Encoding Light — How Fiber Optics Carry the Internet Across Oceans

Article 2 of 8 — Information Infrastructure: The Physical Internet

What Actually Carries the Internet

When a user in São Paulo opens a webpage hosted in Frankfurt, the request does not travel through “the cloud” in any meaningful metaphorical sense. It travels as pulses of infrared light through a strand of doped silica glass roughly 125 micrometers in diameter — narrower than a human hair — buried in armored cable on the seabed of the Atlantic. The entire transaction, comprising perhaps a few hundred kilobytes of TCP/IP packets, transits as photons modulated at frequencies near 193 terahertz, amplified every 70 to 100 kilometers by units powered remotely from shore through a copper conductor wrapped around the same glass strand.

This article describes how that physical process works. It is the second installment in the Information Infrastructure — The Physical Internet series. The first established the global topology of the submarine cable system; this one descends into a single fiber pair and explains the physics, engineering, and architectural choices that determine how much of the world’s data any given cable can carry, and what an adversary destroys when they sever one. The logical routing layer — the protocols that decide which cable a packet takes — is treated separately in BGP Routing.

The Physics of Total Internal Reflection

A fiber optic cable is, at its core, a precision-engineered waveguide that exploits a phenomenon described by Snell’s Law: when light passes from a medium of higher refractive index into one of lower refractive index at a sufficiently shallow angle, it does not refract through the boundary — it reflects entirely back into the denser medium. This is total internal reflection.

A telecommunications fiber consists of two concentric layers of silica glass. The inner core — for long-haul fiber, approximately 9 micrometers in diameter — is doped with germanium to raise its refractive index slightly above that of the surrounding cladding. Any photon entering the core within a narrow range of angles (the “acceptance cone”) strikes the core-cladding boundary at an angle shallow enough to reflect rather than escape. The photon continues bouncing along the length of the fiber, confined within the core, for kilometers.

Data is encoded onto this confined light by modulating one or more of its properties at the transmitter: amplitude (how bright the pulse is), phase (where in its wave cycle the light is at a given instant), polarization (the orientation of the electromagnetic field), or wavelength (the color of the light). Modern long-haul systems modulate all of these simultaneously. The operating wavelengths sit in the infrared C-band (1530–1565 nm) for long-haul DWDM systems, with 1550 nm chosen because silica fiber exhibits its lowest attenuation at this wavelength — approximately 0.2 decibels per kilometer, which is to say that roughly 95% of the light’s power survives every kilometer of glass.

That attenuation figure is the constraint around which everything else in the system is designed.

Single-Mode vs. Multi-Mode — Why Submarine Cables Use a Fiber Thinner Than a Hair

Not all fiber is built the same way, and the distinction matters strategically because it determines what kind of cable can cross an ocean.

Multi-mode fiber (MMF) has a relatively wide core, 50 to 62.5 micrometers, which permits light to propagate along many slightly different geometric paths through the core — different “modes.” Each mode traverses a marginally different distance per unit of cable length, so pulses launched together arrive at the far end smeared in time. This is modal dispersion, and it limits multi-mode fiber to roughly 550 meters at 10 gigabits per second over OM3 or OM4 grades. Multi-mode fiber is confined to data centers, campus networks, and short building runs.

Single-mode fiber (SMF) has a core only ~9 micrometers across, narrow enough that the geometry permits only one propagation mode. With only one path through the core, modal dispersion vanishes. Combined with optical amplification, single-mode fiber can carry signals across entire oceans. All transoceanic submarine cables use single-mode fiber exclusively, manufactured to the ITU-T G.652 standard or one of its successors. (Confidence: High — this is a universal architectural fact of the global cable system.)

The strategic consequence: the fiber inside an 8,000-kilometer trans-Pacific cable is functionally identical, at the level of the glass, to the fiber inside a 2-kilometer terrestrial loop. What differentiates a submarine cable system is everything wrapped around the fiber — the amplifiers, the armor, the power feed, the modulation electronics at each end. Those, not the glass, are the bottlenecks and the targets.

