Johnson–Nyquist noise

These three circuits are all equivalent: (A) A resistor at nonzero temperature, which has Johnson noise; (B) A noiseless resistor in series with a noise-creating voltage source (i.e. the Thévenin equivalent circuit); (C) A noiseless resistance in parallel with a noise-creating current source with $I_{\text{rms}}{=}V_{\text{RMS}}/R$ (i.e. the Norton equivalent circuit).

Johnson–Nyquist noise (thermal noise, Johnson noise, or Nyquist noise) is the electronic noise generated by the thermal agitation of the charge carriers (usually the electrons) inside an electrical conductor at equilibrium, which happens regardless of any applied voltage. Thermal noise is present in all electrical circuits, and in sensitive electronic equipment (such as radio receivers) can drown out weak signals, and can be the limiting factor on sensitivity of electrical measuring instruments. Thermal noise is proportional to absolute temperature, so some sensitive electronic equipment such as radio telescope receivers are cooled to cryogenic temperatures to improve their signal-to-noise ratio. The generic, statistical physical derivation of this noise is called the fluctuation-dissipation theorem, where generalized impedance or generalized susceptibility is used to characterize the medium.

Thermal noise in an ideal resistor is approximately white, meaning that the power spectral density is nearly constant throughout the frequency spectrum, but does decay to zero at extremely high frequencies (terahertz for room temperature). When limited to a finite bandwidth, thermal noise has a nearly Gaussian amplitude distribution.

Thermal noise is distinct from shot noise, which consists of additional current fluctuations that occur when a voltage is applied and a macroscopic current starts to flow.

History of thermal noise

In 1905, in one of Albert Einstein's Annus mirabilis papers the theory of Brownian motion was first solved in terms of thermal fluctuations. The following year, in a second paper about Brownian motion, Einstein suggested that the same phenomena could be applied to derive thermally-agitated currents, but did not carry out the calculation as he considered it to be untestable.

Geertruida de Haas-Lorentz, daughter of Hendrik Lorentz, in her doctoral thesis of 1912, expanded on Einstein stochastic theory and first applied it to the study of electrons. Deriving a formula for the mean-squared value of the thermal current.

Walter H. Schottky studied the problem in 1918, while studying thermal noise using Einstein's theories, experimentally discovered another kind of noise, the shot noise.

Frits Zernike working in electrical metrology, found unusual aleatory deflections while working with high-sensitive galvanometers. He rejected the idea that the noise was mechanical, and concluded that it was of thermal nature. In 1927, he introduced the idea of autocorrelations to electrical measurements and calculated the time detection limit. His work coincided with de Haas-Lorentz' prediction.

The same year, working independently without any knowledge of Zernike's work, John B. Johnson working in Bell Labs found the same kind of noise in communication systems, but described it in terms of frequencies. He described his findings to Harry Nyquist, also at Bell Labs, who was able to explain the results, published in 1928.

Derivation of noise power

Nyquist's 1928 thought experiment deduced the power of the thermal noise of resistors by imagining two equal resistors connected via a lossless transmission line. Each resistor produces a voltage noise signal (represented by voltage sources) which travels across the line. All impedances are identical, so signals aren't reflected. Nyquist then imagined shorting both ends of the line, thereby trapping in-flight energy to remain on the line. All signals will now be completely reflected due to the now-mismatched impedance and form standing waves. For a particular bandwidth (from starting frequency $f_{0}$ to $f_{0}{+}\Delta f$), the number of modes of oscillation are $2\cdot {\text{length}}\cdot \Delta f/f_{0}$, provided that the line is sufficiently long to make this expression a great number (thus approximating a continuous spectrum). Because each mode provides one degree of freedom and $k_{\rm {B}}T$ of energy, the total energy in that bandwidth is thus $k_{\rm {B}}T$ times that number of modes. But since before the shorting there were originally no reflections, that total energy is the combined energy that was transferred from both resistors to the line during the transit time interval of ${\text{length}}/f_{0}$. Dividing the average energy transferred from each resistor to the line by the transit time interval results in $2\cdot {\text{length}}/f_{0}$ being cancelled out, so the resulting average power from each resistor is $k_{\rm {B}}T\,\Delta f$ for that bandwidth.

