APC or UPC: which connector goes in the specification, and why they are not interchangeable

“Single-mode optical connector, SC, compliant with current standards.” That line turns up in a great many fibre network specifications, and it skips the one detail that decides everything else: whether the connector face is flat or cut at a slant. PC and UPC are not two grades of the same object, and neither is APC a step up from either: they are two different geometries, defined by two different documents, updated together in April 2024. Neither can be inferred from the other.

Two families, not a quality tier.

Physical contact (PC) brings two dome-polished fibres together face to face, perpendicular to the ferrule axis: whatever little light reflects at the joint, even with a perfect contact, travels straight back down the same core towards the transmitter. UPC (ultra physical contact) is the same geometry with a finer polish and better reflectance: a finish grade, not a different standard. The angled version (APC) changes the geometry, not the finish: the same dome-polished face is cut at a conventional 8° angle from the perpendicular plane. What little reflects no longer falls within the core’s acceptance cone: it escapes sideways, into the cladding, and is attenuated there. It is the same principle as the blade at the top of this page: a slanted cut does not cut better, it deflects whatever bounces back.

The standards that define the face, updated together on 11 April 2024.

The geometry of the two cylindrical zirconia ferrule connectors, 2.5 mm and 1.25 mm in diameter, is defined by two parts of the same IEC series, published on the same day in their second edition: IEC 61755-3-1:2024 for the non-angled interface (PC/UPC) and IEC 61755-3-2:2024 for the angled one. Both replace their first editions from 2006. These standards are paywalled and we have not read them in full: what follows comes from the public IEC data sheets, not from the full text. Both data sheets declare, among the changes, two optical interface variants — with and without core-axis orientation towards the guide key — and additional requirements for grades B and C, absent from the 2006 edition. A specification that simply writes “connector to IEC 61755-3-2” without an edition can, in theory, still point to the 2006 text: since 2024 the replacement is formal, and the two texts are not identical.

What the transmission standard does — and does not — say.

Anyone looking for a single reflectance figure finds it in ITU-T G.671, Transmission characteristics of optical components and subsystems: we downloaded and read it, it is free, and the edition in force is 9.0, approved on 29 November 2025. Clause 6.7, “Optical connector”, sets a maximum insertion loss of 0.5 dB for single fibre and 1.0 dB for multifibre, and a maximum reflectance of −35 dB, with a note stating that the values already include the effect of mating durability. It is the only connector clause in the document, and it does not distinguish PC from APC. A real APC connector sits well below that limit, but the transmission standard does not demand it. Anyone citing G.671 alone in a specification has written a limit that both families clear without effort: they have not chosen between the two.

The figure that actually forces the choice: the GPON case.

On a passive access network, the number that matters is not that of a single connector: it is that of the whole optical path. ITU-T G.984.2 (edition in force 08/2019, also free: we read it), Table 2c, sets for GPON at 2,488.32 Mbit/s downstream a minimum optical return loss of the ODN at Olu and Old greater than 32 dB, on both single fibre and dual fibre, with a more permissive optional value — 20 dB — allowed only in the cases described in an appendix to the earlier G.983.1. We checked: G.984.2 does not state anywhere “APC connector mandatory”. But those 32 dB are shared across every connector and splice in the path, from the exchange patch panel to the splitter to the customer socket. An angled connector, by construction — the reflected field falls outside the core’s acceptance cone — systematically delivers better reflectance than a non-angled one: that is a consequence of the geometry described in IEC 61755-3-2, not a figure we quote here, since grades B and C remain in the paywalled text. What matters for the specification is that the margin is spent connector by connector: on a path with many patch points, the family that leaves more of it standing is the one that pays off. The same logic — adding up margins along the whole GPON path — applies to the full optical budget, which we covered separately for classes B+ and C+.

The risk on site: the two families do not mate.

An APC connector plugged into an adaptor meant for PC/UPC does not find the physical contact it should: the 8° angle meets a flat face, contact happens on only one side of the core, loss rises unpredictably and reflectance can worsen rather than improve. If the connector is forced, the risk is not only optical: the ferrule can chip at the edge, damage that visual inspection of the face alone only catches afterwards, under a microscope. Hybrid adaptors and patch cords built specifically for the interface between the two families do exist, with a declared and accepted loss written into the design: that is a documented exception, not an improvised fix on site. Housing colour — green for APC, blue for single-mode PC/UPC — is a convention widespread across the industry: we did not find it written as a requirement in any specific standard, and we do not treat it as one. On site, check the marking printed on the connector body, not the colour, which remains a quick visual check and not a substitute for it.

What to write in the specification.

The family for every position, not “optical connector”: PC/UPC or APC, for every patch panel, jumper and outlet.
The interface with its edition: IEC 61755-3-1:2024 or IEC 61755-3-2:2024, not just the part number.
The reflectance limit the network actually needs, not the generic one from G.671: for GPON the reference is G.984.2, for other technologies the applicable recommendation.
A ban on undeclared hybrid adaptors, unless a design waiver accepts the loss in writing.
The test at acceptance: return loss measured with the bench method IEC 61300-3-6 on the component; on the installed plant the method is different — IEC 61280-4-2.

Checking the right family before testing costs one extra line in the specification. Finding out afterwards, on a path already installed, costs a redone patch and a test report nobody wants to sign twice: that is why every specification we review goes through this line as well.

Do you need to write a specification with mixed PC and APC connectors, or check a delivery that has already reached the site? Talk to an engineer: the right family is one line to write, the wrong one is one test bench to discover.