some very sensitive instruments requiring very low-noise power use batteries as their raw power input during the time they're making measurements. Using battery inputs doesn't require that the low-noise regulators exhibit high PSRR, line regulation, etc.

It's easy to assume that 12-volt battery input implies that it's an automotive application. I didn't see anything in the original query that suggests it's an automotive application. I did see that it requires it be a low-noise regulation circuit, however. If the situation is one with a 12-volt battery that's being charged when power is on, and which responds to load, etc, that's changing, then, as Erik points out, the actual voltage is likely to vary widely, thereby suggesting the use of a SEPIC switching supply.

However, if it's a highly sensitive instrument supply requiring a low-noise (total noise amplitude <.05%) supply it's a different problem. The O/P's proposed circuit has one serious weakness, relative to his own statements, and that lies within the fact he expects the output of a regulator to provide GND. In general, positive regulators themselves can source current, but not sink it. Negative regulators can sink current, but not source it. Additionally, the fact that one of the supplies is 3v3 implies that there is a requirement for logic, and logic is noisy. It would be desirable to have a bit of isolation between the bipolar 5-volt supply and the 3v3 logic supply.

The GND should be capable of both sinking and sourcing current. after all, it has to absorb the current spikes normally absorbed by bypass members. Consequently, the GND must be relatively low-impedance, and generated from a push-pull source.

If a power op-amp is used to generate pseudo-GND from the 12-volt input, then linear regulators, both positive and negative can be used, with the caveat that the 12-volt input must be just that, a 12-volt input, within pre-specified limits, and the remaining circuitry must be designed with those limits in mind. If, as Erik suggests, the battery is, in fact, just an input filter, not unlike a capacitor, with a charging circuit always involved, then, yes, indeed, a SEPIC switcher would be the reasonable solution. However, one should consider that the switcher works by running AC current through an inductor, which inherently creates ripple, and, generally, more of it at low currents than at high. Using the power op-amp to provide pseudo-GND allows one to select whatever potential with respect the negative rail that one chooses. The negative rail can then be generated with yet another op-amp with an appropriate reference, as can the two positive rails. None of this is particularly simple, as noise is amplified within each such stage, and with the use of a pseudo-GND, can affect all the supplies, and not just the one generating it. Texas instruments offers TLE2426, which will split the rails, but having 6-volt rails as inputs doesn't solve the problem as it was stated. The power op-amp would do that, but also has its weaknesses.

Kai has often pointed out that he uses switchers as pre-regulators for his LDO's, which he carefully filters with ferrites and carefully chosen bypass members.

Without considerably more target application information as well as battery characteristics, it's hard to come up with a meaningful recommendation.

RE