Nyquist's 1928 paper "Thermal Agitation of Electric Charge in Conductors" used principles of thermodynamics and statistical mechanics to provide a theoretical deduction explaining why Johnson's prior measurements found resistors had a white noise spectrum with a mean square voltage fluctuation of $4k_{\text{B}}TR$. Nyquist provided a thought experiment (key details in figure) that summed the energy of each standing wave mode of oscillation of an imaginary transmission line between two equal resistors ($R_{1}$ and $R_{2}$ which produce noise voltage sources $V_{1}$ and $V_{2}$) using concepts about potential energy and harmonic oscillators from the equipartition law of Boltzmann and Maxwell. Nyquist determined that each mode contributes $k_{\rm {B}}T$ of energy to the total energy, where $k_{\rm {B}}$ is the Boltzmann constant (1.380649×10⁻²³ joules per kelvin) and $T$ is the kelvin temperature. Nyquist concluded that the average power (i.e. energy per time) transferred from each resistor to the line (and consumed by the other resistor) for a bandwidth $\Delta f$ is proportional to temperature and bandwidth:

$P_{\text{R1-to-R2-over-Δ𝑓}}=P_{\text{R2-to-R1-over-Δ𝑓}}=k_{\rm {B}}T\,\Delta f\,.$

To determine the current due to the thermal noise of $R_{1}$ only, imagine that $V_{2}$ is a short. This current according to Ohm's law equals $V_{1}$ divided by the total resistance $2R$:

$$I_{1}={\tfrac {V_{1}}{2R}}\,.$$

Thus the thermal noise power transferred from $R_{1}$ to $R_{2}$ is obtained by multiplying the square of this current by $R$:

$$P_{\text{R1-to-R2}}=I_{1}^{2}R=({\tfrac {V_{1}}{2R}})^{2}R={\tfrac {V_{1}^{2}}{4R}}\,.$$

Noise voltage

Setting ${\textstyle P_{\text{R1-to-R2}}}$ to the earlier ${\textstyle P_{\text{R1-to-R2-over-Δ𝑓}}}$ equation allows solving for $V_{1}^{2}$ over that bandwidth:

$$V_{1}^{2}=4k_{\text{B}}TR\,\Delta f\,.$$

While this above derivation assumed $R_{1}$ and $R_{2}$ were equal impedances, Nyquist added similar reasoning in his paper to show that this result holds when any impedance (including a complex impedance) is used in place of $R_{2}$.

Thus, a noisy resistor represented as a voltage noise source $v_{n}$ in series with a noise-free resistor has a one-sided power spectral density, or voltage variance (mean square) per hertz of bandwidth for a resistance of value $R$ (in ohms) of:

${\overline {v_{n}^{2}}}=4k_{\text{B}}TR\,.$

For the general case, this definition applies to charge carriers in any type of conducting medium (e.g. ions in an electrolyte), not just resistors. The square root of this equation at room temperature (around 300 K) approximates to 0.13 ${\sqrt {R}}$ in units of nanovolts/√hertz. A 10 kΩ resistor, for example, would have approximately 13 nanovolts/√hertz at room temperature.

For a given bandwidth $\Delta f$, the root mean square (RMS) voltage $v_{n}$ is:

$v_{n}={\sqrt {\overline {v_{n}^{2}}}}{\sqrt {\Delta f}}={\sqrt {4k_{\text{B}}TR\,\Delta f}}\,.$

A useful reference point to remember is that 50 Ω over a 1 Hz bandwidth corresponds to 1 nV_RMS of noise at room temperature. Values for $R\,\Delta f$ that are x times bigger than 50 Ω·Hz then make approximately √x nV_RMS of noise at room temperature. For example, 50 kΩ over 1 kHz provides approximately 1000 nV_RMS at room temperature.

Noise current

The Norton equivalent circuit for resistor noise is a voltage noise source $i_{n}$ in parallel with a noise-free resistor. Diving $v_{n}$ by $R$ gives the root mean square value of the current source to be:

$$i_{n}={v_{n} \over R}={\sqrt {{4k_{\text{B}}T\Delta f} \over R}}.$$

Maximum transfer of noise power

The noise generated at a resistor $R_{\text{source}}$ can transfer to the remaining circuit. The maximum noise power transfer happens when the Thévenin equivalent resistance $R_{\rm {th}}$ of the remaining circuit matches $R_{\text{source}}$. In this case, each of the two resistors dissipates noise in both itself and in the other resistor. Since only half of the source voltage drops across any one of these resistors, this noise power transfer is:

$P_{\text{max}}=k_{\text{B}}\,T\Delta f\,.$

This noise power available from a resistor is independent of the resistance.

