how does blackdog's PSU work?

[Cerebus]

Is a separate isolated power supply needed? Can I just use a charge pump to generate some voltage above the output? Let's say I can sacrifice some noise performance (my PSU enclosure is not shielded (plastic), so I'm not going to get microvolts of noise even if I try hard).

Strictly it needs to be floating with respect to the output 'ground' as opposed to fully isolated from all the other supplies, i.e. it needs to move with the final PSU output voltage. Obviously you need galvanic isolation from the primary side, but you'll still get that if you steal a few 10s of mA from the bulk supply. The current demand is modest, but not trivial, so all the usual charge pumping tricks (voltage doublers, IC charge pumps and the like) are going to be a bit marginal. You need current consumption figures for the actual control circuity under worst case conditions to know if you'd be able to get away with it.

A dirty little trick is, if you are using a toroidal transformer for your 'bulk' supply, you can add a couple of minor windings yourself to get your secondary supply. If you've got a 200VA transformer, it's not going to care about a couple of 2VA windings added on the outside. Obviously you've got to use the right materials, right insulation etcetera, etc. Bone up on transformer construction techniques before even thinking about attempting this, so you know how to do it safely. I say toroidal transformer because it's easy to add windings. Obviously you could do the same with an EI transformer or similar, but it would be so much physical work that you'd be better off ordering a custom transformer from the start.

[alm]

So, the question is: What is a reasonable slew rate for an applied step load when testing PSU regulation recovery times? Are there industry standards? (I couldn't find evidence of any.) Or do people just test with whatever slew rate the electronic loads available to them support?

This Keysight application note specifies that the load should have a rise time at least five times faster than that of the power supply. Obviously this is not particularly helpful if you are trying to establish the latter. I guess that means increasing the slew rate until you observe no change in power supply behavior?

[T3sl4co1l]

A charge pump would be workable, but probably too noisy (even if you don't need as low noise as the original design was aimed at!).

If the floating negative supply can be eliminated (say by using a RRIO opamp, and omitting the LEDs, using schottky diodes instead?), then only the positive would be left, and a voltage doubler could be used to make an extra-high supply, which powers a CCS which powers the floating rail.

RRIO amps don't tend to be very low noise, but that's alright.

Tim

[Cerebus]

This Keysight application note specifies that the load should have a rise time at least five times faster than that of the power supply. Obviously this is not particularly helpful if you are trying to establish the latter. I guess that means increasing the slew rate until you observe no change in power supply behavior?

That's helpful, thanks.

The minimum possible rise time of the PSU is pretty much limited by the f_t of the pass transistors and is relatively easy to calculate/measure as is the actual rise time once the whole control loop is taken into account.

My problem was just understanding what was an acceptable load profile for measuring transient response. Obviously the response overshoot/undershoot (and any ringing) of a PSU gets worse with faster edges on the test load, and gets better with slower edges. I suppose that my question really ought to have been "What's the slowest slew rate that is regarded as an acceptable test?". If one sets the criteria as 5x faster than the supply rise time then that seems reasonable to me as you're basically saying "faster than the supply can respond". Kind of obvious really, but experience tells me that I'm more likely to miss the obvious than the subtle.

[T3sl4co1l]

Directly applicable example: I was testing a new DC-DC module against my bench charger/discharger supply.  The former has loop bandwidth in the mid 10s of kHz, while the latter has a rise time of a few milliseconds.  I was unable to detect any perturbation on the module output, while switching the discharger on and off.  In contrast, a MOSFET + resistor + function generator easily shows the step response of the system, which in this case was limited by the module's output filtering network (amazingly, no changes needed on the compensation network!).

Tim

[blackdog]

Hi EXE,


If its only a "normal" hobby supply, you can trow a few extra windings on the power transformer.
But wat is your problem with a extra small transformer for the reference section?

