Polarization-Mode Dispersion and Birefringent Fibers

A single-mode fiber is single-mode only in its transverse shape. The waveguide-modes article establishes that a circular step-index fiber's fundamental $\mathrm{HE}_{11}$ (loosely $\mathrm{LP}_{01}$ ) guided mode is really two modes — two orthogonal linear polarizations that, by cylindrical symmetry, share one propagation constant $\beta$ and travel in lockstep. Polarization-mode dispersion is what happens when that symmetry breaks, which it always does. An elliptical core, frozen-in stress, and every bend, twist and lateral squeeze the cable suffers underground make the glass weakly birefringent, so the two polarizations acquire slightly different propagation constants $\beta_x\ne\beta_y$ and, more importantly, slightly different group velocities. This article takes the group-delay machinery of the chromatic-and-modal-dispersion article and asks what happens when a fiber carries two group velocities at once. The phase difference $\Delta\beta$ sets a beat length $L_B=2\pi/\Delta\beta$ over which the polarization state cycles; the group-velocity difference sets a differential group delay (DGD) $\Delta\tau$ between the fast and slow axes, so a pulse launched on both axes arrives as two time-shifted replicas — that smearing is PMD. In a short uniform fiber the DGD grows linearly with length, like chromatic dispersion. But in a real installed cable the birefringence axes rotate randomly along the route, so the accumulated delay performs a random walk — and a random walk grows as $\sqrt{L}$ , not $L$ . That change from deterministic to stochastic is what makes PMD a statistical, time-varying, outage-defined impairment rather than a fixed penalty a designer can simply subtract.

Key takeaways

A single-mode fiber carries two orthogonal polarization modes; ideally degenerate, but residual asymmetry (core ellipticity, stress, bends) lifts the degeneracy and gives them slightly different phase and group velocities — birefringence.
The differential group delay (DGD) between the fast and slow axes splits a pulse into two time-shifted replicas; this spreading is polarization-mode dispersion.
In real fibers the birefringence axes vary randomly, so DGD accumulates as a random walk: the mean grows as $\sqrt{L}$ (units $\mathrm{ps}/\sqrt{\mathrm{km}}$ ), unlike linear-in- $L$ chromatic dispersion.
Polarization-maintaining fiber builds in a huge, stable birefringence (Panda, bow-tie, elliptical-core) so a pulse aligned to one axis stays there over kilometers — at the cost of a large fixed DGD.
PMD is described by the principal states of polarization (PSPs): the two input states whose output polarization is frequency-stationary to first order, separated by the DGD $\Delta\tau=|\vec\Omega|$ .
Because DGD is Maxwellian and drifts with temperature, PMD sets a statistical outage limit needing adaptive, not fixed, compensation.
Key idea: residual birefringence splits the two polarization modes; the group-velocity difference is a differential group delay $\Delta\tau$ that, because the birefringence axes vary randomly along the fiber, accumulates as a random walk with mean DGD ∝ √L — this stochastic, time-varying √L scaling (versus L for chromatic dispersion) is what turns PMD into a statistical outage phenomenon rather than a deterministic penalty.

The core insight

Residual birefringence splits two nearly-degenerate polarization modes; their group-velocity difference is a differential group delay $\Delta\tau$ that, because the birefringence axes rotate randomly along the fiber, accumulates as a random walk with mean $\langle\Delta\tau\rangle\propto\sqrt{L}$ — this stochastic $\sqrt{L}$ scaling, versus the deterministic $L$ of chromatic dispersion, is what makes PMD a statistical, time-varying outage phenomenon.

Where this lives today

PMD sits at the center of modern long-haul networking. In coherent optical systems the receiver mixes the incoming light with a local oscillator and digitizes both quadratures of both polarizations; a digital signal processor then runs a $2\times2$ butterfly of adaptive finite-impulse-response filters that continuously inverts the fiber's Jones matrix. That single block equalizes chromatic dispersion and PMD together and tracks the DGD as it drifts — which is why a 400 or 800 Gb/s wavelength now runs over legacy fiber that was hopeless with the fixed optics of the 1990s. The same front end enables polarization-division multiplexing: PMD mixes the two orthogonal data streams, but the MIMO equalizer unmixes them, turning the fiber's polarization degrees of freedom into a doubling of capacity. At the other extreme, PM fiber is the enabling component wherever a stable polarization is the whole point. Fiber-optic gyroscopes wind kilometres of PM fiber into a coil so the counter-propagating beams keep a fixed polarization and the Sagnac interference stays sharp; interferometric sensors, coherent LIDAR, and high-power fiber lasers rely on it for the same reason. Commercial PMD analyzers sweep wavelength and read the Poincaré-sphere precession to extract $\vec\Omega(\omega)$ directly — a benchtop realization of the law $\mathrm{d}\hat s/\mathrm{d}\omega = \vec\Omega\times\hat s$ that defines the principal states.

