3. Fiber Materials, Attenuation, and Scattering // Optical Communications // bhaswanth

Now we look at why fibers lose what little they lose, and at how engineers chose the operating wavelengths to minimize that loss.

3.1 Materials: silica and exotic alternatives

The standard material is fused silica (SiO $_2$ ), ultra-pure synthetic glass with parts-per-billion purity at the relevant wavelengths. Manufacturing it is its own hard problem: Corning's modified chemical vapor deposition (MCVD) process oxidizes silicon tetrachloride and germanium tetrachloride inside a rotating silica tube, depositing the doped glass layer by layer, then collapses and draws it into a fiber. Other vendors use OVD (outside vapor deposition) or VAD (vapor axial deposition). All produce silica with sub-parts-per-billion impurity levels.

Dopants tune the refractive index:

Germanium dioxide (GeO $_2$ ) raises index. Used in cores.
Fluorine lowers index. Used in claddings and depressed-cladding designs.
Boron lowers index. Less common.
Phosphorus raises index slightly. Sometimes used in special profiles.
Erbium, ytterbium, thulium are rare earth dopants that absorb pump light at one wavelength and re-emit at another. Used in fiber amplifiers and fiber lasers, not in passive transmission fibers.

Beyond silica, niche materials include:

Halide glasses (fluoride, chalcogenide) for mid-infrared transmission, where silica becomes opaque (above 2.5 µm). Used in mid-IR lasers and gas sensing.
Plastic optical fiber (POF), made from PMMA (polymethyl methacrylate) or perfluorinated polymers. High loss (around 200 dB/km at 650 nm), large core (1 mm), but cheap and forgiving. Used in automotive entertainment buses (MOST), home networks (10 Mbps Ethernet over POF), and short industrial links.
Photonic crystal fibers (PCF), sometimes called holey fibers. The cladding is a periodic array of air holes running along the fiber's length, making a photonic bandgap that confines light by interference rather than TIR. Used in research, supercontinuum sources, and some specialty lasers.
Hollow-core fibers propagate light through air (or vacuum) inside the fiber, surrounded by a structured cladding. The light travels at $c$ instead of $c/n$ , cutting latency by about 30%. Deployed by some financial trading firms for low-latency interconnects.

3.2 Exponential attenuation

Loss in fiber is multiplicative: each kilometer attenuates the signal by the same factor as the previous one. So power versus distance is exponential:

$P(L) = P(0) \cdot 10^{-\alpha L / 10}$

where $\alpha$ is in dB/km. After 100 km of 0.2 dB/km fiber, the signal is down 20 dB, a factor of 100 in power. After 200 km, 40 dB and a factor of $10^4$ . The exponential nature is why amplification spacing is nearly constant in well-engineered systems: every 80 to 100 km, the signal drops below the receiver's sensitivity, so an EDFA boosts it back up.

3.3 Sources of loss

Three families of loss contribute:

Intrinsic absorption, set by the glass itself.

Ultraviolet absorption from electronic transitions in silicon-oxygen bonds. Tail extends into the visible, dominant below about 700 nm.
Infrared absorption from molecular vibrations of the Si-O bonds. Climbs steeply above about 1700 nm.

Between these tails, silica has a very low intrinsic absorption window from roughly 800 nm to 1700 nm. This is the only reason silica fiber is useful at all: the same molecule that transmits poorly in the visible (you cannot see through 100 km of telephone-pole quality glass) transmits superbly in the near-IR.

Extrinsic absorption from impurities.

Hydroxyl ions (OH $^-$ ) absorb strongly at 1383 nm — the famous water peak. Modern low-water-peak fibers (G.652.D and beyond) use a special drying step to eliminate residual water in the silica, opening the E-band (1360-1460 nm) for use.
Transition metal impurities (iron, copper, chromium) used to be a problem; modern fabrication has them well below the detection threshold.

Scattering.

