The layout looks mostly OK, but I have concerns about R1 and R2: they are probably wirewound (and even film resistors are spiralled), and could send a heavily rippled magnetic field into the regulator's circuitry.
Note that if you add the transistor, the denoiser might become unstable.
Use the normal resistor values for the 317, and change the denoiser values to 820 ohm and 82K.
I see the denoising circuit is inside the high current loop, that's not recommended.
For 5V you can even go to lower to 560R/56K combo which would put almost 6mA through the denoising circuit.
If you have a max of 1A draw then I wouldn't use an extra transistor in parallel to LM317, I'd use it in a capacitance multiplier configuration before the regulator. That would decrease the ripple further 56dB or so. And at 1A it would take some of the heat away from the LM317.
edit: attached sim file. also what DACs need 1A for 5V?
For 5V you can even go to lower to 560R/56K combo which would put almost 6mA through the denoising circuit.
If you have a max of 1A draw then I wouldn't use an extra transistor in parallel to LM317, I'd use it in a capacitance multiplier configuration before the regulator. That would decrease the ripple further 56dB or so. And at 1A it would take some of the heat away from the LM317.
edit: attached sim file. also what DACs need 1A for 5V?
Attachments
Last edited:
Thank you very much for your inputs. I too had an idea to use a capacitance multiplier before the denoiser. Was playing around with the idea of using an L Adaptor followed by Denoiser for the DAC and Capacitance Multiplier/Pre-regulator followed by L Adaptor for the Raspberry Pi4B. What do you all think?
I would check if there is some switching PSU on DAC and RPi4B boards, if there is then i'm not sure how much is important to make really silent PSU for it. But then again who am i to judge... I'm gonna use dnoizator for ecc88 heating in phono preamp. 
Talking about overkill...
But i really like this circuit, it has great performance, it's cheap, parts you can get on every corner, what more can you ask for?
Talking about overkill...But i really like this circuit, it has great performance, it's cheap, parts you can get on every corner, what more can you ask for?
Do not stack together too many building blocks having more or less the same function: they will be redundant at best, or degrade something in some instances.
For example, a cap-mult in front of a denoiser will improve the PSRR, but nothing else, and if placed downstream, it will degrade everything except the PSRR, but the improvement will be meaningless because it will be drowned in noise.
Make your choices in a rational way, based on numerical values: you know what a denoiser or cap-mult can achieve, and once you have reached the floor, it makes no sense to go further.
Direct your efforts into other directions: for example, if you use digital devices or a switching supply as a primary power source, it makes sense to include common mode rejection measures, typically one or more CMC, to cover the whole frequency spectrum
An interesting aspect of the CCS is that depending on the denoiser transistor you use, there's a different voltage on the denoiser side of the coupling capacitor. This might be speculated for low voltage output applications.
For example for 12Vout, using the MPSA06 in that spot the voltage sits shy of 10V. Using BC807 makes for around 4.9V. Irrespective of Vout, the DC voltage on the denoiser circuit output sits at the same spot past a certain point. Seems you need around 1-1.5V differential between Vout and output of denoiser circuit to have it perform well.
So for 5Vout for example BC550C seems to do a good job (PSRR wise). I haven't measured this yet, but I suppose noise will be pretty low even with BC550C.
For example for 12Vout, using the MPSA06 in that spot the voltage sits shy of 10V. Using BC807 makes for around 4.9V. Irrespective of Vout, the DC voltage on the denoiser circuit output sits at the same spot past a certain point. Seems you need around 1-1.5V differential between Vout and output of denoiser circuit to have it perform well.
So for 5Vout for example BC550C seems to do a good job (PSRR wise). I haven't measured this yet, but I suppose noise will be pretty low even with BC550C.
Attachments
Seems that higher gain transistors in denoiser spot make for lower DC offset on the output of the denoiser, which allows for more Vdrop across the CCS, for low Vout.
