CAN (Controller Area Network) and In-Vehicle Networks
1. Overview
CAN (Controller Area Network): a multi-master, message-based serial communication protocol in which multiple ECUs (Electronic Control Units) share a single two-wire bus and communicate autonomously by message priority, without a central controller. Its physical and data-link layers are standardized in ISO 11898.
CAN was introduced in 1986 by the German company Bosch to solve the explosion of wiring inside automobiles. As electronic control functions such as engine control, transmission, ABS, and instrument clusters grew rapidly through the 1980s, the traditional approach of connecting each function with point-to-point wiring increased wiring weight, cost, and failure rate exponentially. When the wiring harness of a high-end vehicle reached several kilometers in length and tens of kilograms in weight, a broadcast network in which "every ECU shares one common bus but selectively reads only the data it needs" became necessary. CAN was the answer to exactly this need, and its core value is that it dramatically reduced the number of wires while simultaneously securing real-time behavior and reliability.
The reason CAN has remained the de facto standard for vehicles and industrial sites for decades stems from three design philosophies. First, a message-oriented addressing scheme. A CAN frame contains no address of the sending or receiving node, only the content of the message (an ID). When one ECU sends a message with the ID "engine RPM," every node that needs that value selectively accepts it through a receive filter. As a result, nodes can be added or removed without changing existing communication, giving excellent scalability. Second, non-destructive bitwise arbitration ensures that even when a collision occurs no data is lost and the higher-priority message is transmitted without delay. Third, strong error detection and confinement mechanisms isolate faulty nodes by themselves, protecting the availability of the whole network in a vehicle environment where safety is paramount. These three combined made CAN the archetype of the "low-cost, high-reliability real-time fieldbus."
2. CAN Physical Structure and Layer Model
CAN defines only the Physical Layer and Data Link Layer of the OSI 7-layer model; higher network, transport, and application functions are handled by higher-layer protocols such as CANopen and J1939. The structure diagram below shows the electrical topology and the layer mapping together.
graph TB
subgraph layer["CAN Layer Model (ISO 11898)"]
APP["Application layer<br/>(CANopen·J1939·OBD-II)"]
LLC["Data link - LLC<br/>(acceptance filtering·overload notification)"]
MAC["Data link - MAC<br/>(framing·arbitration·error detection)"]
PLS["Physical layer - PLS<br/>(bit encoding·timing)"]
PMA["Physical layer - PMA<br/>(transceiver·differential signaling)"]
APP --> LLC --> MAC --> PLS --> PMA
end
subgraph bus["Physical bus topology"]
T1["120-ohm termination"] --- BUS["CAN_H / CAN_L twisted pair"] --- T2["120-ohm termination"]
E1["ECU1 - Engine"] --- BUS
E2["ECU2 - Transmission"] --- BUS
E3["ECU3 - ABS"] --- BUS
E4["ECU4 - Cluster"] --- BUS
end
PMA -.connects.-> BUS
The core of the physical layer is differential signaling. CAN expresses bits by the voltage difference between the two lines CAN_H and CAN_L. A state in which there is almost no voltage difference between the two lines is called recessive (logic 1), and a state in which CAN_H is raised and CAN_L lowered to create a large voltage difference is called dominant (logic 0). Because the differential method cancels, at the receiver, the common-mode noise that enters both lines equally, it shows high immunity even in a vehicle environment with severe noise from motors and ignition systems. Both ends of the bus must carry a 120Ω termination resistor to suppress signal reflection; if this termination is missing, communication fails intermittently, so it is the first item checked in practical troubleshooting.
The asymmetry of dominant/recessive is the foundation of all CAN behavior. When several nodes transmit at once, if even one sends a dominant (0), the bus becomes dominant. That is, dominant "overwrites" recessive. This wired-AND electrical characteristic is the physical basis for the non-destructive arbitration and the error signal (error frame) explained later. The maximum transmission speed of Classical CAN is 1 Mbit/s (within about 40m), and speed and distance are inversely proportional. Because the signal must propagate to both ends of the bus so that all nodes "see" the same bit simultaneously for arbitration to hold, the physical constraint that the bit time must be sufficiently longer than the worst-case propagation delay determines the speed/distance limits.
Looking at a single bit in detail, the CAN bit time is divided into a synchronization segment, a propagation delay segment, and phase buffer segments, and each node reads the bus value at the sample point within them. Because all nodes must read the same bit at the same instant for arbitration to hold, the longer the bus with larger propagation delay, the later the sample point must be pushed and the longer the bit time. For this reason, practitioners carefully tune bit timing parameters (SJW·TSEG) considering bus length, number of nodes, and speed together, and parameter mismatch is a common cause of intermittent communication errors.