Wavelength Division Multiplexing — How Multiple Data Streams Share One Fiber

A single fiber strand carries far more than a single data stream. The mechanism is Wavelength Division Multiplexing (WDM): multiple lasers, each tuned to a slightly different wavelength, inject their signals into the same fiber simultaneously, and at the receiver an optical demultiplexer separates them by color. Each wavelength is an independent data channel.

Two grades exist. CWDM (coarse) spaces channels 20 nm apart and supports 8 to 18 channels per fiber — adequate for metropolitan and enterprise applications but coarse and inefficient for long-haul. DWDM (dense) packs channels much more tightly under the ITU-T G.694.1 standard channel grids: a 100 GHz grid yields 40 to 48 channels in the C-band, a 50 GHz grid yields 80 to 96 channels, and modern systems extending into the adjacent L-band (combined C+L operation) push beyond 192 channels per fiber on commercial 2024-2025 deployments.

Multiply channels by per-channel rate and the numbers escalate quickly. Per-wavelength commercial throughput has progressed from 100 gigabits per second (2010s mainstream) to 400 Gbps and now 800 Gbps in mainstream 2024-2025 deployments, with 1.6 Tbps per wavelength at the commercial frontier. (Confidence: High — vendor announcements and operator filings.) The Japanese National Institute of Information and Communications Technology (NICT) demonstrated 301 Tb/s over standard commercially available single-mode fiber in a laboratory environment in 2024, illustrating the headroom that remains as encoding techniques continue to improve. (Confidence: High — primary NICT press release, 2024.)

The decisive analytic point: capacity in a fiber is not a physical given. It is a function of the transmitter and receiver electronics. The glass has been in the seabed since 2017 in many cases; the capacity carried on that glass triples or quadruples without anyone visiting the cable, simply by upgrading the terminal equipment on shore.

Coherent Optical Transmission — The Encoding Innovation That Unlocked Terabit Capacity

The leap from 10 Gbps to 800 Gbps per wavelength rests on a single architectural shift: coherent optical transmission.

Earlier optical systems used direct detection: the receiver simply measured how bright the incoming light was — intensity, on or off, the optical equivalent of Morse code. Coherent receivers, by contrast, measure both the amplitude and the phase of the incoming optical signal by mixing it with a reference laser (“local oscillator”) at the receiver. Phase information is recovered, polarization can be separately decoded, and the resulting electrical signal is processed by high-speed Digital Signal Processing (DSP) that compensates electronically for chromatic dispersion, polarization mode dispersion, and various nonlinear fiber effects that previously had to be managed by costly inline optical components or accepted as hard distance limits.

Coherent detection enables higher-order modulation formats in which each transmitted symbol carries multiple bits:

• QPSK (Quadrature Phase-Shift Keying): 4 states per symbol, 2 bits per symbol.
• 16-QAM (Quadrature Amplitude Modulation): 16 states, 4 bits per symbol.
• 64-QAM: 6 bits per symbol.
• 256-QAM: 8 bits per symbol — frontier for short-reach metro applications.

Combined with polarization multiplexing — using both horizontal and vertical polarization states of the light as independent channels — coherent systems extract roughly an order of magnitude more bits from each hertz of optical spectrum than direct-detection systems did.

The trade-off is reach. Higher-order QAM constellations crowd more bits into the symbol space, making each individual symbol more vulnerable to noise. They require a higher OSNR (Optical Signal-to-Noise Ratio) at the receiver, which is harder to maintain over long spans. Submarine operators therefore tune the modulation format to the route: 16-QAM might suffice on a short trans-Atlantic span; longer trans-Pacific routes step down to QPSK to preserve OSNR margin. The first 400G trans-Atlantic transmission ran over MAREA in December 2018 using the Acacia AC1200 coherent module. (Confidence: High — Microsoft/Facebook/Acacia press release, December 2018.) By 2024-2025, 400G and 800G per wavelength are mainstream commercial offerings.

