{
  "id": "multiplexing",
  "title": "Multiplexing",
  "lang": "en",
  "source_lang": "en",
  "summary": "Multiplexing enables multiple signals, data streams, or activities to share a communication channel or computing resource while preserving their separate identities.",
  "category": "computer-science",
  "subcategory": "networks-security",
  "type": "concept",
  "aliases": [
    "Multiple-channel sharing",
    "Muxing"
  ],
  "generated_at": "2026-11-03",
  "model": "pro/gpt-6.1-sol",
  "url": "https://aiwiki.page/en/multiplexing/",
  "translations": {
    "zh-Hans": {
      "url": "https://aiwiki.page/zh/multiplexing/",
      "json": "https://aiwiki.page/zh/multiplexing.json"
    }
  },
  "sections": [
    {
      "heading": "",
      "text": "Multiplexing is the combination of multiple distinguishable signals, data streams, or activities for transmission or processing through a shared resource. A corresponding operation, demultiplexing, separates the combined representation or directs its components to their intended destinations. In telecommunications, multiplexing allows many conversations or data connections to share a cable, optical fiber, or radio channel. In computing, the term also describes the selection of signals within electronic circuits, the handling of many network connections, and the sharing of processing or input/output facilities.\n\nThe central problem is not simply combining inputs, but preserving enough structure to identify and recover them. Different systems achieve this through frequency bands, time slots, wavelengths, codes, spatial paths, addresses, or stream identifiers. Signal multiplexing emphasizes the physical representation of combined signals; computer-network and software uses often operate on packets, messages, or events instead. These mechanisms can coexist: an application stream may be multiplexed into a transport connection, whose packets share a network link carried on one wavelength of a multiplexed optical fiber.\n\nMultiplexing improves resource utilization and reduces the need for separate physical facilities, but introduces requirements for coordination, synchronization, buffering, and isolation. Its design therefore connects physical transmission theory with digital electronics, network protocols, scheduling, and operating-system interfaces."
    },
    {
      "heading": "Fundamental principles and distinctions",
      "text": "A multiplexing system contains several logical inputs, a shared resource, and a rule associating each input with an identifiable part of that resource. The combining function is commonly called a multiplexer, or mux, and the separating function a demultiplexer, or demux. These may be physical devices, circuits within an integrated circuit, protocol functions, or software components. They need not be colocated: a radio transmitter may combine signals that receivers separate many kilometers away, while a processor may select among signals across a few millimeters of silicon.\n\nSeparability depends on a distinguishing dimension. Frequency-division systems assign different frequency ranges; time-division systems assign different transmission intervals; packet systems attach identifying information to variable sequences of data. Some systems provide channels that are approximately independent, while others permit interference and rely on signal processing to recover the components. A shared transmission medium is therefore not enough to establish that useful multiplexing has occurred: the receiver must also possess the information or physical discrimination needed to distinguish the intended signals.\n\nMultiplexing is related to, but distinct from, several neighboring concepts. Multiple access concerns independent devices gaining access to a shared medium, often through distributed coordination. A multiplexing device may combine inputs already available at one location, whereas a multiple-access system must accommodate separate transmitters. The same physical technique can support both: time-division multiplexing combines channels into a transmission, while time-division multiple access assigns transmission opportunities to different stations. Duplexing instead separates the two directions of a communication link.\n\nMultiplexing also differs from data compression and error correction. Compression reduces the representation required for information; multiplexing arranges several representations within a common resource. Error correction adds structured redundancy to improve recovery from corruption. A system may perform all three, but their purposes and capacity effects are different. Multiplexing alone does not make an arbitrary amount of information fit into a fixed-capacity channel.\n\nA useful abstraction assigns each input \\(x_i\\) an encoding or placement operation \\(E_i\\), producing a shared representation\n\n\\[y=\\mathcal{M}\\bigl(E_1(x_1),E_2(x_2),\\ldots,E_n(x_n)\\bigr).\\]\n\nA receiver applies an appropriate separation operation to recover each component, exactly or within a specified error tolerance. In a digital protocol, \\(E_i\\) may attach a stream identifier; in a radio system, it may shift a waveform onto a carrier frequency. The particular mechanism determines the overhead, synchronization requirements, and consequences of interference or overload."
