Bench power supply design

[Eldi4]

Hello there,

Recently i was trying to design my own next bench power supply, because my previous diy bench power supply is a failure and already dying after 6 month of usage.
I aim at 20V 5A CC/CV, and controlled by STM32F103C8T6.
After looking and trying to understand at other diy bench power supply like :
https://github.com/eez-open/psu-hw and https://www.eevblog.com/forum/projects/diy-bench-power-supply-psl-3604/
I came up with this design for analog parts, however i don't really understand what is the purpose of C8/R16 and C7/R14, as i just copied this from those references, it is definitely something related to feedback but i'm not sure what exactly.

Most of my component choosing is based on availability, so it might be not the best component to use. I might insist to use these components because it's pretty hard to source components.
I don't have oscilloscope to test noise/ripple, CC response, oscillations and other timing critical stuff.

I haven't tested the circuit on real environment though, because breadboarding this would be a pain, and i never got any good result creating my own pcb. But i have tested the circuit with LTSpice, and it seems to be working fine, but i just came here to make it clear, did there are any shortcomings on my circuit?

[David Hess]

I am just going to ramble.

I would work hard to either combine the difference amplifiers with the error amplifiers to reduce each control loop to one stage for easier frequency compensation, or remove the difference amplifiers from inside the control loops.

5 amps at 20 volts is beyond the power capability of a single 2N3055.  There are more powerful parts in the TO-3 and other packages which might be suitable if you do not want to parallel parts.

The drive and output transistors should have base-emitter shunt resistors to improve turn-off performance.  At a minimum, the output transistor needs this to prevent runaway from leakage at high temperatures.  I would like to see a design where the output transistors are driven by class-AB drivers; diamond buffers could be used.

The current sensing arrangement with the sense resistor in series with the output capacitor adds a zero to the frequency response which makes frequency compensation easier if the output capacitor's ESR is too low, but most designs still place the current sense resistor on the high side.  If the current limit amplifier is referenced to the output, then no difference amplifier is required inside the control loop.

[Eldi4]

I would work hard to either combine the difference amplifiers with the error amplifiers to reduce each control loop to one stage for easier frequency compensation, or remove the difference amplifiers from inside the control loops.

The only differential amplifier is the voltage sensing amplifier though, i need it to measure actual voltage applied to load, previously my design just used voltage divider which get interfered by voltage drop on current sense resistor.
I think it will be pretty hard to simplify the op amp stage, since my design is actually just a copy from those reference design, but i will try.

5 amps at 20 volts is beyond the power capability of a single 2N3055.  There are more powerful parts in the TO-3 and other packages which might be suitable if you do not want to parallel parts.

I will use the pretty beefy 2SC2922 with overclocked AMD fan on processor heatsink, i has tested it and it was able to withstand full 100W of dissipation power for hours without problem.

The drive and output transistors should have base-emitter shunt resistors to improve turn-off performance.  At a minimum, the output transistor needs this to prevent runaway from leakage at high temperatures.  I would like to see a design where the output transistors are driven by class-AB drivers; diamond buffers could be used.

Ah yes, thank you very much for pointing that out, definitely going to add it in!
As for the output transistor driver, what could be possible improvement acquired by changing to those drivers?

The current sensing arrangement with the sense resistor in series with the output capacitor adds a zero to the frequency response which makes frequency compensation easier if the output capacitor's ESR is too low, but most designs still place the current sense resistor on the high side.  If the current limit amplifier is referenced to the output, then no difference amplifier is required inside the control loop.

I'm sorry but i don't really get what you said. Is it basically output capacitor should just be placed on OUT+ and GND, instead of OUT+ and OUT- to improve frequency response?
I use low-side sensing and regular non-inverting amp though, so no differential amplifier on current control loop.

[David Hess]

I would work hard to either combine the difference amplifiers with the error amplifiers to reduce each control loop to one stage for easier frequency compensation, or remove the difference amplifiers from inside the control loops.

The only differential amplifier is the voltage sensing amplifier though, i need it to measure actual voltage applied to load, previously my design just used voltage divider which get interfered by voltage drop on current sense resistor.
I think it will be pretty hard to simplify the op amp stage, since my design is actually just a copy from those reference design, but i will try.

I did not write "differential"; I wrote *difference*.  I might also have written instrumentation amplifier but difference fits better.  But the point is that using two stages of the same operational amplifier inside the control loops makes frequency compensation more difficult.

The drive and output transistors should have base-emitter shunt resistors to improve turn-off performance.  At a minimum, the output transistor needs this to prevent runaway from leakage at high temperatures.  I would like to see a design where the output transistors are driven by class-AB drivers; diamond buffers could be used.

Ah yes, thank you very much for pointing that out, definitely going to add it in!
As for the output transistor driver, what could be possible improvement acquired by changing to those drivers?

It was just a thought.  Obviously good performance can be attained with a class-A driver when base-emitter shunt resistors are used and most linear power supplies do it this way.  Higher performance output stages drive the output transistor with a class-AB driver so charge can be quickly removed to turn the transistor off.

The current sensing arrangement with the sense resistor in series with the output capacitor adds a zero to the frequency response which makes frequency compensation easier if the output capacitor's ESR is too low, but most designs still place the current sense resistor on the high side.  If the current limit amplifier is referenced to the output, then no difference amplifier is required inside the control loop.

