CC/CV Benchtop Power Supply Topology Question

[nike9307]

Hi all,

I'd like to design a benchtop power supply with the following requirements

Output Voltage - 0-30V
Output Current - 0-3A. No overshoot on the constant current.
Noise - Less than 50mVp-p. 20 would be better.
No Fan! That's the main reason I'd like to design it in the first place.
Small footprint - Hammond 1455L801BK - I'll mount the power board on the top panel, so it can use it as a heatsink
Limited on device user interface, due to the small form factor. Mostly USB controlled.
If this sounds pretty close to Dave's original uSupply, its because it very much is. I'm not claiming originality here.

Now I'm thinking of a switching pre-regulator followed by a linear stage. I'm still in early idea stage.
I can handle the control section and the switching regulator just fine, but I've never designed a linear regulator from scratch. I've only used linear ICs.

I was reading up on designing the linear section and most designs are similar to this - 1 pass element (BJT or MOSFET is irrelevant) and 2 control loops - one for voltage and one for current.


However i was watching Dave's final uSupply video his schematic was a lot different. He has the 2 control loops, but also a pass element for each loop.


Now i was wondering, what are the benefits of this approach? I cant find anything similar on any other power supply. 
UPDATE: This question was asked on the topic about the uSupplyhttps://www.eevblog.com/forum/blog/eevblog-1561-usupply-usb-power-supply-part-20/msg5026486/#msg5026486. Dave didn't recall why it's like this.

[PCB.Wiz]

However i was watching Dave's final uSupply video his schematic was a lot different. He has the 2 control loops, but also a pass element for each loop.
Now i was wondering, what are the benefits of this approach? I cant find anything similar on any other power supply. 
... Dave didn't recall why it's like this.

Maybe it was simpler to develop ?
At lower voltage outputs you can use the current regulator as a thermal load share by splitting the voltage across them ?
That matters more if you do not have the Switching pre-regulator.

Output Voltage - 0-30V
Output Current - 0-3A. No overshoot on the constant current.
Noise - Less than 50mVp-p. 20 would be better.
No Fan! That's the main reason I'd like to design it in the first place.
Small footprint - Hammond 1455L801BK - I'll mount the power board on the top panel, so it can use it as a heatsink
Limited on device user interface, due to the small form factor. Mostly USB controlled.
If this sounds pretty close to Dave's original uSupply, its because it very much is. I'm not claiming originality here.

Now I'm thinking of a switching pre-regulator followed by a linear stage. I'm still in early idea stage.
I can handle the control section and the switching regulator just fine, but I've never designed a linear regulator from scratch. I've only used linear ICs.

I was reading up on designing the linear section and most designs are similar to this - 1 pass element (BJT or MOSFET is irrelevant) and 2 control loops - one for voltage and one for current.

That's broadly standard.

There are controller chips that do most of that, look at
https://www.st.com/en/power-management/voltage-and-current-controllers/products.html
and
https://www.diodes.com/products/power-management/ac-dc-converters/secondary-side-controllers-and-switchers/#
Some parts include the current sense divider down to sub 100mV and those often sense in the ground side. (eg -50mV - 17mOhms for your 3A)

A switching pre-regulator makes a huge difference to your thermal management challenges.

If you want to go to 0V, that can be done with positive current injection, eg by doubling the reference voltage with a opamp and equal resistors, then a resistor the VFB pin,
In that case, a part with an exposed VREF is useful.

MOSFETS are easier to drive than transistors, but their high Cin can prove a challenge for opamps to drive with stability. You will need a scope.
Audio Amp FETs can have lower Cin and better thermal packaging.

[ledtester]

The original design started by considering using two linear regulators in series - one for CC and the other for CV - and progressed from there:

EEVblog #221 - Lab Power Supply Design - Part 1
https://youtu.be/CIGjActDeoM

I've seen a similar approach taken with this design:

Build a Small Dual Rail Linear Bench Power Supply. -- TheStuffMade
https://youtu.be/jBtNH1EbU8M

[nike9307]

Maybe it was simpler to develop ?
At lower voltage outputs you can use the current regulator as a thermal load share by splitting the voltage across them ?
That matters more if you do not have the Switching pre-regulator.

