Design tool for the erbium-doped fiber amplifier (EDFA) that drives submarine cables and long-haul WDM networks. Sweep the 980 nm pump power, signal wavelength, input level, EDF length and stage topology to watch small-signal gain, output power, noise figure NF, OSNR and saturation output update in real time.
Parameters
980 nm pump power
mW
Output of the 980 nm pump laser that excites the EDF
Signal wavelength
nm
Input signal wavelength across the C+L band
Input signal level
dBm
Signal power at the EDFA input port
EDF length
m
Coil length of erbium-doped fiber
Er concentration
ppm
Erbium ion concentration in the EDF
EDFA stage topology
Single (high power) / dual (low NF) / distributed (hybrid)
The 980 nm pump excites Er³⁺ ions in the EDF coil; the input WDM channels (C+L band) are amplified and exit on the right. Faint green shading shows ASE noise accumulation.
Gain G and noise figure NF. g₀: gain coefficient per metre, L: EDF length, n_sp: spontaneous emission factor (≥1, =1 at full inversion). NF → 3 dB at the quantum limit.
Saturation output power. η_pump: pump-to-signal conversion efficiency (0.5–0.7 at 980 nm), P_pump: pump power in mW.
EDFA Erbium-Doped Fiber Amplifier — Gain and Noise Figure Design
🙋
I keep hearing about EDFAs — basically a "light amplifier", right? Do you boost it with an electronic circuit like an electrical amp?
🎓
Good question. The point of an EDFA is that it amplifies the light without ever converting it back into electricity. You dope the fiber core with erbium ions (Er³⁺), launch 980 nm pump light from a separate laser to excite the Er, and when a 1550 nm signal passes through, stimulated emission copies pump energy onto the signal and you get gain. EDFAs were commercialised in the early 1990s; today's submarine cables and WDM networks simply cannot exist without them.
🙋
OK… but what is this "noise figure NF" on the left? You don't hear that term as much in electronics.
🎓
Optical amplification is haunted by a noise you cannot avoid. When Er ions are excited, they emit photons in random directions even without a signal — that is amplified spontaneous emission (ASE). The ASE rides along the signal and gets amplified too, so the output SNR is always lower than the input SNR. NF in dB is exactly that SNR drop. Quantum mechanically you cannot go below NF = 3 dB (the Caves limit). Real EDFAs land at 4–6 dB; a two-stage design squeezes the total NF down to about 4 dB.
🙋
My OSNR shows something like 37 dB — that sounds like plenty of headroom?
🎓
Right at the output of one EDFA, yes — very comfortable. But a real long-haul link cascades EDFAs every 80 km for 10 to 30 stages. ASE accumulates at each span, so the OSNR collapses much faster than you expect. 10 Gbps NRZ wants OSNR ≥ 20 dB/0.1nm at the receiver; 100 Gbps coherent DP-QPSK gets by with about 14 dB. You design the first-span OSNR and the number of spans so that the final receiver OSNR stays above that target.
🙋
It also shows "Saturation output P_sat = 20 dBm". What is that the ceiling of?
🎓
No matter how much gain an EDFA has on paper, the output power is capped by the pump. Pump-photon energy is converted into signal-photon energy, so η_pump (0.5–0.7 at 980 nm) × P_pump roughly gives the maximum output. A 200 mW pump tops out near 100 mW = +20 dBm. As the input rises and the output approaches that ceiling, gain compresses into the S-curve you see on the "Gain saturation curve" chart. In WDM, all channels share this saturation output power.
🙋
The "stage topology" lets me pick single / dual / distributed. How do I choose?
🎓
By role. A single-stage EDFA is the general-purpose receive-side amp that balances NF and output (NF 5–6 dB). A dual-stage places a low-noise front amp (LN-EDFA, NF 4 dB), then a dispersion-compensating fiber or gain-flattening filter, then a high-power post amp (PA-EDFA) — the standard at intermediate stations in long-haul links. A distributed Raman + EDFA hybrid uses Raman amplification to drop the effective NF to 0–3 dB and shows up in ultra-long submarine and 400G/800G terrestrial systems. The main vendors are Fujitsu, Lumentum, NEC and II-VI (formerly Bookham).
Frequently Asked Questions
Because amplified spontaneous emission (ASE) is intrinsic to optical amplification. Even an ideal EDFA with full population inversion (n_sp=1) has NF = 2·n_sp·(1−1/G) ≈ 3 dB (=10log10(2)) as the lower bound — the Caves quantum limit. In practice, low-noise front-end EDFAs reach 4–5 dB and high-power booster stages reach 5–7 dB. A two-stage layout (low-NF preamp + high-power booster) keeps the total NF around 4 dB.
It depends on the modulation format and bitrate. For 10 Gbps NRZ-OOK at BER 10⁻⁹ a rule of thumb is OSNR ≥ 20 dB/0.1nm. For 100 Gbps coherent DP-QPSK with FEC, OSNR ≈ 14 dB is enough; for 400 Gbps DP-16QAM it climbs to 19 dB or more. Cascading EDFAs in a long-haul link accumulates ASE and degrades OSNR at every span, so the OSNR budget (margin between transmitted and minimum receiver OSNR) must be designed in at the link-engineering stage.
The C band (1530–1565 nm) is the traditional EDFA window that contains the main Er³⁺ emission peak near 1530 nm — high gain and low NF. The L band (1565–1625 nm) is reached by lengthening the EDF (typically 3–5× the C-band length) to lower the inversion. L-band EDFAs run 1–2 dB worse on NF than C-band, but doubling the available WDM capacity makes C+L parallel operation standard on submarine cables and long-haul terrestrial systems.