In simulation 2N5089 performs best in this regard with 2.4V DC offset on the output of the denoising circuit, but for 3.3Vout the performance still suffers as there's less than 1V across the CCS.
But for 5Vout BC550C seems to be a good option.
In simulation 2N5089 performs best in this regard with 2.4V DC offset on the output of the denoising circuit, but for 3.3Vout the performance still suffers as there's less than 1V across the CCS.
But for 5Vout BC550C seems to be a good option.
I made the ripple injector that jackinnj mentioned here and made a few measurements of the single pcb from this post.
I used this mosfet. The regulator PCB was fitted with TI's LM317N.
The first 5 attached measurements are for this circuit:
The last 3 are for this configuration:
PSRR at 15kHz is still over 115dB in both cases.
I reduced the denoising circuit coupling capacitor to 10uF, now the circuit settles faster.
In measurements upper trace is input ripple into the regulator measured with one channel of my ADC, and the lower trace is on the output of the regulator, at the output connector, sensed with the other channel of the ADC + 60dB LNA.
I used this mosfet. The regulator PCB was fitted with TI's LM317N.
The first 5 attached measurements are for this circuit:
The last 3 are for this configuration:
PSRR at 15kHz is still over 115dB in both cases.
I reduced the denoising circuit coupling capacitor to 10uF, now the circuit settles faster.
In measurements upper trace is input ripple into the regulator measured with one channel of my ADC, and the lower trace is on the output of the regulator, at the output connector, sensed with the other channel of the ADC + 60dB LNA.
Attachments
I tried running some sweeps in the whole audio band but didn't have great results. In REW I though about using the Peak Hold trace thing but it's not really adequate, since it doesn't average the trace, and I also can't save the Peak trace as a measurement to compare with others.
This would be great as I could construct a PSRR graph at all audio frequencies.
In the attached photo I blended together the input and output measurements, both were sweeps and used the Peak Hold trace for both.
Sadly they aren't accurate. The input ripple was higher at around -12dBV in reality, and HF output ripple is off by quite a bit.
I think it has to do with the way REW samples the data, FFT length etc.
I did try using 10 minute sweeps but even that didn't make it accurate.
This would be great as I could construct a PSRR graph at all audio frequencies.
In the attached photo I blended together the input and output measurements, both were sweeps and used the Peak Hold trace for both.
Sadly they aren't accurate. The input ripple was higher at around -12dBV in reality, and HF output ripple is off by quite a bit.
I think it has to do with the way REW samples the data, FFT length etc.
I did try using 10 minute sweeps but even that didn't make it accurate.
Attachments
Got to test BC550C in the denoiser position with 5Vout and seems to be working fine.
I used these values for measurement:
I used BC850C on the SMD design that I'll post a bit later, but should be the same as BC550C.
PSRR is a bit lower with BC850C than with BC817 at 12Vout. Noise seems a bit worse at around the 2nV/sqrtHz level (or 270nV total in audio band).
Going from 12Vout to 5Vout (with BC850C) makes for around 1dB lower PSRR, so not much loss. Pretty good for 5Vout.
I attached PSRR measurements for different frequencies, and last photo is a noise measurement.
I used a regular SOT-223 LM317 from TI. Performance could be a bit better for LM317N.
Upper trace is measured on the regulator input, lower trace is the output at the connector.
I used these values for measurement:
I used BC850C on the SMD design that I'll post a bit later, but should be the same as BC550C.
PSRR is a bit lower with BC850C than with BC817 at 12Vout. Noise seems a bit worse at around the 2nV/sqrtHz level (or 270nV total in audio band).
Going from 12Vout to 5Vout (with BC850C) makes for around 1dB lower PSRR, so not much loss. Pretty good for 5Vout.
I attached PSRR measurements for different frequencies, and last photo is a noise measurement.
I used a regular SOT-223 LM317 from TI. Performance could be a bit better for LM317N.
Upper trace is measured on the regulator input, lower trace is the output at the connector.
