Ultralong-Haul Transmission in Submarine Optical Systems

A research-grade technical reference covering noise modeling, nonlinear physics, GSNR framework, and capacity design — with fully interactive formula calculators for design engineers.

OSNR & ASE Modeling Gaussian Noise Model GSNR Framework GAWBS Physics Shannon Capacity SDM Optimization Interactive Calculators

Submarine optical cable systems carry the majority of intercontinental data traffic, carrying the vast majority of all international data traffic across oceans. The engineering challenge — transmitting hundreds of terabits per second across transoceanic distances of 6,000 to 20,000 km using repeaters powered over a single high-voltage conductor — has no parallel in terrestrial networking. Every design decision carries irreversible financial and physical consequences, since deployed cable systems operate on the ocean floor for 25 years or more, far beyond the reach of ordinary maintenance.

The discipline of submarine optical transmission design is an optimization problem. Engineers must simultaneously balance optical signal-to-noise ratio (OSNR), fiber nonlinear effects, cable powering limits, dispersion management, and spectral efficiency — all under the constraint that repeaters cannot be accessed after deployment. This article provides a rigorous technical treatment of the physical models that govern system performance, the analytical frameworks used to predict and optimize capacity, and the transition from single-fiber optimization toward space division multiplexing (SDM) as the architecture of the current decade.

The treatment covers the entire signal impairment hierarchy: amplified spontaneous emission (ASE) noise from cascaded optical amplifiers, fiber Kerr nonlinearities modeled through the Gaussian Noise (GN) model, and the recently characterized guided acoustic-wave Brillouin scattering (GAWBS) effect. These three independent but interacting noise sources define the Generalized Signal-to-Noise Ratio (GSNR), which is the central figure of merit in modern open submarine cable characterization, standardized under ITU-T G.977.1.

Generalized signal-to-noise ratio (GSNR) is the ratio of received per-channel signal power to the total power of every additive impairment the coherent receiver cannot distinguish from noise: amplifier ASE, Kerr nonlinear interference, and guided acoustic-wave Brillouin scattering. GSNR is dimensionless, quoted in decibels, and referenced to the symbol rate rather than to a fixed optical bandwidth.

2.1 Distinctions From the Adjacent Quantities

Three quantities are routinely conflated, and each conflation changes the answer by more than a decibel. OSNR counts only amplifier ASE, measured in a fixed 0.1 nm reference bandwidth, and reports nothing about nonlinear distortion; a planning tool quoting 22 dB OSNR and a receiver reporting 19 dB SNR are measuring different things rather than disagreeing, as the analytical relationship between GOSNR and OSNR sets out. SNR at the receiver is referenced to the symbol rate and includes transponder implementation noise, so it sits below GSNR by the back-to-back penalty of the modem. GSNR sits between them: it is a property of the optical path alone, referenced to the symbol rate, and excludes transponder noise so that a wet plant can be specified independently of whoever terminates it.

Two further pairs matter. Reference bandwidth against symbol rate: OSNR is bandwidth-dependent and SNR is not, which is why the two only reconcile through an explicit conversion, treated at length in the OSNR fundamentals primer. Noise variance against noise power spectral density: the GN model produces a density that must be integrated over the channel bandwidth before it can enter the reciprocal sum above.

2.2 Units and Conversion Arithmetic

  1 / GSNR  =  Σi  1 / SNRi

  OSNRdB  =  SNRdB  +  10 · log10( Rs / 12.5 )

The arithmetic runs one way only. Convert each contribution to linear units, sum the reciprocals, invert, then convert the result back to decibels. Summing decibel values directly is the most common error in this calculation and always overstates the penalty.

