Signal multiplexing is the combination of two or more distinguishable signals or information streams so that they can share a transmission medium, communication link, or processing resource. A corresponding separation process, called demultiplexing, recovers or directs the constituent streams at the receiving end. Multiplexing underlies systems ranging from telephone networks and optical backbones to wireless communications, the Internet, and data paths inside an integrated circuit.
The streams may be separated by frequency bands, recurring time slots, optical wavelengths, spreading codes, spatial paths, or identifiers embedded in digital messages. These techniques do not create unlimited capacity: they organize the available resources so that several communications can coexist with acceptable interference, delay, and cost. Practical systems commonly combine several forms of multiplexing, sometimes at different layers of the same network.
The term covers both physical signal combination and logical sharing of a connection. Its meaning therefore depends on context. An optical multiplexer may combine light from several lasers, a digital link may interleave bits from several sources, and an application protocol may carry several independently identified streams over one connection. Their mechanisms differ, but all require rules that make the combined traffic separable.
Principles and conceptual boundaries
A communication channel is a physical or logical means of carrying information. Multiplexing divides or shares some resource associated with that channel: bandwidth, transmission time, optical modes, antenna dimensions, or the capacity to transmit labeled messages. The transmitting arrangement establishes a mapping from input streams to a composite representation. The receiver must know enough about that mapping to distinguish the streams again.
For a linear signal-combination model, the transmitted waveform can be written as
where is the waveform assigned to stream . Successful separation depends on how those waveforms are constructed and how the transmission channel changes them. Signals occupying disjoint frequency bands can be separated by filters; signals occupying different time intervals can be separated by timed switching. More generally, signals may be distinguished by orthogonal or sufficiently different basis functions. Not every multiplexing system is best described by simultaneous waveform addition: packet interleaving, for example, is often more naturally described as a sequence of labeled data units.
Multiplexing is distinct from modulation, although the two are frequently used together. Modulation maps information onto properties of a waveform, such as amplitude, phase, or frequency. Multiplexing determines how several information-bearing signals share a resource. A radio transmitter can modulate one carrier with one data stream without multiplexing independent sources, or combine several modulated carriers into a frequency-division multiplex. Likewise, compression reduces the representation needed for information, whereas multiplexing organizes multiple representations on a shared facility.
A related distinction separates multiplexing from multiple access. Multiplexing often describes combining streams under a common transmitter or controller. Multiple access concerns several transmitters sharing a medium, potentially with separate clocks and unequal received powers. Frequency-division multiplexing and frequency-division multiple access use closely related separation principles, but the latter must also manage distributed access, user assignment, and interference between transmitters. Duplexing addresses a different question: how opposite directions of communication are separated, whether by frequency, time, or other means.
The receiver also needs control information. This may consist of a fixed channel plan, a frame synchronization pattern, a packet header, a code assignment, or a channel estimate. Such information consumes resources but makes demultiplexing possible. Consequently, the useful payload rate is generally lower than the raw line rate. A complete description of a multiplexing system must specify not only its separation dimension but also its synchronization, addressing, overhead, and behavior when the offered traffic exceeds capacity.
Historical development
Multiplexing emerged from the economic importance of sharing expensive communication infrastructure. Nineteenth-century telegraph wires required substantial investment in installation and maintenance, while the messages sent over them were often short and intermittent. Telegraph engineers therefore developed methods for carrying more than one communication over a wire. Duplex telegraphy enabled simultaneous transmission in opposite directions, while quadruplex arrangements combined directional separation with the ability to distinguish two signals in each direction.
Thomas Edison’s quadruplex telegraph, developed in 1874, used distinctions in current direction and strength to support four simultaneous communications on one circuit, two in each direction. It illustrates an early principle that remained central to multiplexing: signals need not occupy entirely separate physical paths if the receiver can distinguish independent electrical characteristics. These systems depended on careful balancing and adjustment because line resistance, capacitance, and leakage affected the separation of signals.
