Is 550uF too big for a power supply that has CC limit?

[bloguetronica]

Hi,

I'm designing a linear power supply that has both voltage and current control. Through both simulation and experimentation, I've concluded that 270uF is marginal in order to stabilize the power supply. So, I had to double the capacitance just to make sure.

However, by seeing the most recent video on EEVBlog, I've concluded that I might have impaired the CC mode of my supply. Should I be worried?

The attached schematic is just the output stage. There is also a pre-regulator stage, and you can see the discussion here:
https://www.eevblog.com/forum/projects/is-there-any-variable-dc-dc-converter-suitable-for-pre-regulation/

By the way, this project had a few corrections, in order to filter some instability caused by the interaction of the pre-regulator with this final stage. Also, the CC mode was unstable, so I had to add a capacitor (C12) near Q1. The new boards are not made yet, but they are now under way. Lets hope this works, because I've fried the whole control section of my "old" output stage board (a mystery).

Kind regards, Samuel Lourenço

[bson]

The IC4B based sense amplifier lacks external compensation.

[Kleinstein]

How much capacitance is acceptable at the output depends on the current range and application. 500 µF is not good, but about normal for many 3 A supplies.  The capacitor C12 is odd - it slows down the CC mode a lot and has good chances to make it oscillate with an inductive load.  Both the current and voltage sensing are using differential amplifiers in a way that is not good for accuracy and depending on tolerances can cause stability problems. So the whole concept is not that good.

Normally a supply with an emitter-follower  output stage can work with a much smaller ouput capacitor. So with compensation at IC4B it should work with a much smaller (e.g. 10 µF) cap.

[Zero999]

It's all but impossible to design a PSU with constant current and constant voltage and have good load regulation for both. A capacitor will improve the transient response of the CV mode, at the expense of CC. No capacitor an a large inductor will improve the CC mode, at the expense of CV mode.

Normally we optimise for CV mode and accept that current surges beyond the current limit can occur. I've toyed with the idea of building a CV/CC power supply with a three position switch which adjusts the filtering to be optimum for CC, CV or the best of both, because no PSU I've seen offers this, although I've never got round to it.

[T3sl4co1l]

Delete C12, add a resistor in series with C11 and tweak their values; do the same with IC4B, an R+C from out to -in.

Don't expect it to go terribly fast with that 2.7k (R5) supplying a darlington.  I don't know offhand what C and ESR would be necessary to compensate this combination when tuned for best performance, but 100s of uF is probably not unreasonable.

Why two different DACs?  Seems a waste of a BOM line.  Similarly, the different op-amps and comparator used could be simplified.  But these don't affect the correctness of the circuit, in any case.

Tim

[bloguetronica]

The IC4B based sense amplifier lacks external compensation.

The compensation is done externally, via shunts in J5.

How much capacitance is acceptable at the output depends on the current range and application. 500 µF is not good, but about normal for many 3 A supplies.  The capacitor C12 is odd - it slows down the CC mode a lot and has good chances to make it oscillate with an inductive load.  Both the current and voltage sensing are using differential amplifiers in a way that is not good for accuracy and depending on tolerances can cause stability problems. So the whole concept is not that good.

Normally a supply with an emitter-follower  output stage can work with a much smaller ouput capacitor. So with compensation at IC4B it should work with a much smaller (e.g. 10 µF) cap.

I was recalling incorrectly. According to the simulations, 100uF was still marginal. A single capacitor of 150uF up to 270uF might be sufficient, after all. When you say 500uF is not good, is it too big or too small?

It's all but impossible to design a PSU with constant current and constant voltage and have good load regulation for both. A capacitor will improve the transient response of the CV mode, at the expense of CC. No capacitor an a large inductor will improve the CC mode, at the expense of CV mode.

Normally we optimise for CV mode and accept that current surges beyond the current limit can occur. I've toyed with the idea of building a CV/CC power supply with a three position switch which adjusts the filtering to be optimum for CC, CV or the best of both, because no PSU I've seen offers this, although I've never got round to it.

The CV mode takes priority here too. The CC mode is not vwery precise, and it is only there to protect the load. But I might be defeating the purpose if I use a too large filter cap.

Delete C12, add a resistor in series with C11 and tweak their values; do the same with IC4B, an R+C from out to -in.

