Modulation is the controlled variation of one or more properties of a waveform according to another signal. In telecommunications, it usually means representing information through changes in the amplitude, frequency, phase, or timing of a transmitted waveform. The information may be continuously varying, such as speech, or discrete, such as a sequence of binary digits. The reverse operation, recovering the information from the received waveform, is called demodulation.
Modulation makes it possible to place signals in suitable frequency bands, share a transmission medium, and exchange bandwidth, power, reliability, and implementation complexity. It is fundamental to broadcasting, cellular networks, satellite links, cable systems, and optical communications. Related techniques serve radar, instrumentation, and power conversion, where the purpose may be measurement or control rather than communication.
A modulation method cannot be judged independently of its channel and receiver. A waveform that uses spectrum efficiently may demand a highly linear amplifier, while one that tolerates nonlinear amplification may occupy more bandwidth. Modern systems therefore combine modulation with filtering, synchronization, error-control coding, and adaptation to changing channel conditions.
Signal representation and frequency translation
A carrier wave is a periodic waveform whose properties can be varied to represent a message. For a sinusoidal carrier, a general real-valued transmitted signal can be written as
where is the amplitude, is a reference carrier frequency, and is the phase variation. Amplitude modulation changes ; phase modulation changes . Frequency modulation changes the instantaneous frequency, which is related to phase by
These descriptions overlap mathematically but correspond to different relationships between the message and the transmitted waveform.
The message before frequency translation is commonly called a baseband signal. Speech, for example, contains components at audio frequencies, whereas a radio system may transmit its representation at hundreds of megahertz or several gigahertz. Translating the signal makes transmission compatible with the assigned spectrum, propagation conditions, and practical antenna dimensions. Modulation does not necessarily require a visibly distinct carrier spectral line: a suppressed-carrier signal can occupy a radio-frequency band without transmitting an unmodulated sinusoid at its center.
The Fourier transform explains the frequency-domain effect. Multiplying a message by a cosine produces translated copies of its spectrum:
This operation underlies mixers and many analog modulators. A mixer used only to shift an already modulated signal to another frequency performs frequency conversion rather than introducing new information.
A particularly useful representation is
where and are the in-phase and quadrature components. Together they form the complex envelope , expressed using a complex number. Their orthogonal carriers allow independent components to occupy the same frequency band and provide a common description of most modern digital modulation formats.
Modulation is distinct from source coding, which changes how information is represented, and from channel coding, which adds structured redundancy. It is also distinct from multiplexing, which combines streams for shared transport. In actual equipment these operations are closely integrated: a modem may compress data elsewhere in the system, encode bits, map them to symbols, shape the waveform, and translate it to the transmission frequency.
Amplitude modulation and sideband methods
In conventional amplitude modulation (AM), the carrier amplitude follows a message while the carrier frequency and phase remain fixed. For a normalized message , with magnitude no greater than one, a standard expression is
where is the modulation index. When , the ideal envelope does not reverse sign, allowing a simple envelope detector to recover the message. If the envelope crosses through zero because of excessive modulation depth, an ordinary envelope detector introduces distortion.
For a single sinusoidal message at frequency , the spectrum consists of the carrier at and sidebands at and . A message whose highest frequency is therefore produces a conventional AM bandwidth of approximately . The upper and lower sidebands contain redundant representations of the same real-valued message. Their appearance is not a hardware imperfection but a direct consequence of multiplying the message by a real carrier.
Conventional AM sacrifices power efficiency for receiver simplicity. With sinusoidal modulation, the total transmitted power is
At 100 percent modulation, the two sidebands together contain one-third of the total power; the remaining two-thirds is in the carrier, which contains no message variation. This accounting applies to the specified single-tone case, rather than to every possible audio waveform. Nevertheless, it illustrates why transmitting a strong carrier is costly when transmitter power is scarce.
Double-sideband suppressed-carrier modulation removes the carrier component by transmitting a signal proportional to . It retains both sidebands and therefore retains approximately the same bandwidth as ordinary AM, but it requires coherent detection or another suitable reconstruction method. Single-sideband modulation goes further by transmitting only one sideband. For the same message bandwidth , it needs approximately of radio-frequency bandwidth and can concentrate transmitted power in the information-bearing waveform. It is widely used for long-distance voice communication in high-frequency radio.
Vestigial-sideband modulation preserves one sideband and a controlled portion of the other, easing filter requirements near the carrier. It was important in analog television, where a full double-sideband video signal would have consumed excessive spectrum. These related methods demonstrate that “AM” describes a family of spectral and receiver arrangements, not merely the familiar broadcast envelope waveform. Their practical differences concern occupied bandwidth, transmitted carrier power, filter realizability, and the precision required of the receiver’s local reference.
