CAN physical layer: voltages, wiring and termination

01Why CAN uses differential signalling

A vehicle is a hostile place for data. Ignition coils, injector drivers, inverter switching edges and starter-motor currents all inject noise into the harness, and the ground at the front of a truck can sit volts away from the ground at the rear while the engine cranks. Single-ended signalling, which compares one wire against local ground, would read every one of those disturbances as data.

High-speed CAN, specified in ISO 11898-2, avoids the problem by transmitting each bit as the difference between two wires, CAN-H and CAN-L, routed together as a twisted pair. Noise couples almost equally into both conductors, so it moves both voltages up or down together. The receiver ignores that common shift and evaluates only the difference.

The twist matters as much as the pair. Each half-turn reverses the orientation of the loop formed by the two wires, so a magnetic field induces opposing voltages in successive half-turns and they largely cancel. It works in reverse too: equal and opposite currents mean the pair's own fields cancel, so the bus stays quiet towards the radio and every sensor line routed beside it.

02Recessive and dominant: the two bus states

CAN does not drive ones and zeros symmetrically. A logical 1 is the recessive state: every transceiver releases the bus and the termination pulls both lines to the same voltage, nominally 2.5 V. A logical 0 is the dominant state: at least one transmitter actively pulls CAN-H up towards 3.5 V and CAN-L down towards 1.5 V, producing a difference of about 2 V.

Because a released bus cannot hold a dominant level, one dominant transmitter always overrides any number of recessive ones. This wired-AND behaviour is not a side effect. It is the foundation of non-destructive arbitration, of acknowledgement and of error signalling, all described in CAN frames and arbitration.

Receiver thresholds and noise margin

Receivers do not need the full 2 V. ISO 11898-2 requires a differential voltage below 0.5 V to be read as recessive and above 0.9 V to be read as dominant; the band in between is undefined. Even at the bottom of its tolerance, a healthy transmitter delivers 0.6 V of margin above the dominant threshold, and that margin is what absorbs cable resistance, connector losses and reflections on a long harness.

It is also why an extra termination resistor rarely kills a bus outright: the lower load makes the dominant level sag towards the 0.9 V limit, and the network works on the bench but fails intermittently in the vehicle. Measured margins say far more about bus health than a simple “it communicates” check.

Common-mode range and ground offset

Every ECU references its transceiver to its own local ground, and in a vehicle those grounds are never identical. Large currents in the ground return paths create offsets of a volt or more between distant modules, especially during cranking. The differential receiver tolerates this as long as both bus lines stay inside its common-mode range. The early editions of ISO 11898-2 required correct reception from −2 V to +7 V; the 2016 edition widened the receiver requirement to −12 V to +12 V and requires the bus pins to survive −27 V to +40 V, and many automotive transceivers go further still.

That range is generous but not infinite. Ground offsets, alternator ripple and ground loops created by careless installation all consume it. Much of the guidance in Installation best practices exists to keep this budget intact.

03Termination: why 120 Ω at both ends

A CAN bus is a transmission line. A transceiver edge rises in tens of nanoseconds, so even a few metres of cable behave as a line with a characteristic impedance rather than as a simple wire. When an edge reaches an unterminated end, it reflects and collides with the bits that follow. The cure is a termination resistor at each physical end of the backbone, matched to the cable's nominal 120 Ω impedance, which absorbs the travelling wave instead of reflecting it.

Seen from the bus, the two terminators appear in parallel:

The terminators do a second job that is easy to overlook. They are the load that turns the transmitter's drive current into the dominant differential voltage, and they discharge the bus capacitance at the end of every dominant bit so that the recessive state returns quickly. That is why both missing and surplus termination cause trouble, just different kinds of trouble.

Split termination

Many ECUs that sit at a bus end use split termination: two 60 Ω resistors in series between CAN-H and CAN-L, with their midpoint connected to ground through a small capacitor, commonly 4.7 nF. At DC the pair still measures 120 Ω, so the ohmmeter rules above are unchanged. At high frequency the capacitor gives common-mode noise a low-impedance path to ground, which noticeably reduces emissions. Some transceivers add a dedicated split pin that holds the midpoint at 2.5 V to stabilise the recessive level further.

In vehicles the terminators usually live inside two ECUs at opposite ends of a segment, so unplugging one of those ECUs during a repair also removes a terminator.

04Cable, twist and stubs

ISO 11898-2 describes the medium in electrical terms rather than prescribing a cable part number. The reference values that network designers work to are these:

Shielding is optional. Passenger cars overwhelmingly use unshielded twisted pair, while heavy-duty networks offer both: SAE J1939-11 specifies shielded twisted pair and SAE J1939-15 a reduced, unshielded physical layer. Twist quality matters in practice. A tight, consistent twist keeps both conductors equally exposed to interference, and untwisting the pair over a long distance at a connector or splice removes the protection exactly where noise is often worst.

