... my plots.
All the circuits are the symmetrical ones, small BC transistors, biased at some 3-4mA, +/- 19V supply. A +/-2mA dumped into input, 1k2 to develop the voltage at the output. The diamond structure is chosen for driving capability of its "emitter" input. (These circuits are actually I/V converters so they have to deal with real world DAC output impedances.)
First, the simple mirror as shown previously, 1k both resistors (I actually already posted this to my site), less than that makes the current through the output transistor too small and less than 250R with 3mA makes it really not to work, so:
Performance (THD 0.012%):
Practically the same circuit but with four transistors Wilson (“CM3” here) and no emitter degeneration. As opposed to the simple circuit, all the Wilsons here were biased by current sources and not by resistors. The last graph here actually shows a bit better result than that from above, however results I've been getting were not that consistent (I mean between the samples measured, not from time to time), regardless of all the effort put in the parts matching. (Later I moved to Toshiba transistors which proved not only to sound but also to measure better than BC, at least in circuits like this.) Please ignore artifacts at 9kHz, 11kHz and about 20kHz in these graphs, they are not related to this circuit.
Four transistors Wilson, 100R resistors.
Four transistors Wilson, 300R resistors.