Not a voltage regulator but a power BJT PNP. Around 100mA go through the bridge, a low 1W resistor from collector to GND alleviates the load on the transistor (around 100mA). MCU has its own power supply though, a single Li-Ion grounded on the lifted ground, so its current doesn't flow through the bridge. The rest of the system runs on 3 batteries in series, the first of which is used as a voltage reference for the ground lift.
Here is a simulation of the simplified idea.
Hard for me to envision quite yet. Maybe you could label dac ground, output stage ground, and 1W resistor on the simulation or on a simplified schematic? Might help some. Reason I'm asking is because other people have considered doing something like that before. Not sure if you have done it before or not? Usually a careful analysis shows that even though the intent is to do good and simplify the output stage by removing DC offset by offsetting one ground, that in practice it can create more problems than its helping. So it just means if probably a good idea to really think it through and see what kind of errors it might produce. To that end it would help a lot to know more about the exact offset circuit and how it is bypassed to the non-offset ground at RF frequencies, the effective output impedance of the BJT PNP verses frequency, its transient response, etc.
Please do not split grounds unless you absolutely need to. It's just asking for trouble. Which apparently you have got.
Refer to materials like this for more info:
I recommend Phils's mixed signal PCB course:
Phil also makes lots of free tutorials on his channel including audio and MCU combo.
Refer to materials like this for more info:
I recommend Phils's mixed signal PCB course:
Phil also makes lots of free tutorials on his channel including audio and MCU combo.