    },
    {
      "heading": "Historical development",
      "text": "Multiplexing developed from the economic importance of sharing expensive communication infrastructure. During the nineteenth century, telegraph operators sought ways to carry more messages over existing wires rather than construct additional routes. Duplex telegraph systems supported simultaneous communication in opposite directions. Thomas Edison’s quadruplex telegraph, developed in 1874, combined duplex operation with the transmission of two distinguishable signals in each direction. Its use of current polarity and strength illustrates how multiple messages could be separated through different electrical properties.\n\nÉmile Baudot’s telegraph system, patented in 1874, used a five-unit code and synchronized distributors. A rotating distributor assigned successive intervals to different operators, anticipating the regular slot structure of later time-division systems. Such arrangements required the receiving distributor to maintain the correct relationship with the transmitting distributor. A timing error could assign symbols to the wrong channel even when the individual electrical pulses were detected correctly.\n\nThe growth of the telephone created a different requirement: carrying many continuous voice signals rather than sequences of telegraph symbols. Twentieth-century carrier systems translated voice channels into separate frequency bands. Standardized channel groups allowed telephone networks to combine small bundles into progressively larger aggregates. Coaxial cable and microwave radio systems consequently carried large numbers of conversations through infrastructure whose cost could be shared among many subscribers.\n\nDigital telephony established another major lineage. Alec Reeves proposed pulse-code modulation in 1937, representing sampled signal amplitudes as digital values. Practical digital carrier systems followed as electronic components became smaller and more economical. The T1 carrier system entered commercial service in the United States in 1962, carrying 24 digital voice channels at an aggregate line rate of 1.544 megabits per second. Its repeating frames made each channel’s position identifiable without attaching an independent address to every voice sample.\n\nComputer communication introduced traffic whose demand was much less uniform than continuous voice transmission. Packet-switching research during the 1960s, associated with Paul Baran, Donald Davies, and others, developed ways to share communication links among bursts of data. Later optical networks multiplied capacity by transmitting several wavelengths through one fiber, while software protocols introduced independently identified streams within network connections. These developments did not replace one another completely: fixed digital circuits, packet aggregation, optical channels, and application streams continue to occupy different layers of the same communication systems."
    },
    {
      "heading": "Frequency and wavelength division",
      "text": "Frequency-division multiplexing assigns different signals to different portions of a medium’s usable spectrum. Each input is used to modulate a carrier, and the resulting waveforms are combined. At the receiving end, filters and demodulators select the desired band and recover its information. Several channels are transmitted at the same time, but occupy distinguishable frequency ranges. This contrasts with systems that give each channel exclusive use of a broader band for a short interval.\n\nIn conventional frequency-division systems, adjacent channels are separated sufficiently to accommodate imperfect filters and spectral spreading. Unused intervals called guard bands reduce interference but consume spectrum. If channel \\(i\\) requires bandwidth \\(B_i\\), the total allocation must include both the channel bandwidths and the guard intervals. The resulting capacity depends on modulation, noise, channel conditions, and allowable interference, not simply on the number of nominal carrier frequencies.\n\nAnalog telephone carrier systems provide a classic example. A traditional basic group contained 12 voice channels placed in successive 4-kilohertz allocations, occupying a 48-kilohertz band. Further frequency translation combined groups into larger arrangements. Cable television uses the same broad principle: multiple radio-frequency channels coexist on a shared cable. Some systems carry digital modulation within each frequency allocation, showing that frequency-division multiplexing is not inherently analog.\n\nWavelength-division multiplexing applies frequency separation to optical transmission. Lasers generate carriers at different wavelengths, optical components combine them into one fiber, and wavelength-selective devices separate them at the destination. Because optical frequency and wavelength are related by \\(f=c/\\lambda\\) in vacuum, wavelength division is physically a form of frequency division, although optical engineering generally uses its own terminology and component families.\n\nCoarse wavelength-division multiplexing uses relatively widely spaced optical channels, while dense wavelength-division multiplexing supports more closely spaced channels. The ITU-T G.694.1 frequency grid uses a reference frequency of 193.1 terahertz and includes fixed-grid arrangements such as 100- and 50-gigahertz spacing; flexible-grid arrangements permit more adaptable spectral allocations. A channel’s data rate depends on its symbol rate, modulation, coding, and transmission conditions rather than on spacing alone.\n\nOptical channels are not perfectly independent. Fiber nonlinearities, amplifier noise, filter characteristics, and accumulated dispersion can limit performance. Coherent receivers and digital signal processing help recover tightly packed, high-rate signals. Adding wavelengths can therefore increase aggregate throughput without installing another fiber, but only within the optical system’s power, spectral, and signal-quality constraints."
