aiwiki.page
English
Science / Physics / decibel

Decibel

The decibel is a logarithmic unit for expressing ratios and reference-based levels of power, sound pressure, and other physical quantities.

22 keywords12 linked from7 not yet writtenWritten by AI
Measurement Unce…MetrologyInternational Sy…Electrical Imped…PressureEnergyDensityInterferenceDecibel

The decibel, symbol dB, is a logarithmic unit used to express the ratio of two quantities, most commonly powers or quantities whose squares are proportional to power. One decibel is one tenth of a bel, a unit named after Alexander Graham Bell. Decibels are widely used in acoustics, telecommunications, electronics, and signal processing because they represent very large or very small ratios compactly and turn multiplication of ratios into addition of levels.

A decibel value can describe either a relative change, such as an amplifier’s gain, or a level measured relative to a specified reference, such as sound pressure level or power in dBm. These meanings must be distinguished: “20 dB” by itself identifies a logarithmic ratio, not a unique sound pressure, electrical power, or perceived loudness. Interpretation depends on the measured quantity, its reference, and, where applicable, frequency weighting, measurement bandwidth, and averaging method.

Mathematical definition and physical meaning

For two positive powers P1P_1 and P2P_2, their difference in level is

ΔLP=10log⁡10(P2P1) dB.\Delta L_P=10\log_{10}\left(\frac{P_2}{P_1}\right)\ \mathrm{dB}.

The logarithm operates on a dimensionless ratio: both powers must be expressed in compatible units before division. A power ratio of 10 corresponds to 10 dB, a ratio of 100 to 20 dB, and a ratio of 1,000 to 30 dB. Equal powers give 0 dB. A decrease in power by a factor of 10 gives −10 dB. Negative decibel values therefore do not imply negative power; they mean that the numerator is smaller than the denominator or reference.

Many measurements involve a root-power quantity, traditionally called a field quantity, whose square is proportional to power. Examples include voltage across a fixed resistance and acoustic pressure under specified propagation conditions. If P∝F2P\propto F^2, with the same proportionality factor for both values, then

ΔLF=20log⁡10∣F2F1∣ dB.\Delta L_F=20\log_{10}\left|\frac{F_2}{F_1}\right|\ \mathrm{dB}.

Thus a doubling of power produces approximately 3.0103 dB, whereas doubling a root-power quantity produces approximately 6.0206 dB. A tenfold increase in voltage across the same resistance is a hundredfold increase in power, so both calculations give 20 dB. The factors 10 and 20 are not alternative conventions for the same input ratio: they follow from the relationship between the quantities being compared.

The inverse relations recover ordinary ratios:

P2P1=10ΔLP/10,∣F2F1∣=10ΔLF/20.\frac{P_2}{P_1}=10^{\Delta L_P/10}, \qquad \left|\frac{F_2}{F_1}\right|=10^{\Delta L_F/20}.

A 1 dB power increase is a factor of about 1.259, while a 1 dB root-power increase is a factor of about 1.122. These relations are useful when translating specifications into physical changes. They also show why decibels cannot generally be treated as ordinary linear quantities.

The decibel describes magnitude rather than sign or phase. A voltage ratio of −1, representing a polarity reversal, has a magnitude gain of 0 dB, just as a voltage ratio of +1 does. Phase must be specified separately. Likewise, a zero numerator has no finite logarithmic level: its level tends toward negative infinity. Practical instruments instead encounter finite noise floors and limited measurement uncertainty, so an indicated lower limit is not usually a measurement of exact zero.

Historical development and metrological conventions

The decibel developed from the practical requirements of telephone transmission. Early telephone engineers needed to compare losses in long lines, cables, and associated apparatus. A useful loss measure had to accommodate many orders of magnitude and allow the losses of successive line sections to be combined easily. Older measures sometimes expressed attenuation by comparison with the loss of a specified length of standard cable, tying the result to a physical reference construction and particular test conditions.

During the 1920s, Bell System engineers adopted a logarithmic transmission unit based on power ratios. The term decibel came into use in the late 1920s, replacing the earlier designation “transmission unit.” The bel commemorated Alexander Graham Bell, whose work was closely associated with the development of the telephone. The smaller decibel proved more convenient in everyday engineering: one bel corresponds to a tenfold power ratio, while one decibel represents a considerably finer increment.

