Technote 4

May, 1999
Revised 05/08/08

Noise in Electronic Systems

Tim J. Sobering
SDE Consulting
sdeconsulting@pobox.com © 1999 Tim J. Sobering. All rights reserved.
Noise in Electronic Systems Page 2

Noise in Electronic Systems

Noise is a fundamental parameter to be considered in an electronic design as it typically
limits the overall performance of the system. There are a number of good sources of
information on electronics noise and this Technote will not attempt to cover all aspects in
detail. This is simply a starting point and can serve as a reference or summary. Some
types of noise specific to photodetectors are also covered.
Before introducing specific types of noise that are present in electronic systems, it is
necessary to address some common concepts. First, any electronic system will have
multiple sources of noise, both “internal” and “external”. For the purpose of this discussion,
“internal” (or intrinsic) references sources that are a part of the circuit being examined –
resistors, amplifiers, transistors, etc. “External” refers to noise present in the signal being
applied to the circuit or to noise introduced into the circuit by another means, such as
conducted on a system ground or received on one of the many antennas formed by the
traces and components in the system. Any non-signal related voltage or current in a
system has the potential to be defined as noise.
When multiple sources of noise are present, their contributions add in proportion to their
noise powers, not the noise voltages. Uncorrelated noise adds by the sum of the individual
noise powers. This is easy when working with power, but usually we work with voltages or
currents, so we square the values. For example, given uncorrelated voltage noise sources
e1, e2 ... en, the total noise, et, is given by:
2 2 2
eT = e1 + e2 + L + en (1)

Mixed or partially correlated noise is summed using:

2 2
eT = e1 + e2 + 2ce1 e2 (2)

where:
c = correlation coefficient -1≤ c ≤ 1
Many common noise sources are uncorrelated. Correlated noise is less common but can be
present in the form of external noise pick-up or capacitive or inductive coupling between
signals. Use caution when assuming that noise sources are uncorrelated.
Because of the way noise sources add, analyses can be simplified by understanding the
concept of “dominant” noise sources. If a system has two noise sources and one is 25% of
the magnitude (voltage) of the other, the lesser source will only contribute 3% to the total
noise in the system. When performing a noise analysis, smaller sources can quickly be
discarded without significantly affecting the overall performance estimate.
In semiconductor data sheets, noise is typically expressed as a Noise Spectral Density. This
term is used to describe the noise content in a unit of bandwidth. From communications,
noise power was expressed as a spectral plot and the “density” was related to the power in a
1 Hz bandwidth centered on the frequency of interest. If the “resistances” are the same, as
is usually the case in communication (commonly 50 Ω or 75 Ω), the resistance cancels in the
equations the voltage is the square root of the power. But this means that the bandwidth is
now √Hz, or “root hertz”. So you frequently see noise voltage expressed in nV/√Hz or

Copyright © 1999 Tim J. Sobering. All rights reserved.

Noise in Electronic Systems Page 3

pA/√Hz in a data sheet. The math will all work out; just square them before you add them
together1.
While we typically express noise in terms of voltages or currents, you can see that the
definitions all relate back to noise power. Hence the square root term to relate back to
power spectral density. It is common to see various noise specifications in an opamp data
sheet at different frequencies or for different bandwidths. More importantly, plots of
voltage or current noise density as a function of frequency are given. The “shape of these
plots relates to the type of noise. For example, a flat response (constant noise power
distribution at all frequencies) is termed “white” noise. Shot noise and Johnson noise are
examples of “white” noise. It is also common to see a Noise Spectral Density that is
inversely proportional to frequency. For most systems, this is only evident at low
frequencies as other noise sources eventually dominate as frequency increases. This is
termed 1/f noise, or flicker noise and is one of the least understood types of noise.
The final term to be defined is Equivalent Noise Bandwidth, noise bandwidth, or NBW.
This is addressed in detail in a previous Technote2. Equivalent Noise Bandwidth
corresponds to a “brickwall” filter of bandwidth Δf and a noise power equivalent to the
original transfer function. Stated more simply, the NBW is the frequency such that a
rectangle defined by H(max)2 and Δf has an area equal to the area under |H(ω)|2.
The relationship between Δf and f3dB, the 3dB frequency of the system, depends on the
number of poles in the transfer function. Typically, a single pole rolloff (or dominant pole
rolloff) is assumed and Δf is given by:
π
Δf = f3dB = 1.57 f3dB (3)
2
Using 1.57f3dB in all calculations does not introduce a large error for 2nd or 3rd order systems
since the noise is typically a function of √Δf. Use caution if the transfer function exhibits
gain peaking at high frequencies as this will result in underestimating system noise.
There are many types of internal noise sources in electronic systems. This Technote will
address Johnson noise (Thermal noise), Shot noise, 1/f noise and others.
Johnson Noise
Johnson Noise (also known as Thermal Noise or Nyquist Noise) results from the thermal
motion of charged particles in a resistive element.3 This is the most common noise source
in electronics and is present in all conductors. The only parameters in an electronic system
within the designer’s control that influence Johnson noise are the resistance, the
bandwidth, and the temperature. Temperature control is more difficult and less common
but is often used with photodetectors. The noise generated has no relation to the type of
conductor – equal value carbon composition and metal film resistors make the same noise
contribution. However, purely reactive elements do not produce Johnson noise. The
following equations are used for calculating Johnson noise.
Johnson noise power PJ (rms) = kTΔf (4)