    },
    {
      "heading": "Synchronous time-division multiplexing",
      "text": "Time-division multiplexing assigns different inputs successive intervals on a shared transmission facility. In its synchronous form, these intervals occur in a repeating frame, and each channel receives a predetermined position. The receiver identifies a channel’s data from its place in the frame rather than from a separate address attached to every sample. This arrangement is especially suitable for sources that produce information at regular rates.\n\nDigital voice illustrates the mechanism. In the commonly used telephone representation, a voice waveform is sampled 8,000 times per second and each sample is encoded into eight bits, producing a 64-kilobit-per-second channel. A T1 frame contains one eight-bit sample from each of 24 channels plus one framing bit:\n\n\\[24\\times 8+1=193\\text{ bits}.\\]\n\nAt 8,000 frames per second, the resulting line rate is\n\n\\[193\\times 8{,}000=1{,}544{,}000\\text{ bits per second}.\\]\n\nThe E1 system instead operates at 2.048 megabits per second, using 32 eight-bit time slots in each 125-microsecond frame. In a conventional primary-rate telephone arrangement, 30 slots carry bearer channels while the others support framing and signaling functions.\n\nCorrect operation requires both bit timing and frame alignment. Recovering the bit clock identifies individual symbols; recovering the frame boundary determines which symbols belong to which channel. If frame synchronization is lost, correctly detected bits can still be delivered to the wrong output. Framing patterns, alignment procedures, and alarm mechanisms are therefore integral to practical systems rather than optional administrative additions.\n\nHierarchical multiplexing introduces clock relationships between tributary signals and the aggregate. In plesiochronous digital hierarchies, nominally equal-rate inputs may have slightly different actual clock rates. Justification, also called stuffing, accommodates these differences by inserting controlled extra positions. Synchronous digital hierarchy and synchronous optical networking use more coordinated timing structures and pointers to support the mapping and extraction of tributary signals.\n\nThe principal advantage of fixed-slot multiplexing is predictable allocation. Once a channel has a reserved slot, other channels cannot consume that slot merely by becoming busy. Delay and capacity are consequently easier to characterize than in a heavily loaded shared queue. The corresponding disadvantage is that an idle source may leave its allocation unused. Circuit emulation over packet networks can preserve the behavior of a time-division service, but must additionally handle packet-delay variation, loss, and clock reconstruction."
    },
    {
      "heading": "Statistical multiplexing and packet networks",
      "text": "Statistical multiplexing allocates transmission opportunities according to actual demand rather than a fixed reservation for every source. Its effectiveness comes from traffic variability: many sources are not active at the same instant or do not continuously transmit at their peak rates. A shared link can therefore support a collection of intermittent users more economically than a system that permanently reserves each user’s maximum requirement.\n\nPacket networks implement this principle by dividing data into identifiable units and interleaving packets from different conversations. A router receives packets, determines their next destinations, and places them into output queues. Packet headers preserve the information needed to forward and eventually deliver the data. Unlike a synchronous time slot, a packet’s position in the transmission sequence generally does not establish its identity; explicit header information performs that function.\n\nSuppose 100 sources can each generate 1 megabit per second, but each is independently active only 10 percent of the time. Their mean aggregate demand is 10 megabits per second, rather than the 100 megabits per second required to accommodate every source’s peak simultaneously. This does not make a 10-megabit-per-second link sufficient for lossless, delay-free service. The number of simultaneously active sources fluctuates, and periods above the link’s capacity create queues. Correlated activity, such as many clients downloading a newly released file, can greatly weaken the benefit implied by the independence assumption.\n\nBuffers absorb temporary excess demand, but exchange immediate loss for waiting time and memory use. If packets arrive at an average rate \\(\\lambda\\) packets per second and their average size is \\(E[L]\\) bits, a basic offered-load measure for a link of capacity \\(C\\) bits per second is\n\n\\[\\rho=\\frac{\\lambda E[L]}{C}.\\]\n\nFor a simple stationary single-server queue, sustained operation requires \\(\\rho<1\\); even below that threshold, burstiness can produce substantial delay or overflow. Real networks also have finite buffers, changing routes, and traffic with dependencies that a simple average does not capture.\n\nScheduling determines how competing queues share the link. First-in, first-out service is straightforward but provides little differentiation. Priority scheduling favors selected traffic, while weighted fair queuing and deficit round robin distribute service according to configured policies. Packet size also matters: a packet already being transmitted may delay a newly arrived urgent packet until serialization finishes. Statistical multiplexing is thus both an efficiency mechanism and a resource-allocation problem involving fairness, delay, loss, and service guarantees."