2.3 Worked Instantiation

Take the reference case this article returns to throughout: an 8,000 km route, 100 repeaters at 80 km spacing, 0.150 dB/km attenuation, 18 dBm total output power across 100 channels, 4.5 dB noise figure, and 125 µm² effective area. The three contributions evaluate to SNR_ASE = 25.5 dB (linear 356.5), SNR_NL = 17.0 dB (linear 49.5), and SNR_GAWBS = 21.7 dB (linear 149.2). Their reciprocals sum to 0.002805 + 0.020203 + 0.006702 = 0.029709, and the inverse of that is 33.7, or 15.3 dB of GSNR. Nonlinear interference contributes 68.0% of the total noise variance, GAWBS 22.6%, and ASE 9.4% — this case runs above the nonlinear threshold, and the Section 7 optimum moves the split back toward ASE.

A repeatered submarine cable system consists of a small number of fundamental elements whose interaction governs all performance outcomes. Understanding each element — and precisely how it contributes to noise — is the necessary foundation before any formula can be applied meaningfully.

3.1 The Amplified Span

The fundamental building unit is an amplified span: a length of transmission fiber terminated at each end by an erbium-doped fiber amplifier (EDFA). Each repeater in a modern submarine cable houses multiple EDFAs — one per fiber pair direction — that compensate exactly for the span loss accumulated over the preceding fiber section. Span lengths in submarine systems typically range from 50 to 100 km, with the repeater spacing set at manufacture and fixed for the cable lifetime.

EDFAs amplify the signal through stimulated emission in the erbium-doped fiber, but they also generate amplified spontaneous emission (ASE) noise, which accumulates irreversibly with each repeater traversal. The noise figure (NF) of each EDFA — typically 4 to 6 dB in modern submarine repeaters — quantifies how efficiently signal power is added relative to noise power. A lower noise figure directly translates to a higher end-to-end OSNR for a given cable length and span configuration.

3.2 Fiber Types and Their Role

The transmission fiber used in submarine cables directly affects both the linear (ASE-dominated) and nonlinear noise budgets. Two fiber parameters govern system performance:

Effective area (A_eff): A larger effective area reduces the optical power density within the fiber core, proportionally reducing all Kerr nonlinear effects. Premium submarine fibers achieve effective areas of 125–150 µm², compared to 80–110 µm² for standard single-mode fiber. The nonlinear noise power scales approximately as 1/A_eff², so a fiber with 150 µm² effective area generates roughly 4× less nonlinear interference power than a fiber with 80 µm².

Attenuation coefficient (α): Lower attenuation means the signal arrives at each repeater with more power, improving OSNR. Modern ultra-low-loss submarine fibers achieve attenuation around 0.150 dB/km or below, compared to 0.18–0.20 dB/km for standard terrestrial fiber. In a 6,400-km transoceanic link with 80-km spans, reducing attenuation from 0.200 to 0.150 dB/km reduces span loss from 16 dB to 12 dB — a 4 dB improvement that propagates directly into OSNR.

The OSNR is the primary figure of merit for linear noise performance. It expresses the ratio of signal power to noise power integrated in a reference bandwidth — conventionally 0.1 nm (equivalent to 12.5 GHz at 1,550 nm). Every physical impairment in a submarine system is expressed in terms of its contribution to SNR degradation, making the OSNR the natural currency of design.

4.1 OSNR After a Single Amplifier

When a noise-free signal at power P_sⁱⁿ (per channel) enters an EDFA with gain G_s and noise figure NF, the OSNR at the amplifier output, measured in reference bandwidth B_ref, is:

  OSNR1,Bref  =  Pouts  /  (NF · Gs · h · ν · Bref)

                 =  Pins  /  (NF · h · ν · Bref)

When a noisy signal (with input OSNRⁱⁿ_Bref) enters the amplifier, the output OSNR degrades further. The inverse SNR contributions add linearly:

  1 / OSNRoutBref  =  1 / OSNRinBref  +  NF · h · ν · Bref / Pins

4.2 Cascade of N Amplifiers — Coarse OSNR Model

For a transoceanic link composed of N identical amplified spans, each contributing the same ASE noise, the end-to-end OSNR in dB scale at 1,550 nm in a 0.1-nm reference bandwidth takes the classical engineering form:

  OSNRdB  ≈  58  +  TOPdBm  −  10·log10(Nch)  −  SpanLossdB  −  NFdB  −  10·log10(N)

4.3 Signal Droop and the Generalized Droop Model

The coarse OSNR formula (3.3) overestimates performance because it treats signal input power into each EDFA as constant and equal to TOP/N_ch/SpanLoss. In reality, each EDFA generates ASE that fills the amplifier bandwidth and partially depletes the amplified signal. After many repeaters, the signal represents a progressively smaller fraction of total optical power — a phenomenon called signal droop.

The droop coefficient d_k for amplifier k is defined as the ratio of signal power to total output power. For a noise-free input at nominal power, the SNR_ASE° per channel per amplifier is:

  SNR°ASE  =  Pin1channel  /  (NF · h · ν · Δf)

  — where Δf is the channel spacing (symbol rate or grid spacing) —

  Relative droop per amplifier:    1/d  =  1  +  1/SNR°ASE

  In a cascade of N amplifiers, droops multiply:
  1/dtotal  =  ∏k=1..N (1  +  1/SNR°ASE,k)

When optical power is high enough, the refractive index of silica fiber becomes power-dependent through the Kerr effect (n = n₀ + n₂·I, where I is intensity). This intensity-dependent phase modulation generates new spectral components and introduces crosstalk between WDM channels. The principal Kerr effects relevant to submarine systems are self-phase modulation (SPM), cross-phase modulation (XPM), and four-wave mixing (FWM).

Modern coherent DSP-based receivers effectively compensate for deterministic, well-correlated nonlinear effects and treat the residual as a noise-like distortion. This observation motivates the Gaussian Noise (GN) model, which treats the aggregate nonlinear interference noise (NLI) as additive Gaussian noise with a variance that can be computed analytically.

5.1 The GN Model — NLI Noise Variance

For WDM transmission over a dispersion-unmanaged link of N identical spans, the GN model gives the NLI noise variance per channel as:

  σ²NLI,N  ≈  N1+ε  ·  (16/27)  ·  (γ · Pch · Leff)²
                 ·  [ π·|β₂|/α ]  ·  asinh[ (π²/4)·|β₂|/α · N²·B²WDM/Nch·Bch ]
                 ·  Pch,Rx

  Effective length:
  Leff  =  (1 − exp(−α·Lspan)) / α

5.2 NLI SNR

From the GN model NLI variance, the nonlinear SNR is:

  1 / SNRNL  =  σ²NLI  /  Pch,Rx

  Scaling Behavior:
  SNRNL  ∝  1 / P2ch        (doubles in dB for every 3 dB drop in power)
  SNRNL  ∝  A2eff           (doubling Aeff improves NL SNR by 6 dB)
  SNRASE ∝  Pch            (improves linearly with power, +1 dB/dB)

Alongside ASE and Kerr nonlinear noise, a third distinct noise source has been established as measurable in transoceanic submarine systems: guided acoustic-wave Brillouin scattering (GAWBS). Unlike stimulated Brillouin scattering (SBS), GAWBS is a spontaneous, thermally driven acousto-optic effect. Signal photons propagating along the fiber interact inelastically with thermally excited transverse acoustic phonons of the fiber structure, scattering in the forward direction with discrete frequency shifts between approximately 20 and 1,000 MHz corresponding to the resonant frequencies of the transverse acoustic modes.

The scattered field mixes with the signal field as a modulation noise, inducing both phase and polarization distortions. Over transoceanic distances, these distortions accumulate and, when modulated signals are transmitted, the net effect is modeled as additive Gaussian noise.

