The Weak Link

The major problem with hollow state is that the vacuum tube is a high voltage, low current -- and therefore Hi-Z -- device. Electromechanical speakers are just the opposite: requiring high currents at relatively low voltages. The usual solution to this problem is the impedance matching transformer. Unfortunately, no transformer that has a ferromagnetic core is a linear device. Air core transformers that can operate at audio frequencies are not exactly efficient, and what you gain in core linearity you promptly lose in the enormous stray capacitance that the huge number of turns such a thing would require.

So the ideal situation would be to get rid of that transformer completely. There are a couple of factors that go against this: the Hi-Z nature of the device itself, and the fact that there is no such thing as a "P-Channel" VT. Given that, you are limited to some sort of "totem pole" output stage. Here, the problem is the lack of symmetry: one half will be acting as a grounded cathode amp with its characteristic voltage gain, while the other half will be, more or less, working as a cathode follower with its less than unity voltage gain. There have been various attempts to deal with this problem such as asymmetrical drive -- the SEPP output, or the Futterman variations that include feedback to either make the cathode follower half behave more like the grounded cathode half, or the inverted Futterman that applies NFB to the grounded cathode half to make it look more like a cathode follower.

There is another topology that at least takes care of the balance problem: the Wiggins Circlotron design, this being a bridged output stage. Even though originated for use with an OPT for the purpose of eliminating Class AB switching transients from the primary, it also makes for a very good OTL design as well.

The other problem remains: the low current nature of the VT itself. You really don't have much choice here, so far as possible types are concerned.

There have been VTs that were intended to give some serious output currents, these being the horizontal deflection finals. Even though the screen voltage specs limit the screen voltage to rather low levels, these can still pull some serious current at lowish V_PK's. Still, you are going to have to parallel up individual units in order to get the necessary output currents. The 6BQ6GTB can easily pull some 350mA of peak current when the plate and screen both operate at 150V_DC, if you go a bit into Class AB₂. For 30W of output into 8R, it will still require 16 6BQ6GTBs (8 per phase). That's a bit too much.

Other possible types would include the 6C33C (Series pass regulator used on the MiG fighter). There have been a lot of problems concerning this type, as its reliability is questionable, even with the special "burn-in" trick. Also, the pins are way too thin for the heater current requirement. MiG technicians replace not just the 6C33Cs, but also the socket as part of routine maintenance. I would avoid it.

The most likely type is the 6AS7 (and its relatives: 6080, industrial type, or the 6082, 26.5V heater version) -- another type originated as a series pass regulator tube. As such, it has the high current capability with reasonable V_PK's, and it is also a dual triode, so that you require half the number of bottles. Though its plate characteristics aren't spectacular, they aren't half bad either. It's not likely you'd want to use this in a conventional output circuit as there are better power triodes for that. Still, the 6AS7 can pull significant currents without taking the control grid positive. Eight 6AS7s (16 sections, 8 per phase) can get you 30W of output without serious spec busting. The type is also well known for having a high g_m and a very low r_p. The 6AS7 has been used to drive VT Class B audio output stages (AM plate modulators) for a very long time now.

Does getting rid of the OPT really represent an improvement in overall linearity? What you might gain in one area, you might lose back in another: operating the VTs into a very steep loadline (eight sections in parallel to drive 8 ohms is just 64 ohms per section). You might be just as well off in substituting MOSFETs for the hollow state output. There have been some complaints that paralleling finals leads to sonic degradation, since the individual VTs won't all be operating at precisely the same currents with the same exact characteristics. Whether or not this makes a difference is a whole 'nother story. At least the very low (for hollow state) impedance of the 6AS7 does help to mitigate this effect. This is another one of those areas where you just might have to try it to see and make up your own mind if it's worth it.

As for trying this, it is definitely something to take a shot at. It is especially helpful in that the 6AS7 is a current production VT, and at least in this case, the new production is superior to NOS offerings: better matching between sections, and a more robust design that will stand up to the slight spec-busting that can reduce the number of triode sections per phase. The 0.125A plate/cathode current rating is for DC regulator use. That peak current can be increased to 0.375A in audio final use. Music and voice programs won't be spending very much time at the maximum peak current. (It would be a different story for DC use or continuous max P_RMS applications. (If you were doing that, then consider the 0.125A spec to be a DC/RMS value.)

The only other requirements are DC offset detection/correction since you definitely want to keep DC out of speaker voice coils, fault detection in case one 6AS7 (or one section of a 6AS7) develops some fault that causes it to pull more current, and HV delay. Being that the intended purpose was series pass duty, the heater-to-cathode insulation is a good deal thicker to stand up to higher than normal voltages. This also slows down greatly the cathode warm-up, and the specs call for preheating to make sure the cathodes are at operating temp before hitting 'em with the HV.

The specs also strongly discourage the use of fixed bias. The problem of using the much more desirable fixed bias can be solved by mixed fixed/cathode bias, and over current detection. A 6AS7, OTL, Circlotron design is definitely "in the works".