Skip to the content
2 layers100 × 50 mmCan iso, the printed circuit board in 3D

This project is an isolated CAN/J1939 transceiver breakout on a 2-layer board. It includes a DE-9 or terminal block and termination jumper. It demonstrates a complete PCB design workflow: custom symbol/footprint creation, layout, DRC, ordering, hand-soldering, and testing.

Size
100 × 50 mm
Copper layers
2
Thickness
1.6 mm
Parts
41
Nets
28
Pads
105
Tracks, vias
126, 48
Pours
6
Routed
Yes
KiCad
KiCad 9.0 (file 20241229)
Created
8 Sep 2026
Updated
13 Sep 2026

Overview

Can iso is an open-source printed circuit board by karimf1, from the karimf1/J939-transceiver repository on GitHub, published under the MIT licence. It is a 2-layer board of 100 × 50 mm with 41 parts on 28 nets, routed with 126 tracks and 48 vias.

It is built around the ISO1042BDWVR, NXE1S0505MC-R7 and BAT54, with the SMAJ36CA and DMG2301L among its other parts. Its parts include 9 capacitors, 7 resistors, 7 diodes, 3 connectors and 2 ICs. It belongs with the radio and wireless boards of this library.

Main components

5

README

README.md

can-iso-breakout — an isolated CAN / J1939 transceiver, and the barrier measured in millimetres

A galvanically isolated CAN transceiver on a 100 × 50 mm two-layer board: DE-9 wired to CiA 303-1, a 5-way terminal block carrying the same nets in parallel, split 120 Ω termination on a jumper, and bus-side power generated across the barrier.

The circuit is deliberately trivial — one transceiver, one isolated supply, a connector and two resistors. That is the point. A trivial circuit is the only kind that lets you walk the whole hardware pipeline without the circuit itself becoming the project: a symbol and a footprint drawn by hand from a datasheet, a schematic, a netlist, a board, a fabrication package.

What makes it worth reading is what the datasheets said on the way through, and one number that is measured rather than asserted.

Status: designed, not built. No board has been ordered, so there are no bench results here — every number below is derived from a datasheet or read out of the board file.

board


Key features

  • Galvanic isolation between the host logic and the CAN bus, with the barrier's real clearance stated in millimetres rather than claimed from the isolator's headline rating.
  • Two power routes for the bus side: an isolated DC-DC module on-board, or DE-9 pin 9 through a jumper. The second route raises the assembly's isolation from 3 kV to the transceiver's own 5 kV.
  • ±30 V common-mode range and ±70 V bus fault tolerance, with bus protection sized to that range rather than to the usual CAN default.
  • Split 120 Ω termination on a jumper (2 × 60R4 + 4n7), so the board can sit at the end of a bus or in the middle of one.
  • Dual connectors in parallel — DE-9 to CiA 303-1 and a 5-way terminal block — so the same board takes a lab harness or a field pigtail.
  • Level-flexible logic side, 1.71 – 5.5 V, jumper-selected between the host's 5 V and 3V3.
  • Two parts drawn by hand from their datasheets, including a footprint that carries its own copper keepout across the barrier.

The barrier, in millimetres

An isolation barrier is a claim, and claims should have numbers. Sorting every piece of copper onto one side of the board by net and measuring the gap gives:

mmwhat it means
layout clearance, side 1 ↔ side 210.300the closest approach the layout is responsible for
U1 land pattern (TI DWV0008A)9.100the transceiver's own recommended footprint
U2 land pattern (Murata NXE1)4.080the isolated supply module's
shortest path via floating copper6.621two hops, via U2's mechanical pad 14
assembly clearance4.080what the finished board actually is

The isolation rating of an assembly is the minimum over its components, and here that minimum is not the isolator. U1 is 5000 VRMS for one minute with 8.5 mm of creepage; U2 is 3000 VDC with 4.08 mm between its pad rows. Fitting both and calling the board a 5 kV part would be wrong by nearly a factor of two. So it is a 3 kV assembly, and the silkscreen says U1 5kVrms / U2 3kVdc with both ratings side by side.

There is a configuration on the board that gets the transceiver's full rating: leave U2 unpopulated and take bus-side power from DE-9 pin 9 through JP2. Then nothing crosses the barrier but U1, and the clearance becomes 9.1 mm. That is what the JP2 jumper is for.

Full working in docs/isolation.md.