Well that's the thing, the original design does have a switching pre-regulator, its just on a different schematic sheet.
I thought about sharing the thermal load, but only one of the transistors will be dissipating heat at a time, so its not that.

A switching pre-regulator makes a huge difference to your thermal management challenges.

This is why i want one.

MOSFETS are easier to drive than transistors, but their high Cin can prove a challenge for opamps to drive with stability. You will need a scope.
Audio Amp FETs can have lower Cin and better thermal packaging.

I'm aware of the stability issues. I have a scope at work and my boss is cool. He lets me use the work equipment for personal projects.
I'm thinking of using extended SOA N Channel MOSFETs driven by a high current unity gain op-amp with a series resistor on the gate. I've simulated that this driving it like that is simple and causes no issues. \
If you have a recommendation for an audio amp MOSFET I'm happy to hear it
I'd like to use a MOSFET so the voltage drop is minimal. If i can get it stable and low noise at 0.3-0.4V dropout voltage i could probably push the max output current even higher.

[nike9307]

The original design started by considering using two linear regulators in series - one for CC and the other for CV - and progressed from there:

EEVblog #221 - Lab Power Supply Design - Part 1
https://youtu.be/CIGjActDeoM

I've seen a similar approach taken with this design:

Build a Small Dual Rail Linear Bench Power Supply. -- TheStuffMade
https://youtu.be/jBtNH1EbU8M

Yeah the dual LM317 is a classic when its used that way. When its done with discrete transistors I've not seen it done.
So you think this topology is a legacy of the old design and not a contious choice for the new one?

[Kleinstein]

Separate power transistors the CC and CV part is an odd solution, likely based on 2 LM317 regulators in series. There is no real benefit to this, bit some issues.

Having a scope to test the desin is a good idea. For the developement a good tool is a simulation package like LTspice.

No fan and the small case kind of requires a SMPS preregulation, which adds a little extra level of complexity. To a large part one could start with the CV/CC regulator part and add the preregulator later in the developement.

No overshoot for the current is a difficult part, as any capacitance at the output does allow for overshoot in the current. One can mainly keep the overshoot small / limit the capacitance at the output. Most regulators need the capacitance for stablity of the control loop.

[nike9307]

Separate power transistors the CC and CV part is an odd solution, likely based on 2 LM317 regulators in series. There is no real benefit to this, bit some issues.

I see, so its a waist of time to go down that road. Good to know.

Having a scope to test the desin is a good idea. For the developement a good tool is a simulation package like LTspice.

I tend to be kind of a noob in analog design, so i always simulate heavily.

No fan and the small case kind of requires a SMPS preregulation, which adds a little extra level of complexity. To a large part one could start with the CV/CC regulator part and add the preregulator later in the developement.

That was my plan, as the linear CC/CV is more difficult for me. I've designed a lot of SMPS, so i'll leave that for when i'm happy with the performance of the linear one.

No overshoot for the current is a difficult part, as any capacitance at the output does allow for overshoot in the current. One can mainly keep the overshoot small / limit the capacitance at the output. Most regulators need the capacitance for stablity of the control loop.

You are correct. I'm aiming for minimal capacitance on the output. Less than 100uF if possible. If i use an N channel MOSFET, i should be able to stabilize the loop with less capacitance.
I'm also going to try to design an active circuit that draws current from the output cap, so it doesn't go to the output. Crowbar type circuit, that discharges the output. This is however entirely an optional fun challenges and I'll be happy with the performance of 100uF on the output.

[Kleinstein]

100 µF for the output is very feasible. It gets tricky (needs fast regulation) if one aims for less than some 1 µF.

A question for the design is the primary power source. Is this via a classic mains transformer(s) or from a main SMPS with a fixed voltage (e.g. 40 V) ?

The classic form for CV/CC regulation is with a separate supply for the regulator part, so something like 2 transformers. This may make sense also with a preregulator, as the control parts needs a possibly higher voltage anyway.

[andrewtaylor]

Well, the 30V/3A is a 1000x times handled DIY project,
so I`d recommend not to reinvent the wheel.

