How Do You Design an Isolated CAN Bus Transceiver Interface?

Detailed Explanation

Most CAN nodes on a bench prototype share a common ground with every other node, and the transceiver's local ground reference matches the MCU's ground reference directly. Once that same network moves into a deployed product (a long cable run between enclosures, a mix of AC-powered and battery-powered nodes, or a bus segment that reaches into an electrically noisy or high-voltage area of a machine), the assumption of a shared, quiet ground stops holding, and isolating the CAN interface becomes a deliberate design decision rather than an optional extra.

Why CAN Interfaces Get Isolated

Three related problems drive the decision to isolate a CAN transceiver:

• Ground loops. When two nodes on the same bus are powered from separate supplies with different ground potentials, or share a chassis ground path with other equipment carrying significant return current, the resulting ground potential difference appears as a common-mode offset across the bus. On a non-isolated interface this offset adds directly to the differential signal's operating margin; enough of it degrades signal quality or introduces bit errors that look like a wiring or termination fault rather than a ground-reference problem.
• Fault propagation into the host MCU. A CAN bus that reaches into a machine's electrically noisy or higher-voltage sections is exposed to failure modes a bench prototype never sees: a short between CAN_H or CAN_L and a supply rail, an accidental connection to mains-adjacent wiring, or a transceiver damaged by a field event. Without isolation, a large enough fault current has a direct conductive path back to the MCU's ground and supply rails, risking damage well beyond the transceiver itself. Isolation contains that fault to the bus-side circuitry.
• EMC and surge robustness. Automotive and industrial CAN networks are routinely exposed to conducted transients on their supply and signal lines, including load-dump events, inductive switching transients, and coupled surges of the kind characterised by ISO 7637-2 for road vehicles. An isolation barrier, sized with adequate margin, keeps a transient event on the bus side from reaching the host MCU's supply and ground domains, which is a meaningfully different (and stronger) protection than transient suppression diodes alone provide.

Not every CAN node needs isolation. A short, single-enclosure bus with all nodes sharing one clean ground and one supply rarely benefits enough to justify the added bias-supply complexity. Isolation earns its place when the bus physically spans separate ground domains, when a credible fault path exists between the bus and something hazardous, or when a product's target market or certification body expects it.

Isolator Placement Between the CAN Controller and the Bus

The isolation barrier sits between the microcontroller's CAN controller and the CAN transceiver, not between the transceiver and the bus. The MCU-side pins (TXD, RXD, and often a standby or fault-flag pin) cross the barrier as isolated digital signals; the transceiver itself, along with CAN_H and CAN_L, lives entirely on the bus side. This placement means the transceiver's differential driver and receiver, the parts of the circuit actually exposed to bus-side faults and transients, sit on the far side of the barrier from the MCU, which is the whole point: a fault severe enough to damage the transceiver stays confined to the bus-side ground domain.

This creates two ground domains that must be kept genuinely separate on the PCB: an MCU-side ground referenced to the rest of the board, and a bus-side ground that must have no conductive path back across the barrier through a shared shield, mounting hardware, or a copper pour that bridges the isolation gap. The physical spacing across that gap follows the same creepage and clearance principles as any other isolation barrier, sized to the working voltage and pollution degree the design targets under IEC 60664-1.

Powering the Isolated Side: The Bias Supply Problem

Isolating the data lines only solves half the problem. The transceiver on the bus side still needs a supply rail, and that supply cannot be the same rail powering the MCU without a conductive connection that defeats the isolation. This is the detail most often underestimated when a design first adds isolation to an interface that previously just worked off the board's single logic supply.

