{
  "id": "wavelength-division-multiplexing",
  "title": "Wavelength-Division Multiplexing",
  "lang": "en",
  "source_lang": "en",
  "summary": "Wavelength-division multiplexing carries multiple optical signals through a shared fiber by assigning them different wavelengths, increasing capacity and enabling wavelength-selective networking.",
  "category": "computer-science",
  "subcategory": "networks-security",
  "type": "concept",
  "aliases": [
    "WDM",
    "optical wavelength multiplexing"
  ],
  "generated_at": "2026-11-03",
  "model": "pro/gpt-6.1-sol",
  "url": "https://aiwiki.page/en/wavelength-division-multiplexing/",
  "translations": {
    "zh-Hans": {
      "url": "https://aiwiki.page/zh/wavelength-division-multiplexing/",
      "json": "https://aiwiki.page/zh/wavelength-division-multiplexing.json"
    }
  },
  "sections": [
    {
      "heading": "",
      "text": "Wavelength-division multiplexing (WDM) is a method of multiplexing in which several optical signals occupy different wavelength bands while traveling through the same transmission medium, usually an optical fiber. A transmitting terminal combines the signals, and a receiving terminal separates them with wavelength-selective devices. Each optical channel can carry an independently modulated data stream, allowing a single fiber to support many simultaneous connections.\n\nWDM is a foundational technology of modern telecommunications. It increases the capacity of installed fiber without requiring a separate fiber for every channel and allows operators to add, remove, or redirect selected channels within an optical network. Its applications range from short links between equipment to metropolitan networks and transoceanic cable systems supporting the Internet, telephone services, and private data transport.\n\nPractical WDM systems depend on more than assigning different colors to different transmitters. Their operation combines optical filtering, stable light sources, amplification, modulation, digital processing, and network control. Capacity and transmission distance are constrained by usable optical bandwidth, noise, signal distortion, component tolerances, and the interactions among channels."
    },
    {
      "heading": "Physical principle and channel capacity",
      "text": "WDM is the optical counterpart of frequency-division multiplexing. Although its name emphasizes wavelength, engineers frequently specify optical channels by frequency because regular frequency intervals are convenient for channel planning and signal processing. Vacuum wavelength \\(\\lambda\\) and optical frequency \\(f\\) are related by\n\n\\[f=\\frac{c}{\\lambda},\\]\n\nwhere \\(c\\) is the speed of light in vacuum. Light slows when it enters glass, but its frequency remains unchanged at a stationary interface. Telecommunications wavelength specifications therefore conventionally refer to vacuum wavelength rather than the shorter wavelength inside the fiber.\n\nNear a specified operating wavelength, a small frequency separation corresponds approximately to\n\n\\[|\\Delta\\lambda|\\approx\\frac{\\lambda^2}{c}|\\Delta f|.\\]\n\nAt 1,550 nanometers, a 100-gigahertz frequency interval is approximately 0.80 nanometers, and a 50-gigahertz interval is approximately 0.40 nanometers. Equal frequency intervals do not correspond to exactly equal wavelength intervals across a wide band. This distinction matters when configuring filters, specifying lasers, or comparing a wavelength measurement with a frequency-grid allocation.\n\nThe channels coexist because an approximately linear optical medium supports the superposition of electromagnetic fields. A wavelength-selective filter can separate their spectral regions without decoding the information they carry. The signals need not use identical bit rates, modulation formats, or client protocols, provided that each stays within its assigned spectral and power limits. WDM can also be combined with time-division multiplexing: each optical channel may itself contain many electronically aggregated data streams.\n\nFor \\(N\\) channels with net information rates \\(R_i\\), aggregate unidirectional capacity is\n\n\\[C_{\\mathrm{net}}=\\sum_{i=1}^{N}R_i.\\]\n\nThus, 80 channels carrying a net 100 gigabits per second each provide 8 terabits per second in one direction. This arithmetic does not establish that the signals can traverse a particular route. Reach also depends on modulation, amplifier noise, fiber characteristics, filtering, and nonlinear interactions. Published capacity figures must be interpreted carefully because gross modulation rate, payload rate after coding overhead, and bidirectional capacity are different quantities.\n\nWDM does not eliminate the bandwidth required by information. A modulated carrier occupies a finite spectrum rather than a single mathematical wavelength. The useful measure of spectral efficiency is the net information rate divided by occupied or allocated bandwidth, commonly expressed in bits per second per hertz. Increasing the number of channels improves total capacity only when the available spectrum and signal quality can accommodate them. Packing channels more tightly can require sharper filtering, more stable transmitters, better pulse shaping, or reduced operating margins."
