Low-Loss Optical Fibre

Cut a 10 m length of 1 mm PMMA fibre. The end face is the whole measurement, so cut it properly: hold the fibre against a hard flat surface and press a fresh razor blade straight down in one motion. Do not saw and do not use side cutters, which crush the core. Inspect the face with the hand lens. It should be flat, clear, and square to the axis. A chipped or angled face throws a couple of decibels away before the light even enters, and it will not do so repeatably. Wipe both faces with isopropyl alcohol on a lint-free wipe. Butt the input face against the laser module, hold it with tape, and read the output on the photodiode. Note the reading. Do not disturb the launch end again for the rest of this measurement.

Record the received power. Then cut 2 m off the FAR end, re-cut that face cleanly, and record again. Repeat down to 1 m, giving six points. Cutting the far end rather than the launch end is deliberate and is the opposite of the formal standard, which cuts near the launch. Either works provided the LAUNCH is never disturbed, and for a hand-held butt joint the launch is by far the most fragile part of the setup. Convert each reading to dBm and plot against length. The slope is the attenuation in dB per metre, and every fixed loss — the launch, the detector coupling, the source power — sits in the intercept where it cannot contaminate the answer. That separation is the whole point of a cutback. While the fibre is still connected, wrap it once around a pencil and watch the power drop. That is bend loss: past the critical angle the guided ray is no longer guided.

WHAT KAO ACTUALLY CLAIMED. Not a fibre — a number, and an argument about where the loss came from. He and Hockham split the measured attenuation into scattering, which follows a known 1/lambda^4 law and is small, and absorption, which was enormous and tracked the transition-metal content. The conclusion: iron, copper, chromium, nickel and vanadium had to come down to parts per BILLION. No bulk melt had ever reached that, and none could, because the crucible itself contaminates the melt. WHY VAPOUR DEPOSITION WAS THE ANSWER. If you cannot purify a melt, do not make one. Silicon and germanium chlorides distil to extreme purity and are volatile; the chlorides of the offenders are far less so. Build the glass out of vapour and the impurities stay in the bottle. That is the next blueprint, and it is a direct consequence of this one. THE THREE WINDOWS. 850 nm came first because GaAs lasers and silicon detectors already worked there; about 2 dB/km. 1310 nm next, because silica's material dispersion passes through zero there so pulses stop spreading; about 0.35 dB/km. 1550 nm is the loss minimum at about 0.19 dB/km, awkward at first because dispersion is high — until dispersion compensation solved that and the erbium amplifier at the end of this batch made 1550 nm the only wavelength anyone cared about. THE SIBLING, MEASURED. Plastic optical fibre: 1 mm core, NA 0.5, cuts with a razor, couples to a bare LED, terminates by hand, pennies a metre, 200 dB/km because PMMA's own C-H bonds absorb in the red. Silica single-mode: 8 micrometre core, needs a cleaver and a fusion splicer, 0.19 dB/km. POF is not bad silica — it is a different material with a different loss mechanism aimed at a different distance, 100 m inside a car or a machine where termination cost dominates and loss does not. Both are correct answers to different questions. AGAINST COPPER. The 1858 submarine telegraph cable this blueprint links to failed on capacitance: a long conductor in seawater is a distributed capacitor, pulses smear, and the only cure was to send slower. Fibre's enemies are attenuation and chromatic dispersion, and both yield to better glass and a better wavelength. Same ocean, and a limit that could be engineered away rather than merely lived with.