Virtual ground before or after low noise LDO?

Question

I have two boards (microphone amplifier and speaker driver) that require +/-15 VDC. The max current required for the boards is 500 mA. Since low-noise dual power supply devices are hard to come by, I'd like to modify my system to accept a single power supply. Ideally, what I'd like to do is design the following chain:

• Input from the AC mains would be via a high-end SMPS (with switching frequency in the MHz range) with an output of 36 VDC.
• A virtual ground would be provided using a TLE2426. A current boosting BUF634 feedback loop to get up to 250 mA as described in https://tangentsoft.net/elec/vgrounds.html would be used. Note that this can be stacked with a second BUF634 to get up to 500 mA. This would provide +/-18VDC rails.
• I would then use LT3090 and LT3045 LDO regulators to further drop the voltages to +/-15VDC rails that feed into the board. The LDO regulators are very high PSRR devices with ultra low noise outputs.

Questions:

• To minimize cross-talk between the microphone amplifier and speaker drivers, each device should have their own power rail splitter and LDO regulators (this is also necessary because of the max current capabilities of of the regulators and buffers). Can they share a common SMPS, or should each have their own?

• Should this setup work well for powering devices that work in the frequency range 100 Hz to 100 kHz (this is for auditory research experiments at a university, so we have to deal with frequencies above the audible range of humans as well)?

• An alternate approach would be to use the TPS7A47 ultra low noise LDO (which can accept up to 36 VDC input). In this case, I could then use the following chain: SMPS -> TPS7A47 -> TLE2426 + BUF634 -> board. However, would this be OK or would there be higher noise in the circuit due to placing the TLE2426 + BUF634 after the LDO?

Answer 1

To clarify, we want to find a good, clean, low-noise power supply for work with our circuits which require +/-12VDC and 500 mA current. We're just not sure what the simplest, most straightforward approach is.

Okay!

The simplest is a linear power supply: IEC mains filter if required, transformer, rectifier bridge, caps, and a regulator. Really oldskool.

For your measurement stuff (and microphone preamp) the best is to have low interwinding capacitance in the transformer which reduces 50Hz leakage current through your grounds. For this the best is split bobbin EI core, which conveniently is also the cheapest... Toroid has less magnetic flux leakage but if you shove the power supply further away on your lab bench from the sensitive bits then flux leakage does not matter.

OK, you say you don't want to work with mains, but it isn't a problem if you do it right, make sure there is no easily exposed copper at high voltage. If you use a PCB mount transformer, just tape a nice thick piece of photocopier transparent on the back of the PCB. I always do this. The point is that component pins do not puncture it because it is thick. So if a finger wanders in the hot zone... it is safe. If you use a chassis mount transformer, use heat shrink. Or put the thing in a box.

"Low noise" power supplies are not necessarily worth the hassle either. For example, if you use a good old LM317 you can expect a few mV ripple on the output, at frequencies where your opamps will have 80dB PSRR or more...

Since you mention TPS7A47: this is a very low noise regulator... but if the load is opamps, ripple rejection matters more than noise. Opamps have high PSRR: 1µV noise or 100µV noise on the supply matters little if the opamp has, say, 60dB PSRR, which is a factor of 1000.

Usually you will have a few critical bits of the circuit which require a very clean supply but draw low current (like the microphone preamp) and other parts of the circuit which will work fine with a less clean supply but drive high current (like the loudspeaker driver).

You could build a very complicated, low noise power supply like you suggest. But it would still have a non-zero output impedance, if you include the wires and everything. High currents drawn from the power amplifier would thus influence its output voltage.

A much simpler option is just stick some RC filters with large capacitors on the supply of the really sensitive circuits. Or a local regulator. If your microphone preamp draws 10mA, just put a 1000µF low-ESR cap and a 100R resistor. You lose one volt... it costs 1€...

Please understand that the whole point of modern fancy regulators like LT3045 is that they are fast. This means they work with low output capacitance, which allows a small footprint, low height solution which you do not care about. This also means they have good HF PSRR which you do care about. However you can achieve the same with a passive filter. These regulators are not magical, just another engineering tradeoff (space, cost, height, etc).

Hm... I kinda went on a tangent here...