Photonics — Lasers, Fiber Optics, Photodetectors — Engineering Reference

Photonics — Lasers, Fiber Optics, Photodetectors

1. At a glance

Photonics is the engineering of generating, guiding, modulating, and detecting light — the practical discipline that sits between classical optics (lenses, prisms, ray tracing) and electromagnetic field theory ([[Engineering/electromagnetics-engineering]]). Where electronics moves electrons in conductors, photonics moves photons in waveguides; where electronics is limited by RC time constants and conductor loss above tens of GHz, photonics carries hundreds of THz of optical bandwidth on a single fiber strand at 0.2 dB/km. By 2026 the practical reach of photonics covers an extraordinary range of products: datacenter optical interconnect (400 G QSFP-DD shipping in volume, 800 G ramping, 1.6 T sampling), DWDM telecom backbone (terrestrial and submarine), LiDAR and ToF 3D sensing (automotive ADAS, industrial robotics, mobile-phone face authentication), display (LCD backlight, OLED, microLED), laser materials processing (cutting, welding, marking, additive manufacturing), biomedical instrumentation (Raman, fluorescence, OCT, flow cytometry), quantum information (entangled-photon QKD, photonic qubits), and emerging on-package optical I/O for CPU and GPU compute fabrics (Intel, Marvell, Nokia, Ayar Labs, Lightmatter).

This note is the working reference for an engineer who has to specify a transceiver, design a fiber link budget, drive a laser diode, choose a photodetector for a given wavelength and bandwidth, or evaluate a silicon-photonics integration option. The device physics shared with semiconductors (LEDs, laser diodes, photodiodes, APDs) is built on [[Engineering/semiconductor-devices]]; the wave / propagation physics is the optical-frequency specialisation of [[Engineering/electromagnetics-engineering]]. The companion RF reference at [[Engineering/rf-design]] covers the electrical drive and receiver-front-end side of the same engineering problem (a 100 G coherent transceiver has more in common with an RF radio than with a 1980s LED indicator).

2. Why it matters

Over 99 % of intercontinental internet traffic moves on submarine optical fiber. Every hyperscale datacenter spine is built from optical transceivers; the electrical-only era of datacenter networking ended around 2010 with the 10 GBASE-SR transition. Silicon-photonic co-packaged optics (CPO) is moving optical I/O onto the same package as the switch ASIC and, by 2026–2028, onto CPU and GPU packages themselves — driven by the AI compute fabric’s bandwidth-per-watt requirements (NVIDIA NVLink-OS, Broadcom Tomahawk-5 CPO, Ayar Labs TeraPHY). Automotive LiDAR is shipping in millions of units per year (Hesai, Innoviz, Luminar, Aeva), with a continuing migration from 905 nm to 1550 nm for eye safety. Display engineering, additive manufacturing, biomedical optics, and quantum sensing are all expanding photonic markets that did not exist as commercial categories twenty years ago.

For the engineer this means: even non-photonics-focused electrical and mechanical roles routinely touch photonic components. Knowing the basics of a link budget, the difference between an SMF and an MMF, why 1550 nm dominates long-haul, and what an APD does well that a PIN doesn’t, is part of the modern electronics curriculum.

3. First principles

Light as electromagnetic wave

Light is a transverse electromagnetic wave satisfying Maxwell’s equations (see [[Engineering/electromagnetics-engineering]]) in the optical-frequency regime (~100–1000 THz). The wave is fully described by an E field, a perpendicular B field, and a propagation direction; the time-averaged intensity (power per area) of a plane wave in vacuum is:

I = ½ · ε₀ · c · |E|² (W/m²)

The same wave equation that produces radio also produces light — the only difference is wavelength. The historical merger of “optics” and “electromagnetism” by Maxwell (1865) is the foundation on which all of photonics rests.

Photon energy

In the quantum-mechanical view, light energy is carried in discrete packets of energy:

E_photon = h·f = h·c / λ

For wavelength in micrometres, the handy engineering formula is:

E_photon [eV] ≈ 1.24 / λ [µm]

A 1.55 µm telecom photon carries 0.80 eV; a 532 nm green-laser photon carries 2.33 eV; a 405 nm violet diode photon carries 3.06 eV (enough to break C-C bonds, hence its use in resin photopolymerisation and Blu-ray). The detection threshold for a photodiode is set by E_photon ≥ E_g of the semiconductor — Si (E_g = 1.12 eV) responds to λ ≤ 1100 nm; Ge (E_g = 0.66 eV) extends to ~1700 nm; InGaAs to ~1700 nm with telecom-band optimisation; HgCdTe to mid-IR depending on composition.

Refraction (Snell’s law)

At an interface between media of refractive indices n₁ and n₂:

n₁ · sin θ₁ = n₂ · sin θ₂

Common indices: vacuum / air ≈ 1.000; water ≈ 1.33; silica (telecom fiber) ≈ 1.45; BK7 glass ≈ 1.52; silicon ≈ 3.48 at 1550 nm; LiNbO₃ ≈ 2.21 at 1550 nm. The high index contrast of silicon (Δn ≈ 2 against an oxide cladding) is exactly what makes silicon photonics achievable: a 220 nm × 450 nm Si waveguide can confine a 1550 nm mode in cross-section smaller than the wavelength itself.

Total internal reflection

When light travels from a higher-index medium to a lower-index medium and the angle of incidence exceeds the critical angle θ_c = arcsin(n₂/n₁), it is fully reflected — the foundation of fiber-optic and waveguide confinement. For a silica core (n = 1.467) against a slightly down-doped silica cladding (n = 1.462), θ_c ≈ 86° and the numerical aperture is:

NA = √(n_core² − n_cladding²)

Standard SMF-28 has NA ≈ 0.14 — a small acceptance cone, but sufficient because the modal field is confined to a ~10 µm core.

Dispersion

Optical pulses spread in time as they propagate because different spectral components travel at different speeds. Three mechanisms:

• Material (chromatic) dispersion — refractive index n(λ) varies with wavelength. In silica, dn/dλ crosses zero near 1310 nm — the “zero-dispersion wavelength” of standard fiber. Below 1310 nm dispersion is normal; above, anomalous.
• Waveguide dispersion — even at constant material n, the modal field’s effective index depends on λ relative to the waveguide dimensions. Adding waveguide dispersion to material dispersion is what lets fiber designers shift the zero-dispersion point (NZ-DSF fiber per ITU G.655 puts zero somewhere in the C-band).
• Modal dispersion — in multi-mode fiber, different transverse modes have different group velocities. The dominant impairment in MMF; absent in SMF.
• Polarization-mode dispersion (PMD) — random birefringence makes the two polarization states travel at slightly different speeds. Statistical; specified in ps/√km. Limits long-haul 100 G+ coherent links.