Don't expect it to go terribly fast with that 2.7k (R5) supplying a darlington.  I don't know offhand what C and ESR would be necessary to compensate this combination when tuned for best performance, but 100s of uF is probably not unreasonable.

Why two different DACs?  Seems a waste of a BOM line.  Similarly, the different op-amps and comparator used could be simplified.  But these don't affect the correctness of the circuit, in any case.

Tim

Hi Tim,

One DAC is to control voltage, and the other is to control current. I think those resistors that you are suggesting could affect the gain, especially around IC4B (the loop is closed externally). What do you have in mind? Anyway, I've tried putting a 1K resistor in series with C11, and saw no improvement (the OPA705 is already unity gain stable). C12 seems to do the job.

Kind regards, Samuel Lourenço

[T3sl4co1l]

Well, trying a single point doesn't mean anything if you're very far from the local minima.  Try 10k, 100k, 1M even, and different C values.  A random search of the space is not necessary, you just have to get close enough to discover, and follow, the gradient.

A resistor in series with a capacitor has no effect on DC gain.

Tim

[bloguetronica]

Well, trying a single point doesn't mean anything if you're very far from the local minima.  Try 10k, 100k, 1M even, and different C values.  A random search of the space is not necessary, you just have to get close enough to discover, and follow, the gradient.

A resistor in series with a capacitor has no effect on DC gain.

Tim

But wouldn't larger resistors (in series with C11) make IC7 unstable again (and defeat the purpose of C11)? The issue here is not local instability. This instability was caused by IC6, IC7 and Q1 combined. Each sub-system is stable by itself (tested that). C12 is there to slow down the response and make the system stable as a whole (the CC mode is now constant with any load). I don't need a fast response in CC mode, but I do understand your objection.

Kind regards, Samuel Lourenço

[Kleinstein]

C12 is not just making the current regulation slow, it also adds phase shift and thus makes stability difficult. So C12 is definitely a bad idea.
If one really wants it slower it would be a larger cap for C11. Usually with this kind of circuit current regulation is already relatively slow - if one makes it too slow the simulated capacitance may be more than the real 500 µF. One problem it the output of IC7 swinging rather high, well above the active zone that is at some 1-3 V.  In the circuit the OP has only 5 V supply and this already helps, but there is still some delay before the current limit sets in.

A resistor in series with C11 adds a little more higher frequency gain and phase boost. If not too much it can help with stability. One may not need this if C12 is left out.

[blackdog]

Hi,

C12 is there to slow down the response and make the system stable as a whole (the CC mode is now constant with any load). Wrong!
As others have indicated, remove this capacitor.
The very common mistake when designing power supply is to make it very slow to get stability.

You just need Speed...........
All extra RC times, all extra active components like opamp and transistors within a control loop make this loop inherrent unstable (slower).
These components delay this loop and cause the phase spmargin to get smaller and smaller, when you reach 180-degrees you have built a transmitter. 

And the next component that will cause you a headache is Q1 in your schematic.
This transistor gives you extra loop gain that is hard to compensate.

It is clear that you miss a piece of knowledge regarding loopgain, phase margin, phase margin, etc.
Without this knowledge you can't design a good power supply.

Good information about this is available on the TI website: https://training.ti.com/ti-precision-labs-op-amps-stability-1


I hope this helps,

Kind regarts,
Bram

[bloguetronica]

C12 is not just making the current regulation slow, it also adds phase shift and thus makes stability difficult. So C12 is definitely a bad idea.
If one really wants it slower it would be a larger cap for C11. Usually with this kind of circuit current regulation is already relatively slow - if one makes it too slow the simulated capacitance may be more than the real 500 µF. One problem it the output of IC7 swinging rather high, well above the active zone that is at some 1-3 V.  In the circuit the OP has only 5 V supply and this already helps, but there is still some delay before the current limit sets in.

A resistor in series with C11 adds a little more higher frequency gain and phase boost. If not too much it can help with stability. One may not need this if C12 is left out.

In the previous circuit, there was no C12, and in CC mode the circuit oscillated and the current read incorrectly, with spikes of 1.5A when the limit was set to 500mA! Already tried a much bigger C11 (1uF!) and a resistor in series. The capacitor improved, but the resistor didn't helped.

