Skip to content

Connectors, Cables & Physical Interfaces

Intermediate12 min read
Assumes: 0.1 Anatomy of an SoC, 0.6 EMC/EMI Fundamentals
TL;DR

Protocols define what’s in the message; the physical layer defines how it travels. CAN signals ride on differential voltage pairs through transceivers with 120-ohm termination. Automotive Ethernet uses a single twisted pair with BroadR-Reach PHY (100BASE-T1) — not the RJ45 you know from IT. Camera data flows through GMSL2 or FPD-Link III serializer/deserializer pairs at multi-gigabit rates. Connectors — FAKRA, HSD, MQS — are purpose-built for the vibration, temperature, and assembly constraints of vehicles. When your protocol-level debug tools show nothing wrong, the problem is often physical.

You trace CAN frames in Vector CANoe and ping the Ethernet backbone — but the bits only exist because transceivers, cables, and connectors translate protocol logic into voltages on copper. This page is the layer below your stack traces.

If you’ve ever burned a sprint chasing a “CAN stack bug” only to find a 120Ω resistor that wandered off the schematic — or watched three senior engineers argue about frame IDs while an intern wiggles a connector and the bus magically heals — you’ve already lived this page. Welcome to the layer where software is innocent and copper is guilty until proven otherwise.

The CAN controller (inside the MCU/SoC) handles framing, arbitration, and error counters. The transceiver is the chip that actually drives the bus wires. Common parts you’ll see on schematics and BOMs:

Part Vendor Notes
TJA1042 / TJA1043 NXP Workhorse automotive CAN transceivers; 1043 adds standby/sleep modes
TCAN1042 TI Pin-compatible alternative; same differential PHY behavior

CAN uses two wires: CAN_H and CAN_L. The receiver doesn’t care about absolute voltage — it cares about the difference between them.

Bus State CAN_H CAN_L Meaning
Recessive (logic 1) ~2.5V ~2.5V No node driving; differential ≈ 0V
Dominant (logic 0) ~3.5V ~1.5V At least one node driving; differential ≈ 2V

Mnemonic: “Recessive = both rest at 2.5V. Dominant = H high, L low — someone is talking.”

Dominant bits overwrite recessive bits on the wire. That’s how arbitration works — and why “Low and Slow wins” (more on that in the Mental Model below).

A CAN bus is a transmission line. At each physical end of the bus, you place a 120Ω resistor between CAN_H and CAN_L. Two terminators in parallel give a 60Ω characteristic impedance — what ISO 11898 specifies for the twisted pair.

Missing termination kills the bus. Without proper impedance matching, signal edges reflect at stubs and open ends. At 500 kbps you might get away with sloppy termination on a short bench harness. At CAN FD data-phase rates (2–5 Mbps), reflections cause bit errors — often intermittent, temperature-dependent, and maddening to debug in software.

CAN FD uses the same PHY as classic CAN. The transceiver and cable don’t change; only the controller bit-timing in the data phase increases. Physical-layer margin shrinks fast.

flowchart LR
    subgraph ECU1["ECU 1"]
        MCU1["CAN Controller"] --> TR1["Transceiver<br/>TJA1042"]
    end
    subgraph ECU2["ECU 2"]
        MCU2["CAN Controller"] --> TR2["Transceiver<br/>TJA1042"]
    end
    subgraph ECU3["ECU 3 (stub)"]
        MCU3["CAN Controller"] --> TR3["Transceiver<br/>TJA1042"]
    end

    TR1 ---|"CAN_H / CAN_L<br/>twisted pair"| TR2
    TR2 ---|"2m harness"| TR3

    TR1 -.- R1["120Ω"]
    TR3 -.- R2["120Ω"]

    style R1 fill:#f59e0b,stroke:#d97706
    style R2 fill:#f59e0b,stroke:#d97706

Terminators belong at the two ends of the longest path, not at every node. A third ECU on a 2-meter stub changes the electrical topology — if termination wasn’t updated, you just invited reflections.

