Glossary term

Noise Floor

Engineering definition of noise floor covering thermal noise, bandwidth, dBm/Hz, receiver sensitivity, detector limits and validation evidence.

Branch: Engineering Physics
Glossary type: metric
Content: Glossary term
Updated: Jun 26, 2026
Revision: v1.0.0 · reviewed

Definition

metric

Noise floor is the background noise level of a measurement or receiver system over a stated bandwidth, reference point and detection method.

Noise floor sets a lower boundary for weak-signal detection, measurement resolution, dynamic range and receiver sensitivity. It can come from thermal noise, amplifier noise, detector dark current, shot noise, quantization noise, electromagnetic pickup, vibration, background light, phase noise or interference. A useful noise-floor statement must include bandwidth, reference impedance or scaling, detector type, averaging, calibration state and operating condition.

Noise floor is the level of background noise that exists before the desired signal is evaluated. It is the practical lower boundary for weak-signal detection, measurement resolution, dynamic range, receiver sensitivity and many validation decisions. A signal below the noise floor may still exist physically, but the measurement chain cannot distinguish it without more information, narrower bandwidth, longer integration, synchronous detection, better shielding or a different sensor.

A noise-floor value is incomplete unless it states the measurement boundary. The same system can have different noise floors at the antenna connector, amplifier input, ADC input, digital spectrum, image detector, calibrated engineering-unit output or final dashboard value.

Thermal Noise

For an ideal resistor or matched RF input near equilibrium, thermal noise power over bandwidth B is:

P_n=k_BTB

where:

k_B is Boltzmann’s constant;
T is absolute temperature;
B is noise bandwidth.

At approximately 290 K, the thermal noise density is commonly written as:

N_0\approx -174\ \text{dBm/Hz}

For an RF receiver with noise figure NF in dB:

N_{dBm}\approx -174+10\log_{10}(B)+NF

where B is in hertz. This is a receiver-screening formula, not a universal measurement rule.

Voltage And Engineering-Unit Noise

For a voltage noise density e_n over an equivalent noise bandwidth B_n, the RMS noise is:

v_{n,rms}=e_n\sqrt{B_n}

If the calibrated sensitivity is K volts per engineering unit, the equivalent noise floor in the measured quantity is:

\displaystyle x_n=\frac{v_{n,rms}}{K}

This translation is important. A noise floor expressed only in volts may be useless to a mechanical, biomedical or process decision unless it is converted to newtons, pascals, degrees Celsius, optical power, acceleration or another release quantity.

Minimum Useful Signal

A simple detectability screen is:

x_{min}=k_d x_n

where k_d is the chosen detection multiplier. In a communication receiver, a minimum required received power may be stated as:

P_{min}=N_{dBm}+SNR_{req}+M

where SNR_{req} is required signal-to-noise ratio and M is design or implementation margin.

The threshold is application dependent. A slow lock-in measurement, an RF packet receiver, an ECG alarm, a vibration spectral line and an optical power monitor do not use the same detection rule.

Bandwidth Effect

Noise floor depends strongly on bandwidth. If the noise is approximately white over the band, increasing bandwidth raises integrated noise:

\displaystyle \Delta N=10\log_{10}\left(\frac{B_2}{B_1}\right)

Doubling bandwidth raises the noise floor by about 3 dB. Increasing bandwidth by a factor of five raises it by about 7 dB. Narrowing bandwidth can improve weak-signal detection, but it can also remove real signal content, slow response or hide transient events.

Worked RF Example

An RF receiver has bandwidth:

B=200000\ \text{Hz}

and system noise figure:

NF=5\ \text{dB}

The receiver noise floor is:

N_{dBm}=-174+10\log_{10}(200000)+5=-116.0\ \text{dBm}

If the waveform requires 12 dB SNR and the design margin is 4 dB, the minimum useful received signal is:

P_{min}=-116.0+12+4=-100.0\ \text{dBm}

If bandwidth is increased to 1 MHz with the same noise figure, the bandwidth ratio is five:

\Delta N=10\log_{10}(5)=7.0\ \text{dB}

The noise floor rises to about -109.0 dBm, so the same required SNR and margin require about -93.0 dBm. The receiver did not become worse as hardware; the wider measurement admitted more noise.

Difference From SNR And Noise Figure

Noise floor is a level. Signal-to-noise ratio is a ratio between a desired signal and a noise level. Noise figure is a degradation metric that describes how much a receiver chain worsens SNR compared with an ideal reference. Confusing these terms creates weak reviews: a low noise figure does not prove a low absolute noise floor if bandwidth is large, and a measured noise floor does not state whether a particular signal has enough SNR.

Validation Evidence

A defensible noise-floor statement includes bandwidth, filter shape, equivalent noise bandwidth, detector type, impedance, reference plane, gain setting, averaging, windowing, sample rate, calibration state, input termination, environmental condition, shielding, grounding, source-off or dark measurement, raw data record and any known interference. For spectrum measurements, state resolution bandwidth, video bandwidth, averaging mode and detector mode.

Common Mistakes

Common mistakes include quoting a noise floor without bandwidth, mixing dBm and dBm/Hz, measuring with the input floating, using a spectrum analyzer setting that hides peaks, narrowing bandwidth after validation without updating response time, treating interference as random noise, averaging away intermittent bursts, and comparing receiver sensitivity numbers with different bandwidths or target error rates.

The practical rule is to state where the noise floor is measured, over what bandwidth, with what detector and how that level affects the engineering decision.

REF

Disciplines