Anyways, I see conflicting opinions regarding the need of C11 without C12. Taking C11 out solves the current spikes in the totally opposite direction (current is set to low, still oscillating).

Kind regards, Samuel Lourenço

[bloguetronica]

Hi,

C12 is there to slow down the response and make the system stable as a whole (the CC mode is now constant with any load). Wrong!
As others have indicated, remove this capacitor.
The very common mistake when designing power supply is to make it very slow to get stability.

You just need Speed...........
All extra RC times, all extra active components like opamp and transistors within a control loop make this loop inherrent unstable (slower).
These components delay this loop and cause the phase spmargin to get smaller and smaller, when you reach 180-degrees you have built a transmitter. 

And the next component that will cause you a headache is Q1 in your schematic.
This transistor gives you extra loop gain that is hard to compensate.

It is clear that you miss a piece of knowledge regarding loopgain, phase margin, phase margin, etc.
Without this knowledge you can't design a good power supply.

Good information about this is available on the TI website: https://training.ti.com/ti-precision-labs-op-amps-stability-1


I hope this helps,

Kind regarts,
Bram

The fact that Q1 ruins the phase margin is a hint, but I can't get rid of it. Hence, that's why C12 is needed, to dampen the response (I don't see any oscillation at the amp's output with it). Makes sense, and that's why any combinations of C11 plus/minus a 1K resistor in series wouldn't solve this. Now, C11 doesn't need do be huge, although without it the circuit will oscillate, due to the fact that the op-amp is unstable (a different issue here!).

I can only make the necessary changes if I have hard evidence that they will work. As far as I'm concerned, the CC mode was working fine, after the modifications (one of them was the inclusion of C12), until the 5V regulator blew up. I don't like C12 either, but theory doesn't make practice (also it is not as bad as if C12 was there right at the op-amp's output).

Kind regards, Samuel Lourenço

[bloguetronica]

Ok, I think there was an issue with my question, and I got misunderstood. It is not that my circuit was unstable. It worked fine with a 270uF output cap. I don't have stability issues. I just wanted to know of 550uF was too much.

Kind regards, Samuel Lourenço

[Kleinstein]

The IC7 with C11 still has unity gain at high frequency. So C11 can not provide the full compensation for the CC loop. So C12 may indeed be needed. So there is a chance to keep C12 (though maybe a little smaller) as the main compensation for the CC loop and make C11 small enough or provide enough series resistance so that at the higher frequencies it is C12 doing the job and C11 has still finished it's job.

 

[bloguetronica]

The IC7 with C11 still has unity gain at high frequency. So C11 can not provide the full compensation for the CC loop. So C12 may indeed be needed. So there is a chance to keep C12 (though maybe a little smaller) as the main compensation for the CC loop and make C11 small enough or provide enough series resistance so that at the higher frequencies it is C12 doing the job and C11 has still finished it's job.

The OPA705 is unity gain stable. I could try using a 10K resistor in series with C12, which would effectively double the gain at high frequencies, but I don't see this as a substitution for C12.

The issue here is that, I was initially going to use two 47uF caps. But now I found a 270uF cap that has the same footprint, and at some point I've decided to use two of them, as a replacement (the more, the merrier, right?). However, I was alerted by Dave's video that a too big output capacitance might impair the CC limiting of the power supply.

Nevertheless, I saw that a single 270uF cap would be more than enough. And perhaps, I should replace it with a 150uF cap (different footprint).

Kind regards, Samuel Lourenço

[T3sl4co1l]

The remark about unity gain is not about stability of the amp itself, but about its place in the loop.  Namely, even if the amp is severely overcompensated (say by putting 1000uF from out to -in), it still has gain of 1 from +in to out, which means the loop gain has not dropped to an ideal dominant pole as you might've been expecting with such a large cap.

The solution is to apply filtering to +in so it drops at the same rate.

Compare this, which has G = 1 at high frequency,



with this, which has G --> 0 at HF.



For an error amp, R2 and R4 --> infty, so that DC error is as nearly zero as possible.

For a pole-zero compensation of course, a resistor is placed in series with C2 and C4.

The zero itself means HF gain levels off; the point of doing this, versus leaving it alone, is you have control of the gain and frequency where this happens, rather than being stuck with wherever G=1 rolls it off at.