Amplification at Ocean Depth — EDFA and the 25-Year Unmaintained Requirement

If a fiber loses 0.2 dB per kilometer of light, then over 5,000 km of trans-Atlantic cable the signal arriving at Europe would be 10²⁰ times weaker than what left Virginia. Coherent receivers, however sensitive, cannot recover signal from noise that overwhelming. The solution is in-line amplification.

The dominant technology is the Erbium-Doped Fiber Amplifier (EDFA). A short segment of fiber doped with erbium ions is “pumped” by a separate laser at 980 nm or 1480 nm, energizing the erbium ions into an excited state. When the weak DWDM signal at ~1550 nm passes through this excited segment, it triggers stimulated emission: the erbium ions release their stored energy as photons at the same wavelength, phase, and direction as the incoming signal. The signal is amplified — and critically, all DWDM wavelengths in the C-band are amplified simultaneously, without per-channel demultiplexing.

This is the engineering enabler of modern submarine cables. Without optical-domain amplification, every wavelength would need to be individually demultiplexed, electronically regenerated, and re-injected at each amplification site — impossible on the seabed.

Submarine repeaters (the units containing the EDFAs) are spaced approximately 70 to 100 kilometers apart. The near-constant deep-sea temperature of around 4°C is an underappreciated engineering advantage: stable temperature means stable gain, and the cable designer can budget more aggressively. Terrestrial long-haul amplifier spacing is more variable (40-120 km), driven by accessibility constraints rather than purely by physics.

The hardest engineering requirement: submarine EDFAs must operate unmaintained for the cable’s 25-year design life. There is no maintenance window on the abyssal plain at 4,000 meters. Power for these repeaters is delivered as DC current from shore via a copper conductor inside the cable, at voltages reaching 15,000 volts on the longest routes — high enough to compensate for the resistive losses over thousands of kilometers of copper. Some systems supplement EDFAs with Raman amplification, which exploits scattering effects in the transmission fiber itself rather than in a dedicated erbium segment. (Confidence: High — standard submarine cable system design practice, documented in ITU-T standards and operator filings.)

Space Division Multiplexing — The Current Generation Scaling Architecture

There is a physical limit to how much information can be modulated onto a single fiber pair. The Shannon limit of an optical channel, accounting for OSNR achievable in real systems, is approached by current coherent + DWDM systems within a factor of two or three. To grow capacity further, operators have shifted axis: instead of pushing one fiber harder, they bundle more fibers into a single cable sheath. This is Space Division Multiplexing (SDM).

The economics matter. Each additional fiber pair adds capacity roughly linearly, while the dominant cost — laying the cable on the seabed — is paid once. The shift to high-fiber-count cables is therefore the defining capacity story of the late 2020s.

Cable	Fiber Pairs	Design Capacity	Status
MAREA	8	200+ Tbps	In service
Grace Hopper (Google)	16	352 Tbps	In service
Dunant (Google)	12	~250 Tbps	In service
Hawaiki Nui	12	240 Tbps	RFS 2025
SEA-ME-WE-6	10	126 Tbps	RFS Q1 2025
JUNO (trans-Pacific)	20	350 Tbps	Under development
TPU (Taiwan-Philippines-US)	20	260 Tbps	RFS 2025-2026 (first MCF deployment)
Anjana (trans-Atlantic)	24	480 Tbps	Under construction
Medusa (Mediterranean-Atlantic)	24	480 Tbps	RFS ~2025

A critical distinction: these are design capacities — the engineered maximum if every fiber pair is fully lit with the current generation of terminal electronics. Lit capacity — what is actually activated at any given moment — is typically a fraction of design capacity. Operators provision wavelengths incrementally as customer demand and revenue grow, and the design capacity itself rises over time as coherent transceivers improve. A cable built in 2020 with a “200 Tbps” design figure may carry 400+ Tbps a decade later, with no physical change to the cable. (Confidence: High — consistent with operator practice across SubCom, HMN Technologies, NEC, and Alcatel Submarine Networks supply.)

