Bitrates and bus types: from single-wire CAN to high-speed networks

01The bit rates you will meet

ISO 11898 allows any bit rate up to 1 Mbit/s for classical CAN, but practice has converged on a short list. Each value has a history and a typical home:

Above these sit the CAN FD data-phase rates, typically 2, 5 or 8 Mbit/s, which apply only between the bit-rate switch and the CRC delimiter of an FD frame; see Classic CAN vs CAN FD. The arbitration part of every FD frame still runs at a nominal rate from the table, most often 500 kbit/s.

02Inside one bit: time quanta and segments

A CAN controller does not sample a bit at an arbitrary moment. It divides the nominal bit time into time quanta (tq), each a whole number of CAN clock periods, and groups them into four segments:

The complete bit must contain 8 to 25 time quanta. The sample point sits at the boundary between Phase_Seg1 and Phase_Seg2 and is usually quoted as a percentage of the bit time:

Choosing the sample point

A late sample point gives the signal the most time to cross the bus and settle, which favours long networks. An earlier one leaves more room in Phase_Seg2 for clock tolerance and resynchronisation, which favours robustness. CANopen recommends 87.5 %, with roughly 85 to 90 % tolerated between nodes. CAN in Automation's CiA 601-3 advises against nominal sample points later than 80 % in automotive CAN FD networks. Vehicle manufacturers fix exact values for every network in their own specifications.

What matters most is consistency. Every node on a segment must sample at nearly the same point. A node that samples much earlier or later than the others still works on a quiet bench and fails on a long harness, or only when it happens to be the one transmitting.

03Synchronisation: staying aligned without a clock wire

CAN carries no clock line, and every node runs from its own oscillator. Two mechanisms keep them aligned:

• Hard synchronisation at the start of frame: every receiver restarts its bit timing on the recessive-to-dominant edge of SOF, so all nodes begin each frame aligned to the transmitter.
• Resynchronisation during the frame: whenever a recessive-to-dominant edge arrives outside Sync_Seg, the controller lengthens Phase_Seg1 or shortens Phase_Seg2 by up to the synchronisation jump width (SJW), which in classical CAN is between 1 and 4 tq and never larger than Phase_Seg1.

Resynchronisation needs edges, and bit stuffing supplies them: no more than 10 bits can pass between two recessive-to-dominant edges in a valid frame. That bound becomes a requirement on oscillator accuracy, expressed by two classic conditions:

Quartz crystals, whose tolerance is in the tens of parts per million, rarely come close to these limits on classical CAN. The limits start to bite with ceramic resonators, which are far less accurate, and in CAN FD data phases, where every timing error is measured against a much shorter bit.

04Worked example: 500 kbit/s from a 40 MHz clock

A controller with a 40 MHz CAN clock, one of the frequencies CAN in Automation recommends for new designs, has to run a 500 kbit/s bus with a 2 µs bit time. Two valid configurations:

1. 01
   Fix the time quantum

   Divide the 2 µs bit time by a chosen N and check that the result is a whole number of 25 ns clock periods. N = 16 gives 125 ns (BRP 5); N = 20 gives 100 ns (BRP 4).

2. 02
   Place the sample point

   Choose Phase_Seg2 to put the sample point where the network specification requires it: 2 tq out of 16 for 87.5 %, 4 out of 20 for 80 %.

3. 03
   Split the remainder

   Divide the remaining quanta between Prop_Seg and Phase_Seg1, each within its 1–8 tq limit.

4. 04
   Set the SJW

   Use the largest permitted value, up to the smaller of 4 and the two phase segments, to maximise tolerance.

5. 05
   Check the tolerance

   Option A tolerates about 0.49 % of oscillator error and option B about 0.78 %, both far beyond a crystal's error. Option A buys a later sample point; option B buys margin.

Neither option is right in isolation. The correct one is whichever matches the sample point of every other node on the network, which is why bit-timing parameters belong in the network specification rather than in individual ECU designs.

05Bus length versus bit rate

During arbitration, a transmitter must see the bus level that results from every other node's bit before it samples its own. The signal has to reach the farthest node and come back, through two transceivers and the cable, within the part of the bit before the sample point. With a line delay of about 5 ns/m, a 40 m bus at 1 Mbit/s spends 400 ns, 40 % of the bit time, on the cable round trip alone, before any transceiver or controller delay is counted. That arithmetic is why the maximum length falls as the bit rate rises:

Road vehicles rarely approach these figures. A passenger car's networks span a few metres to a few tens of metres, and SAE J1939 limits a heavy-duty backbone to 40 m. In vehicles the binding constraints are usually stub length, topology and EMC, described in CAN physical layer, rather than raw length.

