[KT88]

A very cool implementation of a virtual-GND circuit.
 
As for the latest circuit in goldpoint's post, I see a couple of issues with it:
A. The zener current is only about 0.8mA, and it probably need at least a couple of mA to operate with the nominal voltage across it.
B. The zener voltage can be in the +-5% range, and the internal reference of the LM317 and LM337 can be 1.2V-1.3V which is a +-4% compared to the 1.25V nominal value. That means there could be a difference of up to 225mV between the zener voltage and the reference of the LM317+LM337. So that's up to 225mV across the resistors at the output which yields a current of <=225/(2*R)mA (assuming there's sufficient current in the zener to keep its voltage in specs - no enough current will force a higher voltage across these resistors and therefore more wasted current). The biggest issue is what would happen if the zener voltage is high, and both regulators have a low reference so that Vzener>(Vref317+|Vref337|) - this would make the LM317 output voltage be lower than that of the LM337, and the circuit won't work as its supposed to work. So the designer/builder must make sure this doesn't happen.
Some current at the output is actually a good thing, since the LM317 requires up to 12mA of current at the output to maintain regulation (this is the maximum value, nominal value is about 3.5mA). However since the current will depend on the exact value of the reference voltages it can't be trusted.
C. A different way to look at this circuit is as a push-pull op-amp capable of sourcing/sinking >1.5A (at least that one of ways I see it ). The LM317 sources current, while the LM337 sinks current. As long as you maintain a minimum current at the output of both regulators its basically a class-A/AB buffer with the zener diode trying to compensate for the offset of ~1.25V from the A to O pins of the regulators (1.25V for each regulator). So perhaps instead of trying to use voltage regulators and compensate for the voltage variations which yields great variations in the DC current, it would be best to implement a class-A or AB (probably better) follower with the input generated using a simple resistive voltage divider between the rails as its usually done. This is what most op-amp circuits do, so using an op-amp can work fine, its just a matter of finding one with sufficient current drive capability, and low cross-over distortion - most op-amps now-days actually keep the output transistors operating with a small current at all times, but this current is usually very small to minimize power consumption, and therefore some cross-over distortion can still be noticed.
D. Another point I want to mention is that I think the virtual-GND circuit should be tailored to the amplifier in use. Some amplifiers have a very different PSRR between the two supplies, so in some cases it might be better to implement a "simple" virtual-GND circuit that regulates the virtual-GND with respect to one supply rail or the other (which one would depend on the amplifier used).

 

[KimLaroux]

Thanks for the inputs KT88.
 
I think you have a point with the zener needing more current to operate properly. It's obvious the zener isn't working properly in the last test because it only has 1.5 V across it while it should be 2.5 V. The voltage divider has less than 1 mA going trough it with a 18 V input voltage. Goldpoint, could you lower the value of the 10K resistors and see what happens? Or just raise the input voltage?
 
I also agree that the quiescent current has it's advantage : It takes care of the regulator's minimum load current to maintain regulation. 10 mA should be enough, though.
 
As for everything else... I am not marketing this design as a proper solution for a virtual ground. For me it's nothing more than a novelty hack, an entertaining exercise. 

 Anyone who's doomed to use a virtual ground should look for more elegant, properly designed solutions.
 
Oh and yeah, my math was right last night : You really have a volt between each outputs :
 
U = RI = 20 * 54E-3 = 1.08 V
0.505 + 0.505 = 1.01 V
 
Close enough.
 
More math tells us there's room for improvement :
 
0.505 - (-0.777) = 1.282 V
-0.505 - 0.766 = 1.271
 
Which are the Vref for each regulators, within their published range. So even here, the difference between the adjust pin and the output is equal to Vref. Fixing the problem with the zener not operating properly should prove interesting.

 

[Sonic Wonder]

.

 ​

 ​

 

[Sonic Wonder]

.

 

[Sonic Wonder]

.

 

[Sonic Wonder]

.

 

[KimLaroux]

Looking at the datasheet for the zener you use, it says it's forward voltage is 2.5 V at 20 mA. This seems like a lot of current. I didn't know low voltage zeners needed such high current to maintain their rated voltage.
 
Just looked at the other offerings on Digikey, and they all need 20 mA. How inconvenient.
 
One cheap fix would be to use a higher voltage zener, like 2.7, 2.8 or 3.0 V.
 
The problem I now see with my design can be explained using this drawing from the datasheet:
 

 
As you can see, the forward voltage is dependent on forward current. In other words, it's not a fixed value. This is verified by your latest series of tests. When I thought about using a zener for this application, I took for granted that it would be 2.5 V regardless of the current flowing through it. The current flowing through the zener is the current flowing through the voltage divider, which is dependent on the input voltage. Not good. The point of the design was to make it work with a wide range of input voltages. It turns out that using a zener inside a voltage divider is not a solution.

 

[Sonic Wonder]

.

 

[KimLaroux]

You could probably do it with a pair of Op-Amp. But if you're gonna use op-amps, you may as well use discreet transistors like in Wakibaki's design. You would not need regulators anymore, since the op-amps will do the job. And you'll free yourself of the annoying reference voltage.
 
