Nonlinear Effects — why you cannot just launch more power
On the Link Engineering page you saw that launching a channel a little hotter improves OSNR. That is true — but only up to a point. Past that point the fiber itself starts to distort the signal, and pushing harder makes things worse. This page explains the physics behind that ceiling — the nonlinear (Kerr) effects — and how they combine with OSNR to fix an optimal launch power for every link.
Noise sets a floor under launch power: too little power and OSNR is too low to decode. Nonlinearity sets a ceiling over launch power: too much power and the fiber distorts the signal faster than OSNR improves. The best operating point sits between them — not at either edge. Every design has such a sweet spot, and the whole job of this page is to show you why it exists and how to find it.
Why fiber becomes "nonlinear" at high power
At low power, glass behaves the way you expect: the light passes through and the fiber's properties don't care how bright the light is. At high power that stops being true. The glass's refractive index — the number that sets how fast light travels and how it bends — starts to depend slightly on the intensity of the light passing through it. This is called the Kerr effect. It is tiny per metre, but a long-haul link is millions of metres, so the tiny effect accumulates into a real signal penalty.
Think of the refractive index as the "speed limit" the light obeys. Normally that speed limit is fixed. The Kerr effect makes the speed limit sag wherever the light is brightest. Since a real pulse is bright in its middle and dim at its edges, different parts of the pulse now see slightly different speed limits — so the pulse's own shape starts to warp itself. That self-inflicted warping is the root of every effect on this page.
SPM — self-phase modulation
SPM is the effect above acting on a single channel by itself. Because the channel's own power profile shifts its own index, the channel broadens its own spectrum — the leading and trailing edges of each pulse pick up slightly different phase, which spreads the range of wavelengths the pulse contains. On its own, a wider spectrum then interacts with chromatic dispersion, and the two together can either stretch or compress the pulse depending on the sign of the dispersion.
SPM is the effect you provoke by turning a single channel up too hot. If a lone wavelength on a lightly loaded fiber starts erroring only after you raised its launch power, SPM interacting with dispersion is a prime suspect. Backing the launch power down toward the planned value is the first move.
XPM — cross-phase modulation
XPM is the same index-shifting mechanism, but now the power of neighboring channels does the shifting. When many wavelengths share the fiber, each channel's index — and therefore its phase — is nudged by the fluctuating power of the channels beside it. The result is that channels shift each other's phase. XPM only matters when there are multiple channels, and it gets worse as channels are packed closer together and as more channels are lit.
FWM — four-wave mixing
FWM is the most distinctive nonlinear effect because it does not just distort existing channels — it creates brand-new tones. When two or three channels mix in the glass, the nonlinearity generates additional light at new wavelengths, sum-and-difference combinations of the originals. If one of those newly created tones lands on top of an existing channel, it sits there as interference that no amplifier or filter can remove.
FWM is worst when two conditions are both true: the channels are on equal spacing (so the new tones land exactly on other channels), and the fiber has low chromatic dispersion (so the mixing channels stay in phase long enough to build a strong new tone). A little chromatic dispersion helps here: it makes channels travel at slightly different speeds, so they drift out of phase and the mixing never accumulates. This is exactly why zero-dispersion fiber is a poor choice for dense WDM, and why NZDSF (non-zero dispersion-shifted fiber) deliberately keeps a small, controlled amount of dispersion — enough to suppress FWM without needing heavy dispersion compensation.
SRS and SBS — scattering effects
Two more nonlinear effects come from light scattering off vibrations in the glass. They behave differently from the Kerr effects above but share the same trigger: high power.
SRS — stimulated Raman scattering
SRS transfers power from shorter-wavelength channels to longer-wavelength ones across the band. The visible symptom is a power tilt: the blue end of the spectrum ends up weaker and the red end stronger than they started. On a densely loaded fiber this tilt must be measured and corrected so every channel still meets its per-channel power target. (The same Raman physics is deliberately harnessed for gain on the Amplification page — here it is an uninvited side effect.)
SBS — stimulated Brillouin scattering
SBS reflects power backward toward the transmitter once a per-channel power threshold is crossed, especially for narrow-linewidth sources. The symptoms are a hard cap on how much power you can usefully launch on one channel, plus back-reflected light that can disturb the source. It is one more reason there is a ceiling on launch power that pure OSNR math would not predict.
The penalty grows with power and with channel count
Every effect above shares two dependencies. First, the penalty rises with launch power — brighter light bends the index more (Kerr) and crosses scattering thresholds sooner. Second, XPM and FWM in particular rise with channel count and channel density — more neighbors, packed closer, means more phase-shifting interactions and more mixing products. A launch power that is perfectly safe on a lightly loaded fiber can become a nonlinear penalty once the fiber fills up.
This page names which way each lever moves — more power and more channels mean more nonlinearity — but it deliberately gives no dB thresholds or reach numbers, because those depend entirely on the specific fiber type, channel plan, modulation, and line system. Any real launch-power target comes from the vendor planning tool. Verify against Ekinops docs / BOM and the line-system planner before setting a live launch power.
The bathtub — where OSNR and nonlinearity meet
Now combine this page with the OSNR budget from Link Engineering. Plot the total penalty (or equivalently, the achievable quality) against launch power. Two opposing forces shape the curve into a bathtub:
Start from the OSNR-optimal power (launch as hot as the noise budget wants), then check the nonlinear penalty at that power. If the nonlinear penalty is already significant, back the power down until the sum of the two penalties is smallest — the bottom of the bathtub. In practice the planning tool does this sweep for you and reports a recommended per-channel launch power. Your job in the field is to set the channels to that planned power and equalize them, not to hunt for "more is better."
Channel-count dependence in practice
Because XPM and FWM scale with the number and density of channels, the optimal launch power for one channel drops as the fiber fills. A design validated with four channels lit can slide into a nonlinear penalty once forty channels are lit at the same per-channel power. This is why capacity growth is a design event, not just "light another wavelength" — the whole loaded system's launch plan has to still sit in the bathtub. Cross-reference the Capacity & Design page: leaving spectral and power headroom is partly about keeping the fully loaded system out of the nonlinear wall.
How coherent optics and DBP change the ceiling
Coherent transponders with strong DSP (the subject of the Coherent Optics page) tolerate more nonlinearity than older direct-detect systems, because the receiver can electronically compensate a great deal of the linear distortion and some of the nonlinear distortion. One specific technique, DBP — digital back-propagation, runs a numerical model of the fiber in reverse inside the receiver DSP to partly undo the deterministic part of the nonlinear distortion the signal picked up.
DBP and coherent DSP push the right-hand wall of the bathtub outward, so you can launch somewhat hotter and reach somewhat further before nonlinearity dominates. But a limit always remains: DBP can only undo the deterministic, predictable part of the nonlinearity, not the random interaction with noise, and it costs processing complexity. The bathtub still exists; coherent just widens the good part of it. There is no launch power at which nonlinearity simply goes away.
1. A single channel on a lightly loaded fiber starts erroring right after you raised its launch power to "get more margin." What is the likely cause and the fix?
2. Why does a little chromatic dispersion actually help against four-wave mixing?
3. A launch plan was validated with 4 channels lit and worked fine. After growth to 40 channels at the same per-channel power, links degrade. Why?
4. In one sentence, what sets the lower bound on launch power and what sets the upper bound?
5. Does coherent optics with DBP eliminate the nonlinear limit?