Designing a low-distortion audio output stage - Part 1: Introduction, the problem with push-pull outputs
This material originally appeared as one long article in Linear Audio Volume 2, published in September 2011. Linear Audio is a book-format audio magazine published half-yearly by Jan Didden.
These articles introduce a systematic approach to the design of very low distortion push-pull output stage for audio power amplifiers. Work on this has been going on (well, on and off, to tell the truth) since February 2nd, 1981, according to my notebooks.
The results presented here focus on algebraic analysis and simulation. But a great deal was learned from producing working hardware especially in the initial years. Since this material first appeared in Linear Audio magazine, further feedback – or, rather, error correction! – has thrown additional light on implementation issues, and I've referred to this in the text where it's helpful.
Push-me, pull-you: the core problems presented by push-pull output stages
Conventional push-pull output stages (the current-delivering, unity-gain back-ends of the vast majority of linear amplifier topologies at any power level) aren't very linear. What does that mean? It means that under load (i.e., when significant current of either polarity flows at the output terminal), an error voltage is created between input and output. This error voltage isn't generally related linearly to either the output voltage or the output current, and represents a significant source of non-linearity and of the consequent distortion to signals. Cordell's excellent recent book  has a comprehensive and contemporary study.
So, one goal of power stage design is to create a high output-current circuit that delivers high linearity before the soothing balm of global negative feedback (NFB) is applied. Poor linearity in the output stage is typically the driving factor for the choice of how much NFB to use.
Now, I'm not a "feedback denier," but I like to see it used to make an excellent amplifier 'excellenter' in some way, not to make a marginal amplifier meet a quantitative specification that may or may not have any impact on the sound quality. So I like to begin with an analysis of circuit behaviour that shows where good linearity performance comes from, and quantifies its sensitivity to circuit parameters. This analytical rigour isn't found often enough in audio amplifier engineering discourse, in my view.
The basic linearity issue stems from the imperfect handover between the pushing and pulling halves of the circuit. At some point, this handover turns into a rout, and the current flowing in one of the halves falls to zero. This cessation of correct operation causes dynamic difficulties when the stage needs to 'get going' again as the direction of demanded current changes. Significant distortion and HF stability issues ensue, and it's generally considered a Bad Thing if the current in one half of a push-pull output stage falls to zero.
So the second issue we must contend with is the 'switching distortion' problem. This is the bundle of dynamic effects generated by the cessation of current flow in one half of the push-pull stage.
Linearity and switching effects are bracketed together under the broadly used term "crossover distortion," though several mechanisms cause the difficulties. Margan concentrates on the subtle consequences of switching distortion . It's a goal of this present work to eliminate entirely the performance-limiting behaviour that arises from these mechanisms.
Note that just setting a finite quiescent current (that's the standing current in the output devices at zero signal level) doesn't ensure that the current in one half of the push-pull stage won't fall to zero if you pull enough current out of the other side. In all 'classical' Class AB output stages, the current in one half is certain to fall to zero if you pull enough current out of the other side.
The uneasy truce between the pushing and pulling halves also results in uncertainties in circuit bias currents. The temperature dependencies of multiple devices of varying technologies and power levels make for a thermal stability challenge.
Thank goodness that the affluent consumers that purchase high quality audio amplifiers operate them in centrally-heated (or cooled) living environments. High output current linear amplifiers that need to run over the entire industrial (or even military) temperature range, need a more rigorous and disciplined approach to setting and maintaining critical biasing levels.