An incandescent bulb has a lower resistance when it is cold vs when it is hot. (see attachment) This is not a constant current sink but it does draw more current at low voltages than at high voltages, a function that your circuit requires. It does not introduce offset voltages and it does work down to zero volts. It was used by Bill Hewlett of Hewlett-Packard for their first variable frequency oscillator to stabilize the output voltage in 1940. It does seem like a 50s solution, but it could be the simplest solution. I used it to solve a residual voltage problem on a DC power supply.
Thank you very much, I'm always interested in cool history lessions like that (really!). Awesome solution and I'll definitely keep that in mind. For this project though, where I'm trying to make the PSU as precise as possible (without spending thousands of bucks, though), that seems a bit too unpredictable.
MOSFETs could be substituted for the bipolar transistors in the present current mirror. This would eliminate any current injection into the output. The devices should be matched and tested to ensure that the sink current is actually what is expected.
I'll take a look whether there are matched MOSFETs in one case being sold, I don't really want to match them by hand, to be honest. I might try and breadboard that but I'm kinda sold for the opamp solution already, I have to admit. In my design there is a spare opamp (of a dual package) anyway since I have ditched that sense input idea and I have a suitable MOSFET in my BOM as well already. So, meeh.. :-)
Connecting the emitters of the current mirror to a small negative voltage of -1 V relative to AGND would ensure that the output voltage would be zero when needed. There were some threads on this forum recently on this subject but I can't seem to find them quickly. TI and other manufacturers make some devices for this.
I'd be really interested in those devices. I wasn't really able to find anything besides the LM334 but that does only work from about 1V upwards. I tried to connect -5V out of a TC7660 to the negative pin of the LM334 to shift everything but that didn't work either. Something was drawing current but it was not the LM334. I had my Lab PSU set to Imax of 100mA and those were drawn so I guess that was BS, what I breadboarded there.
There are a number of ways to test a power supply for stability. I would start with using power resistors of differing values for loads and try varying the voltages and currents applied while monitoring the output. I would then use a function generator to inject a low level 10 Hz square wave into IC2A-2 while repeating the above tests. If the output does not overshoot or ring, then it is basically stable.
I have a cheap load from aliexpress which i already used to torture my PSU, so far so good, no issues. Injecting some signal is something I didn't do yet. What amplitude should that square wave have and what current capability?
I was planning to build up an array of mosfets with some power resistors and some atmega to be able to connect and disconnect various resistors very fast and thus change load very fast without switch contact bouncing and look with a scope what the output does.