Pulse broadening per kilometre in standard SMF at 1550 nm: D ≈ 17 ps/(nm·km). A 100 km link with a 0.1 nm-linewidth source spreads pulses by 170 ps — manageable at 10 Gb/s, catastrophic at 100 Gb/s without compensation (DCF or DSP).

Absorption (Beer-Lambert)

Power decays exponentially with propagation distance through an absorber:

I(x) = I₀ · exp(−α · x)

with α the absorption coefficient (m⁻¹). Modern silica fiber at 1550 nm has α ≈ 0.046 km⁻¹ → 0.2 dB/km, approaching the Rayleigh-scattering theoretical limit. Atmospheric absorption peaks (water at 1.38 µm, CO₂ at 2.7 and 4.3 µm) carve out the usable windows for free-space optical links and remote sensing.

Scattering

• Rayleigh — scattering from inhomogeneities smaller than λ; cross-section ∝ 1/λ⁴. Sets the floor for fiber loss and the blue colour of the daytime sky.
• Mie — scattering from particles of order λ; weakly wavelength-dependent. Fog, clouds, milk.
• Brillouin, Raman — inelastic scattering off acoustic phonons (Brillouin) or molecular vibrations (Raman). Brillouin sets the launch-power limit in narrow-linewidth long-haul fiber (SBS threshold ~6 dBm/MHz of linewidth). Raman is exploited in Raman amplifiers and in spectroscopy.

Coherence

• Temporal coherence — the bandwidth-time uncertainty Δν · τ_c ~ 1. A narrow-linewidth DFB laser (Δν = 1 MHz) has τ_c = 1 µs corresponding to a coherence length L_c = c · τ_c = 300 m. An LED (Δν = 10 THz) has L_c ~ 30 µm. Interferometric sensing, holography, coherent optical communication, and OCT all require long coherence length.
• Spatial coherence — single transverse mode → spatially coherent → focusable to the diffraction limit (~λ for a high-NA system). Multimode lasers, LEDs, and arc lamps are spatially incoherent — they cannot be focused to λ-scale spots.

Polarization states

Linear, circular, or elliptical polarization, fully described by the Jones vector (2-component complex), Stokes vector (4-component real), or a point on the Poincaré sphere. Polarization matters because:

• Coherent optical receivers are polarization-sensitive.
• Lithium-niobate Mach-Zehnder modulators are polarization-selective.
• Polarization-maintaining (PM) fiber is required for some sensors and interferometers.
• A randomly birefringent SMF rotates polarization unpredictably — coherent receivers compensate digitally.

Numerical aperture and étendue

A focused or imaged beam carries a conserved geometric quantity, the étendue G = n² · A · Ω (solid-angle × emitting area, with refractive index squared). Étendue is the optical analogue of phase-space volume — Liouville-like; passive optics can compress it in one dimension only by expanding it in another. A high-étendue source (an arc lamp, a Lambertian LED) can never be focused to the diffraction limit; a low-étendue source (single-mode laser) can. This is the engineering reason single-mode fiber communication uses lasers, not LEDs: coupling η is roughly étendue-of-source / étendue-of-fiber, and a Lambertian LED into 9 µm core × 0.14 NA wastes most of its emission. Conservation of étendue also sets the brightness limit of every imaging optical system and the radiance limit of every illumination engine.

4. Light sources

Incoherent: LEDs

Light-emitting diodes are forward-biased pn junctions in direct-bandgap semiconductors (GaN for blue/green/violet, GaAs/AlGaAs for red/IR, InGaN for white via phosphor down-conversion). See [[Engineering/semiconductor-devices]] for the underlying physics. Practical points for photonics:

• Spectrum: 20–50 nm FWHM. Too broad for high-speed digital fiber links but fine for indication, illumination, and short-reach POF systems.
• Modulation bandwidth: typically <100 MHz (carrier lifetime). Visible-light communication (Li-Fi) pushes white LEDs to ~10 MHz; near-IR comms LEDs to ~100 MHz.
• Spatial emission: Lambertian — diffuse, hard to couple into single-mode fiber.
• Power: from mW (indicators) to 10–100 W (lighting); high-brightness blue-pump LEDs for laser-pumped fiber lasers.

Coherent: lasers

A laser is an optical oscillator: a gain medium inside a Fabry-Perot or distributed-feedback cavity, pumped to population inversion, providing stimulated emission at a wavelength set by gain × cavity resonance. Maiman demonstrated the first ruby laser at Hughes Research Labs in 1960; Hayashi and Panish achieved the first room-temperature continuous-wave semiconductor laser at Bell Labs in 1970, the event that made fiber-optic communication commercially viable.

Solid-state lasers — Nd:YAG (1064 nm, frequency-doubled to 532 nm for green), Ti:Sapphire (broadly tunable 650–1100 nm, mode-locked for fs pulses), Er:YAG (2940 nm, water-absorbed for medical), Yb:YAG (1030 nm, high-power industrial).

Semiconductor laser diodes (LDs) — the workhorse of telecom and datacom:

• Fabry-Perot (FP): cleaved-facet cavity. Multi-longitudinal-mode; few mW; cheap. Short-reach datacom (e.g. 1310 nm FP for 2 km links).
• DFB (Distributed Feedback): integrated Bragg grating along the active region; selects a single longitudinal mode. Narrow linewidth (~1–10 MHz). Telecom standard for 10–400 G transceivers, DWDM transmitters, LiDAR.
• DBR (Distributed Bragg Reflector): grating outside the gain section; allows tuning. Coherent transmitters, gas sensing.
• VCSEL (Vertical-Cavity Surface-Emitting Laser): cavity perpendicular to the substrate. Emits round beam, easily array-able, low threshold (~mA). Dominant 850 nm short-reach datacom (10/25/100/400 GBASE-SR), 940 nm 3D sensing (face authentication), 1310 nm long-wavelength VCSEL emerging for datacom.
• EML (Externally-Modulated Laser): DFB + integrated electro-absorption modulator. The 1310/1550 nm transmitter of choice for 25–100 Gb/s/lane datacom.

Fiber lasers — gain in a rare-earth-doped fiber core (Er³⁺ for 1530–1610 nm, Yb³⁺ for 1030–1080 nm, Tm³⁺ for ~2 µm, Ho³⁺ for ~2.1 µm). Single-mode output, high beam quality, kilowatt-scale CW power. IPG Photonics dominates industrial. Used for cutting, welding, marking, and as pumps for other systems.

CO₂ laser — 10.6 µm gas laser. Mature industrial cutting (sheet metal, acrylic), medical (dermatology), used to be ubiquitous before fiber-laser displacement in the 2010s. Still preferred for non-metal absorption (plastics, wood, organic tissue).

Gas lasers — HeNe (632.8 nm) for alignment and metrology; ArF excimer (193 nm) for semiconductor photolithography; KrF (248 nm) and XeCl (308 nm) for industrial UV; argon-ion (488/514 nm) for legacy biomedical (now displaced by DPSS).