For the threadopener:
Please take all goodies from this (three episodes) really well discussed design:

https://www.paulvdiyblogs.net/2015/05/tuning-030v-dc-with-03a-psu-diy-kit.html

It is worth the mindful  reading and follow the pro an cons the author discusses there.

[nike9307]

100 µF for the output is very feasible. It gets tricky (needs fast regulation) if one aims for less than some 1 µF.

This is my starting goal, if i can get it lower i surely will. The requirements are still floating, as i do it to learn something new. I've found that the best way to learn is to make in into a project.

A question for the design is the primary power source. Is this via a classic mains transformer(s) or from a main SMPS with a fixed voltage (e.g. 40 V) ?

For the first prototype its going to be a fixed 36V brick, that can easily be hidden behind my work bench, along with the other bricks and cables. That way I'll also have higher voltages to drive N channel MOSFETS and i already have the brick.
For a later revision I'm thinking USB C input power, with a buck-boost pre-regulator. Unfortunately buck-boost regulators require manual compensation that depends on the output voltage, so its a problem I'd like to leave for future me.
Most modern switchers have internal compensations that should work well over their entire working range and i'll be able to focus on the linear section.

[nike9307]

Well, the 30V/3A is a 1000x times handled DIY project,
so I`d recommend not to reinvent the wheel.

For the threadopener:
Please take all goodies from this (three episodes) really well discussed design:

https://www.paulvdiyblogs.net/2015/05/tuning-030v-dc-with-03a-psu-diy-kit.html

It is worth the mindful  reading and follow the pro an cons the author discusses there.

Reinventing the wheel can be quite fun!
I'll check that blog post out. It looks highly detailed and I'm sure i can learn something out of it.
I'll read it in detail tomorrow, but a cursory glance at the 470uF output capacitor kind of leaves me wanting more ... or less in this case 
Thank you.

EDIT: Turns out he has a newer version here - https://www.paulvdiyblogs.net/2017/07/my-new-power-supply.html. I'll also check that one out as well.

[PCB.Wiz]

I'm aware of the stability issues. I have a scope at work and my boss is cool. He lets me use the work equipment for personal projects.
I'm thinking of using extended SOA N Channel MOSFETs driven by a high current unity gain op-amp with a series resistor on the gate. I've simulated that this driving it like that is simple and causes no issues.
If you have a recommendation for an audio amp MOSFET I'm happy to hear it
I'd like to use a MOSFET so the voltage drop is minimal. If i can get it stable and low noise at 0.3-0.4V dropout voltage i could probably push the max output current even higher.

Another variant topology is to use NMOS with the Drain = GND, so you can better cool the package.
Spice indicates that's a bit fussier than source follower, but it has practical mounting & thermal advantages. You could start with source follower and then change as you get a feel for things ?


Picking a FET these days is more of a challenge. 
The older Audio MOSFETS (Sanken, Toshiba)  are nearing EOL and niche parts, but modern MOSFETS do not come in thermally friendly packages, as they expect switching use, not linear.
Not all of them give SOAR details, and the capacitance curves vary with voltage and vendor.

I guess you start with a package/cooling choice, and work backwards. 
TO3P/TO-247 are still available, and higher voltage parts are low enough RDS and cheap enough to also consider these days, or there are 'big' SMD packages like TOLL etc

eg  A Toshiba TK39N60W5 is 600V, 74mOhms with good SOAR, but Cap values vary widely with bias, in TO-247.

A MOT MOT8125T comes in TOLL-8  85V 220A 2.5mΩ@10V,30A 232W which could be clamp-cooled in a PCB sandwich. 

There are many SSHD coolers on aliexpress that could be useful spreaders.

Or a smaller SMD fets like TO252 (or DFN5x6) 15N10 have lower powers, but also sub nF Ciss, and are very cheap, so you could apply 4-6? of those in a clamp cooling setup.

[Kleinstein]

With a preregulator / SMPS stage the choice of the power FET is not that critical. The power loss is small and high power would only be for a short time (e.g. 1-10 ms rage) at most. 30 V is also not hat high in power. The choice of exact parts is anyway a later step.

