ONU Rx power: the normal range on GPON and EPON

A class B+ GPON ONU works anywhere from -8 to -27 dBm. A standard EPON ONU works from -3 to -24 dBm. Below is where those figures come from, and what to do at each level.

Each figure from a standard was read from its own text on 26 September 2026. ONU and ONT mean the same box here. Sources are at the foot of the page.

What the numbers mean

Optical power is quoted in dBm. 0 dBm is one milliwatt. Every 3 dB down halves the light, so -27 dBm is about two microwatts. A more negative number is weaker light.

ONU Rx is the light the subscriber's ONU receives from the OLT, on 1490 nm. It is what is left of the OLT's launch after the fibre, the splitters, the connectors and the splices.

OLT Rx is the light the OLT receives from one ONU, on 1310 nm. The OLT reads it during that ONU's own burst, so every ONU on a PON has its own OLT Rx.

ONU Tx and OLT Tx are what each laser launches. Launch minus receive is the loss of the path. With both ends in hand you can tell a weak laser from a bad fibre.

• ONU Rx and the OLT's reading of that ONU fell by about the same amount. The path got worse.
• Only the OLT's reading fell, and ONU Tx fell with it. The ONU's laser is ageing.
• Every ONU on a PON fell together. Check the OLT module's Tx and the feeder first.

The range each standard sets

GPON is ITU-T G.984.2. EPON is IEEE 802.3ah, now clause 60 of IEEE 802.3. Each defines classes, and each class fixes what the lasers launch and what the receivers accept.

Overload is the most light a receiver takes without errors. Sensitivity is the least. So the usable ONU Rx window is -8 to -27 dBm on B+, -8 to -30 dBm on C+ and -3 to -24 dBm on PX20.

• GPON B+ is G.984.2 Annex A. The standard recommends it above its other classes for 2.5G GPON.
• GPON C+ is G.984.2 Appendix V. It buys 4 dB more path. It needs a C+ module at the OLT and FEC switched on.
• EPON PX20 is the IEEE class for 20 km. Its most loss is 23.5 dB down and 24 dB up. Many EPON modules are sold as PX20+ or higher. Those are vendor grades, not IEEE classes, and their receivers reach below -24 dBm. Read your ONU's datasheet.
• XGS-PON class N1 (ITU-T G.9807.1) launches +2 to +5 dBm at 1575 to 1580 nm. Its ONU works from -9 to -28 dBm across 14 to 29 dB of plant.

Bands you can act on

These bands are rules of thumb. The healthy band stops 3 dB above the receiver's floor. That 3 dB is the margin the Fiber Optic Association suggests for ageing and repairs. The ONU's own optics class decides in the end.

• Too hot. The receiver is overloaded and errors climb. Nothing is broken in the plant. Fit an attenuator at the ONU on the next visit.
• Healthy. No visit. Note the reading. It is this ONU's normal.
• Watch it. It works with little margin left, and one dirty connector can tip it over. Clean its connectors on the next job in that street. Do not send anyone for this alone unless it keeps falling.
• Send someone. It is at or past the receiver's floor. Expect drops and errors. Send someone with a cleaning kit and a power meter.

A fall of 3 dB or more from an ONU's own normal is worth a look in any band. Half its light has gone somewhere.

If your EPON ONU's datasheet rates its receiver at -27 dBm, use the GPON column. The OLT's reading of each ONU takes the same 3 dB margin above the OLT's own floor: -28 dBm on B+, -27 dBm on PX20.

A simple loss budget

Loss adds up in dB. Take the launch power, subtract every loss on the path, and you have what the ONU should read.

Splitters

The ideal figure is physics: each halving costs 3 dB. A real PLC splitter loses a little more. The limits here are one maker's datasheet, measured without connectors. Add about 0.2 dB for each connector on a pigtailed splitter.

An uneven coupler follows the same rule on each leg. A 70/30 coupler costs at least 1.5 dB on its 70 leg and 5.2 dB on its 30 leg.

Fibre, connectors and splices

The fibre figures are the most a G.652.D cable may lose. New cable usually does better, and 1490 nm loses less than 1310 nm. The connector and splice figures are the Fiber Optic Association's estimates beside the TIA-568 limits.

Worked example: 1:32 at 5 km on GPON B+

B+ allows a launch of +1.5 to +5 dBm, and +3.0 sits near the middle. We took 0.5 dB a connector, halfway between a clean one and the limit, because field connectors get handled.

21.4 dB sits inside B+'s 13 to 28 dB. The ONU reads -18.4 dBm, 8.6 dB above its -27 dBm floor. That is healthy. If the OLT module launches only the lowest +1.5 dBm, the ONU reads -19.9 dBm. Still healthy.

Now change one thing at a time.

