Let’s talk a bit about distortion, particularly in a DHT driver. It defeats the purpose of using a very low distortion 45, 2A3, 300B, or 845 if the driver has more distortion than the DHT itself. You might as well use a much cheaper power pentode or beam tetrode and save yourself a lot of trouble. Not only that, but pentode and ultralinear-connected output sections requires feedback across the whole amplifier, which cleans up the mess from the input and driver sections. This is why the driver section is an afterthought in Golden Age PP pentode amplifiers. In the Dynaco circuit, there isn’t even a driver ... the power tubes are driven from the split-load inverter, or "concertina" stage.
But ... if the goal is the most linear possible amplifying stage, with zero feedback, a completely different approach is required. True, push-pull substantially reduces distortion, but in reality it only reduces even-order distortion ... 2nd harmonic, 4th harmonic, 6th harmonic. etc. It has no effect on odd-order distortion ... 3rd harmonic, 5th harmonic, 7th harmonic, etc. Most tubes have dominant 2nd harmonic distortion. But not all. The 6DJ8/6922, an RF tube, has dominant 3rd harmonic. Tubes designed for video amps can also have dominant 3rd harmonics, since distortion in a video tube has almost no effect on the picture (even 5% distortion would be barely visible).
Aside from the subjective sonics of dominant 3rd harmonic, this has consequences for a PP amplifier. The PP circuit will have no effect on 3rd harmonic; it will not reduce it even 1 dB. So tubes with dominant 3rd harmonic should be avoided.
What about more "normal" tubes with dominant 2nd harmonic? Well, one of the requirements for PP cancellation is the magnitudes and phase of the cancelled harmonics line up with each other. The magnitudes of the (even) harmonics from each tube should be within 1 dB of each other, and the phase of the (even) harmonics should be within 45 degrees of each other. Normally this would be of no concern with reasonably matched tube sections, but it does matter if the capacitive loads are not the same. The odd harmonics, of course, do not cancel.
The notion of harmonics having phase might seem a little odd, but keep in mind that square waves and triangle waves have identical magnitudes of harmonics; the only difference between the two signals is the phase of the harmonics. So the phase of the harmonics is not insignificant, and actually reflects a different transfer curve. In graphical terms, you want complementary transfer curves; if kinks appear in the curve of one tube but not the other, that kink will not cancel.
The full scope of the driver requirements for DHTs is probably now coming into view.
* Three times as much swing is required, compared to pentode or beam tetrode.
* Linearity should be better than the DHT itself.
* The load is mostly capacitive, consisting of the 60 to 80 pF Miller capacitance of the DHT grid. Reactive loads increase distortion at high frequencies, where it is most audible, and also reflects the power back to the plates of the driver tube.
* If the driver is PP, 2nd harmonic should be at least 10 to 20 dB greater than 3rd harmonic, and the two loads should be symmetric to keep the phase of the driver harmonics the same.
* There should be at least 3 dB of driver headroom so the amplifier doesn’t all clip at once. More headroom is desirable.
* Momentary sags, or program-correlated noise, in the power tube B+ supply should not interact with the driver tube B+ supply.