The physical picture

Put two clocks in the fiber, one bolted to the fast axis and one to the slow axis. A pulse launched on both axes starts both together, but they tick at slightly different rates, so at the far end the two copies arrive a time $\Delta\tau$ apart and the received pulse — their sum — has smeared. That is first-order PMD, and unlike chromatic dispersion it does not care about spectral width; it is a purely polarization effect, present even for a perfectly monochromatic carrier.

Figure

Pulse splitting in a birefringent fiber: a pulse launched on both principal axes emerges as two replicas, the fast-axis copy leading the slow by the differential group delay Δτ. For a Panda fiber with beat length L_B ≈ 3.1 mm at λ ≈ 1550 nm the DGD is ≈ 1.7 ps per metre, so a 250 m coil gives Δτ ≈ 425 ps

Now let the axes twist. Every hundred metres or so the local fast axis points in a new, random direction, so the delay banked in one section is not simply added to the next — it can be partly cancelled or reinforced depending on the relative orientation. The total delay is a sum of many small, randomly oriented vector steps: a random walk. Random walks grow as the square root of the number of steps, which is the whole reason a long fiber's DGD scales as $\sqrt{L}$ while its chromatic dispersion, adding deterministically, scales as $L$ . The second figure is what an engineer budgets against.

Figure

Mean DGD grows as the square root of length in a strongly mode-coupled fiber. A legacy fiber with PMD coefficient D = 0.5 ps/√km reaches ⟨Δτ⟩ = 10 ps at 400 km — the 10-percent-of-a-bit budget for a 10 Gb/s (100 ps) system — while a modern spun fiber with D = 0.05 ps/√km stays near 1 ps over the same span

Polarization-maintaining fiber turns the same physics on its head: make one clock run so much faster, and pin the axes so rigidly, that no ordinary bend or twist can phase-match the modes and swap energy — a pulse launched on one axis then keeps its polarization over kilometres, at the price of a large but fixed DGD. The principal states of polarization generalize these two axes to a whole random fiber: the two special inputs that, like a waveplate's eigen-axes, emerge with a stable output polarization, while every other input is a superposition that splits and spreads.

The core equations

The starting point is the phase difference the two modes build up as they propagate. Seeing slightly different effective indices $n_x$ and $n_y$ , they have different propagation constants; we package the difference as a birefringence $\Delta n$ .

\Delta\beta = \beta_x - \beta_y = \frac{2\pi}{\lambda}\,\Delta n \qquad (1)

where: $\beta_{x,y}$ — propagation constants of the two polarization modes; $\Delta n = n_x - n_y$ — modal (effective) birefringence; $\lambda$ — free-space wavelength. Physical reading: $\Delta\beta$ is the rate at which relative phase builds between the modes — ordinary telecom fiber has $\Delta n\sim10^{-7}$ – $10^{-6}$ , PM fiber $\sim5\times10^{-4}$ . As the relative phase sweeps through $2\pi$ , the combined polarization state returns to where it started. That length is the beat length, the single most quoted number characterizing a fiber's birefringence.

L_B = \frac{2\pi}{\Delta\beta} = \frac{\lambda}{\Delta n} \qquad (2)

where: $L_B$ — length over which the polarization state cycles once; $\Delta n$ — modal birefringence. Physical reading: a short beat length means strong birefringence — PM fibers reach $L_B\sim$ a few millimetres, low-PMD telecom fiber tens of metres or more. The beat length is a phase quantity and does not tell you how a pulse spreads; pulse timing is governed by group velocity, so the relevant object is the group-delay difference between the axes — the differential group delay, proportional to length in a uniform fiber.