Émile Baudot’s telegraph work in the 1870s combined a five-unit character code with distributor mechanisms that allocated successive transmission opportunities among operators. Synchronized rotating distributors provided a mechanical form of time sharing. Rather than permitting each operator to occupy a separate wire continuously, the system assigned each one a recurring portion of the shared line’s operation. The need to keep sending and receiving distributors aligned anticipated the framing and clock-recovery problems of later digital networks.
During the twentieth century, carrier telephony made frequency separation central to long-distance communication. Voice channels could be shifted into different frequency bands and combined into larger groups on wire pairs, coaxial cables, and microwave radio links. A nominal 4 kHz allocation per telephone channel accommodated the speech band and the filtering needed to separate adjacent channels. Hierarchical grouping allowed networks to handle many calls while keeping channel arrangements standardized. Work at organizations including Bell Laboratories connected these transmission systems with advances in amplifiers, filters, switching, and communication theory.
Digital transmission brought a major change in implementation. The T1 system, introduced commercially in 1962, carried 24 digitized voice channels at a line rate of 1.544 Mbit/s. Later optical systems combined high-rate digital time-division structures with wavelength multiplexing. Packet networks introduced a different allocation principle: instead of reserving a recurring slot for every source, a shared link could transmit whichever source had a packet ready. These approaches did not replace one another completely. Contemporary networks often retain fixed resource allocations in some layers while sharing those allocations statistically in others.
Frequency and wavelength division
Frequency-division multiplexing (FDM) assigns different signals to different frequency bands. Each source can operate continuously within its allocated band, and the transmitted signal is the sum of the band-limited components. At the receiver, filters isolate the desired band, after which frequency conversion and demodulation recover the information. Traditional carrier telephony, cable television distribution, and many radio systems use this general principle.
The frequency allocation must account for real filters. An ideal filter would pass its assigned band and reject everything immediately outside it, but practical filters have transition regions. Guard bands reduce interference between neighboring channels by leaving some spectrum unused. Wider guard bands ease filtering requirements but reduce the proportion of spectrum available for information. Frequency accuracy, transmitter spectral leakage, and nonlinear distortion also influence how closely channels can be packed.
Nonlinearity is especially important because a shared amplifier does not necessarily preserve the independence of its inputs. If two tones at frequencies and enter a nonlinear device, the output may contain combinations such as and . In a many-channel system, such intermodulation products can fall inside other assigned channels. Designers therefore consider the composite waveform and total power, not just the performance of each channel in isolation.
Wavelength-division multiplexing (WDM) applies a corresponding principle to light transmitted through optical fiber. Different optical carriers, usually produced by different lasers, share the fiber and are separated using wavelength-selective components. Frequency and wavelength describe the same optical carrier through the relation in vacuum, although equipment specifications commonly use wavelength for optical components and frequency spacing for dense channel grids. Around a wavelength of 1,550 nm, a 100 GHz frequency separation corresponds to approximately 0.8 nm in wavelength.
Coarse WDM uses relatively widely spaced channels, while dense WDM supports closer spacing and larger channel counts. Optical add-drop multiplexers can remove selected wavelengths from a route and insert others without converting every channel to an electrical signal. Reconfigurable versions change these optical paths under network control. A wavelength is not intrinsically a single user connection: it may carry a digital aggregate that already contains many lower-rate streams.
Optical multiplexing is constrained by laser stability, filter passbands, amplifier noise, chromatic dispersion, and nonlinear effects in the fiber. Increasing the number of wavelengths or the power per wavelength therefore does not increase useful capacity without limit. Modern coherent optical systems combine wavelength separation with advanced modulation, polarization separation, digital equalization, and error correction, making wavelength division one component of a larger transmission architecture.