Quantum cascade laser (QCL) — intersubband transitions in InGaAs/AlInAs superlattices. Mid-IR emission 3–12 µm tunable, room-temperature CW operation. The enabling source for trace-gas spectroscopy (methane leak detection, breath analysis, atmospheric monitoring).

OLED, microLED, supercontinuum, frequency comb — specialty sources for display, broadband instrumentation, and precision metrology. The optical frequency comb (Hänsch / Hall, 2005 Nobel) converts a single CW laser into ~10⁶ phase-coherent spectral lines spanning an octave — the ruler of optical frequency.

5. Fiber optics

Geometry and modes

A standard optical fiber is a cylindrical silica waveguide: a high-index core (germanium-doped) surrounded by a lower-index cladding (pure or fluorine-doped silica), protected by a polymer coating (acrylate, 250 µm OD).

The V-parameter determines guided modes:

V = (2π · a / λ) · NA

where a is the core radius. Single-mode when V < 2.405 (only the fundamental LP₀₁ mode propagates); multi-mode otherwise.

Fiber types

Type	Core / cladding (µm)	NA	V at 1550 nm	Use
SMF-28 (ITU G.652)	8.2 / 125	0.14	2.3	Standard single-mode telecom/datacom
NZ-DSF (ITU G.655)	8 / 125	~0.14	2.3	Shifted zero-dispersion in C-band; long-haul DWDM
Bend-insensitive (ITU G.657.A1/A2/B3)	8.5 / 125	0.14	2.3	FTTH / building runs; survives 7.5 mm bend radius
OM3 multi-mode	50 / 125	0.20	~30	850 nm datacom to 300 m at 10 Gb/s
OM4 multi-mode	50 / 125	0.20	~30	850 nm datacom to 400 m at 10 Gb/s
OM5 wideband	50 / 125	0.20	~30	850–953 nm SWDM, 400 Gb/s short reach
Polarization-maintaining (panda, bow-tie)	6–10 / 125	0.13–0.18	varies	Interferometric sensing, coherent test, fiber lasers
Photonic-crystal fiber (PCF)	air-hole array	engineerable	varies	Nonlinear optics, high-power delivery, supercontinuum
Hollow-core (HCF, NANF)	air core	low	varies	Ultra-low-latency (≥99 % c), high-power CO₂ delivery
Specialty multimode (HCS, POF)	200–1000 / various	0.37+	very high	Industrial sensing, automotive (MOST), short-reach POF

Corning SMF-28 Ultra is the de-facto reference single-mode fiber; loss spec ≤ 0.18 dB/km at 1550 nm, ≤ 0.32 dB/km at 1310 nm, mode-field diameter 10.4 µm at 1550 nm.

Optical telecom bands (ITU-T)

Band	Wavelength (nm)	Use
O (Original)	1260–1360	1310 nm zero-dispersion; LR/ER/ZR datacom
E (Extended)	1360–1460	Avoided historically due to water-peak loss; modern “ULL” fiber re-opens it
S (Short)	1460–1530	Pumped amplifiers; some CWDM channels
C (Conventional)	1530–1565	EDFA gain peak; DWDM workhorse
L (Long)	1565–1625	Extended-band EDFA / Raman; second DWDM band
U (Ultra-long)	1625–1675	Out-of-band monitoring, future use

The C-band sits on the silica loss minimum (~0.2 dB/km) and the EDFA gain peak — there is no accident in its dominance.

Fiber loss

Modern SMF approaches theoretical limits:

• 0.18–0.20 dB/km at 1550 nm (Rayleigh-scattering limited)
• 0.32 dB/km at 1310 nm
• 0.50 dB/km at 1310 nm in legacy OS1 plants
• ~3 dB/km in OM3/OM4 at 850 nm

Loss mechanisms ranked: Rayleigh scattering (intrinsic, ∝ 1/λ⁴), OH-radical absorption near 1383 nm (mitigated in low-water-peak fiber), waveguide imperfection, microbending, macrobending. A 1 cm bend in standard SMF-28 adds ~0.5 dB; G.657.B3 fiber tolerates a 7.5 mm radius wrap-around.

Splices and connectors

• Fusion splice (Fujikura 90S+, AFL FuseConnect): 0.02–0.10 dB typical, 0.05 dB design value.
• Mechanical splice: 0.1–0.3 dB; field repair.
• LC connector (small form factor, latch): datacom default; 0.2 dB typical, 0.5 dB max.
• SC connector (push-pull, larger): legacy datacom, FTTH.
• FC, ST: legacy; FC remains in PM and lab applications because of its threaded retention.
• MPO/MTP (12 or 24 fiber ribbon): parallel-fiber datacom (40/100/400 GBASE-SR4/SR8/SR16).
• APC vs PC polish: 8° angle-polished connector (APC, green boot) has return loss > 60 dB versus 50 dB for PC. APC is required for FTTH analog video and coherent receiver protection.

6. Detectors

Type	λ range	Bandwidth	Gain	Noise	Use
Si PIN	400–1100 nm	DC–GHz	1	shot/Johnson	Visible & near-IR sensing, 850 nm datacom
InGaAs PIN	900–1700 nm	DC–50 GHz	1	shot/Johnson	Telecom / coherent datacom
Ge PIN	800–1600 nm	DC–GHz	1	high dark current	Legacy 1310 nm
Si APD	400–1100 nm	DC–GHz	10–500	excess multiplication noise	Low-light Si detection, LiDAR 905 nm
InGaAs APD	1100–1700 nm	DC–10 GHz	5–20	excess noise	10 G PON, long-reach telecom
Si SPAD (Geiger)	400–1000 nm	photon counting	10⁶+	dark counts	Quantum, dToF LiDAR, low-light imaging
InGaAs SPAD	1000–1700 nm	photon counting	10⁶	high dark counts (cryocooled)	QKD, eye-safe LiDAR
Photomultiplier tube (PMT)	200–900 nm	DC–GHz	10⁵–10⁷	low	Scientific, scintillator readout
Si CCD / CMOS image sensor	400–1000 nm	frame rate	1	read + dark	Imaging (Sony IMX, OmniVision, onsemi AR)
InGaAs FPA	900–1700 nm	frame rate	1	dark	SWIR imaging (Sensors Unlimited, Xenics)
Microbolometer	8–14 µm	60 Hz	thermal	NETD	LWIR thermal imaging (FLIR, Lynred, Seek)
HgCdTe (MCT)	1–25 µm tunable	varies	1 or APD	cryocooled	MWIR/LWIR scientific, astronomy

Key figures of merit

• Responsivity R (A/W) — output photocurrent per incident optical power. For an ideal detector at λ = 1550 nm, R = η · q · λ / (h · c) = η · 1.25 A/W with quantum efficiency η. Real InGaAs PINs achieve R = 0.9–1.05 A/W at 1550 nm.
• Dark current I_d — current with no light. Si PIN <1 nA; InGaAs PIN typically 1–10 nA at room temperature; APDs much higher and temperature-sensitive.
• NEP (Noise-Equivalent Power) — optical power giving SNR = 1 in 1 Hz bandwidth. W/√Hz.
• D* (specific detectivity) — area-normalised inverse NEP. cm·√Hz / W. Higher is better.
• Bandwidth — set by RC of the diode capacitance (linearly ∝ 1/area), carrier transit time, and the front-end amplifier (TIA bandwidth and feedback).