Noise power in decibel-milliwatts

Signal power is often measured in dBm (decibels relative to 1 milliwatt). Noise power would thus be $10\ \log _{10}({\tfrac {k_{\text{B}}T\Delta f}{\text{1 mW}}})$ in dBm. At room temperature (300 K), noise power for a bandwidth in hertz can be easily approximated as $10\ \log _{10}(\Delta f)$ - 173.8 in dBm. Some example noise powers in dBm are tabulated below:

Bandwidth $(\Delta f)$	Thermal noise power at 300 K (dBm)	Notes
1 Hz	−174
10 Hz	−164
100 Hz	−154
1 kHz	−144
10 kHz	−134	FM channel of 2-way radio
100 kHz	−124
180 kHz	−121.45	One LTE resource block
200 kHz	−121	GSM channel
1 MHz	−114	Bluetooth channel
2 MHz	−111	Commercial GPS channel
3.84 MHz	−108	UMTS channel
6 MHz	−106	Analog television channel
20 MHz	−101	WLAN 802.11 channel
40 MHz	−98	WLAN 802.11n 40 MHz channel
80 MHz	−95	WLAN 802.11ac 80 MHz channel
160 MHz	−92	WLAN 802.11ac 160 MHz channel
1 GHz	−84	UWB channel

Thermal noise on capacitors

Ideal capacitors, as lossless devices, do not have thermal noise, but as commonly used with resistors in an RC circuit, the combination has what is called kTC noise. The noise bandwidth of an RC circuit is Δf = 1/4RC. When this is substituted into the thermal noise equation, the result has an unusually simple form as the value of the resistance (R) drops out of the equation. This is because higher R decreases the bandwidth as much as it increases the noise.

The mean-square and RMS noise voltage generated in such a filter are:

${\overline {v_{n}^{2}}}={4k_{\text{B}}TR \over 4RC}={k_{\text{B}}T \over C}$

$v_{n}={\sqrt {4k_{\text{B}}TR \over 4RC}}={\sqrt {k_{\text{B}}T \over C}}.$

The noise charge is the capacitance times the voltage:

$Q_{n}=C\,v_{n}=C{\sqrt {k_{\text{B}}T \over C}}={\sqrt {k_{\text{B}}TC}}$

${\overline {Q_{n}^{2}}}=C^{2}\,{\overline {v_{n}^{2}}}=C^{2}{k_{\text{B}}T \over C}=k_{\text{B}}TC$

This charge noise is the origin of the term "kTC noise".

Although independent of the resistor's value, 100% of the kTC noise arises in the resistor. Therefore, if the resistor and the capacitor are at different temperatures, the temperature of the resistor alone should be used in the above calculation.

An extreme case is the zero bandwidth limit called the reset noise left on a capacitor by opening an ideal switch. The resistance is infinite, yet the formula still applies; however, now the RMS must be interpreted not as a time average, but as an average over many such reset events, since the voltage is constant when the bandwidth is zero. In this sense, the Johnson noise of an RC circuit can be seen to be inherent, an effect of the thermodynamic distribution of the number of electrons on the capacitor, even without the involvement of a resistor.

The noise is not caused by the capacitor itself, but by the thermodynamic fluctuations of the amount of charge on the capacitor. Once the capacitor is disconnected from a conducting circuit, the thermodynamic fluctuation is frozen at a random value with standard deviation as given above. The reset noise of capacitive sensors is often a limiting noise source, for example in image sensors.

Any system in thermal equilibrium has state variables with a mean energy of kT/2 per degree of freedom. Using the formula for energy on a capacitor (E = 1/2CV²), mean noise energy on a capacitor can be seen to also be 1/2CkT/C = kT/2. Thermal noise on a capacitor can be derived from this relationship, without consideration of resistance.

Noise of capacitors at 300 K

Capacitance	${\sqrt {k_{\text{B}}T/C}}$	${\sqrt {k_{\text{B}}TC}}$	Electrons
1 fF	2 mV	2 aC	12.5 e−
10 fF	640 μV	6.4 aC	40 e−
100 fF	200 μV	20 aC	125 e−
1 pF	64 μV	64 aC	400 e−
10 pF	20 μV	200 aC	1250 e−
100 pF	6.4 μV	640 aC	4000 e−
1 nF	2 μV	2 fC	12500 e−

Thermal noise on inductors

Inductors are the dual of capacitors. Just like kTC noise is independent of resistance, a resistor with an inductor $L$ also results in a noise current that is independent of resistance:

$${\overline {i_{n}^{2}}}={k_{\text{B}}T \over L}\,.$$

Generalized forms

The $4k_{\text{B}}TR$ voltage noise described above is a special case for a purely resistive component for low frequencies. In general, the thermal electrical noise continues to be related to resistive response in many more generalized electrical cases, as a consequence of the fluctuation-dissipation theorem. Below a variety of generalizations are noted. All of these generalizations share a common limitation, that they only apply in cases where the electrical component under consideration is purely passive and linear.