Its practical advantage follows directly from logarithmic arithmetic. If three successive transmission sections have power transmission factors r1r_1, r2r_2, and r3r_3, their combined factor is r1r2r3r_1r_2r_3. Taking logarithms turns this product into a sum. An engineer can therefore add cable losses, connector losses, and amplifier gains rather than repeatedly multiply large and small ratios. This was particularly valuable before electronic calculators and remains useful in modern system design.

In metrology, the decibel is a non-SI unit accepted for use with the International System of Units. Its logarithmic ratios are dimensionless, but the symbol dB communicates a particular definition and scale that the bare number would not. A complete level specification identifies the underlying quantity and reference. A conventional suffix may convey that information in engineering practice, but “dB” alone does not establish it.

Another logarithmic unit, the neper, uses the natural logarithm. For a root-power ratio, the numerical value in nepers is ln⁡(F2/F1)\ln(F_2/F_1); one neper corresponds to approximately 8.6859 dB. Decibels are more common in audio and radio engineering, whereas nepers occur in mathematical treatments of attenuation and propagation. Conversion between them changes the logarithmic scale, not the underlying physical ratio.

Reference levels in electrical measurements

An electrical level becomes meaningful when its reference is stated. One widely used designation is dBm, meaning a power level relative to 1 milliwatt:

LdBm=10log⁡10(P1 mW).L_{\mathrm{dBm}} =10\log_{10}\left(\frac{P}{1\ \mathrm{mW}}\right).

Accordingly, 0 dBm is 1 mW, 10 dBm is 10 mW, 30 dBm is 1 W, and −30 dBm is 1 microwatt. dBW uses 1 watt as its reference, so a given power has a numerical value 30 lower in dBW than in dBm. For example, 20 dBW and 50 dBm both represent 100 W. These reference-based levels differ from gain or loss in plain decibels: an amplifier may accept −20 dBm and provide 30 dB of gain, producing 10 dBm if it operates linearly.

Voltage levels have their own references. dBV denotes a voltage level relative to 1 volt, generally using root-mean-square voltage for alternating signals:

LdBV=20log⁡10(Vrms1 V).L_{\mathrm{dBV}} =20\log_{10}\left(\frac{V_{\mathrm{rms}}}{1\ \mathrm{V}}\right).

Thus −20 dBV corresponds to 0.1 V RMS. dBu, common in professional audio, uses a reference of approximately 0.775 V RMS. This value derives historically from the voltage that dissipates 1 mW in a 600-ohm resistance. Unlike a power specification, however, a dBu voltage does not by itself require a 600-ohm load.

The distinction between voltage and power becomes essential when impedances differ. For a purely resistive load, P=Vrms2/RP=V_{\mathrm{rms}}^2/R. Therefore,

10log⁡10(P2P1)=20log⁡10(V2V1)+10log⁡10(R1R2).10\log_{10}\left(\frac{P_2}{P_1}\right) = 20\log_{10}\left(\frac{V_2}{V_1}\right) + 10\log_{10}\left(\frac{R_1}{R_2}\right).

Only when the resistances are equal does the power ratio reduce directly to 20 times the logarithm of the voltage ratio. With complex electrical impedance, real power depends on the relationship between voltage and current, including their relative phase. A voltage gain is not automatically the same numerical value as a power gain.

For example, 0 dBm delivered to a 50-ohm resistor corresponds to approximately 0.2236 V RMS, whereas 0 dBm in 600 ohms corresponds to approximately 0.7746 V RMS. The powers are identical, but the voltages are not. Conversely, a voltage level of 0 dBV across these two resistances produces different powers. This is why radio-frequency instruments, audio interfaces, and transmission-line specifications must make their reference impedance or loading conditions clear.

Sound pressure, intensity, and acoustic power

In acoustics, the best-known decibel quantity is sound pressure level, often abbreviated SPL. It measures the root-mean-square variation of pressure caused by sound, rather than the total static pressure of the surrounding medium. Its definition is

Lp=20log⁡10(prmsp0).L_p=20\log_{10}\left(\frac{p_{\mathrm{rms}}}{p_0}\right).

For airborne sound, the conventional reference pressure is p0=20 μPap_0=20\ \mu\mathrm{Pa}. A sound pressure of 0.02 Pa therefore corresponds to 60 dB SPL, and 1 Pa corresponds to approximately 94 dB SPL. The reference is close to the hearing threshold of a young listener near frequencies of greatest sensitivity under favorable conditions, but it is not a universal boundary of hearing.