Its also possible to build de Circuits Online power supply, maybe with my modifications?
https://www.circuitsonline.net/forum/view/131554
Google translation time again 

This is a link for a more up to date schematic, it is not a complete schematic, i removed the "difficult" components.
But to make Tim more happy with me, (i mean this friendly) i changed al the diodes who are of a Schottky type and the last MOSfet i dit forget.



Analog Devices has a new opamp, interesting for fast lineair power supply's.
Its the ADA4625-1, minimum 5V en Max. 36V power supply.
Fast, 48V/uSec, low noise, low bias current and relativ good DC specs.
But what is really special, is the phase margin, which is 88 degrees, normaly it is around 65 Degrees for opamps.

If its realy this good (i have to do some test) than i can choose for a lower value electrolitic capacitor on the output or better dynamic performance.
We shall see what the tests will tell me.

Kind regards,
Bram

PS
I started with a TL431 Reference for this power supply, look a the link below, the pictures tells enough if you cant read Dutch.
https://www.circuitsonline.net/forum/view/116156
 

[Jwillis]

Most definitely an interesting design.I will have to give this one a try sometime.I'd like to shake your hand for all the hard work you put into it and would very much like to see the finished design.Bravo!

[exe]

So, the question is: What is a reasonable slew rate for an applied step load when testing PSU regulation recovery times? Are there industry standards? (I couldn't find evidence of any.) Or do people just test with whatever slew rate the electronic loads available to them support?

What I often see in specs is something like "50us recovery time". As I understand, this reads as "in 50us or less the output level gets back to normal (or settles within 5-10%) under any step load". But 50us is waay to slow, imho.

So, what is a good recovery time? My benchmark is LT3080 which I consider a fast regulator (there are some faster, but not as versatile). It has recovery time around a few us (the datasheet has a few pictures). I also check overshooting and, uhm, undershooting at step load. If it's below 100mV  (at 3.3V) for step load 0.1A => 1A, then it's a good power supply. (Of course, slew rate of step load affects results)

So, everything close-enough to LT3080 I consider a good PSU. I assembled my variation of blackdog's PSU, it was good. If I find oscilloscope screenshots, I'll share.

BTW, there are many things affect performance. Like, pass transistor. My 2STA1943 (from ST) showed better performance than 2TA1943 from Toshiba. But I measure the transistor alone, not in the circuit. So, I don't know what would be real impact. Also dropout voltage affects performance a lot. BJTs don't like to work near saturation voltage (imho).

My two eurocents, I'm no expert 

[Cerebus]

So, the question is: What is a reasonable slew rate for an applied step load when testing PSU regulation recovery times? Are there industry standards? (I couldn't find evidence of any.) Or do people just test with whatever slew rate the electronic loads available to them support?

What I often see in specs is something like "50us recovery time". As I understand, this reads as "in 50us or less the output level gets back to normal (or settles within 5-10%) under any step load". But 50us is waay to slow, imho.

You misunderstand, what we're talking about is the slew rate of the current step waveform of the load that is used to test the PSU load effect recovery time. We're not talking about the load effect recovery time itself.

As it is, we already have an answer, one that convinces me anyway, which is 5 times faster than the rise time of the supply.

[blackdog]

Hi EXE, 

Measuring the performance of a fast linear power supply is not easy.
And if you are say above the 2-Ampere output current, you will also have to take the EM field of the all wires  with these current into account.

I already showd a picture of my setup and the BNC connector has a 90 degree angle to the current wiring.
Even de sense wires i try to give them a 90 degrees angel.

The measuring point for performance measurements are the points where the sense wires soldered to the current wires.
That is on the back of the 4mm banana sockets, every mm extra wire give a extra L and R.
I want to measure the performance of the electronics ande not the wiring during testing 

And i explaned already that i also do testing with one of my Dummy Loads for more practical testing.
If you connect your device under test with 1M cables to te Power supply, and most users do not twist the cables, all those nice specs are out of the window 

My standard measuring puls from one of my generators used with my Dummy Loads.



This is a picture of how my 200-Watt Dummy Load is performing in one of my tests of this dummy load afther building him.



