how to design fast bench supply with CC and CV?

[Neomys Sapiens]

In my experience, the LT1468 is as care-free as it is performant. I never experienced instabilities with this OPA.

Was that just on its own, with nothing but the usual resistors in the feedback loop? My concern isn't the op-amp in isolation, but according to the other schematics posted in this thread, it will have a pass transistor on the output, inside the feedback loop. The extra delay due to a transistor, especially a big slow Darlington pair, can cause oscillation, with a fast op-amp. It's possible to avoid this by slowing the op-amp down, with a capacitor between the inverting input and output, but why not just use a slower op-amp in the first place?

As of why the CV opamp needs to be fast, it needs not to introduce much phase lag as it is driven by CC opamps. So, CV opamps needs to be much fast than CC opamps.

Stability is checked in ltspice. In fact, it was the biggest challenge for me. Output stage is a plain push-pull stage with just two transistors, no darlington. This is why I also have to limit maximum output current. I hope I can get 1A, but we'll see. I understand with higher frequencies bjt gain drops. Assuming hfe is 100, getting 1A output requires 10mA drive current from the opamp. Is this too much for lt1468? Typical output current claimed to be 22mA, but, as I understand, higher output current slows opamp down.

I'll make the circuit on a perfboard and report results here.

If you need more current, there might be a solution. It is clear that large Darlingtons are slow and smaller ones are not carefree in parallel operation. But you might consider the LM195/395, whose integral protection makes them safe to operate in parallel arrangements.

[mvs]

As of why the CV opamp needs to be fast, it needs not to introduce much phase lag as it is driven by CC opamps. So, CV opamps needs to be much fast than CC opamps.

Output of your CV opamp swings 1:1 with output of your supply. You may look not only for a gain bandwidth product, but also for a slew rate.
22V/µs is not that fast for 90MHz opamp.

TL082 with GBP of 4MHz has for instance 13V/µs slew rate and this is one of the reasons, why it is quite popular in lab power supplies.

[exe]

As of why the CV opamp needs to be fast, it needs not to introduce much phase lag as it is driven by CC opamps. So, CV opamps needs to be much fast than CC opamps.

Output of your CV opamp swings 1:1 with output of your supply. You may look not only for a gain bandwidth product, but also for a slew rate.
22V/µs is not that fast for 90MHz opamp.

TL082 with GBP of 4MHz has for instance 13V/µs slew rate and this is one of the reasons, why it is quite popular in lab power supplies.

For me bandwidth is a small-signal parameter, and slew rate is a large-signal parameter. I'm fairly sure I need bandwidth for my application.

[mvs]

For me bandwidth is a small-signal parameter, and slew rate is a large-signal parameter. I'm fairly sure I need bandwidth for my application.

Each theoretical model has its assumptions and limitations.
You have a LT-Spice simulation, so it should be quite easy to check the actual slew rate and evaluate how close it gets to max. values from the datasheet.

[exe]

You have a LT-Spice simulation, so it should be quite easy to check the actual slew rate and evaluate how close it gets to max. values from the datasheet.

Actually, I evaluated slew rate of tl071 and tl051, check this thread: https://www.eevblog.com/forum/projects/adding-offset-to-comparator-lm311-doesnt-work-the-way-i-expect/msg3365146/#msg3365146 . The results make me thinking that choosing the best opamp requires a lot of practical knowledge, some of it is, probably, not in datasheets. Like, slew rate vs input overdrive. Or output impedance vs load vs frequency.

[exe]

consider the LM195/395, whose integral protection makes them safe to operate in parallel arrangements.

I think they are just darlingtons with ballast resistor bult-in (+some protection circuitry). Do you think they will bring any advantage over a discrete version? (except for less board space). They also have 2V saturation, which is... kinda expected for three transistor, but a lot for my application.

[magic]

Actually, I evaluated slew rate of tl071 and tl051, check this thread: https://www.eevblog.com/forum/projects/adding-offset-to-comparator-lm311-doesnt-work-the-way-i-expect/msg3365146/#msg3365146 . The results make me thinking that choosing the best opamp requires a lot of practical knowledge, some of it is, probably, not in datasheets. Like, slew rate vs input overdrive.

Slew rate is input stage output current times compensation capacitance (if there are no secondary limitations). The latter is given as 18pF in the datasheet, the former is input voltage times effective input transconductance, which needs to be guessed/calculated.

At unity gain, peak input voltage corresponds to peak output slewing and both amplitudes are equal. So, 100mV input overdrive is expected to produce 0.1V·2π·3MHz = 1.9V/µs. Not far from what you observed.

This is far below maximum slew rate, so input stage is still mostly balanced and transconductance is close to the quiescent value so quiescent transconductance can be derived.

