CANopen, J1939 or Your Own Protocol? Choosing a Vehicle Bus for a Traction System

This decision gets made in an afternoon, early, usually by whoever is writing the first firmware, and then the machine lives with it for fifteen years. It determines what diagnostic tools exist, who can service the vehicle, how painful it is to add a component in year four, and whether your customer’s fleet system can see your machine at all.

It is worth half a day of thought rather than an afternoon of default.

The three options, honestly characterised

CANopen J1939 Proprietary
Identifier 11-bit, allocated by node ID 29-bit extended, PGN based Whatever you choose
Addressing model Node ID per device; message IDs derive from it Source address per device, claimed at startup Yours to define and document
Data model Object dictionary; PDO for cyclic, SDO for configuration Parameter Groups (PGN) containing defined parameters (SPN) None until you write one
Device profiles Yes — DS401 for I/O, DSP402 for drives, others Standardised parameter definitions across vendors No
Typical bitrate Anything from 20 kbit/s to 1 Mbit/s 250 kbit/s classically; 500 kbit/s in J1939-14 Your choice
Strong when A closed machine with a known device list, drives and I/O from CANopen-speaking vendors Mixed-vendor vehicles, off-highway equipment, anything a customer’s telematics will read Tiny closed systems, or when you own every node and always will
Costs you Node ID management, configuration complexity, less natural for multi-master More implementation work up front, heavier identifiers, 8-byte frames need transport protocol for longer data Every tool, every document, every training session, forever

What the hardware you already bought has decided for you

Often most of this is settled before you start. The traction controllers in common use lean CANopen:

  • Sevcon Gen4 implements CANopen with the DS301, DS401 and DSP402 profiles.
  • Curtis 1232E / 34E / 36E / 38E are CANopen compatible out of the box, and explicitly allow other 11-bit or 29-bit protocols to be built in VCL.

If your drive is one of those and your bus is otherwise yours, CANopen on the traction side is the path of least resistance — you inherit a working implementation and the vendor’s own tooling rather than reimplementing their protocol badly.

The five questions that actually decide it

1. Who else is on this bus, and who owns them?

If the answer is “a BMS from one vendor, a charger from another, a display from a third and a drive from a fourth”, you need a protocol they already share, or a gateway. If the answer is “three boards I designed”, proprietary is genuinely defensible.

2. Does your customer need to see this machine with tools they already own?

This is the question people skip, and it is frequently decisive. A customer running a mixed fleet with existing J1939 telematics does not want your proprietary frames. Being readable by the tools they already have can be worth more than any technical elegance.

3. Who services it in year six?

A dealer technician with a generic CAN tool can make progress on CANopen or J1939. On your proprietary protocol they can do nothing without software you have to write, maintain, version, and distribute. Count that cost honestly — it is the one that is always underestimated.

4. How much data, how fast, and over what cable length?

Bitrate and bus length trade against each other, and 8 bytes per frame is a real constraint. If you have a value that does not fit in 8 bytes you need a transport mechanism, and all three options handle that differently. If your harness is long or electrically unpleasant, a lower bitrate buys you robustness — Curtis notes applications running happily at 125 kbit/s even without twisted pair, which tells you how much margin the lower rates give.

5. Is there a standard your market expects?

Marine electrification means NMEA 2000, which is built on J1939 and brings its own certification process. Off-highway and agricultural equipment leans J1939. Industrial drives and material handling lean CANopen. Fighting the convention of your market is a cost with no benefit.

The answer is often two buses

A single bus asked to carry high-rate traction control, battery management, a charger handshake, accessory switching and an operator display is a bus with a loading problem and a blast radius problem. One misbehaving accessory node should not be able to disturb torque control.

A pattern that works well on machines of this size:

  • A traction bus — controller, motor feedback, BMS. Short, fast, few nodes, electrically clean, nothing added after design freeze.
  • A vehicle or accessory bus — display, lighting, body controllers, telematics, charger. Lower rate, more forgiving, allowed to grow over the product’s life.
  • A gateway between them that republishes exactly the values the vehicle side is allowed to see, and nothing else.

That gateway is also where protocol translation lives, which means the traction side can stay CANopen because the drive vendor made it so, while the vehicle side can be J1939 because your customer’s fleet system needs it. You stop having to win the argument.

It costs one more node. It buys you isolation, a clean place to log, a single point to change when a component is superseded, and the ability to answer “can you add telematics” with yes.

If you do go proprietary, do these three things

Proprietary is a legitimate choice for a small closed system. It becomes a liability through neglect, not through the decision itself. So:

  1. Write the specification before the firmware. Every identifier, every byte, scaling, units, sign, endianness, transmit rate, timeout behaviour. A document, versioned, not comments in source.
  2. Reserve space deliberately. Leave identifier ranges and spare bytes for things you have not thought of. Retrofitting structure into a full identifier map is miserable.
  3. Build the diagnostic tool as you go. If reading your own bus requires a developer and a laptop with your toolchain, you have built a product only you can service.

What we would ask you

When someone brings us this decision, these are the questions, and you can work through them yourself:

  • List every node that will ever be on the bus, including the ones in version 3.
  • For each, name the protocol it already speaks and whether that is changeable.
  • Name the longest cable run and the electrical environment it passes through.
  • Name who diagnoses a fault in the field, and what is in their hand when they do.
  • Name the customer’s existing systems that will want to read your machine.

Those five answers usually make the decision obvious, and frequently make it “two buses and a gateway”.

We design these

Multi-bus vehicle architectures, custom CAN nodes and gateways, and the firmware for them — see custom electronics and control system development. Related reading: getting a third-party controller onto your own display and BMS to controller integration.

Email matt@electricdrivesolutions.com or call (717) 421-9988.

Controller protocol support is from the Sevcon Gen4 Applications Reference Manual and the Curtis 1232E family manual. Bitrate conventions for J1939 reflect classic J1939 and J1939-14; confirm against the current standard for your application.

Back to blog