My previous post
#200 was autobias with a nonfloating power supply with common source MOSFET. Adding the mirror introduced a 3rd pole to the autobias loop making it tricky to get it stable.
Here's an alternative autobias using the CFP:
This one uses a floating power supply to get voltage gain in the power stage allowing a CMOS input stage. The advantage of the CFP is the extra drivers reduces the drive current to under 1mA which suits the CMOS stage giving an open loop gain of 50dB with 20dB of overall feedback. The advantage of the autobias is the temperature of the drivers
and power transistors is not an issue because the output currents are determined by the D+R network D1,D2,R14,R15) with Q5 and Q6. Now the drivers and power transistors can be on the same heatsink - while D1,D2 are thermally linked to Q5,Q6 on a separate small heatsink for only 2W total at full output power. R13 provides the keep-on current of 18mA in Q1 and Q2 - even when clipping!!
I realise the CFP is not popular for Class-AB, but when the base-emitter resistors R1,R2 are only 22 ohms the turn-off delay is similar to the popular EF output stage. Crossover current remains at 230mA up to 1MHz. Sim'd full power bandwidth in closed loop is 500kHz. This is the best I have seen so far
for a CFP with nonswitching autobias. The drivers do run a bit hotter with 22 ohm resistors (R1,R2) but with autobias their temperature variations is of no consequence
. (With a CFP the drivers need to be rated for rail-rail voltage so higher voltage rails will require higher voltage drivers than used here).
So even with a CFP this autobias loop is pretty much the same stability as my earlier floating supply version like in
Post 150. But better stability than Post 200 - however, its back to a floating power supply. For those wanting a non-floating supply -- maybe adapt Q7,Q8 to be the VAS stage.
Other sim performance: THD 0.02% at 1W with gain symmetry trimmable by R17. Output resistance 0.7 ohms so it can drive 4 ohm dips. Peak current is limited by Q7,Q8 current and Betas of the drivers and power transistors.
The optimum idle current is reduced from about 350mA to 230mA by adding 0.075 ohms to the diodes. At half full power THD is 0.25% and above half-power the soft clipping starts (trimmable by R24). Sim'd distortion is low-order and quite adequate for good audio IMO.