Time division and synchronous digital hierarchies
Time-division multiplexing (TDM) assigns different input streams different transmission intervals on a common link. In synchronous TDM, those intervals recur according to a predetermined schedule. A frame may contain one slot for each input, followed by the next repetition of the same arrangement. The receiver identifies a stream by its position in the frame rather than by a destination label attached to every sample.
Digital telephony provides a concrete example. In conventional pulse-code modulation for telephone speech, a channel is sampled 8,000 times per second and each sample is commonly represented by eight bits, producing 64 kbit/s. A T1 frame contains 24 eight-bit channel samples plus one framing bit, for 193 bits per frame. At 8,000 frames per second,
The framing bit helps establish the boundaries needed to recover the individual channels. E1 uses a different structure: 32 eight-bit time slots at the same frame rate produce 2.048 Mbit/s, with slots assigned to framing, signaling, or bearer traffic according to the operating arrangement.
The sampling rate is related to the sampling theorem, but a multiplexing link also requires timing discipline beyond the sampling operation itself. The receiver must recover bit timing and determine frame boundaries. If independent tributary clocks differ slightly, a rigid interleaving arrangement cannot indefinitely absorb the mismatch. Older plesiochronous digital hierarchies used justification, or bit stuffing, to accommodate small rate differences. Additional bits and control indications allowed tributaries to be carried within standardized aggregate rates.
The Synchronous Digital Hierarchy and the closely related SONET family introduced standardized synchronous transport structures with extensive management overhead. The basic SDH STM-1 signal operates at 155.520 Mbit/s. Pointer mechanisms identify payload positions and accommodate phase relationships without requiring every payload to remain permanently aligned with the transport frame. These hierarchies supported protection switching, monitoring, and the aggregation of many circuit services over optical networks.
Fixed-slot TDM provides predictable service because a source’s transmission opportunities do not depend directly on the activity of other sources. That property is valuable for continuously timed traffic. Its disadvantage is that a reserved slot may remain unused when its source is idle. Dynamic slot assignment can improve utilization, but it requires additional control and changes the delay characteristics. Time-division multiple access extends time separation to several transmitters, adding requirements such as guard intervals, timing advance, and control over which transmitter may occupy each interval.
Statistical multiplexing and packet networks
Statistical multiplexing allocates shared capacity according to current demand rather than reserving a fixed share for every input. It is particularly effective for bursty sources: an interactive terminal, web application, or sensor may transmit briefly and then remain silent. When their busy periods do not coincide completely, many such sources can share a link whose rate is lower than the sum of their individual peak rates.
Packet switching implements this principle by dividing traffic into separately handled units. Packets or frames from several sources enter queues and are selected for transmission over the same outgoing link. Headers identify their destinations or logical connections, allowing receivers and intermediate devices to separate them. In an Ethernet network, frames carry addressing and protocol information; a router forwards network-layer packets toward their destinations. Neither mechanism requires every source to own a permanently reserved time slot.
The statistical advantage depends on traffic behavior. If sources become active together, the aggregate demand may exceed link capacity. Queues absorb temporary excesses, but finite buffers eventually overflow unless traffic subsides or the sending rate falls. In a simplified single-server model, stability generally requires the average offered workload to remain below the service capacity. This condition does not ensure low delay: variable packet sizes and bursty arrivals can produce substantial waiting times even when average utilization is below 100 percent.
Scheduling policies determine how contention is resolved. First-in, first-out scheduling serves packets in arrival order; priority scheduling favors selected classes; fair-queuing methods approximate controlled sharing among flows. Quality-of-service mechanisms may also police traffic, shape bursts, or reserve resources for particular classes. Statistical multiplexing therefore does not mean the absence of guarantees. It means that guarantees must be established through explicit traffic assumptions and resource management rather than inferred from a permanent physical slot.