    },
    {
      "heading": "Orthogonal, code-based, and spatial techniques",
      "text": "Orthogonal frequency-division multiplexing (OFDM) divides a transmission into many closely spaced subcarriers whose waveforms are orthogonal over a defined symbol interval. Unlike conventional frequency-division arrangements with substantial guard bands between channels, OFDM permits their spectra to overlap. Under ideal timing and frequency alignment, each subcarrier contributes zero interference to the others at the receiver’s corresponding demodulation output.\n\nFor a useful symbol duration \\(T\\), the usual subcarrier spacing is\n\n\\[\\Delta f=\\frac{1}{T}.\\]\n\nTransmitters commonly construct the combined waveform using an inverse discrete Fourier transform, and receivers separate the subcarriers with the corresponding forward transform. A cyclic prefix repeats part of the symbol at its beginning, helping a receiver handle multipath propagation when the significant channel delay spread fits within the designed guard interval. It consumes transmission time, but enables comparatively simple frequency-domain equalization.\n\nOFDM appears in wireless local-area networks, digital broadcasting, and cellular systems. When different groups of subcarriers or time-frequency allocations are assigned to different users, the associated multiple-access method is called OFDMA. Orthogonality is conditional rather than absolute: frequency offsets, phase noise, timing errors, and time-varying channels can cause intercarrier interference. OFDM also commonly produces a high peak-to-average power ratio, placing demands on transmitter amplifiers.\n\nCode-division multiple access separates users through spreading sequences rather than exclusive time slots or frequency bands. A receiver correlates the received signal with the desired user’s sequence. Other users’ signals are suppressed to an extent determined by code relationships, synchronization, received powers, and propagation conditions. Orthogonal codes can lose their orthogonality in multipath environments, while nonorthogonal codes generally create residual multiple-access interference. Power control is important because a strong nearby transmitter can overwhelm a weaker distant one—the near–far problem.\n\nSpatial multiplexing distinguishes signals through separate physical paths or sufficiently different propagation channels. Separate wire pairs and separate optical fibers provide direct spatial separation. Radio systems using multiple-input multiple-output (MIMO) techniques can transmit several streams through multiple antennas in the same time-frequency resource. A simplified model is\n\n\\[\\mathbf y=\\mathbf H\\mathbf x+\\mathbf n,\\]\n\nwhere \\(\\mathbf H\\) represents the propagation channels between transmitting and receiving antennas. Recovering multiple streams depends on the rank and conditioning of this channel matrix, the available signal-to-noise ratio, and receiver processing. Multiple antennas do not automatically yield the same number of independent streams.\n\nThese dimensions can be combined. A cellular system may use spatial streams, OFDM subcarriers, and time scheduling together, while an optical system may combine separate wavelengths with two polarization channels per wavelength."
    },
    {
      "heading": "Multiplexing across Internet protocol layers",
      "text": "The Internet uses multiplexing at several protocol layers, each with its own identifiers and scope. A link carries packets from many network-layer conversations; an Internet Protocol packet identifies its higher-layer protocol; and transport protocols distinguish application endpoints. These nested operations allow a physical interface to serve many applications without requiring one cable or radio channel per application.\n\nIn the TCP/IP architecture, transport-layer port numbers distinguish services and endpoint bindings within a host. An established Transmission Control Protocol connection is commonly identified by its local and remote IP addresses and port numbers, together with the transport protocol context. A server can therefore accept many connections on one listening port because the complete endpoint combinations differ. A port number alone does not identify a globally unique connection, and connection demultiplexing is separate from the routing that delivers a packet to its destination network.\n\nApplication protocols can introduce another level of sharing. HTTP/1.1 commonly used multiple TCP connections to obtain parallel transfers; its pipelining mechanism allowed requests to be sent without waiting for every preceding response, but responses remained ordered. HTTP/2, published in 2015, introduced binary frames and independently identified streams within a connection. Frames from different streams can be interleaved, allowing multiple exchanges under the Hypertext Transfer Protocol to share one transport connection.\n\nHTTP/2 does not remove TCP’s requirement to deliver one ordered byte stream. If a TCP segment is missing, later bytes remain unavailable to the application until the gap is repaired, even when those bytes contain frames for otherwise independent HTTP/2 streams. This is a form of head-of-line blocking. Application-layer multiplexing cannot, by itself, eliminate an ordering dependency imposed by the underlying transport.\n\nQUIC, standardized in 2021, instead provides multiple streams within its own transport design over UDP. HTTP/3 maps HTTP exchanges onto those facilities. Loss affecting one stream does not inherently prevent delivery of available data on another stream merely because it arrived later in the connection. Nevertheless, streams still share connection-level resources, including congestion control and some flow-control constraints, so their performance is not wholly independent.\n\nIdentifiers also define security boundaries. Correct demultiplexing must prevent data for one connection or stream from being delivered to another. Protocol state, connection validation, and encryption mechanisms help enforce those boundaries. Multiplexing can consolidate overhead and connection establishment, but it also concentrates many exchanges within one stateful context whose scheduling and failure behavior affect all participants."