6.1 GAWBS Noise Model

The variance of GAWBS noise is proportional to signal power and propagation distance. The governing relation is:

  σ²GAWBS  =  ΓGAWBS(fiber)  ·  Distance[Mm]  ·  P

  GAWBS coefficient empirical law (from multi-source experiments):
  ΓGAWBS,dB  ≈  −9.8 dB  −  10·log10(Aeff[µm²])     [per 1,000 km]

  GAWBS SNR in signal channel band:
  SNRGAWBS,dB  ≈  9.8 dB  +  10·log10(Aeff[µm²])  −  10·log10(Distance[Mm])

The Generalized Signal-to-Noise Ratio (GSNR) is the central performance metric for modern open submarine cable systems, standardized in ITU-T G.977.1. It aggregates all relevant noise contributions — ASE, nonlinear interference, and GAWBS — into a single quantity that characterizes the cable's delivered signal quality independent of the terminal modem.

7.1 GSNR Definition — Generalized Droop Model

Rather than simply summing noise variances (additive model), the generalized droop (GD) model accounts for the fact that each noise source depletes the undistorted signal, causing further droop for subsequent sources. The GD model expresses aggregation through multiplicative inverse droop terms:

  (1 + 1/GSNRB) = (1 + 1/SNRASE,B°) · (1 + 1/SNRNL,B°) · (1 + 1/SNRGAWBS,B°) · ...

  Simplified additive approximation (valid when all SNR terms >> 1 dB above GSNR):
  1/GSNR  ≈  1/SNRASE  +  1/SNRNL  +  1/SNRGAWBS

7.2 Optimum Power and the Nonlinear Threshold

At the nonlinear threshold (NLT), GSNR is maximized with respect to launch power. Setting ∂(1/GSNR)/∂P = 0 yields the condition that NLI noise variance equals exactly half the ASE noise variance:

  At optimum power:    2/SNRNL(Popt)  =  1/SNRASE(Popt)

  i.e., SNRNL  =  3 dB above SNRASE  (NLI variance = 0.5 × ASE variance)

  Joint SNRASE+NL at optimum:   3/(2·SNRASE)  →  degraded by 1.76 dB vs pure ASE

  Optimum power (3rd-power law):
  (Popt / P°)³  =  (1/2) · (SNRNL° / SNRASE°)

  GSNR at NLT:
  1/GSNRNLT  =  1/SNRGAWBS  +  (3/2) · 1/SNRASE(PNLT)

Shannon's channel capacity theorem provides the ultimate bound on information transmission. For a channel with additive Gaussian noise, the maximum achievable information rate C (in bits per second per channel use) is:

  C  =  log₂(1 + SNR)    [bits/symbol/polarization]

  For dual-polarization coherent WDM, channel capacity in Gb/s:
  Cchannel  =  2  ·  Bs[GHz]  ·  log₂(1 + GSNRlinear)    [Gb/s]

  Cable capacity (N fiber pairs, spectral occupancy χ, bandwidth BEDFA):
  Ccable  =  NFP  ·  2  ·  χ  ·  BEDFA  ·  log₂(1 + SNRTOT(NFP))    [Tb/s]

The most profound insight from the Shannon-based analysis of submarine cable capacity is that concentrating electrical pump power to maximize the capacity per fiber returns less capacity per watt than distributing it. Shannon's law tells us that capacity grows only logarithmically with SNR, while it scales linearly with the number of parallel spatial paths. This observation drives the shift to SDM architectures in modern submarine cables.

9.1 The Power Efficiency Argument

Consider a cable with fixed total electrical power budget feeding N_FP fiber pairs through a shared pump architecture. Doubling the total optical power per fiber to push capacity further yields at most +1 b/s/Hz/polarization increase in spectral efficiency (operating in the ASE-dominated regime). But doubling the number of fiber pairs at the same power per fiber doubles cable capacity — a far more efficient allocation.

