I have a device, that consists of logic controller, with some inputs, outputs, switching power supplies, relays, communication (all the usuall stuff) and an RFID module, rated for some higher powers (up to 2A@5V, but usually consumes 0.5A). All the mentioned stuff is placed on one PCB. There is a big ground plane underneath all the components, apart from the RFID, which sits over it's own copper plane. The two parts are conencted with a bundle of cables (about 15 cm long).
And here is the problem: Two of my RFID modules suddenly stopped working, in a very similar circumstances. I plugged all the cables and checked if everything works, next to it's destination cabinet. I took out a fuse, closed controllers plastic casing, and then put it all inside bigger, metal cabinet. After plugging the fuse back again - RFID did not respond. Two identical failures, in two different sites, happening during installation. Havent had any similar mishap, powering the device on my desk, even though I restarted it dozens of times.
So, I started to look for flaws in the design. First thing that caught my attention was the fact that RFID and the rest don't share the same ground. It was alarming, because RFID module's microcontroller is directly connected to my logic controller's micro. These communicate via UART, and some other lines, without separation, other than a 33 or 50 Ohm resistor (can't remember exactly). So 2 IC's communicate directly, despiter not having common ground. The two grounds are not floating, relative to each othe, because it wouldn't, obviously, work at all. Instead, these are separated by a ferrite bead. So, ciruit's designer wanted to keep all the potential high frequencies inside the RFID and not let any to the rest of the board.
I instantly checked the voltage drop on that bead. During normal opeation, it was around 100mV, peaking at 200mV, at high frequency. So not terrible. Then, I focused on the exact moment when I power the device. To turn it on, I pushed a switch on a 230V power outlet. Oscillosope probe's terminals where put between the two groun planes. Everytime I flicked the switched, the osciloscope triggered. As you can see below, the oscilations have a significant amplitude, easily exceeding IC's absolute maximum ratings. Then I replaced the ferrite bead, between the two copper planes with a solder bridge. As you can see, the amplitude dropped significantly. Next thing I did was to turn two copper solids into one. I just scratched some soldermask and soldered those together. But waveforms still looked the same as it did after getting rid of the ferrite bead. To make sure, that the waveform I measured was not just some error that the scope picked up, I shorted the probe's terminals, put in more or less the same place it was before. Didn't triggered. Then I put some piece of cable (about 30 cm) between the probes - it caught waveform similar to what I saw between copper planes. The longer the cable was, the bigger the amplitude of the signal.
So my hipothesis is: switching the power supply to mains generates some kind of a electromagnetic disturbance. The cables and the board picks it up, and high frequency current induces between part of my device, destroying the more vulnerable RFID module.
As you can tell, I'm rather new to the concept of EMI and so on. The device is my company's prototype and I can make changes to the final design. The designer that made the original PCB was our subcontractor (outsourcing).
I apreciate all of your reponses.