electronics

Review
Photoplethysmography in Wearable Devices: A Comprehensive
Review of Technological Advances, Current Challenges, and
Future Directions
Kwang Bok Kim 1 and Hyun Jae Baek 2, *

1 Digital Health Care R&D Department, Korea Institute of Industrial Technology (KITECH),
Cheonan 31056, Republic of Korea; [email protected]
2 Department of Biomedical Engineering, Soonchunhyang University, Asan 31538, Republic of Korea
* Correspondence: [email protected]; Tel.: +82-41-530-4991

Abstract: Photoplethysmography (PPG) is an affordable and straightforward optical technique used

to detect changes in blood volume within tissue microvascular beds. PPG technology has found
widespread application in commercial medical devices, enabling measurements of oxygen saturation,
blood pressure, and cardiac output; the assessment of autonomic nerve function; and the diagnosis of
peripheral vascular disease. Recently, the growing demand for non-invasive, portable, cost-effective
technology, along with advancements in small semiconductor components, has led to the integration
of PPG into various wrist-worn wearable devices. Multiple sensor structures have been proposed
and, through appropriate signal processing and algorithmic application, these wearable devices can
measure a range of health indicators during daily life. This paper begins by addressing the market
status of wrist-worn wearable devices, followed by an explanation of the fundamental principles
underlying light operation and its interaction with living tissue for PPG measurements. Moving
on to technological advancements, the paper addresses the analog front end for the measurement
of the PPG signal, sensor configurations with multiple light emitters and receivers, the minimum
sampling rate required for low-power systems, and the measurement of stress, sleep, blood pressure,
blood glucose, and activity using PPG signals. Several challenges in the field are also identified,
including selecting the appropriate wavelength for the PPG sensor’s light source, developing low-
Citation: Kim, K.B.; Baek, H.J.
power interpolation methods to extract high-resolution inter-beat intervals at a low sampling rate,
Photoplethysmography in Wearable
Devices: A Comprehensive Review
and exploring the measurement of physiological phenomena using multi-wavelength PPG signals
of Technological Advances, Current simultaneously collected at the same location. Lastly, the paper presents future research directions,
Challenges, and Future Directions. which encompass the development of new, reliable parameters specific to wearable PPG devices and
Electronics 2023, 12, 2923. https:// conducting studies in real-world scenarios, such as 24-h long-term measurements.
doi.org/10.3390/electronics12132923
Keywords: photoplethysmography; PPG; wearable device; wrist wearable; heart rate monitoring
Academic Editor: Giuseppe
sensor; mobile healthcare; wearable healthcare
Prencipe

Received: 1 June 2023

Revised: 21 June 2023
Accepted: 25 June 2023 1. Introduction
Published: 3 July 2023
In the past, healthcare mostly focused on hospital-oriented care, where patients were
treated in hospitals by healthcare professionals. However, in recent years, there has been a
paradigm shift towards individual-oriented care, where patients are encouraged to take
Copyright: © 2023 by the authors.
a more active role in managing their own health [1–3]. Wearable healthcare devices are
Licensee MDPI, Basel, Switzerland.
playing an important role in this paradigm shift. The need for wearable healthcare devices
This article is an open access article is driven by the desire for the continuous monitoring of vital signs and health metrics
distributed under the terms and in a non-invasive and convenient way, which can help to improve health outcomes and
conditions of the Creative Commons quality of life for individuals. Wearable healthcare devices can help to detect and monitor a
Attribution (CC BY) license (https:// range of health conditions, such as heart disease, diabetes, and sleep disorders [4–6]. By
creativecommons.org/licenses/by/ providing continuous monitoring of vital signs, wearable devices can help individuals
4.0/). to detect potential health issues early on and take appropriate steps to manage them.

Electronics 2023, 12, 2923. https://doi.org/10.3390/electronics12132923 https://www.mdpi.com/journal/electronics

