C IRCU IT INTU ITIONS

Ali Sheikholeslami

“Noise and Distortion, Part I”

W
Welcome to the 39th article in the
“Circuit Intuitions” column series.
As the title suggests, each article
provides insights and intuitions into
circuit design and analysis. These
to understand what circuit design
means at the system level.” He then
continued “The other thing you have
to know, for the rest of your life, [is
that] you have to know something
sure over time, causing vibrations in
our eardrums. Alternatively, when
captured by a microphone, this seg-
ment can be converted into voltage
changes over time.
articles are aimed at undergraduate about noise and something about We can amplify a signal by sim-
students but may serve the interests distortion, and then you are in good ply multiplying it by a constant. For
of other readers as well. If you read shape.” This article is an attempt to example, an audio signal, repre-
this article, I would appreciate your write “something about noise and sented as a voltage waveform, can
comments and feedback as well as something about distortion.” How- be fed into an amplifier and then
your requests and suggestions for ever, before writing about noise and to a speaker, resulting in amplified
future articles in this series. Please distortion, I need to briefly intro- air pressure over time. Whether our
e-mail me your comments at ali@ duce the concept of a signal. original signal is audio, tempera-
ece.utoronto.ca. ture, or any other physical quantity,
The circuit design community What Is a Signal? it can always be converted into a
lost one of its pioneers, Willy San- In general, a signal refers to any voltage or current waveform. This
sen [1], on 25 April 2024. Through physical quantity or any attribute of conversion allows us to amplify or
his prolific life, Willy mentored sev- a physical quantity (such as air pres- process the signal in various ways
eral generations of students, cir- sure or temperature) that varies as before converting it back to the
cuit design engineers, academics, a function of time. In electronics, a desired quantity, such as higher
and industry leaders. In 2011, Willy signal can be a variable, such as a powered audio. Figure 1 illustrates
wrote an article titled “Circuit of node voltage or branch current in this concept; an audio signal is con-
Life” [1] in this magazine, describing a circuit as a function of time. An verted into a voltage signal, pro-
the similarities between analog cir- example of a signal is a 30-s segment cessed by an electronic circuit, and
cuit design and real life. Willy will of a song. Initially, this segment is then converted back into the real
be greatly missed by all of us, but represented by changes in air pres- world, either in the same form, like
his influence will remain with us
for many years to come. I would like
to dedicate this article to him as he
VDD
inspired me to choose this topic. Amplified
When asked during the Joy of Cir- Audio
Audio
cuit Design session of the ISSCC2022
s(t) A × s(t)
Circuit Insights [2] to offer his
A
“advice to new and aspiring circuit
designers who are just starting
their careers,” Willy said: “You have
to leave the level of circuit design
and go higher than that; you have
FIGURE 1: An audio signal, in the form of a variation of air pressure as a function of time,
is captured by a microphone, producing a voltage signal s(t). This signal is multiplied by a
Digital Object Identifier 10.1109/MSSC.2024.3419508 factor of A via the amplifier and fed to a speaker, converting the voltage waveform back to
Date of current version: 23 August 2024 air pressure as a function of time.

