When performance requirements are well-defined and margins are tight. The clearest triggers: non-standard wavelength coverage (extended C-band, L-band, or non-standard spectral windows), demanding noise figure targets that catalogue products cannot reliably meet across the full dynamic range, wide gain dynamic range under variable channel loading, and mission-critical reliability requirements with full design documentation and traceability — submarine, defence, and scientific instrumentation being the canonical cases. System-specific integration is another driver: when standalone datasheet performance is insufficient to predict behaviour in the actual link, a design engineered to the system's own link budget is the only defensible approach. For applications where specifications are loose and margins are wide, a catalogue product may be entirely adequate. The case for custom engineering rests on the specificity of the requirement, not the prestige of the application.

Practical EDFAs achieve 4–6 dB noise figure; the theoretical floor set by quantum mechanics is 3 dB. That 1–3 dB gap above the quantum limit is where amplifier design either wins or loses OSNR budget. In a cascaded amplifier chain, noise figure compounds via the Friis formula — each stage adds ASE that no downstream amplifier can remove. A 1–2 dB per-amplifier NF improvement translates directly into meaningful OSNR margin at the receiver, often the difference between a link budget that closes and one that does not. Pump wavelength is a key design variable here. Forward 980 nm pumping achieves lower NF than 1480 nm pumping because it produces more complete population inversion at the signal input end of the erbium-doped fibre. This makes 980 nm forward pumping the preferred configuration for preamplifiers and noise-sensitive in-line stages, where NF is the binding constraint rather than output power.

Not completely. An optical amplifier restores signal power in the optical domain but does not perform 3R regeneration — it does not retime, reshape, or retransmit the signal. Accumulated ASE noise, chromatic dispersion, and nonlinear impairments are not corrected; they are carried forward. What optical amplification does provide — simultaneous amplification of all WDM channels without OEO conversion, full protocol and format transparency, and near-zero latency — makes it the preferred repeater technology for the vast majority of modern fibre-optic links. OEO regeneration at defined intervals remains necessary when accumulated impairments degrade the signal beyond what the receiver can tolerate. In practice, modern coherent DSP at the terminals has substantially extended the reach of optically amplified links, reducing the frequency at which OEO regeneration is required in long-haul systems. The two technologies are complementary, not mutually exclusive — the engineering question is where along the link OEO regeneration, if any, is needed.

An optical amplifier increases the power of an optical signal directly in the optical domain, without converting it to an electrical signal. This transparency preserves the signal's data rate, modulation format, and wavelength simultaneously — the amplifier has no knowledge of what it is amplifying. In practice, optical amplifiers are deployed in three distinct roles. A booster amplifier is placed immediately after the transmitter to maximise launch power into the first fibre span; output power is the priority here. An in-line amplifier sits at repeater sites along the route to compensate for accumulated span loss; it balances output power and noise figure. A preamplifier is placed just before the receiver to improve detection sensitivity; noise figure is the critical parameter, as any noise added at this stage directly degrades the receiver's OSNR. Each role carries different design priorities, which is why a single catalogue specification rarely optimises for all three simultaneously.

The primary advantages are protocol transparency, simultaneous multi-channel amplification, format agnosticism, low latency, and pay-as-you-grow capacity scalability. Protocol transparency means a single EDFA amplifies any signal within its gain band regardless of bit rate or modulation format — upgrading from 10G to 100G to 400G requires only terminal equipment changes, not new amplifiers in the field. WDM cost leverage is substantial: one amplifier simultaneously replaces N OEO regenerators, one per channel, at every repeater site. The cost advantage scales multiplicatively with channel count and span count. Latency is another axis: all-optical amplification adds sub-nanosecond delay per site versus microseconds of processing delay through each OEO regeneration node — a material difference for latency-sensitive traffic. Finally, once fibre and amplifiers are installed, additional WDM channels are activated by upgrading only transmit and receive terminals. The amplifier plant requires no modification, making capacity growth a terminal-side decision rather than a network-wide infrastructure change.

EDFA dominance originates from a physical coincidence: the erbium gain band aligns with the lowest-loss transmission window of standard silica fibre, the C-band (1530–1565 nm). This is not a design preference — it is a material property that makes EDFA the natural amplifier for the wavelengths fibre already propagates most efficiently. Three factors reinforce that foundational advantage. First, noise figure: practical EDFAs achieve 4–6 dB NF, approaching the 3 dB quantum limit, giving them the best OSNR performance of any lumped amplifier type. Second, gain dynamics: the millisecond upper-state lifetime of erbium ions means EDFA gain cannot respond to individual data bit transitions, making the gain effectively constant across all co-propagating WDM channels and eliminating inter-channel crosstalk. Third, ecosystem maturity: three decades of deployment have produced a deep supply base of pump lasers, gain-flattening filters, and control electronics optimised specifically for EDFA operation — a practical engineering advantage that compounds over time.