The more fundamental question to decide on is if a single supply voltage or 2 seprate ones are used. The classic solution, found in most lab supplies has an isolated supply for the regulator part. This would need extra effort when starting with a fixed voltage SMPS as the power source.
It is possible to also use just a single source, but this is than a different configuration, like the kit use in the link  https://www.paulvdiyblogs.net/2017/07/my-new-power-supply.html.
AFAIK the Chinese kit is based on an old design that used 741 type OPs. The Kit has an issue with too high a supply to the OPs. Getting 30 V this way can be a bit tricky as the OP-amp needs to operate on the full voltage. The need for the negative supply can also be a bit unconveninet. With a single supply capable OP-amp one could get around the negative supply or use on a single diode drop as a minimal negegative.
The rather high ripple in the CC mode is due to a layout problem. With some care one should also get away with a smaller output capacitor in the type of circuit. Beside the stablity they also want the capacitor to limit the overshoot for the CC to CV transition. There are other ways to speed up this part and thus get away with less capacitance for this purpose too.

[nike9307]

I'm thinking of using a TH package transistor that is directly mounted to the top aluminum plate. It has very low thermal resistance, so I'm not concerned about heat.

[nike9307]

Interesting development. I wanted to model hot this thing would get, so i know how much losses the design can get. Since i don't have experience with a thermal simulator, I got a aluminum plate with around the same size as the one in the final enclosure, mounted a couple of 7812 on it and started to draw current from them with a DC Power Load. I could configure the losses by adjusting the input voltage and the output current.
The results were disappointing to say the least. i was hoping to get at least 5-6W of loss budget, but the plate got to around 35*C with just 2.4W. Since I'd like to keep this below 40 when i use it in a 25*C room, this is my maximum loss budget.
I'm looking into ways of doing this with just a switch mode convertor + some way to filter out the noise and give me a clean and quick CV-CC mode switch.

[PCB.Wiz]

Interesting development. I wanted to model hot this thing would get, so i know how much losses the design can get. Since i don't have experience with a thermal simulator, I got a aluminum plate with around the same size as the one in the final enclosure, mounted a couple of 7812 on it and started to draw current from them with a DC Power Load. I could configure the losses by adjusting the input voltage and the output current.
The results were disappointing to say the least. i was hoping to get at least 5-6W of loss budget, but the plate got to around 35*C with just 2.4W. Since I'd like to keep this below 40 when i use it in a 25*C room, this is my maximum loss budget.
I'm looking into ways of doing this with just a switch mode convertor + some way to filter out the noise and give me a clean and quick CV-CC mode switch.

Always a good idea to check power vs temp rise in a mock-up   

You could always give two outputs ?
A Switching one for higher voltages / currents, limited only by the  switcher...

eg - I saw this new release from Diodes inc  https://www.diodes.com/part/view/AL8890Q

and a second low noise, more modest average currents from a linear post regulator.

A part like TPS7A4701 has very low noise and high PSRR, with some aggressive copper design it could give a useful quiet linear output.
Or, just stick with a OpAmp+ref+Power Device ?

[David Hess]

AFAIK the Chinese kit is based on an old design that used 741 type OPs. The Kit has an issue with too high a supply to the OPs. Getting 30 V this way can be a bit tricky as the OP-amp needs to operate on the full voltage. The need for the negative supply can also be a bit unconveninet. With a single supply capable OP-amp one could get around the negative supply or use on a single diode drop as a minimal negegative.

Cheap 741s may support 44 volt supplies, but not many single supply parts do.  If they are still available, the LF356 is fast, may support a 44 volt supply, and has a common mode input range to the positive supply.

You are correct. I'm aiming for minimal capacitance on the output. Less than 100uF if possible. If i use an N channel MOSFET, i should be able to stabilize the loop with less capacitance.

Output transistor type does not matter for the needed output capacitance.  There is nothing wrong with adding a little current buffer to help drive the output transistors.

I'm also going to try to design an active circuit that draws current from the output cap, so it doesn't go to the output. Crowbar type circuit, that discharges the output. This is however entirely an optional fun challenges and I'll be happy with the performance of 100uF on the output.