Stub lengths

A stub is the branch between the backbone and a node's transceiver. Because a stub is unterminated, it reflects energy back onto the backbone, and those reflections are harmless only while the stub is short compared with the edge rise time. Faster bits mean tighter limits:

Two practical consequences follow. First, a harness that works at 250 kbit/s can fail when the same topology is reused at 500 kbit/s or with a CAN FD data phase, because the acceptable stub length shrinks as the edges get faster. Second, every device added to an existing bus with a long flying lead creates a new stub. Keep added stubs as short as the installation allows, and never extend a backbone by running a long cable out to a distant node and back.

Passive stars can work at classical CAN rates with careful design, but they fight the transmission-line model at every edge. Architectures that need stars or long branches at CAN FD data rates rely on signal-improvement transceivers, covered in Classic CAN vs CAN FD.

05Transceivers: the analog front end

An ECU's CAN controller speaks in logic levels on two pins, TXD and RXD. The transceiver translates between those pins and the bus: it drives the dominant state, releases the bus for recessive and compares the differential voltage against the receive thresholds. Because the receiver is always active, a transmitter reads back every bit it sends, which is how it knows it has lost arbitration or that a bit was corrupted on the way.

Transceiver generations follow the editions of ISO 11898-2:

Automotive transceivers add features that matter in the field: a TXD dominant time-out that releases the bus if a crashed microcontroller holds TXD low, so one faulty node cannot block the network; slope and symmetry control to limit emissions; and undervoltage protection that keeps the output recessive while the supply is out of range.

Low-power modes and wake-up

A parked vehicle must keep its quiescent current to a minimum, so most ECUs shut down completely and leave only the transceiver listening. In low-power mode an ISO 11898-2:2016 transceiver can release its 2.5 V bias and let the lines fall to about 0 V while it watches for a wake-up pattern, a defined sequence of dominant and recessive phases. Transceivers with selective wake-up go further and decode frames, waking their ECU only for a specific wake-up frame. This partial networking keeps part of the vehicle asleep while another part works, which matters for sleep current budgets and for electric vehicles that run charging functions for hours.

For the installer, the consequence is simple: anything permanently connected to the bus must leave it untouched when the vehicle sleeps. A device that transmits, or holds the lines away from their sleep levels, can keep the whole network awake and flatten a battery over a weekend. Network management and the power states of CAN in electric and hybrid vehicles cover this side in detail.

06EMC: keeping the bus quiet and immune

Electromagnetic compatibility runs in both directions. Emissions come mostly from asymmetry: if CAN-H and CAN-L do not switch at exactly the same moment with mirrored slopes, the common-mode voltage jumps at every edge and the harness radiates. Immunity depends on the same symmetry, so that injected noise stays common-mode and the receiver rejects it. The usual measures are:

• Common-mode chokes on the bus lines, which suppress common-mode current while passing the differential signal. They must suit the bit rate; a poorly matched choke can ring at CAN FD edges.
• Split termination with a midpoint capacitor, which shunts common-mode noise to ground at the ends of the bus.
• ESD protection with low and well-matched capacitance on both lines, particularly on CAN FD buses where every picofarad slows the edges.
• Symmetrical routing: equal lengths, tight coupling and no long untwisted sections at connectors keep the pair balanced.

They are verified against an established family of standards: CISPR 25 for emissions, ISO 11452 for radiated and bulk-current-injection immunity, ISO 7637-2 and ISO 7637-3 for transients on supply and signal lines, and IEC 62228-3 for CAN transceivers themselves. At vehicle level, electromagnetic compatibility is regulated in Europe and many other markets through UN Regulation No. 10.

07Failure modes and their electrical fingerprints

High-speed CAN is not fault tolerant in the way the low-speed bus described in Bitrates and bus types is. A single wiring fault can silence a whole network, and some faults leave it limping in a way that hides the problem. Reasoning from the bus states makes the behaviour predictable: any fault that prevents CAN-H from rising above CAN-L by 0.9 V makes a dominant bit impossible, and the bus falls silent.

1. 01
   Look before you measure

   Inspect the connectors at both ends of the segment and any recently added device. Most physical-layer faults are mechanical: pushed-back pins, corrosion, chafed insulation.

2. 02
   Resistance, unpowered

   With the network asleep or the battery disconnected, measure CAN-H to CAN-L (expect ≈ 60 Ω), then each line to ground and to battery positive (expect a high resistance, in the kilo-ohm range or above).

3. 03
   Voltages, bus awake

   Each line should sit close to 2.5 V to ground on an idle bus. With traffic, a multimeter averages the bits: CAN-H reads slightly above 2.5 V, CAN-L slightly below, and the two add up to roughly 5 V.

4. 04
   Oscilloscope

   Probe CAN-H and CAN-L against ground on two channels, or use a differential probe. Look for clean rectangular bits, a dominant differential of about 2 V, mirrored edges and ringing that settles well before the sample point.

5. 05
   Divide and conquer

   If a fault is present, disconnect ECUs or split the backbone at a connector and repeat the measurements on each half until the fault is localised.

08Frequently asked questions