how to design fast bench supply with CC and CV?

[udok]

In practice dominant pole compensation is good enough, probably there is not much need to improve. 
Its the task of exe to make his requirements clear, he has not defined a test setup yet.  Without more information this is going in circles.

[Zero999]

Attaching the schematic of that anti-soar clamp that I advertised earlier. Hope posting it here falls under fair use.

Here's my shot at the fast current limit.

Thanks a lot for your effort and time, learning this design. Kleinstein's design will be next .

It's not finished. As drawn it will oscillate with a capacitive load. The main weakness it has is the voltage control loop pulls down the current limit, so U2 is inside U1's loop, which isn't ideal. For completeness here's the the design again, with a frequency compensation capacitor on the voltage amplifier and a pre-amplifier for the reference. It will now be more stable for capacitive loads, but you'll need to build it to find out. The op-amp models I've used are fairly slow, with similar speed to the OP07. You can try faster op-amps, but it will be more difficult to stabilise.


Your idea of pulling the output up with a current source and using the op-amps to pull it down should be more stable, because only one amplifier is in the loop. There's no reason why my design can't be modified to use the same technique.

Note how my design senses the current, without having to use a current mirror to move the reference voltage to the output? The current limiting is achieved using a Howland current pump. The downside is it needs closely matched resistors to achieve a high output impedance, but that's probably not a priority here.

Here's a TI application note which covers the circuit in great detail.

[exe]

Here's a TI application note which covers the circuit in great detail.

Sorry, what note? Thanks for the update!

The simple dominant pole compensation with -6dB/octave gives an inductive output impedance. So with an external or internal low ESR cap, one has a resonant circuit and thus quite some ringing.
One may have the 100dB or similar loop gain not at 1 Hz but only well below that.  Still it gets better with a steeper part in between and than a little less than the -6dB for the higher frequencies / cross over region so that the output does not behave like an ideal inductor.

You guys give really useful insights on compensation, thanks a lot! Hopefully I'll able to use this information.



PS I realized how much time I've spent already on this task. It's starts getting frustrating, so I'll relax requirements quite a bit:
1) voltage-controlled CC and CV modes (on the same ground), so I can use a two-channel DAC to do this (I can use a four-channel dac too if this helps)
2) precise current limit, that is: all current to the load goes through the shunt, no parasitic or quiescent current that is unaccounted
3) opamps don't go into hard saturation. This will ensure fast switching. How fast? Doesn't really matter, it's more about flexing brain muscules and feel good.


Ideally I want to use the same design for high-current (3A max), and low current (100-200mA) of power supply. The only different I expect is pass element (darlington output stage for high-current version), and different compensation. I believe low-current version will be inherently fast-enough for all my practical needs.

[not1xor1]

I think the most important features are:
-1) fast CV->CC switch with as little over-current spike as possible
-2) CC-CV switch with no over-voltage spike

In various different circuits I've simulated in past I've got the feeling that the worst overload recovery (i.e. highest relative value voltage spikes) occur when output voltage is set at a few volts (1-2V and below).
Besides that anti-windup diodes (like in those blackdog circuits) do affect load regulation due to leakages (at least in simulation).
For instance simulate a 5-95% load variation and check output voltage with and without the windup diodes.

Oh, wow, I see you are speaking from experience (at least with simulators) . I had issues with diodes too, esp. with LEDs. Definitely rf schottky worked better in the simulator. I'm so happy that I have no shortage of high-performance parts (except some good parts from the past that extinct). It's not like 20 years ago I was limited what was in my local store.

sorry for my late reply, due to the COVID-19 quarantine I cannot hire anybody to take care of my large garden and so I'm just too busy at the moment   
I found that old floating supply (Harrison design) simulation of mine and realized that those poor load regulation problems caused by the anti-windup diodes disappear as soon as the CC/CV OR diodes are replaced by 3 diodes (3 x 1N4148 - in simulation) in series. So I guess that blackdog's circuit is unlikely to be affected by that problem since LED diodes (with higher Vf) are used instead of 1N4148.

In the attachment a working simulation with LT1056, 2N3055 and LED diodes (with 2SC5200 it is much better)

[Zero999]

Here's a TI application note which covers the circuit in great detail.

Sorry, what note? Thanks for the update!

Sorry, I thought I'd pasted the link into my previous post, obviously not. Here it is:
http://www.ti.com/lit/an/snoa474a/snoa474a.pdf

[exe]

those poor load regulation problems caused by the anti-windup diodes disappear as soon as the CC/CV OR diodes are replaced by 3 diodes (3 x 1N4148 - in simulation) in series. So I guess that blackdog's circuit is unlikely to be affected by that problem since LED diodes (with higher Vf) are used instead of 1N4148.

