1. Introduction

Submarine EDFA repeaters do not respond to individual channels. They respond to total input power. Every repeater in a transoceanic chain is designed to receive a specific total optical power at its input, corresponding to the power spectral density (PSD) that results from a fully populated spectrum at the design operating point. Pull that total power down by leaving channels dark, and the amplifier enters a higher-gain regime where the noise figure worsens, gain tilt increases, and the generalized signal-to-noise ratio of every traffic-carrying channel degrades. The amplifier has no way to distinguish between "the spectrum is partially filled for a good reason" and "the system is broken." It amplifies whatever arrives, including its own spontaneous emission.

This is the loading problem, and it is not a corner case. Every submarine system goes through spectrum turn-up, from first light to full capacity, spanning months to years on commercial cables. Every traffic event, whether a capacity upgrade, a partial failure, or a maintenance action, temporarily changes how many channels the fiber carries. Without a mechanism to hold total power constant during these transitions, the amplifier chain degrades all deployed traffic every time an operator tries to add a new channel. On a cable carrying 20 Tb/s of live customer traffic, a 1.5 dB GSNR excursion caused by an unmanaged loading transition can force modulation-format fallback across dozens of channels simultaneously.

Amplified spontaneous emission (ASE) loading solves this by filling the unoccupied spectral slots with optically generated noise, injected at the terminal, so that the repeaters always see the same total power they were designed for. Two distinct implementations exist, continuous broadband ASE and channelized ASE, and the choice between them carries different implications for power management, nonlinear impairment, operational flexibility, and the ability to measure line performance during commissioning. A third option, the continuous-wave (CW) idler, remains relevant on legacy dispersion-managed cables and in specific power-holding roles on modern systems.

This article covers the physics, the mathematics, the architecture, the operational procedures, and the failure modes for all three approaches. It gives specific attention to the differences between legacy dispersion-managed cables and modern uncompensated D+ designs, to the interaction between loading and stimulated Raman scattering in C+L systems, to loading recovery through branching units and ROADMs, and to the division of loading responsibility in spectrum-sharing arrangements where multiple parties operate independent terminal equipment on one fiber pair.

2. Optical Loading and Power Spectral Density Definitions

Optical loading is the deliberate injection of non-traffic optical power into unoccupied frequency slots of a WDM spectrum, at a power spectral density matching that of traffic channels, so that the total optical power presented to every downstream amplifier remains at its design value regardless of how many channels carry live data. The injected power is measured in dBm per unit bandwidth.

Three quantities are routinely conflated in loading discussions, and separating them is the whole of the subject. Total output power (TOP) is the integrated optical power across the amplifier's full gain bandwidth, expressed in dBm. It is a single scalar that a repeater's control loop regulates. Power spectral density (PSD) is power per unit optical bandwidth, expressed in dBm/GHz or dBm/nm. It is the quantity that determines how a channel or an idler affects its neighbors nonlinearly. Per-channel power is the integrated power within one channel slot, expressed in dBm. It equals PSD multiplied by the slot bandwidth, which means two channels at the same per-channel power but different slot widths have different PSDs and therefore different nonlinear behavior.

2.1 Distinguishing the Adjacent Quantities

Occupied bandwidth against slot width. Occupied bandwidth B_occ is the spectral width a modulated signal actually fills, equal to the symbol rate multiplied by (1 + roll-off factor); slot width is the flexible-grid allocation reserved for that channel, always a multiple of 12.5 GHz per ITU-T G.694.1 and always at least as wide as B_occ. A 64 GBd signal with 0.06 roll-off occupies 67.8 GHz and is allocated a 75 GHz slot, leaving 7.2 GHz of guard within the allocation.

Power against power spectral density. Power is what a power meter reads; PSD is power divided by the bandwidth it occupies. Two channels at the same per-channel power but different symbol rates have different PSDs, and the higher-PSD channel imposes more nonlinear interference on its neighbors even though a power meter reads them as equal. Loading correctness is a PSD condition, never a per-channel power condition.

PSD referenced to symbol rate against PSD referenced to slot width. The same channel has two different PSD values depending on whether the denominator is B_occ or the slot width. A 400G channel at −0.5 dBm has a PSD of −18.8 dBm/GHz referenced to its 68 GHz occupied bandwidth, but −19.3 dBm/GHz referenced to its 75 GHz slot. Loading targets must state which reference applies, because the 0.5 dB difference is enough to accumulate into a visible spectral ripple across 100 spans if applied inconsistently between traffic and idler slots.

Idler against guard band. An idler carries power; a guard band deliberately carries none. Idlers hold the amplifier operating point; guard bands isolate spectral neighbors from filter roll-off or nonlinear interaction. Confusing them produces the specific field failure of an operator inserting guard bands where idlers were required, dropping total power below the design point.

3. The Amplifier Power Balance Problem

3.1 Total Output Power Regulation

A submarine EDFA repeater operates under total output power regulation. Pump power, erbium doping concentration, and the automatic control loop are all configured to deliver a fixed total output power, the sum of signal power and ASE power across the entire C-band (approximately 1,525 to 1,568 nm, or across C+L in wider-band systems). In a fully loaded system operating at a per-channel launch power that targets the nonlinear threshold, the total output power is the sum of all channel contributions at their design PSD. When a quarter of those channels are removed because three transponder shelves are unpopulated during initial turn-up, the amplifier still delivers the same total output power. That power is now concentrated over fewer channels, raising per-channel power above the nonlinear threshold and degrading the GSNR of every deployed channel.

The mechanism is gain compression. In the small-signal regime, where total input power is well below the saturation power, the amplifier gain is high, potentially 6 to 10 dB above the nominal operating point for 980 nm-pumped amplifiers and 4 to 7 dB above nominal for 1,480 nm-pumped designs. As more channels are added, each contributing power to the total input, the gain compresses back toward the design value. The amplifier's output power stays essentially constant across this compression range because pump power determines the envelope of available stored energy in the erbium inversion. What changes is how that energy is distributed: over more channels as the system fills, fewer channels as it empties.

The asymmetry between 980 nm and 1,480 nm pumping is worth understanding, because it changes the severity of an unloaded excursion. At 980 nm, the pump excites ions to the ⁴I_11/2 level, from which they decay non-radiatively to the metastable ⁴I_13/2 level. Because the ⁴I_11/2 level is essentially empty at all times, no stimulated emission at the pump wavelength competes with absorption, so pump light is almost fully absorbed even in the small-signal regime. This produces the larger 6 to 10 dB gain compression range. At 1,480 nm, the pump excites directly into upper Stark sublevels of ⁴I_13/2, where stimulated emission at the pump wavelength competes with absorption, limiting compression to 4 to 7 dB. A cable using 980 nm pumping therefore has more gain headroom available when a fault occurs, and correspondingly more gain excursion when loading is lost.