For example, in an actual passenger car it is typical to run the powertrain domain (engine·transmission) on high-speed CAN at 500 kbit/s and the body domain such as doors, seats, and lighting on low-speed CAN at 125 kbit/s. This is an example of a design trade-off that allocates bandwidth to areas where real-time behavior matters and economy to cost-sensitive areas.
3. Non-destructive Bitwise Arbitration (CSMA/CR) and Frame Structure
CAN's medium access method is CSMA/CR (Carrier Sense Multiple Access with Collision Resolution). Unlike Ethernet's CSMA/CD, which discards the transmission upon detecting a collision and retries after a random time, CAN "resolves" the collision and transmits the higher-priority frame to the end without loss. The sequence below shows the arbitration process when three nodes start transmitting simultaneously.
sequenceDiagram
participant A as "Node A (ID 0x100)"
participant B as "Node B (ID 0x010)"
participant C as "Node C (ID 0x111)"
participant Bus as "Bus state"
Note over A,C: After SOF, identifier (ID) bits are sent simultaneously from the MSB
A->>Bus: 0 0 1 ...
B->>Bus: 0 0 0 ...
C->>Bus: 0 0 1 ...
Note over Bus: At the 3rd bit only B sends dominant (0) → Bus=0
Bus-->>A: Sent 1 but Bus=0 → lost arbitration, switch to receive
Bus-->>C: Sent 1 but Bus=0 → lost arbitration, switch to receive
Note over B: Bus matches sent value → won arbitration, keep transmitting
B->>Bus: Completes the rest of the frame
The arbitration principle is as follows. Each node wishing to transmit puts the identifier (ID) on the bus one bit at a time from the most significant bit, while simultaneously reading the bus state. If it sent a recessive (1) but the bus reads dominant (0), it means a node with a lower ID (= higher priority) is transmitting, so it immediately stops transmitting and switches to receive mode. Conversely, the node whose sent value keeps matching the bus value becomes the final winner and continues data transmission right after arbitration ends. The key is that the smaller the ID number, the higher the priority, and even when a collision occurs the winner's frame is not corrupted by a single bit. As a result, the latency of the highest-priority message is deterministically upper-bounded, so the real-time behavior of safety-critical messages such as an emergency braking signal can be analyzed mathematically. This is the decisive reason CAN is preferred in safety systems.
The field composition of the data frame is shown in the table below. Each field shares a distinct role of transmission, arbitration, and error detection.
| Field | Composition | Role |
|---|---|---|
| SOF (Start of Frame) | 1 bit (dominant) | Frame start, synchronization reference for all nodes |
| Arbitration field | 11-bit ID (2.0A) or 29-bit (2.0B) + RTR | Priority determination·arbitration |
| Control field | IDE, DLC (4 bits) | Extended ID flag, data length (0–8 bytes) indication |
| Data field | 0–8 bytes (Classical) | Actual payload |
| CRC field | 15-bit CRC + delimiter | Transmission error detection |
| ACK field | ACK slot + delimiter | A correctly receiving node responds dominant |
| EOF (End of Frame) | 7 bits (recessive) | Frame end |
As the table shows, CAN has no sender-address or receiver-address fields. The ACK slot is an interesting design: the sending node leaves this segment recessive, and any node that received the frame correctly overwrites it dominant to signal "someone received it properly." The fact that this is a collective acknowledgment of the whole bus rather than an individual receive confirmation is consistent with the message-oriented philosophy.
4. Error Detection and Error Confinement
CAN's reliability comes from combining five independent error-detection techniques with an error-confinement mechanism that isolates faulty nodes by itself. The five techniques are ① bit monitoring (the sending node reads back its own transmitted bit and checks for mismatch), ② bit stuffing (when 5 identical bits appear in a row, one opposite bit is inserted—the receiver recognizes a rule violation as an error), ③ CRC check (15-bit cyclic redundancy), ④ frame format check (detects rule violations in fixed-format fields), and ⑤ ACK check (detects absence of a response). Operating complementarily, CAN's undetected error probability is assessed to be extremely low.
A node that detects an error emits an error frame onto the bus. This deliberately violates the bit-stuffing rule with a series of dominant bits, so that all nodes simultaneously recognize that "this frame is invalid," discard it, and let the sender automatically retransmit. Here an important problem arises—if a particular node's transceiver fails and continuously pours out error frames, even normal communication can be paralyzed (the babbling idiot problem). Error confinement is what prevents this.