    },
    {
      "heading": "Coarse, dense, and flexible channel grids",
      "text": "The principal conventional categories are coarse WDM, or CWDM, and dense WDM, or DWDM. Their distinction concerns channel spacing and the component requirements associated with that spacing, not whether the signals carry packets, telephone circuits, or another service. CWDM generally favors fewer, widely separated wavelengths and comparatively simple equipment. DWDM places many channels within a narrower optical band and is particularly suited to amplified, high-capacity transmission.\n\nITU-T Recommendation G.694.2 specifies a CWDM wavelength grid with 20-nanometer spacing, using nominal wavelengths from 1,271 to 1,611 nanometers. This yields 18 nominal channel positions. Actual systems may use only a subset because of fiber attenuation, component availability, or the desired reach. Older fiber can exhibit substantial absorption around the water-related loss peak near 1,383 nanometers, whereas low-water-peak fiber makes more of the CWDM range useful. Wide channel separation can allow operation with less tightly controlled lasers, reducing cost and power consumption.\n\nITU-T Recommendation G.694.1 specifies DWDM frequency grids referenced to 193.1 terahertz, corresponding to approximately 1,552.52 nanometers. Familiar fixed spacings include 100 and 50 gigahertz; the recommendation also provides finer grid definitions. Dense systems require accurate control of carrier frequency and filter passbands. A transmitter drifting toward the edge of its filter can suffer attenuation or distort its modulation spectrum even before it substantially interferes with a neighboring channel.\n\nDWDM commonly operates in the C band, approximately 1,530–1,565 nanometers, and sometimes in the L band, approximately 1,565–1,625 nanometers. These ranges are useful because silica fiber has low attenuation there and suitable optical amplifiers are available. Band boundaries are conventional rather than abrupt physical transitions, and the usable range of a particular line system is determined by its amplifiers, filters, and other components. Wider-band systems can combine multiple amplification bands, but each additional band introduces its own engineering requirements.\n\nFlexible-grid allocation addresses the mismatch between a fixed slot width and signals with different spectral widths. In the G.694.1 flexible grid, nominal center frequencies have a granularity of 6.25 gigahertz, while slot widths are multiples of 12.5 gigahertz. A high-symbol-rate signal can occupy a wider slot rather than being forced into a legacy 50-gigahertz channel. Conversely, narrower signals need not reserve unnecessarily large slices of spectrum.\n\nThese categories are not interchangeable measures of performance. A short-reach DWDM link may use no optical amplifier, while a CWDM link may be limited by loss rather than channel spacing. Likewise, flexible-grid support alone does not guarantee higher spectral efficiency: the transmitter waveform, receiver capability, filter shape, and route quality must all support the proposed allocation."
    },
    {
      "heading": "Transmitters, multiplexers, and receivers",
      "text": "A WDM terminal begins with optical transmitters, usually based on a laser whose emission lies within a specified channel band. In simpler systems, data directly modulate the laser drive current. More demanding systems often use an external modulator to alter the amplitude or phase of a continuous optical carrier. External modulation provides greater control over the waveform and can reduce impairments associated with frequency changes accompanying direct intensity modulation.\n\nLight sources are commonly fabricated from semiconductor materials. Distributed-feedback lasers provide wavelength-selective optical feedback within the device, while tunable lasers can move among many channel frequencies. Tuning simplifies equipment inventory and allows a network to assign a replacement transmitter to the required channel. Frequency accuracy, linewidth, output power, and stability remain important specifications. Narrow linewidth is especially valuable when information is encoded in optical phase.\n\nA multiplexer combines the transmitter outputs into a common fiber. Devices based on thin-film interference filters select wavelength bands through multilayer coatings. An arrayed waveguide grating uses paths of different optical lengths so that interference directs different wavelengths toward different ports. Diffraction gratings and other integrated or free-space optical structures can perform similar functions. Many passive devices are reciprocal: used in the opposite direction, the same underlying structure separates rather than combines channels.\n\nReal filters have finite insertion loss, nonzero leakage, and passbands that are neither perfectly rectangular nor independent of operating conditions. Isolation describes how effectively unwanted channels are rejected. Passband width and shape determine how much of the desired modulated signal is transmitted without distortion. Polarization dependence and temperature sensitivity also matter, particularly when many devices are cascaded. A design that works through a single filter may fail after repeated narrowing along a multistage route.\n\nAt the receiving terminal, a demultiplexer directs each channel toward its receiver. An intensity-detection receiver uses a photodiode to convert incident optical power into electrical current. A coherent receiver instead combines the signal with light from a local laser and measures electrical outputs that retain information about the optical field. The distinction determines which modulation formats can be recovered and how effectively certain transmission impairments can be corrected.\n\nTerminology varies across equipment families. A transceiver contains both transmitting and receiving functions, while a transponder commonly converts a client-side signal into a line-side optical signal with the required wavelength, format, and framing. A muxponder aggregates several client signals into a higher-rate line signal. These functions may reside in separate chassis or in pluggable modules connected directly to a router or switch. Their packaging does not change the optical requirement: every transmitted spectrum must fit the line system’s frequency, power, and signal-quality constraints."