\Delta\tau = \left|\frac{\mathrm{d}\beta_x}{\mathrm{d}\omega} - \frac{\mathrm{d}\beta_y}{\mathrm{d}\omega}\right| L = \frac{L}{c}\,\bigl|\Delta n_g\bigr| \qquad (3)

where: $\Delta\tau$ — differential group delay (DGD) between fast and slow axes; $\Delta n_g$ — group birefringence; $L$ — fiber length; $c$ — vacuum speed of light. Physical reading: the DGD per unit length is just $\Delta n_g/c$ ; the beat length uses phase birefringence, the DGD group birefringence, and the two need not be equal. That last point deserves its own equation: the group birefringence carries an extra term because the birefringence itself changes with wavelength, so a fiber can have a short beat length yet a modest DGD.

\Delta n_g = \Delta n + \omega\frac{\mathrm{d}(\Delta n)}{\mathrm{d}\omega} = \Delta n - \lambda\frac{\mathrm{d}(\Delta n)}{\mathrm{d}\lambda} \qquad (4)

where: $\Delta n_g$ — group birefringence governing the DGD; $\Delta n$ — phase birefringence governing the beat length; the derivative is the wavelength dispersion of the birefringence. Physical reading: phase and group birefringence coincide only if $\Delta n$ is wavelength-independent; in real stress-birefringent fiber they can differ by tens of percent. Everything above is deterministic and holds for a short, uniform, or PM fiber. For a long installed cable the axes rotate randomly, the DGD becomes a random variable, and engineers specify its mean, which grows as the square root of length.

\langle\Delta\tau\rangle = D_{\mathrm{PMD}}\sqrt{L} \qquad (5)

where: $\langle\Delta\tau\rangle$ — mean DGD over random axes; $D_{\mathrm{PMD}}$ — PMD coefficient in $\mathrm{ps}/\sqrt{\mathrm{km}}$ ; $L$ — fiber length. Physical reading: legacy fiber has $D_{\mathrm{PMD}}\sim0.5$ – $2$ , modern below $0.1$ ; the $\sqrt{L}$ (not $L$ ) is the fingerprint of the random walk. The instantaneous DGD fluctuates about its mean as the environment perturbs the fiber. In the strong-coupling limit it follows a Maxwellian distribution — the same law governing a gas molecule's speed — whose long tail forces PMD budgets to be written as outage probabilities.

p(\Delta\tau) = \sqrt{\frac{2}{\pi}}\,\frac{\Delta\tau^{2}}{\alpha^{3}}\,\exp\!\left(-\frac{\Delta\tau^{2}}{2\alpha^{2}}\right) \qquad (6)

where: $p(\Delta\tau)$ — probability density of the instantaneous DGD; $\alpha$ — scale parameter, with mean $\langle\Delta\tau\rangle=2\alpha\sqrt{2/\pi}$ and mean-square $\langle\Delta\tau^2\rangle=3\alpha^2$ . Physical reading: the density vanishes at zero and has a long tail, so DGD several times the mean occurs with small but non-zero probability. Finally, the description that ties it together treats the whole fiber as one frequency-dependent object: as the optical frequency is scanned, the output polarization precesses about a fixed axis — the PMD vector $\vec\Omega$ — whose length is exactly the DGD.

\frac{\mathrm{d}\hat{s}}{\mathrm{d}\omega} = \vec{\Omega}\times\hat{s},\qquad |\vec{\Omega}| = \Delta\tau \qquad (7)

where: $\hat s$ — output Stokes vector on the Poincaré sphere; $\vec\Omega$ — PMD (polarization-dispersion) vector; $\omega$ — optical angular frequency. Physical reading: the two input states that map to $\pm\vec\Omega$ do not precess — they are the principal states of polarization, and a pulse launched on one suffers no first-order broadening.

The core derivations

Derivation ADGD as a random walk and the √L scaling

The goal is to explain the most counterintuitive fact about PMD: why the mean DGD of a long installed fiber grows only as the square root of length, against the strictly linear growth of chromatic dispersion. We model the route as a chain of short birefringent segments with independently random axis orientations and add their contributions as vectors — so that $\sqrt{L}$ appears as the variance-adds-linearly result of a random walk.

Step 1. Break the fiber into $N$ segments of length $\ell$ , each a small waveplate contributing a PMD vector $\vec\Omega_k$ of fixed magnitude $\Delta\tau_1 = \Delta\tau'\ell$ but random orientation: $\vec\Omega = \sum_{k=1}^{N}\vec\Omega_k$ . Why: within a segment the birefringence is uniform and deterministic; the randomness lives entirely in how successive segments are oriented.

