Guided light in fibre

Why does light stay trapped inside a bent glass thread instead of escaping at the curve?

Why the light stays inside the bend is a question about a threshold — total internal reflection at the core-cladding boundary, and the critical angle that decides it — answered in this collection's companion piece on guided light, so take it as settled and ask the engineering question that follows.

Because confinement is not the achievement. A glass thread keeps light in for a kilometre or ten, and then the light is gone — not out of the side, but absorbed and scattered inside the glass itself. Between that and a cable across the Atlantic there is an enormous distance. What has to be built to cross it?

Start with the loss budget, because it settles the argument immediately. The best silica fibre loses about 0.2 decibels per kilometre at 1550 nanometres. Now scale that: six thousand kilometres of ocean is 1,200 decibels, and a decibel is logarithmic, so that is not a large loss but an inconceivable one — ten raised to the power of a hundred and twenty. The signal must be rebuilt along the way, and the only question is how.

The first answer was the obvious one: a regenerator. Detect the light, recover the bits electrically, reclock them, drive a fresh laser. It works, and it has an eventually fatal property — it must understand the signal. A regenerator is built for one wavelength, one bit rate, one modulation format. Put ten wavelengths down the fibre and you need ten at every station; upgrade the terminals to a faster line rate and every regenerator becomes wrong. On the sea floor, changing one means a cable ship.

So consider what a device would need in order to avoid understanding the signal: it would have to amplify light as light, without conversion. That is what an erbium-doped fibre amplifier does — a short length of fibre doped with erbium ions, illuminated by a pump laser at 980 or 1480 nanometres, so a passing signal photon stimulates the excited ions to emit copies of itself. There is a happy coincidence at its heart: erbium's emission band, roughly 1530 to 1565 nanometres, sits almost exactly where silica is most transparent. And because such an amplifier is indifferent to what the light carries, it amplifies the whole band at once — so you can put many wavelengths down one fibre, wavelength-division multiplexing, and one amplifier serves all of them. The amplification chain stops scaling with the number of channels, and upgrading the terminals raises capacity without touching anything in between, which is precisely what the regenerator forbade.

What does amplification not fix? Three things, and they define the rest of the engineering. Every amplifier adds spontaneous-emission noise along with its gain, and that noise accumulates while the signal does not improve, so the span count is limited by signal-to-noise rather than by loss. Second, chromatic dispersion keeps accumulating — wavelengths travel at slightly different speeds, around 17 picoseconds per nanometre per kilometre at 1550 nm, and an amplifier that boosts a smeared pulse returns a brighter smeared pulse; modern coherent receivers detect the light's phase as well as its intensity and undo it digitally at the far end. Third, you cannot simply amplify harder: glass's refractive index changes very slightly with intensity, so at high power the channels distort themselves and each other. There is an optimum launch power, and pushing past it makes things worse.

Before any of that sits the accounting that dominates a short link. Fibre comes in reels, so a long route is many pieces joined. A fusion splice — the cores melted together — costs a few hundredths of a decibel; a demountable connector costs several times more and presents a polished face that reflects and that a fingerprint can ruin on a core nine micrometres across.

Why is a submarine cable a different problem rather than a long terrestrial one? Because of a constraint with no land equivalent: the amplifiers must be powered from the beach. A copper conductor around the fibres carries a constant current, on the order of an ampere, fed at up to some tens of kilovolts from shore stations at both ends, with every repeater in series on that single circuit. Total electrical power is a hard, shared budget.

That one constraint reshapes everything downstream. It sets repeater spacing, typically fifty to a hundred kilometres, since closer spacing gives better signal quality but needs more repeaters on the same feed. It rules out anything adjustable or serviceable, because a fault means a ship, a grapnel and weeks of weather — so components are qualified to a design life of about twenty-five years, pump lasers are duplicated, and there is nothing inside to tune. And it produced the counter-intuitive turn in modern cable design: for years the goal was the most capacity per fibre pair, but once you are power-limited rather than bandwidth-limited that is wrong, because running each amplifier at lower gain is more efficient per watt, so the same shore power buys more capacity spread across many more pairs. The binding constraint moved, and the design followed.

• Why Raman amplification, which uses the transmission fibre itself as the gain medium, complements erbium amplifiers rather than replacing them.
• How the nonlinear ceiling on launch power sets a practical capacity limit for a fibre.

Every term the collection defines is gathered in the glossary.