Virtual ground and stability driving a capacitive load

[motocoder]

I've been working on restoring an old precision current source. One of the portions of the circuit that used components for which direct replacements were not available, is a DC supply that generates +/- 12V. The supply works by rectifying a secondary output of a transformer into ~31VDC, and then running that through a linear regulator to get 24VDC. This 24V is then "split" using an op-amp "virtual ground" circuit (see the portion of the schematic from the manual below).

So because I needed to replace this circuit, I just built something myself out of an LM317 regulator to supply the 24V, and an op-amp and a simple push-pull circuit with discrete transistors to supply the virtual ground. The discrete transistors are there because the op-amps I had available couldn't supply enough current. This all seemed to work great, however, I noticed a little bit of oscillation on the output of the equipment. It didn't occur to me that my virtual ground circuit might be the source of this, but now I am starting to think that it is. I had already ripped up the circuit to rebuild the thing on a separate, small PCB, before I thought of this, so I can't just check on the scope to see. However, I did mock the thing up in LTSpice, and I definitely see oscillation in the simulation.

I think the root cause of that is the capacitive load on the output (C215 and C216 in the original schematic). That's an obvious source of stability problems when working with op-amps.

So my question is: Does this seem plausible? Should I try and compensate the circuit to remove this oscillation, or just remove those two capacitors? My inclination would be to just remove the capacitors, as I don't see that they are adding any value.

Relevant portion of original schematic


Replacement circuit (from LTSpice model)

[rs20]

1. Firstly, it is semi-good practice to have a 10k resistor (rather than the dead short you have) between the VGND line and the inverting input of the opamp, to equalise error due to input bias current. Maybe offset on the virtual ground isn't such a big deal in this application, but it also allows you to do step 2 below:

2. Random, uneducated though: add a capacitor from opamp output to inverting input? (Without the 10k resistor, the cap would just be fighting against VGND). The capacitor will settle down the speed of the opamp, and hopefully bring it to a stable state. Having said that, I'm fairly bad with control loops and the cap might just make things worse.

[coppice]

Op-amp instability while driving capacitive loads is nicely addresses here http://www.ti.com/ww/en/analog/Amplifiers-eBook/#chapter-1-2

[motocoder]

1. Firstly, it is semi-good practice to have a 10k resistor (rather than the dead short you have) between the VGND line and the inverting input of the opamp, to equalise error due to input bias current. Maybe offset on the virtual ground isn't such a big deal in this application, but it also allows you to do step 2 below:

2. Random, uneducated though: add a capacitor from opamp output to inverting input? (Without the 10k resistor, the cap would just be fighting against VGND). The capacitor will settle down the speed of the opamp, and hopefully bring it to a stable state. Having said that, I'm fairly bad with control loops and the cap might just make things worse.

I've actually tried both those things in the simulation, and some others as well - with a range of op-amps it oscillates. It's a fairly low level oscillation, but if it's happening in the real circuit, it would definitely couple through to the output.

[motocoder]

Op-amp instability while driving capacitive loads is nicely addresses here http://www.ti.com/ww/en/analog/Amplifiers-eBook/#chapter-1-2

That basically says stick a resistor between the output of the op-amp/transistor stage and the capacitors/load. Ok, these are also things I have tried in the simulator (I was going through this reference: http://www.analog.com/library/analogDialogue/archives/31-2/appleng.html ), but unless I make the resistor prohibitively high, it doesn't help. And if I do that, I can guarantee it's going to cause issues with the circuit this is going into. The whole idea of using the transistor output stage is to allow it to source and sink enough current. Sticking a 500 ohm resistor on the output will guarantee I can't meet that goal.

[rs20]

If you're just leaving the two 100 ohms resistors in the circuit, and adding your 500 ohm resistor afterwards, then clearly that creates a voltage divider, which is why you need such an unacceptably high resistance there.

I'd be curious to see what happens if you take your negative feedback off the emitter of Q1, rather than from VGND. That will prevent the voltage divider effect. Obviously this will make the VGND line the wrong DC value, but see if it fixes the oscillation? Then again, everything we're doing here is more or less equivalent to my original suggestion :-/

[motocoder]

If you're just leaving the two 100 ohms resistors in the circuit, and adding your 500 ohm resistor afterwards, then clearly that creates a voltage divider, which is why you need such an unacceptably high resistance there.

