LM317/LM337 configurations for rail splitter

[dcf1007]

I am designing a rail splitter to power a high precision board with low current requirements (10-15 mA). Initially I thought about building an opamp-based rail plitter, but then I realized I needed to clean and pre-filter the DC anyhow and opted for Linear Regulators instead.

The main 12V DC power comes from an AC transformer and I clearly see some AC leaking through with the oscilloscope as well as a bit of ripple left.

Right now my question is not about how to filter the ripple or the AC, but rather in which configuration the LM317/LM337 are more stable for the task in hand.

The LM317 is a positive voltage regulator, Vin=12V, Vadj=1.25V,  12V>Vout>0V. The LM337 is a negative voltage regulator, so Vin=-12V, Vadj=-1.25V, 0V>Vout>-12V

By setting LM317 Vout=6 and LM337 Vout=-6, they are effectively pointing to the same voltage, giving us a regulated midpoint (VGND).

I am aware that the LM3x7 can be either regulated with an external voltage in the Vadj pin (in this case, voltage divider between V+ and GND) or with a resistor pair, the first one creating a loop between Vout and Vadj pins and the second resistor between Vadj and GND.

I have found a couple of places where they used the LM317/LM337 to do exactly this using the voltage dividier/external voltage in the Vadj pin here and here but none with the traditional feedback loop of the LM3x7.

I have simulated both circuits with the bare minimum amount of components to make them work (it will be beefed up later with protection mechanisms and noise reduction later), and both seem to behave the same way in simulations.

My question is, which one would be more stable or recommended to continue with the design? Or will both work as intended with similar performance?

Thank you in advance!

[magic]

What's wrong with using one 15V preregulator, then a 6:6:3 divider, filtering capacitors and opamps outputting +6V and +12 rails?

Ripple rejection will be good (particularly at 100Hz), and internal current limiting will protect them sufficiently at such low voltage.

[dcf1007]

My main concern is actually some 50Hz full wave AC which leaks into the system and that I will have to get rid at all costs. But the RC filters, bypass capacitors etc etc will come next. Right now I am rather focused on the basic design.

I did think about both opamp design and opamp + transistors to increase the output current, or simply using a power opamp, but I think the linear regulators provide more flexibility and simplicity to the design while giving enough current supply, and they seem to already fulfill my needs.

Another reason is that the high precision board which should be powered with this circuit has some capacitive loads which could throw off more easily the opamp design than the linear regulators. It also seems to be easier to obtain a solid mid-point that works well with asymmetric loads.

[Peabody]

I'm so glad you asked that, because now I don't feel like such an idiot for thinking the same thing.  Not just real world regulators, but real world resistors as well.  In general, what prevents lots of current being sourced from the 317, all of which is sunk by the 337, but ground stays the same?  Ok, maybe I am an idiot.  I probably just need to understand how these regulators work.

[dcf1007]

With real world regulators, what determines the standing current through the 1 ohm resistors?

The 1ohm is used to improve the current sensing between both regulators and limits the current flow in the occasion of an unbalance.

I'm so glad you asked that, because now I don't feel like such an idiot for thinking the same thing.  Not just real world regulators, but real world resistors as well.  In general, what prevents lots of current being sourced from the 317, all of which is sunk by the 337, but ground stays the same?  Ok, maybe I am an idiot.  I probably just need to understand how these regulators work.

It has been explained here in this forum before  : https://www.eevblog.com/forum/beginners/virtual-ground-circuit/

From what I've seen in the two links I mentioned before (here and here), people using similar approaches with LM3x7 seem to claim a robust operation.

My main doubt is wether the design with the voltage divider between Vin and GND would be more stable than the one with the feedback loop between Vout and Vadj or vice-versa.

I am open to other design ideas if you think they can keep the simplicity, provide a more stable mid-point and deliver between about 50 mA of asymmetric loads.
The board right now has a demand of 10-15 mA but could go up to 50 mA. I don't think the load would ever exceed 100 mA in any of the board revisions/expansions.

[porter]

I have a Power Designs TP340. I was looking at the schematic yesterday and noticed the control circuits have a rail splitter using a zerner diode. Not for you probably, but it is somewhat apropos.

[dcf1007]

I have a Power Designs TP340. I was looking at the schematic yesterday and noticed the control circuits have a rail splitter using a zerner diode. Not for you probably, but it is somewhat apropos.

Actually, in the design using the voltage divider, you could replace R2 for either a 2.5V Zenner diode or an LM336 voltage reference .

I am just wondering which of the two schematics would be more stable. And I was hoping anyone with more experience working with LM3x7 would shine more light on it. Then I will add the rest of the circuitry and build a prototype to try its stability in real life

[dcf1007]

With real world regulators, what determines the standing current through the 1 ohm resistors?

It has been explained here in this forum before  : https://www.eevblog.com/forum/beginners/virtual-ground-circuit/

Yes, where the possible standing current due to regulator voltage tolerance is estimated at 100mA.  Is that really acceptable?

Yes because the connected circuit board draws 10-15 mA and I don't think it will ever reach 50 mA.

[gbaddeley]

If you have a single AC available and low current requirement, I suggest using 2 half wave rectifiers, one for LM317 +ve rail and one for LM337 -ve rail. No need for any rail splitting circuitry or any of its concerns.

[magic]

This still doesn't solve PSRR, although it does give more voltage headroom.
So you could, for example, cascade LM7xL12 → LM7xL09 → LM7xL06 on each rail.

[dcf1007]

If you have a single AC available and low current requirement, I suggest using 2 half wave rectifiers, one for LM317 +ve rail and one for LM337 -ve rail. No need for any rail splitting circuitry or any of its concerns.

My power supply is 12V DC single rail and I can't change that sadly. I can't add a new power supply.

[magic]

This actually simplifies things because it kills any option of active ripple filtering if you need ±6V outputs from that. A passive filter it is, followed by two resistors and one opamp. Maybe with another few passives for stability, if it is to drive a capacitive load.

[dcf1007]

I appreciate all your concerns and pointing out all the possible flaws of this design with lm3x7, but sadly none of them help me with my original question.

Given the 2 schematics attached in my original post, one with voltage divider between 12 and gnd, and the other with the feedback loop from vout to vadj, what I really wanted to know is which of the schematics will work as intended or be more stable (with the limitations and flaws this design has) , because although the feedback loop is the original intended way to connect them, I haven't found a single design like that online or comments about it. I am interested in the real world functional differences of both connection schemes

Can anyone shed some light on that?  :'(

[dcf1007]

This actually simplifies things because it kills any option of active ripple filtering if you need ±6V outputs from that. A passive filter it is, followed by two resistors and one opamp. Maybe with another few passives for stability, if it is to drive a capacitive load.

That was my original thought. Then I thought about combining the rc filters and the bypass capacitors with a linear voltage regulator for even cleaner and more stable DC, and after that it's when I thought about just splitting the rail with the linear regulators. Then I found the sources proving it was actually a viable option and started designing.

[magic]

If 12V is made ripple-free then 6V also becomes so automatically, so 6V is solved simply by solving 12V.

If 12V is long term stable then all you need is an accurate 50:50 voltage divider. If it isn't, then 6V will inevitably drift with respect to 12V, or with respect to 0V if you make it track 12V, or with respect to both - there is nothing you can do unless you only care about stability with respect to one input rail.

As for the two circuits, why can't you analyze or test them yourself? See for example what happens to them with input voltage at 9V.