• Make it 1:64. The splitter costs 3.6 dB more. The ONU reads -22.0 dBm. Still healthy.
• Build the 1:32 as a 1:4 then a 1:8. That is 17.4 dB of splitter plus one more connector pair, 1.1 dB worse. The ONU reads -19.5 dBm.
• Run 20 km instead of 5. The fibre costs 6.0 dB more. The ONU reads -24.4 dBm, in the watch band on its first day.
• Look upstream. Say the ONU launches +2 dBm at 1310 nm through the same 21.4 dB. The OLT reads -19.4 dBm against its -28 dBm floor.
• Put the same plant on EPON PX20. With an OLT module at +4 dBm, the ONU reads -17.4 dBm. A 1:64 split pushes the loss to 25.0 dB, past PX20's 23.5 dB. That run needs a higher grade module or a smaller split.

Low light: what the crew checks first

Start with the cheapest check and work outward.

1. Connectors. Dirt on an end face is the usual first suspect and the cheapest fix. Clean both ends, look through a scope, then reconnect. A green APC plug belongs in a green APC adapter. Mated to a blue UPC one it loses a lot of light.
2. Bends. A drop pulled tight round a door frame, a staple through the jacket, a loop too small in a closure tray. A bend costs more at longer wavelengths. So a bend shows first on the ONU's 1490 nm reading, while the OLT's 1310 nm reading of that ONU barely moves.
3. A bad splice. A closure reopened for another job, a splice left out of its sleeve, a tray packed too hard. An OTDR shows it as a step at a known distance.
4. The wrong splitter port. A drop moved to a deeper cascade, or to the weak leg of an uneven coupler. Each halving costs 3 dB. Check the port against your records.
5. A long drop. The fibre itself costs little, 0.2 dB for 500 m. The joints and bends along a long drop cost more.

One ONU low while its neighbours on the same splitter are fine points at its drop, its connectors or the ONU. Every ONU on a splitter low together points at the splitter or something above it.

Too much light on a short run

B+ assumes at least 13 dB of loss, and PX20 at least 10 dB. A small split close to the OLT can come in under that. Take an EPON OLT at +7 dBm through a 1:4 split, 1 km of fibre and three connector pairs. That is about 9 dB of loss, so the ONU reads about -2 dBm, above its -3 dBm limit. A 5 dB fixed attenuator at the ONU brings it to -7 dBm. Match the attenuator to the connector type.

Where HansaNet fits

HansaNet shows every ONU's optical power, with 30 days of history, on OLTs that report it: C-Data and DBC EPON after a short one-time setup, Syrotech and C-Data GPON, and the Stelfiber STGP08X. It shows none on BDCOM EPON, and other families get it once we have run one. See optical monitoring and the OLT pages.

Limits

• The bands are rules of thumb. The ONU's own optics class and datasheet decide.
• An ONU's reading of its own light is good to about 3 dB, and the OLT's reading of each ONU to about 2 dB, per G.984.2 Annex B. Repeatability is 0.5 dB. So a change against the same ONU's history tells you more than one reading against a table. Confirm with a calibrated meter before you cut or re-splice anything.
• A plain power meter adds up every wavelength on the fibre. If a 1550 nm video signal shares it, use a PON meter or one with a 1490 nm filter.
• The worked example uses our planning figures. Your own meter readings and OTDR traces beat any of them.
• These figures cover 2.5G GPON and 1G EPON. 10G EPON has its own IEEE classes.

See also: monitoring a 1,000-subscriber ISP, and the fibre map for splitter ports and drops.

Sources

All read 26 September 2026.

• Primary. ITU-T G.984.2 (08/2019), GPON physical media dependent layer. Class B+ from Annex A, Tables A.1 and A.2. Class C+ from Appendix V, Tables V.1 and V.3. Measurement accuracy from Annex B, Table B.1. itu.int.
• Primary. IEEE 802.3ah-2004, clause 60, 1000BASE-PX20. Launch power from Table 60-6, receivers from Table 60-8, path loss from Table 60-9. standards.ieee.org.
• Primary. ITU-T G.9807.1 (02/2023), XGS-PON. Class N1 from Annex B, Tables B.9.2 to B.9.4. itu.int.
• Primary. ITU-T G.652 (08/2024), single-mode fibre and cable. The G.652.D cable attenuation limits. itu.int.
• Vendor. FS, PLC fibre splitter datasheet, bare fibre in a steel tube. Insertion loss by split, without connectors. fs.com.
• Secondary. The Fiber Optic Association, loss budgets. The TIA-568 limits of 0.75 dB a connector pair and 0.3 dB a splice, and the 3 dB margin. thefoa.org.
• Secondary. The Fiber Optic Association, loss estimates. 0.3 dB a connector and 0.15 dB a fusion splice. thefoa.org.
• Secondary. Fluke Networks, on testing at two wavelengths. A bend loses more at 1550 nm than at 1310 nm. flukenetworks.com.