Optical Fiber Impairments Explained: CD, PMD, Nonlinearity Interview Prep

Optical Fiber Impairments — Type, Scaling, and Compensation

Sample interview questions

1. A 64 GBaud channel propagates through 1000 km of G.652 fiber (CD coefficient 17 ps/nm/km at 1550 nm). What is the accumulated chromatic dispersion, and how is it compensated in a coherent system?
   • A. Accumulated CD is irrelevant in coherent systems; only OSNR matters.
   • B. Accumulated CD = 17 ps/nm/km × 1000 km = 17,000 ps/nm. For a 64 GBaud signal with ~0.5 nm spectral width, this gives ~8500 ps of pulse broadening — orders of magnitude more than the 16 ps symbol period, so without compensation each symbol would smear across many neighbors. Coherent DSP compensates CD as a frequency-domain phase rotation: the receiver applies the inverse of the CD-induced phase response, undoing the smearing before subsequent DSP blocks. CD compensation is the first major block in the receiver pipeline (see /topics/coherent-optical-detection) because all subsequent blocks assume a CD-corrected stream. ✓
   • C. Accumulated CD reverses at the receiver automatically; no compensation needed.
   • D. CD only matters at FR2 frequencies.

   Option B is correct. Chromatic dispersion (CD) in G.652 single-mode fiber at 1550 nm is approximately 17 ps/nm/km — meaning that two spectral components 1 nm apart propagate at velocities differing by 17 ps per km. CD accumulates linearly with distance, so 1000 km gives 17,000 ps/nm. For a 64 GBaud signal with bandwidth ~64 GHz (~0.5 nm at 1550 nm), the across-band group-delay spread is 17,000 ps/nm × 0.5 nm = 8500 ps. The symbol period at 64 GBaud is 1/64×10⁹ ≈ 16 ps. So the spectral edges of the channel arrive ~530 symbols apart — each transmitted symbol smears across hundreds of received symbols. Without compensation, the channel is unusable. Coherent DSP compensates CD digitally: - The receiver has the full complex baseband (after the optical front end and ADC) - CD is a linear, deterministic, well-modeled impairment (it is a known transfer function H_CD(f) given fiber type and length) - The DSP applies the inverse filter H_CD⁻¹(f) — typically in the frequency domain via FFT — to cancel the CD-induced phase response - The output is a CD-corrected stream that subsequent DSP blocks (polarization de-mux, frequency-offset estimate, phase recovery) can process Before coherent: long-haul systems used in-line Dispersion-Compensating Fiber (DCF) — special fiber sections with opposite-sign dispersion installed at each amplifier site. DCF is bulky, lossy, and has its own nonlinear penalties; it limited reach. Coherent DSP CD compensation replaced DCF and is one of the architectural improvements that made the 100G-and-beyond coherent era possible. Option A misses that CD is among the dominant impairments. Option C is wrong — coherent DSP must actively compensate; physics does not undo CD automatically. Option D is unrelated; CD applies to all optical wavelengths in silica fiber. Production reality: coherent transceivers spec their CD-compensation range (typical 50,000 ps/nm or more) which determines the maximum unregenerated reach the transceiver can support.

2. A WDM system shows degraded BER on a specific subset of channels at the band center but not at the edges. The operator suspects nonlinear effects. Which Kerr-nonlinearity mechanism most likely causes inter-channel-localized degradation, and what is the operator response?
   • A. Self-phase modulation (SPM) — only affects each channel independently, so band-center degradation is impossible.
   • B. Four-wave mixing (FWM) — when three channels at frequencies f1, f2, f3 are present, FWM generates new spectral components at f1+f2-f3 (and permutations) that fall on existing channels. The effect is strongest when channels are equally spaced (the FWM products land exactly on other channels) and worst at the band center where many channel-triples meet the four-wave mixing condition. Operator response: introduce unequal channel spacing (no FWM products on existing channels), reduce per-channel launch power, or migrate to G.654 large-Aeff fiber which reduces nonlinear coefficient. ✓
   • C. Stimulated Brillouin scattering (SBS) — affects only band edges.
   • D. Cross-phase modulation (XPM) — affects only the slowest channel.