APD multiplication and excess noise

An APD multiplies primary photocurrent by avalanche gain M = 10–500 through impact ionisation in the depletion region. The excess noise factor:

F(M) = k · M + (2 − 1/M) · (1 − k)

where k is the ionisation-ratio coefficient (0.02 for Si, 0.3–0.5 for InGaAs). Si APDs at M = 100 have F ≈ 4; InGaAs APDs at M = 10 have F ≈ 5. APDs are useful when the front-end amplifier (TIA) noise dominates — typical for receivers below ~50 µW input.

CCD vs CMOS

In the 1990s–2000s CCD imagers dominated for low-noise / scientific use, CMOS for low-cost / consumer. By 2026 the line has erased — Sony’s stacked back-side-illuminated CMOS (IMX661, IMX904, IMX900) achieves CCD-equivalent read noise (<2 e⁻ RMS) at 100 Hz global-shutter operation. CCDs survive only in scientific / astronomy applications where extreme uniformity is required (e.g. Vera Rubin Observatory’s 3.2 Gpixel LSSTcam still uses CCDs).

Rolling-shutter CMOS suffers geometric distortion of fast-moving objects; global-shutter CMOS solves this at a cost in read noise. ToF/LiDAR sensors are increasingly global-shutter SPAD arrays (Sony IMX556, OmniVision OS08D10, ams OSRAM TMF8828).

7. Components and modulators

Modulators

• Mach-Zehnder modulator (MZM) — interferometer with phase shift on one arm. LiNbO₃ (Sumitomo, Fujitsu, EOSPACE) for high-speed coherent 100/400 G; thin-film LiNbO₃ (HyperLight, Nokia) reaches >100 GHz at lower V_π. Silicon-photonic MZM uses carrier-depletion phase shifters (slower, higher chirp, but CMOS-integrable).
• Electro-absorption modulator (EAM) — Franz-Keldysh or quantum-confined Stark effect in InGaAsP/InP. Compact, low-V_π, fab-co-integrated with the laser (EML). The dominant 25–100 Gb/s/lane datacom modulator.
• Acousto-optic modulator (AOM) — RF-driven traveling acoustic wave in TeO₂ or quartz diffracts light by Bragg scattering. Used for frequency shifting (~100 MHz–1 GHz), pulse picking from mode-locked lasers, and Q-switching.
• Silicon-photonic ring modulator — micro-ring (5–20 µm radius) tuned thermally and modulated by carrier injection or depletion. Compact, low-power, but temperature-sensitive (~80 pm/K).

Couplers, splitters, and passive components

• PLC (Planar Lightwave Circuit) splitter — silica-on-silicon 1:N or 2:N power splitter. The workhorse of FTTH (1:32 and 1:64 splitters in OLT-to-ONT distribution).
• FBT (Fused Biconical Taper) — fiber-fused 1:2 or 2:2 coupler. Low-cost, narrow-band.
• WDM mux/demux — TFF (Thin-Film Filter) for CWDM (20 nm grid), AWG (Arrayed Waveguide Grating) for DWDM (50/100 GHz grid).
• Isolator — Faraday-rotation + polariser; one-way light path; protects DFB lasers from back-reflection. Typically >40 dB isolation.
• Circulator — three-port non-reciprocal element; used in OTDR, bidirectional links, OADMs.
• Optical switch — MEMS (Calient, Polatis 384×384), electro-optic (LiNbO₃), thermo-optic (silicon photonic).

Filters

• Fiber Bragg Grating (FBG) — periodic refractive-index modulation along a fiber, reflecting a narrow band around λ_B = 2 · n_eff · Λ (Λ = grating period). Bandwidths 0.05–10 nm; insertion loss < 1 dB. Used as WDM add-drop, dispersion compensators (chirped FBG), and the sensing element of fiber-Bragg strain / temperature sensors.
• Arrayed Waveguide Grating (AWG) — planar-circuit demultiplexer: an input slab waveguide expands the field into an array of length-graded waveguides whose differential phase steers each wavelength to a different output port. Standard 40/80/96-channel DWDM mux/demux at 100/50/25 GHz spacing.
• Thin-Film Filter (TFF) — alternating-dielectric-stack Fabry-Perot etalon laminated between fibers. Cheap CWDM mux (8-skip-0 or 18-channel grids).
• Fabry-Perot etalon — two parallel mirrors. Used as a frequency reference (ultra-narrow notch) and as the wavelength selector in fixed-wavelength laser external cavities.

Amplifiers

• EDFA (Erbium-Doped Fiber Amplifier) — Er³⁺ in silica fiber, pumped at 980 nm or 1480 nm. C-band gain 30–50 dB, noise figure 4–6 dB; the enabling technology of long-haul DWDM. Extensions: L-band EDFA (1565–1625 nm), S-band TDFA.
• Raman amplifier — pump at λ_pump produces gain peak ~13 THz Stokes-shifted (e.g. 1450 nm pump → 1550 nm gain). Distributed along the transmission fiber; low effective noise figure. Often used as a co-amplifier with EDFA in unrepeatered submarine spans.
• SOA (Semiconductor Optical Amplifier) — InP-based; small, integrable, fast (~100 ps gain dynamics → useful for switching but bad for analog amplification due to crosstalk). Used in PON OLT, datacenter integration.

8. Practical math and worked examples

Photon energy and detector cutoff

E_photon [eV] = 1.24 / λ [µm]. A Si photodiode with E_g = 1.12 eV stops responding at λ = 1.24/1.12 = 1.107 µm — matching the empirical 1100 nm cutoff.

Free-space diffraction

A focused beam reaches a minimum waist:

w₀ ≈ λ · f / (π · D)

for input beam diameter D and focal length f. A 1.55 µm telecom laser through a 6 mm collimator and 25 mm lens focuses to w₀ ≈ 2 µm — comparable to a single-mode-fiber core, which is why fiber-to-fiber coupling works at all.

Worked example A — 100 G QSFP28-LR4 link budget

Specification: 100 GBASE-LR4 (IEEE 802.3ba, 4 lanes of 25 Gb/s at 1295/1300/1305/1310 nm, 10 km target reach over SMF).