Mental Model

Physical interfaces are the plumbing of the vehicle’s nervous system — think Die Hard air vents, not a pristine IKEA bathroom. The protocol (CAN, Ethernet) is the building code: pressure limits, flow rules, who gets to talk first. The physical layer (transceiver, cable, connector) is the actual pipe run through a vibrating metal box for 200,000 km. Perfect regulations don’t matter if the pipe has a crack (bad termination), a kink (pinched harness), or a corroded joint (0.5µm gold plating — we’re looking at you, supplier audit).

Mnemonic: “Low and Slow wins.” In CAN arbitration, the lower ID value wins because the dominant bit (logic 0 — the LOW voltage state on the differential pair) overwrites the recessive bit. The node with the lowest ID number transmits first. “Slow” refers to the lower numeric ID, not baud rate — the quietest ID on the bus, not the slowest talker.

Think About It

3 AM. Pager: “CAN bus flaky on pilot fleet — all three ECUs throwing errors.” Two ECUs worked fine for months. Yesterday, a third ECU went in 2 meters down the harness. No software update, no new DBC, no guilty commit. CANoe still decodes frames like nothing happened. What physical change broke the bus?

Reveal Answer

Adding the third node extended the bus topology and placed a transceiver on a 2-meter stub without re-evaluating 120Ω termination at both physical ends of the full bus. The two-node setup may have masked impedance mismatch on a short harness; the longer path increases round-trip delay and reflections — bus impedance reads 120Ω instead of 60Ω if only one terminator exists. Measure with a TDR or ohmmeter across CAN_H/CAN_L (power off) and confirm terminators sit at the ends of the longest path, not at the new stub node.

CAN sorted? Deep breath — Ethernet in a car is a whole different PHY cosplay.

Automotive Ethernet is not office Ethernet in a different jacket. Different PHY, different cable, different connector, different power management.

Standard Speed Media Reach Role
100BASE-T1 (BroadR-Reach) 100 Mbps Single unshielded twisted pair (UTP) ~15m Camera ECUs, ADAS sensors, zonal gateways
1000BASE-T1 1 Gbps Single twisted pair ~15m Domain backbone — cockpit to zonal controllers

Common PHY chips: Marvell 88Q2112 (100/1000BASE-T1), TI DP83TC811 (100BASE-T1). These integrate the single-pair PHY, cable diagnostics, and often SQI (Signal Quality Index) reporting.

Partial networking is the automotive equivalent of wake-on-LAN: PHYs and switches support sleep/wake patterns so ECUs draw microamps in park mode but wake targeted subnets on ignition or CAN-triggered events — critical for parasitic drain budgets.

flowchart TB
    subgraph Office["Office Ethernet (100BASE-TX)"]
        O_PHY["Standard PHY"] --> O_Cable["4-pair Cat5e/6<br/>8 conductors"]
        O_Cable --> O_Conn["RJ45 jack"]
    end

    subgraph Auto["Automotive Ethernet (100BASE-T1)"]
        A_PHY["BroadR-Reach PHY<br/>88Q2112 / DP83TC811"] --> A_Cable["Single UTP<br/>2 conductors"]
        A_Cable --> A_Conn["HSD connector<br/>locking, sealed"]
    end

    style Office fill:#e0f2fe,stroke:#0284c7
    style Auto fill:#fef3c7,stroke:#d97706

No RJ45. No Cat5/6 four-pair cable. You can’t plug your laptop into the car’s Ethernet backbone — different PHY, different link training, different voltage levels.

Okay, backbone Ethernet is weird. Cameras said “hold my coax” and went full gigabit.

Surround-view and ADAS cameras need gigabit bandwidth but can’t run four shielded pairs to every bumper corner. SerDes (serializer/deserializer) links solve this: one high-speed lane over coax or shielded twisted pair (STP).

Link Vendor Speed Media Power
GMSL2 Maxim / Analog Devices Up to 6 Gbps Coax or STP PoC (Power over Coax)
FPD-Link III TI Up to ~6 Gbps Coax or STP PoC / PoDL variants

Topology: Serializer IC at the camera module → coax/STP cable (often 5–15m) → Deserializer IC at the SoC/ISP side. One cable carries video, control (I2C), and power.

