RS-485 vs CAN: Choosing a Field Bus

Quick answer

RS-485 and CAN both run over a terminated differential pair, but they solve different problems. RS-485 is a physical layer: it moves bytes reliably over long cable, and you supply the protocol, addressing and error handling in software. CAN is a complete data-link layer: it adds framing, arbitration, error detection and automatic retransmission in hardware. Choose RS-485 when you need distance and simplicity, CAN when you need multi-master determinism and built-in fault handling. At 250 kbps, RS-485 comfortably covers 300 m while CAN is already near its 250 m limit — and CAN's real advantage is not distance, it is the protocol.

Executive takeaways

• Distance is a cable and termination problem, not a protocol problem. At 250 kbps, CAN's practical limit is about 250 m; RS-485 can go further with the same pair and a slower bit rate.
• CAN's value is arbitration. Multiple nodes can start transmitting at once and the higher-priority frame wins without collision; RS-485 needs a master or a software protocol to avoid collisions.
• RS-485's common-mode range is wider (−7 V to +12 V) than a typical CAN transceiver's, which matters in industrial installations with ground offsets.
• Both need termination at both ends, short stubs and a bias network; most field failures are wiring and termination, not the transceiver.

1. What each bus actually is

RS-485 (TIA/EIA-485-A) defines the electrical layer only: differential drivers and receivers, a common-mode range, unit-load limits and termination rules. It says nothing about addressing, framing, collision handling or error recovery. A UART plus an RS-485 transceiver gives you a robust byte pipe; the protocol on top is your design.

CAN (ISO 11898-2) defines the physical layer and the data-link layer. Frames carry an identifier, a data field, CRC and acknowledgment; arbitration is bit-wise and non-destructive; errors are detected and frames are retransmitted automatically. That is a lot of behaviour built into the controller and transceiver, and it is why CAN dominates automotive and mobile machinery.

A related standard, RS-422, uses the same differential signalling with a dedicated driver and receiver pair (four wires) and supports higher speed over shorter runs. RS-485 is the multi-drop, two-wire version; when this article says RS-485, it means the two-wire multi-point case unless stated otherwise.

2. Speed versus distance

The table is a planning guide, not a guarantee: cable capacitance, stub length, termination tolerance and node count all move the limit. The important comparison is the shape: both fall roughly with the same physics, but CAN gives up distance earlier because its bit timing and arbitration require tighter signal integrity.

3. Topology, termination and bias

Both buses are multi-drop differential pairs terminated at both physical ends. RS-485 uses a single 120 Ω resistor at each end; CAN often uses a split termination — two 60 Ω resistors in series with the midpoint bypassed to ground — which improves common-mode stability and EMC.

Idle-state bias is the other half. A terminated pair with no active driver floats near 0 V, which is an undefined logic level. RS-485 networks use a fail-safe bias network — typically a pull-up on one line and a pull-down on the other — to hold a known idle state. With two 120 Ω terminations in parallel (60 Ω), a 200 mV idle differential needs about 3.3 mA, which from a 5 V rail means roughly 1.4 kΩ total, or 680 Ω on each side.

CAN transceivers usually provide the bias internally and specify the recessive level; that is one reason CAN wiring is simpler to get right.

4. Protocol: the real difference

RS-485 is half-duplex on a two-wire pair (four-wire full duplex exists). Only one driver may be enabled at a time, so the network needs a master, a token or a collision-avoidance scheme implemented in software. That is fine for Modbus, DMX, Profibus-style protocols and many proprietary links, but the determinism is only as good as the software.

CAN is multi-master by design. Any node can start a frame when the bus is idle; if two start together, the one with the dominant identifier wins arbitration and the other retries automatically. Higher-priority messages therefore have bounded latency, which is why CAN is used for control loops in vehicles and machines.

5. Worked example: 300 m, 250 kbps, 16 nodes

An industrial network needs 16 nodes over 300 m at 250 kbps, with a 5 V transceiver supply. Which bus?

• CAN at 250 kbps is specified to about 250 m, so 300 m is already beyond the planning limit; it would need a repeater or a lower bit rate.
• RS-485 at 250 kbps is comfortable at 300 m. With 50 pF/m cable, the pair presents about 15 nF; with 60 Ω of effective termination the RC time constant is 0.9 µs, and the 10–90% rise time is about 2 µs against a 4 µs bit — acceptable, with margin for stubs and connectors.
• Termination: 120 Ω at each end. Bias: 680 Ω pull-up and pull-down give roughly 200 mV idle differential.