I'd be curious to see what happens if you take your negative feedback off the emitter of Q1, rather than from VGND. That will prevent the voltage divider effect. Obviously this will make the VGND line the wrong DC value, but see if it fixes the oscillation? Then again, everything we're doing here is more or less equivalent to my original suggestion :-/

I think you've made an error there. The feedback is coming off the junction of R5 and R6, which means the op-amp will act to make that node match the voltage on it's non-inverting input (12V, the "virtual ground" midway between 0 and 24V supply). Moving the feedback to the emitter of Q1 will mean that it is trying to keep the emitter of Q1 at 12V, and then the voltage at the load will actually be somewhat less due to the voltage divider between R5, R6, the series resistor you've suggested I add, and the load resistor.

[coppice]

Op-amp instability while driving capacitive loads is nicely addresses here http://www.ti.com/ww/en/analog/Amplifiers-eBook/#chapter-1-2

That basically says stick a resistor between the output of the op-amp/transistor stage and the capacitors/load. Ok, these are also things I have tried in the simulator (I was going through this reference: http://www.analog.com/library/analogDialogue/archives/31-2/appleng.html ), but unless I make the resistor prohibitively high, it doesn't help. And if I do that, I can guarantee it's going to cause issues with the circuit this is going into. The whole idea of using the transistor output stage is to allow it to source and sink enough current. Sticking a 500 ohm resistor on the output will guarantee I can't meet that goal.

It doesn't just say stick a resistor in. It gives you a proper explanation of the root cause of the problem. Adding a resistor alters the relationship between the closed and open loop response curve in a beneficial way. If you can't tolerate the rise in output impedance you need to find an alternative way to achieve the same effect. Its not easy, without adding considerable complexity. These kinds of virtual ground circuits with a single op-amp are common, and are typically poorly done. You often see huge amounts of capacitance, clearly trying to slug the output of the op-amp so hard it can't fight back.:-) . Its not a proper solution, and the amps usually fight harder than you might expect.

[motocoder]

Op-amp instability while driving capacitive loads is nicely addresses here http://www.ti.com/ww/en/analog/Amplifiers-eBook/#chapter-1-2

That basically says stick a resistor between the output of the op-amp/transistor stage and the capacitors/load. Ok, these are also things I have tried in the simulator (I was going through this reference: http://www.analog.com/library/analogDialogue/archives/31-2/appleng.html ), but unless I make the resistor prohibitively high, it doesn't help. And if I do that, I can guarantee it's going to cause issues with the circuit this is going into. The whole idea of using the transistor output stage is to allow it to source and sink enough current. Sticking a 500 ohm resistor on the output will guarantee I can't meet that goal.

It doesn't just say stick a resistor in. It gives you a proper explanation of the root cause of the problem. Adding a resistor alters the relationship between the closed and open loop response curve in a beneficial way. If you can't tolerate the rise in output impedance you need to find an alternative way to achieve the same effect. Its not easy, without adding considerable complexity. These kinds of virtual ground circuits with a single op-amp are common, and are typically poorly done. You often see huge amounts of capacitance, clearly trying to slug the output of the op-amp so hard it can't fight back.:-) . Its not a proper solution, and the amps usually fight harder than you might expect.

Thanks for reviewing and commenting. I do appreciate the help reasoning through the options here.

I actually understand why it oscillates, and I read several articles on driving large capacitive loads with an op-amp before posting. I can't tolerate a high output impedance because this piece of test equipment places the equivalent of a 200 ohm load on this virtual ground, and it switches from sourcing to sinking current when you adjust one of the controls on the front panel (the output polarity). If that wasn't the case, I could just adjust the 24V supply to set the VGND at exactly 12V above the actual ground because the load when it's set on a particular polarity is relatively constant. However, can't do that for the reasons stated, and I need to keep the output impedance low.

So I am not hearing any actual solutions so far that will work for my scenario. It seems like the options are:

- Remove the filter caps, and hope any remaining capacitance in the load is low enough not to cause stability issues.
- Use an op-amp without built in compensation, and then compensate the crap out of it using its compensation pins
- Use some other form of frequency compensation. I did try some ideas there based on the article I posted, but in simulation, I was still seeing oscillations.

[rs20]

If you're just leaving the two 100 ohms resistors in the circuit, and adding your 500 ohm resistor afterwards, then clearly that creates a voltage divider, which is why you need such an unacceptably high resistance there.

