CAN Bus Wiring and Termination: The Complete Guide

This article is part of the CAN & CANopen family and pairs with CAN Bus Protocol Explained, which covers the arbitration, frame formats, and error counters — the logical layer — while this one covers the physical layer. For the full family landing with CANopen, J1939, and DeviceNet, visit the CAN & CANopen category hub.

Most CAN faults are wiring faults. Not protocol bugs, not controller settings — a missing terminator, a stub that's too long, a shield grounded at both ends, a connector that passes a continuity test but fails under vibration.

The frustrating part is that CAN degrades gracefully enough to hide these. A network with one terminator instead of two usually works. It works right up until the day it doesn't, and then the fault log says "bus-off" and tells you nothing about why.

This guide covers what the physical layer actually requires, what to measure and what those readings mean, and the wiring mistakes that produce intermittent failures. The electrical values come from ISO 11898-2, the high-speed transceiver standard.

First, a nuance about what the standard specifies

ISO 11898-2 specifies the transceiver — the high-speed medium attachment. It defines what a transceiver must drive, what it must receive, and how it must behave.

It explicitly does not specify the cable. The physical media dependent sublayer is outside its scope. There is no ISO 11898-2 requirement for conductor size, impedance, or connector type.

So where does 120 Ω come from? It's implied rather than mandated. Every transceiver test condition in the standard uses an effective bus load of 60 Ω — which is exactly what two 120 Ω terminators in parallel produce. The transceiver's entire specification assumes you built the bus that way. Cable and connector specifics come from the higher-layer protocol (CANopen, J1939, DeviceNet each specify their own) or from the equipment vendor.

That distinction matters when someone asks you to cite a requirement. The correct answer is that 120 Ω at each end is what the transceiver specification presumes, and what the higher-layer standard for your industry actually requires.

Termination: two resistors, both ends, no exceptions

A CAN bus is a transmission line. Signals reaching an unterminated end reflect back down the cable and interfere with the bits that follow. At low bit rates over short distances you may get away with it. At 500 kbit/s over 100 m you will not.

The rule is simple and gets broken constantly:

One 120 Ω resistor at each physical end of the bus. Two total. Never more, never fewer, never in the middle.

Measure across CAN_H and CAN_L with the bus powered down and you should read about 60 Ω. That single measurement is the fastest diagnostic in CAN work:

Extra terminators are as damaging as missing ones — they load the bus down, reduce the differential voltage every receiver sees, and push nodes toward the receiver's dominant threshold. The classic cause is a device with a switchable internal terminator that someone enabled "to be safe," in the middle of a line that already had two.

Split termination

A refinement worth knowing: instead of one 120 Ω resistor, use two 60 Ω resistors in series with a capacitor from their midpoint to ground. The standard's own transceiver test circuit includes an optional split-termination capacitor for exactly this reason.

The benefit is common-mode noise suppression. Differential signalling handles differential noise well; common-mode noise — both lines pushed together by an inverter, a contactor, a long parallel cable run — is what the split capacitor drains away. On electrically noisy machines, split termination measurably improves reliability at negligible cost.

The DC measurement is unchanged: still 60 Ω across the pair.

Voltage levels: what to measure and what it means

The transceiver standard gives precise values, and they're what your multimeter or scope should show.

Receiver thresholds define what a node will actually accept:

Dominant requires at least 0.9 V differential. Recessive requires 0.5 V or less. Between those lies the undefined zone — and it's the reason a bus with too many terminators or excessive cable loss gets flaky rather than failing outright. Nodes closest to the transmitter still resolve dominant bits; nodes at the far end drift into the gap and start reporting errors.

Common mode range is ±12 V. Receivers must work correctly with the whole bus shifted that far relative to their local ground. That sounds generous until you meet a machine with a genuine ground potential difference between cabinets, which is why CAN grounding deserves the attention below.

A practical multimeter check on an idle powered bus: both lines should sit near 2.5 V with respect to ground, and near 0 V differential. CAN_H stuck at 0 V, CAN_L at 5 V, or a differential reading of several volts at idle all point at a wiring fault before you touch any software.

Topology: line, not star

CAN is a linear bus. One trunk, terminated at each end, with devices connected along it.

Stubs — the drops from the trunk to each device — are the compromise everyone makes and many get wrong. A stub is an unterminated branch, and it reflects. Keep them as short as the mechanical layout permits; the higher-layer standard for your industry gives the specific limits, and they tighten sharply as bit rate rises.

Star topologies are what break networks. Several long branches from a central point create multiple unterminated ends, and there's no legal way to terminate them all. If a star is mechanically unavoidable, use a repeater or a proper CAN hub that terminates each segment — not a passive junction box.