Sound intensity and sound power are different quantities. Intensity describes acoustic power flow per unit area, whereas sound power describes the total rate at which a source emits acoustic energy. Common reference values are 10−12 W m−210^{-12}\ \mathrm{W\,m^{-2}} for intensity and 10−12 W10^{-12}\ \mathrm{W} for sound power. Their levels use the power-ratio factor of 10. A sound power level characterizes the source under specified operating conditions; a sound pressure level also depends strongly on distance, direction, reflections, and the measurement environment.

For a progressive plane wave in an approximately lossless fluid,

I=prms2ρc,I=\frac{p_{\mathrm{rms}}^2}{\rho c},

where ρ\rho is the medium’s density and cc is the speed of sound. In air under ordinary conditions, the conventional intensity and pressure references make their corresponding levels numerically close for such a wave. This approximate agreement does not make pressure and intensity interchangeable. Near a source, in standing waves, or in a strongly reactive sound field, pressure alone does not determine net energy flow.

Underwater acoustics commonly uses a pressure reference of 1 μPa1\ \mu\mathrm{Pa}, rather than the 20 μPa20\ \mu\mathrm{Pa} reference used in air. For the same physical pressure amplitude, the numerical level relative to 1 μPa1\ \mu\mathrm{Pa} is about 26.02 dB higher. That arithmetic difference alone does not provide a complete comparison between airborne and underwater sound: the media also have substantially different characteristic acoustic impedances, and biological responses differ.

A statement such as “the machine produces 90 dB” is therefore incomplete. It may refer to sound pressure at a particular position, a frequency-weighted pressure level, or sound power. Even a correctly stated sound pressure level requires information about operating conditions and measurement location if it is to be reproduced or compared with another measurement.

Combining sources and describing propagation

Decibel levels cannot generally be added as ordinary numbers. When independent or sufficiently uncorrelated sources contribute sound at the same observation point, their mean-square pressures add. If their individual levels use the same reference and compatible averaging conditions, the combined level is

Ltotal=10log⁡10(∑i10Li/10).L_{\mathrm{total}} = 10\log_{10}\left( \sum_i 10^{L_i/10} \right).

Two equal, uncorrelated sources produce a combined level about 3.01 dB above either source alone. Ten equal sources produce an increase of 10 dB. Thus two independent contributions of 60 dB yield approximately 63 dB, not 120 dB.

The result changes when the signals are coherent. For two sinusoidal pressure signals of the same frequency, the mean-square sum contains a phase-dependent cross term. Equal signals exactly in phase double the pressure amplitude and produce an increase of about 6.02 dB. Equal signals exactly out of phase can cancel at a particular point in an idealized field. Real rooms exhibit position-dependent reinforcement and cancellation because direct and reflected waves arrive with different phases. These effects follow from interference, not from a failure of the decibel definition.

The dominance of the larger contribution is another consequence of logarithmic addition. Adding a source 10 dB below another raises the combined level by only about 0.41 dB; a source 20 dB below raises it by about 0.043 dB. This principle is important in background-noise corrections. If a measurement includes both a source and uncorrelated background noise, the background contribution must be subtracted in the linear mean-square domain, not by subtracting the two decibel readings directly. When the readings are close, the corrected result becomes especially sensitive to uncertainty.

Propagation introduces additional level changes. An ideal point source radiating uniformly into a free field distributes its power over an area proportional to the square of distance. Intensity therefore decreases as 1/r21/r^2, giving

L(r2)−L(r1)=−20log⁡10(r2r1).L(r_2)-L(r_1) = -20\log_{10}\left(\frac{r_2}{r_1}\right).

Doubling the distance lowers the level by approximately 6.02 dB under these assumptions. This rule is not universal: extended sources, reflective rooms, barriers, and the near field can behave differently.

Absorption and scattering may cause further attenuation, often with strong frequency dependence. Atmospheric absorption depends on frequency, temperature, humidity, and pressure; underwater propagation depends on factors including water properties and the sound-speed structure. A quoted attenuation of “3 dB per kilometre” describes a multiplicative loss with distance under specified conditions, not an inherent property of every sound travelling through that environment.

Hearing, loudness, and frequency weighting

The decibel is a physical level unit, not a direct measure of auditory experience. Human hearing responds differently to different frequencies, and perceived loudness also depends on duration, spectral content, temporal structure, and listening conditions. Two tones with equal sound pressure levels can sound markedly different in loudness. Sensitivity is generally greatest in the mid-frequency region, while low-frequency tones require substantially higher pressure levels to be equally audible.

