Is this kind of class D done before?

[wwenze]

The typical class D has a switch for the output stage which connects the output to V+ / V-. So low V*I loss but the PSRR suffers without global feedback.

How about if we don't set output = V+ / V-, instead we have some local feedback which sets output = input (driver's output) times gain which gives a value close to the V+/V- but not touching it. So we trade some loss in efficiency for a more stable output voltage.

 

[wwenze]

Actually come to think about it, this would give an efficiency similar to class H, but something has to be switched either way; in class H the power supply is switched, while in class D the output is switched. I guess it's better to switch the power supply instead and then use the class AB with its better PSRR and lots of other things.

 

[basreflex]

this first idea looks like the labgruppen 10k and 14k amps, where the powersupply is a crude classD swinging a few volts above the needes signal, and the actual amp is a linear one;