Basically, poke and prod the circuit in some way which produces a measurable result. Generally speaking, the dynamic response (overshoot and bounce) will be largely similar anywhere in the circuit, give or take a few poles.
Typical examples:
- Step change in VREF (i.e., op-amp +in)
- Step change in VFB (i.e., op-amp -in)
- Step change in supply or load voltage/current
- Step change in error amp output (usually requires an additional summing stage)
The last two cases need not be good methods, however:
- A step change in load can be handled intrinsically, to some extent or another, by the hardware surrounding that load. Examples: large filter capacitors, low impedance emitter followers, etc. The resulting response (over/undershoot and ringing at the same load node) might be quite small. On the upside, it's exactly one of the figures you're looking for -- Zout = delta Vout / delta Iout.
- Likewise, the error amp output is a naturally "slippery" node, i.e., it adjusts itself, hopefully very quickly, to upsets anywhere in the loop, so that the added error is quickly subtracted out by the feedback mechanism. It should still do something, but you will observe only a glitch at the output. Which should still be good enough to excite some ringing, but the amplitude units should be considered "arbitrary", and used only as ratios to the size of the disturbance.
The step change in VREF/VFB is arguably the most useful, as the voltage gain is as given, or can be calculated from the way you are introducing it, and you see the size of the resulting step (the DC output changes proportionally!), and the ringing can be measured as a ratio to it.
Changing VREF is also the method of measuring amplifier performance, if you are treating your power supply as one. Obviously, this is most useful for amplifiers, as such, but may also be an important figure if you need an agile power supply (examples: AC inverter supply, pre-regulator within a switching power supply control loop, RF modulator, etc.).
For power supply purposes, direct measurement of supply and load transients is important data. The signals can generally be expected to be small (ideally, zero!), so you usually have to put up with that. Still, even if your measurement sucks, you can always report a worst case value -- if a 1V step at the supply input results in less than 10mV (i.e., less than half a div, on the 2mV/div scale, with a 10x probe (20mV/div actual gain)), you can at least say your PSRR is >= 40dB (i.e., 1/100th). That's not tremendous as power supplies go, but specs are better than nothing!
Tim