Step 2. Because distinct segments have independent, isotropic orientations, their dot product averages to zero off-diagonal: $\langle\vec\Omega_j\cdot\vec\Omega_k\rangle = (\Delta\tau_1)^2\,\delta_{jk}$ . Why: an isotropic vector has zero mean, so the cross terms carry no coherent bias — this is the step that kills the linear-in- $L$ growth.

Step 3. The mean-square DGD is then $\langle|\vec\Omega|^2\rangle = \sum_{j,k}\langle\vec\Omega_j\cdot\vec\Omega_k\rangle = N(\Delta\tau_1)^2$ , so the variance grows linearly in the number of segments. Why: only the $N$ diagonal terms survive, and "variance adds linearly for independent steps" is the defining property of a random walk.

Step 4. Taking the square root with $N = L/\ell$ gives $\langle\Delta\tau\rangle \sim \Delta\tau_1\sqrt{N} = \Delta\tau'\sqrt{\ell}\,\sqrt{L}$ , so RMS DGD scales as $\sqrt{L}$ with prefactor $D_{\mathrm{PMD}}$ . Why: chromatic dispersion adds coherently (mean linear in $L$ ), PMD incoherently (variance linear in $L$ ) — coherent versus incoherent accumulation.

Go deeper · Derivation B · 1 minPrincipal states and first-order pulse splitting

Derivation BPrincipal states and first-order pulse splitting

The goal is to show that, however randomly a fiber scrambles polarization, there are exactly two input states that behave like clean waveplate axes to first order, and that any pulse splits into just these two delayed replicas — explaining the received pulse shape and the mitigation trick of launching on a single principal state. We treat the fiber as a frequency-dependent unitary Jones matrix and read off the two states and their delay.

Step 1. Write the fiber as a unitary Jones operator $T(\omega)$ and define the input principal states $|p_\pm\rangle$ by requiring their outputs $|q_\pm\rangle$ to be frequency-stationary to first order: $T(\omega)|p_\pm\rangle = e^{i\phi_\pm(\omega)}|q_\pm\rangle$ , $\mathrm{d}|q_\pm\rangle/\mathrm{d}\omega = 0$ . Why: these are the eigenstates of the frequency evolution — a waveplate's fast and slow axes generalized to the whole random fiber.

Step 2. Each principal state carries a group delay $\tau_\pm = -\,\mathrm{d}\phi_\pm/\mathrm{d}\omega$ , and their difference is the DGD, $\Delta\tau = |\tau_+ - \tau_-| = |\vec\Omega|$ . Why: this identifies the abstract PMD-vector magnitude with a physical time — the gap between the two replicas on a fast oscilloscope.

Step 3. Decompose an input as $|E_{\mathrm{in}}\rangle = \sqrt{\gamma}\,|p_+\rangle + \sqrt{1-\gamma}\,|p_-\rangle$ , so the output is two copies of the envelope $a(t)$ delayed by $\Delta\tau$ : $|E_{\mathrm{out}}(t)\rangle = \sqrt{\gamma}\,|p_+\rangle\,a(t) + \sqrt{1-\gamma}\,|p_-\rangle\,a(t-\Delta\tau)$ . Why: to first order the fiber only splits the input onto its two principal states and delays one — the two-replica picture is exact here.

Step 4. The added RMS width is $\sigma_{\mathrm{PMD}}^2 = \gamma(1-\gamma)\,\Delta\tau^2$ , maximal at an equal split $\gamma=\tfrac12$ and vanishing at $\gamma=0$ or $1$ . Why: launching entirely into one principal state gives a single group delay and no first-order broadening — the basis of PSP-launch and the reason PM fiber works.

Limiting cases

Different regimes of birefringence and length separate deterministic beating from statistical PMD.

