Most integration delays on custom battery packs start in the same place: the data link between the battery management system and the host device. The pack charges, the device runs, and yet the firmware team and the battery team spend weeks arguing about which register carries state of charge, what units it uses, and what the system should do when the link goes quiet. None of that is a chemistry problem. It is a specification problem, and it is avoidable.
This article lays out what belongs in the communication section of a custom smart battery pack requirement. It covers the two buses you are most likely to meet – CAN bus on industrial equipment, AGVs and energy storage racks, and SMBus on portable and embedded products that follow smart battery conventions – plus the field list, scaling rules and fault behaviour that should be written down before firmware work starts.

CAN bus and SMBus solve different problems
CAN bus (Controller Area Network) is a differential two-wire bus designed for multi-node equipment. In an AGV, an energy storage rack or an industrial machine, the pack is one node among several: the host controller, the charger, a display, and other drives all share the same pair of wires. Messages carry prioritised identifiers, so a safety-relevant frame does not wait behind cosmetic traffic.

SMBus (System Management Bus) is a two-wire management bus used between a smart battery and its host and charger, mainly in portable and embedded products. Its value is the shared data model: the host can read voltage, current, temperature and state of charge from any compliant pack without a custom protocol effort. The current release of the specification is version 3.3.1, published in October 2024 by the System Management Interface Forum.
The choice is usually set by the device architecture rather than by preference: one battery talking to one host points to SMBus, while a pack that must coexist with chargers, HMIs and machine controllers on a shared network points to CAN.
| Aspect | CAN bus | SMBus |
|---|---|---|
| Typical setting | Machine networks: AGVs, ESS racks, industrial equipment | Portable and embedded products: pack-to-host and pack-to-charger link |
| Nodes | Multi-node, multi-master | Usually one battery, one host, optionally a charger |
| Message style | Short frames with prioritised identifiers | Register-style reads and writes |
| Fault signalling | Broadcast fault frames plus heartbeat | Alarm bits and status registers |
| Hardware implication | Transceiver per node, bus termination | Two signal lines; defined behaviour if the bus fails |
| Reference documents | CiA 418 / CiA 419 CANopen device profiles | SMBus specification and smart battery data conventions |
What the CANopen battery profile already decided for you
If your project uses CANopen, much of the groundwork exists. The CiA 418 device profile specifies the CANopen interface for battery modules, written so that a compliant module interoperates with a charger built to the companion CiA 419 charger profile. A compliant battery module must provide a defined minimum dataset – battery type, battery capacity, number of cells, the maximum charge current permissible, and battery temperature. Optional data includes identifiers, charge history, battery voltage, state of charge and requested current. A default transmit process data object (TPDO) sends temperature and status on a timer, optional TPDOs carry voltage, current requests and state of charge, and the charger returns its status through a receive PDO. Network management and heartbeat behaviour follow the CANopen application layer defined in CiA 301.
Even if the project does not use CANopen, the mandatory-versus-optional split is a useful template for any CAN-based pack: identify the minimum dataset needed for safe charging, treat everything else as optional, and give each optional field an explicit unit and scaling.
| Data item | CiA 418 status | Why the host or charger needs it |
|---|---|---|
| Battery type | Mandatory | Determines the charge regime |
| Battery capacity | Mandatory | Sets charge current and charge-time expectations |
| Number of cells | Mandatory | Cell-level supervision and balancing limits |
| Maximum charge current permissible | Mandatory | Protects cells from over-current charging |
| Battery temperature | Mandatory | Charge-window supervision |
| Identifiers and charge history | Optional | Traceability and fleet management |
| Battery voltage | Optional | Pack-level monitoring |
| State of charge | Optional | Runtime estimation |
| Requested current | Optional | Battery-led charge control |
Source: CAN in Automation (CiA), CiA 418 device profile for battery modules.
The field list to write down either way
Whatever the bus, the host needs roughly the same dataset: pack identity (manufacturer, chemistry, serial number, cell count), capacity (rated and full-charge), state of charge, state of health, pack voltage, current, temperatures (with the number of probes and their physical locations), protection status flags, cycle count, a remaining-runtime estimate, and fault registers. For the electrical and safety envelope around those values, IEC 62619 governs the safety requirements that a pack for industrial applications must satisfy, which is the context in which most CAN-based packs are specified.