   Option B is correct. Four-wave mixing (FWM) is a Kerr-nonlinear effect specific to multi-channel WDM systems: when three channels at frequencies f1, f2, f3 are present in the fiber simultaneously, the Kerr nonlinearity generates new spectral components at f4 = f1 + f2 - f3 (and the permutations). For evenly-spaced WDM channels these new components land exactly on existing channel frequencies, producing channel-to-channel crosstalk. Why band-center degradation: - At the center of the band, every channel has many neighbors on both sides - Many possible f1+f2-f3 combinations fall on each center channel - The combined FWM products from all the triples accumulate as crosstalk - At the band edges, fewer channel-triples exist, so less FWM crosstalk Why dispersion matters: FWM efficiency drops sharply when the channels have different group velocities (the FWM products dephase and average out). In low-CD fiber regions (G.653 dispersion-shifted fiber, or near zero-dispersion wavelength) FWM is severe; in high-CD fiber (G.652 standard at 1550 nm) it is moderate; in G.654 large-Aeff fiber with low nonlinear coefficient it is mild. Operator responses: 1. **Unequal channel spacing**: spread channels at frequencies that don't produce FWM products on existing channels. This is the classic mitigation but limits how densely the band can be packed. 2. **Reduce per-channel launch power**: FWM scales as P³ (cubic in channel power), so a 1 dB launch reduction gives 3 dB FWM reduction. This is the operational lever in deployed systems. 3. **Upgrade fiber**: G.654 large-Aeff fiber has lower Kerr-nonlinear coefficient, reducing all nonlinear effects including FWM. See /topics/optical-fiber-standards-itu. 4. **Coherent DSP nonlinear compensation**: some advanced transceivers include nonlinear-compensation blocks that estimate and partially undo the deterministic part of nonlinear interactions (SPM and XPM are easier than FWM). Option A misreads SPM — SPM affects each channel's own phase as a function of its own intensity, producing intra-channel pulse-shape distortion, not inter-channel crosstalk. Option C is wrong — SBS is a single-channel effect, not inter-channel. Option D inverts the XPM mechanism — XPM is a phase shift on one channel proportional to another channel's intensity, which scales with channel power and walk-off, not channel speed alone. Production reality: FWM is one of the dominant nonlinear penalties in G.652-based DWDM at high per-channel launch power. The operator response is usually a combination of power reduction and channel-plan optimization rather than fiber replacement.

3. PMD (Polarization-Mode Dispersion) in older deployed fiber can be several ps/√km. Why does PMD accumulate as √L instead of L, and how does coherent DSP handle PMD compared to chromatic dispersion?
   • A. PMD and CD accumulate identically (linearly with L); the formula difference is incorrect.
   • B. PMD comes from random birefringence in real fiber — small variations in the core's circular symmetry along the fiber's length. The local birefringence axes change orientation randomly along the fiber, so the differential group delay (DGD) between the two polarization modes follows random-walk statistics: it accumulates as the square root of length, not linearly. Coherent DSP handles PMD as part of the polarization de-multiplexing stage (typically CMA-based), automatically compensating the DGD as long as it stays within the DSP's tracking range. PMD that exceeds the DSP's buffer becomes an outage event — much rarer in modern low-PMD fiber than in older deployed fiber. ✓
   • C. PMD only affects DP-QPSK systems, not higher modulations.
   • D. PMD accumulates linearly like CD; coherent DSP cannot compensate it.

