The two polarisation states of a signal travelling at slightly different speeds — a statistical, time-varying impairment.
Two Polarisations Travelling at Different Speeds
A single-mode fibre carries one spatial mode but two polarisation states. In perfectly circular, unstressed glass they propagate identically. Real fibre is slightly elliptical and is squeezed by cabling, bends, temperature and wind, so the two states see slightly different indices and arrive at different times. That difference is the differential group delay.
Unlike chromatic dispersion, PMD is not a fixed property of the route. It drifts through the day and the season, which is precisely why no passive component can compensate it the way a dispersion module compensates CD.
Why It Goes as the Square Root
Birefringence in a real cable is randomly oriented and changes along the length, so the delays add as a random walk rather than linearly. Doubling the route length multiplies accumulated DGD by about 1.41, not 2, which is why the coefficient is quoted in ps/√km.
The instantaneous DGD follows a Maxwellian distribution around that mean, with a tail reaching about three times it. Systems are therefore designed to an outage probability rather than a hard limit: the conventional target is on the order of two seconds of outage per year, which is where the rule of keeping mean DGD near 10% of the symbol period comes from.
The Numbers That Decide It
Modern cabled G.652.D carries a link design value of 0.20 ps/√km or better, so 400 km accumulates about 4 ps. PMD is essentially never the limit on new plant. Fibre drawn before spun-preform manufacturing became normal can run 0.5 to 2 ps/√km, and that is a different situation entirely.
At 10 Gbit/s the mean DGD budget is roughly 10 ps. On 0.2 ps/√km fibre that is 2500 km; on 2 ps/√km fibre it is 25 km. A serial 40 Gbit/s signal needs about 2.5 ps and shrinks those distances sixteen-fold. Same cable, entirely different network.
Why Coherent Made It Go Away
A coherent receiver recovers both polarisations explicitly, so the DSP can track the state of polarisation and unwind DGD continuously as it changes. A 400ZR module is specified to hold 10 ps of mean DGD and 33 ps instantaneous with under half a decibel of penalty — and to keep doing so while the fibre moves.
That is why the standard fix for a bad legacy route is not a PMD compensator but a change of transmission technology. If a route carries 10G happily and fails at higher rates with healthy receive power and no dispersion explanation, PMD is the suspect: measure it, and if the coefficient is poor, price coherent pluggables against re-cabling.
PMD reference
| Units | ps/√km, because it accumulates with the square root of length |
|---|---|
| G.652.D link design value | ≤ 0.20 ps/√km |
| Older installed plant | 0.5 to 2 ps/√km, occasionally worse |
| Design rule | mean DGD about 10% of the symbol period |
| 10 Gbit/s NRZ | roughly 10 ps mean DGD |
| 40 Gbit/s serial | roughly 2.5 ps mean DGD |
| Instantaneous DGD | Maxwellian — peaks near three times the mean |
| 400ZR tolerance | 10 ps mean, 33 ps instantaneous |
| Standard outage target | about two seconds per year |
Related Terms
Chromatic dispersion, coherent optics, OSNR, OS2 G.652.D and OTDR.