How far, how fast, which ground

Inside one bit

Every desk on the rail must see the others' bits before it samples its own. During arbitration, and in the acknowledge slot, the far desk's dominant has to travel to you before you read the bit, so the signal has to make the trip there and back inside one bit time. Cable carries a signal at about five nanoseconds a metre, so 100 m costs about a thousand nanoseconds on the round trip.

At 500 kbit/s a bit lasts two thousand nanoseconds, and that round trip takes half of it. At 1 Mbit/s a bit lasts a thousand, and the same cable takes all of it. Each desk samples somewhere between about 75 and 87 per cent of the way through the bit, so the cable alone has to fit well before that. The transceiver chips at each end add delay of their own, which is why the honest limit is shorter than the cable arithmetic suggests.

The rule of thumb

Classical CAN runs up to 1 Mbit/s. The table people use is a rule of thumb, with every length an "about": 1 Mbit/s reaches about 40 m, 500 kbit/s about 100 m, 250 kbit/s about 250 m, 125 kbit/s about 500 m, and 50 kbit/s about 1000 m. The 40 m figure is what ISO 11898 and the CiA 303-1 cabling guideline give. The table in the CANopen documents says 25 m at 1 Mbit/s. Neither is a promise.

Which ground

The other limit is the ground. Each receiver measures both wires against its own ground, and they must stay inside about -2 V to +7 V of it. A dominant bit puts CAN high near 3.5 V, so two grounds 4 V apart push a wire out of the window at the far desk, and that desk may misread bits or stop working.

The industrial guideline CiA 102 asks that ground potentials differ by no more than 2 V. Desks on one supply or chassis usually meet that. Boards on separate supplies with no common ground may not, and where grounds really are far apart, isolated transceivers exist.

Common mistakes

• Reading the table as a guarantee. It is about, and the sources disagree on the top row. Allow margin, especially at 1 Mbit/s.
• Judging a long rail by a short bench test. Two boards a hand apart work at any rate. The delay only shows when the cable gets long.
• Powering far-apart boards from separate supplies. Their grounds can differ by more than the window allows, and the fault looks like a random bit error.