To see how a power supply is reacting on realy fast pulses, i use my Jim Williams dummy load.

Kind regards,
Bram

[Kleinstein]

There is a limited advantage of a super fast regulation for a lab supply: Once you add 1 m of cable the response at the other end of the cable will not be very fast anymore and one would likely see ringing from the cable. So if one has a critical load, there should be local decoupling caps as part of the test circuit. For the lab supply there is limited need to be faster than the cable - it may be even an advantage to show some damping to the cable instead of the virtual short.

This is a little different with the LT3080: this is a voltage regulator that can be placed close to a possibly sensitive load. On the down side the LT3080 will not show such good DC specs - in parts due to not having extra sense lines down to the terminals. This not so good DC performance helps with AC stability and thus allows the fast response. Also the LT3080 is missing the adjustable current limit and the very fast recovery is likely with a small output capacitance, while low drop on fast transients is with more output capacitance.

There are different requirements for a voltage regulator compared to a lab supply:
The voltage regulator needs to get good response with a well behaved load with suitable load capacitance. Performance with a poorly chosen cap can be poor, up to oscillating. Some DC drop is usually acceptable.

A lab supply should react reasonable well with most loads. And should not oscillate with any load (with less than 90 degree phase shift). This often requires the capacitance with ESR at the output that with a well behaved load mainly slows down the response. So optimization is not for the best cast tested with a voltage regulator, but more for the difficult cases. Usually DC load drop should be very low,  which may require some compromises in the medium frequency range.

[Cliff Matthews]

I enjoyed reading that    What's a good view on the necessity of both sinking and sourcing on a bench supply vs lab-grade?

[David Hess]

1) The output voltage is defined by current flowing through resistor P2  (20k multi-turn pot). But why voltage drop on P2 should be constant? I don't see a current source to ensure this. I only see a voltage source which is +5V above the positive output.

The inverting input of operational amplifier IC1a is tied to the output.  The non-inverting input follows the inverting input due to negative feedback.  Usually this is described the other way but the result is the same.  So the top of P2 is equal to the output voltage.  Since the reference produces an output 5 volts higher than the output, R32 which produces the reference current has a constant voltage across it.

2) what is return path for current flowing through P2?

The reference produces a voltage which is 5 volt higher than the output voltage so the return path for current through P2 is the same as the current through the output.

The above is just a repetition of what T3sl4co1l said but may be more understandable if said in a different way.


I like the floating design which allows the use of low voltage operational amplifiers and easy high side current sensing.  If you are going to build a power supply at this power level, then the cost of an extra but small transformer is well worth it.

I like that the output voltage does not rely on modifying the closed loop gain which compromises frequency compensation although not usually to a degree which matters.

The displayed load transient response is poor; power supplies should have a very tame transient response to handle difficult loads.  In production designs, I slightly overcompensate for the *worse* case of component and load variation.

The controlled ESR bulk output capacitance is just a bad idea and indicates a problem with the frequency compensation.  I have seen designs with comparable output voltage and current which were much faster and better behaved that used less than 1 microfarad of output capacitance in series with like 10 ohms although I question whether this level is performance is actually ever needed in such a high power supply unless it is part of a source meter.  Someone else can run the numbers but this looks like dominant pole compensation on the output which largely defeats the purpose of using fast transistors and operational amplifiers.

Why use a Sziklai pair instead of a Darlington pair when not required?  This is especially odd since the output power transistor is a PNP instead of NPN although the difference in price probably does not matter anymore; PNP power transistors used to be much more expensive than NPN power transistors.  I suspect the lack of local feedback around the Sziklai pair is causing problems.  Designs like this using much slower transistors often include local feedback.  Even fast Darlington designs often use local feedback.

I would unload the outputs of the operational amplifiers with emitter followers or even FETs.  Heavy loading compromises precision defeating the purpose of using a good reference and remote sensing.  The ultimate performance of the suggested ADA4077 precision operational amplifier is completely wasted here.