The practical answer is RS-485 unless the application needs CAN's arbitration or its diagnostic stack. Distance alone does not decide it.

Run the same numbers for CAN and the picture is clear: at 250 kbps the CAN limit is 250 m, so 300 m would require dropping to 125 kbps (500 m) or adding a repeater. That is a system change, not a component choice.

6. Isolation, protection and cable

Long field buses pick up surge and ground offsets. At the connector, a TVS or ESD array protects the transceiver from ESD and surge; a common-mode choke reduces high-frequency emissions and improves immunity. Neither replaces the need for a defined ground reference or isolation where the offset is large.

Galvanic isolation is the robust answer for long runs between buildings or between equipment on different grounds. Isolated transceivers combine the driver, receiver and an isolation barrier in one package; discrete digital isolators plus a non-isolated transceiver achieve the same result with more parts. The choice is usually cost and board space, not performance.

Cable matters as much as the silicon. A twisted pair with 100–120 Ω characteristic impedance, 50–60 pF/m and a foil or braid shield is the baseline. Shield grounding at one end avoids ground loops in most installations; the exception is an installation with a dedicated reference conductor, where both ends may be bonded.

7. Protocol examples

On RS-485, Modbus RTU is the most common protocol: a master polls slaves by address, with a CRC per frame. DMX512 uses the same physical layer for lighting, with a fixed frame rate. Both work because the master controls the bus; neither provides hardware collision handling.

On CAN, CANopen and J1939 add addressing and profiles on top of the frame layer. Because arbitration is in hardware, higher-priority messages get the bus first, and error frames force a fast retransmission. That is a different reliability model from RS-485, where a colliding frame is simply lost unless the software detects it.

It is worth saying what these buses are not. Neither is a replacement for industrial Ethernet when the payload is large or when remote diagnostics and time synchronisation matter. Ethernet adds magnetics, PHYs and a switch, and it solves different problems; the field bus remains the right choice for small, deterministic control traffic over a single pair.

8. Choosing

Choose RS-485 when: the run is long, the protocol is already defined (Modbus, proprietary), nodes are few to moderate, and a master-slave structure is acceptable. It is also the simpler choice when the same pair must coexist with other signals or when common-mode offsets are expected.

Choose CAN when: multiple nodes must arbitrate without a master, messages have priorities, and the application benefits from hardware CRC, acknowledgment and retransmission. Automotive, battery management, motor drives and mobile machinery are the classic cases.

Choose CAN FD when: you need the CAN model but more bandwidth — up to 5 Mbps in the data phase with the same arbitration phase.

9. Who should not use which

• Do not use RS-485 for safety-critical multi-master control unless you have proven the software protocol; there is no hardware arbitration or error containment.
• Do not use CAN for very long, low-speed sensor runs where a simple Modbus link would do; the transceiver and controller add cost without benefit.
• Do not use either for a single point-to-point link under a few metres; a plain UART or SPI is simpler and cheaper.
• Do not use either without isolation where ground potential differences exceed the common-mode range; add digital isolators or isolated transceivers.

Rise time and cable capacitance

The pair behaves as an RC line: t_r ≈ 2.2 × R_term × C_cable, with R_term the parallel value of the two terminations (60 Ω) and C_cable the total capacitance. At 50 pF/m and 300 m, C = 15 nF, so t_r ≈ 2 µs — half a bit at 250 kbps. That is why the distance falls quickly as the bit rate rises, and why adding nodes or stubs makes it worse.

Check the result against the bit time, not against the data sheet's maximum rate. A transceiver rated at 10 Mbps will still fail a 300 m link at 10 Mbps because the cable, not the transceiver, sets the limit. This is also why a bit-error-rate test at the real cable length is worth more than any table.

10. Layout and EMC

Both buses are differential, so route the pair together, keep the impedance consistent and avoid splits in the reference plane. Place the termination at the connector, not somewhere in the middle of a stub. Keep stubs short: as a rule, a stub should be a small fraction of the bit length, well under 0.3 m at 1 Mbps.