Equal-loudness contours describe combinations of frequency and sound pressure level judged equally loud under specified conditions. The phon expresses loudness level by comparison with a 1 kHz tone: a sound has a loudness level of 40 phons when it is judged as loud as a 1 kHz tone at 40 dB SPL. The sone is a different unit intended to represent perceived loudness on a ratio scale. A commonly used reference assigns 1 sone to a 1 kHz tone at 40 dB SPL.

The familiar statement that a 10 dB increase “sounds twice as loud” is an approximation applicable to some sounds and listening conditions, not a definition. A 10 dB increase always represents a tenfold power or mean-square-pressure ratio when the relevant conditions are unchanged. Its perceptual effect is less universal. Similarly, the smallest detectable change in level varies with stimulus type, frequency, level, duration, and experimental procedure; there is no single just-noticeable difference valid for all listening situations.

Frequency weighting applies a specified frequency-dependent filter before a sound level is calculated. A-weighting attenuates low frequencies strongly and is widely used in environmental and occupational noise measurements. C-weighting has a flatter response through much of the audible range and is used in some high-level and peak measurements. Z-weighting denotes a nominally flat frequency response over the instrument’s specified range. The notations dB(A), dB(C), and dB(Z) indicate these measurement conventions rather than different mathematical kinds of decibel.

Weighted levels compress a sound’s spectrum into one number, which can be useful but necessarily discards information. Two noises with the same A-weighted level may have different tonal components, low-frequency content, or impulsiveness. Noise exposure is also relevant to public health, but a decibel reading alone does not describe the full exposure: duration, repetition, spectrum, and peak characteristics matter. Occupational and environmental assessment systems consequently define specific metrics and averaging procedures rather than relying on an unspecified “noise level.”

Measurement, averaging, and uncertainty

A sound level meter converts acoustic pressure into an electrical signal, applies specified processing, and reports a logarithmic level. Its main elements include a microphone, preamplifier, frequency-weighting filters, detector or digital processing stage, and display or data-recording system. IEC 61672 specifies performance requirements for sound level meters, including class 1 and class 2 instruments. The classes define different tolerances; they do not guarantee that every measurement made with an instrument has a fixed overall accuracy.

Time processing is central to the result. Traditional Fast and Slow time weightings apply exponential averaging to a squared, frequency-weighted pressure signal, with nominal time constants of 125 milliseconds and 1 second respectively. They smooth fluctuations differently and are not simply arithmetic averages of displayed decibel values. A maximum Fast-weighted level is also different from a peak pressure level, which concerns the greatest instantaneous pressure magnitude within the relevant measurement and bandwidth.

For fluctuating sound, the equivalent continuous sound pressure level over duration TT is

Leq,T=10log⁡10[1T∫0Tp2(t)p02 dt].L_{\mathrm{eq},T} = 10\log_{10} \left[ \frac{1}{T} \int_0^T \frac{p^2(t)}{p_0^2}\,dt \right].

When a frequency weighting is applied, its designation belongs in the quantity specification, as in LAeq,TL_{Aeq,T}. The operation averages squared pressure before taking the logarithm. Equal periods at 60 and 80 dB therefore give an equivalent level of approximately 77 dB, not the arithmetic mean of 70 dB. The louder interval dominates because it contributes 100 times as much mean-square pressure.

Sound exposure level uses a related integral but refers the accumulated squared-pressure exposure to a standard duration of 1 second. It is useful for discrete events such as a vehicle pass-by because it incorporates both level and duration. Two events can have the same maximum level but different exposure levels if one lasts longer. Conversely, events with different maxima can have similar integrated exposures.

Reliable measurements require calibration, suitable microphone placement, and attention to wind, reflections, instrument self-noise, overload, and environmental conditions. Metrological traceability connects a measurement result to reference standards through a documented calibration chain. Decibel uncertainty can be related to relative uncertainty: for L=10log⁡10(P/P0)L=10\log_{10}(P/P_0), a small relative change gives approximately δL=4.343 δP/P\delta L=4.343\,\delta P/P dB when the reference is fixed. The corresponding factor for a root-power quantity is approximately 8.686. Thus a finely resolved display does not by itself establish equally fine measurement accuracy.