The not shown input clamp diodes on the operational amplifiers may be screwing up the performance during transitions between voltage and current mode.  Also, the ADA4077 datasheet is horrible and I would never recommend or use this part because of it.

I am suspicious that the ADA4077 and NE5532 displayed any difference in AC performance with everything else going on.

I think the output TVS should have been an SCR crowbar circuit.  I would expect it to behave like one once and only once if it was actually needed.  I hope it was not needed for handling overshoot.

The preregulator is clever but I am surprised switching artifacts do not show up in the output.  Old designs used phase controlled SCRs and a big inductor, SCRs and transformer taps, or a buck switching regulator.

[blackdog]

Hi David,

Can you explane this?
The displayed load transient response is poor; power supplies should have a very tame transient response to handle difficult loads.  In production designs, I slightly overcompensate for the *worse* case of component and load variation.

David did you read everything?
Wat is bad about this test see the two pictures below, do you find this horrible?

This is the current puls, 0.5 to 2-Ampere.


This is the respons on the power supply connector, about +-3mV abberation, sorry for the noise, the scoop is only 2mV/Div.



Ooo...
Why use a Sziklai pair instead of a Darlington pair when not required?  This is especially odd since the output power transistor is a PNP instead of NPN although the difference in price probably does not matter anymore; PNP power transistors used to be much more expensive than NPN power transistors.  I suspect the lack of local feedback around the Sziklai pair is causing problems.  Designs like this using much slower transistors often include local feedback.  Even fast Darlington designs often use local feedback.

My measurements tell me different.
The rightmost line (F) with the text Double Compound transistor indicates the output impedance of this power supply.
And it is measured with the NE5532A as used opamp, can you point my to a powersupply that has a lower Ri over this frequency range?



Ooo again...
The ultimate performance of the suggested ADA4077 precision operational amplifier is completely wasted here...
Nop, The DC stability is fine with this opamp in use, but the hf performance of the ADA4077 is ofcoure less then with a faster opamp.
Some DC measurements: With the NE5532A at 15V output the DC drift was < 0.2mV, thats less than 0.0015% within a time of 6 hours.

I am suspicious that the ADA4077 and NE5532 displayed any difference in AC performance with everything else going on.
Your susspicion is not correct :-)

Switcher
I had already explained that I do not want a switcher in a low noise power supply.
And that I finally switched to a transformer with tabs and three relays to limit the power loss.

Some attention
The only thing I think about is the opamp of the U control loop, in difficult situations there is quite a lot of energy going to the + input of this opamp.
This can probably be solved with two diodes that are anti-parallel from the + output to the + input of the opamp.

Current limiting
Some pictures about the performance of this power supply and this time the I loop performance.
This is measured over the current sense resistor.
I used one of my dummy loads for a hi current peak.
And this is the result without the transistor for the peak limiting, Within 10usec back to 5-Ampere, fast enough for you?


But I was not satisfied with that yet, so I applied Q6 to further limit the peak current and now its fine by me.


And again, during the development I sometimes test with other conditions than for a 30V and 5-Ampere power supply, which was the starting point.

Some remarks
Let me make it clear again, this is a power supply that has very low noise, good DC stability, and good dynamic performance.
He must meet my requirements, for my purpose.

I do a lot of measurements on sensitive electronics, and sometimes it is necessary that my LAB power supply can deliver more than 2-Ampere with great stability and low noise.
As an example, I develop voltage references that are in small ovens.
And testing these ovens i need a high performance power supply.
I do not want to say that everyone should build this power supply.
I want to show that you can achieve very good specifications with a fairly simple set-up.
If you want, use part of this design for your application.

This design is not meant for production, but only for my LAB.
And if I want to use more expensive parts, I do that because I think this is necessary for my application.

Kind regards,
Bram

[blackdog]

Hi,

Many comments are made about the output network of this power supply.
And yes, this is built differently than with most power supply's.
And Why?...
Perhaps it is good that those who make comments about this configuration read the following document.

www.bramcam.nl/NA/NA-01-PSU/Calex-Power_Impedance-Decoupling.pdf

Perhaps this document makes things more clear.