Reactive impedances

Nyquist's original paper also provided the generalized noise for components having partly reactive response, e.g., sources that contain capacitors or inductors. Such a component can be described by a frequency-dependent complex electrical impedance $Z(f)$. The formula for the power spectral density of the series noise voltage is

$S_{v_{n}v_{n}}(f)=4k_{\text{B}}T\eta (f)\operatorname {Re} [Z(f)].$

The function $\eta (f)$ is simply equal to 1 except at very high frequencies, or near absolute zero (see below).

The real part of impedance, $\operatorname {Re} [Z(f)]$, is in general frequency dependent and so the Johnson–Nyquist noise is not white noise. The rms noise voltage over a span of frequencies $f_{1}$ to $f_{2}$ can be found by integration of the power spectral density:

${\sqrt {\langle v_{n}^{2}\rangle }}={\sqrt {\int _{f_{1}}^{f_{2}}S_{v_{n}v_{n}}(f)df}}$.

Alternatively, a parallel noise current can be used to describe Johnson noise, its power spectral density being

$S_{i_{n}i_{n}}(f)=4k_{\text{B}}T\eta (f)\operatorname {Re} [Y(f)].$

where $Y(f)=1/Z(f)$ is the electrical admittance; note that $\operatorname {Re} [Y(f)]=\operatorname {Re} [Z(f)]/|Z(f)|^{2}$

Quantum effects at high frequencies or low temperatures

Nyquist also pointed out that quantum effects occur for very high frequencies or very low temperatures near absolute zero. The function $\eta (f)$ is in general given by

$\eta (f)={\frac {hf/k_{\text{B}}T}{e^{hf/k_{\text{B}}T}-1}}+{\frac {1}{2}}{\frac {hf}{k_{\text{B}}T}},$

where $h$ is the Planck constant and $\eta (f)$ is a multiplying factor.

At very high frequencies $f\gtrsim k_{\text{B}}T/h$, the function $\eta (f)$ starts to exponentially decrease to zero. At room temperature this transition occurs in the terahertz, far beyond the capabilities of conventional electronics, and so it is valid to set $\eta (f)=1$ for conventional electronics work.

Relation to Planck's law

Nyquist's formula is essentially the same as that derived by Planck in 1901 for electromagnetic radiation of a blackbody in one dimension—i.e., it is the one-dimensional version of Planck's law of blackbody radiation. In other words, a hot resistor will create electromagnetic waves on a transmission line just as a hot object will create electromagnetic waves in free space.

In 1946, Robert H. Dicke elaborated on the relationship, and further connected it to properties of antennas, particularly the fact that the average antenna aperture over all different directions cannot be larger than $\lambda ^{2}/(4\pi )$, where λ is wavelength. This comes from the different frequency dependence of 3D versus 1D Planck's law.

Multiport electrical networks

Richard Q. Twiss extended Nyquist's formulas to multi-port passive electrical networks, including non-reciprocal devices such as circulators and isolators. Thermal noise appears at every port, and can be described as random series voltage sources in series with each port. The random voltages at different ports may be correlated, and their amplitudes and correlations are fully described by a set of cross-spectral density functions relating the different noise voltages,

$S_{v_{m}v_{n}}(f)=2k_{\text{B}}T\eta (f)(Z_{mn}(f)+Z_{nm}(f)^{*})$

where the $Z_{mn}$ are the elements of the impedance matrix $\mathbf {Z}$. Again, an alternative description of the noise is instead in terms of parallel current sources applied at each port. Their cross-spectral density is given by

$S_{i_{m}i_{n}}(f)=2k_{\text{B}}T\eta (f)(Y_{mn}(f)+Y_{nm}(f)^{*})$

where $\mathbf {Y} =\mathbf {Z} ^{-1}$ is the admittance matrix.

See also

• Fluctuation-dissipation theorem
• Shot noise
• Pink noise
• Langevin equation
• Rise over thermal

References

 This article incorporates public domain material from Federal Standard 1037C. General Services Administration. Archived from the original on 2022-01-22. (in support of MIL-STD-188).

External links

• Amplifier noise in RF systems
• Thermal noise (undergraduate) with detailed math