Frequency and phase modulation
Frequency modulation (FM) makes instantaneous frequency proportional to the message:
where is the frequency sensitivity. In phase modulation (PM), the message directly changes phase:
Both are forms of angle modulation and ideally maintain a constant envelope. Their relationship follows from differentiation: the frequency deviation of a PM signal depends on how rapidly the message changes, whereas its phase deviation depends directly on the message amplitude.
For single-tone FM, the modulation index is , where is peak frequency deviation and is message frequency. Unlike sinusoidal AM, sinusoidal FM generally produces many sideband pairs, with amplitudes described by Bessel functions. Its mathematical spectrum extends indefinitely, although distant sidebands may carry negligible power. Carson’s rule estimates a useful occupied bandwidth as
where is the highest significant message frequency. This is an engineering approximation rather than a strict spectral boundary.
Analog FM broadcasting provides a familiar scale. Many systems use a maximum deviation of about kHz; a simple monophonic audio example with kHz gives a Carson estimate of kHz. Stereo and supplementary services complicate the composite baseband spectrum, so that calculation is not a complete emission specification. FM stereo places a sum signal in the ordinary audio band and a difference signal on a suppressed 38 kHz subcarrier, with a 19 kHz pilot assisting stereo decoding.
The constant envelope permits efficient nonlinear power amplification and allows amplitude limiters to suppress some unwanted amplitude variations before demodulation. Above the receiver’s threshold region, wideband FM can deliver improved output signal-to-noise ratio by spending more bandwidth. At low input signal levels, however, performance deteriorates sharply; FM does not provide unlimited immunity to noise. Its capture effect, in which a stronger co-channel signal dominates reception, also differs from the simultaneous interference typical of AM.
Pre-emphasis and de-emphasis help manage the increasing high-frequency noise of FM demodulation. The transmitter boosts higher audio frequencies according to a specified characteristic, and the receiver applies the complementary reduction. Common broadcast time constants are 50 or 75 microseconds, depending on the system. Angle modulation is consequently a coordinated transmitter–receiver technique: deviation limits, baseband filtering, emphasis, and detection must be designed together.
Digital symbols and signal constellations
Digital modulation maps discrete information onto distinguishable waveforms called symbols. A symbol may represent one bit or several bits, and its physical representation can involve amplitude, phase, frequency, or a combination. For a fixed-length mapping with equiprobable alternatives, each symbol represents bits. The symbol rate, measured in baud, is therefore not generally equal to the bit rate. Before coding and framing overhead, 64-symbol modulation at one million symbols per second carries six million bits per second.
Phase-shift keying (PSK) represents symbols through discrete carrier phases. Binary PSK uses two phases separated by . Quadrature PSK uses four phases and can carry two bits per symbol. In a constellation diagram, these alternatives appear as points in the in-phase–quadrature plane. Such diagrams show both the information mapping and the separation available to resist noise. Gray labeling often assigns adjacent points bit patterns differing in one position, reducing the number of bit errors associated with common nearest-neighbor symbol errors.
Quadrature amplitude modulation (QAM) varies both quadrature components to form two-dimensional constellations. Common square formats include 16-QAM, 64-QAM, and 256-QAM, representing four, six, and eight bits per symbol respectively. Increasing constellation size improves the number of bits conveyed per symbol, but at fixed average energy the points become more closely spaced. A given disturbance then has a greater chance of moving a received point into the wrong decision region. High-order QAM consequently requires cleaner channels, more accurate synchronization, and lower transmitter and receiver distortion.
Frequency-shift keying (FSK) represents symbols by different frequencies. Binary FSK uses two alternatives; larger alphabets use more. Depending on frequency spacing and waveform design, FSK may be detected coherently or without a carrier-phase reference. Continuous-phase forms avoid abrupt phase discontinuities and can support efficient power amplification. Minimum-shift keying is a continuous-phase binary FSK format with modulation index ; Gaussian-filtered minimum-shift keying was adopted for GSM cellular radio.
A constellation alone does not fully specify a transmitted signal. Pulse shapes determine how symbols join in time and how much spectrum they occupy. Coding, interleaving, and labeling determine how bit patterns relate to likely errors. Differential modulation can encode information in changes between symbols rather than absolute phase, reducing some reference ambiguities at a possible performance cost. Adaptive systems switch among modulation orders as channel quality changes, but the actual delivered rate also depends on coding rate, retransmissions, pilot symbols, and scheduling.
