Backstory: a couple days ago I had made some changes to my adjustable dummy load: I wanted the fans and op amp to be powered from a separate source, so that I could turn the load current right down to 0 mA. While doing this I also swapped out the op amp for a different one that could run at higher voltages, so I didn't need an onboard 5v regulator. Initial tests showed that it worked, so I didn't think anything more about it.

This evening I had some time to work on hardware again (the last few evenings I had been doing USB programming). I wanted to test some higher currents, with the hope of figuring out a working implementation of current limiting. I started off by calibrating the supply for a LM338 (so that I am not limited to 1.5A). The voltage calibration went fine, but when I tried to calibrate the current measurement, things went wonky.

I pulled out my scope to see what was happening, and to my astonishment saw extreme oscillations, noise, and all sorts of ugly garbage on the Vout line, whenever the current exceeded about 200mA.

Initially I assumed that I did something wrong. I triple checked the breadboard against the schematic, added bypass caps at every place I could think of, looked for shorts in the control circuit, etc. Nothing. Next I started tearing down the supply and going back to basics, all the way back to using a single pot to control voltage. Still, oscillations out the wah-zoo!

My overly paranoid brain next went to ESD... had I zapped the regulator at some point? I pulled out a new LM338, straight from the staic wrapping... AND THE SAME THING WAS HAPPENING.

At this point I really started to doubt myself. Had I actually done the right setup when I originally measured regulation under load? Perhaps I had thought I was testing under load, but was in fact not... who knows?!? By now I had been staring at the same board for hours, with no progress.

So what was different?

Right... my dummy load (stupid load) had changed.

I tried a couple of simple current sinks using a LM338 with a single resistor (trying a few distinct current loads, from 2A down to 500mA-ish). The output was solid as a rock.

Well, now we are getting somewhere.

I did some reading on dummy loads similar to mine, and found some suggestions. Don't drive the transistor directly with a LM324 (use a resistor in series); add decoupling caps between the inverting and non-inverting inputs; add decoupling caps between GND and the base of the transistor, etc. I ended up doing all of them (why go halfway!), and while the load is still a bit noisy, it is much better than before. Furthermore, since I know that the source of the noise is the load, I don't have to worry quite so much about it. (As long as I can try a clean load at various currents to verify load regulation, I am good).

The moral of this story is two-fold: 1) You know what happens when you make assumptions, and 2) Dummy loads can be quite stupid.

Cheers