This is my experience as well. Only after re-designing it from scratch I realized how much effort it was put into the circuit. All those resistors and diodes are carefully arranged and have particular values ensuring good performance and smooth operation. I have the analog part built and quickly tested it, it worked really well, almost as good as lt3080 in terms of transient response (tested at 1A, with 2sta1943 as a pass transistor), which is considered a fast regulator.

Good luck with the garden!

[David Hess]

Chapter 2 of "Analog Circuits - World Class Designs", written by Phil Perkins and edited by Robert Pease, describes the -3dB per octave compensation method.

Ok, i did not get it that you only take the AC feedback from the point before the shunt resistor.  This should work, but you have to design for the worst case capacitor (largest one). 
If the largest cap is 100 mF and Rout is 0.1 Ohm, you pole is at 16 Hz.   This is not promising.

The lead resistance and capacitor ESR add phase lead and if the capacitance is large enough, then dominant pole compensation applies so it is really not too difficult.  Well designed power supplies have no trouble with stability with practical large capacitance loads.  It gets trickier with power supplies designed for fast response.

In practice dominant pole compensation is good enough, probably there is not much need to improve. 
Its the task of exe to make his requirements clear, he has not defined a test setup yet.  Without more information this is going in circles.

A fast design is not going to use dominant mode compensation though.  Most "slow" designs do not either and rely on the output capacitor's ESR for phase lead.

[Zero999]

I think the most important features are:
-1) fast CV->CC switch with as little over-current spike as possible
-2) CC-CV switch with no over-voltage spike

In various different circuits I've simulated in past I've got the feeling that the worst overload recovery (i.e. highest relative value voltage spikes) occur when output voltage is set at a few volts (1-2V and below).
Besides that anti-windup diodes (like in those blackdog circuits) do affect load regulation due to leakages (at least in simulation).
For instance simulate a 5-95% load variation and check output voltage with and without the windup diodes.

Oh, wow, I see you are speaking from experience (at least with simulators) . I had issues with diodes too, esp. with LEDs. Definitely rf schottky worked better in the simulator. I'm so happy that I have no shortage of high-performance parts (except some good parts from the past that extinct). It's not like 20 years ago I was limited what was in my local store.

sorry for my late reply, due to the COVID-19 quarantine I cannot hire anybody to take care of my large garden and so I'm just too busy at the moment   
I found that old floating supply (Harrison design) simulation of mine and realized that those poor load regulation problems caused by the anti-windup diodes disappear as soon as the CC/CV OR diodes are replaced by 3 diodes (3 x 1N4148 - in simulation) in series. So I guess that blackdog's circuit is unlikely to be affected by that problem since LED diodes (with higher Vf) are used instead of 1N4148.

In the attachment a working simulation with LT1056, 2N3055 and LED diodes (with 2SC5200 it is much better)

That schematic made my head hurt. Labels are good, but try to use wires as well, whenever practical, otherwise it means one has to keep searching for them. Try to avoid overlapping text with the symbols. It makes it hard to read. I've rearranged the schematic a bit to make it clearer. I know some of the changes are a bit petty and just personal preference and you and others might not like them, but everyone's different.

 transient_5-95perc-LT1056 + 2N3055 mod1.asc (11.01 kB - downloaded 85 times.)

There were some models which aren't included in the standard LTSpice install, which I had to Google and import. Hopefully I've chosen the same ones as you have on your machine. I'd be nice if LTSpice included a feature to tell you which models aren't included by default and a way of importing symbols would be nice, so people don't have to use zip files.

You have some good ideas there: a current source to provide a minimum load and LEDs doubling as ORing diodes, so the user can tell whether it's in CV or CC mode.

The issue with this design is the fast current limit is defeated by C4 and C2, which will cause cause current surges, when short circuited and could easily toast an LED. Unfortunately removing them isn't an option as it oscillates, without them.

Which three 1N4148 diodes are you talking about in the simulation? Unless they're subject to high temperatures or reverse voltages, the 1N4148 has a fairly low leakage current and shouldn't affect the regulation that much, unless very high value resistors are used. The data sheet specifies a maximum leakage of 25nA, at 25^oC and V_R = 20V, which is a voltage drop of just 25mV across a 1M resistor.
https://www.vishay.com/docs/81857/1n4148.pdf

EDIT:
Note, that in the .asc file attached to this post, the load current pulse is set to 100A, to test the overcurrent protection circuit. The result is posted  here:
https://www.eevblog.com/forum/projects/how-to-design-fast-bench-supply-with-cc-and-cv/msg3019102/#msg3019102

[xavier60]

C1 and R7 do not provide any compensation feedback for U2. The CV response must be very fast and possibly borderline unstable.

[Zero999]

C1 and R7 do not provide any compensation feedback for U2. The CV response must be very fast and possibly borderline unstable.