Per-lane optical budget:

Quantity	Value
TX power per lane (min)	−4.3 dBm
TX power per lane (max)	+4.5 dBm
RX sensitivity (min)	−10.6 dBm
RX max input (overload)	+4.5 dBm
Link budget	6.3 dB (worst-case TX min to RX min)

Allocate this across the link:

• 10 km × 0.32 dB/km @ 1310 nm = 3.2 dB
• 4 connectors × 0.5 dB = 2.0 dB (two patch-cord ends and two enclosure adapter pairs)
• 2 fusion splices × 0.1 dB = 0.2 dB
• Total loss = 5.4 dB; margin = 6.3 − 5.4 = 0.9 dB

That is the spec margin. Real installations add CD penalty (~0.5 dB at 10 km / 25 Gb/s NRZ), connector wear, and aging margin → use −2.5 dBm typical TX and design to a ~3 dB margin for production.

Worked example B — Driving a DFB laser diode

A 1550 nm DFB single-mode LD has threshold I_th = 12 mA at 25 °C and slope efficiency η = 0.20 mW/mA (typical of a 10 Gb/s telecom DFB). Target P_out = 4 mW into fiber pigtail.

Bias current: I_op = I_th + P_out / η = 12 mA + 4 mW / 0.20 mW/mA = 12 + 20 = 32 mA.

Driver design:

• DC bias source: high-side current source (cascode PNP or BJT-+-FET) referenced to a stable voltage; long-term stability requirement ~±100 µA → need a precision reference and a feedback loop (auto-power-control APC).
• Temperature: I_th rises ~1 %/°C in InGaAsP; without TEC stabilisation, P_out drifts by tens of percent over the 0–70 °C industrial range. APC loop adjusts I_bias to hold a fixed photocurrent on the back-facet monitor PD.
• High-speed modulation: bias-tee combines DC current with RF input from the SerDes driver (typically 200–600 mV_pp at 50 Ω); modulation depth chosen to avoid extinction-ratio penalty (typical 4–6 dB ER for NRZ, higher for PAM4).
• TEC (thermo-electric cooler) — required for DFB DWDM transmitters; the 0.1 nm/°C wavelength TC means ±0.01 nm ITU-grid stability needs ±0.1 °C die-temperature control.

Eye safety classification: a 4 mW 1550 nm DFB into a single-mode fiber emits Class 1M (eye-safe to the unaided eye but a magnifier-collected beam is hazardous). Cleaving the fiber and looking into it under a microscope must be avoided.

Worked example C — Receiver sensitivity from photodiode and TIA

A 10 Gb/s direct-detect receiver uses an InGaAs PIN with R = 1.0 A/W, C_d = 100 fF, dark current I_d = 5 nA at room temperature, fronting a TIA with transimpedance R_F = 500 Ω and input-referred current noise i_n = 12 pA/√Hz. Required electrical bandwidth ≈ 0.7 × bit rate = 7 GHz.

• Total integrated noise current: I_n,rms = i_n · √B = 12 pA/√Hz · √(7 × 10⁹) = 1.0 µA RMS (TIA contribution).
• Shot noise at signal P: I_n,shot = √(2·q·R·P·B); at P = 1 µW, I_n,shot = √(2 · 1.6 × 10⁻¹⁹ · 1.0 · 1 × 10⁻⁶ · 7 × 10⁹) = 1.5 nA — negligible vs TIA noise.
• For BER = 10⁻¹², Q-factor ≈ 7 → required signal current I_s = Q · I_n = 7 · 1 µA = 7 µA → required optical power P_min = 7 µA / 1.0 A/W = 7 µW = −21.5 dBm.

That is a TIA-limited (thermal-noise) regime — typical for 10 Gb/s/lane direct-detect. At 25/50 Gb/s the TIA noise rises with √B and sensitivity degrades roughly 3 dB per doubling of bit rate. APD receivers regain ~6 dB by amplifying photocurrent above the TIA noise; coherent receivers regain 15+ dB by mixing with a strong local oscillator that boosts the signal-shot-noise floor above TIA thermal noise.

Worked example D — CCD / CMOS image-sensor SNR

A 12-bit CMOS sensor (representative of a Sony IMX modulus chip used in machine vision): full-well capacity (FWC) = 30 000 e⁻; read noise σ_r = 2 e⁻ RMS at slow scan, 5 e⁻ at high gain; dark current = 0.05 e⁻/pixel/s at 25 °C.

Signal-to-noise at full-well (shot-noise limited):

SNR_max = √FWC = √30 000 = 173 ≈ 44.7 dB

Dynamic range:

DR = 20 · log₁₀(FWC / σ_r) = 20 · log₁₀(30 000 / 5) = 75.6 dB (high-gain mode) DR = 20 · log₁₀(30 000 / 2) = 83.5 dB (low-noise slow scan)

At low light (mean signal S = 100 e⁻): N_shot = √100 = 10 e⁻; N_total = √(10² + 5²) = 11.2 e⁻; SNR = 100/11.2 = 8.9 (≈ 19 dB).

For high-dynamic-range scenes use bracketed exposures or modern HDR sensors (Sony IMX490 has 120 dB DR via multi-exposure stitching in-pixel) — automotive ADAS cameras now ship 140 dB sensors that handle direct sun in one half of the frame and a tunnel exit in the other.

9. Edge cases and gotchas

Laser safety classes (IEC 60825-1). Every laser sold is classified by accessible emission level (AEL):

• Class 1 — safe under all reasonably foreseeable use, including binocular viewing. Most consumer fiber transceivers.
• Class 1M — safe unless magnifying optics are used.
• Class 2 — visible only (400–700 nm), ≤ 1 mW, protected by blink reflex.
• Class 2M — Class 2 + magnification hazard.
• Class 3R — low-risk; visible up to 5 mW. Hand-held pointer regulations.
• Class 3B — direct viewing hazardous; intra-beam exposure of skin permitted briefly. Industrial alignment.
• Class 4 — high-power; diffuse reflection can also be hazardous. Skin / fire risk. Cutting / welding / surgery. Mandatory interlocks, eyewear, beam blocks.

Always design product classification at the lowest practical class and use engineering controls (interlocks, beam dumps, beam expanders) to lower exposure. Telecom EDFAs commonly emit several hundred mW C-band (Class 3B), demanding APC (Auto Power Control) shutdown and fiber-disconnect detection.

Chromatic dispersion management. SMF at 1550 nm has D = 17 ps/(nm·km). For 10 G NRZ (transmitter linewidth ~0.1 nm) the dispersion penalty crosses 1 dB at ~80 km. Strategies:

• Use 1310 nm (zero dispersion in SMF) — limited to 10 G/lane, no DWDM.
• Use NZ-DSF (G.655) — lower dispersion in C-band but not zero.
• Use DCF (Dispersion-Compensating Fiber) — discrete modules with D < 0; trades insertion loss for compensation.
• Use coherent DSP — the modern answer above 10 G/lane. Acacia, Inphi, Ciena DSPs digitally compensate >100 000 ps/nm.

