Transceiver Won't Link: A Field Diagnosis Order for SFP+/QSFP/DD Modules (Clean, Configure, Verify, Replace)

A repeatable diagnostic sequence for optical modules that fail to link: physical inspection, cleaning, speed/FEC configuration, DDM-based fault isolation, and the compatibility faults that mimic hardware failure — with a decision table for 10G through 800G.

Troubleshooting · Updated 2026-10-24 · 3 min read

Conclusion first

"No link" diagnoses fail when engineers jump straight to RMA. The reliable sequence is physical → configuration → telemetry → isolation, and it resolves the majority of cases without touching procurement. This article gives the sequence, the decision table, and the compatibility traps that mimic hardware failure.

1. The diagnostic order

```

1. INSPECT → connector seated? latch closed? dust cap off both ends?
2. CLEAN → one-click cleaner on every ferrule that was ever exposed
3. CONFIGURE → speed, duplex, FEC, autoneg match on both ends
4. TELEMETRY → DDM: Tx power, Rx power, bias, temperature on BOTH ends
5. ISOLATE → swap module / port / cable, one variable at a time

```

Skipping step 2 is the single most common field mistake. Skipping step 3 is the second: 25G SFP28 ports default to autonegotiation that 10G SFP+ optics partially ignore, and PAM4 hosts at 50G+ lane rates have speed and FEC settings that must match module class.

2. DDM decision table

Read DDM on both ends before concluding anything:

3. The compatibility traps that look like dead hardware

These cause a large share of false RMAs:

1. Vendor-coded modules in locked switches. Many platforms reject unknown EEPROM codes at the driver level; the module is electrically fine. Test the same module in an unlocked platform before scrapping it.
2. Speed mismatch between SFP+ and SFP28. A 25G SFP28 in a 10G SFP+ port usually works; a 10G SFP+ in a 25G port will not negotiate 25G and may be reported as "unsupported" rather than "absent."
3. FEC mode mismatch on 25G/50G/100G links. RS-FEC vs FC-FEC disagreement brings the link down at the PCS layer with perfect optics. This shows up when mixing vendor generations.
4. PAM4 lane-rate confusion at 400G+. A 400G DR4 in a QSFP-DD port configured for 4 × 50G breakout will train some lanes and fail others — check the breakout and lane-map configuration before the optics.
5. AOC/DAC confusion. An AOC (active optical) is not a DAC: it needs power, is temperature-sensitive, and fails differently. Treating an AOC like copper hides the real fault.

4. Isolation procedure (five minutes, no guesswork)

1. Move the suspect module to a known-good adjacent port (same switch):

• Links → the original port or its config is at fault.
• Still down → continue.

1. Put a known-good module into the original port:

• Links → RMA the suspect module.
• Still down → the port or path is at fault; continue.

1. Replace the fiber path with a tested patch cord:

• Links → cabling/plant fault; run an insertion-loss test.

1. If all three swaps fail together, suspect configuration or platform firmware — pull port logs and module EEPROM dump.

5. After the link comes up

A link that trains is not a healthy link. Watch for 24 hours:

FAQ

The FAQ block above covers the asymmetric-DDM case, cleaning discipline, and module-vs-port isolation.

Field practices integrated from operator runbooks and MSA/IEEE diagnostic guidance; no vendor documentation copied.

FAQ

The link LED is off but DDM shows good Rx power. What now?

Rx power only proves the receive optical path. Check the far end: Tx power on the remote module, and whether both sides negotiated the same speed and FEC. Asymmetric PAM4/NRZ settings between a 25G port forced on and a 10G module will keep the link down even with perfect optics.

Should I clean fiber connectors every time?

Yes for every re-mating, with one-click cleaning tools. A single dust particle at 1310 nm on a single-mode ferrule costs 1-3 dB — enough to push a DR-class link below its budget without any visible symptom.

How do I tell a bad module from a bad port?

Swap variables one at a time: the module into a known-good port, a known-good module into the suspect port, and the fiber path last. Two of the three swaps isolate the fault to module, port, or cabling in under five minutes.