Attachments
Last edited:
Actually any of the preferred BJTs can be used as long as R3/R10 in the following schematic are adjusted to keep over 1V across the CCS. As long as you're using a CCS version you should be able to get good performance even at 5Vout.
Increasing this resistor value results in higher performance, but increasing it decreases the voltage drop across the CCS. So you'd want to max that value until you get to 1.3-1.5Vdrop across the CCS.
edit: also maybe adjust the LED resistor to keep at least 1-2mA through it. That varies with Vout.
Increasing this resistor value results in higher performance, but increasing it decreases the voltage drop across the CCS. So you'd want to max that value until you get to 1.3-1.5Vdrop across the CCS.
edit: also maybe adjust the LED resistor to keep at least 1-2mA through it. That varies with Vout.
Last edited:
This post is for the latest SMD boards I designed to include the NoNoiser way of controlling the ADJ pin.
This time the boards are really thin. Since the Nonoiser increases the input impedance for the denoising circuit output, the coupling capacitor value can be much smaller.
I used 3528 Tantalum capacitor footprint for both denoiser coupling capacitors since that should be enough for any Vout.
For 12Vout you have 82K and you can use 10uF.
For 5Vout you have 27K and you can use 22uF, maybe even 33uF if you can find it. If we lower Vout then we can lower this capacitor's voltage rating. I've checked on Farnell and I've seen some available in that package.
This is also valid for the sensing cap, but I'd keep that as large as possible (up to 100uF) keeping in mind its voltage rating.
Don't go lower than 22uF on that capacitor.
In the attached measurements for the single LM317 SMD board I used a 10uF/16V 1206 capacitor for the output of the denoising circuit. This doesn't seem an issue, but for the sensing capacitor you'll have a bad experience using a ceramic one, as everything picked up by it gets amplified by a lot. So in consequence a ceramic capacitor in this spot (input of denoising circuit) is very microphonic. Using a Tantalum capacitor in this spot seems to solve this issue.
There's 3x 1206 input capacitor footprints, you don't have to populate all, just have at least 10uF in this spot.
There's a 1210 fuse on the input which can be replaced for an appropriate inductor to be used as a filter with the ceramics on the input.
There's two parallel resistor footprints to help with trimming the output voltage.
You can also use the included jumpers to select between the different topologies discussed so far, with the mention that if you switch from NoNoiser way of controlling the ADJ pin you'll need a larger coupling cap value. Unless you want to solder a THT electrolytic there then use the NoNoiser version.
For the attached measurements I used a regular LM317 from TI in SOT-223 footprint. LM3x7N is also available in this package.
The output RLC network is not yet fully determined. I couldn't get it stable without that extra inductor. The resistor should be around 0.15R to 0.2R depending on the inductor's resistance. I have tried a few combinations for output capacitors and seems it needs at least 3 paralleled. This was valid for 10uF and also 22uF. In measurements I used 3x22uF/16V 0805 capacitors (mounted on the 1206 footprints). 3x10uF/16V 1206 worked as well but would go for higher capacitance, 3x22uF worked fine for me.
I didn't test the LM337 version but the only differences should be 3.3R instead of 1R for denoiser compensation network and 0.03-0.04R for the output RLC network, with same capacitors.
I used a 0805 red LED (1.7Vdrop) in my tests, and worked just fine. I also tested an amber one (1.8Vdrop) and worked fine as well. But I wouldn't go higher Vdrop.
All BJTs used in measurements were BC8x7, works fine.
So I still need to test the range on values where the circuit is stable for both LM3x7 versions. These SMD boards should be considered still experimental at this stage. I'll post some example parts that could be used for output inductor and capacitors.
Ferrite bead is 0805 footprint and I used this one, I don't know if other type would be better here, maybe Elvee can chime in on this. This is what I already had. Maybe that's the reason the output inductor is still needed?
Board sizes are around 37mm x 31mm for the single boards and 46mm x 40mm for the dual one.
