The hacks made to the prototype allowed it to function for testing but the fans limited the input voltage range to less than 14 volts. I wanted a simple & efficient method for regulating the inverter's auxiliary supply. The sine section would need this too.
Since the redesign settled on the inverter operating as a DC transformer the option of a control loop was out. The discrete LDO I used in the start circuit wasn't suitable for delivering 300+mA so I began to look for off-the-shelf solutions. I looked at linear LDOs but their dissipation was too high. Their temperature derating was also a problem. I considered building my own buck/boost - more time consuming than I wanted. I eventually settled on a 6W isolated monolithic switching supply from Mean Well. It has current limiting, short circuit protection, decent transient response, acceptable noise margin, and provides a stable 12V from 9 to 18 volts.
I purchased one for evaluation and characterized it with an electronic load. It performed to spec well with only one anomaly. When running under greater than half load and applying a short, the unit would not restart with the short removed. I wound up contacting the OEM and confirmed it was working as designed. I was a bit surprised to hear that since when testing my own designs I consider that a conditional stability problem. But, this unit was nearly half the cost of the competitor's unit and I didn't anticipate loads or transient conditions that would make this a problem even when considering temperature derating (incidentally, I did test the other unit and it didn't have the problem).
I decided to use the Mean Well and added to the prototype.
Without fans to power, the sine section would have very modest power requirements (100mA max). A switching regulator would be a waste in both cost & space. A linear LDO regulator like the MC7812 is cheap, small, and wouldn't dissipate more than 1W. To provide isolation the redesign will use an auxiliary transformer winding with one additional turn than the primary. This ensures that even at low input voltages the LDO will have several volts of margin to provide a stable 12V.
I didn't make this modification to the prototype since it is basically using this design; the only change is to an LDO that can handle higher input voltages.