Will the fibre you already have take the step up to 100G and 400G?

“We have a run carrying 10 Gig today: can we put 100 or 400 on it without laying anything new?” The right answer today is not the one from ten years ago, and it goes wrong in two directions: “with coherent optics there is nothing left to measure” is as false as “dispersion always has to be compensated”.

Chromatic dispersion is now compensated by the receiver.

With direct detection, chromatic dispersion (CD) was a physical limit: the spectral components of the pulse travel at different speeds, the pulse broadens, and distance called for compensating fibre or dedicated modules. With coherent reception and the DSP on board the transceiver, CD becomes a linear distortion that is inverted in the electrical domain.

How much? Implementation agreement OIF‑400ZR‑03.0 of 8 October 2024 sets, for application code 0x01 (400ZR on the 100 GHz grid, amplified link), a tolerance of 2400 ps/nm, with a reference reach of 80 km minimum and G.652 fibre quoted “for link budgeting purposes only”. At the typical link value in ITU‑T G.652 — D1550 = 17 ps/(nm × km), Table I.1 — that is roughly 140 km of standard fibre.

It is not free, and this is the part always dropped: the same document assigns 0.5 dB of OSNR penalty to that compensation. CD does not disappear: it is paid for in the optical budget.

Inside the data centre the question is different.

The reasoning does not carry into the data centre: there 400G runs on direct detection over short distances, with no coherent DSP, and what decides is fibre type, parallel fibres and the optical budget — as with 800G.

PMD is the real question.

The two orthogonal polarisations travel at slightly different speeds and arrive out of step: the difference is the differential group delay (DGD). It is not a fixed number for the run: it varies randomly with time and wavelength, following a Maxwell distribution, and cannot be cancelled once and for all as CD can.

Hence the most abused attribute in specifications. ITU‑T G.652, edition 08/2024, states in clause 7.2 that cable PMD “shall be specified on a statistical basis, not on an individual fibre basis”. What the manufacturer declares is not the PMD of your fibre: it is PMDQ, the link design value, the statistical upper bound for the PMD coefficient of M concatenated cables, exceeded with probability Q. Tables 1 and 2 set M = 20 cables, Q = 0.01% and a maximum PMDQ of 0.20 ps/√km, the same for G.652.B and G.652.D.

In plain terms: 0.20 ps/√km is not a promise about your drum, but a property of the population of cables it came from, with a stated one-in-ten-thousand probability that a link of twenty sections will exceed it. On installed plant — splices, bends, closures — that number is measured, not inherited.

The age of the fibre matters, and the recommendation says so.

The G.652 change log records that PMD requirements were extended to all categories only with the 2003 edition, when two categories got the reduced limit compared with 0.5 ps/√km: in the 10/2000 edition the clause still opened with “Not all tables include requirements on PMD”. And it warns that products comply with the edition in force when they were made, not with later ones.

Fibre made before 2003 may therefore come from production with no PMD obligation. Table I.2 puts it bluntly: with no PMD specification the supported bit rates stop at 2.5 Gbit/s; at 0.20 ps/√km the table reaches 3000 km at 10 Gbit/s (implied maximum DGD 19.0 ps) and 80 km at 40 Gbit/s (7.0 ps). Beyond 40 Gbit/s the table stops: the reference becomes the transceiver’s tolerance.

The number that separates a reusable run from one that is not.

The sum runs backwards, from the receiver. For application code 0x01, OIF‑400ZR declares 10 ps of average PMD tolerated with under 0.5 dB of penalty, a maximum instantaneous DGD of 28 ps on the channel, and a maximum-to-mean DGD ratio taken as 3.3 (probability 4.1 × 10⁻⁶); it also needs a delivered OSNR of at least 26 dB per 12.5 GHz.

Over 100 km, 10 ps of average PMD is a coefficient of 1 ps/√km: that is the order of the threshold. A G.652.D-compliant cable, over the same length, sits at 2 ps — five times inside. An older, never-characterised run may be inside or outside, and the drawings will not tell you.

Over short distances everything changes. For application code 0x02 — single wavelength, unamplified — the tables drop to 1200 ps/nm of CD and 16 ps of DGD, and the constraint becomes an 11 dB budget minus link impairments: in the metro the optical budget decides, not dispersion.

What to measure before deciding.

ITU‑T G.650.3 (08/2017) calls this fibre characterization and defines it for exactly our case: verifying the attributes of older links to be used at 10 Gbit/s or above. The set covers end-face inspection, insertion loss, return loss, OTDR, chromatic dispersion, PMD and spectral attenuation. With a warning that justifies the quotation: once fibres are in service, they are often no longer accessible for assessment.

Two details from the same source. CD and PMD are measured in one direction only, unlike attenuation, which has to be read both ways. Across concatenated sections PMD values add in quadrature, whereas CD values add algebraically, signs included.

Mind which standard is cited. ITU‑T G.650.2, in force as edition 08/2015 under the title Definitions and test methods for statistical and non-linear related attributes of single-mode fibre and cable, defines the statistical attributes but states methods “suitable mainly for factory measurements”: on installed plant the reference is IEC 61280‑4‑4:2017, edition 2.0 of 7 March 2017.

What to write in the specification.

CD and PMD as separate line items, each priced: if they are “included in the certification”, nobody performs them.
A declared PMD measurement method to IEC 61280‑4‑4:2017, stated in the report with the test conditions: the choice depends partly on whether the cable can move during the test.
Wavelengths matching the intended use: if the run will carry multiplexed channels, the whole band must be covered.
Measurement per fibre, not per sample, with identifiers matching the as-built.
Comparison with the tolerance of the intended transceiver, quoted by application code: the only acceptance criterion that means anything.

The point.

Reusing an existing run is nearly always the right call: coherent optics have moved the bottleneck from chromatic dispersion to OSNR and PMD. But “easier” does not mean “unverified”: characterisation costs a measurement campaign, relaying costs an excavation. That is why in the certifications we carry out CD and PMD are line items of their own; traces and native data stay wherever suits you, on-premise or on dedicated cloud with a VPN and data centres in Italy staffed by us.

Need to decide whether an existing run will carry 100G or 400G? Talk to an engineer: agreeing up front what gets measured, and against which tolerance, costs less than rebuilding a backbone.