I’m always interested in boundary conditions ... what happens when the amp leaves its happy place and a surge of current or voltage is required. Does a circuit saturate and hit the wall? Does a transistor fail? Does it store charge and "stick" for a few milliseconds? How smooth is the transition in and out of the Bad Place?
I mention this because speakers are badly behaved much of the time. They store energy for tens to hundreds of milliseconds, then throw it back to the amplifier. The feedback network might, or might not, keep correcting this, but the error overshoots can be very large and can saturate an input section.
Many power amps do not accept boundary conditions gracefully. Not just the output section, but the driver as well. Driver transistors fail when SOA is exceeded by transient reactive loads (failure to accurately read a SOA graph almost bankrupted Audionics). In tube amps, drivers can’t push enough linear current into the Miller capacitance of the output tubes. The voltage-amp section of a transistor amp can’t charge the dominant-pole capacitor fast enough, resulting in slewing.
These are all boundary conditions, and they are audible not just when they reach 100% failure, but well before that, when nonlinearity just begins. The previous point about Class A operation in a differential stage still holds: what happens when more than 100% of the current programmed in a current source is exceeded?
This is a boundary condition problem. When current is exceeded, what next? What’s after that? Does anything fail? What does current clipping look like? Are there any energy storage mechanisms that result in "sticking", a well-known problem in solid-state power stages. If sticking happens, how long does it take before it gets unstuck? In milliseconds?
The approach in the Blackbird/Karna does not use current sources, nor differential stages. Each side is parallel, but in antiphase, and all phase splitting, and re-summing, is done by passive devices, which do not have slew limitations. The 1:1 interstage transformer makes sure that recovery from A2 grid-current events happen in microseconds, not milliseconds.
Overload happens in the tubes, mostly in the output section, and the overload condition is not affected by local or global feedback, so the overall boundary characteristic is that of a (very) fast-recovery limiter/compressor. There is no hard boundary between Class A, where it remains most of the time, and A2, AB, or AB2, depending on current or voltage demand.
During the development of the Karna amp, I was in a kind of perverse mood, so I was curious just how much abuse the circuit, and the tubes, could take. I set the oscillator level so the scope display was just below clipping, around 20 watts, and the 8-ohm test load was nice and warm. I increased the drive frequency beyond 20 kHz, and as transformer gain started to fall off beyond 50 kHz, I just increased the input level to keep the output at a steady, undistorted 20 watts. Because why not?
I finally lost my nerve at 500 kHz. The scope display was still an undistorted 20 watts, with no sign of triangle waves or flat-topping, but playing around with an AM-band transmitter (with 500 volts inside) was asking for trouble. I wasn’t trying to kill anything, but sooner or later some part was going to fail. (If any of you customers try this stunt, yes, we will void the warranty, so don’t do this. Ever.)
Not many transistor amps would survive full power at 500 kHz. Some would, some wouldn’t. It’s an absurd test, with no relation to audio use. But it’s interesting to know the development prototype survived it. No, I would never do this to the current production model, and don’t you guys try it, either.
