Request an intervention

Operational notes Engineering

What does it mean to write "50G-PON" in a specification, and why that alone isn’t enough

7 min read

Bundle of optical patch cords plugged into a distribution frame, in black and white
Every patch cord goes into the same physical port: which transceiver option sits at the other end isn’t something the connector tells you.

“OLT and ONU compliant with 50G-PON.” It’s a plausible line, and it already turns up in tenders and specifications, and on its own it doesn’t guarantee the two units will link up at all. ITU-T G.9804.3, 50-Gigabit-capable passive optical networks (50G-PON): Physical media dependent (PMD) layer specification, is the recommendation that defines the physical layer: base edition from September 2021, updated by an Amendment 3 approved on 6 February 2026. Where the two texts diverge, the amendment governs, and the latest one has just retired one of the line rates specifications still ask for. “50G-PON” on its own is like writing “optical fibre” in a tender: true, and not enough to make anything work.

The standard keeps moving, and its latest move deletes a line rate.

50G-PON downstream was never in question: 49.7664 Gbit/s, wavelength 1340–1344 nm (Table 9-5). Everything else has moved, and it is worth knowing when. The base text of September 2021, in Table 9-8, said no more than one line: “49.7664 Gbit/s upstream specification is for further study” — the symmetric upstream was not specified. Amendment 1, approved on 22 February 2023, defines it, and adds a third wavelength option; Amendment 2, of 22 March 2024, adds budget classes; Amendment 3, of 6 February 2026, does the thing that matters most to anyone writing a specification right now.

It retires a line rate. Clause 9.2.1 now reads: “The 12.4416 Gbit/s upstream line rate has been deprecated”, and Table 9-6, which held its parameters, carries a single line: “This table has been deprecated”. Two upstream rates remain, 49.7664 and 24.8832 Gbit/s (Table 9-2). A specification still asking for the 12.5 Gbit/s upstream — or an offer proposing it as the economical option — is citing a rate the standard retired five months ago. That is not a matter of form: it is the line on which the price of the ONU is negotiated.

Three upstream wavelength options, not one.

Table 9-8, in the version in force, assigns the 49.7664 Gbit/s upstream three possible bands: Option 1, 12601280 nm; Option 2, 12901310 nm (wideband) or 12981302 nm (narrowband); Option 3, 12841288 nm. These are different bands, not variants of the same channel: the choice determines the spectrum the 50G upstream occupies on the same fibre where, under the coexistence principle set out in [ITU-T G.9804.1], earlier PON generations already installed can go on operating. Transmission stays bidirectional over “1-fibre WDM” (Table 9-4a), on fibre described as “[ITU-T G.652] or compatible”: a single strand of glass, several carriers. A specification that writes “50G-PON” without stating which upstream option leaves the spectrum the ONU will use unspecified — and so leaves unspecified which pre-existing generation that fibre will have to coexist with.

The transceiver has an option of its own, and the standard says so plainly.

Within each ODN class, Table 9-8d lists transceiver pairs: N1/N1b, N2/N2b, E1/E1b, E2/E2b, C+/C+b. The standard itself, in clause 3.2.11, defines “transceiver option” as “The nomenclature given to distinguish two or more optional sets of optical specifications, any of which will satisfy the physical layer requirements of the optical distribution network (ODN) class”, and adds: “there is no requirement for interoperability between multiple transceiver options.” That isn’t a footnote: it’s the definition itself. That’s why the text mandates an active check: “For the 49.7664 Gbit/s upstream line rate, the ONU shall check whether its transceiver option matches that of the OLT.” And if the check fails, the consequence is spelled out without ambiguity: “If the ONU and OLT transceiver options do not match, the link may or may not be operational.” Two products each legitimately “compliant with 50G-PON”, with different options, may fail to talk to each other — and the standard doesn’t even promise the fault will be obvious: it says “may or may not,” it may work or it may not.

The ODN class depends on how you choose to coexist with the past.

The optical budget available isn’t a single figure: it depends on the coexistence method. With OLT MPM (multi-PON module) — the card that integrates several PON generations into a single port, often the most compact solution in the exchange — Table 6-1 allows only two classes: N1, 14–29 dB, and C+, 17–32 dB. With an external coexistence element (CEx) or a direct ODN connection, Table 6-2 opens up to four classes: N1 14–29 dB, N2 16–31 dB, E1 18–33 dB, E2 20–35 dB. Choosing MPM, for the benefit of occupying a single port in the exchange, takes precisely the classes with the widest margin off the table — N2, E1 and E2 are only available with an external CEx or a direct connection. And that margin is what decides how many branches a splitter can carry and how far the ONU can reach: Table 9-4a sets the maximum fibre distance between reference points at DD20: 20 km or, where declared, DD40: 40 km. An ODN class written into a specification without the method that determines it is a promise that can be read two different ways, and the difference between the two is measured in decibels of real margin on the run.

Four items, not a name.

A specification that wants to say something verifiable has to state, for every 50G-PON run:

  • the upstream wavelength option — Option 1, 2 (wideband or narrowband) or 3 — and which pre-existing generation the fibre has to coexist with;
  • the transceiver option — N1 or N1b, C+ or C+b, and so on — because the standard itself warns that if OLT and ONU don’t match, the link “may or may not be operational”;
  • the ODN class and the coexistence method that determines it — MPM, or CEx, or a direct connection;
  • the real attenuation of the run, measured, not the class declared on paper.

The ODN — fibres, splices, splitters, distribution frames — is the infrastructure that lasts twenty years; the electronics in the exchange and on the premises get replaced far sooner. Whoever designs the ODN today decides what can be switched on in 2030, and the four items above are how that decision gets written into the specification instead of staying an intention.

The same four items are what we verify at acceptance: a conformity declaration signed by the supplier isn’t enough — the real attenuation of the run is measured with bench equipment and checked against the declared class, run by run. And every measurement — OTDR trace, splice, splitter position, distribution frame port — ends up on a single network map, the same map where the calculated optical budget for every run can also be read, on which an AI performs the first diagnosis when a link won’t come up. The mismatched-transceiver-option case is the perfect example: without the map, equipment gets swapped by trial and error, one unit at a time, until the right one turns up; with the map, you already know the ODN is sound — attenuation within threshold, class met — and that the problem lies in the declared option, and the team, ours together with CSIDIA, closes the fault without subcontracting. Where an AI is needed on network data, it runs within the client’s perimeter: on autonomous on-premise machines that require no deep integration into the existing network, or in a dedicated cloud with a data centre in Italy, always with shared management between the two teams.

Do you need to write a 50G-PON specification, or check a quote that declares only the technology’s name? Talk to an engineer: the first session comes at no cost, and the four items fit in one paragraph, written before you order the first unit.

Sources