I'd be curious to see what happens if you take your negative feedback off the emitter of Q1, rather than from VGND. That will prevent the voltage divider effect. Obviously this will make the VGND line the wrong DC value, but see if it fixes the oscillation? Then again, everything we're doing here is more or less equivalent to my original suggestion :-/

I think you've made an error there. The feedback is coming off the junction of R5 and R6, which means the op-amp will act to make that node match the voltage on it's non-inverting input (12V, the "virtual ground" midway between 0 and 24V supply). Moving the feedback to the emitter of Q1 will mean that it is trying to keep the emitter of Q1 at 12V, and then the voltage at the load will actually be somewhat less due to the voltage divider between R5, R6, the series resistor you've suggested I add, and the load resistor.

Yeah, I did say the dc value of vgnd would be wrong, and that I was just curious as to whether it fixed the oscillation. If it did, you could generate a feedback signal couple to vgnd in dc and q1e in ac, but like I said that's moving back towards my original suggestion. Don't mind me :-)

[T3sl4co1l]

What was the original op-amp?  LM309 or something?

Part of the reason most voltage op-amps are unstable with capacitive load is they're dominant pole compensated.  The original appears to have pole-zero compensation (R+C across two pins) which might be enough to keep it happy.

Putting a buffer after a generic amp is probably not going to help.  The current capacity of that circuit doesn't look much more than the amp itself, and it needs compensation.

If higher-than-average current capacity is required, you should shop for an op-amp with suitable ratings, and capacitive load drive capability.  A common case is up to 18V supply rating and 60mA output, for something to do with LCD driving.  Or if you need more, there are driver/buffer kinds, or you can go to the trouble of building (and compensating) your own (or using a proper split supply to begin with).

How much load is actually on that rail?  A few mA?

Tim

[dom0]

Op-amp based virtual ground circuits are almost all total crap, since they will always become instable with some capacitive loads.

This one is much better. Unconditionally stable. https://www.mikrocontroller.net/topic/258266#2675651
Stability comes at a price here (lower precision, higher output resistance), but neither matters in usual applications.

[motocoder]

If you're just leaving the two 100 ohms resistors in the circuit, and adding your 500 ohm resistor afterwards, then clearly that creates a voltage divider, which is why you need such an unacceptably high resistance there.

I'd be curious to see what happens if you take your negative feedback off the emitter of Q1, rather than from VGND. That will prevent the voltage divider effect. Obviously this will make the VGND line the wrong DC value, but see if it fixes the oscillation? Then again, everything we're doing here is more or less equivalent to my original suggestion :-/

I think you've made an error there. The feedback is coming off the junction of R5 and R6, which means the op-amp will act to make that node match the voltage on it's non-inverting input (12V, the "virtual ground" midway between 0 and 24V supply). Moving the feedback to the emitter of Q1 will mean that it is trying to keep the emitter of Q1 at 12V, and then the voltage at the load will actually be somewhat less due to the voltage divider between R5, R6, the series resistor you've suggested I add, and the load resistor.

Yeah, I did say the dc value of vgnd would be wrong, and that I was just curious as to whether it fixed the oscillation. If it did, you could generate a feedback signal couple to vgnd in dc and q1e in ac, but like I said that's moving back towards my original suggestion. Don't mind me :-)

Well, I don't have the real circuit to test with, but at least in sim putting a large enough series resistor in before the caps does fix the oscillation - just as the analysis says it will.  Could you sketch out a circuit showing the feedback approach you are proposing? I've been tossing that idea around, but it's challenging because the op-amp is configured as a voltage follower, so no matter what you do, the minimum gain is 1.

[motocoder]

What was the original op-amp?  LM309 or something?

Part of the reason most voltage op-amps are unstable with capacitive load is they're dominant pole compensated.  The original appears to have pole-zero compensation (R+C across two pins) which might be enough to keep it happy.

Putting a buffer after a generic amp is probably not going to help.  The current capacity of that circuit doesn't look much more than the amp itself, and it needs compensation.

Wrong. Using a variety of op-amps, the voltage virtual ground was getting pulled several volts to one side or the other by the load. My circuit, using a 741 op-amp and the transistors as configured, was able to source/sink up to 80mA while keeping the virtual ground within a millivolt.

If higher-than-average current capacity is required, you should shop for an op-amp with suitable ratings, and capacitive load drive capability.  A common case is up to 18V supply rating and 60mA output, for something to do with LCD driving. 

Requirements here are 24V. Read original post for details.

Or if you need more, there are driver/buffer kinds, or you can go to the trouble of building (and compensating) your own (or using a proper split supply to begin with).

Yes, I'll just rip out the transformer and replace it with one that can supply the other 3 rails, plus the split supply for the +/-12V. Let me know where to find one, and make sure it fits the original PCB.

How much load is actually on that rail?  A few mA?

40mA or so, both source and sink. Re-read original post.

