How Fiber Optic Cables Transmit Data

A standard optical fiber is 125 micrometres across, about the width of a human hair, and it carries most of the world's internet traffic. In December 2020, researchers at Japan's National Institute of Information and Communications Technology (NICT) sent 1.01 petabits per second, more than a quadrillion bits every second, through a single fiber over 23 km in a laboratory experiment, using 382 wavelength channels and a special fiber supporting 15 light-guiding modes. Commercial links run far below that, but the experiment shows how much headroom glass still has. The material doing the work is ultra-pure silica. The carrier is light. This article explains how the physics turns that combination into a global data network, and how an idea that most engineers doubted in 1966 became the internet's backbone.

Key takeaways

• Fiber optic cables carry data as pulses of light through a glass core. A surrounding cladding with a slightly lower refractive index traps the light by total internal reflection.
• Signal loss is measured in decibels per kilometre. Modern telecom fiber loses roughly 0.2 dB/km at 1,550 nm, so light travels about 15 km before half its power is gone.
• Single-mode fiber (core about 8–10 µm) carries light over thousands of kilometres; multimode fiber (50 or 62.5 µm core) is cheaper for short links inside buildings and data centres.
• Wavelength-division multiplexing (WDM) sends many colours of laser light down one fiber at once, each an independent channel.
• Erbium-doped fiber amplifiers boost all channels together in the optical domain, which made today's transoceanic cables practical.
• Charles Kao and George Hockham's 1966 paper argued that glass could be made clear enough for communication; Kao shared the 2009 Nobel Prize in Physics for it.

Light in glass: total internal reflection

The core trick of fiber optics can be demonstrated with a glass of water and a torch.

When light passes from one transparent medium into another, it changes speed and bends. How much it bends is set by each material's refractive index, n, which is the ratio of the speed of light in vacuum to its speed in the material. Snell's law describes the bending: n₁ sin θ₁ = n₂ sin θ₂, where θ is the angle measured from the perpendicular to the boundary.

When light travels from a higher-index material into a lower-index one, it bends away from the perpendicular. Tilt the beam further and further, and at one particular angle, the critical angle, the refracted ray would skim exactly along the boundary. Beyond that angle there is no refracted ray at all: the light is reflected completely back into the first material. That is total internal reflection, and unlike a mirror it loses essentially no energy at each bounce.

The critical angle follows directly from Snell's law: sin θc = n₂ / n₁. An optical fiber builds this into its structure. A thin core has a slightly higher refractive index than the cladding around it. Light launched into the core at a shallow enough angle hits the core–cladding boundary beyond the critical angle and is reflected, again and again, all the way along the fiber. The fiber can bend gently and the light follows it.

A worked example

Typical telecom fiber has a cladding of pure silica (refractive index about 1.444 at 1,550 nm) and a core doped with germanium dioxide, which raises the index by only around a third of a percent, to roughly 1.449. Plugging those numbers into the formula gives a critical angle of about 85 degrees from the perpendicular. In other words, only rays travelling within about 5 degrees of the fiber's axis are guided. That small acceptance angle is described by the fiber's numerical aperture (NA = √(n₁² − n₂²)), which for standard single-mode fiber is roughly 0.12. Because the angle is so small, the laser must be aligned to the core very precisely, which is why fiber connectors and splices are engineered to micrometre tolerances.

Why the index difference is so small

A bigger index difference would capture more light, so why keep it tiny? Because a small difference allows a single guided mode in a core large enough to manufacture and connect reliably. The core diameter, the wavelength and the index difference together set how many modes fit, as the next section explains.

Single-mode versus multimode fiber

Light in a fiber does not simply bounce like a billiard ball. Because it is a wave, only certain field patterns, called modes, can propagate. The number of modes depends on the core size relative to the wavelength.

Property	Single-mode fiber	Multimode fiber
Core diameter	about 8–10 µm	50 µm or 62.5 µm
Cladding diameter	125 µm	125 µm
Typical light source	Laser at 1,310 nm or 1,550 nm	VCSEL or LED at about 850 nm
Main limit on distance	Chromatic dispersion and loss	Modal dispersion
Typical reach	Tens to thousands of km (with amplifiers)	Hundreds of metres to about 2 km
Typical use	Long-haul, metro, undersea, fiber to the home	Inside buildings and data centres

Multimode fiber has a large core, so many modes travel at once, each taking a slightly different path. Rays that bounce steeply travel further than rays down the middle, so they arrive later. A sharp pulse of light therefore spreads out in time, an effect called modal dispersion. Spread enough, adjacent pulses blur into each other and bits become unreadable. Graded-index multimode fiber, whose refractive index falls smoothly from the centre outward, reduces the effect by letting the outer, longer paths travel slightly faster, but multimode is still limited to short distances.