Two things the datasheets caught

The board would have destroyed itself on first power-up

The Murata NXE1 is an unregulated, open-loop converter. Buried in its notes:

The minimum load to meet datasheet specification is 10% of the full rated load … Lower than 10% minimum loading will result in an increase in output voltage, which may rise to typically double the specified output voltage if the output load falls to less than 5%.

The ISO1042 draws 2.8 mA on the bus side when the bus is recessive. That is 1.4 % of the module's 200 mA rating — below the 5 % cliff. An unloaded barrier supply would push roughly 10 V into a part whose absolute maximum VCC2 is 6 V. Every board, every time, before a single frame was sent.

The fix is R7, a 240 Ω preload, holding the module at 12.8 % of rating no matter what the bus is doing. It costs 104 mW and it is the difference between a board that works and a board that kills a $5 transceiver on the bench. D7, a 5.6 V zener, is the backstop for the day someone builds one and leaves R7 off. The silkscreen says R7 PRELOAD - DO NOT OMIT.

The usual CAN TVS is the wrong part here

The standard bus-protection diode for CAN is a 24 V stand-off part, chosen for transceivers with a ±12 V common-mode range. The ISO1042's common-mode range is ±30 V. A 24 V TVS would conduct inside the transceiver's own legal operating range — it would clamp the bus during exactly the ground offset this board exists to survive.

So D2 and D6 are 36 V stand-off, bidirectional, clamping below the ±70 V bus fault rating: a window bounded below by the transceiver's common-mode range and above by its fault rating.

What the board is built to prove

A CAN bus is a differential bus with a limited common-mode range. Two nodes on different chassis, different supplies, or 30 m of cable apart do not share a ground — they share a ground offset. A non-isolated transceiver survives that offset up to its common-mode limit and then the link stops working, or the part stops existing.

The board is the isolated node. A $2 non-isolated module is the control group. Sweeping a DC offset between the two nodes' grounds and plotting frame error rate for each should give a flat line at zero for the isolated board and a cliff for the other. That turns received wisdom into a graph, which is the difference between a project and a tutorial.

The firmware is written for it: fw/j1939_rx counts frames against a sequence number the transmitter embeds, so the error rate comes from counting rather than from an assumption about timing.

The two hand-drawn parts

Everything except U1 and U2 comes from KiCad's standard libraries. Those two are drawn from their datasheets into hw/lib/, because "draw a part from a datasheet" is the skill this board exists to demonstrate.

TI_DWV0008A — the ISO1042's 8-pin wide-body SOIC. Every number is off drawing 4218796/A in the datasheet: 10.9 mm land row spacing, 1.8 × 0.6 mm pads, 1.27 mm pitch, 0.07 mm solder-mask expansion, 11.5 mm lead span. The resulting gap between the rows is 9.1 mm, which matches the datasheet's own 9.1 mm NOMINAL CLEARANCE/CREEPAGE note exactly.

The footprint also carries a keepout zone across its own barrier, so the rule that no copper may sit between the pad rows travels with the part instead of living only in someone's memory.

NXE1S0505MC — the isolated supply, including pad 14, which Murata documents as "not available for electrical connection" and which is therefore deliberately left off the netlist rather than tied to ground.

Why the board is generated rather than drawn

The schematic and the PCB are emitted by Python rather than placed in the GUI. The reason is not speed — 25 parts do not need a generator — it is that a generator can carry a rule the human eye does not enforce: no net and no part may touch both sides of the barrier. That invariant is expressed once, in code, and it holds before the board exists. Everything else the generator buys (reproducible geometry, a canonical digest of every part position and track) is a consequence of that choice rather than the reason for it.

Specification

parametervaluewhy
ProtocolCAN 2.0B, 29-bit IDs, J1939-compatible PHYJ1939 is CAN 2.0B on a defined physical layer
Bit rate250 kbit/s (J1939-11/-15), 500 kbit/s (J1939-14)loop delay allows 157 m and 57 m respectively
Logic side1.71 – 5.5 V, JP1 selects 5 V or 3V3 from the hostRXD swings to whatever the host runs at
Bus side5 V, from U2 or from DE-9 pin 9 (JP2)the bus side must not need a second power brick
Isolation3 kVDC as built; 5 kVRMS with U2 unpopulatedset by the weakest component, stated on the silkscreen
Bus fault tolerance±70 V24 V truck electrics; a shorted bus line is routine
Common-mode range±30 Vwhich is why the TVS is a 36 V part
Termination120.8 Ω split (2 × 60R4 + 4n7), jumper-selectableTI's own characterisation network
Host draw164 mA worst case, ~40 mA idlefits a USB-powered dev board
Board100 × 50 mm, 2 layer, 1.6 mmJLCPCB standard process, no special features

