The Decay of Light

In a standard single-mode fiber (G.652), attenuation at $1550\,\text{nm}$ is approximately $0.2\,\text{dB/km}$. After $100\,\text{km}$, the signal power drops by $20\,\text{dB}$ (99%). Without amplification, a signal sent from New York would be mathematically non-existent long before it reached the Mid-Atlantic Ridge.

EDFA: The Heart of the Repeater

The Erbium-Doped Fiber Amplifier (EDFA) revolutionized subsea comms by allowing all WDM channels to be amplified simultaneously in the optical domain.

A short segment of fiber is doped with Erbium ions ($\text{Er}^{3+}$). When 'pumped' with a high-power laser at $980\,\text{nm}$ or $1480\,\text{nm}$, the erbium ions are excited to a higher energy state. When a signal photon passes through, it triggers Stimulated Emission, causing the ions to drop back to a ground state while releasing a new photon identical to the original.

Raman Amplification

While EDFAs use a dedicated doped fiber, Raman Amplification uses the transmission fiber itself as the gain medium. This relies on Stimulated Raman Scattering (SRS).

A high-power pump signal is sent into the fiber (often in the reverse direction). When pump photons collide with silica molecules, they lose energy to molecular vibrations (phonons) and are re-emitted as lower-frequency photons that match the data signal's frequency, effectively boosting it.

The Raman frequency shift in silica fiber.

Spectral Tilt and OSNR

In a cable with 100+ repeaters, even a $0.1\,\text{dB}$ gain imbalance per repeater accumulates into a $10\,\text{dB}$ 'tilt' across the spectrum. High-frequency channels might see massive gain while lower frequencies vanish into the noise floor.

Conclusion

The physics of subsea repeaters is what transforms a simple strand of glass into a global neural network. By manipulating erbium ions and Raman shifts miles below the surface, we effectively negate the laws of attenuation and keep the world connected.

Pump Laser Redundancy and Reliability Engineering

The EDFA within a subsea repeater is only as reliable as its pump laser module. Pump lasers at $980\,\text{nm}$ or $1480\,\text{nm}$ must deliver tens to hundreds of milliwatts of continuous optical power for the operational lifetime of the cable — typically 25 years. Achieving this reliability target requires a multi-layered redundancy architecture that spans from the semiconductor die level up to the module packaging.

The most common configuration is N+1 cold standby: each repeater houses four pump lasers but only three are active at any given time. The fourth is held in reserve, its laser junction biased below threshold. If a primary pump degrades — detected via monitor photodiodes tracking the output power and drive current — the supervisory control system switches to the standby unit within milliseconds. This switching is accomplished by redirecting the pump current via a fail-safe relay network that defaults to the backup path even in the event of a control electronics failure.

Beyond simple redundancy, modern repeaters employ polarization multiplexing of pump light. Each pump laser emits linearly polarized light; by combining two orthogonally polarized pumps through a polarization beam combiner, the effective pump power into the erbium-doped fiber is nearly doubled without exceeding the catastrophic optical damage threshold of the laser facet. This architecture also provides built-in redundancy — if one polarization arm fails, the other continues to provide at least half the required pump power, preventing a complete service outage.

Thermal management is equally critical. Each pump laser generates waste heat on the order of 1–2 W, and the cumulative thermal load inside a repeater housing can reach 20–30 W. The repeater housing is in direct contact with seawater at 2–4°C at depth, which provides an excellent heat sink. A thermal conduction path using beryllium oxide or aluminum nitride ceramics carries heat from the laser submount to the titanium pressure housing. Finite element analysis of the thermal gradient across this path is a standard part of the qualification process, ensuring the laser junction temperature never exceeds 50°C even under worst-case operating conditions.

Gain Transient Control in Dynamic WDM Networks

When a subsea cable is part of a meshed terrestrial-submarine network, the number of WDM channels entering a repeater can change suddenly — for example, when a terrestrial fiber cut triggers automatic protection switching and channels are re-routed. The EDFA must handle these transient events without allowing the surviving channels to experience power surges or dropouts. This is the domain of gain transient control.

Consider a repeater amplifying 80 WDM channels at steady state. If 40 channels are abruptly dropped (e.g., a protection event upstream), the total input power to the EDFA drops by approximately 3 dB. The EDFA gain medium, having a finite excited-state lifetime of approximately 10 ms in erbium, cannot instantaneously adjust its inversion level. In the microseconds following the channel drop, the surviving channels suddenly see all of the available pump power, leading to a gain burst that can increase their output power by 6–8 dB. Such a surge can trigger nonlinear effects in the downstream fiber or even damage the receiver optics at the far end.

To counter this, subsea repeaters implement fast pump power control using feed-forward and feedback loops. The input tap coupler monitors the total input power via a photodiode; when a drop is detected, the pump laser drive current is reduced proportionally within microseconds, before the erbium inversion can change significantly. This feed-forward path provides the primary transient suppression. A slower feedback loop, sampling the output power, then fine-tunes the pump power to restore the exact target gain.

The time-dependent gain expression where $N_2$ and $N_1$ are the excited and ground state populations of erbium ions along the fiber length $L$.

Advanced transceivers at the terminals also participate in transient mitigation through fast power equalization. When a channel drop is detected, the transmitters of the surviving channels can temporarily reduce their launch power in coordination with the repeater control system, a technique known as bandwidth-mediated gain control. The combination of optical-level pump control and electronic-level transmitter coordination ensures that even in the most dynamic network conditions, the OSNR of every channel remains within specification.