Technical Q: Virtual Ground- Opamp or not...

[Breadboard Warrior]

Hi.  I rarely post around here, But I often stop by and browse.  I'm not a beginner with this stuff, but there's something I've always had trouble understanding.

When developing a single rail opamp circuit, when should I use a buffer for the virtual ground, and when can I settle for a pair of resistors and a cap?

I've seen some quite complex circuits with many opamps all getting their bias voltage from a pair of say, 10k resistors and an electrolytic cap.

I've also seen some fairly simple circuits that go the extra mile and use an opamp as a buffer for this 1/2 Vcc.

So how do you decide?  What considerations/compromises are behind this choice?

Thanks.

[EBK]

For me personally, it comes down to this:

Does my circuit require an even or odd number of op amps? 

If the number is odd, I will have an extra, otherwise unused op amp in one of the chips in my design (since I almost always use dual or quad chips, e.g., TL072 or TL074).  The otherwise unused op amp becomes a buffer.

I arrived at this personal design philosophy through a question I asked in my very first post on this forum, by the way (I just gave an overdue "like" to bluebunny for this idea).   

[bluebunny]

Thanks, Eric.  I wasn't really counting the passing months, but you might just be back on my Christmas card list. 

[DrAlx]

I would think it depends on how stiff the bias voltage needs to be for your particular circuit.
To get a stiff bias without an op-amp you would need to use lower R values in the voltage divider.
That wastes more current compared to using higher values resistors in a divider followed by a voltage buffer, but then it's probably not something you would care about unless you were running off a battery and trying to make things as efficient as possible.
Like EBK, if I had a spare op-amp in the circuit, I would probably use it to provide a stiff bias.

[antonis]

And now rises the "all time classic" query:

In case of bias voltage buffer use, should also be there a shunt capacitor..?? 
(provided the exictence of resistive divider junction cap, of course..)

[Phoenix]

In case of bias voltage buffer use, should also be there a shunt capacitor..?? 
(provided the exictence of resistive divider junction cap, of course..)

The voltage divider on the input to the split-supply buffer should be decoupled with a sufficiently large cap to reduce noise (doesn't need to be large anyway, as the resistive divider should be sized for the input bias current of only that single op-amp anyway, 1Meg resistors would be fine for a TL07x for instance, so the cap can be sized to suit), but the output from the split-supply buffer should not, to avoid instability issues (oscillation). It will have very very low source impedance within the op-amps gain bandwidth anyway, so a capacitor wouldn't provide any real additional benefit.

[antonis]

Well said Greg but, even in case of NFB wired op-amp (like buffer..) there still IS some impendance..
Taking in mind the fact that impedance is "offset" by output transistor(s) V_CE, cap placement shouldn't be an improvement..??

(or shoud be just something for nothing..??) 

P.S.
From personal experience, use of that cap adds absolutely nothing..
(not the slightest audible difference in a simple booster w bias buffered..)

[anotherjim]

I'm thinking that an active virtual ground isn't worth doing. An unused op-amp isn't so expensive that it can't just be tied up and kept out of the way.
Local vref divider/decoupling is a better solution:-
Avoids possible stability/interaction/noise from an additional common power rail.
Avoids distribution of an additional power rail.

[Phoenix]

Buffered/active split-supply can be a good idea when you've got large current pulses going into Vref, like if your circuit has an OTA using it as a current source, or you HAVE to dump clipping diodes into it, especially if they're assymetric, which can drag Vref up or down and put signals out of the common-mode input range of other stages, which could be likend to blocking distortion in vacuum tubes/valves, but will generally not sound as pleasant

Antonis, even the lowly TL07x wired as a non-inverting buffer has output impedance below 1ohm over the audible range, and putting a cap on a typical op-amp output is asking for instability issues, might not always happen, but it is sure to rear its ugly head when most inconvenient. You have to get into more exotic op-amps if you want something that will guarantee stability with capacitive loads.

