Kirkus, I'm having a very hard understanding why an all-NPN output stage is very non-linear at the x-over point. Please explain. Thanks.
In a typical all-NPN output stage, the drivers are complementary (NPN/PNP), but the outputs are not (NPN/NPN). In order to make the bias voltages work out, the transistors are connected in such a way so that for the positive pair, there are actually two local feedback loops - one around the driver, and one around the output transistor. But for the negative side, the driver and output are in a single local feedback loop together. This means that in the crossover region, there is an abrupt change in the static gain as the current is transferred from one half of the output stage to the other.
If you ever play around with adjusting the bias on such an output stage (I've actually modded many MC2105s to have adjustable bias), it can be easily seen that as the operating point is shifted, the crossover distortion never goes away, it simply changes shape a bit.
There doesn't seem to be anything primitive about this input stage except that it is (very) low gain since resistors are used as the load to input differential pair.
The following voltage amp is a transresistance amplifier (current in, voltage out) - meaning that the signal doesn't appear as a voltage on the collector of the diff-amp transistor that drives it . . . rather, it's a current. So in the diff-amp, it's the transconductance (set by the transistors' beta and their emitter resistors) that determines its open-loop gain.
And, how do "input stages with current-sources and current-mirrors, and voltage amplifiers with an active load" make the power amp more "linear"?
Using a current source to supply the diff-amp tail current effectively "makes the tail longer", and eliminates the variation in tail current with common-mode voltage. Also, the performance of any diff-amp is VERY dependent on the static balance of current between the two transistors (or tubes) . . . and with resistors setting the current, this critical balance is affected by tolerances in the (carbon!) resistors, and variations in beta and Vbe in the (old!) input transistors themselves, and also the line voltage when using an unregulated supply (as is almost always the case). Using an active current mirror, instead of the resistors, forces the static currents to remain balanced regardless of these variations.
As far as the active load on the voltage amp - the improvement is the same as for any traditional single-ended collector/plate-loaded voltage amp -- that is, by keeping the quescient current constant across the required voltage swing, the amplifier is more linear. And here is one place where increasing the impedance on the collector does indeed increase open-loop gain, which also reduces distortion.
All of this makes the amplifier much more linear in the traditional sense -- its transfer function. And if you ever compare a MC2105 to a MC2255 (an old amp, but which has these refinements) on the test bench - they are worlds apart in the amount of distortion they produce . . . especially in the higher harmonics.
