Macrobend or microbend: how to tell them apart on a cable already in the ground

“OTDR test passed, no anomalies detected.” Then, months after commissioning, a link that passed acceptance starts losing a few extra tenths of a dB, always at the same point, or always over the same section. Nobody has touched the cable. The question that decides the fix is the first one an engineer asks on site: is it a tight bend that formed — or settled — inside a closure or a manhole, or is it a microbend distributed along the section, for instance where the cable is crushed inside an overfilled duct or squeezed by backfill? The two diagnoses call for opposite repairs, and an OTDR run at the wrong wavelength tells them apart not at all.

Two different physical phenomena, not two degrees of the same defect.

Macrobending is a bend with a radius that is large compared with the fibre’s own diameter — a tight loop in a closure, a forced radius at a duct entry, a cable folded past its design minimum. It is a discrete event: on an OTDR trace it shows up as a single non-reflective loss step, localised at one precise point along the link, indistinguishable at first glance from a badly fused splice.

Microbending is something else entirely: thousands of random micro-deformations of the fibre axis, on a scale comparable to the fibre’s own diameter, caused by distributed lateral pressure — a cable crushed inside an overfilled duct, cable ties pulled too tight, ice on an aerial cable, backfill settling on a direct-buried run. It is not an event: it is a rise in the attenuation coefficient (dB/km) over an extended section, with no single point to put a finger on.

The physical law that makes the diagnosis possible.

ITU-T Recommendation G.650.1, Definitions and test methods for linear, deterministic attributes of single-mode fibre and cable, edition 01/2024 approved on 13 January 2024, devotes clause 6.6 to “Test methods for macrobend loss”: the reference method winds a fibre sample around a mandrel of known radius and measures the induced loss. By the Recommendation’s own admission, this is a method meant “mainly for factory measurements” — a factory-level characterisation of the fibre type, not a field diagnostic procedure for a cable already in the ground. But clause 6.6.1.2.4 states the physical principle that field diagnosis borrows: “it should be considered that macrobending losses increase exponentially with the wavelength” — bend loss grows exponentially with wavelength. Clause 6.6.1.2.3 adds the second term: “the macrobending losses increase exponentially as radius decreases”. That is why the standard specifies measurement at 1550 or 1625 nm, not at 1310: a bend that is almost invisible at 1310 nm can cost several dB at 1625.

We have already seen this in figures when discussing fibre choice: a G.657.A1 fibre bent to a 10 mm radius is rated for up to 0.75 dB of loss at 1550 nm and up to 1.5 dB at the very same bend at 1625 nm — double. That is the exponential of clause 6.6.1.2.4, with a number attached.

The signature that separates a bend from a splice.

That physical law is exactly what an OTDR exploits in the field, and it is reported practice among instrument manufacturers, not an ITU-T requirement. A VIAVI Solutions white paper on macrobend detection puts it plainly: the same event — the same splice, the same connector — is measured at two wavelengths, historically 1310 and 1550 nm; the paper adds that “the best measurements to analyze macro bending should be taken between 1310 nm and 1625 nm, or between 1550 nm and 1625 nm”, the wavelengths relevant to a DWDM link. The reading criterion is direct: “for a given event (splice or connector), if there is no macro bend, the loss measurement shall be about the same at any wavelength. If there is a large difference (>0.2 dB) between the two wavelengths, this is due to macro bend” — if the two measurements match, it is a splice or a connector; if they differ by more than roughly 0.2 dB, it is a bend. That is a value declared by a single manufacturer, not a normative threshold: the physical difference is certain, the numeric threshold has to be declared by whoever runs the acceptance test, not left at the instrument’s factory default.

Microbending leaves no single event to measure twice.

Here the multi-wavelength diagnosis changes shape, because there is no single point to compare. The same physical principle — single-mode fibre has a larger mode field diameter at 1550 nm than at 1310, and larger still at 1625, and a wider mode field is more sensitive to any bend — still applies, but it applies to a section, not to an event: what gets compared is the section’s average attenuation coefficient, at the different wavelengths, against the intrinsic value declared by the fibre manufacturer, not the size of a single step. The same white paper describes, for a related phenomenon — residual tension in a fibre along a long span, with no splice or connector event at all — the technique that extends by analogy here too: not a search for events, but a comparison of the trace against a tolerance mask on the average attenuation, typically ±0.1–0.2 dB in reported operator practice.

A factory standard for microbending, in the same series that characterises macrobending at a fixed radius — IEC 60793-1-47:2017, cited by G.650.1 itself for the detail on radius and wavelength — simply does not exist. G.650.1, when describing factory attenuation measurements, only recommends keeping the fibre loose on the drum, because “microbending effects should not be introduced by the drum surface”: an effect to be avoided in the measurement, not a parameter characterised at a fixed radius the way macrobend is.

What to put in the specification.

Bidirectional OTDR testing at a minimum of two wavelengths on every link, not one — 1310/1550 nm at a minimum; 1550/1625 nm where the link will carry DWDM or L-band traffic, or where in-service maintenance testing at 1650 nm is already planned.
Every above-threshold event reported at both wavelengths, with the difference calculated — not the single value: it is the difference that makes the diagnosis, not the absolute figure.
A declared difference threshold in writing, above which an event is classified as a suspected macrobend and must be physically verified — not left at the instrument’s factory default.
The average attenuation coefficient per section, at the same wavelengths, compared against the value declared by the fibre manufacturer — the only way to catch a distributed microbend that no single event will ever reveal.
Physical verification as a closing condition: a suspected bend must be opened, relaxed and re-measured, not simply re-documented as “within threshold”.

How it is checked on acceptance.

It is not enough to hand over two traces, one per wavelength, filed separately: the acceptance record must show the comparison already made, event by event, with the dB difference sitting next to each value — the same rigour that a proper reading of an OTDR report already demands for path bidirectionality needs to extend here to wavelength bidirectionality. Where the difference exceeds the declared threshold, acceptance requires proof that the physical fix was actually carried out: the corrected section is re-measured, and the loss at the longer wavelength must drop by a commensurate amount — if it barely moves, the bend was not the cause, or it was not fixed. On long sections with no discrete events, check that the average coefficient declared at acceptance is the one actually measured at the different wavelengths, not the fibre’s catalogue figure: a systematic deviation across the whole section is the signature of a microbend that no single event would ever have shown. The choice of pulse width needs to be declared for both measurements: two different pulse settings at the two wavelengths make the comparison useless.

The point.

The physics that separates a bend from a splice has been known for twenty years and sits in a clause of a standard that almost no specification cites: bend loss grows exponentially with wavelength, splice loss does not. The rest — the difference threshold, the average-coefficient comparison, the physical check before closing the test — is written in no Recommendation: whoever writes the specification has to write it. That is the work we do when we certify a link with an OTDR at the right wavelengths, and when we turn that difference threshold into a verifiable clause in a specifications and compliance engagement, with the test and the documentation delivered together with the work. Every suspected bend, every out-of-mask coefficient and the outcome of the physical check feed into the same network map — not scattered files — where an AI flags the deviation between wavelengths and one of our teams closes the fault, for an operator, a data centre, an industrial site, a public administration body, a healthcare facility or a defence site. Where an AI is needed on network data, it runs within the client’s perimeter — on-premise or on dedicated cloud with a data centre in Italy — alongside CSIDIA, the group’s other company.

Do you have a link losing more than it should, and you cannot tell whether it is a bend or a splice? Talk to an engineer: the site visit costs nothing, and the two-wavelength diagnosis is run with the same OTDR used for acceptance, before anyone has to reopen the route to find out where to look.