Nonlinear effects in fiber. As launch power increases past ~5 dBm/channel in DWDM:

• SPM (Self-Phase Modulation) — intensity-dependent phase, chirps the pulse, interacts with CD.
• XPM (Cross-Phase Modulation) — neighbouring channels modulate each other’s phase.
• FWM (Four-Wave Mixing) — generates intermodulation tones on the ITU grid. Equally-spaced DWDM channels are particularly vulnerable; NZ-DSF was designed with a slight dispersion offset to suppress FWM.
• SBS (Stimulated Brillouin Scattering) — power threshold ~6 dBm/MHz of source linewidth. Above threshold, light back-scatters off acoustic phonons. Mitigated by linewidth broadening (low-frequency dither on the laser bias).
• SRS (Stimulated Raman Scattering) — transfers power from short-λ to long-λ channels; also the gain mechanism of Raman amplifiers.

Connector contamination. A single 1 µm silica dust speck in a 10 µm SMF core blocks half the optical power. Field statistics: contamination causes >50 % of fiber-link installation failures. Always inspect with a fiber microscope (Viavi P5000i) before mating, and clean with one-click cassettes or proper fiber wipes + IPA. Never re-use a wipe; never blow on a connector.

Bend loss. Standard SMF-28 starts to leak appreciably at radii < 15 mm; below 7 mm it loses tens of dB. G.657.B3 fiber (Corning ClearCurve, OFS EZ-Bend) survives 5 mm radius wraps for FTTH staple installation. Macrobend (smooth large radius) is reversible; microbend (sharp small-radius pressure from a clamp, cable-tie, or tray edge) is often permanent.

Wavelength drift with temperature. DFB lasers shift ~0.1 nm/°C; ITU-grid DWDM channels are spaced 0.4–0.8 nm. Without TEC stabilisation, DWDM transmitters drift off-channel within tens of degrees. Tunable transmitters in modern ZR coherent modules wavelength-lock to wavelength references (etalons) for absolute stability.

Eye safety in handheld LiDAR. 905 nm is silicon-detectable and cheap but reaches the retina (silicon’s eye-response cutoff is ~1100 nm). Compliance is achieved via low duty cycle (short pulses) and beam divergence engineering. 1550 nm is intrinsically eye-safer (water absorption in the cornea / vitreous limits retinal energy) but requires InGaAs detectors at higher cost. The 2026 automotive LiDAR market is splitting: short-range (Hesai AT128, Innoviz One) using 905 nm at high duty-cycle control; long-range (Aeva Aeries II, Luminar Iris) using 1550 nm FMCW for >300 m range.

EDFA gain flatness, spectral hole burning, transient. EDFA gain is not flat across the C-band; gain-flattening filters (GFF) typically reduce ripple to ±0.5 dB. When a channel is added or dropped, the remaining channels experience a power transient — automatic gain control (AGC) loops in modern EDFAs respond in microseconds. Spectral hole burning at high inversion adds wavelength-selective dip in gain that’s invisible to the AGC’s average-power loop.

OSNR vs electrical SNR. Optical Signal-to-Noise Ratio is measured in a 0.1 nm bandwidth (12.5 GHz at 1550 nm). For coherent 100 G PM-QPSK, OSNR ≈ 13 dB is the BER = 10⁻³ threshold (FEC-corrected). For 400 G 16-QAM, OSNR ≈ 21 dB. Each EDFA span adds noise that degrades OSNR; the per-amplifier NF and span loss determine reach.

Polarization fading. Direct-detect intensity systems are polarization-insensitive (PD responds to total power). Coherent receivers mix the signal with a local-oscillator laser at a polarizing beam splitter, and slow polarization rotation in the fiber causes the two polarization components to drift — DSP equalisers in coherent receivers update polarization tracking at kHz–MHz rates.

Rolling-shutter artifacts. CMOS image sensors that read out row-by-row distort fast-moving subjects (propellers, rotating shafts, falling drops). Global-shutter CMOS or strobed illumination is required for ITS/ADAS or industrial vision.

Reflectometric ghosts and FP cavities. Two parallel reflective surfaces in an optical path (a fiber connector and a downstream APD facet, an EAM and a back-facet AR coating, two FBG sensors) form an unintended Fabry-Perot etalon. The transmission ripple in wavelength is ν_FSR = c/(2·n·L); for L = 1 m of fiber, FSR = 100 MHz. Coherent receivers see this as relative-intensity noise (RIN). Mitigation: APC connectors (8° angle), antireflection coatings on every facet, isolators at sensitive interfaces.

Coherence collapse. A semiconductor laser driven by even tiny back-reflection (~ −30 dB) into its facet undergoes coherence collapse — the linewidth broadens from MHz to GHz and intensity noise increases by 20–30 dB. This is why every DFB transmitter sits behind an isolator (typically integrated in the TO-can or butterfly package, ≥30 dB) and why APC connectors are mandatory near coherent transmitters.

Mode partition noise. Multi-longitudinal-mode FP lasers shuttle power between modes randomly while keeping total power constant; combined with chromatic dispersion this becomes an intensity-noise term at the receiver. The dominant reason DFBs (single mode) replaced FPs for any link beyond a few km.

Differential group delay (DGD). PMD has a mean μ = D_PMD · √L (typical 0.1 ps/√km in modern fiber, 1 ps/√km in legacy 1990s fiber) but instantaneous DGD follows a Maxwellian distribution — outliers can be 3.5× the mean. Reliability calculation must use the outage probability, not the mean. Coherent DSP equalisers handle DGD up to a few symbol periods; beyond that, FEC and ARQ pick up.

Laser diode ESD. Semiconductor laser diodes are extraordinarily ESD-sensitive — a 100 V human-body-model discharge through the facet often kills the device by catastrophic optical damage (COD). Always store LDs in conductive packaging, handle with grounded wrist straps and ionised air, and never connect to a bench power supply without a current-limited driver and slow-start ramp.

Wavelength windowing in CWDM vs DWDM. CWDM uses 20 nm channel spacing (ITU G.694.2, 1271–1611 nm, 18 channels typical) and uncooled DFB lasers (~6 nm drift over operating range). DWDM uses 50/100 GHz spacing on the C/L band and requires temperature-stabilised TX, locked to the ITU grid. The economic split is roughly: ≤ 18 channels and < 80 km → CWDM; > 18 channels or > 80 km → DWDM.

10. Specialized topics

Silicon photonics. Monolithic photonic integrated circuits in CMOS-compatible silicon. Light at 1310/1550 nm sees Si as transparent (E_g = 1.12 eV → opaque only below 1100 nm). Waveguides are sub-micron Si strips on buried oxide (SOI); passive components, modulators, and Ge-on-Si photodetectors integrate on the same die. Lasers (Si is indirect-bandgap, doesn’t lase efficiently) are externally hybrid-bonded (Intel/II-VI heterogeneous integration) or wafer-bonded InP (HP/Aurrion, now Aerora). Foundries: GlobalFoundries 45RFSOI photonics, TSMC, AIM Photonics (USA), IMEC iSiPP (Belgium), AMF (Singapore), Tower Semiconductor. Major product categories: 100/400/800 G transceivers (Acacia/Inphi, Lumentum, Cisco/Acacia), co-packaged optics for switch ASICs (Broadcom Tomahawk-5 CPO, Marvell), optical I/O for compute (Ayar Labs TeraPHY, Lightmatter Passage).

