Attenuation and transmission windows
Light gets weaker as it travels through glass, and how much weaker depends on the wavelength — a curve with two humps and three valleys that decided which wavelengths the whole industry uses. This page explains where the loss comes from (scattering, absorption, the water peak), quantifies it per kilometre for each fibre type and window, adds the discrete losses of connectors and splices, and shows how the total turns into the "dB" column of every link budget.
Where the loss comes from
| Mechanism | Physics | Wavelength dependence | Share at 1550 nm |
|---|---|---|---|
| Rayleigh scattering | density fluctuations frozen into the glass scatter light in all directions | ∝ 1/λ⁴ — dominates at short wavelengths | ~0.15 dB/km of the 0.2 |
| Infrared absorption | silica lattice vibrations | rises steeply above 1 600 nm | small at 1550, closes the window beyond 1 700 |
| OH⁻ (water) absorption | hydroxyl ions from manufacturing | peaks at 1 383 nm (and 1 240, 950 nm) | 0 in low-water-peak fibre; up to 1–2 dB/km in old fibre |
| Metal-ion absorption | Fe, Cu, Cr impurities | — | negligible in modern fibre |
| Bending | macro- and microbends | grows with wavelength | design-dependent (Light in fibre) |
| UV absorption | electronic transitions | far below 800 nm | irrelevant for telecom |
The minimum of the sum is at ≈ 1 550 nm (0.17–0.20 dB/km in the best fibre) — the reason long-haul and DWDM live there.
The windows
| Window | Wavelength | Typical attenuation (SMF) | Typical attenuation (MMF) | Who uses it |
|---|---|---|---|---|
| 1st, 850 nm | 770–910 | — (multi-mode only) | 2.5–3.5 dB/km | VCSEL-based SR optics, 1000BASE-SX to 800G-SR8; SWDM |
| 2nd, 1310 nm (O-band) | 1 260–1 360 | 0.32–0.40 dB/km | 0.6–1.5 (MMF at 1300) | LX/LR/DR/FR, CWDM low band, PON upstream, zero dispersion |
| E-band | 1 360–1 460 | 0.35 (only low-water-peak fibre) | — | CWDM channels 1371–1451 |
| S-band | 1 460–1 530 | 0.25 | — | PON XG-PON upstream? no — GPON downstream 1490, CWDM 1471–1511 |
| 3rd, 1550 nm (C-band) | 1 530–1 565 | 0.18–0.22 dB/km | — | ER/ZR, DWDM, EDFA gain window, PON video/1550 |
| L-band | 1 565–1 625 | 0.20–0.25 | — | DWDM extension, PON XG-PON downstream 1577, monitoring 1625 |
| U-band | 1 625–1 675 | 0.3+ | — | OTDR monitoring of live systems |
Band names and channel plans: Wavelength bands.
Attenuation by fibre type
| Fibre | 850 nm | 1310 nm | 1383 nm | 1550 nm | 1625 nm | Standard maximums |
|---|---|---|---|---|---|---|
| OM1 (62.5 µm) | 3.5 | 1.5 (1300) | — | — | — | TIA-568 |
| OM2–OM5 (50 µm) | 3.0 (2.3 typ.) | 1.5 (0.7 typ.) | — | — | — | TIA-568 |
| OS1 (indoor tight buffer) | — | 1.0 | — | 1.0 | — | ISO 11801 |
| OS2 / G.652.D (low water peak) | — | 0.35 (0.4 max) | ≤ 0.4 | 0.20 (0.3 max) | 0.23 | ITU-T G.652.D |
| G.652.B (older, water peak) | — | 0.35 | 1–2 | 0.22 | 0.3 | G.652.B |
| G.655 NZ-DSF | — | — | — | 0.22 | 0.25 | G.655 |
| G.654.E ultra-low-loss | — | — | — | 0.17 | 0.19 | G.654 |
| G.657 bend-insensitive | — | 0.35 | — | 0.20 | 0.23 | G.657 |
Field rule: use the cable manufacturer's specified maximum for design (0.35/0.22), the measured value for troubleshooting.
Discrete losses
| Element | Typical | Design allowance | Notes |
|---|---|---|---|
| Connector pair, single-mode UPC, clean | 0.1–0.3 dB | 0.5 dB (TIA), 0.75 dB (old designs) | grade B connectors ≤ 0.25 dB typical |
| Connector pair, APC | 0.1–0.3 dB | 0.5 dB | plus return loss ≥ 60 dB |
| Connector pair, multi-mode | 0.1–0.5 dB | 0.75 dB | — |
| MPO-12/16 pair | 0.2–0.5 dB (low-loss ≤ 0.35) | 0.5–0.75 dB | per mated pair, all fibres |
| Fusion splice, single-mode | 0.02–0.1 dB | 0.1–0.3 dB | measured by OTDR |
| Fusion splice, multi-mode | 0.05–0.2 dB | 0.3 dB | — |
| Mechanical splice | 0.2–0.5 dB | 0.5 dB | temporary repairs |
| Dirty connector | 0.5–3 dB | — | the most common loss of all (Plant faults) |
| 1:2 splitter | 3.5 dB | 3.9 dB | each doubling adds ≈ 3.2–3.5 dB |
| 1:32 splitter | 17 dB | 17.5–18 dB | PON (ODN classes) |
| CWDM mux (8-ch) | 1.0–2.0 dB | 2.5 dB | per pass; ×2 for mux + demux (CWDM budget) |
| DWDM AWG (40-ch) | 3–5 dB | 5–7 dB | per pass |
| Fixed attenuator | 1–20 dB nominal | ± 0.5 dB | to prevent receiver overload |
Putting it together
Total loss = Σ(fibre length × dB/km) + Σ(connector pairs) + Σ(splices) + Σ(passive devices).
| Example link | Fibre | Connectors | Splices | Devices | Total |
|---|---|---|---|---|---|
| 10 km SMF at 1310 nm, 4 connector pairs, 4 splices | 3.5 | 2.0 (design) / 0.8 (typ) | 0.4 / 0.2 | — | 5.9 design / 4.5 typical |
| 40 km SMF at 1550 nm, 4 pairs, 12 splices | 8.8 | 2.0 | 1.2 | — | 12.0 |
| 300 m OM4 at 850 nm, 2 pairs | 0.9 | 1.5 | — | — | 2.4 |
| 20 km GPON at 1490 nm, 1:32 split, 4 pairs, 6 splices | 5.0 | 2.0 | 0.6 | 17.5 | 25.1 (class B+ 28 dB) |
| 40 km CWDM at 1550 with mux/demux | 8.8 | 2.0 | 1.0 | 5.0 | 16.8 |
Compare with the module's budget: Link budget engineering.
Wavelength matters twice
- Loss per kilometre — a 1550 nm ER module loses 0.2 dB/km where a 1310 nm LR loses 0.35; over 40 km that is 6 dB of difference, the reason ER and ZR sit at 1550.
- Bend sensitivity — losses from bends grow with wavelength; a plant that passes at 1310 can fail at 1550 or 1577 (XG-PON), and a 1625 nm test wavelength is used to find such bends (Testing).
In CodingBox
The module's DDM gives the two ends of the equation — Tx power here, Rx power there — and the difference, read on the bench with a known-good jumper and then in the rack, is the plant loss; comparing it with the table above tells you whether the plant is what the design assumed (DDM in the app, Rx power & budget).
Light that comes back instead of going forward — Fresnel reflection, reflectance and return loss per component and per technology: Reflections & return loss; how attenuation grows with age — hydrogen, water, radiation, microbending: Reliability & ageing.