06Three physical layers, one protocol

The same CAN protocol runs over three standardised physical layers in road vehicles, plus a dedicated variant for truck–trailer links. They differ in voltages, wiring and termination and, crucially, in what happens when a wire fails.

High-speed CAN (ISO 11898-2)

The dominant physical layer in every vehicle class today. Its strengths are speed and simplicity: two wires, two terminators and an inexpensive transceiver. Its weakness is fault tolerance. A short between the lines, and most shorts to ground or battery, stop communication on the segment, so manufacturers split functions across several networks joined by gateways and one fault never takes down the whole vehicle. Every CAN FD network uses this physical layer.

Low-speed fault-tolerant CAN (ISO 11898-3)

Designed for comfort electronics whose wiring runs through doors, seats and mirrors and is exposed to chafing and crushing. In the recessive state, termination resistors in each node pull CAN-H towards ground and CAN-L towards the 5 V supply; a dominant bit drives them to about 3.6 V and 1.4 V. The transceiver monitors both lines continuously. If one wire breaks or shorts to ground or battery, or the two lines short together, it switches to single-wire transmission and reception on the healthy line and returns to differential mode once the fault clears. All nodes keep communicating, with reduced noise immunity.

The price is a ceiling of 125 kbit/s and a different termination scheme: each node carries its own resistors, typically from 500 Ω to several kilo-ohms, sized so that the network totals about 100 Ω per line. The familiar 60 Ω reading across CAN-H and CAN-L means nothing on this bus. Low-speed fault-tolerant CAN was widespread on comfort networks and has largely given way to high-speed CAN and LIN on newer platforms.

Single-wire CAN (SAE J2411)

The minimum-cost option: one signal wire referenced to ground. Normal communication runs at 33.3 kbit/s with dominant levels of about 4 to 5 V. A raised dominant level of about 10 to 12 V acts as a wake-up broadcast for sleeping nodes, and a high-speed mode at 83.3 kbit/s serves diagnostics and reprogramming. Without a differential pair, emissions and immunity depend on slow, shaped edges, which is why the bit rate stays low. Single-wire CAN is most closely associated with General Motors' GMLAN body networks, where up to 32 nodes share the line.

Truck–trailer CAN (ISO 11992-1)

Commercial-vehicle combinations add one more variant. ISO 11992-1 defines a point-to-point CAN link between towing vehicle and trailer at 125 kbit/s for 12 V and 24 V systems, which keeps communicating on one wire if the other is interrupted. Brake and running-gear data travel over it through the ISO 7638 connector, with the messages defined in ISO 11992-2, and other equipment in ISO 11992-3. Because the link is strictly point-to-point, each trailer in a combination has its own link to the vehicle in front.

07Why a bit-rate mismatch is never harmless

A node configured for the wrong bit rate does not simply fail to understand the traffic. Its controller sees stuff, form and CRC errors in almost every frame and, if it is allowed to transmit, answers each one with an active error flag. Those flags destroy valid frames for every other node on the bus. Transmitters retry, error counters climb, and the vehicle's ECUs start storing communication faults; on a busy bus, warning lamps can appear within seconds.

The same applies, more subtly, to the correct bit rate with an incompatible sample point or synchronisation setting: the node works most of the time and corrupts frames only with particular bit patterns or at particular temperatures. That is why the full bit timing, not just the headline rate, has to match the network specification.

08Verifying bit rate and timing on an oscilloscope

An oscilloscope confirms that a bus runs at its documented rate and that its bits are clean. Because stuffing limits runs of identical bits to five inside a frame, the narrowest pulse in any frame is exactly one bit time:

1. 01
   Probe

   Connect CAN-H and CAN-L to two channels referenced to chassis ground, or use a differential probe. Set the time base to show a whole frame, about 500 µs across the screen at 500 kbit/s.

2. 02
   Measure the narrowest pulse

   Place the cursors on the shortest dominant or recessive pulse. Its width is the bit time; compare it with the documented rate.

3. 03
   Check the edges

   Rising and falling edges should be clean and mirrored, and any ringing should have died away well before the sample point late in the bit.

4. 04
   Look for error flags

   Runs of six or more dominant bits mean that a node is signalling errors. A burst of them straight after a new device is connected points directly at that device's configuration.

5. 05
   Check the FD data phase

   On CAN FD, zoom into the data phase: the bits narrow after the BRS bit and widen again at the CRC delimiter. Data-phase edges are the most sensitive indicator of stub and topology problems.

A CAN FD data phase at 5 Mbit/s needs real bandwidth from the instrument: at least 100 MHz and several hundred megasamples per second, so that you see actual edges and ringing rather than a rounded approximation. Further patterns are collected in Field diagnostics.

09Frequently asked questions