And it'll be so much more elegant. 

 
I have a question regarding your original design using fixed regs. Do you power an amplifier directly from it, or do you add regulators after it to regulate each rails separately? 

 

[wakibaki]

I had one of the opamps upside down in the original drawing...
 

 
You should be able to see (I think), if you expand the picture, that the legend in the bottom left hand of the screen shows 498.301nA. This is the current in the ballast resistor. The offset voltage (from midpoint) is ~4mV. This takes no account of opamp input offset voltage, however, but the error due to variation in regulator voltage references and consequent quiescent current is likely to be considerably higher than the opamp circuit error. 
 
Of course the total quiescent current is higher than this. There is the current down through the resistor divider to account for, and there is the current drawn by the opamps.
 
The quiescent current can be minimised by picking a low-power opamp such as the LM2904 (0.7mA) and increasing the divider resistors to (say) 100k. The LM2904 will put out 40mA, so if you pick transistors with a gain of 100, we're talking 4A source/sink capability.
 
All-in-all it can be made to outperform the regulator circuit by a considerable factor. It's probably cheaper too.
 
w

 

[Sonic Wonder]

Kim,
 
The headphone amplifier is powered directly from it. Some of the things I really like about the circuit I use (we're speaking of the 3 regulator "Regulated Virtual Ground" circuit at the beginning of page 1 of this thread) is that the TO-220 devices do not need any heat sinks. The circuit is rugged, dependable, inexpensive, and even fairly simple. But mainly, the ground is SOLID = unbudgeable. Using that circuit will wow you when you hear the lower/lowest octaves.

 

[KT88]

The current solution of using a zener + regulators has the very big problem of being affected by variations (I work mainly on IC circuit design, so variations is always in the back of my head ). The current design requires a manual adjustment of the zener reference voltage to fit the reference of the generator. As I've stated earlier, there's a 50% chance the circuit won't work at all as expected (if Vzener>Vref317+Vref337). KilLaroux, because of that its impossible to use a higher voltage zener without modifications to the circuit. Using higher resistors will minimize the maximum current, but will affect output impedance at lower frequencies + wont solve the issues involved with the variations such as minimum load current required and maintaining Vzener low enough. A possible solution is to use parts with tighter tolerances. The zener could be replaced with a precision band-gap IC (the additional noise that band-gap has over buried-zeners is not of much importance due to the internal band-gap of the regulators which adds lots of noise already). These IC's also need less current than a zener diode. The regulators could be replaced with a more modern equivalents with tighter tolerance as well (and lower minimum load current while were at it), this way it is possible to have a lower offset voltage between the output of both regulators which will allow lowering the value of the resistors and minimizing current flow at the same time.

Keep in mind there's also the issue of the tempco of the parts. The band-gap has a very low tempco (the exact value will depend on the order and quality of the band-gap used in the regulators), while the zeners have a very significant tempco (and of course they aren't at the same temperature which is an issue as well).

If you really want to stick to a zener diode I think some modifications should be made, I would suggest replacing it with a higher voltage zener (~5-6V zeners usually have the lowest tempco due to the combination of two different breakdown mechanisms at work at these voltages), this will also allow using a somewhat lower current in the zener. To handle the issue of increased voltage compared to the reference of the regulators (which is a big issue as I've posted earlier) it is possible to put a resistive voltage divider in parallel to the zener to drop its voltage from ~5-6V to ~2.5V (1K-1.5K-1K will work for ~6V). This voltage divider could have a potentiometer in it to allow adjustment for variations in the zener voltage and the reference voltage of the regulators (IMO this is actually a good thing to have if using a zener):
 

The trimmer allows the user to set the voltage difference between the two regulators (a capacitor could be added in parallel to the trimmer for improvef noise performance - adding a cap in parallel to the zener is a waste of capacitance due to the low incremental resistance of the zener that requires a huge cap), and as long as the tempco of the parts used is low enough it'll work with minimal change in the DC current in the resistors.
 
That's just my 2 cents

 

[Sonic Wonder]

.

 

[KT88]

Are you sure the diodes are in the correct polarity? Looks like they are connected backward.
 
Also, it will probably be best not to use the pot in series with the diodes. This means the voltage divider is (3*rd+Rpot)/(3*rd+Rpot+20K). If the pot is at ~1K this means there's about 5% transfer from the supply voltage to the voltage between the two adjust pins. This will make the circuit behave very differently with changes in the power supply. Putting the voltage divider in parallel to the low resistance part (diodes or zeners) as in the picture in my last post will eliminate this issue.
 

 

[Sonic Wonder]

Hello KT88,
 
<>  What we're trying to do is create a simple, inexpensive "self adjusting" virtual ground which has very low quiescent current - for use with batteries.
 
<>  Self adjusting - so that a wide range of battery voltages could be used with it.
 
<>  The virtual ground can be perhaps 0.1V or so above or below the exact "1/2 rail-to-rail voltage".
 
<>  It seems using the readily available, inexpensive LM317/337 complimentary regulators is one good place to start from - high precision is not a goal on this...