Tim

[blackdog]

Hi  bloguetronica,

For example, assume that IC7 is set as a 1x amplifier and has 60 degrees of phase space.
Of these 60 degrees little remains, this because of the following components in de current loop: IC6, R7+C12, Gain Q1 and Q4 your power stage, which will eat all phase margin.
45 degrees is the absolute minimum you need for a good en stable power supply, this under all conditions.

That at a certain capacitor value at the output connector, and the power supply does not generate, does not mean that it is stable...
You will have to test this circuit at various output voltages and also at various currents.
Test it dynamically with short pulses, pay special attention to the transition area when your power supply changes from CV to CC.

Kind regards,
Bram

[xavier60]

From my experience, designs that have high impedance output stages, also called transconductance or voltage to current transfer function are easy to compensate for good CV  and CC stability using what TI calls "Type II Compensator".
This typically applies to so called floating type designs which tend to give very good performance.
With a low impedance or voltage follower type output stage, in CC mode the loop gain would increase steeply with lowering load resistance.

http://www.ti.com/lit/an/slva662/slva662.pdf

[xavier60]

Have a close look at a typical LM723 implementation. Although the output stage is voltage follower in CV mode, it becomes transconductance in  current limiting mode.

[floobydust]

550uF is too much.
Prove it by setting any PSU to say 15V at 10mA and testing some LED's, as a real world example.

Energy is stored in the output capacitor which will be charged at 15V which dumps into the LED load down to 3V. You thought the LED was 12V but it's a 3V part. The CC mode transition takes some time as well.

So the LED might just look like a strobe light (bright) flash happened, or the LED might be dead now due to the impulse.

Some lab power supplies I can't even test LED's, they surge so bad. This is my criterion, a decent PSU is at most ~100uF but if you work with discrete semi's and small parts, 550uF is too big.

[bloguetronica]

The remark about unity gain is not about stability of the amp itself, but about its place in the loop.  Namely, even if the amp is severely overcompensated (say by putting 1000uF from out to -in), it still has gain of 1 from +in to out, which means the loop gain has not dropped to an ideal dominant pole as you might've been expecting with such a large cap.

The solution is to apply filtering to +in so it drops at the same rate.

Compare this, which has G = 1 at high frequency,



with this, which has G --> 0 at HF.



For an error amp, R2 and R4 --> infty, so that DC error is as nearly zero as possible.

For a pole-zero compensation of course, a resistor is placed in series with C2 and C4.

The zero itself means HF gain levels off; the point of doing this, versus leaving it alone, is you have control of the gain and frequency where this happens, rather than being stuck with wherever G=1 rolls it off at.

Tim

Op-amp stability is a science in itself. I confess that it is hard for me to understand.

Hi  bloguetronica,

For example, assume that IC7 is set as a 1x amplifier and has 60 degrees of phase space.
Of these 60 degrees little remains, this because of the following components in de current loop: IC6, R7+C12, Gain Q1 and Q4 your power stage, which will eat all phase margin.
45 degrees is the absolute minimum you need for a good en stable power supply, this under all conditions.

That at a certain capacitor value at the output connector, and the power supply does not generate, does not mean that it is stable...
You will have to test this circuit at various output voltages and also at various currents.
Test it dynamically with short pulses, pay special attention to the transition area when your power supply changes from CV to CC.

Kind regards,
Bram

Testing is planned for sure. Lets hope I don't fry another board. I'll have to test this using a 150uF output capacitor, see if it is stable, and do some improvements. I hope there is no instability due to the interaction and feedback with the DC-DC pre-regulator, and for that R15 and C16 will have to serve (those were added too).

Have a close look at a typical LM723 implementation. Although the output stage is voltage follower in CV mode, it becomes transconductance in  current limiting mode.

Thanks! The LM723 implementation is no different that the typical implementation that employs a series sensing resistor that will pull the BE junction of a transistor, which will in turn pull down the output of the error amplifier. I've used a 2.7K here to avoid shorting the amplifier output. Their approach is simple and effective, but only works if you want a fixed current limiting (something that I may have to do if this doesn't work).

550uF is too much.
Prove it by setting any PSU to say 15V at 10mA and testing some LED's, as a real world example.

