Yamaha MX-10000
(1987 - 1988)

Up to 1,2kW output at 1 Ohm, yes : your windows will rattle. And in Class A mind you !
That feat is possible by way of the Hyperbolic Conversion Class A circuit which you can real all about just below thanks to Mirko Essling :

"HCA actually is not HCA, because it is used and implemented differently in the MX-10000 compared to it's smaller MX sisters !

Here is, how I understand it, how and why it is used in the MX-10000. The MX-10000 consists basically of three stages :
a) A small signal voltage amplifier stage, whose purpose is to provide the voltage amplification to the level needed at output. This stage has its own NFB loop.
b) A class AB output buffer stage behind the voltage amplifier, which provides a large voltage swing (roughly +-85V) at the positive loudspeaker terminal.
c) A class A error correction power amplifier, with it's output at the negative speaker terminal, which provides the difference (distortion + voltage drop, max. +-7V or so) between the input of the buffer and it's output.

There is no global feedback loop ! Because of this special asymmetrical bridge mode of operation the channels of the MX-10000 can't be bridged.

If there is any voltage drop between the Class AB buffer input and it's output (e.g. distortion), exactly the same amount of this error signal is reproduced at the output of the very accurate Class A stage and thus at the negative loudspeaker terminal, which means, that virtually any error of the Class AB stage is canceled out, as seen from the speaker.
Finally the only distortion the speaker sees, is the combination of distortions of the Class A output stage and small signal voltage amplifier stage (which runs Class A as well).

The only drawback of this approach (if HCA wouldn't be used to improve it) is, that the small Class A error correction amplifier sees the total output current of the speaker and thus had to run with at least 50% of this maximum current as idle current.

In case of the MX-10000, the needed idle current for a classical pure Class A stage at 2 Ohms would be 85V/2Ohms/2 = 21.25A, which is a lot !!! This would lead to an idle dissipation in the small Class A amplifier of roughly 7V*2*21.25A = 297W per Channel.
Here now HCA steps in, because it allows <1/3 of the above idle current (roughly 7A), while still providing full Class A operation into any load at full output, so even into 1 Ohms, which means 1200W/1Ohm in pure Class A with the MX-10000. The idle power in the small Class A stage is thus reduced to roughly 100W, which is a much more handy amount.

The trick with HCA is that the push pull transistors themselves don't operate in a quasi linear mode as with any other conventional Class A stage, they more or less operate in a special (hyperbolic) transfer mode, which has the very interesting feature, that the combination of these both transfer functions (voltage input against current output of the both push and pull sections separately) combine together again into an overall linear transfer function, while the currents within the output devices never reach zero, which means they never switch off.
This very transfer function is not a feature of the output transistors themselves : it is an artificially calculated one by means of an analog computer circuit and feedback from the output current of the power devices, so the output devices don't influence the actual transfer function anymore, thus eliminating strict matching (N to P), temperature and aging effects.

The HCA in the MX-10000 is related to the Class A error correction output stage alone (!!), the main feature of this amplifier nevertheless is the true feed forward error correction function.

Pioneer used a very similar approach as the Yamaha HCA in its Zero Switching amplifiers, but ran it at much lower idle currents, which meant less linearity and partial Class AB operation at higher powers as well.

A somehow similar circuit as the HCA was already used by Nelson Pass for his Threshold 400A and 800A models (called here dynamic biasing circuit) in the late seventies, but his circuit included the output transistors within the analog computer's transfer function and thus didn't work as good as expected (due to thermal distortions induced by the heat changes in the output devices, which altered the transfer function and other effects).
Another disadvantage was, that parts of the analog computer ran in Class AB, which means, that the switching distortions are transferred from the output stage to somewhere else (where it's impact is nevertheless much smaller, because of higher speed). The non-switching output stage of the Technics SE-A3 e.g. is another not so perfect solution, because the switching distortion happens within a tiny pair of ultra fast Germanium (!!!) diodes.

All other Yamaha MX amplifiers didn't use a second output stage at the negative speaker terminals or feed forward correction and thus featured HCA in the "main" output stage and because of the much higher operation voltages needed for these (rather +-70V instead of only +-7V) features here much less idle currents.
These lower Yamaha MX amps even don't use the original HCA (with current feedback from the output transistors) at all and only a dynamic bias current circuit, which shares the same disadvantages as the Threshold circuit or any other zero switching circuit (like the one e.g. used in the Technics SE-A3, Pioneer M-90, Denon PRA's, Victor M-L10 etc).

I'm not sure, if all of this was understandable, but the topic is difficult, because it relates to the absolute state of the art of linear amplifier design (which is still valid today)..."

Also part of this extremely rare monster are 100µ thick copper film / glass epoxy boards, hi fT (power buffer amplifiers) and MOS-FETs (outputs), four ELNA Great Supply caps, strictly non-magnetic metals for the enclosure, chassis and 2,5cm extruded aluminium front, walnut sideburns, carefully designed heatsinks for optimal cooling and absence of resonance, gold-plated brass terminals, oxygen-free copper bus bars and large logarithmic compression peak level meters. And even a photocoupler remote terminal so the MX-10000 can be powered on by the CX-10000.

And specs to die for.

Pity only around 250 MX-10000 were made.