Error Correction — How FEC Makes 800G Transoceanic Transmission Viable

Long fibers attenuate. Amplifiers, while restoring signal power, add their own noise — amplified spontaneous emission (ASE), the optical equivalent of microphone hiss. Over a transoceanic cascade of 50 to 70 EDFA stages, the accumulated ASE noise produces a raw bit error rate that would be entirely unusable for data — perhaps one error in every thousand or ten thousand bits transmitted, far worse than Ethernet’s 10⁻¹² target.

The bridge between unusable raw transmission and reliable user-grade throughput is Forward Error Correction (FEC). FEC algorithms add redundant parity bits to the data stream at the transmitter, structured so that the receiver can detect and correct errors mathematically without retransmission. The cost is overhead; the benefit is coding gain, expressed in decibels — effectively, how much noise the link can tolerate while still delivering error-free data to the user.

The progression in submarine systems is significant:

• ITU-T G.975 (baseline, 1996): Reed-Solomon RS(255,239) — 239 data bytes plus 16 parity bytes per codeword, coding gain approximately 6 dB.
• ITU-T G.975.1 (Enhanced FEC, 2004): coding gains of 8 to 10+ dB, enabling longer spans or higher spectral efficiency on existing cables.
• Soft-Decision FEC (SD-FEC): modern submarine systems use proprietary soft-decision algorithms achieving coding gains in excess of 11 dB. SD-FEC makes 400G and 800G transoceanic transmission commercially viable.

The analytic point: an apparently incremental innovation in mathematics — better codes, more efficient decoders — translates directly into multiplied capacity on the seabed. The cable does not change. The DSP firmware on shore changes. (Confidence: High — ITU-T standards documents and vendor technical briefs.)

Submarine Cable Architecture — The System Around the Fiber

A complete submarine cable system, beach-to-beach, is a stack of distinct engineering layers:

1. Armored nearshore segments — heavy double or single armor for the first kilometers from each coast, where fishing trawls, anchors, and tidal scour pose the greatest risk.
2. Lightweight mid-span cable — for the deep ocean floor, where physical risk is lower and weight matters for ship deployment.
3. Repeater units every 70–100 km — pressure-housed cylinders containing EDFAs and pump lasers, spliced inline with the cable.
4. Branching units (BUs) — passive or active optical splitters that allow a single cable trunk to land at multiple stations.
5. Landing stations — coastal facilities housing the Power Feed Equipment (PFE) that pushes 15,000 V DC down the cable to the repeaters, plus the coherent transceivers and DSP that constitute the actual capacity of the system.

The strategic geography of a cable system is determined less by the route of the glass than by the location of these landing stations and the terrestrial backhaul connecting them to the internet’s interconnection points. Sever the glass and the system has redundancy; disable the landing station’s power feed and a 480 Tbps cable becomes a 480 Tbps inert filament. This asymmetry between physical and functional vulnerability is treated at length in Economic Chokepoints — Coercive Statecraft.

What This Means for Infrastructure Security — Why Cutting One Cable Can Matter More Than Cutting Ten

The technical architecture described above produces an outcome that consistently surprises non-specialist commentary on cable incidents: capacity is wildly unevenly distributed across the cable system.

A modern 24-fiber-pair cable like Anjana or Medusa carries roughly 480 Tbps of design capacity. A first-generation cable built in 2001 with two or four fiber pairs carries perhaps 5 to 20 Tbps. Both are “cables” in the dataset, both appear as a single line on TeleGeography’s map, but the loss of one Anjana represents the loss of capacity equivalent to 25 to 100 of the older cables. Cable counts therefore mislead. What matters is the capacity at risk, not the line count. (Confidence: High — direct corollary of design capacity figures above.)