What changed from the plan

PLAN.md was written first and mostly held. Four deviations, all decided for a reason:

  • The board is 100 × 50 mm, not 55 × 35. A right-angle DE-9's courtyard is 22 × 32 mm. The plan had not looked one up.
  • The schematic is generated, not drawn in the GUI. §13 argued that 25 parts do not need a generator. True — but a generator can carry the barrier invariant, and that is the one mistake on this board a human reviewer would not see. Automation earned its place somewhere the plan did not expect.
  • The common-mode choke is gone, not fitted-as-0 Ω-links. It was first on the plan's own cut list, and dropping it keeps the bus stubs short and the pair symmetric — which matters more than the experiment it would have enabled.
  • R7 and D7 did not exist in the plan. They come from §17 row 9, the fill-in-the-blank datasheet table: "Isolated DC-DC: load regulation at your current." That row is why the board works.

The plan's §3.2 called the unregulated module "the single most likely 'it works but marginally' failure in this design." It was not marginal, and it was found by reading the notes section.

What this project does not claim

  • Not a certified isolation product. No hipot test, no creepage sign-off against a stated working voltage. The goal was narrower and is met: the board is not the weakest link in the barrier its parts provide.
  • Not a J1939 stack. Three real parameter groups on a J1939-compatible physical layer. No transport protocol, no address claiming.
  • The DE-9 is not a J1939 connector. A Deutsch pigtail on the terminal block is a cable, not a board feature.
  • Nothing here has been measured. No board exists yet.

Possible improvements

  • Build one and take the ground-offset sweep. Everything above is a prediction. The frame-error-rate-versus-offset curve for the isolated board against the non-isolated control is the single result that would turn this from a design exercise into a measurement.
  • Replace the unregulated DC-DC with a regulated one. R7 is a good fix for the part chosen, but it burns 104 mW permanently to work around a module that cannot hold its own output. A regulated isolated converter deletes both R7 and D7 — at higher cost, and with a new set of datasheet notes to read.
  • Pick a supply module with wider pad spacing. U2's 4.08 mm is what caps the whole assembly at 3 kV. A module with pad rows matching U1's 9.1 mm would let the board claim the transceiver's full 5 kV with the DC-DC still fitted, which is the configuration most people will actually want.
  • Add a bus-side supervisor. Right now nothing on the board notices if VCC2 drifts. A cheap UVLO or a zener-plus-LED would make a failed barrier supply visible instead of silent.
  • Fit the common-mode choke footprint back as 0 Ω links. It was cut to keep the stubs short, but leaving the pads would make the with/without EMI comparison a populate-or-don't choice rather than a respin.
  • Bring the address claiming and transport protocol into the firmware. Three PGNs on a compliant PHY is enough to prove the hardware; it is not enough to put the board on a real vehicle bus alongside other J1939 nodes.
  • A four-layer version. Two layers forced the ground pours into a shape the stitching vias have to compensate for. On four layers the return path under the bus pair stops being a routing problem.