Multiplexing also occurs within end-to-end protocols. TCP uses port numbers, together with network addresses and other connection context, to distinguish connections at endpoints. HTTP/2, standardized in 2015, carries multiple application streams over one TCP connection using stream identifiers. However, loss of TCP data can delay delivery of later bytes belonging to different HTTP/2 streams, a form of head-of-line blocking. QUIC, standardized in 2021, provides multiple independently ordered streams so that missing data in one stream does not inherently prevent delivery of available data in another. Those streams still share connection-level resources such as congestion control, so independence of ordering does not imply independence of capacity.
Orthogonal multicarrier transmission
Orthogonal frequency-division multiplexing (OFDM) transmits data over many closely spaced subcarriers whose waveforms are orthogonal over a specified observation interval. Unlike conventional FDM with substantial separation between individual channel spectra, OFDM permits subcarrier spectra to overlap. Orthogonality allows an ideal receiver to separate them despite that overlap, provided timing and frequency conditions are satisfied.
For a useful symbol interval of duration , a common subcarrier spacing is
The complex exponentials representing different subcarriers then satisfy
The receiver exploits this relation by projecting the received symbol onto the subcarrier basis. Practical digital systems implement the operation using the discrete Fourier transform, usually calculated with a fast Fourier transform algorithm. An inverse transform at the transmitter constructs a time-domain block from the frequency-domain symbols.
Multicarrier techniques were investigated before integrated digital processors made them economical. A significant development was the 1971 work of Stephen Weinstein and Paul Ebert on discrete Fourier transform implementation for frequency-division data transmission. Transform processing replaced large banks of separately implemented modulators and demodulators with a structured computation. Subsequent advances in digital signal processing helped make multicarrier methods practical in digital subscriber lines, broadcasting, wireless local-area networks, and cellular systems.
A principal advantage is the handling of frequency-selective channels. Reflections can cause delayed copies of a radio signal to overlap, producing intersymbol interference. OFDM divides a wideband channel into narrower subchannels and usually adds a cyclic prefix: a copy of the end of a symbol placed at its beginning. If the prefix is long enough for the channel’s significant delay spread and the receiver is correctly aligned, equalization becomes approximately a separate complex scaling operation on each subcarrier. The prefix consumes transmission time and therefore reduces payload efficiency.
OFDM does not necessarily imply that each subcarrier carries a different user. A single user’s coded data can occupy the entire set. Orthogonal frequency-division multiple access assigns different time-frequency resource elements to different users. The distinction matters because a multicarrier waveform and multiuser resource allocation are separate design choices.
Important limitations include sensitivity to carrier-frequency error, phase noise, and rapid channel variation, all of which can destroy orthogonality and cause intercarrier interference. The sum of many subcarriers can also produce a high peak-to-average power ratio, demanding amplifier headroom or distortion-control techniques. Windowing, filtering, synchronization, pilot placement, and coding are therefore integral to practical implementations.
Code division and spread-spectrum separation
Code-domain multiplexing distinguishes streams by assigning different sequences or signatures rather than exclusively separate time intervals or frequency bands. In code-division multiple access (CDMA), users may transmit over the same nominal bandwidth at the same time. A receiver extracts a desired signal by exploiting knowledge of its code and the statistical or algebraic relationships between that code and the others.
A common implementation uses direct-sequence spread spectrum. Each data symbol is multiplied by a faster sequence of chips, spreading its energy over a wider bandwidth. For a simplified synchronous model with normalized code vectors and transmitted symbols , the received vector is
where represents a channel gain and represents noise. Correlation with the desired code produces a desired term plus contributions proportional to cross-correlations with the other codes. Orthogonal codes eliminate those contributions only when the assumed alignment and channel conditions hold.
Processing gain is commonly expressed as the ratio of chip rate to data-symbol rate, or equivalently as a related spread-bandwidth ratio. It describes how spreading and correlation distinguish a desired low-rate signal from some forms of interference. It is not free capacity: spreading uses more bandwidth, and a growing population of simultaneous users raises the interference level. The achievable number of users depends on received powers, activity factors, coding, propagation, and the required error rate.
