vref. I suspect there won't be any between IR LED and UV LED as Vref.
Using a capacitor in parallel with Vref seems to help with output impedance (from simulation), it does lower it further. There shouldn't be much if any PSRR gain (with MOSFET as pass transistor).
Here's a more clear comparison between both red and blue LEDs as Vref, with denoiser. Measurement is positive rail output. Traces are color coded for LED color.
The very same noisefloor, and blue LED offers a bit higher PSRR at 50Hz. The way I see it there's no difference in the resulting regulator noisefloor between either red or blue LEDs. That is not to say that they themselves don't have different noise levels, it's just the difference becomes irrelevant when installed in this circuit with a denoiser.
There may be a difference in the noisefloor without a denoiser, just the discrete regulator by itself, for example in this measurement:
This was with a blue LED, maybe using a red LED would make for a lower noisefloor, but you have bigger problems than noisefloor without a denoiser. Namely the high ripple content. Using a red LED instead of a blue one won't do you much practical good. Any lowering in noisefloor is irrelevant in practical terms.
But using the denoiser the noisefloor is as low as it goes even with a blue LED. The total noise in audio spectrum, with the attenuated 50Hz ripple included, is 125-150nV, total. As seen from measurements I posted there's no real difference between using red LED and blue LED as Vref. I suspect there won't be any between IR LED and UV LED as Vref.
Using a capacitor in parallel with Vref seems to help with output impedance (from simulation), it does lower it further. There shouldn't be much if any PSRR gain (with MOSFET as pass transistor).
Here's a more clear comparison between both red and blue LEDs as Vref, with denoiser. Measurement is positive rail output. Traces are color coded for LED color.
The very same noisefloor, and blue LED offers a bit higher PSRR at 50Hz. The way I see it there's no difference in the resulting regulator noisefloor between either red or blue LEDs. That is not to say that they themselves don't have different noise levels, it's just the difference becomes irrelevant when installed in this circuit with a denoiser.
Last edited:
Made a set of measurements for Vref (with blue LED) with 1000uF in parallel, and without. I looked at noisefloor PSRR and output impedance.
Regulator was loaded with 100mA (15Vout) static in load resistor (75R with 7.5V across it) + around 12mA RMS signal load for measuring output impedance.
First the PSRR/noisefloor for both no cap and 1000uF in parallel with Vref blue LED:
There's maybe just a little bit lower noisefloor 20Hz-100Hz but that's nothing. 50Hz ripple is maybe 1-2dB worse with 1000uF. Could be due to the capacitor's body picking up extra mains stuff.
These are output impedance measurements at various frequencies, both with and without cap. Bottom of photos are trace names which include 1000uF or no cap.
Denoiser addon was sensing on power supply board output connector, LNA sensing on the same spot (tried to get them as close as possible).
70Hz:
Upper trace is ADC sensing across the 75R load resistor. Lower traces are LNA on regulator output.
Both lower traces are similar with or without cap. Output impedance works out at 185uOhm.
120Hz:
Output impedance - 180uOhm.
500Hz:
175uOhm
1kHz:
175uOhm
5kHz:
205uOhm
10kHz:
305uOhm (here the 1000uF does something still too little)
20kHz:
No cap - 650uOhm
1000uF - 438uOhm
There's barely anything better at 20khz with 1000uF. In this design this capacitor has no place and does nothing really, isn't worth its price, the PCB space or even the trouble to install it on the backside of the PCB. The circuit is good enough already, that cap fixes nothing, at least for PSRR/noise/output impedance. With MOSFET as pass transistor.
With a denoiser this circuit should only be used with a single higher Vf LED. Blue for 12-15V and up. Maybe even UV (but I'll first have to validate this). Adding a second junction will alter tempco and won't give you anything in return. A blue LED works fine.
edit: The 1000uF cap also delays startup, at first I though I burnt the LED. Stays off for 2-3 seconds until cap charges up I think? The negative rail (with no 1000uF across Vref) started faster.
Regulator was loaded with 100mA (15Vout) static in load resistor (75R with 7.5V across it) + around 12mA RMS signal load for measuring output impedance.
First the PSRR/noisefloor for both no cap and 1000uF in parallel with Vref blue LED:
There's maybe just a little bit lower noisefloor 20Hz-100Hz but that's nothing. 50Hz ripple is maybe 1-2dB worse with 1000uF. Could be due to the capacitor's body picking up extra mains stuff.
These are output impedance measurements at various frequencies, both with and without cap. Bottom of photos are trace names which include 1000uF or no cap.
Denoiser addon was sensing on power supply board output connector, LNA sensing on the same spot (tried to get them as close as possible).
70Hz:
Upper trace is ADC sensing across the 75R load resistor. Lower traces are LNA on regulator output.
Both lower traces are similar with or without cap. Output impedance works out at 185uOhm.
120Hz:
Output impedance - 180uOhm.
500Hz:
175uOhm
1kHz:
175uOhm
5kHz:
205uOhm
10kHz:
305uOhm (here the 1000uF does something still too little)
20kHz:
No cap - 650uOhm
1000uF - 438uOhm
There's barely anything better at 20khz with 1000uF. In this design this capacitor has no place and does nothing really, isn't worth its price, the PCB space or even the trouble to install it on the backside of the PCB. The circuit is good enough already, that cap fixes nothing, at least for PSRR/noise/output impedance. With MOSFET as pass transistor.
With a denoiser this circuit should only be used with a single higher Vf LED. Blue for 12-15V and up. Maybe even UV (but I'll first have to validate this). Adding a second junction will alter tempco and won't give you anything in return. A blue LED works fine.
edit: The 1000uF cap also delays startup, at first I though I burnt the LED. Stays off for 2-3 seconds until cap charges up I think? The negative rail (with no 1000uF across Vref) started faster.
Last edited:
A cap in parallel with Vref might help at higher Aout, like 5A:
Sim with 1uF/10uF/100uF/1000uF. I think 100-220uF should be fine. But I have not tested such high output current.
Sim with 1uF/10uF/100uF/1000uF. I think 100-220uF should be fine. But I have not tested such high output current.
Last edited:
Hi, Im looking for a low noise regulator to supply Vop to Purify 1ET400A D class board & buffer - and wondered if there are any reasons the above circuits wouldn't be suitable - or if any of the variants are likely to be more or less suitable for this application.
I'm taking supply from the unregulated output of a Hypex SMPS1200A400 which is listed as +/-17V-25V and looking for a Vop of +/- 12V.
Thanks for any advice you may have.
I'm taking supply from the unregulated output of a Hypex SMPS1200A400 which is listed as +/-17V-25V and looking for a Vop of +/- 12V.
Thanks for any advice you may have.
That looks like a high power application, only tested this for few hundred mA. I might try testing the circuit for higher output current but don't know when I'll find the time for that.
Its just for the supply on 2 OPA1612 gain stages - not the power stage. The 1ET400A DS lists a current requirement of 27mA for the second gain stage, and i'd have thought the buffer gain stage would use a similar amount, so unless i'm missing something we should be well under 100mA.
I'm just not sure whether there is likely much to gain by improving regulation on supply to OPA1612s. I cant find much discussion on how sensitive they are to regulation quality.
I'm just not sure whether there is likely much to gain by improving regulation on supply to OPA1612s. I cant find much discussion on how sensitive they are to regulation quality.