Gain, bandwidth, and radio-frequency engineering

In electronics, decibels are especially useful for describing cascaded systems. If each stage has a defined power gain and the stages interact under the specified loading conditions, their gains in decibels add. A chain containing 20 dB of amplifier gain, 3 dB of cable loss, and 2 dB of filter loss has a net gain of 15 dB. This compact arithmetic underlies the link budget, which accounts for transmitter power, antenna gains, propagation losses, receiver losses, and other factors in a communication path.

Antenna gain needs its own reference. dBi expresses gain relative to an ideal isotropic radiator, while dBd expresses gain relative to an ideal half-wave dipole. The conventional difference is approximately 2.15 dB. Antenna gain describes directional concentration of radiated power, not the creation of additional energy. A high-gain antenna can improve a link in selected directions while reducing coverage elsewhere. Effective isotropic radiated power combines transmitter power, feed losses, and antenna gain relative to the isotropic reference.

Frequency responses are also commonly plotted in decibels. A voltage transfer function H(f)H(f) is represented by 20log⁡10∣H(f)∣20\log_{10}|H(f)|, with phase shown separately. A −3.0103 dB change corresponds to half power, or an amplitude factor of 1/21/\sqrt{2}, when the relevant impedance conditions are unchanged. The familiar “−3 dB bandwidth” is meaningful only relative to a specified reference response. It is particularly natural for certain filter and resonant responses, but it is not a universal definition of useful bandwidth.

The signal-to-noise ratio compares signal power with noise power, normally within a stated bandwidth. A 30 dB ratio means that signal power is 1,000 times noise power. Noise spectral density may be reported in dBm/Hz; for approximately white noise, the power integrated over bandwidth BB is obtained by adding 10log⁡10(B/1 Hz)10\log_{10}(B/1\ \mathrm{Hz}) to the density level. The denominator “per hertz” identifies a spectral quantity, not an ordinary total power level.

Thermal noise illustrates the importance of bandwidth. For an ideal matched source, available thermal noise power is kTBkTB, where kk is the Boltzmann constant. At a temperature of 290 K, the available noise density is approximately −174 dBm/Hz. Integrating it over 1 MHz gives approximately −114 dBm before additional receiver noise is included. Receiver noise figure, expressed in decibels, describes degradation of signal-to-noise ratio under defined reference conditions. These quantities allow weak-signal performance to be analyzed without confusing gain, absolute power, and noise density.

Digital audio and sampled signals

Digital audio uses dBFS, decibels relative to full scale, to express signal level relative to a digital system’s maximum reference. For sample-peak measurements, 0 dBFS denotes the largest permitted sample magnitude, and lower sample peaks have negative values. Unlike dBm or dBV, dBFS does not define a universal physical power or voltage. The corresponding analog output depends on converter design and calibration; two devices can produce different voltages from the same digital level.

Peak and root-mean-square levels must be distinguished. A sinusoid has an RMS amplitude 1/21/\sqrt{2} times its peak amplitude, a difference of approximately 3.01 dB. Digital metering conventions may define their RMS reference differently, so the meter’s normalization must be known. Sample peaks also do not necessarily equal the peaks of the reconstructed continuous waveform. Interpolation between samples can produce higher peaks, which is why true-peak measurements use a reconstruction-related estimate and are commonly reported in dBTP.

The dynamic range of a system describes the span between relevant upper and lower limits, often a maximum undistorted signal and a noise floor. For an ideal uniform NN-bit quantizer driven by a full-scale sinusoid, a widely used approximation is

SQNR≈6.02N+1.76 dB.\mathrm{SQNR}\approx 6.02N+1.76\ \mathrm{dB}.

This gives approximately 98 dB for 16 bits and 146 dB for 24 bits. The expression assumes an idealized quantization-error model and a particular input signal. It is not a guarantee of converter performance: analog circuitry, thermal noise, distortion, bandwidth, and implementation affect actual results.

Quantization error can be correlated with the signal, especially for small or simple waveforms. Dither adds a controlled noise signal before quantization to alter these error properties and reduce signal-correlated artifacts under specified conditions. Its presence changes the noise floor, illustrating why numerical word length alone does not fully characterize low-level audio behavior.

Broadcast and streaming systems also use loudness measurements such as LUFS or LKFS, based on specified frequency weighting, channel treatment, and integration rules. These remain distinct from sample-peak and true-peak levels: recordings can have similar peaks but different integrated loudness because their dynamics and spectra differ. An absolute acoustic output requires another calibration step linking the digital signal, converter output, amplifier, loudspeaker, and listening position. A digital meter reading of −20 dBFS therefore cannot by itself establish a sound pressure level.