Pulse modulation, sampling, and switching
Pulse modulation represents information through properties of pulses rather than necessarily through a continuous sinusoidal carrier. Pulse-amplitude modulation varies pulse height; pulse-width modulation varies duration; pulse-position modulation varies timing. These can carry continuously varying or discrete information. Their names describe the physical waveform, not whether the original source was analog or digital. A sampled but unquantized pulse-amplitude signal, for example, is discrete in time while retaining a continuous range of amplitudes.
Sampling connects continuous waveforms to digital representations. Under the ideal conditions of the sampling theorem, a strictly band-limited signal with highest frequency can be reconstructed from uniformly spaced samples at a rate greater than . Real systems require anti-aliasing filters and transition bands rather than an unattainable ideal cutoff. Sampling alone does not create binary data: quantization assigns samples to a finite set of levels, after which those levels can be encoded.
Pulse-code modulation (PCM) combines sampling, quantization, and coding. Conventional digital telephony commonly samples speech at 8,000 samples per second and represents each sample with eight bits, producing a 64 kbit/s channel before additional framing. Companding laws such as A-law and μ-law allocate quantization precision nonuniformly to accommodate the wide amplitude range of speech. For an ideal uniform quantizer driven by a full-scale sine wave, a frequently used approximation is
where is the number of quantization bits. This formula depends on assumptions about signal amplitude, quantization error, and the absence of overload.
PCM is not, by itself, a complete radio modulation scheme. PCM bits may subsequently be transmitted using QPSK, QAM, or another waveform. Conversely, multilevel pulse-amplitude modulation can directly carry symbols over a wire or an optical intensity channel. In that setting, each amplitude level represents a symbol rather than an unquantized sample. The same broad terminology therefore spans distinct stages of a communication chain.
Pulse-width modulation (PWM) has a particularly important role in power electronics. A switch alternates between states while its duty ratio controls average voltage or transferred energy. For an ideal two-level waveform switching between zero and , the period-average voltage is , where is the fraction of the period spent high. Inductors, capacitors, or mechanical inertia smooth the switching waveform. Inverters, motor drives, and lighting controllers exploit this principle, although dead time, switching losses, electromagnetic interference, and load dynamics prevent the average-value model from describing every detail.
Multicarrier modulation and spread spectrum
Multicarrier modulation divides a stream among several subcarriers rather than sending every symbol through one high-rate waveform. Its major advantage appears in channels with delayed echoes: a long symbol on each narrowband subcarrier is easier to manage than a short broadband symbol distorted by the same delay spread. The subcarriers need not be separated by unused guard bands if their waveforms are orthogonal over the receiver’s observation interval.
Orthogonal frequency-division multiplexing (OFDM) commonly uses subcarrier spacing , where is the useful symbol duration. Over this interval, ideal subcarriers do not interfere at the designated demodulation frequencies even though their spectra overlap. An inverse discrete Fourier transform efficiently synthesizes the transmitter waveform; a corresponding transform separates subcarriers at the receiver. Each subcarrier may carry a QAM or PSK symbol, potentially with a different loading according to channel conditions.
A cyclic prefix copies the end of an OFDM symbol to its beginning. If it is sufficiently long relative to the channel’s effective delay spread, and synchronization is suitable, it suppresses interference from the preceding symbol within the useful observation interval. It also makes the channel convolution behave cyclically over that interval, allowing a frequency-selective channel to be equalized largely through one complex coefficient per subcarrier. The prefix consumes time and energy without adding new payload information, so longer protection reduces efficiency.
OFDM is used in Wi-Fi, digital broadcasting, and cellular systems. Orthogonal frequency-division multiple access allocates subsets of time–frequency resources to different users; it is a multiple-access arrangement built on multicarrier modulation. OFDM also has costs. The sum of many subcarriers can exhibit a high peak-to-average power ratio, demanding amplifier back-off or waveform processing. Frequency offsets and rapid channel variation can destroy orthogonality and produce intercarrier interference. Thus, resilience to multipath does not imply equal resilience to every impairment.
Spread-spectrum methods approach bandwidth use differently. Direct-sequence spreading multiplies a lower-rate information waveform by a faster spreading sequence, producing a wider signal that a matched receiver can despread. Frequency hopping changes the operating frequency according to a sequence. These methods can distribute interference effects and support particular access arrangements, but the spreading sequence is not inherently encryption. Chirp-based signaling deliberately sweeps frequency and supports matched-filter processing, with applications in both communications and radar. Multicarrier and spread-spectrum methods are not mutually exclusive: they describe different aspects of waveform construction and can be combined.
