Discrete, linear power supply design

[ifonlyeverything]

For my next project I'm interested in building a discrete linear power supply. I am targeting something with CC/CV limiting (set by potentiometer or external ADC) adjustable down to 0, and I'd like a +/- 30V/3A output... maybe with the ability to do series or parallel operation if it won't massively complicate the control loop. Anyways, I've looked at various designs online and I've noticed some variation in how linear power supplies are constructed:




First screenshot (Post Apocalyptic Inventor) uses a current source to turn on the pass transistor, which is controlled by both op amps sinking current. Second screenshot (Kerry Wong) is turning on the pass transistor via IC1A through T6/T5, and then managing current limiting via sinking through D1 and IC3B.

I've read that feedback networks which pass through multiple layers of transistors (T5/T6 in the second screenshot) can yield instabilities due to gain and non-linearity, although this design does seem simpler as the op amp power rails are easier for me to understand, and I can use lower voltage op amps. I've played around with LTSpice simulations some and I can get something like the second screenshot to work with various resistive loads. In the real world, however, how likely is this design to work without oscillations or other problems? How are professional lab bench linear power supplies designed -- what sort of op amp topologies and feedback networks are commonly used? Are there any good reference designs that I should study to better understand this?

[Kleinstein]

Many commercial power supplies follow the concepts similar to the upper circuit. So with an extra supply for the regulator part. A nice point of this design is that it is pretty flexible and can be used both for low voltage with high current and high voltages at low current. One can also use MOSFETs instead of the darlingtons with very little change.  There are a few details that are still a bit odd in the voltage adjust part.
The regulator may also want a few more details to improve stability.
For a high power one would usually add some transformer tap switching. This often are relays at the AC side. Alternatively one can use a 2nd set of power transistors and a diode for an electronic "switch" over.

The lower circuit principle is possible too. There is a pretty long thread in the forum on a similar supply with a pretty well developed circuit for this (https://www.eevblog.com/forum/projects/linear-lab-power-supply/msg2388873/#msg2388873).

The circuit as shown has a few off point that could cause trouble:
The usually needs to be some ouput capacitance, even if just in the 1 µF range, often more like 10-100 µF.
C1 to a certain degree does the job, but is much better placed at the output.
T6 is a rather high power type - this can be a smaller one to also be faster.
R21 is calling for trouble - it should not be there, so that the current regulation gets full priority.
IC3 B and R20 are not really needed - they only add delay.
One usually needs some minimum load (e.g. a constant current), as the transistors get slow at very low current. The circuit with more transistors can be a bit more nonlinear, but not that bad. The minimum load is also missing in the upper circuit.
The current sense part can be slow and tricky. It would also include some of the control current and thus not be super accurate. Current sensing at the low side may be easier.

Doing a simulation is a good idea to get an idea if the circuit tends to oscillate and how fast the CC-CV transition is expected to be. This may determain if more output capacitance is needed to limit voltage overshoot.

[pqass]

How are professional lab bench linear power supplies designed -- what sort of op amp topologies and feedback networks are commonly used? Are there any good reference designs that I should study to better understand this?

See attached schematic of the HP E361xA power supplies.  It definitely works well and is fairly easy to understand and build.

Since the schematic is for several models, E3610a, E3611A, and E3612A, there are different components fitted for a given model making it appear more complicated than it is especially around the series pass transistors.  See the service manual on the differences between models and component choice (pages A-5 thru A-9): http://www.doe.carleton.ca/~nagui/labequip/powersupply/E361XA_Operating%20Manual.pdf  Unfortunately, the manual doesn't specify Q1, Q3, Q6, Q7, however, in my HP3611A, there are just two MJ3042 (darlingtons, not MOSFETs) nor are there any Q4 and Q5.

An a-ha moment for me was when I realized that the ground on the floating Reference and Bias Supply (+/- 12V, +5V, CREF) for the op amps is attached to the positive binding post (labeled "+S")!
You might consider it a negative supply (assume the red binding post is 0V) with the voltage control knob controlling how negative the black binding post goes.

Also note, S1A switch allows the user to see the current (actual vs set) by attaching the meter (7107 ADC-based) to the shunt or the current control op amp (without needing to short the output posts).

[ifonlyeverything]

How are professional lab bench linear power supplies designed -- what sort of op amp topologies and feedback networks are commonly used? Are there any good reference designs that I should study to better understand this?

See attached schematic of the HP E361xA power supplies.  It definitely works well and is fairly easy to understand and build.

Since the schematic is for several models, E3610a, E3611A, and E3612A, there are different components fitted for a given model making it appear more complicated than it is especially around the series pass transistors.  See the service manual on the differences between models and component choice (pages A-5 thru A-9): http://www.doe.carleton.ca/~nagui/labequip/powersupply/E361XA_Operating%20Manual.pdf  Unfortunately, the manual doesn't specify Q1, Q3, Q6, Q7, however, in my HP3611A, there are just two MJ3042 (darlingtons, not MOSFETs) nor are there any Q4 and Q5.

An a-ha moment for me was when I realized that the ground on the floating Reference and Bias Supply (+/- 12V, +5V, CREF) for the op amps is attached to the positive binding post (labeled "+S")!
You might consider it a negative supply (assume the red binding post is 0V) with the voltage control knob controlling how negative the black binding post goes.

