A camera in a modern vehicle sends gigabits of image data per second towards the central computer — and receives only a trickle of configuration and control messages in return. Asymmetrical Ethernet is built around that imbalance. It is the sensor interface the NAV Alliance promotes, and it is now being standardized in IEEE P802.3dm.
Key takeaways
- Sensor traffic is extremely asymmetric: multi-gigabit upstream, megabits downstream.
- Asymmetrical Ethernet uses a high-speed uplink and a low-speed back channel on the same single pair or coax, saving power, cost and heat at the sensor.
- Video enters the vehicle Ethernet network directly — switchable, time-synchronized, secured and diagnosed like any other traffic.
- Energy-Efficient Ethernet (EEE) further reduces power when the sensor idles.
- IEEE P802.3dm (Asymmetrical Electrical Automotive Ethernet) passed IEEE 802.3 working-group ballot in 2026 and is heading to IEEE SA ballot.
The problem: symmetric links for asymmetric traffic
Standard Ethernet PHYs are symmetric: a 10GBASE-T1 link offers 10 Gb/s in each direction at the same time. That is ideal between switches and computers, where traffic flows both ways. A camera, however, mostly talks and rarely listens. Its downstream traffic — register writes, exposure settings, frame triggers, time-synchronization messages, occasional firmware updates — needs perhaps tens of megabits per second at most. A symmetric PHY at the camera still has to build a full-rate transmitter and receiver for both directions, with the associated silicon area, power and heat inside a small, sealed housing that is often mounted behind a hot windshield.
That is why the industry has relied on point-to-point serializer/deserializer (SerDes) links for cameras: they are inherently asymmetric and efficient. But SerDes creates separate islands outside the Ethernet network, requiring bridge chips at the ECU, proprietary configuration channels and separate diagnostics.
How Asymmetrical Ethernet works
An asymmetrical link pairs a high-speed transmitter at the sensor with a low-speed transmitter at the ECU or zonal controller. Both share the same physical medium — a single shielded pair or a coaxial cable — and the PHYs separate the two directions with echo cancellation and frequency planning. Conceptually:
| Uplink (sensor → ECU) | Downlink (ECU → sensor) | |
|---|---|---|
| Purpose | Video, point clouds, radar cubes | Control, configuration, triggers, time sync, updates |
| Data rate class | Multi-gigabit (target speeds up to ~10 Gb/s in P802.3dm) | Tens to hundreds of Mb/s |
| Transmitter complexity | High (at the sensor) | Low (at the ECU) |
| Power focus | Minimize at the sensor end | Not critical |
Because the frames are Ethernet frames, the sensor can be a normal network endpoint with a MAC address. Its stream can be switched to several consumers — the perception computer, a redundant safety computer and a data logger — without special splitter hardware.
Why Ethernet end-to-end matters
- Switching and replication. Streams can be routed or duplicated by standard Ethernet switches, enabling redundancy and flexible software-defined architectures.
- Time synchronization. gPTP (IEEE 802.1AS) timestamps let the computer align frames from many sensors precisely for sensor fusion. See TSN in the vehicle.
- Security. MACsec can authenticate and encrypt sensor data on the link, defending against spoofed camera feeds. See our security article.
- One diagnostic model. Link quality, cable faults and errors are reported the same way as everywhere else in the network.
- Fewer bridge chips. Removing SerDes-to-Ethernet bridges in the ECU saves cost and board space.
Energy-Efficient Ethernet at the sensor
Even a busy camera does not transmit continuously: there are gaps between lines and frames, and many sensors are only needed in certain driving modes (for example parking cameras). Energy-Efficient Ethernet allows the PHY to enter low-power idle states when there is nothing to send and to wake within microseconds when traffic resumes. At the system level, OPEN Alliance TC10-style sleep/wake lets entire links power down when a function is not in use. Less power means less heat in the camera module, which in turn improves image-sensor noise and reliability. Read more in PoDL, EEE and sleep/wake.