Why this matters for you: A surround-view system with four cameras would need impractical wiring without SerDes. The ISP on your Qualcomm SoC receives MIPI CSI-2 from the deserializer chip — your software sees a camera pipeline; hardware sees a multi-gigabit serial link that fails silently when cable impedance or PoC decoupling is wrong.

flowchart LR
    CAM["Camera Module<br/>image sensor"] --> SER["Serializer<br/>GMSL2 / FPD-Link III"]
    SER -->|"Coax / STP<br/>data + PoC power"| DES["Deserializer"]
    DES --> ISP["SoC ISP<br/>MIPI CSI-2"]

SerDes is the magic trick. Connectors are where the magic meets a technician with gloved hands and a torque spec.

Connectors are where physics meets assembly-line reality. Vibration, temperature cycling, and mate/de-mate cycles define the part — not just pin count.

Connector Primary Use Key Properties
FAKRA RF — antenna, V2X, radar Color-coded by frequency (e.g., coral = GPS, bordeaux = V2X); 50Ω impedance; mate cycles rated
HSD (High-Speed Data) Ethernet, LVDS, USB, SerDes Locking mechanism; IP67 sealing options; designed for high-bandwidth differential pairs
MQS (Micro Quadlock System) Low-current signals Compact; used in body/chassis modules for sensor and actuator connections
USCAR Power connections High-current battery and distribution; vibration-rated latching

Spec sheet items that matter: mate/de-mate cycle rating (e.g., 10–50 cycles for service connectors), vibration class (ISO 16750), IP rating, and contact plating thickness — more on that in the war story below.

Connectors are the breakup text of automotive debug — “it’s not you, it’s the fretting corrosion.” Time to stop guessing and start measuring.

At high speeds — Ethernet PHY, SerDes, MIPI — you stop trusting “does it ping?” and start measuring signal integrity. The eye diagram overlays thousands of bit transitions on one plot.

What You See Meaning
Open eye (wide, tall diamond) Clean signal — adequate voltage and timing margin
Closed eye (collapsed opening) Bit errors imminent — reflections, crosstalk, or bad termination
Eye height Voltage margin — how far the signal is from the decision threshold
Eye width Timing margin — jitter tolerance before sampling errors
Jitter Horizontal spread of transitions — clock recovery stress

You’ll encounter eye diagrams in Ethernet PHY compliance testing (1000BASE-T1 conformance suites), SerDes validation for camera links, and supplier DV reports. Software can’t fix a closed eye.

Common Gotcha

Beginner thinks automotive Ethernet is just regular Ethernet with different connectors. (Same energy as “it’s just Linux” on a QNX cluster — technically related, practically a different planet.)

Reality: different PHY (single pair vs four pairs), different connector (HSD vs RJ45), different power management (partial networking), different EMC requirements (CISPR 25 limits on the harness). Link training, cable diagnostics, and SQI thresholds are automotive-specific. You can’t plug your laptop into the car’s Ethernet backbone.

Common Gotcha

“Missing one of two CAN termination resistors doesn’t matter.” (One sock in the dryer theory — surely the other one is doing enough work.)

Reality: with only one 120Ω terminator, bus impedance is 120Ω instead of 60Ω. Reflections increase. At 500 kbps this might work on a short bench harness. At CAN FD data rates (2–5 Mbps), bit errors become frequent and intermittent — the worst kind to debug because CANoe shows correct frame definitions while the bus physically corrupts every third edge.

Signal Integrity Basics

Impedance matching means the source, cable, and load all present the same characteristic impedance (60Ω for CAN, 100Ω for Ethernet pairs). Mismatch → reflections → ringing on edges → sampling errors.

