An amplifier with an open-loop distortion of 1% and an open-loop gain of 100,000 has a closed loop distortion of almost nothing when negative feedback reduces the gain to 20.

That's only harmonic distortion. I don't want to get into a 'better amplifier' argument, but there are other considerations.

TIM, slew rate limiting, and so on. A feedback loop takes a finite time to exercise control. Moreover there has to be an error or there is no control signal. Any signal in a feedback loop cannot by definition be perfect. One of those catch22s.
When the signal is crossing zero there is (or should be) zero output. Therefore there is (or should be) zero feedback!

 

First, there is no crossover distortion -- at least I see none on the scope even under a full load and maximum gain.

I don't see how increasing the 100 ohm resistor will help. It was actually found experimentally as the maximum value resistance for a minumum current required to drive the Class AB amp.

A Darlington pair was chosen in order to increase the input resistance so that a weak voltage amp could drive it. I don't think there's any way to get around the need for Darlington or quasiDarlington configurations in high power amps (not that my 2 Watt amp is high power).

 

I almost forgot: I understand the need for feedback if I am to use discreet components throughout the amplifier (voltage gain, driver, power output). I wrestled with noise and feedback for weeks without finding any success. I've not given up though. This amplifier we are speeking of is made in modules 1. Voltage Amp 2 Power Amp 3 Speaker with zobel circuit. I plan on using these to study feedback. The real test comes when I use an unregulated supply and a big enough current draw to get a substancial amount of ripple on my supply. That problem was driving me nuts!

 

I haven't got around to testing this design out, it is on my list of to do's.

Medium inpedance in (600Ω), low impedance out. Should work OK for DC applications too.

Don't forget the Sziklai Pair configuration, it is supposed to be better for some audio applications. Me, not a clue either way.

 

I've been looking at your circuit and I'm not suprised it has a non linear response.

The 33μ capacitor coupling the driver to the output emitter followers has impedances

50Ω @ 100Hz
5Ω @ 1000Hz
.5Ω @ 10Khz

Compare this with the 100Ω resistor in the driver and you will see that they form a significant potential divider at mid to low audio frequencies.

Thus any non pure sinusoidal signal will have its harmonics amplified differently. This will appear as inadequate low frequency rsponse.

It is normal to connect the driver as a common emitter, not common collector (emitter follower) as you have done. This also makes DC biasing possible, eliminating the capacitor altogether.

A further result of your arrangement is that the output is in phase with the input, since both stages are CC. This will generate difficulties if you come to consider applying feedback.