Closing the loop: from emitting light to reading it
In rung 3 you ran a semiconductor junction *forwards* to make light: pour current into a light-emitting diode or a laser diode and electrons drop across the band-gap, each one shedding its energy as a photon. The whole story of this rung is one beautiful symmetry — run a junction the *other* way and it does the reverse. Let a photon fall onto a reverse-biased junction and, if the photon carries more energy than the band-gap, it knocks an electron loose and creates a tiny pulse of current. Light in, current out. That device is the photodiode, and it is the receiver that completes every optical link.
EMITTER (rung 3) DETECTOR (this rung)
─────────────── ────────────────────
forward-biased junction reverse-biased junction
current ──►──┐ photon ────►──┐
│ electron + hole h ν │ frees an
▼ recombine ▼ electron-hole pair
┌─────────┐ ┌─────────┐
│ P │ N │ ── photon out ──► light │ P │ N │ ──► photocurrent
└─────────┘ (LED / laser) └─────────┘
electrical → optical optical → electrical
Same PN junction. Same band-gap E_g. Run it forwards to SHINE,
run it backwards to SEE.A good photodiode is astonishingly sensitive — it can register a handful of photons — and astonishingly fast, switching in tens of picoseconds, which is why fibre links run at tens of gigabits per second. Two refinements matter in practice. The PIN photodiode adds an undoped *intrinsic* layer between P and N to widen the light-catching region and speed up the response. The avalanche photodiode (APD) biases the junction so hard that each freed electron triggers a cascade — built-in gain, like a microphone with its own pre-amp — buying extra sensitivity when the arriving light is desperately faint.
How glass traps light: total internal reflection
Now the channel — and it rests on a trick you've half-seen at a swimming pool. Look up at the underside of the water surface from below at a shallow angle and it turns into a perfect mirror; you can't see out at all. That mirror is free, requires no silvering, and loses almost nothing. It is called total internal reflection, and it is the entire reason a strand of glass can pipe light around corners and across oceans.
Light bends when it crosses between materials of different *refractive index* n — glass (n ≈ 1.47) slows light more than air (n ≈ 1.0). When a ray travelling in the denser material hits the boundary at a shallow enough angle, Snell's law has *no real solution* for the refracted ray: there is nowhere for the light to go but back inside. Past a sharp threshold called the critical angle, 100% of the light reflects. A optical fiber engineers this on purpose: a glass core is wrapped in a cladding of slightly lower index, so any ray launched within a narrow cone stays forever trapped, bouncing down the core in a zig-zag, kilometre after kilometre.
CROSS-SECTION LIGHT BOUNCING DOWN THE CORE
───────────── ───────────────────────────
┌───────────────┐ cladding n2 (lower index)
│ cladding │ n2 ╔═══════════════════════════════╗
│ ┌─────────┐ │ ║ ║
│ │ core │ │ n1 > n2 ║ ●─►╲ ╱╲ ╱╲ ╱─► ║ ──► out
│ │ (glass)│ │ ║ ╲ ╱ ╲ ╱ ╲ ╱ ║
│ └─────────┘ │ ║ ╲__╱ ╲__╱ ╲__╱ ║
└───────────────┘ ║ each bounce: angle > θ_crit ║
╚═══════════════════════════════╝
core ~9 µm (single-mode) TOTAL internal reflection → ~0 loss per bounce
to ~50 µm (multi-mode) core n1 (higher index)
Critical angle: sin θ_crit = n2 / n1
Acceptance cone: NA = √(n1² − n2²) ("numerical aperture")What the channel does to your light: attenuation & dispersion
No channel is perfect, and fibre has its own version of the comms track's enemies. The two that set every limit are attenuation — the light gets dimmer — and dispersion — the pulses get blurrier. Master these two and you can predict how far a link reaches and how fast it can run.
Fiber attenuation is the steady loss of optical power as light travels, caused mostly by Rayleigh scattering off the glass's microscopic structure and by absorption from trace impurities. Crucially, it depends on *wavelength*. Early fibre lost light fast, but engineers found magic windows where glass is astonishingly clear. At 1550 nm — in the infrared, invisible to your eye — the best fibre loses only about 0.2 dB per kilometre. Let that sink in: light entering one end is still at 1% of its power after 100 km. No copper wire on Earth comes close.
