Memoirs of an overgrown I²C bus

In 2009, I worked on a instrument which had a large and complex I²C bus.

• Four PIC microcontrollers: one is a master, three slaves. "Dumb" peripherals too: EEPROMs, I/O expander, digital potentiometer.
• 1.8m [6ft] cable. One of the slave PICs was at the end of the cable. The cable is hot-pluggable.
• AC power going through the same cable at the same time as I²C communication. 500kHz sine wave. 300W power. High voltage in some modes of operation, high current in others. That’s interference from the I²C’s perspective.
• A section of the I²C bus was galvanically isolated.

Yes, we were able to make all those complications work reliably. But, was the I²C the right communication bus for that instrument?

I²C is not intended to operate over cables, and it's not intended to operate in presence of EMI. Differential buses, such as RS-485 or CAN, are better suited for this kind of scenario. Such bus is inherently more EMI-resistant. It would work well for communication across the cable and between the boards, while I²C could provide convenient short range communication within the immediate neighborhood of each microcontroller.

Related

Sending I²C bus signals via long communications cables. Application note AN10658 by NXP (2008).

Very large I²C bus systems and long buses. Application note AN11084 by NXP (2011).

These app notes were helpful. At the same time, instead of stretching I²C way beyond its intended use, one should consider switching to a communication bus that's actually intended for longer range.