Paralleling OP amps

[IvoS]

I am building a virtual ground splitter. I have limited space on PCB that is why I decided to go with OPamp. I need around 30-40mA of current to be available for the other circuit that it will be powering. So paralleling two OPamps would be sufficient. I breadboarded this "A" and "B" version and differences are as follow:

-Version "A" works very well for voltage regulation and it can cource current of approx. 100mA. But when it runs unloaded it consumes 40mA by itself.

-Version "B" does not regulate voltage well enough due to feedback resistors tied right at output pins. There is like 10-100mV drop at VG output due to 5 ohm resistors voltage drop. This version does not cunsume extra current. All it consumes is the op amp idling current, which is around 2mA per 1 opamp.

Why is so much extra current consumed in version "A" when it runs unloaded?
Thanks.

[wraper]

You get high current consumption because outputs fight each other. Also you should not use resistors in series with inverting input, they won't do any good, and likely will even do bad. Better use opamp which can supply such current without paralleling.

[Benta]

I'm trying to understand why you need two opamps for generating a virtual ground voltage between +12 and -12 V.
A single high-current opamp will do the same with much better results.

Configured as buffer, the non-inverting input connected to your voltage divider should do the job.

Are there some special requirements that you have not mentioned?

Benta.

[Zero999]

It should work. Perhaps the current sharing resistors aren't big enough?

At only 40mA the current sharing resistors can be much higher. Try 47R.

And refer to the application note linked below:
http://www.intersil.com/content/dam/Intersil/documents/an11/an1111.pdf

[Benta]

It could probably be made to work.
But again, why complicate your life? There are plenty of high-current opamps or buffers out there at the same price.

Here you go:

LMH6321: http://www.ti.com/product/lmh6321

Same price range as the amps you've chosen. Much easier.

Benta.

[IvoS]

It should work. Perhaps the current sharing resistors aren't big enough?

At only 40mA the current sharing resistors can be much higher. Try 47R.

And refer to the application note linked below:
http://www.intersil.com/content/dam/Intersil/documents/an11/an1111.pdf

Thanks, I stumbled upon this note a few days ago. This is pretty much what I did. Mine connection doesn't have any gain. I will try to increase output resistors to 47 ohm and see what happens.
I will also try to connect it as shown in figure one. That may do the trick.

[IvoS]

It could probably be made to work.
But again, why complicate your life? There are plenty of high-current opamps or buffers out there at the same price.

Here you go:

LMH6321: http://www.ti.com/product/lmh6321

Same price range as the amps you've chosen. Much easier.

Benta.

Thank you. Reviewing its data...... That may be a good candidate.

[BrianHG]

I cant seem to find the old Motorola 150 ohm driver opamp I used to use, however, here is the equivalent:
http://www.digikey.com/product-detail/en/njr-corporation-njrc/NJM4556AD/NJM4556AD-ND/673767
Though it's only 70ma per channel, it's under 1$ a piece at digikey.

[rs20]

...you should not use resistors in series with inverting input, they won't do any good, and likely will even do bad.

Isn't it sometimes a good idea to make sure the op amp sees the same impedance (or as similar as can reasonably be achieved) on both inputs, so that the bias currents sunk by both inputs produce equal (and thus cancelling) drops? This is the reason op-amp followers have a resistor on their feedback path, even though it appears totally useless (under the idealistic but actually incorrect assumption that opamp bias currents equal zero).

[wraper]

...you should not use resistors in series with inverting input, they won't do any good, and likely will even do bad.

Isn't it sometimes a good idea to make sure the op amp sees the same impedance (or as similar as can reasonably be achieved) on both inputs, so that the bias currents sunk by both inputs produce equal (and thus cancelling) drops? This is the reason op-amp followers have a resistor on their feedback path, even though it appears totally useless (under the idealistic but actually incorrect assumption that opamp bias currents equal zero).

Really? I'll tell you what, it was good for old opamps with BJT input and without internal bias current cancellation. They had high input bias current, but it was almost equal on both inputs. So by keeping the same impedance on both inputs you could cancel them out.  OPA2134 has FET input. And input bias currents may be not only non equal but in opposite directions. So by adding resistor you will only increase offset (depending on particular opamp specimen) and noise.
http://www.analog.com/media/en/training-seminars/tutorials/MT-038.pdf

Most modern precision bipolar input stage op amps use some means of internal bias current
compensation, examples would be the familiar OP07 and OP27 series.
Bias current compensated input stages have many of the good features of the simple bipolar
input stage, namely: low voltage noise, low offset, and low drift. Additionally, they have low
bias current which is fairly stable with temperature. However, their current noise is not very
good, and their bias current matching is poor.
These latter two undesired side effects result from the external bias current being the difference
between the compensating current source and the input transistor base current. Both of these
currents inevitably have noise. Since they are uncorrelated, the two noises add in a root-sum-ofsquares
fashion (even though the dc currents subtract).
Since the resulting external bias current is the difference between two nearly equal currents,
there is no reason why the net current should have a defined polarity. As a result, the bias
currents of a bias-compensated op amp may not only be mismatched, they can actually flow in
opposite directions! In most applications this isn't important, but in some it can have unexpected
effects (for example the droop of a sample-and-hold (SHA) built with a bias-compensated op
amp may have either polarity).