Kind regards,
Bram

[T3sl4co1l]

Another way to look at it: a diffusion network.  This gives an approximately Z ~ 1/sqrt(F) characteristic, which has equal components of resistance and capacitance.  In a sense, it's the most lossy capacitor you can have, over a wide range of frequencies, and therefore at the least risk of underdamped loads above the controller's cutoff frequency.  (The purpose of the control is to set Zo very low, so within the passband, one should not expect damping to occur -- attaching an LC network to the output, such that the resonance falls in the passband, should indeed be poorly dampened, an indication of the low output impedance of the supply.)

Tim

[Wortel]

Hoi Bram,
Ik zie dat je de LM317 current source als voorbelasting van een hele tijd geleden er weer ingezet hebt, en volgens mij, net als vroeger, met de uitgang aan de verkeerde pin. De constante stroom loopt bij zo'n 3-pins regelaar toch van de output pin,  *door* de weerstand, dus rechtstreeks aangesloten op de sense (of ground)-pin?

Ik heb je NA-01 blog op CO helemaal gelezen, groot respect! maar is dat nu doodgebloed?

Grtz,
Wortel

[blackdog]

Hi Carrot,

No, the project was not dead, but when I was busy with the extended testing of the 230V net filtering, it went very badly with one of my family members.
So it is still a bit of a sensitive point...

But I had already decided to make a smaller finnishd version first (2.5A), so that I can place it better between my other equipment in my LAB.
And 2.5-Ampere is sufficient for most of my applications, i started building of the next test version, see the picture below.
The 5V Reference on this board is a MAX6350, also a fine part.



The basic schematic is ready, the reference section, Power Off Glitch Protection, and control loops and power section are extensively tested and ready for use.
Depending on the used opamps, the compensation capacitors can be slightly adjusted for optimization. (C16 and C16)
Because there is now a somewhat smaller transformer for me, the 230V grid filtering will also be adjusted a bit.

And if I have tested with the new Analog Devices opamps: ADA4625-1 then together with the lower 2.5-Ampere output current, also the 220uF capacitor over the output can be smaler.
This in turn will make the peak currents in the connected D.U.T smaler, in fault situations.
But these are no big changes.

Anyone who still wants to apply DAC control can connect the top of R32 to the U-DAC (0 to +5V).
You will have to calculate the voltage divider for the I-DAC yourself(0 to +5V).
The simplest method is to replace P1 with a 1K resistor and R31 on the top of this new 1K resistor.
If you want two current ranges, connect R27 or R28 to the output of the I-DAC.
You will have to do it with this information, I will not give any further advice about it, you are on your one 

Wortel, thanks for detecting the error in the schematic, it is now Ok (i hoop)

Kind regarts,
Bram

[David Hess]

Can you explane this?
The displayed load transient response is poor; power supplies should have a very tame transient response to handle difficult loads.  In production designs, I slightly overcompensate for the *worse* case of component and load variation.

David did you read everything?
Wat is bad about this test see the two pictures below, do you find this horrible?

The first test you showed indicates a problem in the small signal response.

The test conditions for the others are not clear enough to rely on but they do seem to show a large signal response problem which might be caused by the error amplifier not being able to pull charge out of Q3 and Q6 quickly enough.  I would usually expect a base-emitter shunt resistors on Q3 and Q6, capacitive bypassing to the base of Q3 and Q6, and much stiffer drive to Q3 and Q6 which goes along with unloading the outputs of the operational amplifiers to improve precision.

The precision design shown at the end of my reply for instance would nominally use twice the drive current to the pass element while having 2 orders of magnitude less loading on the error amplifiers.  This actually understates the difference because an integrated pass element has much higher current gain; idle current through the buffer is high to keep the buffer transistor characteristics consistent.

