A length of erbium-doped fibre that amplifies every C-band wavelength at once, optically, without conversion to electrical.
A Coil of Doped Fibre and a Pump Laser
An EDFA is a few metres of silica fibre doped with erbium ions, spliced into the line and pumped by a laser at 980 or 1480 nm. The pump lifts the ions into an excited state; an arriving signal photon stimulates them back down, emitting a photon identical in wavelength, phase and direction. Gain happens in glass, with no conversion to electrical and back.
The erbium transition happens to have gain across roughly 1530 to 1565 nm, which is why the DWDM grid lives in the C-band rather than anywhere else. One device lifts every channel at once, with no per-channel electronics and no knowledge of what those channels carry.
Three Places to Put One
A booster sits after the mux, takes a strong input and produces +17 to +23 dBm total to launch into the span. Its noise figure barely matters because the signal is already strong; what matters is not launching so much per channel that four-wave mixing starts.
A pre-amplifier sits before the demux and lifts a very weak input, so its noise figure is the number on the datasheet worth reading — 4.5 to 5.5 dB against 6 or 7 for a booster. An in-line amplifier at a mid-span hut does both jobs badly enough to need care with both. Keep the pre-amp's input inside its window; a fixed attenuator on an over-strong input is normal practice, not a bodge.
Gain Is Free, Noise Is Not
Amplified spontaneous emission rides along with the gain, so each amplifier costs OSNR roughly equal to its noise figure. A useful approximation for a single span is OSNR = 58 + per-channel launch power in dBm - span loss - noise figure, and every doubling of the span count costs a further 3 dB.
That is why an amplified system runs out of OSNR rather than out of power. Adding gain at the end amplifies the accumulated noise along with the signal and changes nothing. Model the chain before buying the amplifiers, not after.
Flatness, Tilt and Transients
Erbium gain is not flat. A gain-flattening filter brings it to about 1.0 to 1.5 dB across the band, but the residual tilts with total input power, so a partly filled system does not have the same per-channel gain as a full one. Fill plans and dummy loading exist for this reason.
Transients are the other trap. Drop forty channels out of forty-eight and the survivors briefly get the pump energy the departed channels were using, which can spike power hard enough to hurt downstream receivers. Gain-clamped or transient-controlled amplifiers handle it; unmanaged ones do not. The pluggable EDFA guide covers where the port-sized units fit.
EDFA reference
| Gain band | C-band, about 1530 to 1565 nm |
|---|---|
| L-band variant | about 1570 to 1605 nm, separate amplifier |
| Pump wavelengths | 980 nm (low noise) or 1480 nm (high power) |
| Typical gain | 20 to 30 dB |
| Noise figure | 4.5 to 6 dB |
| Gain flatness | 1.0 to 1.5 dB across the band |
| Polarisation-dependent gain | 0.3 to 0.5 dB |
| OSNR rule of thumb | 58 + per-channel dBm - span loss - noise figure |
| Cannot amplify | anything outside the C band — only two CWDM lanes, 1531 and 1551 nm, fall inside the gain |
Related Terms
DWDM, OSNR, OADM and coherent optics. Use the optical power converter for dBm and milliwatt arithmetic, and our 100 GHz mux/demux for the channel plan it feeds.