CANCELING THE EFFECTS OF BIAS CURRENT (EXTERNAL TO THE OP AMP)
When the bias currents of an op amp are well matched (the case with simple bipolar input stage
op amps, but not internally bias compensated ones, as noted previously), a bias compensation
resistor, R3, (R3=R1||R2) introduces a voltage drop in the non-inverting input to match and thus
compensate the drop in the parallel combination of R1 and R2 in the inverting input. This
minimizes additional offset voltage error, as in Figure 3. Note that if R3 is more than 1 k? or so,
it should be bypassed with a capacitor to prevent noise pickup. Also note that this form of bias
cancellation is useless where bias currents are not well-matched, and will, in fact, make matters
worse.

In many cases, the bias current compensation feature is not mentioned on an op amp data sheet,
and a simplified schematic isn't supplied. It is easy to determine if bias current compensation is
used by examining the bias current specification. If the bias current is specified as a "±" value,
the op amp is most likely compensated for bias current. Note that this can easily be verified, by
examining the offset current specification (the difference in the bias currents). If internal bias
current compensation exists, the offset current will be of the same magnitude as the bias current.

[bktemp]

OPA2134 datasheet, page 9:

SOURCE IMPEDANCE AND DISTORTION
For lowest distortion with a source or feedback network
which has an impedance greater than 2k, the impedance
seen by the positive and negative inputs in noninverting
applications should be matched. The p-channel JFETs in the
FET input stage exhibit a varying input capacitance with
applied common-mode input voltage. In inverting configurations
the input does not vary with input voltage since the
inverting input is held at virtual ground. However, in
noninverting applications the inputs do vary, and the gateto-
source voltage is not constant. The effect is increased
distortion due to the varying capacitance for unmatched
source impedances greater than 2k.
To maintain low distortion, match unbalanced source impedance
with appropriate values in the feedback network as
shown in Figure 3. Of course, the unbalanced impedance
may be from gain-setting resistors in the feedback path. If
the parallel combination of R1 and R2 is greater than 2k, a
matching impedance on the noninverting input should be
used. As always, resistor values should be minimized to
reduce the effects of thermal noise.

So using a matched resistor in the feedback path is actually recommended for the JFET amplifier, but in this case it doesn't care, because the input voltage is constant. Therefore using a small or no resistor at all is probably the best solution, otherwise the resistor forms a low pass filter, making the amplifier potentially unstable.

[Kleinstein]

To get low output impedance and current sharing, there is a 3 rd, better version of paralleling the OP:
Use on OP like in Version A (so getting the feedback from behind the resistor). Use the second OP as a dedicated slave, with the positive input from the output of the 1 st OP, and than wired like Version B.
This is working very well at low frequencies, but may be limited at high frequencies.

This way the 2 nd OP is only increasing the current, the signal output is only determined by the 1.st one.

With stiff virtual GND like in version A (or better C), the capacitors have very limited effect. So one might use less here. With the not so stiff virtual ground in version B, the capacitor can take over much of AC currents and thus reduce the power consumption.

Instead of making the virtual ground so powerful, I would rethink if so much current is needed and maybe avoid it in the first place.

[Brutte]

Being strict that is not a virtual ground splitter. Rail splitter is the thing that splits unregulated supply in half (be it 3V or 30V) while your circuit does not have such requirements as it is a fixed 24V.
So you only need a 12V source and 12V sink regulators.
Depending on your requirements on load, line stabilization, power consumption and price, that can span from resistor divider, 12V zeners, ... , up to a DC/DC switchers.

If that is for powering sth that can push and pull from GND then I suggest two 12V zeners in miniMELF. Under full load that is 12V*0.04A<0.5W

[David Hess]

I prefer to add a bipolar current buffer but the circuit shown here on the right work well:

https://e2e.ti.com/blogs_/archives/b/thesignal/archive/2013/03/26/paralleling-op-amps-is-it-possible

[IvoS]

Just tried what Kleinstein described, which is the same as Intersil note posted by Hero999. I burned one OPA2134    replaced with OP275 and it can deliver plenty of current, more than I need. 110mA sourced and 80mA sinked current.
However, the voltage is dropping with increasing current. It is not rock solid.  There was like 12mV drop.
I searched Mouser database and discovered LM7171 which is single high speed OpAmp.
http://www.ti.com/lit/ds/symlink/lm7171.pdf
I don't need high speed but it can do up to 100mA. I think I will abandon all this paralleling idea and go with single OpAmp instead. I need only 2 of them for my application, so price is not an issue.

[Benta]

LM7171 is a good find.
Enough output current, voltage feedback and unity-gain stable. Go for it.