    },
    {
      "heading": "Amplification and propagation impairments",
      "text": "Fiber attenuates optical power, and connectors, splices, multiplexers, and switching elements add further loss. Around 1,550 nanometers, conventional low-loss single-mode fiber commonly has attenuation on the order of 0.2 decibels per kilometer, although the exact value depends on fiber type and wavelength. An 80-kilometer span at that attenuation introduces approximately 16 decibels of fiber loss before other components are counted. Long routes therefore require amplification or regeneration.\n\nAn erbium-doped fiber amplifier amplifies signals in suitable wavelength bands by transferring energy from pump light to erbium ions and then to the passing optical signal through stimulated emission. Its importance for WDM is that it can amplify many channels simultaneously without first converting each one to an electrical signal. However, gain is wavelength-dependent, amplifier output power is finite, and amplified spontaneous emission adds noise. Gain-flattening filters and power-control schemes help maintain an acceptable balance among channels.\n\nRaman amplification transfers power from pump light to longer-wavelength signals through stimulated Raman scattering. It can be distributed along the transmission fiber or implemented in a more localized arrangement. Distributed amplification can improve the signal-power profile along a span and reduce effective noise penalties in suitable designs. Its implementation requires careful treatment of pump power, pump-signal interactions, reflections, and operating safety. Neither Raman amplification nor erbium amplification restores a waveform to its original noiseless state.\n\nDispersion distorts signals even when sufficient optical power reaches the receiver. Chromatic dispersion causes different frequency components to travel with different group delays, spreading a modulated pulse over time. Polarization-mode dispersion introduces differential delay between polarization states because real fiber is not perfectly symmetric. The magnitude and time variation of these effects depend on the installed fiber and its environment. Modern coherent receivers can compensate substantial linear dispersion electronically, but compensation capability is not unlimited.\n\nFiber also exhibits an intensity-dependent refractive index, producing the optical Kerr effect. Self-phase modulation affects a channel through its own power, while cross-phase modulation couples its phase to power variations in other channels. Four-wave mixing can generate new spectral components from combinations of existing optical frequencies. These effects depend on launch power, dispersion, wavelength spacing, modulation, and the distribution of power along the route. Consequently, raising transmitter power eventually worsens performance rather than continuously improving it.\n\nLong-haul design balances amplifier noise against nonlinear distortion. Low launch power makes noise comparatively dominant; excessive launch power increases nonlinear penalties. Amplifier gain changes when channels are added or removed, potentially causing transients on surviving channels. Operating systems therefore use channel equalization, automatic gain control, monitoring, and sometimes substitute loading signals. These measures make multichannel operation manageable, but every supported channel loading and route still has to remain within the system’s engineering limits."
