>i have done extensive thermal tests of the LM3046, PMP4201, THAT300, and MAT14 in exponential converter applications, and found that the high emitter resistance and low beta of the LM3046 made it the worst of the bunch, regardless of its monolithic process. the PMP4201 is so small, that another layer of insulation around the whole IC makes it pretty much the same temperature inside the plastic packaging. the PMP4201 and THAT300 performed identically, with the MAT14 being miles ahead of the rest.

>It's lower beta and (seemingly) higher emitter resistance than the duals only introduces a static linear error which can be perfectly compensated for.

i'm not sure i want to learn all this transistor stuff right now when i'm really only tweaking existing circuits. is LM3046/CA3046 sloppy enough to warrant an upgrade or even just replacing it with a newer manufactured equivalent that might have been manufactured with tighter tolerances than what they had 40-50 years ago? for example pic related is probably some averaging and logarithmic conversion of the full wave rectified audio signal to use for sidechain detection, i have no idea if replacing the CA3046 would improve performance.