The discipline that prevents integration failures is simple: every field gets a unit, a scaling, an offset, a byte order, an update rate, and a defined behaviour when the value is unavailable. Two classic failures show why. A state of charge sent as 0 to 100 when the host expects 0 to 1000 will read as a nearly empty pack for weeks. A temperature sent in 0.1 kelvin increments when the host assumes 0.1 degree increments will read as impossibly hot. Both are caught by a single sentence in the specification and both are expensive to find during commissioning.
| Item | What to specify | Failure mode it prevents |
|---|---|---|
| Parameter name and ID | Register address, object index or CAN identifier | Host polling an ID the BMS never sends |
| Unit and scaling | Unit, multiplier, offset, signed or unsigned | State of charge or temperature misread by an order of magnitude |
| Byte order | Big-endian or little-endian for each multi-byte field | Swapped current polarity or garbled values |
| Update rate | Period for cyclic data; on-change rules for events | Stale data mistaken for a frozen pack |
| Validity rules | What the BMS sends when a measurement is unavailable | Host acting on stale or default values |
| Fault flags | Bitmap semantics, latching or momentary, who clears them | Faults missed, or cleared by the wrong side |
Fault flags, heartbeats and timeout behaviour
Fault reporting deserves the same rigour as the measurement list. Define the fault bitmap: which bits exist, what each one means, whether it latches or clears automatically, who is allowed to clear it, and how (host command, fault removal, or power cycle). A fault bit that two sides each assume the other will clear is a support ticket waiting to happen.
Heartbeat and timeout behaviour matters just as much. Specify who produces the heartbeat, who consumes it, the timeout value, and the action on each side. On the host side, silence from the pack might mean derating, a safe state, or an orderly shutdown – that choice is an application decision and belongs in the document, not in firmware folklore. On the pack side, a charger that goes quiet mid-charge needs a defined response, for example reverting to a reduced current or stopping charge, with the condition stated precisely.
On SMBus-based systems the battery is the data authority and the host reads on its own schedule, so the specification should state the alarm mechanism, what triggers it, and how the host reacts. The same timeout discipline applies: a stuck bus is a foreseeable condition, and the SMBus specification itself is maintained and versioned precisely so that such behaviours can be referenced rather than reinvented.
Some products end up with both buses: a pack that reports smart battery data to the host over SMBus while also answering the machine network on CAN. Dual-interface packs are workable, but they double the specification surface, and the two interfaces must agree on every shared value – a state of charge that differs between buses turns the disagreement into the product’s most visible bug. If a dual-interface requirement exists, it should be stated as such in the requirement document, with one side named as the data authority for each parameter.
Verify the interface before firmware integration starts
A communication specification is only as good as its verification. Four bench steps catch most problems early:
- Decode live traffic with a bus analyser and compare every field against the BMS display and a calibrated meter – not just against the specification document.
- Inject faults (a disconnected temperature probe, a simulated over-temperature) and confirm that the correct bit is set with the agreed severity and latching behaviour.
- Run the timeout test: silence each side in turn and confirm the specified safe response happens, on both sides.
- Archive the captures as acceptance evidence, so the interface has a baseline record for future firmware revisions.
Decoding proves the transport is honest; it does not prove the numbers are accurate. Verifying that reported state of charge tracks reality needs its own bench procedure, described in our article on BMS state of charge accuracy.
Where the communication section belongs in the pack requirement
The communication specification should not live in a firmware folder; it belongs in the pack requirement document that the RFQ is built around. Our custom battery pack RFQ checklist treats the data interface as one of the engineering inputs that should be fixed before quotes are requested, because suppliers price and design the BMS around it. A plain-language overview of what the BMS itself contributes is in our guide to the battery management system.
If you are defining the data interface for a new pack and want it reviewed before the firmware sprint starts, contact our engineering team with your host-side requirement and we will respond with the matching pack-side dataset.

