Oh yeah - more on "linearity"
If you look at the transfer curve of any device, they tend to be most linear (straight line) in the middle of the curve. That means that for a given input increase, the output changes the same amount for each increase/decrease.
(ie. "linear" - a 1:1 relationship being maintained)
At the ends of the curves (all the way on, and all the way off) the relationship is not so good. In fact the transfer curve looks like an "S."
What this means is that when almost off - at the bottom of the curve the device is not very linear - you need much more drive to get not so much output, UNTIL you get past the curve of the "S" and it gets linear.
So, two things are important here: 1) for class AB and B circuits, you're almost off (near "cut-off") so the curve to begin with is not linear. 2) Class A is in the middle of the most linear portion of the curve.
In class AB, the idea is usually to bias so that you start out just above the non-linear portion, so that all positive going drive sends you only up through the linear portion of the curve. Of course, when any one side of a Push-Pull pair is driven toward "off" it travels back down the non-linear area. Thus the gains do not sum perfectly, and the earlier comment about "crossover distortion" applies.
In class A, you can drive out of the linear region on peaks! This is *part* of the sound you get from pure triode Class A ZERO feedback amps. There is effectively some *compression* when driven hard into the ends of the linear region (more signal does not get you equally more output). Which in part accounts for why small pure triode amps can sound like they play louder than very linear solid state amps.
Feedback changes this, since it forces the input to be whatever is needed to keep things linear (in this example, more drive is required until a 1:1 gain relationship is acheived).
Here when I say "1:1" gain relationship it does not mean that the gain is unity, but if it is 1 input unit then you get out "N" output units, 5 input: 5N output.
Which is best? That's entirely unclear.
To add to the design mix - not all devices, tubes or transistors *ARE LINEAR* by themselves! Most are NOT. Most
are either entirely non linear everywhere or linear over only a small portion of their range. This is a large part of why feedback of all sorts is employed in practice. Keep in mind that the main reason a cathode follower is linear, is that it uses 100% feedback! :- )
