Physical-layer mismatches

A surprising share of dead links are two perfectly good parts that were never meant to meet: a multi-mode module on single-mode fibre, two BiDi modules of the same colour, an angled connector in a flat port. Nothing is broken, nothing alarms — and the link never comes up. This page lists the mismatches, how each one shows up, and how to catch it before installation.

Fibre type

Mismatch	What happens	Signature
Multi-mode module (850 nm) on single-mode fibre	almost no light couples into the 9 µm core	Rx at the floor (−40 dBm) or barely above
Single-mode module (1310/1550) on multi-mode fibre	may link over short distances, unstable; modal noise and DMD	link up at 10 m, errors or down at 100 m
62.5 µm (OM1) patch in a 50 µm plant	2–4 dB loss at the joint in one direction	asymmetric Rx, marginal link
OM1/OM2 on 10G+ SR	bandwidth-limited	works at 1G, errors at 10G, dead at 25G

Check the module's length fields and compliance codes against the plant: an SR module declares OM3/OM4 lengths (bytes 16–19), an LR module declares km (byte 14) (memory map). Fibre types and colours: Connectors & fibre.

Wavelength

Mismatch	What happens	Signature
BiDi pair wrong — both ends the same colour (two -U or two -D)	each transmitter lands on a receiver filtered for the other wavelength	Rx −40 dBm both ends, fibre proven good
CWDM module in the wrong mux port	the mux filter blocks the channel	that channel dark, others fine
DWDM tunable on a channel off the plan	filtered or attenuated	low Rx or dark
PON GPON module on an XGS-PON OLT (or vice versa)	different wavelength plans never meet	ONU never sees the OLT
1310 vs 1550 on a long span	1310 loses ~0.35 dB/km, 1550 ~0.2 dB/km	budget miss on the 1310 side

The wavelength is in the identity block (SFF-8472 bytes 60–61, SFF-8636 186–187) and on the Check transceiver screen — read it, don't trust the label. Plans: Wavelength bands & channel plans, PON wavelengths.

Connectors and polish

Mismatch	What happens	Signature
APC (green) mated to UPC (blue)	angled and flat faces cannot touch	3–10 dB extra loss, possible end-face damage; high return loss on the UPC side
Wrong MPO polarity (Type A/B/C mix)	Tx lands on Tx	link dead both directions on parallel optics, per-lane Rx at the floor
MPO pinned to pinned or unpinned to unpinned	cannot mate / no alignment	mechanical — or huge loss
MPO-12 cable on MPO-16 optics (or 8-fibre vs 12)	lanes unmapped	some lanes dark
Dirty or scratched ferrule	0.5–2 dB per particle	marginal, flapping (Link flapping)

400G DR4/DR8 and InfiniBand NDR use APC MPO — mixing in a UPC patch is the classic first-install failure (InfiniBand optics).

Tx/Rx crossed

A duplex link needs Tx→Rx in both directions. If a patch cord or panel is "straight" where it should be crossed, both ends see LOS while each Tx is happily lighting the other Tx. Test: swap the two fibres of the duplex pair at one end. On MPO this is the polarity problem above.

Too much light

Long-reach optics on a short span are a mismatch too: a ZX/ER module rated for 80 km launches +3…+5 dBm; on a 500 m link the receiver sees more than its overload point (typically −3…0 dBm) and errors or refuses to link. Fix with an inline attenuator (5–15 dB) — and read Rx power on the bench first to size it (Receivers).

Port and module class

1G SFP in a 10G-only port, or 10G SFP+ in a port that does not do 1G — see Speed & rate.
QSFP+ in a QSFP28 port works (backward compatible) if the host supports 40G; QSFP28 in a QSFP+ port does not.
OSFP and QSFP-DD do not interchange mechanically (Form factors).

Catching it before installation

Read both modules in CodingBox: wavelength, fibre type/lengths, connector code (byte 2 / 130) and rate codes.
Compare with the plant record: fibre type, connector polish, mux channel plan.
Label modules by wavelength and polish colour before they leave the bench.

If a mismatch is found after the fact and the hardware is right but the identity is not (e.g. a module coded with the wrong wavelength field for its mux plan), correct the identity in the EEPROM editor — but never code a wavelength the laser does not have.