Below is an example of a power supply which meets all of your requirements.  I am not suggesting this design, but gives some idea of what is required.

D2 clamps the LM101A current error amplifier for faster response.  That plus careful frequency compensation allows for an output capacitance of just 0.22 microfarads in series with 10 ohms despite supporting a 10 amp output.

I would move the current control loop from the collector to the emitter to further improve the frequency compensation.

[iMo]

I (with help of xavier60) spent a week messing with a cheapo PSU based on the Harrison type (basically the schematics at the top of this thread) and I have to admit it works nice. I decreased the output capacitor from 470u to 47u without a problem (some minor changes in the original design). The rather old schematics has been replicated perhaps X-million times with many almost identically wired PSUs sold in last 25 years. So no need to "reinvent" the wheel. What would be required is simply to a) adjust the loops better based on the modern opamps and transistors used, b) perhaps a better voltage reference, some other minor changes. Also from the principle it does not require opamps for 30-50V voltage, afaik. It requires several floating sources, however.
The tricks used for example above with the LM101/201/301 etc. are, imho, not easy to replicate as those opamps are obsolate..

PS: below a typical "cheapo" psu design (the same architecture as the schematics in the first post above).
In the top there is the floating source for +12V, -12V, -6V (not needed, imho) and the 2.5Vref. Its ground is at +output post. The opamps are powered from that source.

The "plus" of the 0.. to +30V regulated ouptut is at the "gnd" output post (the virtual ground for the voltage reference and opamps).

The relays are switching the main transformer secondary windings based on the set voltage (to lower down the power loss at the pass transistor).

With higher Amps PSUs they just add more pass transistors in parallel.

The two bottom floating sources are for the digital meters (V/A) which hang on the Amp shunt and the output voltage posts.

The CC and CV 741s have to be replaced with something more modern and the capacitors around them (the feedbacks) have to be adjusted/finetuded accordingly (C or RC etc)..
You may go down with the original 470uF output capacitor after some finetuning (like I do use 47uF there instead).

[David Hess]

What would be required is simply to a) adjust the loops better based on the modern opamps and transistors used,

"Modern" operational amplifiers and transistors perform no better than old ones in this application.  Ring emitter transistors, power MOSFETs, and fast operational amplifiers have been around for 2 generations now.

My point is that nothing about modern parts will make a low output capacitance design easier.  The lack of an alternative to the "obsolete" LM301A actually makes it more difficult.

The tricks used for example above with the LM101/201/301 etc. are, imho, not easy to replicate as those opamps are obsolate..

I used that example because it shows clamping, but clamping does not require a 301A type of operational amplifier.  Clamping can also be implemented as part of the feedback loop.  A faster operational amplifier is not a solution because the frequency compensation requirements and network constrain speed leading to excessive recovery times.  Most power supply designs rely on the output capacitance to smooth transitions between voltage and current mode, but if the output capacitance is minimized, then clamping becomes necessary.

The highest performance solution is to use operational transconductance amplifiers which recover instantly because their current outputs are self clamping, and this is what regulator ICs like the UC3834 and LM723 do, but this will almost always mean going with a discrete design.  I have seen one design, shown below, which uses a pair of LM723s just for the operational transconductance amplifier in each one.

[PCB.Wiz]

The highest performance solution is to use operational transconductance amplifiers which recover instantly because their current outputs are self clamping, and this is what regulator ICs like the UC3834 and LM723 do, but this will almost always mean going with a discrete design.  I have seen one design, shown below, which uses a pair of LM723s just for the operational transconductance amplifier in each one.

You can also find dual OTA in the Feedback controllers like AP4310/AP4320/ME431x series.
This family of parts have a Voltage reference plus Dual Amplifiers intended for Voltage and Current control loops.
There seem to be two groups, some 18-20V and others 36-40V rated.