IEEE
Authorized licensed use limited to: ARAMCO HQ. Downloaded on February 06,2025 at SOLID-STATE
11:02:31 UTCCIRCUITS MAGAZINE
from IEEE SU M M apply.
Xplore. Restrictions ER 2024 15
sound (as in this example), or in a difficult for others to hear us. How- whose job is to multiply this input by a
different form. As an example of the ever, there are circumstances, such constant factor A (usually larger than 1),
latter, consider the task of creating as an approaching train at a station ideally producing an output in the
light intensity proportional or a loud fire truck form of v out ^ t h = A # ^s ^ t h + n ^ t hh . In
to our voice volume. In this on the street, where the ideal case, the SNR at the ampli-
scenario, a microphone Two crucial these unwanted sig- fier output is identical to the SNR at
translates the air pressure
phenomena nals can overpower the its input simply because both the
affect the quality
into a voltage waveform. desired signal, disrupt- signal and the noise are multiplied
of our signal
An interface circuit then ing conversations. Let by the same factor A. Thus, when
processing and the
adjusts this voltage to a resulting output us represent the desired calculating the SNR, the multipli-
level suitable for a light- signal: noise and signal by s ^ t h and the cation factor A cancels out in the
emitting diode (LED), distortion. noise by n ^ t h . The total ratio. In practical implementations,
which in turn converts signal (referred to as the however, the electronic amplifier
the input voltage (or noisy signal in this article) will generate its own electronic
current) into varying light intensi- can be written as the sum of the noise and will add it to the output
ties. This illustrates the art of con- two: v ^ t h = s ^ t h + n ^ t h . signal. Consequently, even if the
trolling electrons for the purpose of To quantify the strength of the input has zero noise, the output
signal processing [3]. However, two signal relative to the noise in a will contain the noise generated by
crucial phenomena affect the qual- noisy signal, we typically measure the amplifier. As a result, the SNR
ity of our signal processing and the the mean-square value of the signal at the output is lower than the SNR
resulting output signal: noise and (also referred to as the signal power) at the input, indicating that the sig-
distortion. We will focus on noise and the mean-square value of the nal quality is degraded at the output
for the remainder of this article and noise (also referred to as the noise of the amplifier.
address distortion in the next article power). The ratio of these values is You might be curious about the
in this series. called the signal-to-noise ratio (SNR) origins of the electronic noise and
and is expressed in decibels (dB). how we can potentially reduce it.
What Is Noise? In the remainder of this article, we
Noise is an unwanted random signal SNR = 10 log 10 e S o will intuitively address these two
N
that is added to the desired signal, questions.
degrading its quality. An example of where S and N represent the signal
noise is the humming sound of a flu- power and the noise power, respec- Origins and Properties
orescent light bulb or a fridge. This tively. For example, a noisy signal of Electronic Noise
background sound is added to your with an SNR of 20 dB has a signal Any electronic component, by defi-
voice before it reaches a listener or power that is 100 times greater than nition, utilizes electrons for its func-
microphone. This is illustrated sym- the noise power. To recap, noise refers tion. Consider a linear resistor as
bolically in Figure 2, where the air to an unwanted signal that typically an example; if a current i ^ t h flows
pressure reaching the microphone adds to the signal of interest, reduc- through the resistor, a voltage signal
is the sum of the air pressure pro- ing its quality by contaminating v ^ t h = R # i ^ t h will appear across the
duced by your voice (the desired sig- the useful information contained in resistor. It is the flow of electrons
nal) and the air pressure produced the signal. that creates this voltage. But what if
by the fridge and the light bulb the applied current i ^ t h = 0? In this
(the unwanted signals). We usually What Is Electronic Noise? case, we expect the voltage across
speak loudly enough to overpower The noisy signal ^s ^ t h + n ^ t hh may the resistor to be zero—unless we
the noise; otherwise, it would be form the input to a voltage amplifier consider the random movement of
electrons inside the resistor due to
temperature. In fact, the only way to
completely stop the thermal move-
Noise
ment of electrons is to keep the resis-
tor at 0 K. At any other temperature,
there is a random thermal movement
v(t) = s(t) + n(t) of electrons, which will contribute
Desired
Signal + to a current within the resistor. This
current will create a random voltage
across the resistor. As a result, the
total voltage across the resistor can
FIGURE 2: The desired audio signal and noise will add up before converting to a voltage be written as v ^ t h = R # i ^ t h + v n ^ t h,
signal by the microphone. where i ^ t h represents the current