Transmission line effects matter when cable length approaches λ/10 at the signal frequency (rule of thumb). At 500 kbps CAN, wavelength is hundreds of meters — a 2m harness isn’t a transmission line problem for propagation delay, but stub length and termination still matter for edge integrity. At 100 Mbps Ethernet, λ/10 is roughly 7 meters on UTP — automotive harness lengths enter the danger zone. At 6 Gbps SerDes, even centimeters of unmatched trace length on the PCB matters.

Why controlled impedance matters for Ethernet and SerDes but not CAN at 500 kbps: higher bit rates have shorter bit periods and faster edge rates. The same reflection that smears a 2 ns edge at 1 Gbps is invisible when your bit is 2 µs wide at 500 kbps. CAN FD closes that gap — treat it with the same termination discipline as a high-speed bus.

Where You'll See This

When to suspect the physical layer before opening a software ticket:

Symptom Physical-Layer Clue
Intermittent errors, no pattern in logs Loose connector, cracked solder joint, marginal termination
Failures only hot or only cold Thermal expansion on contacts, humidity ingress past seal
Errors after connector mate/un-mate Bent pin, insufficient insertion depth, fretting corrosion
Works on bench, fails in vehicle Harness length, routing next to power return, missing termination

Tools you’ll reach for: oscilloscope with differential probe for CAN (measure CAN_H − CAN_L, not either line to ground), PHY SQI/cable diagnostic registers for Ethernet, TDR/cable tester for impedance faults.

War story: An OEM had a fleet issue — 5% of vehicles had intermittent CAN errors after 18 months in service. Root cause: a connector supplier changed gold plating thickness from 0.8µm to 0.5µm without notification. Fretting corrosion developed at the contact interface under vibration. CAN frames looked valid in CANoe on a cold bench; failed under thermal cycling in the field. Fix: supplier audit, incoming inspection spec for plating thickness, and connector lot traceability. Software was innocent the entire time.

Connect the Dots

Module 1.2 (ECU Networks): Protocol-level CAN and Ethernet — frame formats, addressing, topology — all ride on the transceivers and cables described here. Module 3.1 (Signal-Based): Vehicle signals travel through these physical paths before they become service definitions. Module 7.4 (ADAS Sensors): Camera links use GMSL2/FPD-Link SerDes from this page. Module 10.3 (Debug Interfaces): UART and JTAG are also physical interfaces with specific connectors, voltage levels, and probe requirements.

  1. Protocols are the message; PHY is the medium — transceivers, cables, connectors, termination.
  2. CAN: differential CAN_H/CAN_L, 120Ω at both bus ends, dominant/recessive voltage levels.
  3. Automotive Ethernet: single-pair 100/1000BASE-T1, HSD connectors — not RJ45.
  4. Cameras: GMSL2/FPD-Link SerDes with PoC — one coax carries data and power.
  5. When software looks clean and the bus doesn’t, measure the physical layer first.

Check Your Understanding

1. A CAN bus with two nodes works fine. Adding a third node 2m away causes bit errors. Most likely physical cause?

Applied

2. What makes automotive Ethernet physically different from office Ethernet?

Recall

3. In CAN arbitration, 'Low and Slow wins' — what does 'low' refer to?

Recall

4. What is the correct CAN bus termination configuration per ISO 11898?

Recall

5. CANoe shows valid frame definitions but the bus has a 15% error rate at 2 Mbps CAN FD data phase. Classic CAN at 500 kbps on the same bus works. First physical-layer check?

Applied

6. A surround-view camera shows green-tinted frames after a harness redesign routed GMSL2 coax alongside a high-current 12V return. Scope shows noise spikes synchronized with PWM loads. CANoe shows clean vehicle-speed frames. Root cause?

Synthesis

Connects to: Module 0.6 — EMC/EMI · Module 7 — Connectivity & Telematics

Module Review: Hardware Foundations

Question 1 of 10

Post-mortem on a field failure: 5% of vehicles develop CAN errors after 18 months. Root cause is fretting corrosion at a connector with reduced gold plating. CANoe showed valid frames on cold bench. Why did software teams waste 3 weeks first?

🔗 Connects to: Module 10 — Debug Decision Tree

Further Reading