This shapes the security picture across three threat surfaces:

• Baltic Sea incidents (2023-2025): Cables severed in the Gulf of Finland and surrounding waters were largely older, lower-capacity systems serving regional traffic. The physical damage was real and the geopolitical signal was loud, but the bit-per-bit impact on global capacity was modest. See Baltic Sea Cable Incidents 2023-2025.
• Red Sea cuts (2024): The Red Sea is a chokepoint of unusual capacity density — multiple cables, several of them high-fiber-pair next-generation systems, share a narrow corridor. A single anchor incident there imposed a measurable hit on Europe-Asia capacity for weeks. See Red Sea Cable Cuts 2024.
• Trans-Pacific corridor: The next generation (JUNO, TPU, and successors) concentrates very high design capacity into a small number of cables crossing waters increasingly contested by Chinese maritime operations. Loss of two or three of these cables would be disproportionate to a “two of N” count.

This is the Dual-Use Technology dimension of the cable system: the same SDM and coherent-optics innovations that make global cloud services affordable also concentrate strategic value into fewer, denser physical assets.

Strategic Implications

1. Capacity is software-defined on hardware-deployed glass. A cable built in 2020 can multiply its lit capacity over its 25-year life with no physical intervention, simply through upgrades to shore-side coherent transceivers and DSP firmware. Threat assessments based on cable age understate current capacity.

2. Counting cables is the wrong metric. Capacity-at-risk per route, weighted by fiber pair count and modulation generation, is the analyst’s metric of interest. A single 24-pair cable can dominate a corridor’s capacity even alongside a dozen older systems.

3. Landing stations are softer targets than the glass. The cable is buried, armored, redundant, and in international waters; the landing station is a fixed terrestrial facility with addressable power feed equipment. Functional denial does not require severing fiber.

4. The amplification supply chain is a concentrated dependency. Submarine EDFAs, pump lasers, and pressure housings are produced by a small number of suppliers — primarily SubCom, NEC, Alcatel Submarine Networks (ASN), and HMN Technologies. Supply-chain access to these components is a strategic question independent of any individual cable.

5. Coding-gain improvements are quiet capacity events. Each generation of soft-decision FEC firmware upgrade silently re-rates the capacity of existing cables. Capacity forecasts that miss this re-rate cadence systematically underestimate the resilience of the deployed base.

The next article in this series turns from the fiber itself to the routing decisions made above it: how packets choose between the cables described here, how BGP advertises and withdraws those choices, and how the same logical layer can amplify or contain a physical-layer outage.

→ SYNTHESIS → Next: BGP Routing — the logical layer that runs atop this physical layer.

Sources

• ITU-T G.652 — Characteristics of a single-mode optical fibre and cable. International Telecommunication Union, current revision. (Confidence: High, primary standard.)
• ITU-T G.694.1 — Spectral grids for WDM applications: DWDM frequency grid. ITU. (Confidence: High, primary standard.)
• ITU-T G.975 / G.975.1 — Forward Error Correction for high-bit-rate DWDM submarine systems. ITU, 1996 / 2004. (Confidence: High, primary standard.)
• NICT — “301 Tb/s transmission over standard commercially available optical fibre.” National Institute of Information and Communications Technology, Japan, primary press release, 2024. (Confidence: High, primary source.)
• Microsoft / Facebook / Acacia Communications — MAREA 400G coherent transmission announcement, December 2018. (Confidence: High, joint operator press release.)
• TeleGeography Submarine Cable Map and database — cable status, fiber pair counts, design capacities, RFS dates. (Confidence: High for in-service cables; Medium for forward RFS projections subject to schedule slip.)
• Operator/vendor technical filings and white papers from SubCom, NEC, Alcatel Submarine Networks, and HMN Technologies — cable architecture, repeater spacing, EDFA design. (Confidence: High for system-level specifications.)
• Capacity figures for cables under construction (JUNO, TPU, Anjana, Medusa) — operator announcements and consortium press releases, 2023-2025. (Confidence: Medium-High — design capacities are credible engineered targets; lit capacity at RFS will be a fraction.)

---

Series: SYNTHESIS — Article 2 of 8.