Parts

41
QtyValueFootprintReferences
3100nCapacitor_SMD:C_0805_2012MetricC2, C4, C9
34u7Capacitor_SMD:C_0805_2012MetricC5, C6, C7
21k5Resistor_SMD:R_0805_2012MetricR3, R6
24n7Capacitor_SMD:C_0805_2012MetricC1, C8
260R4 1%Resistor_SMD:R_0805_2012MetricR1, R2
2BAT54Diode_SMD:D_SOD-123D1, D5
2SMAJ36CADiode_SMD:D_SMAD2, D6
2greenLED_SMD:LED_0805_2012MetricD3, D4
1100kResistor_SMD:R_0805_2012MetricR4
11MResistor_SMD:R_0805_2012MetricR5
11uCapacitor_SMD:C_0805_2012MetricC3
1240R 1%Resistor_SMD:R_1206_3216MetricR7
15V6Diode_SMD:D_SOD-123D7
1600R@100MHzInductor_SMD:L_0805_2012MetricFB1
1CANHTestPoint:TestPoint_Pad_D1.5mmTP1
1CANLTestPoint:TestPoint_Pad_D1.5mmTP2
1DE-9 male, CiA 303-1Connector_Dsub:DSUB-9_Pins_Horizontal_P2.77x2.84mm_EdgePinOffset9.90mm_Housed_MountingHolesOffset11.32mmJ2
1DMG2301LPackage_TO_SOT_SMD:SOT-23Q1
1GND1TestPoint:TestPoint_Pad_D1.5mmTP7
1GND2TestPoint:TestPoint_Pad_D1.5mmTP3
1Host 1x06Connector_PinHeader_2.54mm:PinHeader_1x06_P2.54mm_VerticalJ1
1ISO1042BDWVRcan-iso:TI_DWV0008AU1
1MKDS 1,5/5-5.08TerminalBlock_Phoenix:TerminalBlock_Phoenix_MKDS-1,5-5_1x05_P5.00mm_HorizontalJ3
1NXE1S0505MCConverter_DCDC:Converter_DCDC_Murata_NXExSxxxxMC_SMDU2
1RXDTestPoint:TestPoint_Pad_D1.5mmTP6
1SHLD hard bondConnector_PinHeader_2.54mm:PinHeader_1x02_P2.54mm_VerticalJP4
1TERM 120RConnector_PinHeader_2.54mm:PinHeader_2x02_P2.54mm_VerticalJP3
1TXDTestPoint:TestPoint_Pad_D1.5mmTP5
1VCC1 = 5V / 3V3Connector_PinHeader_2.54mm:PinHeader_1x03_P2.54mm_VerticalJP1
1VCC2TestPoint:TestPoint_Pad_D1.5mmTP4
1VCC2 = DCDC / busConnector_PinHeader_2.54mm:PinHeader_1x03_P2.54mm_VerticalJP2

Similar boards

  • 4 layers95 × 60 mmRubyLink revA, the printed circuit board in 3D

    RubyLink revA

    Misaka0x2730/RubyLink

    Hardware platform for Miyako-Link based on RP2040

    219 parts2
  • 4 layers160 × 100 mmMastertemp esp32, the printed circuit board in 3D

    Mastertemp esp32

    ryanbjohnson/PentairMasterTempESP32

    This folder is the KiCad project area for the MasterTemp 125 ESP32 interface board.

    95 partsMIT
  • 2 layers68.5 × 34.3 mmCurrent Sensor, the printed circuit board in 3D

    Current Sensor

    ExcessiveMotion/drive-hardware

    36 parts175CERN-OHL-W-2.0
  • 4 layers154.5 × 112.9 mmLibrePLC32, the printed circuit board in 3D

    LibrePLC32

    AndreiVladescu/LibrePLC32

    The goal of the project is to create a compact, hackable, and well-documented PLC platform for industrial automation experiments, prototyping, education, and open hardware development.

    295 parts14CERN-OHL-P-2.0
  • 4 layers69.5 × 62.6 mmSensors Module, the printed circuit board in 3D

    Sensors Module

    Queens-Rocket-Engineering-Team/Avionics-23-26

    QRET Avionics hardware, firmware, and documentation for 2023 to 2025.

    197 parts4
  • 4 layers143.1 × 87.6 mmThermistor Expansion Module, the printed circuit board in 3D

    Thermistor Expansion Module

    Gad9889/Thermistor-Expansion-Module

    KiCad project files remain at the repository root. Hierarchical schematic sheets are organized under schematics/, with connector sheets in schematics/connectors/ and gate sheets in schematics/gates/.

    114 parts

Questions

How big is the Can iso PCB?

It measures 100 × 50 mm and has 2 copper layers, 1.6 mm thick.

What parts does the Can iso use?

41 parts, built around the ISO1042BDWVR, NXE1S0505MC-R7, BAT54 and SMAJ36CA. The full list of its parts is on this page, with links to each part's symbol and footprint where PCBKit's library has them.

Where can I download the Can iso KiCad files?

From the karimf1/J939-transceiver repository on GitHub, at the commit this page was made from: the board is can-iso.kicad_pcb and its schematic can-iso.kicad_sch. PCBKit opens the board in the browser, with nothing to install.

Can I use the Can iso design in my own project?

Its repository publishes it under the MIT licence: you may reuse it on that licence's terms (keep the author's notice, and for a copyleft licence share your changes alike).