   Option B is correct. PMD comes from random birefringence in real fiber — manufacturing tolerances and installation stress produce small deviations from circular core symmetry, with the local birefringence axes randomly oriented along the fiber length. The two orthogonal polarization modes travel at slightly different velocities, but the differential delay direction changes randomly along the fiber. The mathematics: at each short fiber segment, the local DGD adds vectorially in a random direction. Over a long fiber the total DGD follows random-walk statistics — the magnitude grows as √L rather than L. The standard PMD coefficient is expressed in ps/√km: a 0.1 ps/√km fiber gives 1 ps DGD at 100 km, ~3.2 ps at 1000 km. Why √L matters: - Modern fiber (post-2000 manufacturing) has very low PMD (under 0.1 ps/√km), so even 1000 km links have manageable PMD - Older deployed fiber (especially pre-1995) can have several ps/√km, accumulating to 30+ ps over 1000 km - 30+ ps PMD is comparable to a 64 GBaud symbol period and constitutes a real impairment Coherent DSP handles PMD: - The polarization-de-multiplexing block (typically CMA — Constant Modulus Algorithm — for fixed-modulus formats or DD-LMS for QAM) automatically tracks and separates the two polarization streams - The CMA equalizer can handle DGD up to its tap-count budget (typical: 32-128 taps, supporting tens of ps of DGD) - PMD that fluctuates faster than the CMA can track produces a momentary "outage" — the equalizer loses lock and reacquires - Modern coherent systems handle typical deployed PMD seamlessly; PMD outages are rare on new fiber Comparison with CD: - CD is deterministic and accumulates linearly with length; CD compensation is a one-shot frequency-domain filter - PMD is stochastic and accumulates as √L; PMD compensation is part of the adaptive equalizer chain - Coherent DSP handles both, but PMD is more challenging because it varies with time (mechanical stress, temperature) Option A is the textbook error. Option C is wrong; PMD affects all coherent modulations because it is a polarization-mixing impairment. Option D is wrong on both counts. Production reality: PMD-outage incidents on older fiber are the main reason some long-haul routes upgrade to new fiber even when PMD-tolerant coherent transceivers are deployed.

Frequently asked questions

What are the main fiber impairments in optical transport?

Optical fiber impairments fall into three categories: linear deterministic (chromatic dispersion CD), linear stochastic (polarization-mode dispersion PMD), and nonlinear (Kerr effects — SPM, XPM, FWM — plus stimulated scattering — SBS, SRS). CD and PMD spread pulses in time; nonlinear effects produce phase modulation, crosstalk, and channel-to-channel interaction. In coherent systems, CD is fully DSP-compensated (a deterministic linear filter), PMD is handled by the adaptive equalizer, and nonlinear effects are the dominant remaining impairment at high launch power. Operators tune launch power, channel spacing, and fiber choice to manage nonlinear effects within the OSNR budget.

What is chromatic dispersion?

Chromatic dispersion (CD) is the wavelength-dependent group velocity in optical fiber: different spectral components of a pulse travel at slightly different velocities, so the pulse spreads in time as it propagates. The CD coefficient for G.652 single-mode fiber at 1550 nm is approximately 17 ps/nm/km — two components 1 nm apart accumulate 17 ps of relative delay per km. CD accumulates linearly with length, so a 1000 km link has 17,000 ps/nm of total CD. For a 64 GBaud signal this spreads each symbol across many neighbors, making the channel unusable without compensation. Coherent DSP compensates CD as a frequency-domain filter (the inverse of the CD-induced phase response), enabling thousands of km of unregenerated reach.

What is PMD (Polarization-Mode Dispersion)?

PMD is the dispersion that comes from random birefringence in real fiber — manufacturing tolerances and installation stress produce small deviations from circular core symmetry with local birefringence axes randomly oriented along the fiber. The two orthogonal polarization modes travel at slightly different velocities, but the direction of the differential delay changes randomly along the fiber. PMD follows random-walk statistics: differential group delay (DGD) accumulates as the square root of length, not linearly. Modern fiber has PMD coefficient under 0.1 ps/√km; older deployed fiber can have several ps/√km. Coherent DSP handles PMD as part of polarization de-multiplexing in the adaptive equalizer, automatically tracking DGD within its budget.