Free-space optical (FSO). Atmospheric optical links at 1550 nm: 1–10 Gb/s, line-of-sight, weather-sensitive (fog can attenuate 100+ dB/km). Niche terrestrial use; emerging satellite inter-satellite links (Starlink laser crosslinks, ~200 Gb/s).

Quantum optics. Single-photon sources (heralded SPDC, quantum dots, NV centres), single-photon detectors (SPAD, SNSPD — superconducting nanowire), entanglement-based QKD (Ekert E91, BBM92), prepare-and-measure QKD (BB84). Commercial deployments: ID Quantique, Toshiba, QuantumCTek (China). Integrated photonic QKD modules emerging for fiber-trunked quantum networks.

Integrated photonics platforms. SOI silicon photonics is one of several:

• InP — direct bandgap → on-chip lasers; mature for optical telecom.
• Si₃N₄ (silicon nitride) — ultra-low loss (< 0.1 dB/m), 400 nm–4 µm transparency; used for microcombs and ultra-narrow-linewidth lasers (LIGENTEC, Aluxa).
• LiNbO₃ (thin-film, TFLN) — high-speed (>100 GHz) modulators in CMOS-compatible form (HyperLight, Nokia).
• AlN, AlGaAs-on-insulator — emerging for nonlinear optics.

Photonic computing. Optical matrix-vector multiplication using cascaded MZIs (Lightmatter Envise) or microring weight banks (Lightelligence). Targets analog inference workloads with 100× efficiency over GPUs at the cost of bit-depth (4–8 bit). Production silicon shipped in 2024–2025. The matrix multiply maps onto a unitary network of beam-splitters and phase shifters (a Reck or Clements decomposition); precision is set by phase-shifter resolution and is fundamentally analog, so digital post-processing handles the higher-significant bits. Inference-only — training stays on GPUs.

Optical clocks and frequency combs. Stabilising a self-referenced frequency comb (octave-spanning supercontinuum + f-2f interferometer) to an optical reference (Sr or Yb optical-lattice clock, NIST F2) realises absolute frequency references at parts in 10⁻¹⁸. The 2026 redefinition of the SI second is expected to use an optical transition. Time-distribution applications: optical fiber links carrying < 10⁻¹⁵ frequency stability over hundreds of km, used by VLBI radio astronomy and chronometric geodesy.

Biophotonics. Confocal microscopy (Olympus, Zeiss, Leica), super-resolution (STED, STORM, PALM — diffraction-limit-bypassing techniques; Hell, Betzig, Moerner 2014 Nobel), optical coherence tomography (OCT — interferometric depth imaging, retina + cardiology), flow cytometry, Raman spectroscopy of cells and tissues, photoacoustic imaging. Two-photon excitation microscopy (Denk, Strickler, Webb 1990) penetrates ≥ 1 mm into scattering tissue using sub-100-fs near-IR pulses.

AR/VR waveguides. Diffractive surface-relief gratings (Magic Leap, Microsoft HoloLens 2) or volume-Bragg holograms (DigiLens, Snap Spectacles 2026) couple light from a near-eye microdisplay into a glass waveguide that totally-internal-reflects to the user’s pupil. The optical engineering is dominated by eye-box, field-of-view, and pupil expansion trade-offs.

LiDAR techniques.

• dToF (direct Time-of-Flight) — emit a pulse, measure the SPAD return time-stamps. Sub-cm range resolution. Used in Hesai AT128, Innoviz One.
• iToF (indirect Time-of-Flight) — phase-modulated CW illumination; measure phase delay on a demodulating sensor (e.g. Sony DepthSense). Lower range and resolution, very cheap. Used in iPhone face-ID front cameras, robot vacuums.
• FMCW (Frequency-Modulated Continuous-Wave) — coherent detection of a chirped laser; range from beat-frequency, velocity from Doppler. Aeva Aeries II (1550 nm), SiLC. Inherently immune to interference from other LiDARs.
• Structured light — projected pattern decoded geometrically (Microsoft Kinect, original; now mostly obsolete in favour of ToF).

11. Tools and software

Design / simulation

Tool	Vendor	Scope
OptiSystem	Optiwave	System-level optical link / DWDM simulation
VPIphotonics Design Suite	VPI	System + DSP + coherent link modelling
Lumerical FDTD	Ansys	Device-level 3D FDTD (modulators, waveguides)
Lumerical MODE	Ansys	Eigenmode / propagation in waveguides and fibers
Lumerical INTERCONNECT	Ansys	Photonic-IC schematic + circuit simulation
Synopsys OptoCompiler / OptSim	Synopsys	Silicon-photonic PDK + foundry-tape-out flow
COMSOL Wave Optics / Ray Optics	COMSOL	General-purpose multi-physics (thermal-optical)
Zemax OpticStudio	Ansys	Lens and imaging-system design (sequential + non-sequential)
CODE V	Synopsys	Lens design / aberration analysis
FRED / LightTools	Photon Engineering / Synopsys	Non-sequential stray-light / illumination
RSoft (BeamPROP, FullWAVE)	Synopsys	BPM and FDTD device design
MEEP / Tidy3D	MIT / Flexcompute	Open-source FDTD; Tidy3D is cloud-accelerated commercial

Measurement

• Optical Spectrum Analyser (OSA) — Yokogawa AQ6370D (600–1700 nm, 0.02 nm resolution), Anritsu MS9740B, Apex AP207x (sub-MHz resolution for laser linewidth).
• Optical Time-Domain Reflectometer (OTDR) — EXFO MaxTester, Anritsu MT9085, Yokogawa AQ7280, Viavi T-BERD. Locates events along the fiber (splices, bends, connector reflections).
• Optical power meter — Thorlabs PM100D + S132C, Newport 1936-R + 818-IS-1. Calibrated to NIST traceability ±2 %.
• Optical Modulation Analyser (OMA) — Keysight N4391/N4392A, R&S RTO-K137. Constellation and EVM for QPSK/16-QAM coherent.
• Wavemeter — HighFinesse WS6-200 (200 MHz absolute accuracy), Bristol 871A.
• Coherent Receiver Test Set — Keysight N4391A. Reference for transceiver characterisation.
• VOA / variable attenuator, polarisation controller, polarimeter — Thorlabs, EXFO, JDSU.