EDFAs cover 1530–1610 nm with 25–40 dB gain and 4–6 dB NF, making them the workhorse for long-haul WDM. SOAs (semiconductor optical amplifiers) are compact and work across all telecom bands but with NF 7–9 dB — better suited to ROADMs and intra-PIC applications. Raman amplifiers provide distributed gain in the transmission fiber itself with an effective NF of 0–3 dB and are combined with EDFAs in ultra-long-haul, high-capacity links. Each type has its sweet spot, and real systems use them in combination.
Real-World Applications
Submarine cable repeaters: Backbone submarine systems such as TPC-5 (Pacific), AAE-1 (Asia–Africa–Europe) and FASTER (Japan–US) use undersea EDFA repeaters spaced 60–80 km apart. A single fiber pair may carry nearly 100 cascaded EDFAs over 10,000+ km, so a 0.5 dB improvement in per-stage NF translates directly into OSNR budget — and ultimately into traffic capacity (20–40 Tbps with C+L).
Terrestrial long-haul WDM backbones: Carrier backbones (NTT, KDDI, AT&T, China Telecom and others) link data centres over 100–1000 km hops by cascading EDFAs to boost 80–96 WDM channels. Combined with gain-flattening filters around ROADM nodes, the per-channel gain ripple is held within 1 dB. With 400G/800G coherent transmission tightening the OSNR budget, dual-stage layouts with Raman hybrids are becoming the new normal.
Data Centre Interconnect (DCI): Metro 40–100 km DCI links built by hyperscalers (Google, Meta, AWS) are well served by inexpensive single-stage EDFAs (NF ~6 dB). Integrated transponder + EDFA modules optimised for cost dominate, including pluggable QSFP-DD ZR+ variants with on-board amplification.
Optical instrumentation and sensing: EDFAs are key components in OTDR (optical time-domain reflectometers), Brillouin / Raman distributed sensing and fibre-laser pumping. Long-range distributed fibre sensors can interrogate strain and temperature over tens of kilometres with EDFA-boosted optical signals, enabling structural-health monitoring of bridges, pipelines and power cables.
Common Misconceptions and Pitfalls
The biggest pitfall is treating "more gain = more link budget" as a free lunch. An EDFA boosts the signal, but it boosts ASE noise just as eagerly. Cascading twenty 4-dB-NF EDFAs builds the accumulated NF up to about 17 dB (10log20 + 4). Engineer the link by OSNR budget, not by gain. In long-haul designs each span typically carries 20–25 dB of gain — just enough to undo 80 km × 0.2 dB/km ≈ 16 dB of fibre loss. Excessive gain only invites saturation and runaway ASE.
Next, the myth that "more pump power always lowers NF". Higher pump does pull the inversion factor n_sp closer to 1, asymptotically driving NF toward the 3 dB quantum limit. But at low pump levels, ground-state absorption by unpumped Er³⁺ dominates and the NF actually worsens; at high pump levels, excited-state absorption (ESA) and thermal effects plateau the performance. The optimum pump power depends on the EDF length × Er-concentration combination, and typically sits around 150–250 mW. Bumping the pump higher mostly burns electricity for no NF gain.
Finally, using "OSNR in dB/0.1nm" without thinking about the reference bandwidth. Telecom convention measures OSNR over a 0.1 nm (≈12.5 GHz) reference band — that is also the default setting on most OSAs (optical spectrum analysers). But the effective noise bandwidth of a 100 Gbps coherent signal is several tens of GHz, and some papers quote OSNR per signal bandwidth instead. Confusing the two costs 5–6 dB and turns "should reach" into "does not reach". Always check the reference bandwidth (nm or GHz) before comparing OSNR figures.
How to Use
Set pump power at 980 nm between 100–800 mW; higher pump power increases gain and reduces noise figure but risks ASE saturation.
Enter signal wavelength (typically 1530–1565 nm C-band); longer wavelengths toward 1565 nm exhibit lower inversion and higher NF.
Configure signal input power (−40 to +3 dBm); weaker signals amplify with higher NF; strong signals compress gain and degrade OSNR.
Adjust erbium-doped fiber length (5–50 m); longer fibers increase gain but accumulate ASE noise at the output.
Read small-signal gain, output power, noise figure, OSNR, saturation output, and pump conversion efficiency instantly.
Worked Example
Submarine cable application: 980 nm pump = 500 mW, signal wavelength = 1550 nm, input signal = −25 dBm, EDF length = 20 m. Simulator returns small-signal gain ≈ 38 dB, output power ≈ 13 dBm, noise figure ≈ 4.8 dB, OSNR ≈ 38 dB/0.1 nm, saturation output ≈ 18 dBm, pump efficiency ≈ 65%. These values match long-haul C-band EDFA performance in terrestrial and submarine networks.
Practical Notes
Reduce 980 nm pump below 200 mW for low-cost metro links; NF rises to 5–6 dB but power consumption halves versus high-gain submarine designs.
Wavelength choice: 1530 nm peak gain suits older terrestrial systems; 1545–1555 nm flatness improves DWDM channel uniformity across 96-channel grids.
Input power sweeps identify saturation knee; −20 dBm input typically maximizes OSNR in regenerator spacing calculations for 100 km spans.
Pump efficiency below 50% signals excessive ASE; increase EDF length incrementally or raise pump to 600+ mW for transoceanic 9,000 km routes.