Also note, S1A switch allows the user to see the current (actual vs set) by attaching the meter (7107 ADC-based) to the shunt or the current control op amp (without needing to short the output posts).

Thanks for posting this. Still trying to wrap my head around some of this.

What's the purpose of JP1/JP2?

How exactly is the current sensing working here? I understand that U4A sets the current reference (CREF) which is the reference for the current error amplifier U4B, but I'm not understanding how this is all tying together. U4B has one input tied to the center tap (+S) and the other input to a node with CREF and R27/R34 divider. How does this work? I was expecting to see something similar to the first screenshot in my OP.

[mariush]

In case it helps, here's more schematics of power supplies.

Also DC Power Supply Handbook (AN 90) from HP : https://www.changpuak.ch/electronics/PowerSupply/HP_AN90B.pdf

and the service manual with schematic for a TTi  'S Series' power supplies (too big to attach, 7 MB) : https://fastupload.io/bKPNqWsnSDkwknr/file

[pqass]

How exactly is the current sensing working here? I understand that U4A sets the current reference (CREF) which is the reference for the current error amplifier U4B, but I'm not understanding how this is all tying together. U4B has one input tied to the center tap (+S) and the other input to a node with CREF and R27/R34 divider. How does this work? I was expecting to see something similar to the first screenshot in my OP.

CREF isn't like it sounds; a stable reference. It moves according to the front panel current setting knob R19 by -1.00V per 0.25A set (eg. at 0.25A setting it's -1.00V, at 0.50A it's -2.00V, ... at 1.00A it's -4.00V, at 1.5A it's -6.00V).

The R27+R34 (6.8R, 681K) divider is so lop-sided (10 ppm) that the center node is effectively tied to TP3 (left side of R2).

CREF is scaled down by 4.76% via R23+R24 (2.5K, 50K) divider.  So if the current setting calls for 1A, CREF would be -4.00V, we then multiply by 4.76% to get -0.19V at the -in of U4B (when no load on output).

As the load draws current, there is a voltage drop over R2. This causes the U4B -in node to creep toward 0V until -in > +in when the output of U4B flips to -12V which turns on Q2 which then shuts off the series pass transistors..

Note: values for R2,R22,R23,R27,R34 are specified for the E3611A; other models use different values.

What's the purpose of JP1/JP2?

In the E3610A (max out at 15V) and E3611A (max out at 35V) models, JP2 is jumpered, and JP1 is open.
In the E3612A (max out at 120V) JP2 is open, and JP1 is jumpered.

The jumpers decide to set the voltage comparison point after the shunt (E3610A, E3611A) vs before the shunt (E3612A).  Normally, the voltage comparison feedback is from the output terminals (after the shunt). But given the shunt value (R2=1.78R for the E3612A) at its max current of 0.25A, that's only 445mV difference in voltage (no load vs max load). That's effectively "in the weeds" at 120V. I guess they did that so any fluctuating current doesn't mess with the voltage regulation which is harder to do at high output voltages.

[xavier60]

For my next project I'm interested in building a discrete linear power supply. I am targeting something with CC/CV limiting (set by potentiometer or external ADC) adjustable down to 0, and I'd like a +/- 30V/3A output... maybe with the ability to do series or parallel operation if it won't massively complicate the control loop. Anyways, I've looked at various designs online and I've noticed some variation in how linear power supplies are constructed:




First screenshot (Post Apocalyptic Inventor) uses a current source to turn on the pass transistor, which is controlled by both op amps sinking current. Second screenshot (Kerry Wong) is turning on the pass transistor via IC1A through T6/T5, and then managing current limiting via sinking through D1 and IC3B.

I've read that feedback networks which pass through multiple layers of transistors (T5/T6 in the second screenshot) can yield instabilities due to gain and non-linearity, although this design does seem simpler as the op amp power rails are easier for me to understand, and I can use lower voltage op amps. I've played around with LTSpice simulations some and I can get something like the second screenshot to work with various resistive loads. In the real world, however, how likely is this design to work without oscillations or other problems? How are professional lab bench linear power supplies designed -- what sort of op amp topologies and feedback networks are commonly used? Are there any good reference designs that I should study to better understand this?

Something like that popped up on a thread a while back. It evolved into a design that works well.
https://www.eevblog.com/forum/projects/linear-lab-power-supply/msg2020999/#msg2020999
This is the last version, https://www.eevblog.com/forum/beginners/lm324-power-supply-with-variable-voltage-and-current/msg3582664/#msg3582664

[iMo]

Also here is a several years long thread where many participants spent a lot of time designing/simulating/testing of a similar design..
https://www.eevblog.com/forum/beginners/lm324-power-supply-with-variable-voltage-and-current/

[MathWizard]

To make a few things, I've been saving old stereo's from the landfill. I made a 20V 1A CV-CC PSU with a big multi-tap transformer. The trans. can do a lot more current than that too. That PSU was all BJT's except for a TL431 I used. I have to learn a few more math tricks before I try making it with op-amps.

I kept the trans, in the stereo case, and added connectors on all the rails. It has a nice big heatsink too. So I can also it in other projects, like a curve tracer. Swapping stuff on/off the heatsink will be a pain, but for now it will be ok.