A GPON network fails slowly, not suddenly. From the OLT to the ONU, the ODN is all splitters and fiber with no active node in the middle to self-report, so a micro-bend, an oxidized connector, or a damp splitter drags receive power down gradually until a user finally calls. Turn inspection from "fix after outage" into "check on a schedule" and most faults get caught before they happen.
Read Rx power first: GPON Class B+ sensitivity is −28 dBm (ITU-T G.984.2), but alert below −25 dBm and treat below −27 dBm as cliff-edge. OLT Tx sits at +2 to +5 dBm.
Clean light ≠ clean link: check CRC errors on the OLT and uplink switch — GPON runs 2.488 Gbps down / 1.244 Gbps up (G.984.1), and a few CRC errors mean mosaic and stutter on video.
Temperature kills optics: commercial grade is 0 to 70 °C, industrial −40 to 85 °C; above 60 °C check the airflow and dust filter.
Let SNMPv3 (RFC 3414) watch for you: at 100+ ONUs, threshold alarms beat manual walks.
A switch network has active nodes that raise traps; a PON network's ODN does not. Fiber micro-bends, oxidized connectors, and damp splitters never cut service outright — they erode receive margin until a user drops off the edge. Scheduled proactive checks pull those hidden faults out before they become outages.
Log into the OLT and export the receive (Rx) of every ONU under each PON port. Class B+ receive sensitivity is −28 dBm (ITU-T G.984.2), but do not wait for that number — below −25 dBm is already a warning, below −27 dBm is walking the cliff. OLT transmit (Tx) on the OLT side is normally +2 to +5 dBm.
At a 1:128 pon splitter, far-end ONU receive usually lands at −20 to −24 dBm, a healthy band. The first read of any GPON network inspection is always this power table.
Good power does not guarantee a clean link. Open the interface counters on the OLT and the uplink switch and watch CRC errors and code errors climb. GPON down is 2.488 Gbps, up is 1.244 Gbps (G.984.1); a little CRC on HD video shows up as mosaic and stutter.
Occasional CRC is worth watching; sustained growth almost always means compressed fiber, a dirty connector, or an aging optic — trace that strand.
Commercial optics are rated 0 to 70 °C, industrial to −40 to 85 °C. Kill the AC and the OLT board temperature climbs, laser wavelength drifts, and receive starts to float. During inspection, read the device's own temperature sensor; above 60 °C check the airflow path and dust filter before a fan stops.
In the OLT alarms, LOS (loss of signal) and LOSi (loss of frame) are the clearest. An ONU flapping up and down is usually a loose fiber connector or unstable ONU power. Sort offline ONUs over the last 7 days and prioritize the top few for a site visit. SNMP monitoring pushes these LOS and offline-ONU alarms to Zabbix automatically instead of waiting for a manual walk.
After inspection, do not leave it in your head. Export the OLT current config, the ONU LOID bindings, and the splitter-port map. Next cutover or hardware swap, that baseline saves half a day of cross-checking.
| Metric | Healthy | Warning | Danger |
|---|---|---|---|
| OLT Rx from ONU | −8 to −22 dBm | < −25 dBm | < −27 dBm |
| OLT Tx | +2 to +5 dBm | < 0 dBm | — |
| Port CRC errors | 0 growth | occasional | sustained growth |
| Device temperature | < 55 °C | 55 to 60 °C | > 65 °C |
| ONU offline | 0 | single occasional | multiple / repeated |
Inspect the pon splitter port map together with the config backup so a replaced splitter keeps its LOID bindings.
Manual inspection fits small sites of dozens of units. Past 100 ONUs, SNMP monitoring with threshold alarms is the only sane path. The OLT in Rayin's PON lineup supports standard SNMPv3 (RFC 3414) with MIB reads; paired with open-source Zabbix, an over-limit Rx power emails you automatically — inspection shifts from "people hunting problems" to "problems finding people."
How often should a GPON network be inspected? A stable site under 50 users is fine monthly; a new build or one that recently touched fiber warrants weekly for the first three months. Enterprise lines with symmetric bandwidth demands justify weekly inspection plus daily threshold monitoring.
Can −26 dBm receive still hold? It passes, but margin is only 2 dB. One more connector oxidation or one winter fiber shrink and it drops to the −28 dBm sensitivity ledge. Schedule remediation; do not wait for the break.
Power looks fine but the user says it stutters — what to check? First CRC errors and upstream bandwidth use. GPON upstream is 1.244 Gbps shared across a hundred users by TDMA; a few heavy upstream users at evening peak stall the rest. Then check whether the OLT uplink is maxed on GE.
ONU keeps dropping — where to look first? Check ONU power and the fiber connector. Logs showing repeated LOS are 90% a loose SC/APC connector or an aged cold splice; unstable power shows as regular drops you can read off the timestamp.
Where to store the inspection config backup? One local, one offline (USB or NMS server each). The backup holds LOID, VLAN, and DBA templates; losing one makes recovery painful, so bind it to the same ledger as the inspection record.
Written by Sara, Customer Manager at Rayin — over 10 years in communications, focused on helping ISPs run proactive PON and industrial-switch maintenance for emerging markets.
About Rayin → https://www.szrayin.com/Profile/