They are DIY-able with the traces on the top-side. A single bottom-side link is needed for the single versions and two passes for the dual version. All are GND traces and for DIY you need to install the wires, for fab-house the traces are included in the gerbers.
As far as power goes, these could be used for up to 200mA output depending on Vdrop across the regulator. I wouldn't put more than 0.5W on it as it is.
I thought about adding a cooling plane on the bottom of the pcb, for fab-house version only, but I don't know if I ruin the performance or not. Depending on LM317 or LM337 I'd either risk picking up crap, or radiating it.
Maybe a copper plane connected via thermal jumpers on the bottom? I could add vias to get some heat on the backside.
These designs are released for personal use only, no commercial applications.
This time the boards are really thin. Since the Nonoiser increases the input impedance for the denoising circuit output, the coupling capacitor value can be much smaller.
I used 3528 Tantalum capacitor footprint for both denoiser coupling capacitors since that should be enough for any Vout.
For 12Vout you have 82K and you can use 10uF.
For 5Vout you have 27K and you can use 22uF, maybe even 33uF if you can find it. If we lower Vout then we can lower this capacitor's voltage rating. I've checked on Farnell and I've seen some available in that package.
This is also valid for the sensing cap, but I'd keep that as large as possible (up to 100uF) keeping in mind its voltage rating.
Don't go lower than 22uF on that capacitor.
In the attached measurements for the single LM317 SMD board I used a 10uF/16V 1206 capacitor for the output of the denoising circuit. This doesn't seem an issue, but for the sensing capacitor you'll have a bad experience using a ceramic one, as everything picked up by it gets amplified by a lot. So in consequence a ceramic capacitor in this spot (input of denoising circuit) is very microphonic. Using a Tantalum capacitor in this spot seems to solve this issue.
There's 3x 1206 input capacitor footprints, you don't have to populate all, just have at least 10uF in this spot.
There's a 1210 fuse on the input which can be replaced for an appropriate inductor to be used as a filter with the ceramics on the input.
There's two parallel resistor footprints to help with trimming the output voltage.
You can also use the included jumpers to select between the different topologies discussed so far, with the mention that if you switch from NoNoiser way of controlling the ADJ pin you'll need a larger coupling cap value. Unless you want to solder a THT electrolytic there then use the NoNoiser version.
For the attached measurements I used a regular LM317 from TI in SOT-223 footprint. LM3x7N is also available in this package.
The output RLC network is not yet fully determined. I couldn't get it stable without that extra inductor. The resistor should be around 0.15R to 0.2R depending on the inductor's resistance. I have tried a few combinations for output capacitors and seems it needs at least 3 paralleled. This was valid for 10uF and also 22uF. In measurements I used 3x22uF/16V 0805 capacitors (mounted on the 1206 footprints). 3x10uF/16V 1206 worked as well but would go for higher capacitance, 3x22uF worked fine for me.
I didn't test the LM337 version but the only differences should be 3.3R instead of 1R for denoiser compensation network and 0.03-0.04R for the output RLC network, with same capacitors.
I used a 0805 red LED (1.7Vdrop) in my tests, and worked just fine. I also tested an amber one (1.8Vdrop) and worked fine as well. But I wouldn't go higher Vdrop.
All BJTs used in measurements were BC8x7, works fine.
So I still need to test the range on values where the circuit is stable for both LM3x7 versions. These SMD boards should be considered still experimental at this stage. I'll post some example parts that could be used for output inductor and capacitors.
Ferrite bead is 0805 footprint and I used this one, I don't know if other type would be better here, maybe Elvee can chime in on this. This is what I already had. Maybe that's the reason the output inductor is still needed?
Board sizes are around 37mm x 31mm for the single boards and 46mm x 40mm for the dual one.
They are DIY-able with the traces on the top-side. A single bottom-side link is needed for the single versions and two passes for the dual version. All are GND traces and for DIY you need to install the wires, for fab-house the traces are included in the gerbers.