    },
    {
      "heading": "Multiplexers in digital hardware",
      "text": "In digital electronics, a multiplexer usually means a combinational circuit that selects one of several inputs and connects it logically to an output. An \\(n\\)-to-1 multiplexer has \\(n\\) data inputs and enough control inputs to identify the selected source. When \\(n=2^k\\), \\(k\\) binary select signals can encode all choices. This selection function differs from continuously transmitting several recoverable channels together, although repeated selection can implement time-division multiplexing.\n\nA two-input, one-output multiplexer can be expressed as\n\n\\[Y=(\\neg S\\land A)\\lor(S\\land B),\\]\n\nwhere \\(S\\) is the select input. When \\(S=0\\), the output follows \\(A\\); when \\(S=1\\), it follows \\(B\\). The circuit may be implemented using logic gates, transmission gates, or technology-specific cells. A wider datapath uses multiple corresponding selection elements under common control: selecting between two 64-bit words is logically a bank of 64 two-input multiplexers.\n\nMultiplexers are fundamental to processor datapaths. They select register operands, choose between an arithmetic result and loaded data, and determine whether a program counter receives a sequential address or a branch target. In a central processing unit, the control logic produces the selection signals appropriate to the current instruction and execution state. A register file’s read circuitry likewise performs selection among stored values, although its physical implementation may be more specialized than a simple tree of textbook muxes.\n\nA demultiplexer routes one input to one selected output. A decoder is related but not identical: it activates an output corresponding to an encoded input value, without necessarily distributing a separate data signal. Engineers combine selection and decoding functions to control buses, memory access, and peripheral connections. Programmable logic devices also rely heavily on configurable routing and selection structures.\n\nHardware selection has timing and power costs. Larger multiplexers often require multiple logic levels or heavily loaded control signals, increasing propagation delay. Glitches can occur when data and select inputs change along paths with different delays. A combinational multiplexer does not make an asynchronous signal safe to sample: crossing between clock domains requires suitable synchronization or handshaking.\n\nAnalog multiplexers extend the selection idea to continuously varying voltages or currents. A data-acquisition system may select several sensors in turn for one analog-to-digital converter. Switch resistance, leakage, charge injection, source impedance, and settling time then affect accuracy. The converter must receive a sufficiently settled signal after each selection change, so the advertised conversion rate alone does not determine the achievable per-channel measurement rate."