Each node maintains a Transmit Error Counter (TEC) and a Receive Error Counter (REC), which increase when an error occurs and decrease on successful transmit/receive. Depending on these counter values, a node moves among three states.
stateDiagram-v2
[*] --> ErrorActive
ErrorActive: Error Active (normal)
ErrorPassive: Error Passive (caution)
BusOff: Bus Off (isolated)
ErrorActive --> ErrorPassive: TEC or REC > 127
ErrorPassive --> ErrorActive: counter <= 127 recovers
ErrorPassive --> BusOff: TEC > 255
BusOff --> ErrorActive: re-enter after observing 128 sequences of 11 recessive bits
A normal node in the Error Active state intervenes actively with a dominant error flag upon finding an error. When the counter exceeds 127 it falls to Error Passive, signaling errors but using only a recessive flag so as to disturb the bus less. When TEC exceeds 255 it becomes Bus Off, completely detaching itself from the bus and no longer able to transmit. Thanks to this staged isolation, a single failed node cannot bring down the whole network. For instance, even if the communication chip of the ABS ECU degrades and produces intermittent errors, only that node exits to Bus Off, while engine and transmission communication continues and the vehicle can behave in a minimal safe state. This is a typical implementation of the "fault containment" principle required in safety-critical systems.
5. Classical CAN · CAN FD · CAN XL Comparison
CAN has evolved through three generations to meet increasing bandwidth demand. Beyond a simple comparison, the key is what limit each generation broke through and how.
| Category | Classical CAN | CAN FD | CAN XL |
|---|---|---|---|
| Standard | ISO 11898-1 | ISO 11898-1:2015 | ISO 11898-1:2024 / CiA 610-1 |
| Max payload | 8 bytes | 64 bytes | 1–2048 bytes |
| Data-phase speed | 1 Mbit/s | ~5–8 Mbit/s | up to 20 Mbit/s (10 in some sources) |
| Arbitration method | Bit arbitration | Bit arbitration (arbitration stays slow) | Bit arbitration + priority |
| Main use | Body·powertrain | High-resolution sensors·gateways | ADAS·zone architecture backbone |
The core idea of CAN FD (Flexible Data-rate) is to separate the speed of the arbitration phase and the data phase. Because arbitration requires all nodes to see the bit simultaneously, its speed is hard to raise under the propagation-delay constraint. So CAN FD raises the bit rate only in the data phase after arbitration ends and the winner is fixed, and also expands the payload from 8 to 64 bytes. As a result it greatly increased throughput while keeping the existing topology, and today most new vehicle models adopt CAN FD. Concretely, transmitting 64 bytes of data—which used to be split into eight 8-byte messages—in a single frame measurably reduces both overhead and latency.
CAN XL goes one step further, increasing the payload up to 2048 bytes and the speed to the order of 20 Mbit/s, aiming to be an intermediate backbone that fills the gap between low-speed CAN and high-speed Automotive Ethernet. However, because physical-layer requirements are strengthened—such as the need for SIC (Signal Improvement Capability) transceivers in data phases above 10 Mbit/s—cost and maturity must be weighed together when applying it. All three generations took backward compatibility and coexistence as design goals, which is a strength of the CAN ecosystem.
6. Higher-Layer Protocols and Industrial Application
CAN itself defines only "send 8 (or 64) bytes to anyone," so real applications need higher-layer protocols that specify the meaning of the bytes, the addressing scheme, and network management. Representatively, J1939 uses 29-bit extended IDs in heavy commercial vehicles (trucks, buses, construction machinery) to standardize thousands of vehicle parameters with the PGN (Parameter Group Number) and SPN (Suspect Parameter Number) scheme. CANopen provides an Object Dictionary and PDO/SDO communication model in industrial automation, medical devices, and robotics. OBD-II is a legally mandated diagnostic protocol for emissions diagnosis, the basis on which a repair-shop scanner reads fault codes (DTCs) through a standard connector.
From a diagnostic perspective, OBD-II gives direct value on the repair floor. When the engine warning light comes on, the whole sequence of reading a standard DTC with a scanner (e.g., P0300 misfire detection) to pinpoint the fault, and observing the real-time sensor stream (live data) to track the reproduction condition, all takes place on top of CAN higher-layer diagnostic protocols.
As a concrete example showing the breadth of industrial application, a single mass-produced passenger car typically has dozens of ECUs (70–100 or more in high-end models) split across several CAN segments and connected through gateways. Agricultural machinery such as tractors and combines achieves interoperability between implements and tractor with ISOBUS (ISO 11783, based on J1939), and building and factory automation uses DeviceNet (CAN-based) as a sensor/actuator network. In this way CAN has spread beyond automobiles to every industry that needs a "low-cost deterministic fieldbus," and this versatility is the background of its vitality sustained for nearly 40 years.