    },
    {
      "heading": "Coherent modulation and digital signal processing",
      "text": "Coherent detection greatly expanded the information that can be carried within each WDM channel. Direct detection normally measures optical intensity and discards the carrier phase. A coherent receiver mixes the incoming signal with a local oscillator laser, allowing recovery of in-phase and quadrature components. With polarization-diverse reception, it can recover two independently modulated polarization components, increasing the amount of information transmitted within the same nominal spectral band.\n\nQuadrature amplitude modulation represents information through combinations of amplitude and phase. Quadrature phase-shift keying, usually abbreviated QPSK, uses four phase states, corresponding to two bits per symbol per polarization before overhead. Dual-polarization QPSK therefore carries four bits per combined symbol interval. Dual-polarization 16-QAM carries eight, and dual-polarization 64-QAM carries twelve. Higher-order constellations increase spectral efficiency but require better signal quality because their decision regions are more closely spaced.\n\nFor symbol rate \\(R_s\\), constellation size \\(M\\), and two independently used polarizations, a simplified gross bit-rate expression is\n\n\\[R_{\\mathrm{gross}}=2R_s\\log_2 M.\\]\n\nA dual-polarization 16-QAM signal at 64 gigabaud therefore has a gross rate of 512 gigabits per second. The available client payload is lower after framing and coding overhead. Probabilistic shaping and other adaptations can make the actual information rate differ from the simple fixed, equally likely constellation calculation.\n\nDigital signal processing performs tasks that previously required restrictive optical conditions or dedicated optical compensation. Receiver processing can estimate and correct chromatic dispersion, separate mixed polarization components, recover symbol timing, and track frequency and phase offsets between transmitter and local oscillator. High-speed analog-to-digital converters provide the sampled electrical signals. Specialized application-specific integrated circuits execute the required processing at line rate within practical power and latency budgets.\n\nAn error-correcting code adds structured redundancy so that a receiver can correct some transmission errors. Soft-decision decoding uses reliability information rather than only hard binary decisions and can operate at lower signal quality than simpler decoding schemes. This improvement comes at a cost in overhead, processing complexity, energy, and sometimes latency. A quoted line rate is therefore incomplete unless its relationship to payload capacity and coding is understood.\n\nChannel bandwidth is closely related to symbol rate and pulse shaping, not simply to the nominal number of bits per second. For idealized raised-cosine shaping with roll-off factor \\(\\alpha\\), the occupied bandwidth is approximately \\(R_s(1+\\alpha)\\). Increasing bits per symbol can raise capacity without a proportional increase in bandwidth, but usually reduces reach. Probabilistic constellation shaping offers a finer adjustment of information rate and noise tolerance. Operators can therefore trade capacity against distance, choosing different operating modes for short metropolitan routes and demanding long-haul paths."
    },
    {
      "heading": "Wavelength routing and optical network control",
      "text": "A point-to-point WDM link multiplexes channels between two endpoints. A wavelength-routed network adds intermediate nodes that can direct selected channels toward different destinations. A reconfigurable optical add-drop multiplexer, or ROADM, allows channels to pass through, be removed locally, or be inserted onto an outgoing fiber without requiring manual replacement of wavelength-specific filters. A wavelength-selective switch is a central building block in many such nodes.\n\nOptical transparency means that an express channel can pass through intermediate nodes without undergoing optical-electrical-optical conversion. Avoiding conversion can reduce the amount of transit equipment and allow the optical layer to transport different client formats. Transparency does not imply unlimited reach or complete independence from signal format. Every traversed amplifier and filter changes signal quality, and a path may eventually require regeneration to restore timing, reshape the waveform, and retransmit it.\n\nThe simplest wavelength-routing problem requires the same wavelength to remain available along an entire transparent path. This is the wavelength-continuity constraint. Wavelength conversion or electrical regeneration can relax it, but introduces additional resources and cost. In flexible-grid systems, the corresponding resource is a contiguous spectral interval. A connection must obtain a sufficiently wide interval along the route, unless suitable conversion is available.\n\nRoute selection and spectral allocation are coupled. A topologically short path may have insufficient free spectrum or excessive optical impairment. A longer path may offer a usable slot but require a more robust, lower-rate modulation format. Network planning therefore combines graph-based routing with transmission feasibility. The shortest-path problem is relevant, but distance alone is not a complete objective. Spectrum fragmentation can also leave enough unused bandwidth in total while providing no sufficiently large contiguous interval for a new signal.\n\nROADM capabilities are often described as colorless, directionless, and contentionless. Colorless operation allows an add/drop port to support different wavelengths rather than a permanently assigned one. Directionless operation allows a local transceiver to reach different outgoing fiber directions. Contentionless architectures reduce restrictions that would otherwise prevent simultaneous use of the same wavelength on different directions or local connections. These capabilities depend on the internal switching structure and are not automatic properties of every ROADM.\n\nControl software maintains inventory, provisions channels, evaluates feasible paths, and coordinates restoration. Optical-layer protection may use reserved resources, while restoration can search for new resources after a fault. These mechanisms interact with packet-layer rerouting and service-level agreements. A cable cut, amplifier failure, and failed client interface have different scopes, so failure localization matters. Optical power monitoring, spectral measurements, and receiver telemetry provide evidence for deciding whether to reroute traffic, adjust a channel, or replace a component."