You're right and D10 will also short circuit U2, causing it to current limit. I can't believe I spent so long looking at the circuit and rearranging it, without noticing. 

[exe]

Which three 1N4148 diodes are you talking about in the simulation?

It's where you have LEDs.

[Zero999]

Which three 1N4148 diodes are you talking about in the simulation?

It's where you have LEDs.

But there are only two LEDs and he mentioned three diodes.

[exe]

Which three 1N4148 diodes are you talking about in the simulation?

It's where you have LEDs.

But there are only two LEDs and he mentioned three diodes.

Afaik he meant three diodes in series to create a needed voltage drop. Or this can be replaced with one LED, that creates voltage drop of 1.8V+.

[not1xor1]

That schematic made my head hurt. Labels are good, but try to use wires as well, whenever practical, otherwise it means one has to keep searching for them. Try to avoid overlapping text with the symbols. It makes it hard to read. I've rearranged the schematic a bit to make it clearer. I know some of the changes are a bit petty and just personal preference and you and others might not like them, but everyone's different.
(Attachment Link)
(Attachment Link)

There were some models which aren't included in the standard LTSpice install, which I had to Google and import. Hopefully I've chosen the same ones as you have on your machine. I'd be nice if LTSpice included a feature to tell you which models aren't included by default and a way of importing symbols would be nice, so people don't have to use zip files.

You have some good ideas there: a current source to provide a minimum load and LEDs doubling as ORing diodes, so the user can tell whether it's in CV or CC mode.

The issue with this design is the fast current limit is defeated by C4 and C2, which will cause cause current surges, when short circuited and could easily toast an LED. Unfortunately removing them isn't an option as it oscillates, without them.

Which three 1N4148 diodes are you talking about in the simulation? Unless they're subject to high temperatures or reverse voltages, the 1N4148 has a fairly low leakage current and shouldn't affect the regulation that much, unless very high value resistors are used. The data sheet specifies a maximum leakage of 25nA, at 25^oC and V^R = 20V, which is a voltage drop of just 25mV across a 1M resistor.
https://www.vishay.com/docs/81857/1n4148.pdf

I apologize, but when I wrote yesterday (and now too) I was physically exhausted so I just modified an old test circuit and tried to remove all non standard models but as you noticed I forgot some of them.

The original circuit diagram used other components. I just modified a bit the compensation network to avoid oscillations.
LEDs as ORing diodes are taken from Blackdog's circuit. I had noticed that with just 1N4148 as OR diodes, when diodes were used in the opamp feedback to prevent windup, load regulation failed misearbly, but the problem was solved with LEDs or 3 1N4148 diodes as OR devices.

Regarding current surge when using a LED as load, I think one should consider the energy stored in the capacitor. Let's make it 50µF, at 30V of output voltage there is a (50*30*30)/2e6 = 45000/2e6 = 22.5mJ.
I suspect that would not damage a LED.

[not1xor1]

C1 and R7 do not provide any compensation feedback for U2. The CV response must be very fast and possibly borderline unstable.

You're right and D10 will also short circuit U2, causing it to current limit. I can't believe I spent so long looking at the circuit and rearranging it, without noticing. 

You probably missed that the opamp supply is floating and positive output is ground. That is a simplified version of Blackdog's circuit.

[Zero999]

C1 and R7 do not provide any compensation feedback for U2. The CV response must be very fast and possibly borderline unstable.

You're right and D10 will also short circuit U2, causing it to current limit. I can't believe I spent so long looking at the circuit and rearranging it, without noticing. 

You probably missed that the opamp supply is floating and positive output is ground. That is a simplified version of Blackdog's circuit.

No, I get that. The problem is, when the current limit kicks in, U2's output tries to go high, but is short circuited to 0V, via D10. In the .asc file attached previous post, I set the load current to 100A, just to see what would happen. I should have posted the result and made it clear that I had done that. Here's the plot of the current through D10, whilst it's in CC mode.


Regarding xavier60's remark about C1 and R7: he's right they can't provide any compensation or feedback, because the inverting input is just connected straight to 0V. To get negative feedback, the output needs to be able to affect the voltage on the non-inverting input.

Which three 1N4148 diodes are you talking about in the simulation?

It's where you have LEDs.

But there are only two LEDs and he mentioned three diodes.

Afaik he meant three diodes in series to create a needed voltage drop. Or this can be replaced with one LED, that creates voltage drop of 1.8V+.

How does having a higher forward voltage help with the transient response? I can't see how it makes any difference and would have thought, the lower the forward drop the better. Silicon diodes should be better than LEDs.

[xavier60]

After the U2"s compensation feedback is fix with a resistor between the Inverting input and ground, it should be tested with C1 connected from the ORing node. It will allow U2 to operate in open loop while the PSU is transitioning from CC back to CV witch should be fast with that op-amp.
I feel that C1 should be something larger for now.