7. Advanced: Evolution of Vehicle Networks and Security
(1) Bandwidth limits and Automotive Ethernet. In the era of autonomous driving and ADAS, cameras, lidar, and radar pour out hundreds of Mbit to Gbit per second of data, so CAN at a few Mbit/s alone cannot cope. Accordingly, single-pair Automotive Ethernet such as 100BASE-T1/1000BASE-T1 is introduced as a high-bandwidth backbone, and TSN (Time-Sensitive Networking) is combined for deterministic real-time behavior. But CAN does not disappear—CAN(FD) is still optimal for actuator control that needs low speed, low cost, and high reliability, so a real vehicle is designed as a heterogeneous hybrid network in which CAN, CAN FD, and Ethernet are connected by gateways. Recently the trend of consolidating numerous distributed ECUs into a few high-performance zonal controllers—zonal architecture and SDV (Software-Defined Vehicle)—is pronounced, and CAN XL is drawing attention as a mid-layer backbone candidate in this structure.
(2) Security vulnerability and countermeasures. Because CAN assumed a "closed network" at design time, it has no authentication or encryption whatsoever. Messages carry no sender identification and anyone can spoof an arbitrary ID, so if an attacker penetrates the internal bus through the diagnostic port, telematics, or infotainment, they can even forge steering and braking commands. The 2015 remote hacking demonstration of the Jeep Cherokee is a representative case that impressed this risk on the public, and afterward automotive cybersecurity became subject to regulation (UNECE WP.29 R155, ISO/SAE 21434). As countermeasures, AUTOSAR's SecOC (Secure Onboard Communication) adds a MAC (message authentication code) and freshness value to messages to detect tampering, domain-separation gateways isolate external contact points from safety-critical buses, and a CAN IDS (intrusion detection) detects abnormal periods and IDs. The expanded payload of CAN FD/XL provides room to carry a MAC, which is also advantageous for security. However, given real-time constraints, an authentication-centric design is more realistic than full encryption, which is a key trade-off from the professional engineer's perspective.
8. Considerations and Implications
The strategic issues to consider when designing and adopting CAN from a professional engineer's perspective are as follows.
- Triangular trade-off of bandwidth, real-time behavior, and cost: Mix Classical CAN, CAN FD, and Ethernet by function, but guarantee determinism for safety-critical signals with the highest-priority IDs and worst-case latency analysis. Unifying everything into a high-speed network just by looking at bandwidth loses on cost and power.
- Security by design: Since it is hard to retrofit security onto legacy CAN, integrate SecOC, gateway separation, and an IDS into the architecture at the new-design stage, and satisfy the type-approval requirements of ISO/SAE 21434 and UNECE R155 over the full design/operation lifecycle (TARA·monitoring).
- Linkage with functional safety: CAN's error confinement is strong but does not by itself satisfy the ASIL requirements of ISO 26262. E2E (End-to-End) protection (sequence number·CRC·timeout) must be added at the application layer so that communication faults are handled at the level of the safety goal. That is, network reliability and functional safety must be reinforced at separate layers.
- Evolution roadmap and backward compatibility: The transition from CAN FD→CAN XL→Ethernet/TSN is realistically a gradual shift through coexistence, not a one-time replacement. A transition strategy that evaluates the gateway's protocol-conversion load, transceiver (SIC) maturity, and supply-chain cost in stages is needed.
- Expansion of application domains: As CAN has spread beyond automobiles to agricultural machinery (ISOBUS), industrial automation (CANopen/DeviceNet), medical devices, and railways, compliance with domain-specific higher-layer protocol standards and interoperability verification determine the success or failure of adoption.
References
- CAN in Automation (CiA), "CAN XL (extended data-field length)": https://www.can-cia.org/can-knowledge/can-xl
- CAN in Automation (CiA), "CiA 610-1 and CiA 610-3": https://can-cia.org/news/cia-in-action/view/cia-610-1-and-cia-610-3/
- Bosch Semiconductors, "CAN XL": https://bosch-semiconductors.com/ip-modules/can-protocols/can-xl/
- ISO 11898-1:2015 (Road vehicles — CAN — Part 1: Data link layer and physical signalling): https://www.iso.org/standard/63648.html
In one line: CAN is the standard vehicle/industrial fieldbus that realizes low-cost, high-reliability real-time communication over a two-wire bus through message-priority-based non-destructive bitwise arbitration and multiple error confinement, and it is evolving into a heterogeneous vehicle network that expands bandwidth with CAN FD/CAN XL and combines with Automotive Ethernet and security (SecOC).