Biasing a DIY current feedback opamp

Question

For a personal project, I'd like to make a discrete current feedback opamp. Here's a simplified schematic:

The question is about how to bias Q3/Q4.

For a nice class-AB output, the sum of both Vbe's should be constant. In a common collector output stage (like Q9/Q10), that's easy enough since both Vbes are on top of each other, all that's needed is a bias voltage generator that follows their temperature drift.

But in the common emitter configuration (Q3/Q4), these Vbe's are quite far away from each other. How to keep their sum constant? The usual solution is to put emitter resistors and run them at high bias current, but that means high idle dissipation and low gain.

There are plenty of integrated rail to rail output opamps. If I remove Q9/Q10 and squint a little, the schematic looks very much like a rail to rail output current feedback opamp.

So how do they do it? How to bias Q3/Q4 so they run in class-AB without the current running away?

Usually datasheets show a magic box, they're not telling about the secret sauce inside...

Answer 1

First attempt at self-answer:

Q20-Q21 make a shunt regulator that biases the output transistors Q18-Q19 (thermal compensation not shown). Q68-69 measure the current through this via R62 and tweak Q70-Q71 to adjust the driver transistors Q3-4 so the average current through the aforementioned shunt regulator remains civilized. C23 should correspond to the thermal time constant of said transistors.

This would not work in an opamp which has to process DC signals.

It works perfectly (ie, Q3-4 are biased as I wanted them) but it also doesn't work at all: the output THD is not better than an EF2 output without further complications.

I'm still interested in how the output stage of a RRO opamp is biased, though.

Answer 2

The general idea is below, the IN+/- are the differential outputs of the input differential stage. Global feedback assures that these two inputs are identical in equilibrium and thus the ground current of OA1 is ~0.

In this case, the quiescent current of OA1 and R4 current keep the output transistor on the Class-AB knive's edge.

The resistor values still must be finetuned for the amplifier to neither succumb into class B mode nor run hot with excessive dissipation. I guess rail-to-rail ICs have more circuitry to prevent the runaway.

^{simulate this circuit – Schematic created using CircuitLab}