The near–far problem is a central engineering issue. A nearby transmitter received at much greater power can overwhelm a weaker desired user if code cross-correlation is not zero. Cellular CDMA systems therefore use power control to regulate received signal strengths. In multipath environments, delayed signal copies have different code alignments. RAKE receivers exploit resolvable delayed components by correlating against them separately and combining their contributions, although multipath can also weaken orthogonality between users.
Code-based separation appeared in systems including IS-95 cellular service and later third-generation mobile technologies. Satellite navigation provides another example: GPS receivers distinguish signals using known spreading codes and estimate their arrival times. In that setting, code structure supports identification and ranging as well as coexistence.
Code assignment is not equivalent to encryption. A publicly known spreading code can provide separation without confidentiality. Even when a sequence is difficult to detect or predict, security claims require a separate analysis of key management and adversary capabilities. Spread spectrum likewise improves resistance to certain interference conditions without making a link immune to sufficiently strong or appropriately structured jamming.
Spatial and polarization multiplexing
Spatial multiplexing separates streams through distinct propagation paths or channel responses. Its simplest form uses separate physical media, such as several wire pairs or several fibers in one cable. These channels share installation infrastructure but not necessarily the same electromagnetic propagation mode. More elaborate systems transmit several streams over the same nominal frequency band and time interval while relying on antennas, optical modes, or other spatial structure to distinguish them.
Multiple-input multiple-output (MIMO) communication uses multiple transmit and receive antennas. A narrowband baseband model is
where contains transmitted signals, contains received signals, and describes the channel gains between antenna pairs. Recovering several independent streams depends on the number and strength of independent dimensions in , not simply the antenna count. Two antennas with nearly identical channel responses may provide little additional spatial-multiplexing capability.
Channel estimation is therefore essential. Receivers use known pilot signals to estimate the channel and then apply detection or equalization methods to separate streams. When suitable channel information is available at the transmitter, precoding can choose transmit combinations matched to the propagation environment. Singular-value decomposition provides a mathematical way to describe independent spatial modes, although practical systems use finite feedback, standardized codebooks, and constrained computational procedures rather than ideal unconstrained operations.
Beamforming is related but not identical to spatial multiplexing. Beamforming combines antenna signals to shape transmission or reception in particular directions, potentially improving signal strength or suppressing interference. An antenna array can beamform a single data stream, multiplex several streams, or do both. Multiuser MIMO directs different streams to different users, making scheduling and the accuracy of channel knowledge important to separation. Movement, feedback delay, and channel similarity can reduce the effectiveness of the intended isolation.
Polarization provides another separation dimension. Two orthogonal electromagnetic polarizations can carry separate signals in the same nominal frequency band. Practical radio links experience polarization mismatch and depolarization, which cause leakage between the channels. Coherent optical communication commonly uses polarization-division multiplexing, with digital processing tracking changes in polarization and separating the received components.
Optical spatial-division techniques include multiple cores within one fiber and multiple guided modes within a core. These arrangements can increase aggregate transport capability but may introduce coupling between channels and require additional signal processing. Whether an implementation uses separate antennas, cores, or modes, spatial multiplexing depends on distinguishability after propagation. A nominally separate launch path does not guarantee an independently recoverable received channel.
Multiplexers in circuits and layered systems
In digital electronics, a multiplexer is a circuit that selects one of several inputs and connects it to an output. For a two-input, one-output binary multiplexer with selection input , the output can be expressed as
It may be built from logic gates or transistor-level switching structures. A larger multiplexer uses additional selection bits to choose among more inputs. The term describes a selection function; the device does not necessarily transmit several independent signals simultaneously.
Time-controlled selection turns this circuit function into a means of combining streams. A controller can select different inputs on successive clock cycles, while a receiving demultiplexer routes the corresponding output values to their destinations. Parallel-to-serial conversion and serial-to-parallel conversion are closely associated operations, although their implementation may use shift registers rather than a single explicit selection tree. High-speed serializer/deserializer circuits combine these functions with clock recovery, equalization, and line coding.