Daisy chaining through devices is cleaner than a trunk-and-stub arrangement when connectors allow it, since it eliminates stubs entirely. The trade is that removing a device breaks the bus unless the connector maintains continuity.

Bit rate and bus length

Arbitration is what fundamentally limits length. During arbitration, a bit must propagate from the transmitting node to the far end of the bus and back within a single bit time — that's how a node learns it lost arbitration before the next bit begins. Longer bus, longer propagation, longer minimum bit time, lower maximum bit rate.

This produces the familiar inverse relationship: 1 Mbit/s is short-range, and lower rates buy distance. The transceiver standard specifies bit rate tiers at the device level — up to 1 Mbit/s for classic operation, with optional support for up to 2 Mbit/s and up to 5 Mbit/s in the CAN FD data phase — while the achievable length at each rate comes from your cable's propagation delay and your network's topology.

Two field notes. Every node must be configured for the same bit rate. A single mismatched node generates error frames continuously and can drive the whole bus into fault states — and its own controller into bus-off. And the transceiver has a loop delay of up to 255 ns, which is real budget consumed before the cable even starts, and is why long buses at high rates fail even when the cable maths looks fine.

Grounding and shielding

Shield at one end only. Grounding both ends creates a loop that couples current into the shield and, through it, into the pair. One end, at the point with the most reliable ground reference.

A ground reference between nodes is required, not optional. The ±12 V common mode range is generous but finite. Nodes on separate power systems with no common reference can exceed it, at which point receivers stop resolving bits correctly while every DC measurement still looks fine. Where a common ground genuinely can't be established, galvanic isolation is the answer.

Twisted pair matters. CAN's noise immunity comes from differential signalling, and differential signalling relies on both conductors picking up the same interference. Untwisted parallel conductors don't, and the noise becomes differential — the one kind the receiver can't reject.

The faults that actually happen

A field checklist, ordered by how often each one turns out to be the answer.

Termination wrong. Measure 60 Ω first, always. Missing, extra, or wrongly placed terminators cause more CAN problems than everything else combined.

One node in bus-off, blamed for the network. A node that removed itself is often the victim of a wiring problem, not its cause. Check the bus before replacing the device.

Intermittent errors under vibration or temperature. A marginal connector. Continuity tests pass at rest; the fault appears only when the joint moves or the cabinet warms. Wiggle-test connectors while monitoring error counters.

Errors that appear only when a specific machine runs. Common-mode noise from a VFD, contactor, or welder. Check shield grounding, cable routing away from power conductors, and consider split termination.

Everything worked until a device was added. Either its internal terminator got enabled, or its stub is too long, or the added length pushed the bus past its limit for the configured bit rate.

One node never communicates, others are fine. Bit rate mismatch, swapped CAN_H and CAN_L on that drop, or a bad transceiver. Check the differential voltage at that node's connector while the bus is active.

Whole bus dead, 60 Ω reads fine. Look at the supply and the ground reference before the data pair — and confirm the two lines aren't swapped somewhere along the trunk.

FAQ

Why 120 ohms? It matches the characteristic impedance of the twisted pair used for CAN, so signals reaching the end are absorbed rather than reflected. The transceiver standard doesn't mandate it directly, but its entire specification is written around a 60 Ω effective bus load — which is two 120 Ω terminators in parallel.

Can I use one 60 Ω resistor in the middle instead of two 120 Ω at the ends? No. The DC measurement would match, but termination works by absorbing signals at the point where the cable ends. A resistor in the middle leaves both physical ends unterminated and reflecting.

What should I measure across CAN_H and CAN_L? About 60 Ω with the bus unpowered. Roughly 120 Ω means one terminator or a break; around 40 Ω means three. This is the single most useful CAN measurement.

Do all CAN devices need a terminator? No — exactly two devices or connectors on the entire bus carry termination, at the two physical ends. Many devices include a switchable internal terminator, which is convenient at the ends and a problem everywhere else.

How long can a CAN bus be? It depends on bit rate, because arbitration requires a signal round trip within one bit time. Lower bit rates permit longer buses. The specific limits for your installation come from your higher-layer standard and your cable's propagation delay, with the transceiver's 255 ns loop delay consuming part of the budget.

Should the shield be grounded at both ends? No — one end only, to avoid a ground loop. The nodes still need a common ground reference through the power system or a dedicated conductor, since the receiver's ±12 V common mode range is finite.

What causes bus-off? A node's error counters passed the limit and it removed itself from the bus. The trigger is usually physical — bad termination, noise, a bit rate mismatch, or a failing transceiver. Treat bus-off as a symptom and investigate the wiring rather than replacing the node first.