The physical frontier

Every few years a vendor announces a link technology that sounds like science fiction — light through air, terahertz radio, quantum-secured keys. Most engineers feel they have to relearn the physical layer each time. They don’t: one model decodes them all.

Long-haul fibre is an analog fight

Light through 5,000 km of single-mode glass is not a clean digital channel — it is an analog signal being slowly destroyed. Three impairments dominate. Chromatic dispersion: different wavelengths travel at slightly different speeds, so a pulse spreads out and bleeds into the next symbol. Polarisation-mode dispersion (PMD): the two orthogonal polarisations of light arrive at slightly different times. Four-wave mixing: in a DWDM system packing 60+ wavelengths into one fibre, nonlinear effects make adjacent channels generate spurious tones that interfere with their neighbours.

For decades the fix was physical: dispersion compensation modules (DCMs) — spools of specially designed fibre that bend the dispersion back. Modern long-haul has moved the fight into DSP. Coherent detection recovers not just the amplitude of the light but its phase, the same way a radio receiver mixes against a local oscillator. Once you have phase, you can run higher-order modulation (16-QAM, 64-QAM on light) and let a digital signal processor undo chromatic dispersion and PMD in software after the fact. The payoff is dramatic: a modern coherent transceiver carries 800 Gbps on a single wavelength, and 60+ wavelengths share one fibre — tens of terabits per fibre pair.

FEC is what makes the SNR margin real

The raw optical channel on a long submarine span has a bit error rate around 1e-6 — one bad bit in a million, far too noisy to ship. Yet the delivered BER is better than 1e-12. The entire gap is forward error correction. Long-haul optics use a concatenated scheme: an LDPC (Low-Density Parity Check) inner code with iterative soft-decision decoding, wrapped in a Reed-Solomon outer code that mops up residual burst errors. The result is 10 dB or more of coding gain — the channel behaves as if it had ten decibels more signal than it physically does, which translates directly into more bits per symbol or longer reach between amplifiers.

The trade is latency and silicon: FEC adds redundant bits (overhead, typically 7–25%), the decoder burns power and adds a fixed processing delay, and once you are FEC-limited the channel fails as a cliff — degrade the SNR slightly past the correction threshold and BER jumps from 1e-12 to unusable, with no gradual warning.

The radio frontier: Wi-Fi 7, 5G-Advanced, 6G

Radio keeps mining the same Shannon levers — more bandwidth, more bits per symbol. Wi-Fi 7 (802.11be) doubles the channel width to 320 MHz, pushes modulation to 4096-QAM (12 bits per symbol versus 10 for Wi-Fi 6), and adds Multi-Link Operation — one client transmitting on 2.4, 5, and 6 GHz concurrently rather than picking one. 5G-Advanced (3GPP Release 18, finalised in 2024) is the mid-life upgrade to 5G: AI/ML optimisation of the air interface, 20%+ RAN power reduction, and tighter satellite (non-terrestrial network) integration. 6G research (2030+) is reaching for terahertz frequencies, an AI-native air interface designed around learned components from day one, and integrated sensing — the network using its own radio to map the physical environment, not just carry data.

Co-packaged optics, LEO, and the quantum question

Two more shifts worth tracking. Co-packaged optics (CPO) moves the optical transceiver off the pluggable front panel and integrates it directly into the switch ASIC package — cutting the electrical distance signals travel, slashing per-port power, and enabling denser fabrics as port speeds climb past 800G. LEO satellite constellations (Starlink and peers) reshape the latency-versus-coverage trade: an orbit at ~550 km gives 20–50 ms RTT, an order of magnitude better than GEO’s ~600 ms, while reaching places no fibre will ever go.

Then the quantum question, where it pays to separate hype from practice. Quantum key distribution (QKD) uses the physics of photons to detect any eavesdropper on a key exchange — but it runs over fibre with a hard range limit near 100 km, because you cannot amplify a quantum state and quantum repeaters are still a research problem. The practical thing here now is the opposite: post-quantum cryptography. ML-KEM and ML-DSA (NIST’s standardised lattice algorithms, FIPS 203/204) are classical software that resists attack by a future quantum computer, and they are already shipping inside TLS handshakes. QKD is a decade-plus from changing day-to-day networking; post-quantum crypto is changing it this year.

The durable model

Standards documents for link technologies run to hundreds of pages, and they keep arriving. You do not need to read them cover to cover. When you hit an unfamiliar link technology — a new Wi-Fi version, an optical standard, an industrial fieldbus — three sections tell you 80% of what you need:

1. Physical medium — copper, fibre, or a radio band. This sets the propagation speed (latency floor) and the noise environment.
2. Encoding scheme — modulation order, line coding, and FEC. This sets bits per symbol and how gracefully the link degrades.
3. Frame format — header, payload boundaries, MTU. This sets overhead and how the link layer hands off to the network layer.

Cross-reference those three and you can estimate bandwidth, latency, error behaviour, and where real deployments will struggle — without reading the spec front to back.