  C(NFP)  =  NFP  ·  2  ·  χ  ·  BEDFA  ·  log₂[ 1 + SNRTOT(NFP) / pen ]

  Pump sharing model — power per amplifier:
  Pp,amp(NFP)  =  ηmux · Ppump,rep / (2·NFP)

  Result: doubling N_FP while keeping total pump power constant
  → amplifier output power decreases by ~3 dB
  → ASE variance doubles (×2), NLI variance decreases by 4× (×0.25)
  → net total noise change depends on operating regime
  → cable capacity increases approximately linearly with N_FP

The capacity gains from SDM are substantial but saturate at high fiber pair counts, because the ASE noise increases with each additional fiber pair (lower power per amplifier) and eventually dominates. The optimal fiber pair count depends on link length, target reach, and GSNR requirements. Shorter links can support more fiber pairs at higher GSNR, while very long transoceanic systems have tighter constraints.

The open submarine cable model has changed how capacity is deployed and upgraded. Unlike turnkey systems — where the cable supplier also provides the terminal transmission equipment (SLTE) including coherent transceivers — an open cable separates the wet plant (fiber, repeaters, branching units) from the terminal equipment. This allows multiple suppliers to provide SLTE independently, enabling competitive capacity upgrades over the cable lifetime.

Open cables require a standardized metric that characterizes cable performance independent of any specific transceiver technology. The GSNR fulfills this role. Once the GSNR of an open cable section is known as a function of wavelength, any SLTE provider can compute their achievable capacity by combining the cable's GSNR with their transceiver's implementation characteristics.

10.1 GSNR Measurement and the Open Cable Budget

Measuring GSNR on a deployed cable requires calibrated coherent test transponders and ASE loading to fill the optical amplifier bandwidth. The measurement procedure, standardized in ITU-T G.977.1, derives GSNR from the total external SNR (SNR_EXT) and an estimate of implementation noise SNR_i:

  1/GSNR  =  1/SNREXT  −  1/SNRi

  where:
  SNREXT — total SNR measured by the test transponder (from Q² conversion)
  SNRi  — implementation noise of the test transponder (back-to-back calibration)
  GSNR  — cable contribution only (ASE + NLI + GAWBS), modem-agnostic

  Electrical SNR with modem:
  1/eSNR  =  ε · (1/SNRASE  +  1/SNRNL  +  1/SNRm)

11.1 Sensitivity Relations

For practical design, the sensitivity of GSNR and capacity to each design parameter governs the outcome. The following table summarizes the most important relations derived from the analytical models above:

11.2 Modulation Format Requirements

Despite the remarkable progress captured in the models above, submarine optical systems face a set of challenges that define the research frontier for the coming decade.

12.1 Cable Powering Constraints

The single greatest constraint on cable capacity growth is electrical power delivery. All repeater amplifiers are fed by a single high-voltage DC conductor running through the cable from the shore stations. The maximum current is limited by resistive losses in the cable conductor, while the maximum voltage is limited by the risk of transient surge current during shunt faults. As fiber pair counts increase in SDM cables, the pump power demand grows proportionally, making the powering constraint increasingly binding. This is why efficient pump-sharing architectures — where a small pool of pump lasers distributes power to many amplifiers — set the practical limit on fiber pair count. Increasing the power feed equipment (PFE) voltage from legacy 15 kV to 18 kV or higher provides some relief but is technically challenging.

12.2 GAWBS as an Irreducible Floor

Because SNR_GAWBS is independent of launch power, it cannot be improved through any optical power optimization. The only mitigation paths are: using fiber with larger effective area (GAWBS ∝ 1/A_eff), reducing cable length, or applying digital post-processing to characterize and subtract the deterministic GAWBS spectrum — a technique demonstrated experimentally but not yet deployed at scale. As GSNR budgets are pushed ever closer to the Shannon limit, GAWBS will take an increasing fraction of total noise.