16 SU M M E R 2 0 2licensed
Authorized 4 IEEE
useSOLID-STATE CIRCUITSHQ.
limited to: ARAMCO MAGAZINE
Downloaded on February 06,2025 at 11:02:31 UTC from IEEE Xplore. Restrictions apply.
applied to the resistor, and v n ^ t h carefully considered when determin- as electrons) causes fluctuations in
represents the thermal noise voltage ing the signal quality at the circuit out- the current flowing through a circuit.
of the resistor. This is also known as put. Given that an amplifier consists of The combined noise cannot be
Johnson-Nyquist noise, named after many transistors with different sizes, fully characterized by its average
the two scientists who first char- bias currents, etc., it produces a more power alone; it requires the noise
acterized this noise in their 1928 complicated noise waveform. Addi- power spectral densit y. Let us
­articles [4], [5]. tionally, besides the thermal elaborate on this a bit
The power of the thermal noise noise discussed earlier, there more.
voltage is proportional to the abso- are other types of electronic The generated
lute temperature, the value of the noise, such as flicker noise noise must How to Characterize
resistor, and the bandwidth. This (also known as 1/f noise be carefully Noise
considered when
relationship is typically expressed or pink noise) and shot Given that noise is a
determining the
as v 2n = 4kTRTf, where v 2n denotes noise, each with its own random signal, it does
signal quality at
the mean-square value of the noise unique ­characteristics. the circuit output. not have a deterministic
voltage, k is the Boltzmann constant, Flicker noise is char- waveform. To illustrate
T is the absolute temperature, R is acterized by a power this, Figure 3 shows the
the resistance, and Tf is the noise spectral density that decreases with noise voltage waveforms gener-
measurement bandwidth. increasing frequency; hence the name ated by 100 resistors (only three
It is interesting to note that at 1/f noise. This type of noise is preva- are shown), each with a resistance
room temperature (with T = 290 K), lent in many electronic devices and of 1 kX, limited by a 10-kHz band-
kT is approximately 4.0 # 10 -21 J. is significant at low frequencies. Shot width at T = 290 K. By examin-
This means that a 1k - X resistor with noise arises from the discrete nature ing these waveforms, we aim to
a 10-kHz signal bandwidth will pro- of electric charge. It occurs when the find common characteristics among
duce a noise power of 1.6 # 10 -13 V 2, finite number of charge carriers (such them. These are as follows:
or 0.4 nVrms . Assuming that our
desired signal is a 100-mV sinu-
soidal voltage across this resis-
tor, the signal power is 3.1 # 10 10 1
Amp (µv)

times larger than the noise power

(SNR = 105 dB) . 0
The fact that this noise power is
–1
proportional to the bandwidth implies (a)
that the noise power spectral density 1
Amp (µv)

(PSD), which is the noise power per

unit frequency of bandwidth, is con- 0
stant. For the linear resistor, the noise
–1
PSD is simply 4 kTR. This means that
(b)
for every 1 Hz of bandwidth, the noise
1
has a constant power of 4 kTR. This
Amp (µv)

is very similar to white light, where 0

the power spectral density is constant
across the frequency spectrum (or –1
(c)
color). Due to this similarity, the noise
1
produced by a resistor is also referred
Amp (µv)

to as white noise. 0 97 Other Waveforms (Not Shown)

It turns out that the random move-
ment of electrons exists in all elec- –1
(d)
tronic components, not just resistors.
1 Average of 100 Noise Waveforms
This includes diodes and transistors
Amp (µv)

(both BJT and MOSFET). However, 0

deriving their equivalent noise voltage
or noise current involves more physi- –1
cal phenomena and ­underlying math- 0 2 4 6 8 10
ematics. For now, it suffices to say Time (ms)
(e)
that any electronic circuit will generate
noise in addition to its intended func- FIGURE 3: (a)–(e) Noise waveforms are generated by 100 identical resistors (only three are
tion. The generated noise must be shown), and the average of the 100 noise waveforms is shown as function of time.

IEEE
Authorized licensed use limited to: ARAMCO HQ. Downloaded on February 06,2025 at SOLID-STATE
11:02:31 UTCCIRCUITS MAGAZINE
from IEEE SU M M apply.
Xplore. Restrictions ER 2024 17
1) All of the 100 waveforms exhibit as we found earlier. This charac- having a uniform power distribution
a zero (or close to zero) time av- teristic is important as it directly across the bandwidth. However, we
erage, indicating that, over a long influences the SNR. note that this is only a characteristic
period, the positive and negative 3) The average of the power spec- of white noise. In general, the noise
fluctuations of the noise voltage tral densities of the 100 noise PSD is not constant, and therefore, it
cancel each other out. This is a waveforms is constant across fre- will affect different frequency compo-
typical characteristic of a purely quency at about -168 dBV 2 /Hz. nents of the signal to varying degrees.
random signal. This last property is significant To recap, we characterize noise by
2) The average power of each wave- because it confirms the white noise analyzing its time average, average
form is approximately ^0.4 nVh2, characteristic of thermal noise, i.e., power, and power spectral density.
Now, let us feed one sample noise
waveform (taken from Figure 3) into
a noiseless bandpass filter, as shown
–160 Input PSD 10
in Figure 4, and observe the noise at
–170 0 the filter output. Given that the filter
PSD (dB v2 /Hz)

