Encircled flux: the launch condition the data centre specification doesn’t write

A data centre specification often includes a line like this: maximum attenuation 3.5 dB/km at 850 nm. It looks like a closed number, verifiable with any instrument. It isn’t. Measure the same multimode run with two different sources — one that floods the entire fibre core, one that injects a narrow, centred beam — and you get two different values on the same link, on the same day. The difference isn’t measurement error: it’s the launch condition, the setting that decides which modes of the fibre get excited and how much light is lost. A specification that writes the number and stays silent on the launch condition has written a threshold that two honest laboratories can legitimately confirm and contradict.

The number isn’t enough: what changes with the source

A multimode fibre carries light along dozens of possible paths, each with its own attenuation: how many get excited at the input depends on how the light goes in. For decades the reference source was the LED, which floods the whole core: the overfilled launch, a severe condition that’s reproducible across different instruments. The transceivers that now equip data-hall switches — the ones we wrote about with OM4 and OM5 heading to 800G, in the same cabinets of the data centres we serve — no longer use LEDs but VCSEL lasers, which excite a different, more concentrated subset of modes: measuring with an overfilled launch a fibre that will only ever see a laser in service gives an unrepresentative picture. That’s where encircled flux comes from: a way of defining how much power a laser source must inject, and where, so the measurement stays reproducible across laboratories. We haven’t opened the document that sets its normative definition — it lives in a paywalled IEC standard — so here we treat it only as a technical concept, without attributing thresholds or procedures we haven’t read.

What the free recommendation states, and where it points instead

The primary source we read in full is Recommendation ITU-T G.651.1 (11/2018), “Characteristics of a 50/125 µm multimode graded index optical fibre cable for the optical access network”, edition 2.0 approved on 29 November 2018 by Study Group 15, verified in force on the official ITU page. It covers 50/125 µm multimode fibre equivalent to the OM2 category — Gigabit Ethernet up to 550 metres — the historical base on which the OM3, OM4 and OM5 classes of modern data halls are built.

Clause 3, Terms and definitions, is the hinge of the whole document, verbatim: “For the purposes of this Recommendation, the definitions and the guidelines to be followed in the measurement to verify the various characteristics are given in the IEC standards series [IEC 60793], [IEC 60794] and [IEC 61280-4-1]. Values shall be rounded to the number of digits given in Table 1 before conformance is evaluated.” Two cross-references in two sentences: the measurement method sits in paywalled IEC standards, not in the text anyone can download for free; and conformance is judged after a rounding step whose rule — round up, round to even, truncate — is never specified.

The value in the table, and the one explicit mention of launch

Clause 6.1, Attenuation coefficient, sets the limit: “The attenuation coefficient is specified with a maximum value at one or more wavelengths in both the 850 nm and 1300 nm regions. The optical fibre cable attenuation coefficient values shall not exceed the values recommended in clause 7. For measuring the value of this attribute, reference is made to [IEC 60793-1-40].” The value, in Table 1: maximum 3.5 dB/km at 850 nm, 1.0 dB/km at 1300 nm. No launch condition in this clause: for attenuation, the recommendation points to the IEC method and stops there.

The launch condition appears explicitly only once in the whole document, and not for attenuation: it’s in the Table 1 row on the modal bandwidth-length product, “for overfilled launch” — lower case, exactly as printed — minimum 500 MHz·km at both 850 and 1300 nm. The recommendation knows the launch condition is a variable that needs declaring: it does so once, for bandwidth, and stays quiet elsewhere. A note to the same table, on macrobend loss, adds that its launch conditions “shall be used” are those of the attenuation measurement in IEC 61280-4-1: even a seemingly mechanical attribute imports its launch condition from a document we haven’t read.

The rounding rule that decides, before the threshold even applies

One point the text leaves open — what follows is our own observation, not a statement from the recommendation. Table 1 gives the values to a single decimal place (3.5; 1.0), so Clause 3’s rule requires rounding the measurement to that digit before judging conformance, but it doesn’t say which convention applies. A value of 3.46 dB/km and one of 3.54 both round to 3.5 and pass; 3.55, depending on the rule used, may round to 3.5 or to 3.6 — and in the second case fail a run that a specification writing only ≤ 3.5 dB/km never expected to have to judge to the millimetre.

Where the evidence sits, and the piece that’s missing

To say a multimode run is conformant, the evidence sits scattered across at least six places: the instrument’s calibration certificate, which the specification often forgets; the test report; the datasheet of the cable installed; the as-built register; the specification, with the limit value written in; the reference standard, in the applicable version — here the 11/2018 edition, not an earlier one. None of the six normally records the launch condition actually used that day: the instrument applies it but rarely logs it, the report states the value and almost never the method. It’s the same gap we find when the OTDR pulse width used in a test goes unrecorded: a setting that decides the outcome, and that nobody notes down.

What to write into the specification

Four lines, for anyone writing or checking a data-hall specification:

the limit value and wavelength, taken from Table 1 of the applicable G.651.1 edition (or from the cabling’s OM class, if higher), as a line item in the link’s optical budget;
the declared launch condition — overfilled, or the laser-source equivalent — with the IEC method invoked to obtain it, not just the instrument’s name;
the instrument’s calibration certificate, tied to the serial number used that day;
the rounding rule applied before the conformance verdict, written down before the measurement, not left to whoever signs the report.

What we haven’t opened

IEC 61280-4-1 (2009), “Fibre-optic communication subsystem test procedures – Part 4-1: Cable plant and links – Multimode fibre-optic cable plant attenuation measurement”, is not freely accessible: we know the title and year because they appear verbatim in G.651.1’s bibliography, not the content. The same goes for IEC 60793-1-40 (2001), on attenuation measurement, and IEC 60793-1-41 (2010), on bandwidth measurement: we know they exist and what they cover from the title, not their procedures. We don’t describe encircled flux as the standard defines it, because we haven’t read it: we treat it as a general concept, with the limit stated.

The first pillar, on this case: limit value, wavelength, launch condition, IEC method and applicable edition become lines in the specification, with the rounding rule written down before the measurement — ready to hold up a disputed test.

The second pillar is the single map: calibration certificate, test report, cable datasheet and as-built converge into one reference against which AI agents, with the operator in command, check that every multimode run has all the documents — not just the final number. Always within the client’s own perimeter: on-premises on self-contained machines, or a dedicated cloud with a dedicated VPN and a data centre in Italy, always with shared management.

Does your specification write only the attenuation number? Talk to an engineer: the review, at no cost, lists the clauses — launch condition, method, rounding — that usually go missing.