TI offer these type of controller too, and the new TI TL103W 'B' version claims this

The upgraded TL103WB features improvements such as a wider supply range (up to 36V), lower supply current (275μA/amp) and tighter voltage regulation.
This regulation can be achieved through low offset voltages for both operational amplifiers (0.3mV typical) and tight tolerances for the voltage
reference (0.44% at 25°C and 1.04% over operating temperature range).

TI show SO8 and SOT23-8 packages. 
Note tho this part has a shunt reference that needs user to supply 1mA current, many others include the Ref current in their supply budgets.

[David Hess]

You can also find dual OTA in the Feedback controllers like AP4310/AP4320/ME431x series.

TI offer these type of controller too, and the new TI TL103W 'B' version claims this

The AP4310 and TL103WB use voltage feedback operational amplifiers, but the AP4320 uses OTAs.  I could not find the ME431x, Chinese?

The choice between operational amplifiers and operational transconductance amplifiers is just an implementation detail.  OTAs make a fast response design simpler, but either can be used.

723s currently cost $0.65 each, and AP4320s cost $0.46, but I would use a pair of 723s because of their uncommitted pinout, or just use the operational amplifiers of your choice.

Another interesting option is the LM13700 OTA which costs $1.28, but only one is required because it has 2 uncommitted OTAs.  I will have to try this out some day, maybe soon.

[jonpaul]

All was detailed in 1960s by Hewlett Packard.

(HP) Harrison Tech Let 2   1967
Constant Voltage/Constant current Regulated Power Supplies

http://hparchive.com/Application_Notes/Harris-Tech-Letter-02.pdf

and
HP AN 90   DC Power Supply Handbook ( 1967)
http://hparchive.com/Application_Notes/HP-AN-90-DC-Power-Supply-Handbook.pdf

Enjoy,

j

[nike9307]

You guys really went all out on the discussion.

Why would an OTA give better performance? As i mostly work in digital, I've never even seen them as components.

For far my favorite design to draw inspiration from has been this https://github.com/eez-open/modular-psu/blob/master/dcp405/EEZ%20DIB%20DCP405%20r3B3.pdf.
Its a part of a larger power supply, that looks absolutely awesome, but is massive (and not sold in Europe). Its a standard switching pre-regulator followed by a linear one, but its designed really nicely.

Currently my biggest problems is the loss budget. Given that its only around 2W, I'm considering going to a fully switched architecture, without a linear stage. This will add output ripple, but it will stay small and quiet.

[Kleinstein]

It is not that an OTA per se gives better performance. It is just convenient to link 2 current output, to do the minimum function from current / voltage regulation. This way the compensation for the 2 loops is linked and this can help with the cross over bettern CC and CV mode.

The classic solution with 2 largely seprate regulators / compensation can show over-shoot from some kind of integrator wind-up. It is however also possible to use a single compensation in this case (e.g. usually done in SMUs this way). It is just extra effort.

With only 2 W for the heat budget one may indeed not have much spare for a linear stage.

[David Hess]

Why would an OTA give better performance? As i mostly work in digital, I've never even seen them as components.

With a constant current and constant voltage regulator, there are separate error amplifiers for the current and voltage control loops with their outputs combined.  When one error amplifier is controlling the output, the other error amplifier saturates with its output stuck high or low.

When standard operational amplifiers are used, then the saturated amplifier takes time to recover because it must charge or discharge its internal frequency compensation network.  This leads to their limited slew rate specification.  If external compensation is added, then the external compensation network must also be charged or discharged, so faster operational amplifier do not necessarily make for faster response because they require more external compensation.

Operational transconductance amplifiers have no internal compensation limiting their slew rate, and external compensation may be applied *after* their current outputs are combined, so recovery time can be hundreds of times faster because the saturated amplifier is disconnected from the external compensation until it takes control.  Integrated regulator controllers, for example the 723, TL494, and various SG and UC parts, usually use OTAs.

An OTA may be thought of as a normal operational amplifier with the VAS (voltage amplifier stage) and output buffer removed, exposing the transconductance output (current) from the input stage.  Since the output is a current, it may be clamped directly.  In the first example I gave, pin 8 of the LM101A sort of represents this point, which is why it can be used to clamp the output of the LM101A to improve recovery time.  No modern operational amplifiers have this capability.