Electronics 2023, 12, 2923 2 of 24

Wearable healthcare devices allow individuals to track their health metrics on a daily
basis, providing valuable insights into their health status and helping them to make
informed decisions about their lifestyles and habits. This tracking can be particularly
useful for individuals with chronic conditions or those looking to improve their overall
health and wellness. They can help to improve patient outcomes by providing real-time
monitoring of vital signs, allowing healthcare providers to quickly detect and respond to
changes in a patient’s condition. Among the various types of available wearable healthcare
devices, wrist-worn wearables have emerged as the most popular, with a wide variety of
products now on the market [7]. The prospects of future wrist-worn health management
devices are promising, with ongoing advancements in sensor technology, signal processing
algorithms, and data analysis techniques. These devices have potential for personalized
health monitoring, disease detection and prevention, and remote patient monitoring. These
developments could lead to improved healthcare outcomes and greater accessibility to
healthcare for individuals.
Wrist-wearable devices typically include several sensors that enable them to track
various aspects of the wearer’s health and fitness. Inertial measurement unit (IMU) sensors
are often included in wrist-wearable devices for healthcare applications to track the move-
ment and orientation of the device in three-dimensional space. An IMU typically includes
a combination of accelerometers, gyroscopes, and magnetometers. Together, these sensors
provide valuable information about the wearer’s physical activity and movement patterns,
which can be used for healthcare applications, such as steps taken, distance traveled, calo-
ries burned, and fall detection [8–10]. Recently, photoplethysmographic sensors (PPG) have
been widely implemented. PPG is a non-invasive optical technology that measures changes
in blood volume in tissue by detecting changes in the amount of light absorbed by the
tissue [11]. In the context of wrist-wearable devices, PPG sensors are used to measure the
pulsatile blood flow in the arteries of the wrist. The PPG sensor consists of a light source,
typically an LED, and a photodetector. The LED emits light that is absorbed by the tissue in
the wrist, and the photodetector measures the amount of light that is transmitted through or
reflected from the tissue. As blood flows through the arteries, it absorbs more light, causing
a decrease in the amount of light that reaches the photodetector, resulting in a pulsatile
waveform that corresponds to the cardiac cycle. PPG sensors can be used to measure a
variety of physiological parameters, including the heart rate, heart rate variability, blood
pressure, blood oxygen saturation, and sleep quality [12,13]. To measure these parameters,
the PPG sensor typically collects data continuously or intermittently over a period of time,
and the data are then processed using algorithms to extract the desired information. While
PPG technology has many benefits for wearable applications due to its non-invasive and
convenient nature, there are also several challenges and limitations associated with PPG in
wearable applications. The movement of the wearer can cause motion artifacts in the PPG
signal, which can lead to inaccurate measurements of the heart rate and other physiological
parameters [14,15], which is particularly problematic for applications where the wearer may
be engaging in physical activity. While wearable PPG can provide valuable information on
the heart rate, oxygen saturation, and other physiological parameters, its accuracy may be
limited for certain measurements. For example, wearable PPG may not be as accurate as
electrocardiography (ECG), even as a finger-clip-type transmittance-mode PPG to measure
heart rate variability. Wearable PPG devices rely on batteries, which limits their operating
time. Frequent charging or battery replacement may be necessary, especially for continuous
monitoring applications. These limitations have been addressed through the development
of new sensor designs, algorithms, and data processing techniques that aim to improve the
accuracy and reliability of PPG measurements.
Wearable PPG technology has been accessible for several decades, but it was not until
the early 2010s that it became more widely commercialized and popularized. This phe-
nomenon was largely due to the development of more accurate and reliable PPG sensors, as
well as advances in miniaturization and wireless connectivity that made the incorporation
of PPG sensors into small, wearable devices practicable. The first commercially available
Electronics 2023, 12, 2923 3 of 24

wrist-wearable device with a PPG sensor, the Mio Alpha, was released in 2012, followed
shortly thereafter by other popular devices, such as the Fitbit Charge HR and the Basis
B1 Band. In the smartwatch product line, after the first PPG sensor was introduced in the
Samsung Galaxy Gear2 in 2014, the PPG sensor was also installed in the Apple Watch in
2015. Since then, wearable PPG technology has continued to evolve and improve, with
many new devices incorporating more advanced sensors and algorithms to monitor a
wider range of physiological parameters. Today, wearable PPG technology is broadly used
in a variety of applications, including health and fitness monitoring, clinical research, and
medical diagnostics. In early 2010, as wrist-wearable devices with PPG sensors began to be
commercialized, a review of wearable PPG systems was published in the journal Electron-
ics [16]. In the review paper, various wearable form factors, such as rings and earphones,
as well as wrist-worn devices, were discussed, and motion noise cancellation methods
were carefully reviewed. The functionality of PPG sensors in wrist-wearable devices has
evolved significantly over the subsequent years. The earliest function of PPG sensors in
wrist wearables was to solely measure the heart rate, which has since become a standard
feature in most wrist-wearable devices. More advanced PPG technologies are now being
used to monitor various other health metrics. This functionality is still relatively new and
requires more rigorous testing and validation, but it has the potential to greatly enhance the
capabilities of wrist-wearable devices as health monitoring tools. As the last comprehensive
review of wearable PPG technology was published a decade ago, this article aims to review
the latest research developments that have enabled various health monitoring functions
through PPG sensors.