I'm sorry but i don't really get what you said. Is it basically output capacitor should just be placed on OUT+ and GND, instead of OUT+ and OUT- to improve frequency response?

I am saying that the arrangement you have shown has the current sense resistor adding to the ESR of the output capacitor and this can be taken advantage of to improve the frequency compensation by taking AC feedback from the other side of the sense resistor.  This trick is used when the ESR of the output capacitor is very low which would cause stability issues.

But most designs ignore this and you can also.

I use low-side sensing and regular non-inverting amp though, so no differential amplifier on current control loop.

That works and has advantages, but most power supplies put up with using high side current sensing.  Either way works.

[Eldi4]

I did not write "differential"; I wrote *difference*.  I might also have written instrumentation amplifier but difference fits better.  But the point is that using two stages of the same operational amplifier inside the control loops makes frequency compensation more difficult.

I see, i'm sorry for misunderstanding. i don't know but it seems really hard to simplify it to one-stage, i could only combine the current and voltage loop so total of 3 amp, but can't really get it to only 2 amp.

It was just a thought.  Obviously good performance can be attained with a class-A driver when base-emitter shunt resistors are used and most linear power supplies do it this way.  Higher performance output stages drive the output transistor with a class-AB driver so charge can be quickly removed to turn the transistor off.

Ah i see, dumb me can't even get a grasp of the advantage of simple class A vs class AB driver.

I am saying that the arrangement you have shown has the current sense resistor adding to the ESR of the output capacitor and this can be taken advantage of to improve the frequency compensation by taking AC feedback from the other side of the sense resistor.  This trick is used when the ESR of the output capacitor is very low which would cause stability issues.

But most designs ignore this and you can also.

I'm not gonna say i understand this perfectly, but that do give me some more insight on how it work

That works and has advantages, but most power supplies put up with using high side current sensing.  Either way works.

The main reason i use low side sensing is because common-mode voltage might be a big problem, i just want to use 5V supply on my sensing amp, because it will feed to ADC of my MCU. and 5V supply LM324 nicely clamp it's own output to around 3.5V in case of some current/voltage spike.

Anyway, thanks a lot for the insights, exactly what i want from these kind of thread.

[David Hess]

The main reason i use low side sensing is because common-mode voltage might be a big problem, i just want to use 5V supply on my sensing amp, because it will feed to ADC of my MCU. and 5V supply LM324 nicely clamp it's own output to around 3.5V in case of some current/voltage spike.

The LM324 output has very limited current sinking below 0.6 volts so low current through the shunt could be a problem.  One solution for this is to configure U5 as a difference amplifier with the reference going to a low voltage virtual ground, but U6 and the control signal would also have to use it also so it complicates things.

[Kleinstein]

The capacitors C7 and C8 are there to kind of bypass the difference amplifier for the high frequency part. With these extra caps in place the extra OP in the loop should not be such a big deal.  C7 is kind of optional - without it one would take the AC feedback from the other side of the shunt ( = ground), which usually helps stability.

A part one may add is a diode to prevent to much reverse BE votlage for the ouput transistors. This may other wise happen of the votlage overshoots with enough capacitance at the output. The added diode would make the regulator part also sink some current from the output. The LM324 is current limited and shound not have a problem going into that limit.

A shunt at the low side is also conventient for digital control, as both the votlage and current set signal can be relative to a common ground.

[Eldi4]

The LM324 output has very limited current sinking below 0.6 volts so low current through the shunt could be a problem.  One solution for this is to configure U5 as a difference amplifier with the reference going to a low voltage virtual ground, but U6 and the control signal would also have to use it also so it complicates things.

Didn't powering LM324 with dual supply solve this problem already?

The capacitors C7 and C8 are there to kind of bypass the difference amplifier for the high frequency part. With these extra caps in place the extra OP in the loop should not be such a big deal.  C7 is kind of optional - without it one would take the AC feedback from the other side of the shunt ( = ground), which usually helps stability.

Thanks for explaining it, it's pretty clear now!

A part one may add is a diode to prevent to much reverse BE votlage for the ouput transistors. This may other wise happen of the votlage overshoots with enough capacitance at the output. The added diode would make the regulator part also sink some current from the output. The LM324 is current limited and shound not have a problem going into that limit.

I see, reversed regular diode between base and emitter seems to fix this problem, the diode will sink some current from output only when output somehow had overshoot right?.

[Kleinstein]

The extra reverse diode would give the regulator a limited sinking capability. This is not just for overshoot, but also when turning down the set voltage with a larger cap at the output.
The LM324 has a limited supply range, so there is not much room for a negative supply. One may get away with some -1 V or so. With 3 transistors for the output and one diode for the wired OR, the regulators only have to go down to about 1 V.

[David Hess]

The LM324 output has very limited current sinking below 0.6 volts so low current through the shunt could be a problem.  One solution for this is to configure U5 as a difference amplifier with the reference going to a low voltage virtual ground, but U6 and the control signal would also have to use it also so it complicates things.

Didn't powering LM324 with dual supply solve this problem already?

It does, but you mentioned using the 5 volt logic supply for the sensing amplifier.