    },
    {
      "heading": "Operating-system and software multiplexing",
      "text": "An operating system multiplexes hardware resources among programs. Processor scheduling allocates execution time, virtual memory separates address spaces, and input/output subsystems coordinate access to devices. These are broader forms of resource sharing rather than necessarily signal-combining operations. Time-sharing systems make the processor available to successive tasks quickly enough that many users or programs can appear to progress together.\n\nIn network programming, I/O multiplexing has a narrower meaning: one execution context monitors multiple input/output endpoints and discovers which can make progress. On Unix-like systems, select and poll report readiness conditions across collections of file descriptors. The program can then read from a socket with available data or write to one with buffer space. This differs from creating a dedicated blocking thread for every connection, although practical systems often combine event-driven processing with multiple threads.\n\nDifferent interfaces have different state and delivery models. In Linux, epoll maintains an interest set and reports events for registered descriptors. BSD-derived systems provide kqueue, while Microsoft Windows supports completion-oriented mechanisms such as I/O completion ports. Readiness notification says that an operation is expected to make progress; completion notification reports the result of an operation already submitted. Treating these as interchangeable can lead to incorrect assumptions about buffer ownership and operation lifetime.\n\nAn event loop coordinates notifications, timers, and application work. The loop waits for events, dispatches the corresponding handlers, and returns to monitoring. Asynchronous programming frameworks often present this behavior through futures, promises, or coroutines, allowing sequential-looking code to suspend while an operation is pending. The underlying system may still use readiness notifications, completion queues, worker threads, or a combination of mechanisms.\n\nCorrectness depends on details beyond receiving an event. A read can return less data than requested, a write may transmit only part of a buffer, and another participant can change an endpoint’s state before a handler runs. Edge-triggered notification commonly requires nonblocking processing until no further immediate progress is possible. Long-running callbacks can delay unrelated connections even when the operating system has reported all their events promptly.\n\nTerminal multiplexers provide another software use of the term. Programs such as tmux maintain multiple terminal sessions accessible through a shared client interface, often allowing sessions to persist after a client disconnects. Their function is session and display management rather than physical bandwidth division. Across these examples, the common principle is maintaining distinct logical activities over fewer underlying interfaces or execution resources."
    },
    {
      "heading": "Capacity, isolation, and design limits",
      "text": "The performance of a multiplexed system depends on what resource is shared and how the allocation is enforced. Relevant quantities include aggregate throughput, per-stream goodput, delay, delay variation, loss, fairness, and implementation cost. Goodput counts useful delivered application data, excluding such items as headers and retransmitted information. A scheme that increases aggregate throughput can still worsen the latency or reliability experienced by an individual stream.\n\nFor a noisy, band-limited channel under the assumptions of the Shannon–Hartley model, the theoretical capacity is\n\n\\[C=B\\log_2(1+S/N),\\]\n\nwhere \\(B\\) is bandwidth and \\(S/N\\) is the signal-to-noise power ratio. Multiplexing partitions or exploits the available communication resources; it does not bypass this limit. A simplified division of a fixed resource gives a payload constraint of the form\n\n\\[\\sum_i R_i\\leq C-H,\\]\n\nwhere \\(R_i\\) are payload rates and \\(H\\) represents equivalent overhead. Real systems require a more detailed model when scheduling, coding, interference, and resource allocation change the achievable rates.\n\nIsolation is often as important as utilization. Reserved time slots provide strong separation of transmission opportunities, while statistically shared queues expose each stream to the others’ demand. Per-stream buffers, rate limits, admission control, and fair schedulers can reduce interference between users. In cloud computing, similar mechanisms address the noisy-neighbor problem, in which one tenant’s resource consumption degrades another’s performance. Sharing without a defined allocation policy does not imply fair sharing.\n\nBuffering presents a recurring trade-off. More storage can absorb larger bursts, but persistent queues increase delay without increasing the bottleneck’s capacity. Backpressure propagates information about downstream limits toward producers, preventing an unconstrained source from overwhelming finite storage. In a multiplexed protocol, both per-stream and connection-wide flow control may be needed: the first limits one stream’s consumption, while the second bounds the aggregate resources associated with the connection.\n\nFailure behavior also changes when many channels share one facility. A failed aggregate link can interrupt numerous services at once; a software connection failure can terminate many application streams. Separate logical channels do not necessarily constitute separate failure domains. Redundant paths, protection switching, replication, and load distribution can reduce this concentration of risk, but require additional coordination and capacity.\n\nEngineering choices therefore depend on workload and service requirements. Continuous fixed-rate signals favor predictable allocations; bursty data often benefits from statistical sharing; weakly separated physical signals demand more capable receivers; and very large connection populations motivate scalable event notification. Practical systems combine mechanisms so that each layer supplies the identification, timing, allocation, and isolation properties needed by the layer above it."
    }
  ],
  "links": [
    "signal-multiplexing",
    "integrated-circuit",
    "data-compression",
    "error-correcting-code",
    "telephone",
    "frequency-division-multiplexing",
    "wavelength-division-multiplexing",
    "router",
    "internet",
    "tcp-ip",
    "hypertext-transfer-protocol",
    "encryption",
    "logic-gate",
    "central-processing-unit",
    "operating-system",
    "time-sharing",
    "thread-computing",
    "linux",
    "event-loop",
    "asynchronous-programming",
    "cloud-computing"
  ],
  "sources": [],
  "license": "https://creativecommons.org/licenses/by/4.0/"
}