Dispersion is the subtler foe. It doesn't weaken the light — it *spreads each pulse out in time* until neighbouring pulses overlap and the receiver can no longer tell a 1 from a 0. Two flavours dominate. Modal dispersion (only in multi-mode fibre) is the many-paths smearing from the last section. Chromatic dispersion is sneakier: a real laser pulse isn't a single colour but a narrow spread of wavelengths, and glass carries each wavelength at a slightly different speed, so the pulse stretches even in single-mode fibre. Dispersion is exactly the distortion enemy you met in communications, wearing a photonics costume.
ATTENUATION (power fades) DISPERSION (pulses blur)
───────────────────────── ────────────────────────
P_in ████ → ██ → ▓ → ░ P_out sent: ▐▌ ▐▌ ▐▌ (clean)
0 km 40 80 120 km │
after long fiber:
loss(dB) = α(dB/km) × L(km) recv: ▟▙ ▟▙ ▟▙ (smeared,
at 1550 nm, α ≈ 0.2 dB/km overlapping → errors)
Attenuation caps the REACH (distance). Dispersion caps the BANDWIDTH (bit rate).
Fix reach with amplifiers/repeaters. Fix blur with single-mode + dispersion
compensation.The engineer's sum: a fiber link budget in dB
Here is the question every optical engineer must answer before laying a single metre of cable: *will enough light reach the far end for the photodiode to read it?* You answer it with a link budget — a running tally, in decibels, of every gain and loss between transmitter and receiver. Decibels turn the whole calculation into simple addition and subtraction, which is the entire reason engineers love them: a series of multiplying loss factors becomes a column of numbers you just add up.
FIBER LINK BUDGET — 80 km single-mode link at 1550 nm
════════════════════════════════════════════════════════
Transmitter (laser) launch power ............. +3.0 dBm (≈ 2 mW)
LOSSES:
Connector, Tx side ........................ −0.5 dB
Fiber: 80 km × 0.2 dB/km .................. −16.0 dB
2 fusion splices × 0.1 dB ................. −0.2 dB
Connector, Rx side ........................ −0.5 dB
─────────
Total loss ................................ −17.2 dB
Power at receiver = +3.0 − 17.2 ............ −14.2 dBm
Receiver sensitivity (min. for low error) .. −22.0 dBm
────────────────────────────────────────────────────────
LINK MARGIN = −14.2 − (−22.0) = +7.8 dB ✓ link works,
with headroom for ageing
& repairs- Start from the launch power the transmitter puts into the fibre, in dBm (here a +3 dBm laser, about 2 mW).
- Subtract every loss along the way: connectors, splices, and the big one — fibre length × its dB/km. Add a few dB of safety margin for components ageing and future repair splices.
- Compare the power at the receiver to the photodiode's sensitivity — the minimum power it needs to keep errors below target. If received power is above sensitivity, the difference is your link margin.
- If the margin goes negative, the link won't close. Your levers: a stronger laser, a more sensitive receiver (an APD), fewer/cleaner splices, or — past ~80–100 km — an optical amplifier or repeater to top the light back up.
Why glass beat copper
In earlier tracks you met the copper transmission line — the coax and twisted pair that carry signals as voltage waves. Copper is brilliant up close, but stretch it far and it collapses, for reasons that are now easy to name. A copper line's loss climbs steeply with frequency (the skin effect crowds current into a thinner and thinner shell), so a fast signal fades far faster than a slow one — and the faster you push it, the shorter the reach. Fibre's loss, by contrast, barely cares about the modulation rate: 0.2 dB/km whether you send 1 Gb/s or 100 Gb/s. That single fact is why glass won the long haul.
COPPER LINE OPTICAL FIBER
─────────── ─────────────
carries signal as voltage / current wave pulses of light
loss at high speed rises steeply w/ freq ~flat, ~0.2 dB/km @1550nm
reach @ 10 Gb/s a few metres → ~100 m tens → 100s of km no repeater
bandwidth MHz–low GHz over distance THz of optical window
crosstalk / EMI picks up & radiates noise immune (glass is a dielectric)
weight / size heavy, bulky bundles hair-thin, light, dense
electrical isolation conducts (ground loops) none — galvanically isolated
Copper still wins for: power delivery, very short links, and the
last cheap centimetres to a chip. Fiber wins everything long & fast.Step back and the whole rung resolves into one familiar diagram. A laser driven by current flashes bits of light (transmitter); a fiber guides them by total internal reflection, dimming them through attenuation and blurring them through dispersion (channel); a photodiode turns the surviving flickers back into current for an amplifier to read (receiver). Make the light, send the light, read the light — and balance it all with a link budget. That is photonics, and it lights the backbone of the entire internet.