The fibre link that fails from too much power

The sensitivity margin reads 30 dB, so power is the first thing ruled out. That is exactly why this costs so many hours.

Two thresholds, not one

Everyone knows a fibre link fails if too little light arrives. Fewer people design around the other end of the range.

Every optical receiver has an overload threshold — sometimes called maximum input power or saturation — above which the photodiode stops behaving linearly. The eye closes, errors appear, and the link is out of specification. Not marginal: out.

The window between sensitivity and overload is often only twenty decibels or so. A transceiver specified for eighty kilometres of fibre assumes about twenty decibels of loss on the way. Patch it across a rack into an identical unit and roughly none of that loss happens.

Why it costs so much time

Because the diagnostic instinct is exactly wrong.

The first thing anyone checks is received power. It reads +1 dBm against a sensitivity of −28, which is a margin of twenty-nine decibels — enormous, obviously healthy — so power gets ruled out immediately and attention moves to transceivers, patch leads, line cards, configuration, firmware.

The number that matters was on the same screen. Nobody compared it with the maximum, because the maximum is not what a power reading is normally checked against.

The fix is an attenuator. A passive in-line unit costing a few pounds, in the receive path, and the link comes up. It is counterintuitive enough to be worth saying explicitly: the link needs less signal, not more.

Where it happens

• Rack-to-rack patching with long-reach optics, because they were what was in stock.
• A route that got shorter. A new duct halves the path, the optics stay, and a link that worked for years starts erroring after an improvement.
• Lab and staging. Two long-reach transceivers on a two-metre lead, which is how a great many people first meet this.
• After a repair that removed a splice enclosure and its accumulated loss.

What the budget is actually for

A loss budget is a column that adds up: fibre attenuation, splices, connector pairs, splitters and muxes, plus a margin. Subtract it from transmit power and you have the level at the far end. Compare that with both thresholds.

Structuring it as a readable column matters as much as the total. A budget you can read down is one you can check against a route — and errors here are the ordinary kind: a patch panel counted at one end and not the other, a splice count from the wrong drawing, a splitter left out.

The margin is not padding

Three decibels is the usual planning figure and it covers things that are certain to happen without being individually predictable.

Repairs. A cable cut adds two splices, permanently. Over twenty years, a route through a city will be cut more than once. This is the main reason the margin exists.

Ageing. Connectors degrade with mating cycles. Splices drift slightly. In some conditions hydrogen ingress raises attenuation over years.

Measurement reality. The attenuation on a datasheet is a specification, not a measurement of your particular drum, and a route measured at 0.25 dB/km against a specified 0.22 is completely normal.

Later work. A mid-span joint added to serve a new building is loss nobody budgeted for.

A link commissioned with no margin will fail during its service life, at a moment nobody chooses.

On short routes, the glass is the minority

Over two kilometres at 0.35 dB/km, the fibre costs 0.7 dB. Six connector pairs cost over two. Reducing patch panels buys far more than better cable would — and this inverts on long routes, where the glass dominates and a connector or two is noise.

It is worth knowing which regime a route is in before optimising the wrong end of it.

Why 1550 nm is the quiet window

Two mechanisms cross there.

Rayleigh scattering comes from microscopic density variations frozen into the glass as it cooled. It falls as the fourth power of wavelength, so it dominates at short wavelengths — and it is why 850 nm multimode is an order of magnitude worse than 1550 nm single-mode.

Infrared absorption is the glass itself beginning to absorb, and it rises with wavelength, dominating beyond about 1600 nm.

The minimum sits where the curves cross, near 1550 nm. That is not a coincidence of engineering — it is a property of silica, and it is why long-haul systems live there and why erbium-doped amplifiers, which happen to work in that band, mattered as much as they did.

What to measure rather than assume

The datasheet figure is a starting point. The route's actual loss is an OTDR trace and an end-to-end power measurement, and those are the numbers to commission against. A budget's job is to tell you whether to expect a problem, and to give you something to compare the measurement with when there is one.

Frequently asked questions

Can a fibre link fail because of too much power?

Yes. Every receiver has an overload threshold as well as a sensitivity, and above it the photodiode saturates and the link errors. Long-reach optics patched across a rack are the classic case — and the sensitivity margin reads 30 dB, so power is the first thing ruled out.

How do I fix an overloaded receiver?

An in-line optical attenuator in the receive path, sized to bring the level inside the window. It is counterintuitive — the link needs less signal, not more — which is why the failure survives so long undiagnosed.

What margin should I allow?

Three decibels is a common figure. It covers repairs — a cable cut adds two permanent splices — connector ageing, later mid-span joints, and the gap between a datasheet specification and your actual drum. A link with no margin will fail during its service life.

Why is 1550 nm lower loss than 1310 nm?

Rayleigh scattering falls as the fourth power of wavelength and dominates at short wavelengths; infrared absorption rises with wavelength and dominates beyond about 1600 nm. The minimum is where they cross, near 1550 nm.

Should I use datasheet attenuation or measured?

Measured, wherever it exists. The datasheet is a specification, not a measurement of your cable, and 0.25 dB/km on a route specified at 0.22 is unremarkable. Commission against an OTDR trace and an end-to-end power measurement.

Why does reducing connectors help more than better cable?

On short routes only. Over two kilometres the glass costs under a decibel while six connector pairs cost over two. On long routes it inverts and the glass dominates — so it is worth knowing which regime you are in before optimising.

How much loss does a splice add?

Around 0.1 dB is a reasonable planning figure for fusion. A poor field splice is several times that, mechanical splices are higher and less stable, and the OTDR trace beats any table once a route is built.