how to design fast bench supply with CC and CV?

[exe]

Some updates on the project: this week I spent quite some time on the project. It seemed close to be done, but I stunk with an unexpected problem: current readback.

To my surprise, most current sense amplifiers have one or more of those unwanted properties:
1) Cannot measure when voltage close to ground (common-mode voltage does not include ground)
2) imprecise (CMRR ~80db)
3) do not work high common-mode voltage (15V in my case)
4) exhibit increased error when measuring close to the ground
5) have high input current
6) confusing datasheet

My workaround is I'll use an instrumentation opamp (such as ad8422). Their CMRR is not that great, so I'll use two current ranges to mitigate that. I wanted to have two current ranges anyway.

Overall, I think for best precision the current-sensing should on the low side of the measurement circuit, otherwise there is a huge common-mode voltage that is hard to deal with if precision is needed.

Where would you point someone who's interested in designing op-amp circuits and feedback loops like whats in this post?
I'm not an absolute beginner but I don't have much experience with it  .

I'd start here: https://www.allaboutcircuits.com/technical-articles/negative-feedback-part-1-general-structure-and-essential-concepts/ . I think it's one of the best introductions into the subject.  I'll post more links once I found them. The topic is quite hard to grasp, expect to spend some time. I also suggest not just read it, but practice it in a spice simulator.

[xavier60]

Overall, I think for best precision the current-sensing should on the low side of the measurement circuit, otherwise there is a huge common-mode voltage that is hard to deal with if precision is needed.

Problem then is the CS will measure the current drawn by the control and measurement circuitry unless an independent control rail is provided.
A work around I used is  powering the LED panel meter with a shunt regulator to prevent fluctuating CS current.

[David Hess]

Overall, I think for best precision the current-sensing should on the low side of the measurement circuit, otherwise there is a huge common-mode voltage that is hard to deal with if precision is needed.

Instead of using an instrumentation amplifier, move the current limit amplifier to the high side and have it follow the output.  Now the common mode rejection is only limited by the amplifier itself.

The above means level shifting the control signal to the current limit amplifier's common mode level however this is a low frequency signal so this is easily accomplished with high precision.

[exe]

Quick update: I'm finalizing the schematic and revising for things that can be done better. Particularly layout. I want to sense on the terminals, which is in my case meaning that the "-" output should be a star ground. But I'm worried that it will be too many wires: reference ground, analog ground, digital ground, power ground... I can, say, simplify a little bit and separate only power ground and control ground, but that might compromise precision and noise.

Instead of doing a star ground on the "-" terminal, I consider using an instrumentation amplifier for voltage sensing. What do you think? My current candidate is AD8421.

[David Hess]

Instead of doing a star ground on the "-" terminal, I consider using an instrumentation amplifier for voltage sensing. What do you think? My current candidate is AD8421.

That is feasible but it complicates the frequency response and precision.

[exe]

That is feasible but it complicates the frequency response and precision.

Is there anything specific to in-amps? I understand why adding an opamp buffer makes it harder: it adds offset voltage, phase lag and may limit slew rate. For in-amps, in addition to that,  I only know their CMRR is about 80-90db, and gain accuracy can be a problem. Is there anything else I should know about them?

[Neomys Sapiens]

I found AD's AN-539 quite enlightening about them.

[David Hess]

That is feasible but it complicates the frequency response and precision.

Is there anything specific to in-amps? I understand why adding an opamp buffer makes it harder: it adds offset voltage, phase lag and may limit slew rate. For in-amps, in addition to that,  I only know their CMRR is about 80-90db, and gain accuracy can be a problem. Is there anything else I should know about them?

No, you covered it; the added phase lag requires extra attention to frequency compensation and DC errors affect the accuracy.  Obviously it is feasible to do because it is a common way to implement Kelvin connections or remote sense, but often there is a way to implement this without a separate amplifier.

The fastest operational amplifiers which would add a minimum of phase lag tend to have poorer DC performance so that is no solution except where lower precision is acceptable.  Fortunately however in many regulators, the output stage is what limits frequency response so cascading a couple of amplifiers is not a problem, but this is application dependent.

[xavier60]

Everyone would agree that the HP/Harrison topology allows the best performance because all of the important bits are ground  referenced to one point, the positive output terminal. But having to ground test equipment to the positive rail makes testing and debugging awkward.