What are the Kerr-nonlinear effects?

The Kerr effect in silica fiber is the intensity-dependence of the refractive index: n(I) = n₀ + n₂·I, where n₂ is the nonlinear-index coefficient. This produces several effects: Self-Phase Modulation (SPM) where a pulse modulates its own phase via its own intensity (intra-channel impairment); Cross-Phase Modulation (XPM) where one channel's intensity modulates another channel's phase (inter-channel); Four-Wave Mixing (FWM) where three channels generate a new spectral component that can fall on existing channels (inter-channel crosstalk). All scale with launch power and fiber nonlinear coefficient. Larger effective area (G.654 fiber) reduces all Kerr effects by lowering optical-power density. See /topics/optical-fiber-standards-itu.

What is FWM (Four-Wave Mixing)?

FWM is a Kerr-nonlinear effect specific to multi-channel WDM systems: when three channels at frequencies f1, f2, f3 are present in the fiber simultaneously, the nonlinearity generates new spectral components at f1+f2-f3 (and permutations). For evenly-spaced WDM channels these new components land exactly on existing channel frequencies, producing channel-to-channel crosstalk. FWM is worst at the band center where many channel-triples exist and in low-CD fiber regions where the products do not dephase. Operator mitigation: reduce per-channel launch power (FWM scales as P³), use unequal channel spacing, or migrate to G.654 large-Aeff fiber.

What is stimulated Brillouin scattering (SBS) and how is it suppressed?

SBS is a third-order nonlinear effect where a forward-propagating optical signal couples to a counter-propagating Stokes wave via acoustic-phonon interaction. Above a power threshold (typically 5-7 dBm per channel in standard SMF with narrow-linewidth lasers), most of the signal energy reflects backward, severely degrading transmission. SBS is the main reason coherent transmitters limit per-channel launch power — beyond the SBS threshold, the link fails not because of forward-propagating impairments but because the signal energy is lost to the backward Brillouin wave. Suppression: spectral spreading via FM dither or via the wide spectrum of modulated coherent signals (DP-QPSK at 32+ GBaud has a much wider spectrum than a CW carrier, so SBS threshold is much higher), large-Aeff fiber, or fiber with intentionally varying dopant profile.

What is stimulated Raman scattering (SRS) and why does it cause band tilt?

SRS is the nonlinear effect where high-power channels transfer energy to lower-frequency (longer-wavelength) channels via optical-phonon interaction in silica. The energy transfer is largest between channels separated by ~13 THz (the Raman gain peak) and accumulates over fiber length. In C+L band DWDM, the C-band channels (higher frequency, ~193 THz) can pump energy into L-band channels (lower frequency, ~188 THz). The result is "gain tilt" across the band — long-wavelength channels gain power, short-wavelength channels lose power. SRS adds to per-channel power imbalance on top of EDFA gain tilt and contributes to the channel-power-equalization challenge in cascaded amplifiers (see /topics/edfa-vs-raman-vs-soa).

How does coherent DSP compensate fiber impairments?

Coherent DSP handles fiber impairments in a defined pipeline order (see /topics/coherent-optical-detection): (1) chromatic-dispersion compensation as a frequency-domain filter (deterministic, well-modeled); (2) polarization de-multiplexing via CMA or DD-LMS that also compensates PMD within the equalizer's budget; (3) frequency-offset estimation to remove LO-laser carrier offset; (4) carrier-phase recovery to track residual LO phase noise; (5) adaptive equalization for residual ISI and some nonlinear effects. The order matters because each block assumes prior compensation. CD is fully recoverable digitally (deterministic, one-shot inverse filter); PMD is largely recoverable but only within the adaptive equalizer's tracking budget — fast or large PMD transients can still outrun the equalizer and cause momentary outages. Nonlinear effects are partially compensable — some advanced transceivers include nonlinear-compensation blocks that estimate and undo the deterministic part of SPM and XPM, but FWM is largely uncompensated and remains as residual penalty.

Related topics