G.652 or G.657: Which Fibre to Specify for Tight Bends?

A technician closes the last splice-tray connection in a tight cabinet: to reach the enclosure, the patch cord must make a bend tighter than the design allowed. The next day’s test result shows a loss no one had budgeted for — not a poor splice, not a dirty connector: it is the fibre bent beyond what its category allows. It happens every time a specification simply states “fibre compliant with G.652.D” and stops there, while the installer threads it into a space designed for a very different fibre.

What bend radius means, in one line.

Every optical fibre has a minimum bend radius: the point below which light starts leaking out of the core instead of staying trapped by total internal reflection. The effect is called macrobending and it is a continuous loss, not a one-off event: the tighter the bend, the higher it climbs, and past a certain threshold it climbs almost vertically. A fibre designed for a 30 mm radius, installed at 10 mm, doesn’t “perform a bit worse”: it can lose several dB in a single loop — enough to fail an entire link that looked perfect on paper.

G.652 and G.657: the same glass, different tolerance to bending.

Standard single-mode fibre, ITU-T G.652.D, is what backbones and long runs are built from: excellent in a straight line, but historically characterised only at relatively wide bend radii, around 30 mm. That gap produced a family purpose-built for the places where a bend is unavoidable: ITU-T Recommendation G.657, “Characteristics of a bending-loss insensitive single-mode optical fibre”, with four sub-categories and as many minimum design radii:

G.657.A1 — 10 mm, fully compliant with G.652.D: it splices into standard fibre without any special precaution.
G.657.A2 — 7.5 mm, still fully compliant with G.652.D.
G.657.B2 — 7.5 mm like A2, but only “compatible” with G.652.D: small differences in chromatic dispersion and PMD that normally cause no issues, but do not guarantee formal interchangeability on a mixed run.
G.657.B3 — 5 mm, built for the last metres inside a building: dense patch panels, apartment entries, interconnects inside a data centre rack.

The difference between category A and category B isn’t only the radius: it’s where you can use it without thinking twice. Category A is built for the whole access network; category B for the last metres, in the “bend-rich” environments the recommendation itself describes.

The same 10 mm bend, three very different losses.

This is why the bare label “G.657” on a specification, with no sub-category, isn’t enough. At the same 10 mm radius and a single turn, the macrobend loss limits manufacturers declare — consistent with the tests set out in IEC 60793-2-50 — are:

G.657.A1: up to 0.75 dB at 1550 nm (1.5 dB at 1625 nm).
G.657.A2: up to 0.1 dB at 1550 nm (0.2 dB at 1625 nm).
G.657.B3: up to 0.03 dB at 1550 nm (0.1 dB at 1625 nm) — a radius that, for this category, is generous.

Twenty-five times the difference between the first and the last, on the exact same physical bend. Tighten to 7.5 mm and the gap widens further: an A2 fibre can reach 0.5 dB, a B3 stays under 0.08 dB. At 5 mm — the radius typical of a shallow wall box — only B3 has a guaranteed limit: 0.15 dB at 1550 nm, 0.45 dB at 1625 nm. An A1, or a G.652.D, bent to that radius has no declared limit at all: no one ever promised it would hold.

Not every “G.652.D” on a specification is the same.

Late-generation G.652.D fibres, sold as “low-loss”, often beat the old G.657.A1’s bend performance straight off the factory floor, while remaining fully compliant with G.652.D. A specification that only says “compliant with G.652.D”, with no macrobend requirement of its own, can just as easily be filled with a fibre like that as with older stock never characterised below 30 mm: the category label alone guarantees nothing at a tight radius. It needs a maximum loss value the manufacturer declares, at the radius actually planned for installation.

The fifth term in the optical budget.

We worked out here the optical budget of a link by adding up cable attenuation, splices, connectors and splitters. There is a fifth line item that calculation doesn’t list because it is normally zero: bend loss, which becomes real every time the design calls for a tighter radius than the fibre chosen can guarantee — a compact cabinet, an unavoidable cable gland, a shallow wall box. Ignoring it doesn’t make it disappear: it just moves it to test day, where an OTDR trace shows it as a loss event indistinguishable, at first glance, from a poor splice. It’s the same physical principle behind the controlled bending used to tap a fibre, around 1 dB: there, the bend is deliberate; here, it’s an installation error — but the physics that produces it doesn’t care which.

Checklist: what to put in the specification.

Always specify the sub-category (A1, A2, B2, B3), never just “G.657” or just “G.652.D”.
In racks, dense patch panels and the last metres inside a building, require G.657.B3 or at least A2.
On long backbone runs, where radius isn’t an issue, G.652.D or G.657.A1 remain the cheaper choice and splice fully into the existing base.
Have the supplier declare the macrobend loss limit at the minimum radius actually planned for installation, not just the category label.
On a mixed run with existing G.652.D segments, prefer A1/A2 over B2/B3 wherever full declared compliance is required, not just compatibility.
Add bend loss to the project’s optical budget whenever a section of the run calls for a radius under 15 mm.

The bottom line.

The right fibre for the right bend almost always costs the same as the wrong one: the difference sits in the label written into the specification, not in the price per metre. It’s one of the first things we check when we design a cable run: wherever the route forces a tight bend — a data centre rack, a building entry, a dense patch panel — the fibre category gets chosen before the cable is ordered, not after the first failed test.

Do you have a project with tight enclosures or dense patch panels and aren’t sure which fibre category to specify? Talk to an engineer about it: checking bend radii is part of every site survey, free of charge.