Paul Keinanen wrote:
Try to take two CAN nodes and make them work together without connecting
the GND. That may even work at the bench top conditions and for the
moderate speed and the short wire length.
I have been using CAN at 50-250 kbit/s up to several hundred meters in
industrial environments, with dozens of nodes on the bus with or
without a signal ground. I haven't seen big differences one way or the
other. Of course the CAN transceiver on each node is isolated from the
rest of the node.
Practically, the GND has to be
connected for good, and the GND problems is the most common reason for
the CAN not working properly.
The signal ground may help in some situations or it may worsen the
situation, especially when a separate wire is running parallel to the
twisted pair carrying the data (i.e. not cancelling any external
field).
CAN transmitter is pulling in one direction only. If there is a CM,
there will be a huge CM glitch with ringing when the dominant level is
changed to the recessive.
When isolated transceivers are used, they usually tolerate 0.5-2.5 kV
common mode voltages. In order to cause data integrity problems, the
common mode voltage would have to be translated to a differential
voltage. When the transmitter goes to recessive state, the current
stops flowing and no voltage drop is generated across the termination
resistance.
Only if the stray capacitance from CAN-H to external ground is much
different from the stray capacitance from CAN-L to external ground, a
voltage difference could be generated across the termination
resistance, when current from the wire with lower stray capacitance
flows through the termination resistance to the other side with a
larger stray capacitance.
CAN is bidirectional bus with the arbitration at the bit level. It is
fairly sensitive to the fast CM glitches. The OP question was about
operating CAN in the proximity of the powerful alternator.
Such alternators generate very strong slowly varying magnetic fields,
but it will generate high frequency interference only if the slip
rings are sparking (assuming synchronous generator).
If a separate signal ground is used, use a wire from the same quad
pair in order to cancel out any voltage induced between the signal
pair and the ground wire. However, when using a quad pair, the balance
in the actual CAN-H and CAN-L lines may be worse than when using an
ordinary twisted pair.
Paul