Differential interfaces have a rather limited common mode voltage range. This is of course depending on the receiver, yet many CAN transceivers can handle just -2 to +7V.
On some sites, ground potential differences of several volts can easily build up over longer distances. Thus it certainly is a good idea to have all receiver grounds connected together by means of the ground wire of the communication cable. This does not create a ground loop since no signal is referenced to ground. That’s (one part of) the real beauty of differential pairs.
There might be some compensation current flowing through the ground wire, this should be low enough to not produce any significant voltage drop.
In noisy environments, shielded cabling is good practice anyway. This also helps to improve ESD immunity. The shield can only be effective when there is a low impedance path to the ground of the circuit that shall be protected. One single ground connection in an installation with several tenths of meters cable running is certainly not low impedance for higher frequencies. If galvanic isolation is really required for some reason, then each node should at least be connected to the shield by a HV capacitor (e.g. 10nF/2kV).
Well, that’s how I handle that very issue in my current project – it’s not proved in practice so far, as I’m still busy getting one single device working with all its features, so external connections over longer distances are designed in (CAN being one of them), yet of no relevance right now.
Hope this helps a bit to make up your mind!