I wonder if it would all be easier with the pass transistor in the negative rail allowing the negative output to be the ground reference point.
Up until a year ago, my only bench supply was something I built 40 years ago. Its 2N3055 was in the negative rail with its Collector bolted directly to the aluminum front panel. It works just fine.

[Kleinstein]

One can mirror the HP/Harrison topology, just use  PNP or P-MOS power devices. These are however often a little slower or high input capacitance. So the regulator floating on the positive side is a little more attractive.

[David Hess]

Everyone would agree that the HP/Harrison topology allows the best performance because all of the important bits are ground  referenced to one point, the positive output terminal. But having to ground test equipment to the positive rail makes testing and debugging awkward.

I wonder if it would all be easier with the pass transistor in the negative rail allowing the negative output to be the ground reference point.
Up until a year ago, my only bench supply was something I built 40 years ago. Its 2N3055 was in the negative rail with its Collector bolted directly to the aluminum front panel. It works just fine.

Sometimes that is done so that the uninsulated collector/source can be bolted directly to the heat sink.  Or maybe only NPN/N-Channel transistors are desired because of their higher performance and lower cost.  In the later case, you can find old designs were the NPN transistor is used on the high side as an emitter follower, and then its output is grounded to make a negative supply.

[exe]

One can mirror the HP/Harrison topology, just use  PNP or P-MOS power devices. These are however often a little slower or high input capacitance. So the regulator floating on the positive side is a little more attractive.

I wanted to do this way, but this means that the pass element on the ground. What if output is disabled and the pass element not conducting? Doesn't this mean there is no return path for current anymore? Or it depends if aux supply is floating or not?

I also wanted to "invert" the control signal so that opamps see the grounded load, but afaik it's less stable ("ldo topology") because output stage has gain (on the pic).

[xavier60]

I also wanted to "invert" the control signal so that opamps see the grounded load, but afaik it's less stable ("ldo topology") because output stage has gain (on the pic).

This kind of level shifter?, https://www.eevblog.com/forum/beginners/lm324-power-supply-with-variable-voltage-and-current/?action=dlattach;attach=1000379;image
It can be made to work very well.

[xavier60]

To avoid unwanted current flow through the CS resistor, two control rails are needed. One can be regulated from the main unreg  in.
The other has to be floating.

[namster]

Hi ,
I have found this article that describe the schematic of Error Amplifier with Forced Equilibrium Adaptor , it give a very good switch from CV to CC maybe it can help https://www.kepcopower.com/equibm2.htm , the schematic is given in figure 3

[exe]

I almost finished the design, but then I decided to add a current sinking capability*, which makes things more complicated... Meanwhile I found a few good opamps with good slew rate.

What if instead of fighting the problem I just take an opamp with fast recovery? Like opa189 which I wanted to use anyway. Usually people say auto-zero opamps go nuts when saturated, but this one seems to be an exception. This will greatly simplify the circuit and will let me use a ground shunt. Although, even with a ground shunt I think it still will be a challenge to route the pcb.

The topology i want to use is attached. It's Keithley 236/237 (pic is taken from the blog linked below).

* After re-reading https://poormanssmu.wordpress.com/research/ , which makes much more sense to me now.

UP:

Hi ,
I have found this article that describe the schematic of Error Amplifier with Forced Equilibrium Adaptor , it give a very good switch from CV to CC maybe it can help https://www.kepcopower.com/equibm2.htm , the schematic is given in figure 3

Oh, just saw your post. Thanks, I'll read it.

[bikeNomad]

How about something like the HP 6826a or 6827a? As I recall, they have a fast (~100µs) switchover between CV and CC modes, and have 4-quadrant outputs.

[exe]

How about something like the HP 6826a or 6827a? As I recall, they have a fast (~100µs) switchover between CV and CC modes, and have 4-quadrant outputs.

Thanks, I'll have a look. Never heard of this models before.

100us is I'd say quite slow, with slew rate of even a few volts per us, the power supply can dump a lot of energy into DUT before switching before CV/CC. I did a simulation with OPA189 without anti-windup diodes, for example. It showed a 2V overshoot when switching from CC to CV mode in less than 3us. So, a lot can happen during these 100us.

Although I start to question how fast a power supply should really be. Would this ultra-fast switching really help when testing sensitive semiconductors?

[bikeNomad]

I just tested my (newly refurbished) HP 6826A going from 4V constant voltage to 20mA constant current (using a 100Ω resistor) and found that it took 400µs to switch to CC mode.

[Kleinstein]

How about something like the HP 6826a or 6827a? As I recall, they have a fast (~100µs) switchover between CV and CC modes, and have 4-quadrant outputs.

Thanks, I'll have a look. Never heard of this models before.

100us is I'd say quite slow, with slew rate of even a few volts per us, the power supply can dump a lot of energy into DUT before switching before CV/CC. I did a simulation with OPA189 without anti-windup diodes, for example. It showed a 2V overshoot when switching from CC to CV mode in less than 3us. So, a lot can happen during these 100us.

Although I start to question how fast a power supply should really be. Would this ultra-fast switching really help when testing sensitive semiconductors?

A fast transition from CC to CV mode limits the overshoot. In designs with a slow switch over one may need additional capacitance at the output to limit the overshoot.  Relaying on the capacitor to limit the rise may need quite some extra capacitance. With 1 A into some 100 µF (usually plenty for small signal stability) and some 0.5 V accepted overshoot is would be some 50 µs to react. The other way around some 400 µs reaction time would require substantial capacitance to keep the overshoot in the voltage low.

The CV to CC more transition may not need to be so fast, as there always is the charge from the output capacitor. A slow reaction kind of looks like additional simulated capacitance. It can make a difference if there is an additional fast (but usually not accurate) current limit so to avoid a large current spike in case of a short. Especially a source follower output stage can be really bad in this respect. For many uses one can accept and may even want not too fast a reaction, so the short current spikes don't let the supply drop even though the average current is still well below the limit.

For really sensitive parts the actual capacitance at the output alone can be deadly. It would be still nice to protect things like To92 transistors or SOIC8 chips from a thermal damage.

[exe]

May it's worth posting a small update on the project. The project is not dead, but I took a break. So I investigated how much I need an active rectifier for the power supply. I think I can get away with this approach: https://github.com/kopchik/semi_active_rectifier . Basically, I replaced normal diodes with FERD diodes, and bottom diodes are "shunted" with n-fets.

I have pcb already, waiting for Mouser to deliver missing parts and I'll do evaluation. I have a nice DIY power supply which is currently limited to ~0.7A max output due to the diode bridge not mounted on a heat sink (yeah, stupid thermal design, I know). The surrounding plastic (PETG) cannot handle more than 70C. After upgrade I expect it to handle about 1.5A or so.

[exe]

I didn't really make much progress, but I wanted to share some thoughts on power supplies.

I wanted to solder something over the weekend, so I chose that national's lab supply (https://www.ti.com/lit/an/snoa692/snoa692.pdf). But I think I'll solder something else, here's why.
1. The output is made of 10xLM395, they have 2mA+ quiscent current. All this goes through the shunt. Perhaps, it's not a problem for a 10A power supply, but I wanted some precision.
2. lm101a is hard to get. I wonder if I could replace it with lm308 . Nevermind, I think I forgot to look for lm301a.
3. lm395 are sorta expensive (~3.5euro).  Probably could be replaced with a low-power darlington (mpsa14) driving a beefy bjt. This way base current could be down to microamps, similar to what lm395 offer. Not sure about performance of the replacement.

Overall, I feel like it's hard to beat Harrisson topology. I was hoping to find an opamp or something with clipping. And, believe or not, I found one, which is lt6372. Unfortunately, it's not an opamp, it's an instrumentation amplifier. But might be handy one day . Loosy specs on clipping voltage

One interesting circuit I found in this post: https://www.eevblog.com/forum/projects/is-550uf-too-big-for-a-power-supply-that-has-cc-limit/msg2161846/#msg2161846 . I mean the one with lt1010. I like that current-limit amplifier controls the voltage-set amplifier. This way only one opamp needs to be clipped. I can't prove yet, but it feels like it's much easier to make stable crossover betwen CC and CV modes when only one opamp goes into saturation. The circuit still suffers from typical problems, such as CC-opamp steals (or injects) some output current, and it's also high-side current sensing. I played in ltspice with the circuit, switch to CC mode within one or two microseconds look impressive, as well it is stable without output capacitor (or requires very little). I want to build one. Probably, I'll replace lt1010 with cheaper and beefier buf634a (https://www.ti.com/lit/ds/symlink/buf634a.pdf).

Any thoughts and/or opinions?

[Kleinstein]

Having the CC regulator reduce the set point to the CV regulator has some nice features.  It makes the CC to CV transition easy, but it can make the stability in the CC mode a little tricky.  The speed of the CC and CV modes are kind of couples.

There is a common Kit sold from china that uses this type of control - the main weak point with the kit is using OPs beyound there voltage specs. So it needs some mods, but than it can make a useful simple circuit.

The LM395 is mainly used for protection. AFAIR one could replace it with a LM317 with not so much change.

[Zero999]

I didn't really make much progress, but I wanted to share some thoughts on power supplies.

I wanted to solder something over the weekend, so I chose that national's lab supply (https://www.ti.com/lit/an/snoa692/snoa692.pdf). But I think I'll solder something else, here's why.
1. The output is made of 10xLM395, they have 2mA+ quiscent current. All this goes through the shunt. Perhaps, it's not a problem for a 10A power supply, but I wanted some precision.
2. lm101a is hard to get. I wonder if I could replace it with lm308 . Nevermind, I think I forgot to look for lm301a.
3. lm395 are sorta expensive (~3.5euro).  Probably could be replaced with a low-power darlington (mpsa14) driving a beefy bjt. This way base current could be down to microamps, similar to what lm395 offer. Not sure about performance of the replacement.

Overall, I feel like it's hard to beat Harrisson topology. I was hoping to find an opamp or something with clipping. And, believe or not, I found one, which is lt6372. Unfortunately, it's not an opamp, it's an instrumentation amplifier. But might be handy one day . Loosy specs on clipping voltage

One interesting circuit I found in this post: https://www.eevblog.com/forum/projects/is-550uf-too-big-for-a-power-supply-that-has-cc-limit/msg2161846/#msg2161846 . I mean the one with lt1010. I like that current-limit amplifier controls the voltage-set amplifier. This way only one opamp needs to be clipped. I can't prove yet, but it feels like it's much easier to make stable crossover betwen CC and CV modes when only one opamp goes into saturation. The circuit still suffers from typical problems, such as CC-opamp steals (or injects) some output current, and it's also high-side current sensing. I played in ltspice with the circuit, switch to CC mode within one or two microseconds look impressive, as well it is stable without output capacitor (or requires very little). I want to build one. Probably, I'll replace lt1010 with cheaper and beefier buf634a (https://www.ti.com/lit/ds/symlink/buf634a.pdf).

Any thoughts and/or opinions?

The one with the LT1010 uses a Howland topology for current limiting, like my designs. It should be possible to increase the current rating, by adding a pass transistor, reducing the value of R4 and adjusting the resistor ratio of R4, R5, R6 & R7.

[exe]

There is one thing I don't like about differential current sensing: it injects (or steals) some output current. Probably, not a big deal in 99% of cases, but I wanted to have an accurate CC mode. Also wanted to avoid resistor matching etc, although vishay MPM series would probably solve this for me. So, I ended up with circuit on "simplified.png".

Then I added a voltage reference referenced to the same node as current setting voltage. I used an instrumentation opamp. This way I can set current and voltage from the same DAC (but Vset and Iset control circuit needs to be floating, which is fine). I include the *.asc file in case someone is curious to play with the circuit.

After posting I realized that I kept shunt value of 1k. This is because I'm trying to understand if it's possible with current design to have minimal source current of 1uA. I doubt, but "it works in simulation". It's not accurate due to input bias current of CC-opamp, but probably could be calibrated out. I'm yet to think what opamps to choose, for now I left old LT118A because I don't know what would be better. I randomly tried a few fast opamps (10-60MHZ bw, slew rate > 20V/us), not all of them gave stable circuit and it's beyond my understanding why. Probably, different opamps have different small-signal response (several opamps were oscillating with a small amplitude), or may be at such speeds things like input capacitance start to play role.

Next step I want to try to make it sink current too. The lt1010 datasheet says it's "easy", I just need to add another opamp that would limit maximum sinking current. I'm too tired to try it now.

So, what do you think of current schematic? Is it yea or nay?