Demodulation, synchronization, and equalization
A receiver must estimate the information-bearing waveform despite noise, interference, distortion, and uncertainty about timing and frequency. Demodulation therefore involves more than applying the inverse of the transmitter’s nominal mapping. A typical receiver performs channel selection, amplification, frequency conversion, filtering, sampling, synchronization, channel estimation, and symbol detection. Their order and implementation vary, but errors in one stage can limit every later stage.
Coherent detection uses a phase-related carrier reference. Multiplying a received signal by properly aligned sine and cosine references and filtering the results recovers its quadrature components. The reference may come from pilot signals, known training sequences, or carrier-recovery algorithms. A phase-locked loop continuously adjusts an oscillator according to measured phase error and is widely used in carrier and clock recovery. Frequency error rotates a received constellation over time, while phase noise creates fluctuating angular errors that are especially troublesome for closely spaced constellation points.
Noncoherent methods avoid the need for an exact carrier-phase reference. An AM envelope detector and certain FSK energy detectors are examples. Differential detection compares successive symbols, transferring the problem from absolute phase estimation to relative phase estimation. These methods can simplify acquisition or hardware, but their noise performance and channel sensitivity differ from those of coherent detection. The appropriate comparison depends on the waveform and the information available to the receiver.
For known signals in additive white Gaussian noise, a matched filter maximizes the output signal-to-noise ratio at the sampling instant. Its impulse response is proportional to a time-reversed, conjugated version of the expected signal. Practical systems often split a desired raised-cosine response between root-raised-cosine transmitter and receiver filters. With suitable timing and an undistorted channel, the combined response satisfies a zero-interference condition at the symbol decision instants. Bandwidth limitation between those instants remains compatible with clean decisions at the correct times.
Real channels introduce intersymbol interference when delayed or filtered contributions from other symbols overlap the current decision. Equalizers counteract this effect using estimated channel properties. They may operate in the time domain or frequency domain and may use linear filtering, decision feedback, or sequence estimation. In a software-defined radio, many of these functions are implemented through programmable signal processing, while the antenna interface and some frequency-selective and amplification stages remain physical hardware. Soft-output detectors further report confidence information rather than only hard symbol decisions, allowing subsequent decoders to exploit uncertainty.
Performance limits and engineering trade-offs
Modulation performance is commonly described through bit-error rate, symbol-error rate, spectral efficiency, required signal-to-noise ratio, and occupied bandwidth. No single measure captures every system objective. A satellite transmitter may prioritize energy efficiency and amplifier operation; a dense terrestrial network may prioritize spectrum reuse and throughput; a battery-powered sensor may prioritize total energy consumed during acquisition, transmission, and reception. These objectives can favor different modulation formats even at the same nominal data rate.
A central theoretical reference is the Shannon–Hartley theorem for an additive white Gaussian noise channel:
where is channel capacity in bits per second, is bandwidth in hertz, and is the signal-to-noise ratio over that bandwidth. The theorem describes a limit attainable asymptotically with appropriate coding under its assumptions. It does not state that any named modulation scheme achieves that rate, nor that arbitrary real channels can be reduced to the same simple expression.
For coherent binary PSK with ideal synchronization in additive white Gaussian noise, the uncoded bit-error probability is
where is energy per bit, is noise spectral density, and is the standard Gaussian tail function. Such formulas isolate specific impairments and make comparisons possible. They must not be transferred unchanged to channels with fading, phase uncertainty, nonlinear distortion, or coding. Comparing formats also requires clarity about whether energy per bit refers to information bits or transmitted coded bits.
Error-correcting codes add redundancy that permits a receiver to infer transmitted data despite some unreliable observations. Modern systems jointly select modulation order and coding rate: 64-QAM with heavy redundancy does not deliver six information bits for every transmitted symbol. Pilots, guard intervals, control signaling, and retransmissions further reduce payload efficiency. Probabilistic constellation shaping changes the frequency with which constellation points are used, often favoring lower-energy symbols, and can improve the match between signaling statistics and a channel constraint.
Hardware imposes additional trade-offs. High-order QAM is sensitive to gain compression, quadrature imbalance, and phase noise; a constant-envelope waveform can better tolerate certain nonlinear amplifiers. Error-vector magnitude measures the deviation of received or measured symbols from ideal reference points, while adjacent-channel leakage measures unwanted radiation outside the assigned channel. A system may therefore satisfy an error-rate target yet fail a spectral-emission requirement. Practical selection balances these constraints with bandwidth, power, latency, implementation cost, and the reliability demanded by the service.
