Vince,
What did you have in mind for the SOZ changes?
I was looking at the critter last fall, trying to figure out a way to reduce the waste heat so I could justify building one. Heat sinks are next to impossible to come by around here (this is why I'm sitting next to a water-cooled Aleph 2 cunningly disguised as an octopus--still working on it--early indications are good, although it won't be practical for most people--I'm not yet sure it'll be practical for me, for that matter), so I was trying to figure ways around the problem.
I offer the following:
1) Current source instead of the resistor array at the bottom. Current sources ain't that hard to do. No, this isn't exactly the most electrifying brainstorm, but it would help.
2) Here's one no one else has mentioned that I've seen--asymmetrical rails. Say you were going for a 15V rail on top. Try a 5 or 10V rail (with current source) on bottom. The gates are tied to ground, and the bias would adjust itself voltage-wise to produce the same offset at the source. The savings heat-wise would be in the reduced rail. Current would be the same, but you'd end up with less watts, since the VI product would be lower. This one will depend on what voltages you intend to swing.
3) All right, fasten your seatbelt, this one's sick...a differential Aleph. Not with a full Aleph up each side of the differential--that's way too easy. Attack it at the current source. Make the differential current source variable (ala Aleph) with a sense resistor in series with the load. Alternate hookup--sense network reads the incoming signal voltage. I prefer the output network idea, as it affords a degree of feedback as regards to what the speaker is doing. This is still in the thought experiment stage; it may take an absolute value (i.e. bridge rectifier) circuit to set the bias current properly, but that wouldn't be all that difficult to implement.
The objection: But, Grey, if the bias is bouncing around, won't that show up in the signal?
The answer: Who cares? It's common mode. It'll cancel out, in the same manner that distortion cancels in Nelson's X topology, or in any differential, for that matter.
If there's no signal, drop the quiescent bias current back to, say, half. If there's a signal, jack it up. Dissipation is cut, yet operation remains push-pull class A.
There are a few variations that I've come up with, but they're pretty obvious once you start thinking in terms of a variable current source for an output differential.
4) Okay, okay...but for the masochist, only. Variable current source at the bottom, as per 3, but with current sources up top, as well, also driven by the varying current scheme so that they wouldn't try to supply/draw (depending on whether you're thinking current goes from + to - or vice versa; also dependent on whether the output devices are N or P channel) more current than the 'downstairs' current source wants to supply/draw. While you're at it, make the top current sources do the Aleph thing, too; or call it a cascode...I don't care. This topology might be suitable only in a parallel output configuration, as I think the damping factor would be fairly low. Gang enough of these arrays, and you'd have plenty of power in addition to getting a decent damping factor.
5) For that matter, there's nothing to stop you from paralleling normal SOZ arrays, ditto with any of the above ideas. Individual device dissipation comes down. Power as high as you want. Distortion and noise lower without all that pesky feedback.
What do you say, Vince (& Petter, et. al.)?
Grey