Filter Gain (dB )

–180 –10 exhibits varying degrees of attenu-
ation at each frequency, the output
–190 Output PSD –20
will no longer have a constant power
–200 Filter –30 density across the frequency spec-
Gain
–210 –40 trum. Instead, it will be shaped by the
filter, as illustrated by the red curve
–220 –50
in Figure 4. The noise at the output is
0 2 4 6 8 10
Frequency (kHz) no longer white, but colored.
How does this coloring change
FIGURE 4: The PSD average of the 100 noise waveforms of Figure 3 (shown in orange), the the noise waveform in the time
bandpass filter gain (shown in blue), and the filter output corresponding to the first noise domain? As shown in Figure 5, the
waveform in Figure 3 (shown in red). time average of the output noise
is still zero, but its amplitude (and
hence its power) is reduced. Further-
more, due to its limited bandwidth,
1 the output noise waveform shows
Amp (µv)

less variation (less activity).

0

–1 How to Separate Signal From Noise

0 2 4 6 8 10 Once noise is added to a signal, it often
Time (ms) becomes indistinguishable from the
signal and very difficult to separate,
FIGURE 5: The time-domain noise at the filter input (in orange) and output (in red). unless we “know something about
the signal and something about the
noise.” We will illustrate this through
4 a simple example here.
Signal + Noise
2 Consider a noisy audio signal, as
Amp (mv)

shown in Figure 6, with the signal

0
content in the range of 4–6 kHz (i.e.,
–2
a 2-kHz bandwidth) but with the
–4
thermal noise content limited to the
(a)
4 20-kHz range. Since we “know” that
Original Signal Recovered Signal the signal is in the 4–6-kHz band, we
2
Amp (mv)

can simply design a bandpass filter

0 with a 2-kHz bandwidth to let the
–2 entire signal power through while
–4 allowing only 10% (2 kHz/20 kHz)
0 0.2 0.4 0.6 0.8 1
of the noise power to pass. As a
Time (ms)
(b) result, this simple act of filtering
is expected to improve the SNR by
FIGURE 6: (a) and (b) By bandpass filtering the noisy signal, only a small amount of noise
remains in the recovered signal. (continued on p. 26)

18 SU M M E R 2 0 2licensed
Authorized 4 IEEE
useSOLID-STATE CIRCUITSHQ.
limited to: ARAMCO MAGAZINE
Downloaded on February 06,2025 at 11:02:31 UTC from IEEE Xplore. Restrictions apply.
 summing junction. We take for times it’s okay if you have loop References
[1] C. Mangelsdorf, “Solving tough feedback
granted that the positive and neg- gain greater than one, but in some problems (without hair loss) [shop talk:
ative terminals of any op-amp will instances, you need to make sure what you didn’t learn in school],” IEEE
Solid State Circuits Mag., vol. 14, no. 2,
be at nearly the same voltage. This that never, ever happens. Be cer- pp. 18–26, Spring 2022, doi: 10.1109/
is not true with positive feedback. tain your design incorporates all MSSC.2022.3167301.
[2] B. Razavi, “The cross-coupled pair - Part I [a
With loop gain near one, the error process and environmental toler- circuit for all seasons],” IEEE Solid State Cir-
voltage (the output of the sum- ances that might affect the gain. cuits Mag., vol. 6, no. 3, pp. 7–10, Summer
2014, doi: 10.1109/MSSC.2014.2329234.
ming junction) is the same order [3] C. Mangelsdorf, “Metastability: Deeply
as—or larger than!—the output Summary Junction misunderstood [shop talk: what you didn’t
learn in school],” IEEE Solid State Circuits
signal. Normal amp input stages Positive feedback is a challenging Mag. , vol. 16, no. 2, pp. 8–15, Sping 2024,
designed for large negative feed- technique. Using it correctly will tax doi: 10.1109/MSSC.2024.3395109.
[4] B. Razavi, “The cross-coupled pair? Part III
back cannot handle this: they have all your circuit skills and raise your [a circuit for all seasons],” IEEE Solid State
a tiny input range beyond which design level. But remember: positive Circuits Mag., vol. 7, no. 1, pp. 10–13, Win-
ter 2015, doi: 10.1109/MSSC.2014.2369332.
they rapidly distort and saturate. feedback is a powerful tool, and [5] A. S. Sedra, “The current conveyor: History
■ Manufacturability: One of the nic- like any powerful tool, you can hurt and progress,” in Proc. IEEE Int. Symp. Circuits
Syst., Portland, OR, USA, 1989, vol. 3, pp.
est things about negative feedback yourself with it. Stay alert! 1567–1571, doi: 10.1109/ISCAS.1989.100659.
is that you can’t have too much Please share your thoughts, exam- [6] C. Mangelsdorf, “Brokaw bonus tracks: A
few lesser-known tunes,” IEEE Solid State
gain at low frequencies. So, you ples and experiences by ­ writing to Circuits Mag., vol. 13, no. 1, pp. 22–27, Win-
make more gain than you need shifobrains@ieee.org. If you’re inter- ter 2021, doi: 10.1109/MSSC.2020.3037387.
[7] T. H. Lee, The Design of CMOS Radio-
and forget about it. Some posi- ested, we can go even deeper into Frequency Integrated Circuits. Cambridge,
tive feedback circuits are equally this topic. Want to see more? Let U.K.: Cambridge Univ. Press, 2004, p. 444.
tolerant, but many are not. Some- me know. 

