I'm building a 7kW bidirectional two-stage switching converter. Each stage has its own control board and processor. The entire system runs off a common 24V rail.
Two H-bridges on either side of a transformer, running at 18 kHz, 50% duty cycle, both sides switching. Big dumb isolator, no regulation. Processor just feeds the pulse widths, phases them in on startup, and provides appropriate delay across the transformer. Running a dsPIC 30F4013, 5V rail from a switching regulator.
Two IGBTs in series, running at 12 kHz. If the lower switches, it boosts out of the isolation stage. If the upper switches, it bucks into the isolation stage. Running a dsPIC 33EP128MC506, 3.3V rail off a switching regulator.
In boost mode, the entire system works fine, no problems across the entire power range. In buck mode, I get processor resets when the system reaches about half rated current. Specifically, the isolation stage processor resets pretty consistently when I reach a particular current range. (It's definitely current-related. If I tell my control software to limit the current, and keep raising the input voltage, I get no faults.) If I alter the configuration bits, moving the brownout detect level from 4.5V to 2.0V, this processor stops resetting. This strongly indicates to me that the processor is resetting due to brownout. RCON register contents are consistent with this as cause of reset.
So now I can raise the current further, However, when I do, the regulation stage processor now starts to reset! That particular processor doesn't have any way to alter the brownout level, but it's a reasonable assumption that's the cause. RCON register is again consistent with brownout as cause of reset.
So I've got two processors resetting as a response to how much current is running through my converter, but only in one direction. It's reasonable to guess that both are resetting on brownout, due to some shared environmental factor. I've tried disconnecting all common points between them, disconnecting the down-stream H-bridge drive, and rerouting some cables. Nothing I've described changes. The only things both processors still have in common are their common power rail, and physical proximity.
Scope shows no deviation on the logic power rails. There's some 18 kHz noise, but that's there even if I connect both ends of the scope probe to the common test point. Otherwise, the rails look great. I've got .1 uF multilayer ceramic caps at every power input to both processors, and a 100 uF tantalum also immediately adjacent. Increasing these capacitances has thus far had no effect. Both circuit boards are two-layer, with a copper area tied to common on the bottom layer directly under each processor. (Processor is defined as being on the top layer, so there's circuit board material in between.)
All I can think to try is some sort of physical rearrangement involving shielding the processors from the power components. Does anyone have other suggestions?