[amptramp]

An op amp that is stable with capacitive loads sounds like a regulator to me.  Have your divider resistors then put that into an adjust pin of an LM317 and have its output feed the adjust pin of an LM337.  The output will be within a few millivolts of the divider voltage and you can load it down with as much capacitance as you want.  You will need nominal resistive loads for both regulators since they only pull in one direction.

[R.G.]

When developing a single rail opamp circuit, when should I use a buffer for the virtual ground, and when can I settle for a pair of resistors and a cap?

I've seen some quite complex circuits with many opamps all getting their bias voltage from a pair of say, 10k resistors and an electrolytic cap.

I've also seen some fairly simple circuits that go the extra mile and use an opamp as a buffer for this 1/2 Vcc.

So how do you decide?  What considerations/compromises are behind this choice?

You decide on the basis of (1) how much current is to be inserted/removed from the artificial ground by the sum total of everything connected to it (2) the degree of low impedance required by the biased circuits, and (3) the frequency response limits of what's being done with the bias.

An opamp bias circuit that feeds only non-inverting inputs can be minimal, as all the bias circuit supplies is the presumably small bias currents of the opamps. It really, really behooves you to know what opamps you're using, and what their input bias currents and offsets are; something like FET input opamps will have nano-amp or smaller biases, some bipolars will need microamps. The devil is always in the details.

An opamp bias circuit that feeds the shunt resistance/etc. to an inverting input may need to source/sink substantial currents, as that is where the current balancing for making voltage gain happen in a stock opamp gain stage comes from. This can drive the need to have much lower impedances at both AC and DC for the bias node. There is a write up on this topic at geofex.com. Again, the devil is in the details, and if your circuit is whacking the bias node with inverting input currents or bias currents for other parts of the circuit, the amount that the bias node gets waved around is fed right back into those non-inverting inputs to be amplified. Mayhem can result. Again, you have to know your circuit; the devil is again in the details.

Finally, you have to understand what frequency range you want your bias node to be an "AC ground" within. If it must go down to DC with low impedance, you must use some kind of regulator, which can be an opamp output or other regulator circuit. This is very rarely absolutely necessary. Sometimes it might be if your circuit necessarily needs DC accuracy for some reason. Most effects circuits don't. If your circuit can live with only AC decoupling of the bias point and the DC impedance of half of the bias resistors is good enough, then you only need to decouple the bias node with a cap, making that cap's impedance significantly less than the parallel combination of all impedances feeding into it. So if the bias net is 10K/10K, and you want to run several opamps with 3.3K resistors to their inverting inputs, you could see loading impedances in the 1K or less range from all the loading. So you'd need a cap that had "small" impedance at the lowest freguency of interest and that loading. If you want to get down to, say, 40Hz for bass, and you have a parallel impedance of 1k total to be decoupled, you need a cap of

C = 1/(2*pi*1K*4Hz) = 39.8uF ; that's a minimum value. If the loading was only several noninvering inputs, you could get by with only 2.2uF or so.

As noted, you can run into trouble with feedback through the bias node, and one solution is to not have a THE bias node, but several, each running its own few inputs. Works well as long as you don't try to DC switch the outputs; you'll get popping.

A buffered bias point solves many of the problems of having to know what you're doing, but introduces, again as noted, the problem of oscillating with capacitive loads. This is pretty soluble with series resistors and shunt caps from the bias output to each bias point. I've done some pretty complex bias setups this way.

[Rob Strand]

I'm thinking that an active virtual ground isn't worth doing. An unused op-amp isn't so expensive that it can't just be tied up and kept out of the way.
Local vref divider/decoupling is a better solution:-
Avoids possible stability/interaction/noise from an additional common power rail.
Avoids distribution of an additional power rail.

Agreed.   Unless you need a DC path there's really no point.  Except in an few special cases it doesn't add any goodness.  It does add noise.