C IRCU IT INTU ITIONS (continued from p. 18)

10 dB. Indeed, the recovered signal the scope of this article. If you for his assistance with preparing
looks much closer to the original find this topic fascinating, you are simulation results for this article.
signal after filtering (see Figure 6). encouraged to consider graduate
However, some deviation from the studies in circuit design. References
[1] W. Sansen, “Circuit of life,” IEEE Solid
original signal remains because the In summary, any electronic cir- State Circuits Mag., vol. 3, no. 1, pp. 6–21,
noise is not completely eliminated. cuit, regardless of its function— Winter 2011, doi: 10.1109/MSSC.2010.
939568.
Intuitively, we understand that be it amplification, integration, or [2] ISSCC 2022 Circuit Insights: The Joy of
the signal of interest has a rather ­filtering—generates noise in addition Circuit Design. (2022). Accessed: Jun. 10,
2024. [Online Video]. Available: www.
limited rate of change (due to its lim- to performing its intended operation. youtube.com/watch?v=zgpsQSHgzhw
ited bandwidth), whereas the rate of This added noise degrades the signal [3] A. Sheikholeslami, “The art of controlling
electrons [circuit intuitions],” IEEE Solid
change of the noise signal can be quality at the output, resulting in an State Circuits Mag., vol. 14, no. 2, pp. 11–
well above and well below that of SNR that is lower than the input SNR. 17, Spring 2022, doi: 10.1109/MSSC.2022.
3164811.
the signal. The act of bandpass fil- Over the years, circuit designers have [4] H. Nyquist, “Thermal agitation of electric
tering (in the frequency domain) is strived to understand, characterize, charge in conductors,” Phys. Rev., vol. 32,
no. 1, pp. 110–113, Jul. 1928, doi: 10.1103/
similar to limiting the rate of change predict, mitigate, and, in some cases, PhysRev.32.110.
of the signal in the time domain. cancel noise. [5] J. B. Johnson, “Thermal agitation of elec-
tricity in conductors,” Phys. Rev., vol. 32,
The characterization of noise no. 1, pp. 97–109, Jul. 1928, doi: 10.1103/
behavior in a general circuit that Acknowledgment PhysRev.32.97.
includes several transistors is quite I would like to thank my Ph.D. stu-
interesting; however, it is beyond dent, Mohammad Emami Meybodi, 

26 SU M M E R 2 0 2licensed
Authorized 4 IEEE
useSOLID-STATE CIRCUITSHQ.
limited to: ARAMCO MAGAZINE
Downloaded on February 06,2025 at 11:02:31 UTC from IEEE Xplore. Restrictions apply.