Hi, Alex!
The original Karna, designed by me, and built in four-chassis format by Gary Pimm (of Portland, Oregon) in 2003, used a 5687 class of input tube. (The 6900, 7044, and 7119 all have the same pinout and similar operating points. The current production JJ ECC99 is similar but has a different pinout.)
I selected that tube for the Amity, back in 1997, and also for the Karna because it had a low plate impedance ... around 2K ... and pretty decent linearity, much better than a 12AU7, which is quite poor and not really suited to driver duty. But I was never entirely happy with the 5687 or the other similar types. I tried just about all of them ... I have quite a stash of 5687, 6900, 7044, and 7119 tubes ... but there was always a bit of glassy, hard quality, nowhere as bad as a 6DJ8, but still there.
There’s nothing wrong with them, again, far better than any 6DJ8, but these are commercial tubes never intended for audio use, and never used in any Golden Age amplifiers, tuners, or TV sets. They were designed for analog computers, commercial radio relay use, and aerospace ... high-end commercial and military applications, at high prices, and not sold in consumer retail channels.
These days they come from military surplus stocks, and only produced in consumer format by JJ as the ECC99. So supplies are getting a little dodgy, twenty years on. Not really suitable for consumer use unless you already have a substantial stash of them, in the hundreds, and all tested and matched, of course.
The 6SN7, and its single-triode predecessors, like the 6J5, 6C5, etc. etc. are famous for their linearity, and they were designed for radio applications in the audio sections of the receiver and power amplifier. Millions were made, in varying quality, but all of them were more linear than the 12AU7 successor, or the quite different 6DJ8 (which was an RF tube never intended for audio). So there’s nothing rare or exotic about the 6SN7, unlike the 5687 family.
I mention "designed for audio" as if it is something special. Well no, not really. But if a tube was originally designed as an RF amplifier, it would never be checked in production for linearity, since RF circuits don’t care about linearity. Nowadays, of course, 6DJ8’s are never used for RF circuits, and only for audio, mostly high-end audio, not guitar amps.
This has the practical effect that vintage (NOS) stocks of authentic 6DJ8’s can be all over the place for in terms of linearity, since that’s not a controlled manufacturing parameter and would have no effect on its performance plugged in to a 1965 RCA color TV set or FM tuner, the task for which it was designed.
In practice, using Gary Pimm’s custom-designed spectrum analyzer with 140 dB resolution, we found that upper-harmonic (5th on up) spectral shapes mostly reflected a given manufacturer, and was surprisingly consistent from year to year. Gary Pimm and I have both worked in manufacturing for big and small companies, and we surmised that consistency reflected the special jigs that aligned the grids, and different manufacturers used slightly different techniques to align the inner structure.
Although tube models are intellectually useful in a design phase, they model ideal tubes that are only available as beautiful Platonic Ideals in a store somewhere in Heaven. Sadly, we humans on Earth have no access to that store. No Platonic Ideals for us.
The tubes we can actually buy were, and are, hand-made by skilled human beings, not robots. The grid pitch is not perfectly uniform, the grids are all tilted just a little bit, electrons escape out each end of the structure, the list of imperfections (and departures from ideal models) goes on and on. These tiny imperfections result in high-order harmonics that can be seen in a high-resolution spectrum analyzer, and heard in a good audio system.
Surprisingly, these departures from perfection are consistent with the manufacturer. That’s why Gary and I surmised it came down to small variations in assembly technique, or even the individual assembler. Again, tubes were never assembled by robots, and still aren’t today. The assembly was, and is, semi-automatic at best.
Frame-grid tubes, like the 6DJ8 or more exotic WE417A, are even more difficult to make consistently, and it doesn’t matter in a high-gain RF circuit anyway. Using them in an audio circuit is a roll of the dice, especially if there is no feedback to tidy up the mess. Harmless in a preamp at millivolt levels, not so good in a power amp.
For all these reasons, Don and I decided to move away from the 5687 family. (Neither Don nor I are fans of miniature 9-pin tubes anyway.) True, the 5687 family greatly simplifies the interstage transformer design, since the plate impedance is about three times lower than 6SN7, but that low plate impedance is the result of high transconductance and more difficult assembly procedures. Part of the reason that direct-heated triodes have a much cleaner spectra is they are big and easy to assemble ... as dumb as that. We’re talking late Twenties to late Thirties technology here ... precision assembly was very difficult back then, especially on a production basis.
Effectively, Don and I took the ultra precision out of the tube and put it into the transformer designer and assembler. That’s where the 21st Century tech comes in. These transformers could not have built in 1939, when the 6SN7 first came on the market (replacing single triodes). The 5687 family dates from the mid-Fifties, with transformer design still in the build-and-try phase, like the loudspeakers of the day. Computer modeling was still decades in the future.