[motocoder]

Op-amp based virtual ground circuits are almost all total crap, since they will always become instable with some capacitive loads.

This one is much better. Unconditionally stable. https://www.mikrocontroller.net/topic/258266#2675651
Stability comes at a price here (lower precision, higher output resistance), but neither matters in usual applications.

But they do matter here. Thanks.

[motocoder]

There's a good article at the link below. I am going to try the compensation technique they suggest.

http://www.maximintegrated.com/en/app-notes/index.mvp/id/5597

[T3sl4co1l]

Wrong.

Read original post for details.

Re-read original post.

If you merely didn't want help, just say so...

[c4757p]

Op-amp based virtual ground circuits are almost all total crap, since they will always become instable with some capacitive loads.

This one is much better. Unconditionally stable. https://www.mikrocontroller.net/topic/258266#2675651
Stability comes at a price here (lower precision, higher output resistance), but neither matters in usual applications.

That won't be unconditionally stable with some parasitic layout inductance thrown in the mix. Chances are the ESR of the capacitors will dampen it, but that might not be the case anymore when you get some nice ceramic decoupling caps and whatnot in there too.

[motocoder]

Wrong.

Read original post for details.

Re-read original post.

If you merely didn't want help, just say so...

I do want help, but your suggested solutions didn't meet the project requirements. No offense intended, and I appreciate the suggestions anyway.  I shouldn't post before my morning coffee

[motocoder]

If you're just leaving the two 100 ohms resistors in the circuit, and adding your 500 ohm resistor afterwards, then clearly that creates a voltage divider, which is why you need such an unacceptably high resistance there.

I'd be curious to see what happens if you take your negative feedback off the emitter of Q1, rather than from VGND. That will prevent the voltage divider effect. Obviously this will make the VGND line the wrong DC value, but see if it fixes the oscillation? Then again, everything we're doing here is more or less equivalent to my original suggestion :-/

I think you've made an error there. The feedback is coming off the junction of R5 and R6, which means the op-amp will act to make that node match the voltage on it's non-inverting input (12V, the "virtual ground" midway between 0 and 24V supply). Moving the feedback to the emitter of Q1 will mean that it is trying to keep the emitter of Q1 at 12V, and then the voltage at the load will actually be somewhat less due to the voltage divider between R5, R6, the series resistor you've suggested I add, and the load resistor.

Yeah, I did say the dc value of vgnd would be wrong, and that I was just curious as to whether it fixed the oscillation. If it did, you could generate a feedback signal couple to vgnd in dc and q1e in ac, but like I said that's moving back towards my original suggestion. Don't mind me :-)

So, I thought about your suggestion for separate AC and DC feedback paths, as well as similar comments in the Maxim app note I posted earlier in the thread. This lead to the modified circuit below, which seems to be stable in the simulator. I'll breadboard it up and try it out and see if that is the case in the real world too. Thanks for the idea!

[dom0]

Op-amp based virtual ground circuits are almost all total crap, since they will always become instable with some capacitive loads.

This one is much better. Unconditionally stable. https://www.mikrocontroller.net/topic/258266#2675651
Stability comes at a price here (lower precision, higher output resistance), but neither matters in usual applications.

But they do matter here. Thanks.

That it needs to be precision wasn't stated by you, but good to know

btw. with "higher output resistance" I mean something like a few Ohms.

c4757p: I couldn't see any instability even with long leads and ceramic chip caps on the output.

[motocoder]

btw. with "higher output resistance" I mean something like a few Ohms.

I know simulation in this scenario can be misleading, but it had to be a lot more than a few ohms to stabilize it in LTSpice. However, with the separate DC/AC feedback paths, this isn't an issue.

For those that are wondering why this is a requirement, the Keithley 225 reference circuit generates a 1V reference above this VGND or below this VGND depending on which way the output polarity selector switch is set. This VGND, and hence the + and - 12V off of it, "float" on a bus that sits between the gross current limit sense resistors and the final output current sense resistors  on the high side. So it is important that VGND is clean and doesn't change much when the output polarity is changed as any mismatch there leads to mismatched output current when you toggle the polarity.

[dom0]

btw. with "higher output resistance" I mean something like a few Ohms.

I know simulation in this scenario can be misleading, but it had to be a lot more than a few ohms to stabilize it in LTSpice. However, with the separate DC/AC feedback paths, this isn't an issue.

I referred to the circuit I linked above ; I only realized later that saying "higher output resistance" in this thread might lead people to believe the linked circuit has 100+ ohms or something like that output resistance.