Having local Vrefs is probably a better solution however you should put AC coupling between VREF domains as the VREFs can be at slightly different DC levels.

[Breadboard Warrior]

Thanks all for your excellent replies.  There's much to think over...

Especially R.G.  Very informative.  (Although, your math is for 4Hz instead of 40Hz, but I get your point.) 

C = 1/(2*pi*1K*4Hz) = 39.8u

OT:  I've read your site for many years.  Thanks for your efforts and insight!

As for just using a spare opamp, well of course that's crossed my mind, but as I'm moving towards surface mount stuff, that's not really an issue.  There are plenty of single opamps in the smt world so an odd number of opamps is fine.  Plus dedicated rail splitter ics.

So it really comes down to how much the current demand on that bias point is likely to push and pull the DC at that node around, right?  And how much fluctuation here is acceptable for the opamps you've chosen.

[PRR]

> how much fluctuation here is acceptable for the opamps you've chosen.

The *circuits*. Not usually the opamps.

If all that Vref is doing is setting-up a bunch of +In pins, current is teensy and interaction may be small.

If Vref is also being the "audio ground" reference for a bunch of 470 Ohm gain-set resistors spanning many stages, big signal through 470r into a wimpy Vref will swing it like fishing-pole and ripple through the whole system.

[R O Tiree]

Antonis, even the lowly TL07x wired as a non-inverting buffer has output impedance below 1ohm over the audible range, and putting a cap on a typical op-amp output is asking for instability issues, might not always happen, but it is sure to rear its ugly head when most inconvenient. You have to get into more exotic op-amps if you want something that will guarantee stability with capacitive loads.

... and yet we see the vast majority of schems all over teh interwebs, showing a cap at the output of opamps, followed by a resistor (or pot) to GND.  Should we be slapping a little 100R or so in between the cap and the output pin?  Or does the resistor to GND after the cap contribute to the party?

[antonis]

I presume it doesn't mind resistor's party contribution as long as it's invited to..

[Phoenix]

... and yet we see the vast majority of schems all over teh interwebs, showing a cap at the output of opamps, followed by a resistor (or pot) to GND.  Should we be slapping a little 100R or so in between the cap and the output pin?  Or does the resistor to GND after the cap contribute to the party?

You're quite right, the input resistor to ground of the subsequent pedal or amp prevents the load from being purely capacitive, and maintains the phase margin of the op amp, preventing it from becoming an oscillator under "normal" situations.
Adding an extra 100R is prudent though (I always add that to everything I build), it helps prevent instability in the "edge" cases where say you have a very long, capacitive cable, or the next pedal in line has a capacitor to ground on the input (low pass filter), or just the simple case where the output might get shorted to ground during switching, which could cause oscillation, and then take a while to recover from it when you switch back on.

[PRR]

> we see the vast majority of schems all over teh interwebs, showing a cap at the output of opamps, followed by a resistor (or pot) to GND.

10uFd and 100K, the opamp only "feels" 100K of loading, not a problem.

If you short the out jack, now the opamp feels a short plus 10uFd. It is OK at DC but gain falls-off by 1KHz (10uFd is 17 Ohms @ 1KHz; most opamps can't drive 17r well; and worse as F goes up). This roll-off adds to the internal roll-off from 10Hz to 1MHz. The double-steep roll-off above 1KHz is liable to make a 2-pole resonator. We need a 3rd pole for sure oscillation, but roll-offs are everywhere.

Adding 100r outside the feedback loop but before any Unknown Loads means the worst the opamp feels is 100 Ohms. The beefier ones will do the right thing. Our friend TL072 is weak in 100r, I would lean to 220r in any adverse situation.

[R O Tiree]

Thanks, chaps.  Five years ago, when I stopped making pedals completely (her indoors annoyed at my habit of doing this stuff at my desk in the lounge, then bought a small farm, plus the day job...) I would have either known that, or been able to deduce it.  A good lesson (re-)learned.

Onward and upward.