Let’s discuss driver tubes. Back in the Nineties, my Tektronix friend Gary Pimm built a distortion analyzer that could measure each distortion harmonic, out to the 10th harmonic, down to an astounding -120 dB. Maybe even -140 dB. It was way, way down there.
So naturally Gary and I went and measured a bunch of tubes, and ran them at realistic drive levels (50 volts RMS or so). I knew from previous experience that audibility of 2nd harmonic is quite low ... I can barely hear 1% second harmonic, and audibility of 3rd harmonic is around 0.25%, if it’s mixed with somewhat more 2nd harmonic. (3rd by itself is harsh, but sounds pretty nice mixed 1:4 with 2nd harmonic, giving a richer, more fun sound.)
But the high-order harmonics are not nice sounding. Crowhurst suggested back in the Fifties weighting audibility of the order by the square of the order, or even by the cube of the order. High-order harmonics are what give electronics their nasty "electronic" coloration, and contribute very strongly with a forest of IM distortion clutter on the noise floor.
Unfortunately, unless the design is very poor, the relatively innocuous 2nd and 3rd order harmonics will dominate the THD measurements, to the point where THD is nearly useless for assessing sonics. When 2% THD is just barely audible, no, that is not a useful measurement. By that standard, all amplifiers are perfect.
But a spectral analysis is a lot more interesting, providing you pretty much discard the 2nd harmonic, and take a good, hard look at the rest, particularly the 4th on up. In a good amplifying device, you want 4th through 10th as low as possible.
We measured 300B’s from several vendors and vintage 45’s. Of all the tubes, these were closest to perfect. High-order harmonics were nearly absent, and of all the 45’s, none were visible at all. For 300B’s, they were way down, -60 dB or less, depending on the vendor.
That was one surprise. Most tubes, with a few exceptions, had comparable THD and 2nd harmonic, but the high-order spectra would vary within a given tube type by 20 dB, a huge difference. Twiddling the operating current made about 2-3 dB change ... not much ... but changing the brand would show much larger changes, 10 to 15 dB. We were plainly seeing differences in manufacturing technique ... uneven grid structure, tilted grids, etc.
The DHT’s were the clear winners in all categories, but the vintage 2A3’s were 10 to 20 dB worse than the 300B’s and 45’s. And the indirect-heated tubes were another 10 dB worse, or comparable to the 2A3’s. The worst tubes of all were the 9-pin miniatures, with the 12AU7 and 6DJ8 at the bottom of the heap. The 12AU7 had the decency to have lots of 2nd harmonic, which masked the clutter higher up, but the 6DJ8 had less 2nd harmonic and a harmonic spectra more like a transistor, with abundant high-order terms.
The 6DJ8 was a frame-grid tube designed for RF front ends in TVs and FM tuners, where linearity is of no concern at all, so you can’t really hold that against it. And the 12AU7 was designed as a compact replacement for the older 6SN7, operating at the same bias point and voltage. But ... high-order distortion is 10 to 20 dB worse than any 6SN7, even the worst. But you can’t hold that against it; 20 dB of feedback was universal when the 12AU7 was introduced. Loop feedback will reduce distortion in all tubes in the circuit, so it hardly matters.
The 6SN7 family is the tube of choice if you care about high-order distortion, and a DHT alternative is out of the question due to cost, size, practicality, microphonics, and severe filtering requirements for the filament supply. Of the 6SN7 family, the antique single triodes were the best ... 6J5, 6C5, etc. but there certainly were good 6SN7’s, both new and vintage. And differences were consistent within brands, reflecting manufacturing technique, and not always visible from the outside. Were the fancy MILSPEC ones better? Not really. More consistent, which is what MILSPEC is really about.
Don is correct that octals are (nearly always) better, unless you need a mike or phono preamp tube. That’s the 6DJ8 (or WE 417A) niche, where signal levels are very low, and noise and low microphonics are the primary concern.