Closer to hard slewing, transconductance will fall, so this doesn't scale perfectly linearly.

You can take the ratio of SR to GBW as a figure of merit for overdrive tolerance.

Also,
http://d3i5bpxkxvwmz.cloudfront.net/articles/2011/06/16/predicting-opamp-slew-rates-1308237089.pdf

Or output impedance vs load vs frequency.

Anything with emitter follower outputs ought to be tens of ohms maximum open loop, and divided by feedback ratio in closed loop.

edit
On second thought, perhaps it's not so certain because impedance driving the emitter follower may not be low enough to hold it stiff. But I remember seeing that sort of numbers in some datasheets which bother to specify it. For instance, OP07 says 60Ω (no data about frequency) and OPA627 has a plot showing 55Ω from 2Hz to 20MHz.

[David Hess]

Slew rate is usually limited by other factors like external frequency compensation or recovery from saturation, so changing the circuit has more benefit than using a faster operational amplifier.  It is a pity that precision differential pairs are only available in a practical sense as part of operational amplifiers because they are very useful for high performance regulators.

[exe]

changing the circuit has more benefit than using a faster operational amplifier

I like the current architecture because CV opamp is always active and I don't need to think about transition between CC and CV mode. At least I think so, time will tell.

I started building the prototype. I deliberatly used long unoptimized connections on the board to check if it's still stable. Guess what, it oscillates at ~10MHz 100mV p-p . Waveform is sawtooth-like. I already destroyed two leds this way that I powered which I tried to drive by voltage. For an led +100mV means substantial growth in current, so leds appear to work normally (presumably because of fast oscillation), but they die soon. So, I guess I need to double-check the circuit. If there is no obvious problems Z(like, it pick up the noise from my noisy power supply), then I need to slow it down.

[Zero999]

It is a pity that precision differential pairs are only available in a practical sense as part of operational amplifiers because they are very useful for high performance regulators.

Are there op-amps with sink only outputs, or at least a much weaker pull-up, than sink, which work up to 30V? I had a quick look and couldn't find any. There are comparators of course, such as the LM393, but they're uncompensated and difficult to stabilise. It would be nice to be able to connect the outputs together in OR configuration, without diodes.

[magic]

NE5534

I could swear I have seen some primitive CMOS R2R amps too where the high side is just a CCS.

[David Hess]

consider the LM195/395, whose integral protection makes them safe to operate in parallel arrangements.

I think they are just darlingtons with ballast resistor bult-in (+some protection circuitry). Do you think they will bring any advantage over a discrete version? (except for less board space). They also have 2V saturation, which is... kinda expected for three transistor, but a lot for my application.

That is not far from the truth, but they have advantages over Darlingtons besides protection.  National released an application note discussing the LM195/395.

Input current is a relatively constant 3 microamps and they are fast.

[David Hess]

It is a pity that precision differential pairs are only available in a practical sense as part of operational amplifiers because they are very useful for high performance regulators.

Are there op-amps with sink only outputs, or at least a much weaker pull-up, than sink, which work up to 30V? I had a quick look and couldn't find any. There are comparators of course, such as the LM393, but they're uncompensated and difficult to stabilise. It would be nice to be able to connect the outputs together in OR configuration, without diodes.

An operational amplifier with a transconductance output can be used that way by adding a diode in series with the output, which is how IC regulators with voltage and current control loops do it.  Unfortunately there are no good choices for operational transconductance amplifiers either unless the 723 is acceptable.  It might be possible to use the compensation pin of some operational amplifiers as a transconductance output but I have not tried it, and that still leaves very few options.

I have not tried this in a power supply, but either supply connection of an operational amplifier can be used as a current output by connecting the normal output to a low impedance load.

[exe]

I've built a prototype. Many mistakes were made, one of them is I bought lt1468-2 (decompensated opamp) instead of lt1468. It oscillates wildly, even with lots of feedback capacitance. Can it be because it's compensated? Like, if it's not unity-gain stable, then I cannot use it my power supply no matter how much feedback it gives.

So for now I put tl051, and did a few experiments. Here are two screenshots, showing load transient ~10mA -- 1A and back. First screenshot shows decoupling only 100n, 5cm or so from the load. The second one shows 1u film capacitor as close to the load as possible. The load is 1Ohm resistor and a mosfet driven by AD2.

[exe]

Wow, I found the reason why it oscillates even with lt1468 (which has just arrived). At least one of them. I need to decouple bjt push-pull pair with a ceramic cap right near the bjts. Just 0.1u did the job. Otherwise it was oscillating at ~2.4MHz when sourcing more than 0.5A current. So, decoupling is not a myth .

This makes me thinking that I'd probably want to build a slower version of this power supply. Like two fast channels, and two like 10x slower bandwidth.