Why use a Sziklai pair instead of a Darlington pair when not required?  This is especially odd since the output power transistor is a PNP instead of NPN although the difference in price probably does not matter anymore; PNP power transistors used to be much more expensive than NPN power transistors.  I suspect the lack of local feedback around the Sziklai pair is causing problems.  Designs like this using much slower transistors often include local feedback.  Even fast Darlington designs often use local feedback.

My measurements tell me different.
The rightmost line (F) with the text Double Compound transistor indicates the output impedance of this power supply.
And it is measured with the NE5532A as used opamp, can you point my to a powersupply that has a lower Ri over this frequency range?

It is not a major factor that I have ever noticed other than more care being needed in controlling the local frequency compensation.  I always found transient response to be dominated by the overall frequency response which is the point here; your first test indicates a problem in this area so a low output impedance from the pass element is more important than it otherwise should be.

Some audio amplifiers use a couple of small signal transistors to add local feedback to the output transistors and reduce output impedance even further.

The ultimate performance of the suggested ADA4077 precision operational amplifier is completely wasted here...
Nop, The DC stability is fine with this opamp in use, but the hf performance of the ADA4077 is ofcoure less then with a faster opamp.
Some DC measurements: With the NE5532A at 15V output the DC drift was < 0.2mV, thats less than 0.0015% within a time of 6 hours.

How good was the load regulation though?  Using remote sense, I would expect the load regulation using the ADA4077 to be almost unmeasurable.  An ADA4077 is not required to deliver a DC drift of less than 0.2mV.

Some attention
The only thing I think about is the opamp of the U control loop, in difficult situations there is quite a lot of energy going to the + input of this opamp.
This can probably be solved with two diodes that are anti-parallel from the + output to the + input of the opamp.

Internally they already have anti-parallel diodes (which is one reason I don't like AD's datasheet; it does not show them) but this is shorting the inputs during large signal conditions and may be upsetting the filter networks for the current and voltage setpoints.  If this is happening, then it is going to show up as long settling time and poorer large signal response which does appear to be a problem.  Low differential input voltage operational amplifiers are almost always a poor choice where open loop operation is possible but in this case even with the feedback clamp network, this may be causing a different problem.

Some remarks
Let me make it clear again, this is a power supply that has very low noise, good DC stability, and good dynamic performance.
He must meet my requirements, for my purpose.

The performance is good.  I just think it should be much better given the complexity and parts selection.

Let me make it clear again, this is a power supply that has very low noise, good DC stability, and good dynamic performance.
He must meet my requirements, for my purpose.

...

This design is not meant for production, but only for my LAB.
And if I want to use more expensive parts, I do that because I think this is necessary for my application.

If it is good enough, then it is good enough and best is the enemy of good.

I have designed and built my fair share of special purpose power supplies for internal production and testing.  My favorite involved the design shown below but modified for extreme precision and low noise using an LT1007 precision operational amplifier, precision reference, and lead-lag frequency compensation which involved 2 more resistors and 2 more capacitors not shown in both the feedback and input networks to optimize dynamic performance.  Load and line regulation from 0 to 1 amp ended up being too good to accurately measure at better at 1 ppm or better.

Note that I am not suggesting this design to you or anybody else.  I just used it as a handy example.

I have been considering doing a precision floating variable output tracking implementation but at lower maximum current but the problem is handling power dissipation at low output voltages without excessive power dissipation.

[blackdog]

Hi DAvid,

Thank you for your remarks, it is late now here in Amsterdam and tomorrow i wil explane some more things.
Like the photo you probably saw first, that is the photo with some ringing, that is from my "torture test" for power supply's :-)

Talk to you later.

Kind regards,
Bram

[exe]

Thank you very much, guys, for all the information you provided.

I entered EE about three years ago (as a hobby). My first device supposed to be a DIY power supply. Now, three years later, after hundreds of hours of reading forums, tech literature, experimenting and building multiple prototypes I'm nowhere near finishing my very first project . But, anyway, please keep discussing. Topics such as frequency compensation or choosing the pass element are quite rarely discussed (at least I didn't find much info, yet alone measurements). Who knows, may be after this thread I'll finish designing my "ideal bench power supply".