Case	Result	Meaning
Short or uniform fiber (deterministic)	$\Delta\tau = \Delta\tau'\,L$	For $L$ below the correlation length $L_c$ the birefringence is constant and DGD grows linearly; $\Delta\tau'$ is the DGD per unit length.
Long, strongly mode-coupled fiber	$\langle\Delta\tau\rangle = D_{\mathrm{PMD}}\sqrt{L}$	For $L\gg L_c$ the DGD is random; its mean grows as $\sqrt{L}$ with PMD coefficient $D_{\mathrm{PMD}}$ in $\mathrm{ps}/\sqrt{\mathrm{km}}$ .
Polarization-maintaining fiber	$L_B\ll L_{\text{perturbation}}$	A mm-scale beat length makes intrinsic birefringence dominate perturbations; polarization is preserved but the DGD is large and fixed.
Launch on a single PSP	$\sigma_{\mathrm{PMD}}\to 0$	With $\gamma=0$ or $1$ the broadening $\gamma(1-\gamma)\Delta\tau^2$ vanishes; the pulse travels with one group delay.
Ideal isotropic fiber	$\Delta\beta\to0,\ \Delta\tau\to0$	Perfect symmetry gives degenerate modes and no PMD; unreachable, so residual PMD always exists.

Common trapTreating PMD like chromatic dispersion, as a fixed penalty that adds linearly in length

The most expensive mistake is to budget PMD the way one budgets chromatic dispersion — a deterministic number that grows as $L$ and can be cancelled once and for all. In a strongly mode-coupled fiber the DGD is a random variable with a Maxwellian distribution; its mean grows as $\sqrt{L}$ , not $L$ , and the instantaneous value drifts with temperature, vibration and handling over seconds to hours. Because the Maxwellian has a long tail, the instantaneous DGD exceeds several times its mean a small but non-zero fraction of the time, so the budget must be an outage probability, not a worst-case delay. And since the DGD wanders, a fixed compensator that cancels one realization is wrong an hour later — adaptive tracking is essential.

Worked examples

1 · Beat length and DGD of a polarization-maintaining fiber (confidence builder)

A Panda PM fiber has a measured polarization beat length $L_B = 3.1\ \mathrm{mm}$ at $\lambda_0 = 1550\ \mathrm{nm}$ and a group birefringence $\Delta n_g = 5.1\times10^{-4}$ . Find the phase birefringence, the DGD per metre, and the DGD accumulated in a 250 m sensor coil.

Step 1. Phase birefringence from the beat length (equation 2 rearranged): $\Delta n = \lambda_0/L_B = (1550\times10^{-9})/(3.1\times10^{-3}) = \mathbf{5.0\times10^{-4}}$ — about a thousand times the residual birefringence of ordinary telecom fiber, the whole point of PM fiber.

Step 2. DGD per unit length from the group birefringence (equation 3): $\Delta\tau/L = \Delta n_g/c = (5.1\times10^{-4})/(2.998\times10^{8}) = 1.70\times10^{-12}\ \mathrm{s/m} = \mathbf{1.70\ ps/m}$ . We use the group birefringence here, not the phase birefringence that set the beat length — close for this fiber but conceptually distinct.

Step 3. Over the 250 m coil: $\Delta\tau = (1.70\ \mathrm{ps/m})(250\ \mathrm{m}) = \mathbf{425\ ps}$ . Because this is a short uniform PM fiber, the DGD is deterministic and scales linearly with length — no random walk, no $\sqrt{L}$ . Δn = 5.0×10⁻⁴, DGD = 1.70 ps/m, and a 250 m coil gives Δτ = 425 ps.

2 · Maximum reach and the outage probability

A 10 Gb/s system (bit period 100 ps) keeps the mean DGD below 10 % of the bit period on legacy fiber with $D_{\mathrm{PMD}} = 0.5\ \mathrm{ps}/\sqrt{\mathrm{km}}$ . Find the reach, then — since the DGD is Maxwellian — estimate how often the instantaneous DGD exceeds three times its mean, and convert to minutes per year.

Step 1. The budget is $\langle\Delta\tau\rangle \le 0.1\times100 = 10\ \mathrm{ps}$ ; inverting equation 5, $L = (\langle\Delta\tau\rangle/D_{\mathrm{PMD}})^2 = (10/0.5)^2 = \mathbf{400\ km}$ — the marked point on the second figure. (Modern fiber at $0.05\,\mathrm{ps}/\sqrt{\mathrm{km}}$ reaches 40 000 km; at 40 Gb/s the legacy fiber falls to 25 km, since reach scales as $(1/B)^2$ .)