Powering the sensor over the same cable
Cameras usually receive power over the same cable that carries their data, avoiding a separate power wire. On coaxial links this is commonly called power over coax; on twisted or parallel pairs, IEEE 802.3bu Power over Data Lines (PoDL) defines the approach. The PHY and the power-coupling network must be designed together so the power filter does not degrade the high-frequency data signal.
Standardization: IEEE P802.3dm
The IEEE 802.3 working group launched the P802.3dm project, titled Asymmetrical Electrical Automotive Ethernet, to add physical layer specifications and management parameters optimized for automotive end-node links such as cameras. The task force adopted a timeline that took its draft through IEEE 802.3 working-group ballot in 2026 — the first recirculation closed successfully in August 2026 — with IEEE SA ballot planned to follow and final approval targeted for 2027. NAV Alliance members are active contributors, consistent with the alliance’s long-standing promotion of Asymmetrical Ethernet for the sensor interface.
Sizing a camera link
The data rate of a camera depends on resolution, frame rate, bits per pixel and compression. A 2.9 MP front camera at 30 fps with 12-bit raw output needs about 1.1 Gb/s; an 8.3 MP camera at 60 fps with 12-bit raw needs roughly 6 Gb/s. Our interactive camera bandwidth calculator lets you explore these trade-offs and see which link speed fits.
Asymmetrical Ethernet vs. SerDes vs. symmetric Ethernet
| Criterion | Proprietary / standard SerDes | Symmetric Multi-Gig Ethernet | Asymmetrical Ethernet |
|---|---|---|---|
| Power at sensor | Low | Higher | Low |
| Network integration | Bridge required | Native | Native |
| Switching / replication | No | Yes | Yes |
| Standard security (MACsec) | Vendor-specific | Yes | Yes |
| Ecosystem | Per-vendor or per-standard | Broad Ethernet | Broad Ethernet |
Design considerations for camera modules
Moving from SerDes to Asymmetrical Ethernet changes how camera modules and ECUs are designed. A few practical points engineers should consider early in a program:
- Image-sensor interface. Most image sensors output MIPI CSI-2. The camera therefore needs a bridge function that packetizes CSI-2 data into Ethernet frames — for example using IEEE 1722 video formats — ideally integrated into the PHY or a small companion device. Our article on protocol encapsulation explains the options.
- Control path. Sensor registers are traditionally accessed over I2C. With Ethernet, those transactions travel in the low-rate back channel, either encapsulated or translated into a higher-level protocol. Latency for control is rarely critical, but frame triggers and synchronization are.
- Thermal budget. Every hundred milliwatts saved at the camera reduces sensor temperature and image noise. Asymmetric PHYs with EEE put the power where it matters.
- Diagnostics and functional safety. Link-quality monitoring, frame counters and timestamps let the perception software detect a frozen or corrupted image stream — a key safety mechanism.
- Harness reuse. Choosing a medium (shielded pair or coax) and connector system that the vehicle already uses for other high-speed links simplifies the harness. See cables and connectors for Multi-Gig.
What it means for the vehicle architecture
When cameras are Ethernet endpoints, architects gain freedom. A camera can be connected to the nearest zonal controller rather than routed all the way to the central computer, cutting cable length. Its stream can be forwarded to the compute cluster over the backbone, mirrored to a data logger during development and removed again in production by configuration rather than hardware changes. That flexibility is a cornerstone of the software-defined vehicle.
Frequently asked questions
Is Asymmetrical Ethernet only for cameras?
Cameras are the main use case, but radar, lidar and displays (where the traffic flows the other way) have similar asymmetric patterns.
Can asymmetrical and symmetric links coexist in one vehicle?
Yes. Sensors connect asymmetrically at the edge; zonal controllers and computers connect symmetrically on the backbone. All carry standard Ethernet frames.
Where can I learn more?
Visit our Asymmetrical Ethernet page or contact our member companies through the member directory.
Updated November 2026 to reflect the IEEE P802.3dm ballot status.