I've been playing with (or at least hypothesizing about) distributed amplifiers lately. Mainly because, for the hardware I have on hand, it's the easiest way to get a fairly large amount of RF power (for wideband lab applications)
Which, because common power BJTs and MOSFETs have internal frequency limitations (distributed resistance, excessive junction capacitances, and recombination in the case of BJTs), means vacuum tubes (which tend to have a higher internal frequency limit, but have relatively high capacitance and very high output impedance in comparison).
So I'm wondering a few things:
1. Is it practical to build one with triodes (grounded cathode)?
2. Is it practical to build one with grounded grid (cathode input) design?
3. Can a push-pull amplifier be built? I don't mean a balanced amp (such as Tektronix used extensively, for obvious reasons), I mean class AB, lots of harmonics each side.
Partial answers:
1. The general advice seems to be, avoid output-input coupling at all costs, because you'll screw it up. Hence, almost unanimously, pentodes or cascodes have been preferred. This persists today with lots of articles on monolithic FET amps using cascodes. (Though with the C(V) variation of semiconductors, that's probably even more important than with tubes.) I've read a patent:
http://www.google.com/patents/US2778886 that says it's perfectly fine, as long as the triangle of transmission lines (GND-grid, grid-plate, and plate-GND) has equal propagation velocities. But anyway, adjusting such a circuit seems... ridiculous. So I'd probably want to avoid that, anyway.
2. Most sources say that, if you have a lossy input line (some FETs, most BJTs), you're limited in the number of stages. Which implies that a lossy input device simply can't provide additional GBW, no matter what. But this looks stupid to me.
You can easily build a tapered transmission line structure, that effectively connects all those loads in parallel. At each node, you have three connections: the input-side filter section, an amplifier load, and the filter section to the next amplifier stages. The input side has an impedance of the parallel combination of the amplifier and the next filter section. The last section is simply matched into the last amplifier load, no termination resistor. (Or you could add a termination resistor, but this would only have the effect of reducing the entire network impedance, while wasting power.) Who needs termination when you can get distributed termination for free, right?
The impedances work out, so it should work correctly. But I don't want to go and build one, and spend weeks trying to tune it, only to discover it's impossible.
3. I don't know, but suspect, that, if you take two matched but otherwise independent dist amps, and split and merge them with a 180 degree hybrid on the input and output, you won't get a PP system, because the energy from one must go down and reflect off (or be absorbed by) the other, as well as go to the output. Which means the pushing and pulling has to be done on a more fundamental level, such as using tightly coupled inductors in a stage-by-stage complementary circuit. Is this right?
Interesting, I see a ref that says it really can be that simple,
https://www.google.com/patents/US4446445On the other hand, this one uses hybrids on each tube,
http://www.google.com.na/patents/US3571742Perhaps it's not exclusive?
Tim