[Kevin.D]

The first spice circuit you posted has some issues. 1/ The two monsters R3,R4  in the
emitters cause the output impedance of the followers  to be raised from ~ 1
 ohm upto 100 ohms.(simply work out what that does to the time constant of any capacitive loads on the output)
2/ in your spice sim  did you right click and give those Electrolytic's some esr ?.this has to be included because the inherent ESR of large electrolytics makes them stable in these type of circuits and some circuits rely on this to create stable conditions (the original you posted relies both on the opamp comp cap and also the esr of the output caps to provide it's loop stabilty ).

Also when using largish caps on the output it's not really needed to bias the push pull as class AB since this is a Voltage regulator and not an Hi fidelty amplifier, so crossover distortion doesn't matter (it  takes a few micro seconds for the opamp to swing and bring the complementary transistor into conduction and when there is also a largish capacitor on the output this will dominate the output slewing rate during any load changes as it will hold the output V steady by sinking/sourcing current during fast load changes, (some spice load stepping experiments here and you might notice  the transient load response is barely effected by class AB, or B biasing ).
 I have included an alternate  simple circuit below for you compare with your version, the loop response plot is included if your interested, it's got good GBW with < 10 uS recovery . (the components I chose are just example g.p and most general purpose opamp in lm324 class will do the job, and BJT again general purpose medium power but will need to handle your ~ 2 Watt dissipation so to220 devices or somthing. Note
the preload resistor which I included to always insure a small UNBALANCED load is present (if  by a long shot you happen to have a perfectly balanced loads on both rails  (within 1 mA) then both output transitors switch off which might cause some very minor ringing during fast transients (says spice but I doubt in a real circuit it would  ).
On the output caps if you choose > 10uF low esr aluminiums with esr in 1-6 Ohm esr range then you should not need any Feedback cap at all but if you wanted to include one for peace of mind or perhaps sub zero temp working then use a ~ 330p from OP out to - OP in  ( esr of alco's goes up ALOT  when very cold ).

Regards   

[motocoder]

The first spice circuit you posted has some issues. 1/ The two monsters R3,R4  in the
emitters cause the output impedance of the followers  to be raised from ~ 1
 ohm upto 100 ohms.(simply work out what that does to the time constant of any capacitive loads on the output)

I think you are forgetting that the high gain op-amp with negative feedback lowers the effective output impedance (look where the negative feedback is coming from - it's after the 100 ohm resistors).

2/ in your spice sim  did you right click and give those Electrolytic's some esr ?.this has to be included because the inherent ESR of large electrolytics makes them stable in these type of circuits and some circuits rely on this to create stable conditions (the original you posted relies both on the opamp comp cap and also the esr of the output caps to provide it's loop stabilty ).

Ideal caps with zero ESR isn't the issue here. It's a widely understood fact that op-amps have stability issues when driving large capacitive loads.

Also when using largish caps on the output it's not really needed to bias the push pull as class AB since this is a Voltage regulator and not an Hi fidelty amplifier, so crossover distortion doesn't matter (it  takes a few micro seconds for the opamp to swing and bring the complementary transistor into conduction and when there is also a largish capacitor on the output this will dominate the output slewing rate during any load changes as it will hold the output V steady by sinking/sourcing current during fast load changes, (some spice load stepping experiments here and you might notice  the transient load response is barely effected by class AB, or B biasing ).

I couldn't disagree more. Since the circuit is designed to sit exactly midway between the two input rails, without the biasing scheme the transistors would continually be sitting right in the dead zone. This definitely has a detrimental effect on stability and output ripple of the circuit.

I have included an alternate  simple circuit below for you compare with your version, the loop response plot is included if your interested, it's got good GBW with < 10 uS recovery . (the components I chose are just example g.p and most general purpose opamp in lm324 class will do the job, and BJT again general purpose medium power but will need to handle your ~ 2 Watt dissipation so to220 devices or somthing. Note
the preload resistor which I included to always insure a small UNBALANCED load is present (if  by a long shot you happen to have a perfectly balanced loads on both rails  (within 1 mA) then both output transitors switch off which might cause some very minor ringing during fast transients (says spice but I doubt in a real circuit it would  ).
On the output caps if you choose > 10uF low esr aluminiums with esr in 1-6 Ohm esr range then you should not need any Feedback cap at all but if you wanted to include one for peace of mind or perhaps sub zero temp working then use a ~ 330p from OP out to - OP in  ( esr of alco's goes up ALOT  when very cold ).

Thanks, Kevin. I will take a look at this when I get home this evening.