Step 2. Now the statistics. Work in units of $\langle\Delta\tau\rangle$ : the Maxwellian scale parameter is $\alpha = \langle\Delta\tau\rangle/(2\sqrt{2/\pi}) = 0.6267$ and the threshold is $T = 3$ . The outage probability is the Maxwellian tail $P(\Delta\tau>T) = \sqrt{2/\pi}\,(T/\alpha)\,e^{-T^2/2\alpha^2} + \operatorname{erfc}\!\bigl(T/\alpha\sqrt2\bigr)$ .

Step 3. With $T/\alpha = 4.787$ : the exponent is $-(4.787)^2/2 = -11.46$ , so $e^{-11.46} = 1.05\times10^{-5}$ ; the first term is $0.798\times4.787\times1.05\times10^{-5} = 4.0\times10^{-5}$ , the erfc term adds $\approx0.2\times10^{-5}$ , giving $P \approx 4.2\times10^{-5}$ . Over a year of $3.15\times10^{7}$ s that is $\approx 1300\ \mathrm{s}$ . Adding margin — threshold at $3.5\times$ the mean — drops the outage below $10^{-6}$ ; this steep sensitivity is exactly why PMD is budgeted probabilistically. Reach 400 km; threshold at 3× the mean gives ≈ 4.2×10⁻⁵ outage, about 22 minutes per year.

Physics of the portrait

PMD is a rare case where the theory is younger than the systems it explains: engineers fought polarization drift in the first coherent links before anyone had the language for it — the principal states of polarization.

Origins · 1 minprincipal states of polarization and how this was discovered

principal states of polarization

the two frequency-stationary input states of a random birefringent fiber · fiber-optic polarization

The principal states of polarization are the concept that made PMD tractable. Through the 1970s single-mode fiber was known to be weakly birefringent, and the output polarization of a long fiber wandered unpredictably with temperature and time; there was no clean way to say what a pulse would actually do. The breakthrough was to stop describing the fiber segment by segment and ask a global question: for a given carrier frequency, are there input polarizations whose output polarization does not change as you tune the frequency slightly? Charles Poole and Robert Wagner, at Bell Laboratories in 1986, showed that for any fiber there are exactly two such states — antipodal on the Poincaré sphere — behaving like the fast and slow axes of an equivalent waveplate for the whole random link. Their delay difference is the DGD, and every input pulse splits into just these two replicas to first order. A few years later Poole and colleagues added the statistical layer — that in a strongly mode-coupled fiber the DGD is Maxwellian and its mean grows as $\sqrt{L}$ — turning PMD from a qualitative nuisance into a budgetable, outage-defined impairment. Modern PMD measurement, compensation, and coherent-receiver equalization all rest on this two-state picture.

History

1669

— Bartholinus reports the double refraction of Iceland spar, the first birefringence

1852

— Stokes introduces the four parameters that describe partially polarized light

1892

— Poincaré maps every polarization state onto the surface of a sphere

1982

— stress-birefringent polarization-maintaining fibers demonstrated (bow-tie, Panda)

1986

— Poole and Wagner introduce the principal states of polarization for fibers

1991

— Poole and colleagues show mode-coupled DGD is Maxwellian and grows as √L

Going further

Texts. Saleh and Teich, Fundamentals of Photonics, 3rd ed., Ch. 9–10 (fiber optics; dispersion and polarization) · Agrawal, Fiber-Optic Communication Systems, 4th ed., §2.3 and the PMD sections · Hecht, Optics, 5th ed., Ch. 8 and Ch. 5 · Foot, Atomic Physics, appendix on polarized-light optics · Loudon, The Quantum Theory of Light, 3rd ed. · Demtröder, Laser Spectroscopy, Vol. 1.

Provenance. The beat-length, birefringence, DGD, mean-DGD $\sqrt{L}$ , Maxwellian, and PMD-vector precession relations are standard; a verifier should recompute $L_B=\lambda/\Delta n$ , $\Delta\tau=(L/c)\Delta n_g$ , the random-walk variance $\langle\Delta\tau^2\rangle=N(\Delta\tau_1)^2$ , the Maxwellian mean $\langle\Delta\tau\rangle=2\alpha\sqrt{2/\pi}$ with $\langle\Delta\tau^2\rangle=3\alpha^2$ , and the worked numbers ( $\Delta n=5.0\times10^{-4}$ , $1.70\ \mathrm{ps/m}$ , $425\ \mathrm{ps}$ ; 400/25/40 000 km reach; $4.2\times10^{-5}$ outage) against Saleh and Teich and the Poole–Wagner PSP literature. The quoted PMD coefficients (legacy $\sim0.5$ – $2$ , modern $<0.1\,\mathrm{ps}/\sqrt{\mathrm{km}}$ ) are order-of-magnitude industry ranges, not catalogue figures.