12.3 Wideband Amplification

Extending the amplification bandwidth beyond the C-band into L-band (adding 4–5 THz of additional spectrum) is an active area of development. C+L-band submarine systems can in principle double capacity per fiber pair without any increase in fiber count or cable infrastructure. However, L-band amplifier designs face challenges including gain equalization across 10 THz combined bandwidth, nonlinear interactions between C and L bands through stimulated Raman scattering (which causes systematic power tilt), and repeater design complexity.

12.4 Multicore Fiber for SDM

Weakly coupled multicore fiber (MCF), where multiple fiber cores are placed within a single 125-µm-diameter cladding, offers a path to raising fiber pair count severalfold per cable cross-section without proportionally increasing cable diameter. The inter-core crosstalk in weakly coupled MCF is low enough to be modeled as an additional SNR penalty proportional to propagation distance, analogous to GAWBS. Experiments have demonstrated that the GAWBS coefficient per core in weakly coupled MCF is similar to that of standard single-mode fiber, suggesting no fundamental penalty from the multi-core geometry.

Submarine optical transmission design rests on a small number of foundational physical models that connect fiber and amplifier parameters to system performance with quantitative precision. The coarse OSNR model (Formula 4.3) provides the starting point — accurate to within 0.5 dB for most engineering purposes. The GN model (Formula 5.1) adds the nonlinear dimension, showing why increasing launch power has diminishing and eventually counterproductive returns. The GAWBS model (Formula 6.1) adds a fixed noise floor proportional to distance and inverse effective area. The generalized droop framework (Formula 7.1) combines all three into the GSNR, which is the standardized metric for open cable characterization per ITU-T G.977.1.

Together, these models reveal the central insight of SDM design: cable capacity scales far more efficiently with spatial parallelism (fiber pairs, cores) than with per-fiber optical power. The Shannon capacity formula (Formula 8.1) quantifies the exact trade-off: each doubling of fiber pairs at constant total cable power yields roughly a 1.7–1.9× increase in total cable capacity, while doubling optical power per fiber yields at most +1 b/s/Hz/polarization. This fundamental asymmetry will continue to drive submarine cable architecture toward higher fiber pair counts, shared pump architectures, and wideband amplification through the coming decade.

• [1] ITU-T Recommendation G.977.1 — Characteristics of optically amplified optical fibre submarine cable systems: Open cable systems, ITU-T Study Group 15.
• [2] ITU-T Recommendation G.972 — Definitions of terms relevant to optical fibre submarine cable systems, ITU-T Study Group 15.
• [3] ITU-T Recommendation G.978 — Characteristics of optically amplified optical fibre submarine cable systems, ITU-T Study Group 15.
• [4] ITU-T Recommendation G.976 — Test methods applicable to optical fibre submarine cable systems, ITU-T Study Group 15.
• [5] P. Poggiolini, G. Bosco, A. Carena, V. Curri, Y. Jiang, F. Forghieri, The GN-Model of Fiber Non-Linear Propagation and Its Applications, Journal of Lightwave Technology.
• [6] J.-C. Antona et al., Analysis of 34 to 101 GBaud Submarine Transmissions and Performance Prediction Models, Optical Fiber Communication Conference.
• [7] M. A. Bolshtyansky, Impact of Spontaneous Guided Acoustic-Wave Brillouin Scattering on Long-Haul Transmission, Optical Fiber Communication Conference.
• [8] E. Rivera Hartling et al., Design, Acceptance and Capacity of Subsea Open Cables, Journal of Lightwave Technology.
• [9] P. Serena, A. Carbo Meseguer, F. Poli, A. Bononi, J.-C. Antona, Scaling Properties of Guided Acoustic-Wave Brillouin Scattering in Single-Mode Fibers, Optics Express.
• [10] OIF Implementation Agreement for 400ZR Interfaces, Optical Internetworking Forum.
• [11] SubOptic Open Cables Working Group, Translating GSNR into Capacity, SubOptic Association.