Add common-mode chokes and TVS protection at the connector for surge and ESD, and a ground reference conductor if the installation cannot guarantee a common ground. The transceiver's common-mode range is not a license to ignore grounding; it is a margin, not a design method.

11. Five mistakes

1. One termination resistor. A single 120 Ω at one end leaves a reflection at the other; both ends need it.

2. No fail-safe bias. The idle level floats, and receivers pick up noise as data.

3. Long stubs to each node. Stubs add reflections and limit the usable bit rate far more than the cable length does.

4. Mixing RS-485 and CAN transceivers on one pair. The signalling and termination expectations differ; the network will be unreliable.

5. Assuming CAN's distance is the same as RS-485's. At the same bit rate, CAN reaches less than half the distance in many cases.

One more practical difference: RS-485 transceivers are available with slew-rate limiting, which reduces EMI at the cost of maximum bit rate. CAN transceivers are specified for a fixed bit-rate range and their symmetry is part of the standard. Do not assume the two are interchangeable even when the pinout looks similar.

12. Checklist

• Required bit rate and maximum cable length.
• Node count and unit-load budget.
• Multi-master or master-slave architecture.
• Arbitration and latency requirements.
• Common-mode voltage between nodes.
• Termination at both ends, with the correct value and placement.
• Fail-safe bias (RS-485) or recessive level (CAN).
• Isolation requirement.
• Surge/ESD protection at the connector.
• Verification: eye pattern or bit-error test at the real cable length.

Ten checks before layout, and one after: run a bit-error test or an eye-diagram capture at the real cable length. A network that looks correct on a two-metre bench cable can still fail at the installed length, and the failure mode — occasional bit errors, not a dead bus — is easy to misdiagnose as a software problem.

FAQ

Can RS-485 and CAN share the same cable?

No. They use different signalling, termination and frame formats. Running both on one pair will produce errors on both networks.

Why does CAN reach less distance than RS-485 at the same bit rate?

CAN's arbitration and bit timing need tighter signal integrity, and its transceivers have a narrower common-mode range. RS-485 is more tolerant because the protocol layer is outside the transceiver.

How many nodes can RS-485 support?

32 unit loads with standard transceivers, or up to 256 with 1/8-unit-load parts. The cable, termination and bias network must be sized for the count.

Do I need a separate ground wire?

If the nodes can have a significant ground offset, yes — a reference conductor or isolation is safer than relying on the transceiver's common-mode range.

Can I bridge RS-485 and CAN with a gateway?

Yes, and it is common when a legacy RS-485 device must join a CAN network. The gateway runs both protocols and maps addresses to identifiers; the two physical layers stay separate.

Do termination resistors dissipate much power?

Only when the driver is active: with 60 Ω effective and a 5 V differential drive, the termination current is tens of milliamps. Use 1/4 W or larger resistors and place them where the heat is acceptable.

Can I use a CAN transceiver for RS-485?

No. The signalling levels, termination expectations and frame formats differ. Use an RS-485 transceiver for RS-485 and a CAN transceiver for CAN.

How does node count affect distance?

Every node adds capacitance and leakage. A 32-node network at 250 kbps behaves differently from a 4-node network at the same rate; derate the distance for high node counts and long stubs.

Is CAN FD worth it for new designs?

If the application needs more than 1 Mbps of payload and the controllers support it, yes. The arbitration behaviour is unchanged, so the system model carries over.

Glossary

Unit load

A normalized input impedance for RS-485 receivers; 32 unit loads is the classic limit.

Fail-safe bias

Resistors that hold a defined idle level on an undriven RS-485 pair.

Arbitration

CAN's method of resolving simultaneous transmissions by identifier priority.

Split termination

Two 60 Ω resistors with a midpoint capacitor, used on CAN for common-mode stability.

Stub

The branch from the main bus to a node; long stubs cause reflections.

Sources

• Texas Instruments, The RS-485 Design Guide (SLLA272) — termination, bias and cable length.
• Texas Instruments, RS-422 and RS-485 Standards Overview (SLLA070) — standards and common-mode range.
• Texas Instruments, Overview of 3.3 V CAN Transceivers (SLLA337) — CAN physical layer and levels.
• ISO 11898-2 for CAN high-speed physical layer, and TIA/EIA-485-A for RS-485.