Prerequisites. Single-Mode and Multimode Operation · Waveguide Modes and the Wave Equation · Chromatic and Modal Dispersion in Fibers. Leads to. Dispersion Management and Compensation · Coupled-Mode Theory and Directional Couplers · Nonlinear Optical Fibers and Effects.

Ask the tutor: Ask me to walk through why a random walk of birefringent segments turns the deterministic $L$ of chromatic dispersion into the stochastic $\sqrt{L}$ of PMD, and how a coherent receiver's DSP undoes it in real time.

Is PMD the same thing as chromatic dispersion?

No, and conflating them causes real design errors. Chromatic dispersion comes from the wavelength dependence of the group velocity, so it broadens a pulse in proportion to spectral width and grows deterministically as the length $L$ . PMD comes from two polarization modes with different group velocities, so it broadens even a monochromatic carrier, and in a real cable its mean grows as $\sqrt{L}$ because the axes rotate randomly. Chromatic dispersion is fixed and cancelled once; PMD is a random variable that drifts in time — a fixed module suffices for the former, but PMD needs adaptive optics or digital equalization.

What is the difference between phase birefringence and group birefringence?

Phase birefringence $\Delta n = n_x - n_y$ measures how fast relative phase builds between the modes and sets the beat length $L_B = \lambda/\Delta n$ . Group birefringence $\Delta n_g$ measures the group-velocity difference and sets the DGD $\Delta\tau = (L/c)\Delta n_g$ — how far apart the two replicas arrive. They differ because $\Delta n_g = \Delta n - \lambda\,\mathrm{d}(\Delta n)/\mathrm{d}\lambda$ includes the wavelength dispersion of the birefringence itself, so a fiber can have a short beat length yet a modest DGD — you cannot read off one from the other.

What exactly are the principal states of polarization?

For any fiber at a given carrier frequency there are two special input states — antipodal on the Poincaré sphere — whose output polarization is stationary to first order as you tune the frequency. They are the eigenstates of the fiber's frequency evolution, the whole-link generalization of a waveplate's fast and slow axes. A pulse launched on one principal state travels with a single group delay and suffers no first-order broadening; a mixture splits into two replicas separated by the DGD. The PMD vector $\vec\Omega$ points along them with length equal to the DGD.

Why is the instantaneous DGD Maxwellian rather than just its mean?

Because the PMD vector is a sum of many small, randomly oriented contributions, one per correlation length. In the strong-coupling limit its three Cartesian components each become independent, zero-mean Gaussian variables, and the DGD is the length of that 3D vector. The magnitude of an isotropic 3D Gaussian follows the Maxwell distribution — the same statistics as a molecule's speed in a gas. So the instantaneous DGD fluctuates about its mean and has a long tail: several times the mean occurs rarely but not never, which is exactly what forces PMD budgets to be written as outage probabilities.

If modern fiber has such low PMD, why do we still care?

Two reasons. First, an enormous amount of older, higher-PMD fiber is already buried and still carries traffic; upgrading those routes runs straight into PMD, since reach scales as $(1/B)^2$ and collapses as the bit rate climbs. Second, even on good fiber the drive to higher symbol rates and tighter margins means residual PMD is no longer negligible. PMD feels solved today not because it went away but because coherent receivers with adaptive DSP equalize it in real time; remove that block and the old reach limits return.

How does polarization-maintaining fiber keep the polarization fixed?

By making its intrinsic birefringence so large that external perturbations cannot compete. Stress rods (Panda, bow-tie) or an elliptical core give a beat length of only a few millimetres, so the modes are strongly separated in propagation constant. Bends, twists and temperature changes vary on scales of centimetres to metres — far too slowly to phase-match the tightly spaced modes and swap energy between them. Light launched along one principal axis therefore stays there over kilometres. The price is a large, fixed DGD of order picoseconds per metre — fine for sensors and jumpers, but unusable as long-haul transmission fiber.