1
Reference “Technote 5 – Opamp Noise Analysis”, Tim J. Sobering, 6/1999
2Reference “Technote 1 – Equivalent Noise Bandwidth”, Tim J. Sobering, 5/1991.
3See “Optical Radiation Detectors”, Eustace L. Dereniak and Devon G. Crowe, John Wiley & Sons, 1984, pp. 39-
40 for derivations. Much of this material is directly from that source.

Copyright © 1999 Tim J. Sobering. All rights reserved.

Noise in Electronic Systems Page 4

Johnson noise voltage VJ (rms ) = 4kTRΔf (5)

4kTΔf
Johnson noise current iJ (rms ) = (6)
R
Where:
k = Boltzmann’s constant = 1.38 x 10-23 J/K
T = Temperature in K
R = Resistance in Ω
Δf = Effective Noise Bandwidth in Hz
Johnson noise is “white noise” meaning that it is spectrally uniform or flat. Said another
way, the noise spectral density is constant with frequency. The amplitude of Johnson noise
has a zero mean Gaussian distribution and is difficult to measure directly due to internal
noise in voltmeter or oscilloscope, but it can be measured in the presence of gain i.e. in an
amplifier. As an example, consider a 50 Ω resistor at room temperature and a
measurement system with a 1 MHz equivalent noise bandwidth.
4kT = 1.61 × 10 −20 J at room temperature (290 K) (7)

v j (rms ) = (1.61 × 10 −20 J )(50Ω)(1MHz ) = 897nV (8)

Expressed in terms of a noise spectral density (removing the effect of the bandwidth of the
measuring device)

v j (rms) = 0.897 nV (9)

Hz
When performing measurements with an oscilloscope, it is most common to estimate the
peak-to-peak noise and convert this to an RMS measurement. Table 1 shows various
rations that can be used in making this estimation.
Nominal % of the Time Gaussian Noise with Exceed
Peak-to-Peak the Nominal Peak-to-Peak Value
2 × RMS 32%
3 × RMS 13%
4 × RMS 4.6%
5 × RMS 1.2%
6 × RMS 0.27%
6.6 × RMS4 0.10%
7 × RMS 0.046%
8 × RMS 0.006%
Table 1. Ratios for computing RMS from Peak-to-Peak Values5

Johnson noise is typically modeled as a noiseless resistor either in series with a noise
voltage source or in parallel with a noise current source, as shown below. Note that stray
shunt capacitance limits noise voltage since as R increases, Δf decreases.

4 The most common conversion factor is 6.6

5 “Opamp Applications”, Walter G. Jung, Editor, pp. 1.83, Analog Devices, 2002.

Copyright © 1999 Tim J. Sobering. All rights reserved.

