From office Ethernet to the vehicle harness

The single-pair journey began outside the IEEE. In 2011 Broadcom developed a proprietary 100 Mbit/s technology called BroadR-Reach specifically for in-vehicle use. BMW were the first to use it in 2013 to connect park-assist cameras, where the cost and weight of four-pair Cat-5 simply did not fit the application. The OPEN Alliance SIG was founded to promote and standardize the approach, and the IEEE picked it up shortly after.

The result is a family of single-pair standards. 100BASE-T1, essentially the hardened BroadR-Reach approach, was ratified as IEEE 802.3bw in 2015 and delivers 100 Mbit/s. 1000BASE-T1 followed as IEEE 802.3bp in 2016, applying the same single-pair, full-duplex philosophy at ten times the data rate. The lower end is covered by IEEE 802.3cg (2019), which defines both 10BASE-T1S for short-reach automotive multidrop links and 10BASE-T1L for long-reach industrial communication of up to 1,000 m. Higher data rates are addressed by IEEE 802.3ch (2.5/5/10 Gbit/s) and the more recent IEEE 802.3cy (25/50 Gbit/s) for next-generation ADAS and sensor backbones.

Figure 1: Comparison of Single-Pair and Standard Ethernet within the OSI model

The crucial point about all these standards is what they do not change. Only the physical layer of the OSI model is new. Above the PHY, everything is plain, unchanged Ethernet: the same MAC, the same frame format, the same EtherTypes. The same software stacks, the same protocol analyzers (e.g. Wireshark), the same higher-level protocols (TSN, MACsec etc.) work without modification. For transferring between 1000BASE-T on one side and 1000BASE-T1 on the other a simple media converter that swaps PHYs is sufficient - the bridging in between is ordinary Ethernet switching, with no protocol translation. This seamless interoperability with everything that already exists on the IT side is the single biggest reason single-pair Ethernet has been adopted so quickly: it inherits the complete Ethernet software and tools ecosystem essentially for free.

Single Pair Ethernet vs. Automotive Ethernet:
Clearing up the terms

The two labels are often used interchangeably, but they describe different things - and the second one is used quite loosely in practice. Single Pair Ethernet (SPE) is strictly a physical-layer family: every Ethernet variant that runs over one twisted pair instead of two or four is Single Pair Ethernet. The "T1" variants - 10BASE-T1S, 10BASE-T1L, 100BASE-T1, 1000BASE-T1, and the multi-gig successors - are all SPE.

Automotive Ethernet is broader and context dependent. Narrowly, it refers to SPE PHYs that meet the temperature, EMC, and connector requirements of the car industry - essentially 10BASE-T1S, 100BASE-T1, 1000BASE-T1 and their faster successors. In everyday OEM and Tier-1 usage, however, "Automotive Ethernet" usually means the whole stack that runs on top of these PHYs: AUTOSAR-aligned network management, SOME/IP service-oriented communication, DoIP diagnostics, security layers like MACsec, and the in-vehicle integration around them. Used in this wider sense, Automotive Ethernet is less a wire than a complete IP-based in-vehicle networking ecosystem.

For comparison: 10BASE-T1L is SPE, but it is normally not called Automotive Ethernet. It targets industrial field-bus replacement, with reach in the hundreds of meters and the option to carry power on the same pair (Power over Data Line, PoDL). Same physical-layer family, very different use case - and a good illustration of why the terms are not always interchangeable.

The communication foundation:
PAM-3 on a shared twisted pair

The line code used on 100BASE-T1 and 1000BASE-T1 is PAM-3 - Pulse Amplitude Modulation with three voltage levels on the wire: −1, 0, +1. For readers coming from a CAN background: CAN signals over a differential pair (the CAN High and CAN Low wires), are driving the bus into one of just two states - dominant and recessive - so each symbol is one bit and bit rate, and baud rate are interchangeable. With PAM-3 they are not. Each ternary (the physical transmission of the data is not binary like with CAN) symbol carries more than one bit, so the wire ticks slower than the data rate would suggest: 100BASE-T1 delivers 100 Mbit/s at a symbol rate of only 66.67 MBd, and 1000BASE-T1 delivers 1 Gbit/s at 750 MBd.

To make it concrete, imagine sending three payload bits, say 101, over each system. On CAN those 3 bits become 3 dominant/recessive symbols on the wire - one level per bit. On 100BASE-T1 the 3B2T mapping turns the same 3 bits into a code group of only 2 ternary symbols, each at one of three voltage levels. Same data, two-thirds the symbols on the wire.

Figure 2: Signaling method of 100BASE-T1

For a given data rate, 100&1000BASE-T1 runs at a lower baud rate on the wire than CAN-style binary signaling would, which helps keep the fundamental frequency - and therefore the radiated emissions - down. The cost is paid on the receiver side: with three voltage levels, the spacing between symbols is tighter than the dominant/recessive gap on CAN, so the PHY needs adaptive equalization, precise level slicing, and the echo cancellation to recover the signal. SPE is friendlier to EMC limits but less tolerant of channel imperfections. Noise, reflections, and impedance mismatches all interfere directly with the level spacing the receiver depends on. This is one reason why 1000BASE-T1 typically calls for a shielded twisted pair, where a CAN bus is happy on a much rougher cable.

