Here's a little more brainstorming on the OTL power amp.
As the Bottlehead reference system has gotten more refined we have been finding limitations with transformers, but also with cathode followers. And we have long felt that paralleled tubes have limitations when pursuing the highest degree of resolution, and so does feedback in general. OTLs keep coming up, to get around the transformer problem. However, the fact is that no tube has a low enough impedance to drive an 8-ohm speaker with a reasonable damping factor, without resorting to heavy feedback or massively parallel tubes. And I mean heavy; a cathode follower is not good enough. In addition, in order to get more than a watt you need more current than any practical tube can supply.
Both problems can be addressed with parallel tubes. Here's an example:
The 6AS7 or 6080 can have a plate resistance as low as 280 ohms according to the specification. Since cathode followers and other feedback approaches are out, that would be the output impedance. An 8 ohm output impedance would give a damping factor of 2 into a 16-ohm speaker, and would require 35 triodes in parallel. A stereo pair would be 35 tubes, drawing 8.75 amps at 135 volts, for a total dissipation of 1181 watts. Heater power would be another 87.5 amps at 6.3 volts, or 551.25 watts. Of course, if you are avoiding transformers you probably don't want to use plate chokes, so figure another 350 watts or so for the current source plate loads. For this, you'll get ideally 150 watts out; in practice due to the inability to find a set of 35 matched 6080 tubes you might get 100 watts - of parallel tube sound. The chassis plate could be as small as 20 square feet. You'll need a separate 20- or 25-amp circuit just for the amp.
This does not sound to me like a Bottlehead amp at a Bottlehead price. If anyone has a spare $10k-$15k to build one as an experiment, it would be interesting to find out whether it sounds OK or not.
At lower power, you need feedback to get the output impedance down. I must have been thinking of that when I suggested a single 6C33 for a 1-watt amp. That could be much easier to build - but, again, with heavy feedback it will not reach the highest level of resolution. I suppose we should try it; at least it won't be as expensive as the massively parallel approach.