Rayleigh scattering from microscopic density fluctuations frozen into the glass when it solidified. Density variations on the scale of nanometers couple a small fraction of the propagating light into all other directions. The cross-section scales as $1/\lambda^4$ , the same dependence that makes the sky blue: blue photons scatter much more than red ones off air molecules. In silica, Rayleigh dominates at short wavelengths and falls rapidly with longer ones. This is the fundamental lower bound on fiber loss; you cannot do better even in a perfectly pure fiber.
Mie scattering from impurities or defects larger than a wavelength. Less wavelength-dependent than Rayleigh. Eliminated by good manufacturing.

Where does the $1/\lambda^4$ come from? A density fluctuation of size $d \ll \lambda$ acts like a tiny induced electric dipole in the field of the passing wave. The dipole oscillates at the wave's frequency $\omega$ and reradiates power proportional to $\omega^4$ — a result that goes back to Lord Rayleigh in 1871. Frequency $\omega$ is $2\pi c/\lambda$ , so $\omega^4$ becomes $1/\lambda^4$ . The dependence is an immediate consequence of Larmor's formula for the power radiated by an accelerating charge. This same $\omega^4$ explains why the sky is blue: short-wavelength photons scatter off N $_2$ and O $_2$ molecules far more than long-wavelength ones, sending blue light in every direction while letting red continue through to make sunsets.

Bending losses.

Macrobending: a continuous bend with radius small enough that the evanescent tail in the cladding exceeds a critical condition and light radiates out of the curve. For standard SMF, bend radii below 30 mm cost noticeable loss; below 10 mm causes severe loss. Bend-insensitive fibers (G.657) tolerate down to 5 mm.
Microbending: small random deformations along the fiber from cabling stress. Each tiny kink couples the propagating mode into radiation modes. Controlled by careful cable design.

3.4 The classic attenuation curve

Plotting all of these loss sources versus wavelength gives the classic fiber-attenuation curve:

The dashed curves are the individual loss mechanisms; the heavy solid curve is the sum. The three highlighted bands are the historical operating windows.

3.5 The three (and now five) wavelength windows

Three classical telecommunication windows, each chosen to dodge a specific loss feature:

First window (around 850 nm): high-ish loss (about 2.5 dB/km), but cheap GaAs-based LEDs and lasers, large multimode core (50 µm) for easy alignment. Used for short-reach datacenter optics and within-building links.
Second window (around 1310 nm): about 0.35 dB/km loss, and zero material dispersion in standard SMF, so pulses do not spread out from material dispersion at this wavelength. Used in metropolitan-area networks and FTTH.
Third window (around 1550 nm): the lowest loss in silica, around 0.20 dB/km. The narrow gain band of erbium ions falls right here (1530-1565 nm), and EDFAs work cleanly. Used in long-haul, submarine, and DWDM systems.

Modern systems also use:

E-band (1360-1460 nm): opened by low-water-peak fiber, useful for additional CWDM channels.
L-band (1565-1625 nm): extends past the C-band into longer wavelengths. EDFAs designed for L-band operation (slightly different doping) are common in modern long-haul systems.
O-band (1260-1360 nm) is the formal name for the 1310 nm window.
S-band (1460-1530 nm): between the water peak and C-band, sometimes used in CWDM grids.

Why two main long-distance windows? Because at 1310 nm standard SMF has zero material dispersion (so pulses do not spread) but moderate loss; at 1550 nm it has lowest loss (so pulses go further) but nonzero dispersion. You cannot have both in standard fiber. Engineers use 1310 nm for medium-distance metro links where you want cheap directly-modulated lasers and you do not care about EDFAs, and 1550 nm for everything long-haul where you need the lowest loss and you can pay for dispersion compensation. Dispersion-shifted fibers (G.653) tried to move the zero-dispersion point to 1550 nm; they worked, but the resulting fiber turned out to suffer from severe four-wave mixing in WDM systems, and have been largely deprecated in favor of NZDSF.