The standard solution is an isolated DC-DC bias supply: either a small transformer-based isolated regulator module, or a discrete transformer-driver IC and transformer generating the isolated rail from the MCU-side supply. How this is handled depends on the transceiver chosen:

• Transceivers requiring an external isolated bias supply. The ISO1042, for example, isolates TXD/RXD but expects its own separate 5 V supply (VCC2) on the bus side; the datasheet's reference design uses either a small isolated DC-DC module or a transformer-driver IC such as the TI SN6505B feeding a small transformer to generate that isolated rail, typically with a target current budget in the low tens of milliamps range once transceiver quiescent and transmit current, decoupling, and any other bus-side load are accounted for. Always check the specific part's datasheet for its actual supply current figures rather than assuming a generic value.
• Transceivers with an integrated isolated DC-DC converter. The ADM3053 integrates its own isolated bias supply (Analog Devices' isoPower technology) inside the same package as the isolator and transceiver, so the whole part runs from a single 5 V supply on the MCU side with no external isolated DC-DC stage needed. This simplifies the board at the cost of some flexibility: the ADM3053's data rate tops out at 1 Mbit/s, sufficient for classic CAN but not for CAN FD's faster data phase.

Either way, size the bias supply for the transceiver's worst-case current, not its typical or quiescent figure, and add margin for anything else drawing from the same isolated rail. An under-sized bias supply is a common, easy-to-misdiagnose source of intermittent bus errors that present as a signal-integrity or firmware problem rather than the power-budget shortfall they actually are.

Common-Mode Choke Placement Alongside the Isolator

A common-mode choke on CAN_H/CAN_L, placed close to the bus connector on the isolated side, is common practice on industrial and automotive CAN interfaces regardless of whether the interface is isolated. It attenuates common-mode noise current coupling onto the bus pair from the cable itself, improving both radiated emissions performance and the bus's tolerance to externally injected noise.

It is important to keep the choke's job and the isolator's job conceptually separate, because they are frequently, and incorrectly, treated as redundant. The choke reduces common-mode noise current on a bus pair that still shares a continuous conductive connection between its two ends; it does nothing to stop a genuine fault, a large ground shift, or a surge event from reaching whatever is on the other end of that connection. The isolator removes the conductive connection entirely, which a choke cannot do regardless of its impedance rating. A design with only a choke and no isolation still has a direct electrical path for a fault to follow; a design with isolation but no choke is protected against faults but gets no benefit against everyday common-mode noise coupling onto the cable. Where both problems exist, which is the normal case for a bus reaching outside a single enclosure, both components belong in the design.

CMTI Selection for CAN's Differential Signalling

Common-mode transient immunity (CMTI), the maximum rate of change of common-mode voltage the isolator can tolerate without corrupting its output, is the isolator specification that most directly governs whether the isolated interface stays reliable under real-world transient conditions. For CAN specifically, it helps to separate two different sources of common-mode dV/dt that the isolator might see:

• CAN's own signal transitions. The bus's own recessive-to-dominant and dominant-to-recessive edges, running at up to 1 Mbit/s for classic CAN or into the low megabits per second for CAN FD's data phase (ISO 11898-1), are comparatively gentle events by isolator standards. They are not the demanding case the way a fast-switching power converter's gate drive signal is for a gate-drive isolator; CAN's physical layer was never designed to switch at the edge rates that stress CMTI.
• External fault and transient events. The demanding case is a large, fast common-mode step imposed on the bus side by something external to normal bus operation: an ESD discharge into the connector, an automotive load-dump or inductive switching transient of the kind ISO 7637-2 characterises, or a sudden ground potential shift from a fault elsewhere in the system. These events can present a genuinely fast edge across the isolation barrier, and it is this worst case, not the routine bus traffic, that the isolator's CMTI rating needs margin against.

As a concrete reference point, the ISO1042 specifies a minimum CMTI of 85 kV/μs; this is representative of the CMTI class typically found in isolated CAN and RS-485 transceiver parts, and it comfortably exceeds what CAN's own signalling demands, leaving the margin available for transient events. Treat the datasheet CMTI figure as the number to check against the transient environment the product will actually see (automotive load dump, ESD exposure at an exposed connector, or a specific surge immunity level required by a customer or standard) rather than against the bus's own bit rate, since the bit rate is rarely the limiting factor.

