[Sonic Wonder]

RESULTS:
 
I got lazy and simplified the circuit again (practical considerations).
The Voltage reference is an LM336Z. (KT88's suggestion)
 
You can see results here similar to those obtained earlier using 4ea 1N4148 diodes in series to create a 2.69V voltage drop. The main difference is that the voltage divider is programmed for just about 1mA of current with the 8K resistors.
 
a)  With no adjustment trimpot:
 

 ​

 
 
b)  With a trimpot to center the virtual ground closely:
 

 ​

 

 ​

Wondering if we would get better results with more current in the voltage divider, I upped it to about 4mA of current with 2.21K resistors. All that did was to use up more current, increasing the voltage drop across the output resistors slightly.
 
I suspect that the different voltages measured at the adjust pins of the regulators are coming from tolerance differences amongst the positive & negative regulators themselves, so using different ones should yield different voltage measurements at the adjust pins.
 
(a) above is the quick and dirty solution.
 
(b) above is if you need/want a closely centered virtual ground point.  [Reduce the value of the associated resistor connected to the trimpot so that it, added to the midpoint value of the trimpot, is equal to the resistance of the fixed resistor at the other end of the voltage divider stack, of course.]
 
Either way, this circuit should make a good battery powered virtual ground, drawing only a few milliamps (2 or 3mA) of quiescent current - while also able to handle 50mA (without heatsinks) - or up to 1.5A per rail with heatsinking the two TO-220 devices.
 
This note is being added a few weeks after posting the above: Further experiments with the two "self-adjusting" virtual ground circuits above show various instabilities. Therefore, use instead the circuits shown in the first article of this thread (on page one of this thread). If anyone works out a good way to use three terminal fixed or adjustable voltage regulators to make a "self-adjusting" "rail-splitter virtual ground, I would like to know about it. Thanks!
 
UPDATE to the comment above. The self-adjusting circuit now included in the first writing of this whole thread on page 1 works just fine! I love it! I had a wiring error in my prototype where I had connected the Adjust pin of the LM336 to its + pin. After removing THAT error, all is well. (So see first page - updated 15 April, 2013 - it's now showing the corrected circuit.) Also shown here:
 

​

 

[wakibaki]

I'm happy to leave it to readers to judge.
 
wakibaki versus goldpoint, same capacitance, ideal voltage source in goldpoint...
 
 

 
 
w

 

[KT88]

You are basically measuring output impedance, the 1R resistors in goldpoint's circuit will make this value very high compared to the 0.1R in the other circuit.
 
BTW, can you measure the VBE of the transistors in the same simulation?

 

[wakibaki]

Quote:

You are basically measuring output impedance, the 1R resistors in goldpoint's circuit will make this value very high compared to the 0.1R in the other circuit.
 

Exactly.
 
w

 

[KT88]

Since you didn't simulate it, I ran a similar simulation of your circuit in ORCAD (I don't use LTspice). I've used a different BJT since I didn't have the model for the 2n3906, but its quite similar.
 
Here's the output waveform:

which is very similar to what you get, so the simulation is fine.
 
Now here is the base-emitter voltage of the transistor (the npn in this case):

It gets to -6V, which is too much for most BJT's.
And this is at 1KHz with a cap at the output for help. If you change the amplitude of the sin-wave source, or use a smaller cap/lower frequency it gets even worse.
 
Here's the same circuit exactly at 100Hz:

It clips to the supply rail, no BJT can survive 20V of revere voltage on its Base-Emitter junction.
 
So in effect, the circuit destroys itself.
And again, the same thing will happen without any load if you just change the polarity of the offset voltage of the op-amps.

 

[wakibaki]

What are you saying, the regulator circuit doesn't have output transistors?
 
w

 

[KT88]

I'm saying the circuit you've posted will destroy itself every single time.
 

 

[wakibaki]

And I'm saying you're categorically wrong when you say that -6V will destroy a 2N3904
 

 
 
w

 

[KT88]

VEBO Emitter-Base Breakdown Voltage 6.0 V 
 
So yes, it'll destroy it. And again, that's at 1KHz, at 100Hz it'll die with the full supply voltage across it.
 

 

[wakibaki]

But not now. Thanks...
 
 
 
 

 
 

 
w
 
 
 
 

 

[KT88]

Exactly, its a very simple fix, but one the must be made before the circuit can really work.
 
Now we can simply move these diodes to the bases, get rid of one of the op-amps, and wrap it all in the feedback of the single op-amp. This will be in effect an op-amp with a discrete class-AB buffer after that. It'll have much better performance than at the moment (the output resistance will much lower than it is now since the 0.1R resistors will be within the loop). Its the straight-forward way to do this. Some people (like tangent in the Pimeta) have used an IC buffer at the output of the op-amp instead of the discrete transistors.
 

 

[KimLaroux]

Quote:

RESULTS:
 
I got lazy and simplified the circuit again (practical considerations).
The Voltage reference is an LM336Z. (KT88's suggestion)
 
You can see results here similar to those obtained earlier using 4ea 1N4148 diodes in series to create a 2.69V voltage drop. The main difference is that the voltage divider is programmed for just about 1mA of current with the 8K resistors.
 
a)  With no adjustment trimpot:
 

​

 
 
b)  With a trimpot to center the virtual ground closely:
 

​

 

 ​

[...]

 
That's interesting. I didn't know such devices even existed. They seem to have a wide range of applications. There's even an example in the datasheet where they use one instead of resistors to adjust an LM317. I'll have to remember those chips...
 
But I think you forgot that the regulators need 3.5 mA to stay stable. They only have 1.2 mA flowing trough them in this test.

 

[wakibaki]

Thanks, KT88, but what took you so long? We could have been here about 60 posts back...
 

 

 
w

 

[Sonic Wonder]

Sour grapes - porque so sour? How very childish some of these comments are! We apparently allow unsupervised, spoiled kids to post on this site, but something should be done about that...
 
The load itself (not shown here) demands more than enough quiescent current for the regulators to be happy..
 
Um, I like testing for sonic qualities - actually listening to circuits - not just saving a few pennies with minimal parts counts or whipping out a theoretical "best circuit". Sketches and "computer simulations" only don't cut it completely. Know what I mean Sparkie and Wobblie?

 

[wakibaki]

Quote:

The load itself (not shown here) demands more than enough quiescent current for the regulators to be happy..