    },
    {
      "heading": "Applications across access, metropolitan, and submarine networks",
      "text": "WDM is used at several network scales, but its purpose and constraints differ among them. In access and enterprise networks, the primary objective may be conserving scarce fibers rather than maximizing spectral efficiency. A small number of wavelengths can connect buildings, aggregate mobile-network traffic, or carry separate services over an existing cable. Passive multiplexers are attractive where electrical power, environmental control, and maintenance access are limited.\n\nA passive optical network distributes signals between a central terminal and multiple endpoints through passive optical infrastructure. Some access systems use different wavelength bands for upstream and downstream traffic; others use multiple wavelength channels to increase capacity or assign resources among users. Wavelength-selective coexistence also allows newer and older generations of access equipment to share a distribution network. The presence of multiple wavelengths does not necessarily mean that each subscriber receives a permanently dedicated wavelength.\n\nMetropolitan networks interconnect central offices, mobile aggregation sites, enterprises, and data centers. Their routes are usually shorter than intercontinental links but may contain numerous add/drop nodes and complicated service patterns. CWDM can be economical for limited channel counts and modest reach, while DWDM supports larger capacities and more sophisticated optical switching. The appropriate architecture depends on the installed fiber plant, growth expectations, node losses, and operational requirements rather than distance alone.\n\nData-center interconnect emphasizes large capacity between a relatively small number of sites. Coherent pluggable modules have made it possible to place long-reach optical transmission functions directly in packet equipment. The Optical Internetworking Forum’s 400ZR implementation agreement, published in 2020, established an interoperable 400-gigabit-per-second coherent interface aimed at this market. Such an interface standardizes a particular operating envelope; it should not be interpreted as a universal replacement for all metropolitan or long-haul transponders.\n\nTerrestrial long-haul networks use amplified spans and, where needed, electrical regeneration to connect cities and regions. The expense of installing cable makes increasing the usable capacity of existing fiber economically important. WDM permits incremental deployment: an operator can activate additional channels as demand grows, provided that the underlying line system was designed to support them. However, changes in channel loading can require power rebalancing and validation of existing services.\n\nIn a submarine communications cable, WDM carries many channels across each fiber pair through chains of submerged optical amplifiers. Cable-system design must account for electrical power supplied from shore, repeater reliability, oceanic repair conditions, and total system capacity. Modern high-capacity approaches may use more fiber pairs rather than maximizing power or channel rate in each pair. This illustrates an important distinction: the best spectral efficiency for one fiber is not always the best capacity or cost outcome for the complete cable system."
    },
    {
      "heading": "Link engineering, measurement, and interoperability",
      "text": "Engineering a WDM route begins with a physical inventory: fiber type and length, splice and connector losses, amplifier locations, node architecture, and available optical bands. A basic power budget compares launch power with total attenuation and receiver sensitivity, but this is sufficient only for relatively simple links. Amplified coherent systems additionally require noise, nonlinear, dispersion, and filtering assessments. Practical designs reserve margin for aging, repairs, component variation, and measurement uncertainty.\n\nOptical signal-to-noise ratio, or OSNR, compares signal power with optical noise power in a specified reference bandwidth. A common reference bandwidth is 0.1 nanometer near 1,550 nanometers, but a measurement or specification must state its convention. The same numerical OSNR does not guarantee equivalent performance for signals with different symbol rates or modulation formats. Coherent receivers may report additional quality estimates derived from received samples, including generalized signal-to-noise measures and error-related metrics.\n\nAn optical spectrum analyzer displays power as a function of wavelength or frequency. It can reveal missing channels, power imbalance, unwanted emissions, and spectral encroachment. Measuring noise accurately in a densely occupied spectrum is more difficult than inspecting an empty interval between widely separated channels. Filtered amplified noise and signal spectra can overlap, requiring appropriate measurement methods rather than a simple reading from the visible gap.\n\nReceiver performance is often characterized using bit-error ratios before and after forward error correction. A pre-correction error ratio provides information about operating margin, but the relationship between that number and successful decoding depends on the code and channel statistics. Corrected-error counts, uncorrectable blocks, received power, carrier-frequency offset, and estimated dispersion can all support fault diagnosis. A link carrying apparently error-free traffic may still be close to a failure threshold if its reserve margin is small.\n\nInteroperability has several distinct levels. Mechanical compatibility of a pluggable module does not guarantee electrical compatibility with its host, nor does either guarantee optical compatibility with a remote terminal. Line-side agreement may need to cover carrier frequency, modulation, symbol rate, coding, framing, launch power, spectral shape, and management behavior. Standards and implementation agreements define particular combinations, while proprietary operating modes may provide different reach or capacity without cross-vendor compatibility.\n\nAn alien wavelength is a channel supplied by equipment outside the line-system vendor’s integrated terminal family. Such operation can separate transponder selection from the amplifiers and ROADMs already installed. It requires clear responsibility for launch conditions, monitoring, impairment limits, and troubleshooting. Channels sharing a fiber can affect one another through amplifier behavior and nonlinear interactions, so acceptance testing must consider the full loading arrangement. Optical safety is also relevant: communications wavelengths are generally invisible, and amplified systems can present hazardous power at open connectors even when no visible light is apparent."
