Hardware & electrical · topic E18 · 25 hours
Field buses: CAN / RS485 / CAN FD
The wire between joints sets the whole machine's real-time ceiling. CAN's non-destructive arbitration is its most elegant design, and the most common interview question: why does the smaller ID win without losing any data?
What to learn
- Electricals: differential twisted pair, dominant (0) vs recessive (1), and wired-AND implementing arbitration
- Non-destructive arbitration: the smaller the ID, the higher the priority; the loser backs off automatically and no data is lost
- Bit stuffing, CRC15, the ACK slot, error frames and error counters (error active → error passive → bus off)
- Termination: 120 Ω at each end; the classic symptom of a missing end is «occasional errored frames, collapse under load»
- Bus load arithmetic: frame length (with stuffing) + bit time → worst-case latency; why 10 axes at 1 kHz do not fit on classic CAN
- CAN FD: arbitration and data phases at different rates, variable-length CRC — why it goes faster
- RS485: half-duplex, differential, termination and bias resistors, direction-switch delay; Modbus RTU rides on it
- Be able to use a USB-CAN analyzer to watch arbitration, error frames, and the bus load curve
The lesson · plain words
CAN is the «nerve bus» of vehicles and robots: noise-immune, multi-master, arbitration without loss. Chassis control, joint drives, and AGV command streams almost all run on CAN / CAN FD. This page covers its soul: arbitration.
Arbitration: the smaller ID speaks first, the loser waits its turn
CAN uses wired-AND levels: a dominant bit (0) overrides a recessive bit (1). When several nodes transmit simultaneously, they compare bit by bit — whoever sends 0 keeps going, and whoever sends 1 and hears 0 on the line drops out (it heard itself being overridden). Watch the animation and drag the two IDs: the smaller ID wins the bus, and the losing message does not have to resend a single bit (arbitration is lossless).
That is why CAN needs no master: priority is baked into the ID. Safety messages (e-stop) get small numbers, logs get large ones — natural prioritization.
Interactive figure · drag it and see
Electricals and error handling
Two differential wires (CANH/CANL), 120 Ω termination at both ends (matching the bus impedance: missing it causes reflections, doubling it adds load). Wiring topology: a trunk with short stubs (stubs <30 cm); messy star wiring is a bit-error factory.
The error frame mechanism: any node that detects an error broadcasts an error frame, voiding the message for everyone. Error counters are graded: error active → error passive → bus off (automatic offline). «A node suddenly drops off» usually means its own error counter blew past the limit into bus off.
CAN FD and RS485
Classic CAN carries 8 data bytes and tops out at 1 Mbps. CAN FD speeds the data phase to 5 Mbps and 64 bytes — one full frame carries a complete joint state. New projects go straight to CAN FD; the cost is nearly identical.
RS485 lives in another world: half-duplex master-slave polling, no arbitration, cheap and rugged — it is Modbus's carrier. Good for low-speed configuration / telemetry, wrong for real-time control. Do not make RS485 do CAN's job.
How deep to go
Explain the arbitration process and its losslessness; compute bit timing (sample point at 75–80%); read error frames with candump / a logic analyzer; build a test bus with 120 Ω at both ends. With these, you can handle CAN-layer problems independently.
Three questions — do you really get it?
1. During CAN arbitration, the node with the smaller ID...?
- AWins the bus with priority, and arbitration is lossless
- BTransmits afterwards
- CNeeds to retransmit
Show answer
A. Dominant overrides recessive, bit by bit, and the smaller ID wins. The losing node automatically waits for the next round; its data is never corrupted.
2. What do the 120 Ω resistors at both ends of a CAN bus do?
- ALimit current
- BPull the level up
- CImpedance matching — absorb reflections and keep the signal intact
Show answer
C. The twisted pair's characteristic impedance is about 120 Ω. Matched termination prevents reflections — without it, a long bus produces errors outright.
3. A node «suddenly goes offline and stops transmitting». Most likely cause?
- AError counter overflow into the bus off state
- BPower cut
- CID conflict
Show answer
A. error passive → bus off is the protocol's self-protection. Investigate why it kept sending error frames (usually wiring or baud rate).
Keep going deeper
Links without a marker stay on this site; the rest point to the Chinese edition until its translation batch lands.
How it fails in the field
- Installing only one 120 Ω: a short bus may «look fine», then collapse once it gets longer or more nodes join
- Assigning IDs carelessly: low-priority frames starve under load, showing up as «one joint occasionally stops responding»
- Using CAN like Ethernet: chunking a large payload across frames sends latency straight out of control
- Ignoring error counters: a node's behavior changes after entering error passive, and you cannot find the cause
You have learned enough when
The Chinese edition carries the full topic page for Field buses: CAN / RS485 / CAN FD, including the complete outline, depth requirements per career track, and further derivations: robotguide.dev/k/E18/.