[Zero999]

Regarding current surge when using a LED as load, I think one should consider the energy stored in the capacitor. Let's make it 50µF, at 30V of output voltage there is a (50*30*30)/2e6 = 45000/2e6 = 22.5mJ.
I suspect that would not damage a LED.

t depends on the LED. I think a 5mm LED might go pop, if connected to a 50µF capacitor, charged to 30V. The only way of knowing is to test a wide range of LEDs.

A capacitor only helps with a constant voltage source. If you want the power supply to be a good current source, then the output capacitance should be zero. In fact an inductor should be used, rather than a capacitor, for a current source.

Unfortunately in this case we can't have both. A relay is far too slow to switch the output between an inductor and capacitor. A simple workaround would be a manual switch to select between optimal current or voltage regulation.

[xavier60]

About C7 and R5. They help prevent overshoot after the PSU's output has been held low for some time but don't help at all when recovering from a  brief overload that doesn't cause a large drop in output voltage. I eventually omitted these parts from my design. The mod I mentioned earlier greatly reduces overshoot.

[xavier60]

Windup problems can be minimized by not having the Vcc much higher than needed.

[exe]

Windup problems can be minimized by not having the Vcc much higher than needed.

Interesting thought, came to me yesterday too . Blackdog's version has a pre-regulator, so we can supply opamps from it. It maintains voltage ~2-3V above the current output (at least in the circuit I've built). So, actually, can be a solution.

A second approach I came with is to modify blackdog's circuit to eliminate sources of parasitic current. That is:
1) set voltage not with a voltage divider (not sure if "divider" is a correct term, more like a summing node?), but with a reference. Yes, this adds complexity, but it's gonna be complicated anyway.
2) buffer voltage sense pin, this way when CV opamp is clamped, it cannot pump current into the load. To avoid performance loss due to additional phase shift on the buffer, I'm willing to try, uhm, bypassing buffer with a small cap. I think this way we'll get two feedback paths: 1) for fast AC 2) slow for DC servo. This is inspired by earlier comments in the thread.

I'll start designing the circuit tonight.

As of why not using big resistors to minimize all those parasitic currents. I think they 1) add noise 2) add offset voltages because of opamp bias current 3) may introduce a pole because with parasitic capacitance they form an RC-filter. So, I'd like to limit resistor values to 20k or below. That's why I learned from "Art of Electronics", hope this is not complete nonsense. I'm yet to do noise simulation in LTSpice, may be I overestimate the noise from resistors.

[Kleinstein]

C7 and R5 not only help to avoid overshoot. They can also help with the general compensation with larger capacitive load. One may want to change the values though (e.g. like 10 time smaller values). It is more like that C6 and R27 have marginal effect and could be removed.

A reduced (not much higher than needed) supply to the OPs can be the simplest from to limit windup. It works kind of similar to the diodes by setting an upper limit to the OPs output.

[not1xor1]

OK I spotted a few mistakes. There was a missing resistor on the inverting input of the voltage regulation opamp. I also had to add a 1N4148 in series with the LEDs to avoid load regulation problems and slightly changed the compensation network.
I also rearranged a bit the schematic.



I've not yet tested current regulation, loop stability, etc... I'll try to do that tomorrow.

I'll try to read the other messages (and reply) later.

[Zero999]

OK I spotted a few mistakes. There was a missing resistor on the inverting input of the voltage regulation opamp. I also had to add a 1N4148 in series with the LEDs to avoid load regulation problems and slightly changed the compensation network.
I also rearranged a bit the schematic.

I've not yet tested current regulation, loop stability, etc... I'll try to do that tomorrow.

I'll try to read the other messages (and reply) later.

You forgot to put the model for the 1N4002 in the .asc file.

Just one comment about the presentation: please put the trailing zeros or use R notation for component values <1, i.e. 0.22 or 0R22, rather than just .22. It's very easy to miss the fact you've got 0.22, especially with the grid setting enabled, it looks like 22, at a quick glance.

Are you using the default font settings? When I load that file, lots of the text overlaps with the lines. Is it just because I'm running it under WINE? Your file looks like this on my system. The only thing I've changed on my install is the resistor symbol to the European one because I prefer it to the American one, but the fonts are default.

[not1xor1]

Are you using the default font settings? When I load that file, lots of the text overlaps with the lines. Is it just because I'm running it under WINE? Your file looks like this on my system. The only thing I've changed on my install is the resistor symbol to the European one because I prefer it to the American one, but the fonts are default.
(Attachment Link)

I'm sorry, my LTSpice installation (WINE running on kubuntu 18.04) is heavily customized and unfortunately I forgot that.