    },
    {
      "heading": "Historical development and emerging directions",
      "text": "The conceptual basis of WDM predates its dominance in telecommunications: separate optical frequencies can share a medium just as radio frequencies share a transmission path. Its large-scale adoption depended on the development of a practical low-loss fiber system. In 1966, Charles K. Kao and George A. Hockham analyzed the prospects of optical fibers for communication and identified material purity as central to reducing attenuation. In 1970, researchers at Corning demonstrated silica fiber with attenuation below 20 decibels per kilometer, an important step toward practical fiber transmission.\n\nEarly commercial fiber systems commonly operated near 850 nanometers, followed by systems near 1,300 nanometers, where conventional single-mode fiber has low chromatic dispersion. Two-window arrangements could combine signals from widely separated optical bands. Dense channel packing around 1,550 nanometers became more attractive as stable lasers, selective filters, and optical amplification improved. The erbium-doped fiber amplifier, developed during the 1980s, was especially consequential because it enabled simultaneous amplification of numerous channels in a low-loss fiber window.\n\nDuring the 1990s, commercial DWDM systems increased the number of optical channels that could occupy a fiber, expanding backbone capacity without a corresponding multiplication of installed cables. This growth supported the rapid expansion of Internet traffic. Channel counts alone do not capture the improvement: systems also increased the bit rate per wavelength, improved amplifier engineering, and developed more capable monitoring and protection arrangements. Work in industrial laboratories, including Bell Labs, contributed to advances across optical transmission, devices, and networking.\n\nAround the late 2000s and early 2010s, digital coherent transmission became central to high-capacity long-haul systems. High-speed converters and signal-processing chips made phase-sensitive reception practical at commercial line rates. Coherent detection enabled polarization multiplexing and powerful electronic compensation of linear impairments, reducing reliance on optical dispersion-compensation modules. Subsequent development increased symbol rates and introduced more flexible modulation and coding choices rather than relying only on additional wavelengths.\n\nCurrent capacity expansion follows several partly independent directions. Higher symbol rates put more information into each carrier but require faster electronics and broader, well-controlled optical passbands. Wider usable optical bandwidth adds spectrum beyond established bands but requires suitable amplifiers and components. Space-division multiplexing adds parallel optical paths through additional fiber pairs, multiple cores, or multiple spatial modes. It complements WDM rather than replacing it: each spatial path can itself carry many wavelength channels.\n\nIntegration is another major direction. Silicon photonics and other photonic platforms can integrate modulators, waveguides, couplers, and related functions, while electronic-photonic co-packaging seeks to reduce difficult high-speed electrical interconnections. Optical frequency combs offer a potential source of many regularly spaced carriers, although source generation, channel control, and system-level performance remain distinct engineering problems. Across these approaches, progress is judged not only by laboratory capacity records but also by achievable reach, energy per delivered bit, manufacturability, repairability, and compatibility with deployed infrastructure."
    }
  ],
  "links": [
    "multiplexing",
    "optical-fiber",
    "internet",
    "frequency-division-multiplexing",
    "speed-of-light",
    "laser",
    "semiconductor",
    "photodiode",
    "router",
    "application-specific-integrated-circuit",
    "error-correcting-code",
    "shortest-path-problem",
    "service-level-agreement",
    "data-center",
    "bell-labs",
    "silicon-photonics"
  ],
  "sources": [],
  "license": "https://creativecommons.org/licenses/by/4.0/"
}