Analog multiplexers switch continuously varying voltages or currents. A data-acquisition system may use one analog-to-digital converter to sample several sensors in turn. This reduces converter count but creates constraints that do not arise with independent converters. Switch resistance, leakage, charge injection, and settling time can alter the measurement. If the source impedance is high or the converter input must charge a sampling capacitor, the system may need a longer acquisition interval after changing channels. Moreover, sequential channel sampling is not simultaneous sampling; this distinction matters when comparing rapidly changing signals.
Physical implementation also raises signal-integrity issues. On high-speed interconnects, impedance discontinuities, reflections, crosstalk, and attenuation can corrupt the combined stream. A single line fault may then affect every logical channel carried on that line. Conversely, using fewer physical connections can reduce connector size, pin count, and routing complexity. Multiplexing therefore trades hardware duplication against timing, control, and bandwidth requirements rather than eliminating cost outright.
Complete systems commonly nest several multiplexing layers. A web exchange can occupy a protocol stream, share a transport connection with other streams, traverse packet-switched links, and ultimately be carried within a wavelength on an optical fiber. A wireless device may combine application-stream multiplexing, packet scheduling, OFDM, and spatial multiplexing. Each layer has its own identifiers and recovery rules. Correct separation at one layer does not automatically provide fairness, confidentiality, or fault isolation at another, and troubleshooting requires identifying which layer’s sharing mechanism is responsible for the observed behavior.
Capacity, delay, and engineering trade-offs
Multiplexing improves the organization and utilization of resources, but it remains subject to communication limits. In information theory, the Shannon–Hartley theorem gives the capacity of an ideal band-limited additive white Gaussian noise channel as
where is bandwidth in hertz and is the signal-to-noise power ratio over that bandwidth. Dividing a fixed channel among more streams does not remove this constraint. If additional antennas, fibers, or usable frequency bands are introduced, capacity may increase because the underlying physical resources have changed.
The relevant efficiency measure depends on the system. Spectral efficiency expresses information rate per unit bandwidth, often in bit/s/Hz. Payload efficiency compares useful information with total transmitted bits or symbols. Statistical utilization measures how effectively capacity is occupied over time. A design can improve one measure while worsening another: smaller packets may reduce the time an urgent message waits behind a long packet, but their greater header overhead lowers payload efficiency.
Delay also has several components. A fixed-slot system can impose a wait until the next assigned slot, while a packet system adds a workload-dependent queueing delay. Interleaving data over a larger block may improve resistance to burst errors but increase latency. Guard intervals, framing, code acquisition, channel estimation, and retransmission introduce further costs. For interactive audio, industrial control, or distributed computation, variation in delay can be as important as the average rate.
Reliable demultiplexing requires synchronization and isolation margins. Clock errors can move a receiver away from the correct time slot; oscillator errors can disturb frequency orthogonality; propagation changes can invalidate a spatial channel estimate. Error-correcting codes help recover corrupted data but cannot always repair the loss of framing or the persistent misidentification of streams. Systems therefore combine coding with synchronization markers, pilot signals, integrity checks, and procedures for reacquiring the channel after disruption.
Resource sharing also creates failure and security relationships. One failed shared amplifier, scheduler, or fiber can interrupt many logical services. Traffic from one user can increase another user’s delay, and shared queues may reveal information through timing or congestion behavior. Logical channel identifiers are not authorization mechanisms, so access control and cryptographic protection must be supplied where required.
The appropriate multiplexing arrangement consequently depends on the traffic and the physical medium. Continuous fixed-rate sources favor predictable allocations; highly intermittent sources benefit from statistical sharing; frequency-selective radio channels benefit from suitable multicarrier processing; and separable antenna or optical paths can support spatial streams. Practical engineering selects combinations under constraints on power, spectrum, equipment complexity, latency, resilience, and the degree of isolation required between services.