[Wortel]

Hi Bram,
Sorry to hear about your family going bad, hopefully things are going better now.
You mentioned a smaller version of the NA-01 low noise PSU on Circuits Online if I remember correctly. Is the schematic available somewhere? That would be nice!

Wortel

[Zero999]

Hi Carrot,

No, the project was not dead, but when I was busy with the extended testing of the 230V net filtering, it went very badly with one of my family members.
So it is still a bit of a sensitive point...

But I had already decided to make a smaller finnishd version first (2.5A), so that I can place it better between my other equipment in my LAB.
And 2.5-Ampere is sufficient for most of my applications, i started building of the next test version, see the picture below.
The 5V Reference on this board is a MAX6350, also a fine part.

Oh no, don't build it like that. The creapage between the primary and secondary is very unlikely to be sufficient, for isolation from the mains. All that's needed is a voltage spike and the tiny gaps on the perf board will arc over, connecting the secondary side to the mains.

Put the mains and secondary sides on different boards, that way you can be sure they're properly isolated.

[blackdog]

Hi Hero999,

Thanks for the warning
But i already now that, i dril out the holes around the 230V powerline connections before i do the real testing, this is only the reference section build.
This is not a final circuit board, i drill a lot in these boards, like in one of my high performance preamps to get the paracitics down.

If you want, take a look at this design of one of my low noise preamp, design for measuring at power supply's and some extra holes in the circuits board!
Date/time: vrijdag 17 maart 2017 16:35:45
https://www.circuitsonline.net/forum/view/135863

Back on topic,
The most important point in the development of a linear power supply is the phase margin of the control loop.
A standaard opamp has about 60 degrees phase margin, every component you place in the loop will bring this phase margin down.
Placing a power transistor(s) in the loop brings it down, if don't place a capacitor over your feedback resistor, almost certainty there will be instability horror.

Look at the many power circuits on the Internet, most of them are compensated to "dead"...
This is because the person who developed the circuit has not understood how the circuit works.
Two examples, Dave Jones fist power supply, I knew from the beginning that it would fail.
He finally learned his lesson with version-2 :-)

The second example is e.g. Gerry Sweeney power supply, look at his video's.
Hi did a fine job, except for the loop stability...

Do not think that I am a gentleman who knows everything, as someone declaimed here on the forum...
In the beginning of the 1974 i started with opamps like the uA709,
my college and i were complaining about it alway was oscillating, what a bad opamp!

But of course we were bad technicians, we did not read the datasheet and aplication notes, we were fools.
Take a look at the LM709 data sheet, and see how universel this opamp is, even gains of 80 dB are possible with reasonable bandwidth.
Would I still use the uA709, no there are now much better opamps for sale, this was only just to make a point clear.
And regularly I have to learn, make stupid mistakes, I am just a human being. 

Thus e.g. only using a fast opamp does not make a good power supply.
The intention is that when you develop the power supply, and that all parts of the schematic are well matched to each other.

The Harrison design i use (from the late 1950's) is really beautiful.
And my part of it, is using modern parts and tuning these for optimum performance.

All designs with extra transistor and opamps in the loop will never reach the performance of this simple design.
And again, most LAB power supply do not need this High End performance.
I have a Dutch famos brand switching power supply, 30V and 5-Ampere, the dynamic performance is "not so good", why?
There is a lot of capacitance on the output en the loop is not fast.
But i still use it regular wen i need higer currents then 1-Ampere and when its not critical.

The time is up, have to work, laters more.

Kind regards,
Bram

[Cerebus]

Topics such as frequency compensation or choosing the pass element are quite rarely discussed (at least I didn't find much info, yet alone measurements).

Don't forget that a power supply is just - in essence - a DC coupled power amplifier. So, texts covering power amplifiers go over much of the same ground. If you want a discussion of pass element choice that's a place to look for it. Just make sure you stick to reliable sources, otherwise you'll find yourself going down the audiophool rabbit hole.