Single-mode fiber has a core so narrow that, at telecom wavelengths, only one mode can propagate. Modal dispersion disappears, leaving smaller effects. The main one is chromatic dispersion: even a laser emits a narrow range of wavelengths, and different wavelengths travel at slightly different speeds in glass. Standard single-mode fiber has nearly zero chromatic dispersion around 1,310 nm, and modern systems compensate for dispersion at 1,550 nm digitally. Practically all long-distance links, including every transoceanic cable, use single-mode fiber.

Attenuation: how clear is the glass?

Even the purest glass slowly absorbs and scatters light. Engineers measure the loss in decibels per kilometre (dB/km). The decibel is logarithmic: 3 dB means half the power is gone, 10 dB means 90% is gone, 20 dB means 99% is gone.

Attenuation	Fraction of light remaining after 1 km	After 10 km
1,000 dB/km (best glass of the early 1960s)	effectively zero	effectively zero
20 dB/km (Kao's 1966 target)	1%	10⁻²⁰
17 dB/km (Corning, 1970)	about 2%	about 10⁻¹⁷
4 dB/km (Corning, 1972)	about 40%	about 0.01%
0.2 dB/km (modern fiber at 1,550 nm)	about 95%	about 63%

At 0.2 dB/km, a signal loses 3 dB, half its power, every 15 km, and about 20 dB over 100 km. That is still a hundredfold drop, which is why long links need amplification, but it is extraordinarily transparent: a window pane made of such glass would have to be tens of kilometres thick before it looked noticeably darker.

Where the loss comes from

• Rayleigh scattering. Glass is frozen liquid, and tiny random fluctuations in density scatter a little light in all directions. Rayleigh scattering falls with the fourth power of wavelength, so longer wavelengths scatter far less. The same physics makes the sky blue, as Tyndall's light-scattering experiment showed in the nineteenth century.
• Infrared absorption. Beyond about 1,600 nm, vibrations of the silicon–oxygen bonds begin to absorb light.
• Impurities. Traces of metal ions and, above all, hydroxyl (OH) groups from water absorb at specific wavelengths. Early fibers had a strong "water peak" around 1,380 nm; modern "low water peak" fibers have largely removed it.
• Bending. Tight bends push light past the critical angle so it leaks out. Bend-insensitive fibers for homes and offices use refined index profiles to tolerate tighter curves.

The balance between falling scattering and rising infrared absorption puts the lowest-loss window around 1,550 nm. That is why long-haul systems operate there, and why the light in a fiber network is infrared and invisible to the eye.

The idea that started it: Kao and Hockham, 1966

In the early 1960s, optical fiber existed as a curiosity used for short medical imaging bundles, but glass lost light so quickly, often 1,000 dB/km or more, that sending a signal even a few hundred metres seemed hopeless. At Standard Telecommunication Laboratories in Harlow, England, Charles Kao and George Hockham examined why glass was so lossy. In a paper presented to the Institution of Electrical Engineers in January 1966 and published that July, "Dielectric-fibre surface waveguides for optical frequencies", they argued that the losses were caused mainly by impurities, not by glass itself, and that purified glass could fall below 20 dB/km, enough for practical communication.

Many engineers were sceptical. The British Post Office, however, was interested, and the challenge was taken up by laboratories in Britain, Japan and the United States. At Corning Glass Works in New York, Robert Maurer, Donald Keck and Peter Schultz used a vapour-deposition process to make extremely pure fused silica. On 7 August 1970, Keck measured a fiber with a titanium-doped core at about 17 dB/km, beating Kao's threshold. By 1972 a germanium-doped core had reached about 4 dB/km, and by 1977 Corning fiber was carrying live commercial traffic.

In 2009, Kao shared the Nobel Prize in Physics, cited "for groundbreaking achievements concerning the transmission of light in fibers for optical communication". The other half of the prize went to Willard Boyle and George Smith for the CCD image sensor. The longer history of how glass replaced copper, from Victorian light-guiding demonstrations to the first commercial links, is told in the fiber optic cable.

Turning light into data

A fiber link has three main parts: a transmitter, the fiber, and a receiver.

The transmitter is a semiconductor laser, a tiny chip descended from the first lasers of 1960, whose output is modulated to encode bits. The simplest scheme switches light on and off: on for 1, off for 0. That is how early systems worked, and short links inside data centres still often use simple intensity modulation.

