As a minor diversion, I should describe the "Golden Age" amplifiers I keep referring to. This aren’t just the amplifiers made in the Fifties and Sixties; it describes the majority of PP tube amps made since then, including today.
There were only a few basic Golden Age circuits, or topologies, as we like to call them. (Topologies omit circuit values, but are easily worked out once you know the tubes.) The first was the Williamson of 1948, but it had the drawback of marginal stability. Still, it dominated the US market until 1955 or so, when the much simpler Dynaco variant came in. (The Dynaco topology simply omits the driver stage of the Williamson and uses the phase splitter to drive the output tubes. More distortion but more stable.)
The Mullard became the prototype of many tube amps as the better-performing alternative to the Dynaco circuit, and is still widely used today. Let’s walk through it.
There’s a high-gain input tube, typically either a 12AX7 or a pentode like an EF86. This is direct-coupled to one half of a differential stage, with the other grid AC-coupled through a cap to ground. Because the grid of the diff stage is at 150 volts or so, the cathode is a little bit higher, maybe 155 volts. This requires a large value resistor that goes all the way to ground, so the diff stage is frequently called a "long-tailed pair". A current source could replace the resistor, but in practice, the performance is very similar to a current source, so it’s rarely done even in modern amps.
The diff pair are a pretty good phase splitter, and unlike the split-load inverter of the Dynaco circuit, audio-frequency balance is not too sensitive to load. It also has more drive capability than the split-load inverter, and unlike the split-load inverter, it has some gain, too. So a win all around.
And we’re not talking about a lot of parts here: 3 triode sections, and the output pair. A Dynaco is even simpler, with 2 triode sections, and the output pair. The only coupling caps with either circuit are between the grids of the output pair and the preceding circuit, so not really complex, and simple enough that a stereo chassis, running off a single B+ supply, is quite practical.
The point of the high gain (in the input section) is to give feedback something to work with. Feedback requires "excess gain" to work its magic; you need 20 dB of excess gain to get 20 dB of feedback, which will reduce overall distortion tenfold. In a pentode or ultralinear connected amplifier, the output impedance is way too high to use with most speakers. The feedback really comes in handy here: 20 dB of feedback reduces output impedance tenfold.
What limits applicability of feedback is loss of stability if too much is used (I’m not going to get into Nyquist Stability Criteria here, nor phase margin, settling time, etc.) In other words, if we slap in another gain stage and try for 40 dB of feedback, it will just oscillate. At full power. And take out a tweeter before damaging itself and letting the smoke out.
A more clever approach is wrapping local feedback around the most distorted stages, like the output section, and then add overall global feedback on top of that. This was done in the McIntosh, Citation II, and a few other amplifiers. This really gets the distortion numbers down, but clipping can get ugly, and settling time from transients can be an issue. Multiple feedback amplifiers can be quite sensitive to operating conditions. It’s more often seen in modern transistor amps as "two-pole compensation", and is not trivial to design.
Note: To puzzle out a schematic, by convention, signal flow is left to right, just like you’re reading this. To see what a tube is doing, look what the grid (the dotted line) is connected to. Often, there will be a coupling cap, typically 0.1uF. If it is much smaller than that, like 30 mmF or 30 pF, it is bypassing RF or has something to do with stability. Larger caps are cathode bypasses or power supply. The plates (the flat-topped dingus) is the output of the tube and typically heads to the right side of the schematic.
You usually have to stare at a phase splitter quite a while before the function becomes obvious. One side is quite simple, coming directly from the input tube, but the other side can be pretty weird. A diff stage can be puzzling, because the DC connection is a high-value resistor going to the other grid, and the AC connection just goes to ground through a 0.1uF cap. The "other half" is actually driven from its cathode, not the grid.
What gives away a split-load inverter, or "concertina" stage, are the equal cathode and plate resistors. This is a dead giveaway you are looking at an inverter, since no other tube stage uses equal resistors ... for one thing, it’s kind of useless for anything else, since gain is a bit less than unity.
I leave the "floating paraphase" as an exercise for the reader. I kind of like them, actually, because current drive for the power tubes is pretty good, although balance is only so-so.