Isolated CAN Transceiver IC Options: Integrated vs Discrete

Two architectures are used for isolated CAN interfaces, and both are legitimate, verifiable design choices rather than one being a shortcut around the other:

• Fully integrated isolated CAN transceiver ICs. These combine the digital isolator and the CAN transceiver in a single package. The Texas Instruments ISO1042 (up to 5 Mbit/s, supporting CAN FD, with a withstand isolation rating of 5000 Vrms and requiring a separate isolated bias supply for its bus-side VCC2 pin) and the Analog Devices ADM3053 (up to 1 Mbit/s, classic CAN only, with an integrated isolated DC-DC converter needing only a single-side supply) represent the two common variants of this category: isolated signal only, or isolated signal plus integrated isolated power. Both simplify the schematic and reduce the number of interface timing parameters that need independent verification, at the cost of being locked into that specific part's data rate and isolation rating.
• A general-purpose digital isolator paired with a standard CAN transceiver. A multi-channel digital isolator with sufficient CMTI and speed, wired to carry TXD and RXD (and, where used, a standby or fault pin) across the barrier to a conventional external CAN transceiver such as the TJA1050 or SN65HVD230, is the older and still entirely valid approach. This gives more flexibility in transceiver selection (bus fault voltage rating, standby current, CAN FD support) independent of the isolator vendor's specific transceiver offering, at the cost of an extra component and an extra interface to verify.

Either architecture still needs the isolated bias supply covered above for the bus-side transceiver, and both benefit from the same common-mode choke and creepage/clearance treatment. The choice between them typically comes down to board space, whether an off-the-shelf part already meets the required data rate and isolation voltage, and how much flexibility the design needs in transceiver selection. Zeus Design designs isolated CAN and other isolated interface boards, from transceiver and isolator selection through to bias supply sizing and isolation-barrier layout.

Design Considerations

• Confirm the transceiver's data rate against the actual CAN or CAN FD requirement before committing to a part. An isolated transceiver limited to 1 Mbit/s, such as the ADM3053, is not usable for CAN FD's faster data phase; check this before the schematic is finalised, not after bring-up.
• Treat the isolated bias supply's current budget as a hard design input, not an afterthought. Read the transceiver datasheet's bus-side supply current figures (quiescent and active) and size the isolated DC-DC stage with margin for the worst-case combined load, including anything else sharing the isolated rail.
• Size the isolation voltage rating to the application's actual fault and safety requirements, not a default part choice. Automotive and industrial designs have different working-voltage and transient exposure expectations; check the isolator or isolated transceiver's withstand and working isolation voltage against IEC 60664-1's requirements for the design's pollution degree and working voltage, and against any customer or automotive-standard requirement that applies.
• Keep the choke and the isolator as two separate line items in the design, each justified on its own. Do not treat a common-mode choke as a substitute for isolation, or assume isolation removes the value of a choke; see the discussion above for why they solve different problems.

Common Mistakes

• Selecting an isolated transceiver by isolation voltage alone and skipping the CMTI check. A part with plenty of isolation voltage margin but a CMTI rating not checked against the design's actual transient exposure (automotive load dump, exposed-connector ESD) can still corrupt data during exactly the events isolation was meant to protect against.
• Under-sizing the isolated bias supply, producing intermittent bus errors or nodes dropping under load that look like a signal-integrity or termination problem rather than the power-budget shortfall they actually are.
• Allowing a stray conductive path to bridge the isolation barrier through a shared shield, a mounting screw crossing both ground domains, or a copper pour that spans the barrier gap, silently defeating the isolation with no obvious symptom until a dielectric-withstand test or a field fault event reveals it.
• Assuming a common-mode choke provides the same protection as galvanic isolation, or the reverse: omitting a choke because isolation is already present. Both components address genuinely different problems on a bus that reaches outside a single clean ground domain, as covered above.

For general isolation technology selection and CMTI fundamentals, see optocoupler vs digital isolator. For the CAN protocol fundamentals this isolation sits on top of, see What Is CAN Bus?.