Noise in Electronic Systems Page 5

Figure 1. Johnson noise models

Shot Noise
Shot noise results from the flow of current across a potential barrier6. This is a statistical
effect of the random emission of electrons (and holes) or the production of photoelectrons. It
is found in vacuum tubes, transistors, and diodes. Shot noise is given by:

ish (rms) = 2qI DC Δf (10)

IDC can be due to signal, bias currents, background radiation (photodetectors7), and leakage.
Like Thermal noise, Shot noise is proportional to √Δf meaning that there is constant noise
power per Hz bandwidth, i.e. it is white noise.
1/f Noise
Also referred to as contact noise (when found in detectors), excess noise (resistors), or
flicker noise (vacuum tubes). 1/f noise is not well understood. It increases without limit as
frequency decreases. It has been measured at frequencies as low as 6 × 10 −5 Hz (≈ 5
cycles/day). Note that while noise density approaches infinity, total noise does not because
the decades keep getting narrower. If an amplifier is 1/f noise limited, measurement
accuracy cannot be improved by increasing the length of the measurement (averaging),
where with white noise, the precision increases with the square-root of the measurement
time. In detectors, it is related to the quality of ohmic contacts and surfaces states. It also
appears in composition-type resistors, carbon microphones, switch and relay contacts,
transistors and diodes and therefore all amplifiers. The equations vary depending on how
the noise was modeled for the application. For a photovoltaic detector

ibα Δf
iRMS = k (11)
fβ
Where:
k = proportionality constant
ib = current through detector
α = typically 2
β = typically ~1
6 Another good reference on dealing with different types of noise is “Noise Reduction Techniques in Electronic
Systems”, Second Edition, Henry W. Ott, John Wiley & Sons, 1988
7 Shot noise applies to photovoltaic detectors. For photoconductive detectors, use G-R noise.

Copyright © 1999 Tim J. Sobering. All rights reserved.

Noise in Electronic Systems Page 6

Note that a photovoltaic detector at zero bias exhibits no 1/f noise.

In general,

⎛f ⎞ ⎛ Δf ⎞ Δf
v f = K ln⎜⎜ h ⎟⎟ = k ln⎜⎜1 + ⎟⎟ ≅ k (12)
⎝ fl ⎠ ⎝ fl ⎠ f
While 1/f noise is often ignored in noise computations where the system bandwidths are
high, it is the dominant noise source in low-frequency applications (e.g. seismic detectors).
Photon Noise
Photon noise is the result of the discrete nature of a radiation field, i.e. random arrival of
protons. This is a fundamental noise source, rising from the detection process. Photon
noise results from both signal radiation and background radiation.
Given an incoming photon stream containing an average n photons in time Δt ,
2
σ p = n = Φ p Δt (13)

Where:
σp2 = variance of the photon flux
Φ = photon flux
While this is normally only applicable to detector systems, one caution is when using glass-
encased diodes. Under the right circumstances, these make good photodetectors.
Generation-Recombination Noise
This noise is due to fluctuations in current carrier generation and recombination. It occurs
in photoconductors.

iG − RRMS = 2qG (ηE p Ad Δf )

1
2

Where:
G = photoconductive gain
q = charge on an electron
η = quantum efficiency
Ep = photon irradiance
Ad = detector area
Δf = effective noise bandwidth
Popcorn Noise
Also known as Burst noise, this type of noise is found in tunnel diodes, junction diodes,
junction transistors, IC’s, and certain resistors. It is proportional to 1/fα for 1 ≤ α ≤ 2. It is
believed to be the result of defects or metallic impurities in the junction which results in
charge filling and discharging from surface traps in a semiconductor. It can only be
improved by improving the manufacturing processes.

Copyright © 1999 Tim J. Sobering. All rights reserved.

Noise in Electronic Systems Page 7

Temperature Noise
This noise is caused by temperature variations of the detector and applies to thermal
detectors such as bolometers, but can also apply to electronic systems in general as there
are many parameters that are temperature sensitive. Usually in electronics this is termed
“drift” and is not technically a noise source, but an error source.
Microphonics
This noise is caused by mechanical displacement of wiring and components when system is
subjected to vibration or shock. The primary cause is capacitance changes due to a change
in wire spacing. Most people ignore or are unaware of this type of noise, but it is easy to
“see” this by tapping on a component or PC board. This is the reason that we encapsulate
all of the die we wirebond. That way, the wires are held in a fixes orientation and no
mechanical capacitance changes can occur.

Copyright © 1999 Tim J. Sobering. All rights reserved.