Humm.. so I've been thinking about SMPS's. In particular the topology used in most AT supplies: half bridge, choke input filter, blah blah.
Well, say you have a load on the output, the system is in equilibrium, and you suddenly decrease the load. If the controller responds slowly (as it must in voltage mode), then the duty cycle will remain constant, pumping more charge into the filter inductor, also ramping up transformer current as well. Eventually, the filter capacitor will respond, but only after roughly an LC time constant, perhaps a milisecond or so (say, 50 cycles of the PWM frequency), at which time the inductor will theoretically have a metric shitload of amps through it! This energy then flows into the capacitor, so voltage keeps rising, a lot, and the controller chokes off the PWM until capacitor voltage falls to a reasonable level, at which point the voltage continues to drop, because it then takes more time to recharge the inductor.
If the controller is current-mode, duty cycle can be controlled directly by current and somehow secondarily by voltage, but voltage will still overshoot as the inductor's excess energy plops into the capacitor.
But neither the KA7500 or SG3524 (both used in AT supplies) are naturally current mode. I don't see obvious evidence of a current shunt and amplifier to convert output current to error voltage, though I haven't thoroughly traced the schematic.
What gives?
Tim