As far as power goes, these could be used for up to 200mA output depending on Vdrop across the regulator. I wouldn't put more than 0.5W on it as it is.
I thought about adding a cooling plane on the bottom of the pcb, for fab-house version only, but I don't know if I ruin the performance or not. Depending on LM317 or LM337 I'd either risk picking up crap, or radiating it.
Maybe a copper plane connected via thermal jumpers on the bottom? I could add vias to get some heat on the backside.
These designs are released for personal use only, no commercial applications.
Attachments
Last edited:
Don't forget that Rds(on) of the MOSFET will influence the measurements if there is an input capacitor ahead of the regulator (suggested by Texas Instruments)
In which way will it affect the measurement? I measure the signal that's on the input of the regulator. I used my phone as signal gen in some tests, hence the -12.2dBV or so of input signal. But I tried with my DAC for 0dBV and worked fine.
I did use the 10uF/100nF combo before the regulator, and had to use something on the input of LM317 or else it would not be stable. I used 10uF, doesn't seem to attenuate the signal much. And would it matter as long as I am measuring the input of the LM317?
I did use the 10uF/100nF combo before the regulator, and had to use something on the input of LM317 or else it would not be stable. I used 10uF, doesn't seem to attenuate the signal much. And would it matter as long as I am measuring the input of the LM317?
I did manage to find a working RLC network for the SMD board.
I used one of these capacitors, tried a 200mOhm and 50mOhm (worked with both) and for the inductor seems a single loop (?) works. Doesn't work without it. I tried replacing it with a 50mOhm resistor thinking it's the extra resistance but no, it seems it wants that at least single loop.
The diy inductor looks like this:
In the end I used two loops. It can be made from normal copper wire, no need for enameled wire in this case, since there's two loops, just stretch them a bit so they don't touch and should work fine.
I'd go for around 150mOhm for the resistance.
I will try and do the same tests for LM337 version so the SMD boards all have known working values.
I also tried again playing with LT3082 since the SMD LM317 single board is compatible with it as well, and I used the CCS dienoiser and seems to be working, and with good performance.
Noise is identical to the LM317 version, and I think it's limited by BC817. Around 100nV total noise in audio spectrum (with LNA's noise included, which is around 60nV).
PSRR is pretty high, I measured over 160dB, and was this high towards 1kHz. I only got it stable with a THT output capacitor, that Panasonic FR 470uF/25V, I couldn't emulate the RLC network with SMD parts yet. But seems to be working with that particular capacitor. Same 12Vout, 150R load.
edit:
this time I left it working for around 30 minutes and performance was still there. tried it in this configuration:
I used one of these capacitors, tried a 200mOhm and 50mOhm (worked with both) and for the inductor seems a single loop (?) works. Doesn't work without it. I tried replacing it with a 50mOhm resistor thinking it's the extra resistance but no, it seems it wants that at least single loop.
The diy inductor looks like this:
In the end I used two loops. It can be made from normal copper wire, no need for enameled wire in this case, since there's two loops, just stretch them a bit so they don't touch and should work fine.
I'd go for around 150mOhm for the resistance.
I will try and do the same tests for LM337 version so the SMD boards all have known working values.
I also tried again playing with LT3082 since the SMD LM317 single board is compatible with it as well, and I used the CCS dienoiser and seems to be working, and with good performance.
Noise is identical to the LM317 version, and I think it's limited by BC817. Around 100nV total noise in audio spectrum (with LNA's noise included, which is around 60nV).
PSRR is pretty high, I measured over 160dB, and was this high towards 1kHz. I only got it stable with a THT output capacitor, that Panasonic FR 470uF/25V, I couldn't emulate the RLC network with SMD parts yet. But seems to be working with that particular capacitor. Same 12Vout, 150R load.
edit:
this time I left it working for around 30 minutes and performance was still there. tried it in this configuration:
Last edited: