fast CC/CV power supply take two

[exe]

Hello my friends,

In my previous attempt https://www.eevblog.com/forum/projects/how-to-design-fast-bench-supply-with-cc-and-cv/ I didn't reach my performance goals. Namely, fast operation under wide range of current, while having minimal output capacitance. I decided that output current range (0-2A) is too much for the topology I chose. So, here is another project that aims at low currents. Desired characteristics: 0.5-12V 0-50mA, stretch goal: 0-30V, 200mA.

The circuit I draw is a bit of a mess (I blame kicad ), so I describe it here in a few words. It's a shunt voltage regulator with a constant current source. This way two loops (CC and CV) are always active. When output voltage drops below set voltage and then goes back, U5 has to slew a little bit. To reduce slewing time, there is voltage shifter V1 on schematic. I used a photovoltaic cell for that, hope it will work. I'll post later more details how the circuit operates, and why I think it should be faster than a typical mosfet current source (spoiler alert: using mosfet in saturation region for intrinsic current regulation and cascode).

I just assembled the board, and did a very quick test: output current set to 10mA, load step 840 Ohm to 5k Ohm and back. Rise time is 0.85us, clean step response, which is pleasantly good). There is no output capacitor except parasitics. The load step is generated by TS555 onboard timer.

This is my first four-layer PCB, and my first use of kicad. Mistakes were made, I screwed voltage regulator part, namely swapped opamp IN+ and IN-. I will have to think how to bodge it. Another problem, minor but annoying: in Art of electroncs 555 timer circuits don't have reset pin connected. So I left it floating. I guess you know that was a bad idea. A bigger issue: the timer has output voltage of 10V, which is probably too much for the fet that does step load. Idk which fet I used, probably FDV303N (Vgs max 8V ). So, fet may die, but hey, it's a proto board. The fet has zenner protection on the gate, which I think might explain why circuit suddenly stops working when I raise power rail beyond 10V.

Question: do I need to short my ground clip? Is it ok for raise times around 300-500ns? I use 10x probe.

Challenges:
- the biggest one is accurate current measurement as now current depends on two resistors: the one that sets current, and the one that shunts output.
- voltage doesn't go all the way down to zero, because of the drop on shunt resistors. I cannot remove shunt resistors (forgot to show them on schematic) because I need to measure current. So, in the worst case in CV-mode minimal voltage is 0.6V. In CC it goes down zero no problem (I think, need to test).

That's it for now, off I go to celebrate my first PCB in a while.

[T3sl4co1l]

Probe clip should be fine around there. Can always check both ways and see how it affects the edge.

Tim

[exe]

I'm now testing voltage regulation section. And I've got problems with it. Ignoring huge voltage spikes when switching CC/CV, the voltage sags quite a bit, and slow to fully recover. I figured I put a wrong capacitor in feedback loop: 1u instead of 1n. I fixed that, now new problems)

One is overshooting, and another one is capacitive load. Anyway, here are some scopeshots to make this thread a bit less boring. One the first one you can see two responses when stepping CC/CV from 1 to 5V, one from pfet-bjt complementary feedback pair (CFP, Q6/Q7, blue trace), driving voltage 1.7V or there about), the other one is just plain bjt (Q8, yellow trace). Overshoot is significant in both case. CFP also has significant ringing, suggesting it needs more compensation or something.

The second shot show help of photovoltaic level shifter. Unsurprisingly, a 650mV shift corresponds to 650mV less overshoot. As of why the drop is 650mV, I'm a bit puzzled. I use an IR led, the forward voltage drop last time I measured was 1V. But in circuit it's only 0.65V. I guess that's because photovoltaic cell outputs only 1.24uA of current?

PS I included schematic of the PSU.

[exe]

My friends, I think I found a flaw in current regulator. It works the best when the voltage across shunt is large. Which makes sense, I just didn't take this into account. This limits minimum current on the range. I think about 0.1V is the minimum voltage for this approach. This can be seen on the following two plots showing 1V-10V step response for 25mA load:
100ohm_25ma.png vs 5ohm_25ma.png . The later has twice lower rise time. Also notice that top of waveform is not flat. That's because the mosfet current source is not ideal, and an overcompensated opamp has to adjust voltage a little bit:




Finally, I include 100mA step response. This is about maximum I can get from sot-723 pass fet (RZM001P02T2L) without risking to burn it. Not that I didn't try to push thru it 200mA, it just started to smell bad, so I had to reduce the current). Here it is:



Next is try to get voltage regulation. Last time I tried I got nasty overshoots and oscillations.

[exe]

So, I have problems with the circuit, so I decided to first make CC mode working reliably. So, what I did is:
1) tried different pass transistors and select the one with fastest response
2) optimize opamp feedback.

It seems the smaller gate capacitance, the better performance. So, I chose tp0606n3. It requires a heatsink. Fortunately, I have one. The heatsink on TO-92 looks a bit ridiculous, but it works, lowering temperature from over 100C to about 66C under worst condition (100mA at 10V).

As of opamp feedback, it turned out I can just drive the fet directly, no oscillation so far. I include two plots for 1V-10V step response in CC mode for 100mA and 10mA loads. Imo looks fine. Idk why there is a small step, that's currently not an issue, though might be an indication of something interesting.

Currently, the biggest issue is that CV mode is slow and oscillates (screenshot attached). I'm investigating this. I really hope it's not two loops fighting, as I don't have much skills to resolve that.

[exe]

Hello my friends. It's been a while since I posted here. The project proved to be more complicated than I expected, but I'm making progress.

I think I'm (mostly) done with constant current source. At the end, the best performance I got from mosfets with minimal gata capacitance, so I just used TP0606N3 with input capacitance about 80p (value from datasheet). With that, I could just drive it from NE5532 directly, no compensation or anything is needed.

So, onto the voltage regulating section. This is where I've got the most troubles. I've got some good results with bjt as a pass transistor, but that wouldn't allow for accurate current measurement due to base current. So, I, again, replaced it with a small-signal pfet. And it worked, kinda. Look at the first plot where I do CV/CC switch at 20mA current. It looks very neat!

But then I increased the current to 100mA things got ugly. Look at the second screenshot. The blue trace shows step output without any capacitors. There is like 1V overshoot, which is, imo, too much. The yellow plot shows output with 50nF output capacitor. Better, but still not purfect.

"What's going on here?" (c) Dave. I believe different output currents make the circuit operate at different operating points. I measured voltage at the gate of voltage regulating fet, and it changes noticeably with output current. With higher output current the CV-opamp has to swing more to maintain regulation (540mV@20mA, 1V@100mA). Not sure how I'm going to solve this.

[xavier60]

You are trying to build a fast PSU. I can't remember all the details. A few years ago, someone posted an off-site link to their Harrison style PSU project. The CV compensation didn't look right, seemed to be under compensated, so I breadboarded just the CV loop to find that it was surprisingly fast and stable. 1 or 2 microseconds load transient recovery, as far as I can remember.

[xavier60]

I put the Harrison CV loop together again using a TL072, BD135 and TIP35C. I put a 33Ω in series with the Base of the BD135 to stop it from oscillating.
I took a snap of the output response of the opamp when a 1A transient load was applied. It's quick.

[exe]

You are trying to build a fast PSU. I can't remember all the details. A few years ago, someone posted an off-site link to their Harrison style PSU project. The CV compensation didn't look right, seemed to be under compensated, so I breadboarded just the CV loop to find that it was surprisingly fast and stable. 1 or 2 microseconds load transient recovery, as far as I can remember.

Ah, probably this one, or similar thread https://www.eevblog.com/forum/beginners/how-does-blackdog_s-psu-work/ . It's a good power supply indeed. There are one or two things that I tried to improve. One is current accuracy, Harrisson topology has problems with that as its floating CV opamp is an inverting amplifier, and it draws current from the output proportional to output voltage. I'm aiming of minimal output current of 1uA, so that doesn't work for me. Anyway, I tried to resolve that (say, buffering output with another opamp), but I wasn't able to design a power supply that would be fast for wide range of currents, even with switching shunts. Hence this project, a small power supply for low end of currents (say, 1uA-100mA) that would be an addition for Harrison-like for currents, say, above 10mA.

I put the Harrison CV loop together again using a TL072, BD135 and TIP35C. I put a 33Ω in series with the Base of the BD135 to stop it from oscillating.
I took a snap of the output response of the opamp when a 1A transient load was applied. It's quick.

It is indeed. The tricky part is to make CC also quick, without much glitching when switching between CC and CV modes. For CC we need to add a shunt, which will reduce phase margin when there is an output capacitor. The bigger the shunt, the faster reaction time, but slower CV mode . Shunt topology, in theory, doesn't have this problem.

Anyway, just in case, this project is more like an exploration. The only sensible objective I could think of is, when, say, a power supply drives an opamp beyond rails, it should be fast-enough not to burn ESD diodes.

[exe]

Small video from me demonstrating the importance of low output capacitance:

[xavier60]

I went back to experimenting with my old design,
https://www.eevblog.com/forum/beginners/lm324-power-supply-with-variable-voltage-and-current/msg3582664/#msg3582664
I was able to make it go much faster, mainly by removing the TIP35C's, leaving the D45H11 as the output. I fitted a 1uF with series 4.7Ω across the output.
D5 in the original design greatly speeds up the CC response to sudden output overloads. Might be a while before I add the CC loop and test.

[iMo]

Small video from me demonstrating the importance of low output capacitance:

Yep, so called "Red LED Test". I've just tried with a 15+y old HY1503 cheapo PSU, and my red led survives, no problem (18.7V/20mA set).. Btw, the output capacitor there is 470uF, in parallel with 1k resistor, afaik. The voltage on the led drops in aprox 25ms (from what I see on my scope). I will post the schematics for reference (I messed with it as it was broken, the schematics fits, the values of the pots are different in the real hw, not easy to follow the design with internally "grounded" positive output node, I had to do some sims in the LTspice to get how it works).. Almost the same as the above xavier60 has linked.. 

[xavier60]

Small video from me demonstrating the importance of low output capacitance:

Yep, so called "Red LED Test". I've just tried with a 15+y old HY1503 cheapo PSU, and my red led survives, no problem (18.7V/20mA set).. Btw, the output capacitor there is 470uF, in parallel with 1k resistor, afaik. The voltage on the led drops in aprox 25ms (from what I see on my scope). I will post the schematics for reference (I messed with it as it was broken, the schematics fits, the values of the pots are different in the real hw, not easy to follow the design with internally "grounded" positive output node, I had to do some sims in the LTspice to get how it works).. Almost the same as the above xavier60 has linked.. 

Actually, your PSU is Harrison topology. I'm surprised that the red LED survives that discharge from 470uF.
Ill do some tests on mine later. It's output cap is 47uF.

[iMo]

..
Actually, your PSU is Harrison topology. I'm surprised that the red LED survives that discharge from 470uF.
Ill do some tests on mine later. It's output cap is 47uF.

I've been surprised too (done perhaps 100 firings)..

[T3sl4co1l]

D44H11 are nice transistors, one of the better, older and cheap yet still widely available, types.  They're... not perforated/ring emitter, I think, but finely interdigitated at least.  (Yay, citing @Noopy 's work in a post. )  maybe some mfgs used the other structures too.

Keep in mind that the circuit can only transition between CC and CV as fast as the control loop(s) can act.  Which is less and less as you go up in frequency.  Which means there must necessarily be a point where neither CC nor CV can truly control, and the output impedance trends towards some asymptotic default open-loop value -- determined by the output device(s) themselves.  Which... asymptotic really isn't the right word for it, because the impedance up there can still be plenty weeble-wobbly; more to say that, the influence due to control decreases asymptotically, leaving whatever the impedance spectrum inherent in the structure is; which might be series inductance, collector capacitance, transconductance (including fT effect), etc.

Since the control must start from such a point, then settle out over time, it stands to reason that the ideal output impedance is midway between the two, i.e. Z ~ Vmax/Imax.  Sometimes it would be better higher or lower (depending on actual CC/CV settings and desired operating mode), but this is presumably the best median approximation for a general-purpose design.  (Of course, if you know you're going to operate more often in some region of the V-I operating area, impedance can be optimized towards that direction; an LED supply for example might make such a compromise, uh... assuming one needed a super-fast LED supply, that is.)

This might not mean much for, like, conventional output structures -- you're going to have an emitter follower and whatever else as usual, and that's just kind of that; but given some creativity, the output impedance of the level shifter / amplifier / follower can be tuned by joint combination of shunt (voltage) and series (current) feedback; speed (of the output stage itself) can also be optimized by distributing voltage and current gain between stages, rather than having all in one (volt amp) and the other (emitter follower).  Granted, this tends to cost some bias current, and is more of an RF/wideband class A approach, but some class AB-ness can still be had from it and if you really need the speed, bias current is likely an agreeable compromise.

Tim

[xavier60]

The old design I linked earlier and the modified version both have Hi-Z output stages as do Harrison topology.
The tricky part is when an overload initially starts to pull down the output voltage, the currently active CV loop immediately begins to rapidly ramp up drive to the output, while the CC opamp is sitting there usually in some saturated state and gets caught off guard, needing time to discharge the compensation cap before it finally takes control of the output stage.
 D5 in the linked design allows the CC opamp to directly discharge the cap, greatly reducing the CV to CC transition time.
Will it be quick enough to save a poor red LED is the question.

[exe]

About LED test: I found that different LEDs have different overload capabilities. I've got some LEDs that are "indestructible", and I've got a back of cheap LEDs from aliexpress that are very sensitive. I'll try to make a video comparison if I find parts (most likely next week).

given some creativity, the output impedance of the level shifter / amplifier / follower can be tuned by joint combination of shunt (voltage) and series (current) feedback; speed (of the output stage itself) can also be optimized by distributing voltage and current gain between stages, rather than having all in one (volt amp) and the other (emitter follower).

I spent a lot of time trying to create something like that: hybrid feedback. But that's beyond my skills. Not even sure where to start. I tried to imagine a three-input opamp, but couldn't). I also tried to make something with TIA, also no success. So, now I'm experimenting with simple circuits.

[T3sl4co1l]

How about, y'know... not burning devices of unspecified burn rate?

A large-ish MOSFET, optional series limiting resistor (consider that LEDs have pretty substantial internal resistance), and a 555 timer set for a low duty cycle, will get you identical conditions with respect to the power supply, but with repetitive trigger for the oscilloscope to read the voltage (and current via shunt resistor) during the transient.

Tim

[magic]

I wonder if this could be done with a pair of OTAs. Lacking internal compensation, they ought to have very fast slew rate if biased generously. One could first diode-OR their outputs, and only then compensate the resulting output with some capacitance to ground or a bit of Miller across a common emitter power stage, or whatever.

[xavier60]

I wonder if this could be done with a pair of OTAs. Lacking internal compensation, they ought to have very fast slew rate if biased generously. One could first diode-OR their outputs, and only then compensate the resulting output with some capacitance to ground or a bit of Miller across a common emitter power stage, or whatever.

That would be fast. A possible problem might be in finding a compensation that suits both loops. I find that the CV and CC loops need to be compensated differently.

[T3sl4co1l]

Dual input stage OTA would be ideal, yes. You can also assign output clamping by simply placing diodes on the gain node (if you don't mind they're a bit inaccurate, being diodes), which isn't relevant here, but good when you need to set the output range of a controller generally speaking.  LM13700 is the only extant and cheap example though, and is relatively slow (~3MHz GBW?) for what we're talking here.  Convenient, but maybe not actually much of an improvement.

Definitely worth testing and playing with though.  Old bit of analog magic, they are!

OTOH, voltage mode op-amps are cheap and abundant, and you could simply plop in a 100MHz part to significantly reduce slewing, even without the help of saturation-clamping diodes.

Personally, where top precision isn't required, I've used something like this before,



the discrete diff pair has low offset (at most 10s mV), and local compensation handles its gain relative to the op-amp it's controlling.  The specific application was for voltage error amp in a current-mode control: the input +/- is reference and output voltages, and "Max" is the current reference; "Out" is the current setpoint for the current loop, which in turn runs PWM and a buck converter stage.

The equivalent for present purposes would be a gm output stage (i.e. common emitter/source plus op-amp for stability and linearization) with this feeding it.

Tim

[magic]

AD829 looks like it might work as a poor man's fast OTA.
There's no bias control, but that's not important for applications like here.

[xavier60]

This shows an idea I used in a 20A PSU. Q2 isolates the output of the CC opamp from the compensation cap C1. When the CC threshold is reached, the opamp's output can swing low at it's full slew rate until B-E of Q2 and the CC ORing diode D9 are forward biased. So at the same time of the CC opamp taking control of the output MOSFET, C1  becomes connected via Q2.
To get a smooth CV to CC transition, C1 needs to be pre-charged to a voltage that suits the CC setting.
I found no simple way to do this and finally used a crude PWM DAC from the PSU's microcontroller to set C1's voltage according to the CC setting.

[iMo]

I wonder whether somebody tried a PSU with for example the TDA2020 - that is basically an opamp +/-22V, 3.5A output, 20W, with two pins for controlling the output transistors (two internal "power and current limiting circuits").

[xavier60]

I wonder whether somebody tried a PSU with for example the TDA2020 - that is basically an opamp +/-22V, 3.5A output, 20W, with two pins for controlling the output transistors (two internal "power and current limiting circuits").

That could be a quick way to make a 4 quadrant PSU if CC mode could be added.