Historical development
Modulation developed alongside the transition from simple electrical signaling to continuous-wave radio and then digital communications. Nineteenth-century telegraphy already represented messages through controlled changes in current. Early wireless spark transmitters produced damped oscillations and were suited chiefly to keyed signaling. Continuous-wave sources made it possible to control a sustained carrier more precisely and to transmit speech or music by varying its amplitude.
Reginald Fessenden achieved wireless speech transmission in 1900 and pursued high-frequency alternators for continuous-wave operation. Valdemar Poulsen’s arc transmitter, introduced in the early twentieth century, provided another continuous-wave source. Vacuum-tube oscillators and amplifiers subsequently made electronic generation and modulation more practical. Broadcasting in the 1920s established conventional AM as a major public medium, partly because receivers could recover audio with relatively simple circuits.
Single-sideband methods developed as engineers sought to reduce the bandwidth and power consumed by long-distance voice transmission. John Renshaw Carson’s 1915 patent application is an important early contribution to this history. Edwin Howard Armstrong’s wideband FM work, including patents granted in 1933, demonstrated a different route to high-quality radio reception through angle modulation and wider bandwidth. Adoption depended on transmitters, receivers, frequency allocations, and commercial infrastructure, not just the merits of an isolated waveform.
Digital modulation drew on developments in sampling, information theory, and computing. Alec Reeves proposed pulse-code modulation in 1937, although economical large-scale implementation required later electronic technology. Harry Nyquist’s 1920s work addressed signaling rate and interference, while Ralph Hartley’s 1928 paper examined information transmission quantitatively. Claude Shannon’s 1948 information theory established a framework for distinguishing channel capacity from the performance of particular transmitters and receivers. These contributions made bandwidth, noise, redundancy, and information rate parts of a common mathematical problem.
As digital circuits became faster and cheaper, modems progressed from relatively simple keyed waveforms to sophisticated phase and amplitude constellations with adaptive equalization. Robert Chang’s 1966 work on orthogonal multichannel data transmission and the 1971 work of Stephen Weinstein and Paul Ebert on discrete-Fourier-transform implementation were important steps toward practical OFDM. Later advances in integrated circuits enabled mass-market cellular and wireless-network equipment to perform synchronization, multicarrier processing, and decoding in real time. Analog and digital modulation consequently coexist rather than forming a simple replacement sequence: the preferred technique continues to depend on service requirements and available equipment.
Optical, measurement, and control applications
Optical communication applies the same broad principles to electromagnetic waves at much higher frequencies. In intensity modulation, transmitted optical power varies with the information signal, and a photodiode detects the resulting power changes. Direct detection does not preserve the optical field’s phase in the way a coherent receiver does. Because optical intensity cannot be negative, electrical baseband techniques must be adapted to the nonnegative transmitted-power constraint. Binary on–off keying and multilevel pulse-amplitude modulation are important examples.
Coherent optical systems mix the received light with a local laser and recover information associated with field amplitude and phase. They can use QPSK or higher-order QAM, frequently with independent data on two polarizations. Digital processing compensates for impairments such as chromatic dispersion and estimates carrier-frequency and phase differences. In optical fiber systems, achievable performance also depends on attenuation, amplifier noise, polarization effects, and optical nonlinearities; increasing launch power does not indefinitely improve reception. Wavelength-division multiplexing combines channels at different optical wavelengths and is distinct from the modulation used within each channel.
Measurement systems use modulation to make otherwise difficult signals easier to distinguish. A weak optical or electrical signal may be modulated at a known frequency and recovered through phase-sensitive detection, reducing interference from unrelated frequencies and low-frequency drift. The benefit depends on the noise spectrum and the reference stability; modulation does not remove noise that falls within the detection bandwidth. In spectroscopy, controlled wavelength or amplitude variation can reveal small absorption changes against a much larger background.
Radar and lidar use deliberately structured waveforms to estimate range and velocity. In an ideal linear frequency-modulated continuous-wave radar, a stationary target’s round-trip delay produces a beat frequency approximately , where is the sweep rate. Since ,
Target motion introduces a Doppler contribution, requiring suitable processing or multiple sweeps to separate range and velocity. Here modulation helps encode propagation delay into a measurable frequency difference rather than carrying a conventional message.
In industrial control and energy conversion, the information-bearing aspect may be subordinate to regulating physical output. PWM drives an electric motor through controlled switching, while modulated excitation can help separate a sensor response from offsets and interference. These applications connect modulation with control theory, but the concepts remain distinct: modulation constructs a waveform, whereas a controller determines how the waveform’s parameters should change in response to a desired state and measured behavior.