Long-haul systems now use coherent optics, which encode information not only in the brightness of the light but also in its phase and in two independent polarisations. Formats such as 16-QAM pack four bits into each symbol on each polarisation. A powerful digital signal processor at the receiver then undoes chromatic dispersion and other distortions mathematically. Coherent transmission is a major reason capacity per wavelength rose from 10 gigabits per second in the early 2000s to hundreds of gigabits, and beyond, today.

The receiver is a photodiode that converts arriving photons back into an electrical current, followed by electronics that recover the bits. Errors are caught and corrected by forward error correction codes added at the transmitter.

WDM: many colours, one fiber

A single laser uses only a sliver of the glass's transparent spectrum. Wavelength-division multiplexing fills the rest. Several lasers, each at a slightly different wavelength, are combined into one fiber with a multiplexer; at the far end a demultiplexer separates them for individual receivers. Each wavelength is an independent channel, unaware of the others.

Dense WDM (DWDM) packs channels tightly, commonly on a 50 GHz or 100 GHz grid standardised by the ITU. The C-band (about 1,530–1,565 nm) holds roughly 80–96 channels at 50 GHz spacing; adding the L-band (about 1,565–1,625 nm) roughly doubles that. Multiply dozens of channels by hundreds of gigabits per channel, and a single fiber pair carries tens of terabits per second. A modern submarine cable may carry many fiber pairs.

Newer systems use a flexible grid, assigning each signal exactly the spectrum it needs, and research is pushing into additional bands (O, E, S and U) and into space-division multiplexing, using multicore or few-mode fibers. The NICT petabit experiment combined both ideas: hundreds of wavelengths sent through a fiber guiding 15 modes.

Amplifiers: keeping light alive over distance

Even at 0.2 dB/km, a signal fades after about 100 km. Early systems used regenerators: every few tens of kilometres, the light was converted to electricity, cleaned up, and converted back to light by a new laser. Regenerators were expensive, consumed power, and worked on one wavelength at a time, which made WDM impractical over long distances.

The breakthrough was the erbium-doped fiber amplifier (EDFA), developed in the late 1980s, with key work by David Payne's group at the University of Southampton and by Emmanuel Desurvire and colleagues at Bell Labs. An EDFA is a few metres of fiber whose core contains a trace of erbium ions. A pump laser, usually at 980 nm or 1,480 nm, lifts the erbium ions into an excited state. When signal photons near 1,550 nm pass through, they stimulate the ions to release identical photons, the same physics as in a laser, and the signal grows. Crucially, an EDFA amplifies every wavelength in the C-band at once, entirely in the optical domain.

Amplifiers also add a little noise each time, so a transoceanic link is a careful balance: amplifiers spaced roughly every 50–100 km, enough launch power to stay above the noise, but not so much that nonlinear effects in the glass distort the signal. Raman amplification, which uses the fiber itself as the gain medium, is often added for extra reach.

Inside a cable

A fiber by itself is fragile, so cables are built in layers. A land cable typically groups many coated fibers in gel-filled tubes or ribbons around a central strength member, inside a plastic jacket. Some contain more than a thousand fibers.

A submarine cable in deep water is surprisingly slim, about as wide as a garden hose. From the inside out it typically contains:

1. A small number of fiber pairs, often a few up to around two dozen in newer designs
2. A protective tube, often filled with water-blocking gel
3. Steel wires for tensile strength
4. A copper conductor that carries high-voltage direct current to power the undersea amplifiers
5. A polyethylene insulating jacket
6. Additional steel armour in shallow water, where anchors and fishing gear are a threat

As of early 2026, TeleGeography counted more than 600 active and planned submarine cables, totalling well over 1.5 million kilometres. Roughly 200 faults occur each year, about two-thirds caused by fishing and ships' anchors, with natural hazards such as earthquakes and underwater landslides, and component failures, accounting for much of the rest. Repair ships grapple the damaged cable, haul it to the surface, splice in a new section and lower it again, which can take days to weeks. For more on routes, landing stations and repairs, see how undersea internet cables work.