Energy is stored in the output capacitor which will be charged at 15V which dumps into the LED load down to 3V. You thought the LED was 12V but it's a 3V part. The CC mode transition takes some time as well.

So the LED might just look like a strobe light (bright) flash happened, or the LED might be dead now due to the impulse.

Some lab power supplies I can't even test LED's, they surge so bad. This is my criterion, a decent PSU is at most ~100uF but if you work with discrete semi's and small parts, 550uF is too big.

Yup, even 270uF is enough to damage a LED in CC mode. Tested that yesterday. I'll have to consider the minimum value of 150uF just to keep the supply stable.

Kind regards, Samuel Lourenço

[bloguetronica]

Hi,

After seeing a few videos, it seems that I'll have to test the transient response of IC7 to see if there is any ringing. I'll try to test for both transients on the inverting and non-inverting input. I'll also check the stability of IC4B.

Kind regards, Samuel Lourenço

[David Hess]

10 to 100 microfarads per amp is good for a simple but well designed constant voltage and constant current power supplies.  Anything higher indicates problems with the frequency compensation.  Usually these problems come about because of additional phase lag from multiple cascaded stages within the control loops or excessive or uncontrolled voltage gain in a level shifter.

High performance constant voltage and constant current power supplies can do much better than this with only say 0.022 microfarads per amp but use faster output transistors and have provisions to prevent integrator windup in the error amplifiers.  They also may use a class-ab output stage with some active pull-down capability.

[bloguetronica]

10 to 100 microfarads per amp is good for a simple but well designed constant voltage and constant current power supplies.  Anything higher indicates problems with the frequency compensation.  Usually these problems come about because of additional phase lag from multiple cascaded stages within the control loops or excessive or uncontrolled voltage gain in a level shifter.

High performance constant voltage and constant current power supplies can do much better than this with only say 0.022 microfarads per amp but use faster output transistors and have provisions to prevent integrator windup in the error amplifiers.  They also may use a class-ab output stage with some active pull-down capability.

I think that this other power supply, that has the corresponding schematic attached, might have stability issues too. I ran into oscillations caused by a  misplaced ground connection (www.eevblog.com/forum/projects/usb-controlled-precision-power-supply-(or-voltage-reference)/msg2007782/#msg2007782). Normally, this supply is stable, but the constant current load that that you see is being supplied via a SMPS and causes some oscillation, as long as it is not grounded to earth. I'll have to do some tests tomorrow.

Never had such issue with the FAU200 model, that uses a simple buffer to control the pass transistor.

Kind regards, Samuel Lourenço

[David Hess]

I think that this other power supply, that has the corresponding schematic attached, might have stability issues too. I ran into oscillations caused by a  misplaced ground connection (www.eevblog.com/forum/projects/usb-controlled-precision-power-supply-(or-voltage-reference)/msg2007782/#msg2007782). Normally, this supply is stable, but the constant current load that that you see is being supplied via a SMPS and causes some oscillation, as long as it is not grounded to earth.

Maybe I am reading the schematic wrong but it looks to me like the remote sense to the voltage feedback loop is backwards, but it reminds me of a circuit I was studying a couple weeks ago from the LT1010 datasheet shown below.  The OPA703 is slow enough that it should be able to control the output through a TIP31 emitter follower without problems.

The first example shown below is a good example of what I was talking about.  D2 clamps the current control operational amplifier to prevent integrator windup.  C1 and to a lessor extend C2 maintain stability despite the current control loop being in series with the voltage control loop with two relatively fast operational amplifiers.  This design requires no output capacitance but has what is effectively a fast full class-ab output stage in the LT1010.  Note that the 2 ohm resistor in series with the output of the power stage aids stability when a capacitive load is present and this is an advantage of having the current shunt in series with the output.

The second example is from National Semiconductor is the low output capacitance design I was thinking of.  D2 takes advantage of the external compensation feature of the LM301A current control operational amplifier to clamp it preventing integrator windup preserving fast response.  Q2 makes the output stage more like class-ab with the ability to pull the output down.

The third example uses 100 microfarads of output capacitance for a 1/2 amp output current to control transient response because no clamping of the operational amplifiers is used.  In practice I think that capacitance could be 22 microfarads without problems but they wanted extra stability.