Optical Amplifiers Compared: SOA vs EDFA vs Raman
The three amplifier technologies compared on the numbers that decide between them: gain, noise figure, saturation power, bandwidth, and dynamics. With the use cases where each is the only sensible answer.
Scope
Three ways to amplify light in the optical domain, three very different machines: the semiconductor optical amplifier (a laser diode with the mirrors deleted), the erbium-doped fiber amplifier (a pumped ion population in glass), and the Raman amplifier (nonlinear scattering in the transmission fiber itself; see Raman scattering). This comparison covers the deciding numbers and the dynamics that datasheets bury. Foundations: optical amplifier, noise figure, saturation power, ASE.
The comparison table
Representative current figures (generational products vary):
| Property | SOA | EDFA | Raman (distributed) |
|---|---|---|---|
| Gain | 10–25 dB | 20–40 dB | 10–20 dB (on-off) |
| Noise figure | 6–9 dB | 4–6 dB (near quantum limit ~3 dB for the best) | Effective NF can beat lumped EDFA; the headline trick |
| Output saturation power | ~+5 to +18 dBm | +17 to +26 dBm | Pump-limited; distributed by nature |
| Wavelength coverage | Anywhere the material is designed: O, S, C, L, U | C-band (~1530–1565) + L; that's it | Anywhere: gain sits ~13.2 THz (~100 nm) below the pump; pick pumps, place gain |
| Gain dynamics | ~0.1–1 ns carrier lifetime → fast, pattern-dependent | ~ms upper-state lifetime → transparent to modulation | Effectively instantaneous but pump-mediated; needs low-RIN pumps |
| Polarization sensitivity | Managed by design (~0.5–1 dB PDG typical) | Negligible | Pump-polarization dependent → depolarized/dual pumps |
| Form factor | Millimeter chip; integrable (heterogeneous/hybrid) | Module: pump lasers + doped fiber + WDM couplers + isolators | Your span fiber + watt-class pumps at the terminal |
| Cost profile | Lowest per unit; scales like semiconductors | The workhorse; moderate | Highest entry (safety-interlocked W-class pumps) |
The dynamics row is the real dividing line
The EDFA's millisecond upper-state lifetime means its gain cannot follow data. It averages over millions of bits, amplifying 96 DWDM channels with no crosstalk. That single property built long-haul transmission as it exists today.
The SOA's nanosecond carrier lifetime means its gain does follow data. Drive it near saturation with one channel and every other channel sees the gain ripple: cross-gain modulation, pattern effects, inter-channel crosstalk. That disqualifies it for multi-channel line amplification. The same vice, inverted, is a feature: XGM/XPM in SOAs performs wavelength conversion, fast gating, and all-optical signal processing. And run well below saturation, a single-channel SOA behaves itself, the basis of its honest jobs.
Raman gain is nonlinear-instantaneous, but what matters practically is that the medium is the transmission fiber itself: gain distributed along the last tens of km of a span, lifting the signal before it hits the noise floor. That's why the effective noise figure of a Raman-assisted span beats any lumped amplifier bolted to its end. The signal never got as low. The bills: watt-class pumps with laser-safety engineering (automatic shutdown on fiber break), pump RIN transferring to signal (co- vs counter-propagating pumping trades gain for noise transfer), and double-Rayleigh backscatter setting the multipath floor.
Selection logic
Amplifying many C-band channels in a line system: EDFA, no contest; add distributed Raman when reach/OSNR margins demand the effective-NF boost (submarine, long-haul, unrepeated spans). Outside the erbium window (O-band datacom reach extension, S-band experiments): SOA or Raman, because erbium simply doesn't play there. On-chip or in a co-packaged transceiver: SOA, the only one that integrates. Single-channel booster/preamp at lowest cost: SOA below saturation, with its NF penalty priced into the link budget. Burst-mode PON upstream amplification: SOA again (fast gain recovery handles bursts; an EDFA's slow gain would surge between them).
One spec-reading caution transfers across all three: NF and saturation power are quoted at particular gains, wavelengths, and input powers, and an amplifier "into compression" (gain saturation) has neither its small-signal gain nor its quoted NF. The OSNR arithmetic that stacks amplifier noise down a link is its own topic. It is also the reason the 1–2 dB NF differences in the table compound into reach differences measured in hundreds of kilometers.