None of this is exotic for Ethernet. 100BASE-TX, the classic office "RJ45 100 Mbit/s", already uses a three-level line code called MLT-3 on top of 4B5B coding, on two pairs with one direction each. 1000BASE-T goes a step further with PAM-5 (five levels) on all four pairs simultaneously. The T1 variants apply the same well-known techniques to a single twisted pair.

Two senders, one wire, how to avoid chaos

100BASE-T1 and 1000BASE-T1 use one pair of wires for both directions, simultaneously. Both PHYs drive the wire at the same time, with the same signaling, with no time-slotting or frequency separation.

If you put a wide-band oscilloscope probe across an active 100BASE-T1 link, what you see on the wire is the sum of the master’s transmitted PAM-3 signal, the slave’s transmitted PAM-3 signal, the channel’s frequency response applied to both, and reflections from impedance mismatches. Looking at the trace alone - without knowing what at least one side is transmitting - it is fundamentally not possible to tell the two streams apart. 

So how are the PHYs able to do that? The trick is that each PHY already knows what it is transmitting. That known signal - and its expected channel response and reflections - can be modelled and subtracted from what the PHY samples on the line. Everything that does not match the model is, by definition, the signal from the other PHY.

It uses an adaptive digital filter, the echo canceller, that takes the local transmit symbols as input and produces an estimate of how its own signal appears at its own receiver, including reflections. The receiver subtracts this estimate from the sampled wire signal. The remainder is the far-end PHY’s data, additionally processed by an equalizer (DFE/FFE) that compensates for the cable’s frequency response.

Two further conditions make this possible. First, one common bit clock: the link is asymmetric in role. One side is master, the other is slave. The master derives its symbol clock from its own oscillator; the slave recovers timing from the incoming master signal and uses the same clock for its own transmissions. Both directions are therefore symbol-synchronous, which is essential for the echo canceller to model the local signal precisely. Second, a training phase: the canceller’s filter coefficients are not known in advance - they depend on the actual cable, connectors, and PCB - and must be learned during a structured link startup. In practice this means configuring one participant as Master and one as Slave (PHY config).

On PHYs that support it, Auto-Negotiation (IEEE 802.3 Clause 98) can resolve both the link speed and the master/slave role automatically - each side advertises its supported speeds and its master/slave preference, and the link comes up at the highest mutually-supported rate with the roles assigned. Auto-Negotiation is optional, though, and only works if both PHYs are configured to use it for speed and master/slave selection; if one side is set to a fixed speed and role and the other to Auto-Neg, the link will not come up.

Strengths and limitations

A single unshielded twisted pair has the potential to cut connectivity cost dramatically compared to classic Ethernet. It reduces harness weight and meets the CISPR-25 and OEM EMC limits required for in-vehicle use. And, because everything above the PHY is the same as in any other Ethernet, the entire IT toolchain applies directly: simple media converters bridge SPE to RJ45 Ethernet with no protocol translation; Wireshark dissects the traffic; IP-based diagnostics and service frameworks (DoIP, SOME/IP) work natively; Linux, AUTOSAR, and POSIX networking stacks need no protocol-level changes to run on the new wire.
The limitations are equally real. Both 100BASE-T1 and 1000BASE-T1 are strictly point-to-point; any branching, even a simple stub, requires a switch. Reach is automotive-scaled: roughly 15 m on Type-A automotive cable, with an optional 40 m Type-B variant for 1000BASE-T1. The IEEE specifications make the channel-quality requirements explicit: the cable-plus-connector path has to meet defined limits on insertion loss (the frequency-dependent attenuation along the channel), return loss (how well the channel impedance matches the nominal 100 Ω, which keeps reflections small) and mode conversion (how cleanly the differential signal stays differential, which drives both radiated emissions and noise pickup). These are qualification limits the channel itself must meet - far stricter than anything CAN wiring has to deliver.  1000BASE-T1 is markedly more sensitive to channel quality than 100BASE-T1 and typically requires shielded twisted pair to meet emission limits, while 100BASE-T1 still works on UTP. And link startup, with its tightly choreographed master/slave handshake, makes bring-up debugging considerably less trivial than on a CAN bus.

100BASE-T1 and 1000BASE-T1 at a glance
 

Table 1: Comparison of 100BASE-T1 and 1000BASE-T1
 

Where 100/1000BASE-T1 is used

100BASE-T1 has become the default link for cameras, infotainment connections, and OBD/diagnostics-over-IP, replacing MOST and LVDS in those roles. 1000BASE-T1 carries higher-bandwidth ADAS data - front cameras, lidar pre-processing, radar fusion - and serves as the typical zone-to-central-compute backbone in modern E/E architectures. The faster 2.5G–10G variants are starting to appear in test benches and pilot vehicles as central HPC platforms consolidate sensor traffic, while 10BASE-T1S fills in the low end as a multidrop replacement for legacy CAN segments where Ethernet’s homogeneous addressing simplifies the gateway story.

What looked, ten years ago, like an experimental detour from "real" Ethernet is now an established in-vehicle networking technology - and the same ideas, with different reach and EMC profiles, are repeating themselves in industrial automation under the broader Single Pair Ethernet umbrella.