A Classic Audio Amplifier

I have always wanted to design a "classic" audio amplifier. One that is "no compromise", meaning that it has everything it needs, and nothing it doesn't. Due to cost or time constraints, I have designed amplifiers in the past that have been sub-optimal. I remember one in particular: It was a studio amplifier for a radio station. It was about 20 watts, and used the usual differential input, a voltage amplifier, and a current amplifier output stage. But, it was the simplest thing that could possibly work -- there were compromises.

• The diff amp used a simple resistor current source, and it had a simple resistor output stage. 
• The voltage amplifier was connected to the diff amp's non-inverting output; the inverting output was ignored (a common practice).
• The voltage amplifier used a resistor as the current source; it did not have an active current source, or a "bootstrap" to regulate the current.
• The voltage amplifier also did not have a Cdom capacitor to establish a dominant HF pole (for feedback stability).
• The output stage used a couple of internally connected Darlington transistors instead of discrete parts for the emitter follower output.

Despite all this, the amplifier had very low distortion numbers, it sounded fantastic, and it was unconditionally stable. All in all, it was a very successful amplifier. Still, I always wanted to revisit that design, and add the missing elements:

• Add a current source to the diff amp. 
• Add a current mirror to the diff amp (to use the output from both transistors in the long-tailed pair).
• Add a current source to the voltage amp. I prefer a current source over bootstrapping, although I have seen some very good amplifiers use bootstrapping. I just think a current source may be more reliable and less dodgy to analyze.
• Add Cdom to the voltage amp, to establish a dominant HF pole to design around, and set predictable phase margins for feedback.
• Use Sziklai configuration for the current output stage, using discrete parts. 

There are many advantages to the Sziklai configuration that I had not been aware of until I started studying it  recently. One of the huge advantages is that the final output device is inside the feedback loop of the driver (pseudo-emitter follower). In other words, the driver behaves like an emitter-follower, but the output device does all the heavy lifting. This means when the gain increases on the final device due to high temperature, it doesn't cause thermal runaway. The bias network doesn't have to track the final BE drop; only the driver BE drop. Since the driver doesn't heat up nearly as much as the final, bias compensation is much simpler, and almost unconditionally stable. I have an idea for a unique way to track temperature that would be very easy to implement.

The Sziklai configuration does have the potential to oscillate, whereas a Darlington emitter follower doesn't, but that is fairly easy to control, and the thermal stability advantage seems to more than make up for it.

I will be blogging the development of a 40 ~ 60 watt design in the upcoming weeks and months. I'll post a preliminary schematic soon.