Fiber compared with copper and wireless

Medium	Typical capacity per link	Distance before regeneration or amplification	Sensitivity to electromagnetic interference
Copper twisted pair (DSL)	Tens of Mbit/s, falling with distance	about 1–5 km	Susceptible
Copper coaxial (cable broadband)	Up to a few Gbit/s shared	a few km with amplifiers	Partly shielded
Wireless (5G, Wi-Fi)	Up to Gbit/s, shared among users	metres to kilometres	Shares radio spectrum
Single-mode fiber (DWDM)	Tens of Tbit/s per fiber pair	about 50–100 km per amplifier	Immune

Fiber's advantages come from physics. Light at around 193 THz has an enormous range of usable frequencies compared with electrical signals, so bandwidth is vast. Glass does not conduct electricity, so fiber is immune to electrical interference and lightning-induced surges, and it is hard to tap without detection. Its main disadvantages are the cost of installation, the skill required to splice and connect it, and the need for electronics at each end. Even wireless networks depend on fiber: every mobile mast needs a "backhaul" connection, usually optical, as explained in how Wi-Fi and 5G work.

Common misconceptions

"Fiber is fast because light is fast"

Light in glass travels at about two-thirds of its vacuum speed, roughly 200,000 km per second, which is similar to electrical signals in many copper cables. Fiber's real advantage is bandwidth and low loss, not raw signal speed. For latency-sensitive uses, the route length matters more than the medium, which is why some financial firms have paid for microwave links that travel through air in straighter lines. The history of measuring that speed runs from Rømer's observations of Io to modern lasers.

"The light zig-zags like a ball off walls"

The ray picture is a useful simplification for multimode fiber. In single-mode fiber, the light is better described as one guided wave, part of which actually extends a little into the cladding.

"Fiber optic signals are visible light"

Telecom systems use infrared light at around 850, 1,310 and 1,550 nm, outside the visible range. That is also why you should never look into a live fiber end: the light is invisible but can still damage the eye.

"Satellites carry most international traffic"

They carry a very small share. Submarine fiber cables carry the overwhelming majority of intercontinental data, because their capacity and cost per bit are far better.

Frequently asked questions

Q: How can one fiber carry TV, phone calls and internet at the same time? A: Every service is converted into digital data, so the fiber carries only bits. Wavelength-division multiplexing separates traffic onto different colours of light, and within each wavelength the bits from many users are interleaved as packets.

Q: Why does fiber outperform copper for broadband? A: Copper signals weaken quickly at high frequencies and pick up electrical interference, so speeds fall with distance. Fiber has very low loss across an enormous frequency range and is immune to electromagnetic interference, so it supports much higher speeds over much longer distances.

Q: What is fiber to the home (FTTH)? A: FTTH runs fiber all the way to each building instead of switching to copper for the last stretch. It supports high symmetrical speeds and low latency. The main barrier is the cost of digging or stringing new cable to every home.

Q: What happens when an undersea cable breaks? A: Traffic is rerouted through other cables, sometimes with reduced capacity or higher latency. A repair ship locates the fault, lifts the cable, splices in new fiber and lowers it back to the seabed, typically within days to weeks.

Q: Is fiber better than 5G for home internet? A: Usually. Fiber gives dedicated capacity, symmetrical speeds and low latency. 5G fixed wireless can be a good option where fiber is unavailable, but the radio spectrum is shared, so performance varies with distance and congestion.

Q: Will fiber ever run out of capacity? A: There is a theoretical ceiling for any one fiber, set by noise and nonlinear effects, and the best commercial systems are approaching it. Engineers respond by using more spectral bands, more fibers per cable and multicore or few-mode fibers, which multiply capacity in parallel.

The bottom line

Fiber optics rests on a simple piece of physics, total internal reflection, and an extraordinary feat of materials science: glass so pure that light crosses kilometres of it with barely any loss. Kao and Hockham's 1966 insight that impurities, not glass itself, were the problem set the target; Corning's 1970 fiber hit it; erbium amplifiers and wavelength-division multiplexing turned single strands into multi-terabit highways. Every video call and web page you load spends most of its journey as infrared light in a strand of glass. For the bigger picture of how those strands connect into a network, see how the internet actually works.

Sources

• NICT press release: 1 petabit per second transmission over a 15-mode fiber (18 December 2020).
• IEEE Engineering and Technology History Wiki: Milestone, World's First Low-Loss Optical Fiber for Telecommunications, 1970.
• Corning Museum of Glass: Fiber Optics at 50.
• TeleGeography: Submarine cable frequently asked questions.
• Wikipedia: Charles K. Kao, for the 1966 paper citation and 2009 Nobel Prize citation.
• K. C. Kao and G. A. Hockham, "Dielectric-fibre surface waveguides for optical frequencies", Proceedings of the IEE 113(7), 1151–1158 (1966).

---

Read next:

• How the Internet Actually Works
• How Undersea Internet Cables Work
• History of Electricity: Faraday, Tesla, Edison, and the Force That Powers the World