Foundry and component

• Photonic-IC foundries: AIM Photonics (USA), GlobalFoundries 45RFSOI photonics, TSMC, IMEC iSiPP (Belgium), Tower Semiconductor, AMF (Singapore), CompoundTek, LIGENTEC (SiN), SMART Photonics (InP), HHI (InP).
• Lasers: Coherent (acquired II-VI/Finisar), TRUMPF, IPG Photonics (industrial fiber), nLight, Lumentum (telecom + DPSS), Thorlabs, RPMC, Eblana, OEwaves, II-VI/Coherent integrated lasers.
• Fiber: Corning (SMF-28, ClearCurve, Vascade submarine), OFS Optics, Sumitomo Electric, Fujikura, Prysmian, YOFC.
• Connectors / patch cords: Senko, US Conec (MPO), Diamond, Sumitomo, Fujikura.
• Detectors: Hamamatsu (Si APD, MPPC SiPM, InGaAs), First Sensor (Si APD), Excelitas (SPAD, PMT), Thorlabs, Onsemi (SiPM), Sony Semiconductor Solutions (SPAD ToF), Lynred (microbolometer LWIR), Sensors Unlimited (InGaAs FPA).
• Transceiver modules: Lumentum, Coherent, Cisco/Acacia, Innolight, Eoptolink, AOI, Source Photonics, FS.com.

12. Cross-references

• [[Engineering/electromagnetics-engineering]] — Maxwell’s equations and wave propagation; photonics is the optical-frequency specialisation of that foundation.
• [[Engineering/semiconductor-devices]] — pn-junction physics behind LEDs, laser diodes, photodiodes, APDs; bandgap drives detector cutoff wavelength.
• [[Engineering/rf-design]] — companion note from the same batch; coherent optical transceivers are RF radios with an optical mixer.
• [[Engineering/pcb-design]] — photonic-electronic co-packaging and high-speed transceiver host boards.
• [[Engineering/signal-processing-dsp]] — coherent receiver DSP, chromatic-dispersion compensation, polarisation tracking, carrier-phase recovery.
• [[Engineering/op-amps]] — TIA (transimpedance amplifier) design at the photodiode output.
• [[Engineering/microcontrollers]] — embedded control of LD drivers, TECs, APC loops, fiber-network management.
• [[Engineering/digital-control]] — APC and TEC loops use the same control machinery developed there.
• [[Robotics/sensors-perception]] — LiDAR, ToF cameras, structured-light depth sensors built on photonic primitives in this note.
• [[Languages/Tier3/3d-scene]] — point clouds and depth maps that LiDAR and ToF cameras produce.
• [[Languages/Tier3/network-protocol-dsls]] — physical-layer optical Ethernet and OTN frame the upper protocol stack.

13. Citations

• Saleh, B. E. A. & Teich, M. C. (2019). Fundamentals of Photonics (3rd ed.). Wiley. The canonical engineering-graduate text covering everything in this note in depth.
• Hecht, E. (2016). Optics (5th ed.). Pearson. The standard undergraduate optics reference; geometric, wave, and modern optics.
• Agrawal, G. P. (2021). Fiber-Optic Communication Systems (5th ed.). Wiley. The definitive fiber-comms text — dispersion, nonlinearity, amplifiers, coherent.
• Agrawal, G. P. (2019). Nonlinear Fiber Optics (6th ed.). Academic Press. SPM, XPM, FWM, SBS, SRS — the standard reference for fiber nonlinearities.
• Born, M. & Wolf, E. (1999). Principles of Optics (7th ed.). Cambridge University Press. Classical optics foundation; coherence, polarisation, diffraction.
• Yariv, A. & Yeh, P. (2007). Photonics: Optical Electronics in Modern Communications (6th ed.). Oxford. Engineering-focused photonics with strong device coverage.
• Verdeyen, J. T. (1995). Laser Electronics (3rd ed.). Prentice-Hall. Laser physics and rate-equation modelling for engineers.
• Reed, G. T. & Knights, A. P. (2008). Silicon Photonics: An Introduction (2nd ed.). Wiley. The starter on the SOI platform.
• Coldren, L. A., Corzine, S. W. & Mašanović, M. L. (2012). Diode Lasers and Photonic Integrated Circuits (2nd ed.). Wiley. Definitive on LD design and PIC integration.
• Becker, P. C., Olsson, A. A. & Simpson, J. R. (1999). Erbium-Doped Fiber Amplifiers: Fundamentals and Technology. Academic Press. The EDFA reference.
• Kaminow, I. P., Li, T. & Willner, A. E. (2013). Optical Fiber Telecommunications VIA / VIB. Academic Press. Two-volume comprehensive survey at the start of the coherent era.
• Maiman, T. H. (1960). “Stimulated optical radiation in ruby.” Nature 187: 493–494. First operational laser.
• Hayashi, I., Panish, M. B., Foy, P. W. & Sumski, S. (1970). “Junction lasers which operate continuously at room temperature.” Applied Physics Letters 17(3): 109–111. First CW room-temperature semiconductor laser.
• Kao, K. C. & Hockham, G. A. (1966). “Dielectric-fibre surface waveguides for optical frequencies.” Proc. IEE 113(7): 1151–1158. Proposed low-loss silica fiber for communication; Kao 2009 Nobel.
• Schawlow, A. L. & Townes, C. H. (1958). “Infrared and optical masers.” Physical Review 112(6): 1940–1949. Theoretical foundation of the optical maser; led to the 1960 laser realisation.
• Hänsch, T. W. (2006). “Nobel Lecture: Passion for precision.” Reviews of Modern Physics 78: 1297. Optical frequency comb.
• IEC 60825-1:2014+AMD2:2021. Safety of laser products — Part 1: Equipment classification and requirements. The international laser-safety classification standard.
• IEC 60793-1, -2. Optical fibres — Measurement methods and test procedures / Product specifications. The fiber-product standard family.
• ITU-T G.652 (2016), G.655 (2009), G.657 (2016). Characteristics of single-mode optical fibres and cables. The single-mode fiber recommendations.
• IEEE 802.3 (2022). IEEE Standard for Ethernet. PHY clauses 86, 88, 121, 124, 137, 138 (10/25/40/50/100/200/400 G) and 802.3df (800 G, 2024) for datacom optical PHYs.
• TIA-526 / IEC 61280 family. Fiber-optic test procedures (FOTP). Insertion-loss, return-loss, OTDR, optical-power measurement methods.
• Lumerical, Optisystem, VPIphotonics, Synopsys OptoCompiler product documentation. Vendor references for the design-tool ecosystem used in section 11.
• Vendor datasheets and application notes: Corning (SMF-28 Ultra, ClearCurve, Vascade), Coherent/II-VI/Finisar, Lumentum, IPG Photonics, Hamamatsu (S13720 SPAD, S5972 PIN, R9880U PMT), Sony Semiconductor Solutions (IMX661, IMX556 ToF), Wolfspeed/Cree photonics divisions, Acacia/Inphi coherent DSPs, Marvell Inphi Porrima/Spica, Broadcom Tomahawk-5 CPO. Always the first stop for parametric data on specific parts.