2. Working Principle
Human tissue exhibits a significant degree of light scattering [17]. When photons
emitted from a light source impinge on tissue, they can take several different paths [18].
As shown in Figure 1a, some photons are absorbed into the tissue (p1), and some pass
through the tissue through several pathways to reach the photodetector located on the
opposite side of the light source across the tissue (p2). In addition, some photons pass
through the tissue but do not reach the photodetector, resulting in a loss of light (p3), and
some are scattered and reflected (p4). Transmissive PPG, widely used in clinical practice
for pulse oximetry, involves measuring p2. This type of PPG sensor is typically placed
on the fingertip, with the light source and detector positioned on opposite sides of the
finger. When properly positioned and secured, finger-based PPG can yield high-quality
signals with a good signal-to-noise ratio. However, it may cause discomfort or restrict hand
movement during usage. Reflective PPG, used in wearable devices, records p4. In this
setup, both the light source and the detector are located on the same side of the biological
tissue. Consequently, reflectance PPG sensors can be easily integrated into wearable devices
such as smartwatches or fitness trackers, allowing continuous monitoring without the need
for specific sensor placement on opposite sides of the tissue.
As represented by p4 in Figure 1a, some of the scattered photons emitted from the
light source return to the tissue surface and reappear for detection. This phenomenon of
scattered photons suggests that the light reflected by the tissue back to the detector has
partially undergone a combination of scattering and absorption. Therefore, by measuring
the change in the intensity of the reflected light as a function of the wavelength, the relative
change within the tissue can be determined. According to the Beer–Lambert law, the path
of most photons encountered by the detector generally follows an arc shape, as shown
in Figure 1b [19,20]. The wavelength of the light source and the size of the curve have
a proportional relationship. In other words, the longer the wavelength, the deeper into
the skin the light penetrates. Photons emitted from the light source undergo processes of
scattering and absorption in human tissue, blood vessel walls, and blood and are attenuated
along a curved path according to the wavelength, while the total reflectance, R, can be
simplified as the absorbance, A, through Equation (1) [21].
Electronics2023,
Electronics 2023,12,
12,2923
x FOR PEER REVIEW 44 of 24
of 24

Figure 1. (a)
Figure (a)Absorbed
Absorbedand andreflected
reflectedlight in living
light tissue
in living for PPG
tissue model.
for PPG (b) the(b)
model. measurement model
the measurement
of reflectance
model type PPG
of reflectance typethrough the skin
PPG through microvascular
the layer, with
skin microvascular layer,emphasis on the on
with emphasis rolethe
ofrole
wave-
of
length. The amount
wavelength. of absorbed
The amount light correlates
of absorbed with the
light correlates pulsation
with of arterial
the pulsation blood. In
of arterial the systolic
blood. In the
phase, the
systolic diameter
phase, of the arterial
the diameter vessels is
of the arterial maximal
vessels and therefore
is maximal the absorbance
and therefore due to arterial
the absorbance due to
hemoglobin is also maximal and the amount of detected light is low, which corresponds to a sensor
arterial hemoglobin is also maximal and the amount of detected light is low, which corresponds to a
peak.
sensor peak.

As represented by p4 in Figure 1a, some of the scattered photons emitted from the
light source return to the tissue surface and reappear 1 for detection.IThis(λ) phenomenon of
A(λ) = −log10 R(λ) = log10 , where R(λ) = (1)
scattered photons suggests that the light reflected
R(λ) by the tissue back Io (λto) the detector has
partially undergone a combination of scattering and absorption. Therefore, by measuring
The term Io (λ) is defined as the total photon flow emitted from the light source, and
the change in the intensity of the reflected light as a function of the wavelength, the rela-
I (λ) is the total photon flow entering the detector. Typically, reflectance measurements
tive change within the tissue can be determined. According to the Beer–Lambert law, the
are calculated for standard reflective surfaces, known as relative reflectance. This mea-
path of most
surement, photonsinencountered
expressed Equation (2),by the ratio
is the detectorof the generally follows
reflectance R of an thearc shape,
tissue to theas
shown in Figure 1b [19,20]. The wavelength of the light source and
reflectance R0 of the white reflectance standard or background, such as a reference surface the size of the curve
have
or a proportional
instrument relationship. In other words, the longer the wavelength, the deeper
calibrator.
into the skin the light penetrates. Photons emitted R from
(λ) the light source undergo processes
A(λ) = −log10 (2)
of scattering and absorption in human tissue, blood R0 (vessel
λ) walls, and blood and are atten-
uated along a curved path according to the wavelength, while the total reflectance, 𝑅, can
As shown in Figure 1b, the light source and detector are separated such that the average
be simplified as the absorbance, A, through Equation (1) [21].
path of light in the tissue follows a curved trajectory before reaching the detector. From
these reflectance arrays with different source–detector 1 𝐼(𝜆)
spacing, the penetration depth of
A(λ) = −𝑙𝑜𝑔 𝑅(λ) = 𝑙𝑜𝑔 , 𝑤ℎ𝑒𝑟𝑒 𝑅(λ) = (1)
light becomes approximately one10third to one10half 𝑅(λ) of the separation distance
𝐼𝑜 (𝜆) into the tissue.
Photoplethysmography measures the effect of arterial blood in tissue on the intensity
The term 𝐼𝑜 (𝜆) is defined as the total photon flow emitted from the light source, and
of the transmitted or reflected light [11]. As shown in Figure 2, the volume of blood in the
𝐼(𝜆) is the total photon flow entering the detector. Typically, reflectance measurements
tissue is a function of the arterial pulse, the AC component, with a greater volume present
are calculated for standard reflective surfaces, known as relative reflectance. This meas-
at the systole and a smaller volume present at the diastole, which is the DC component.
urement,the
Because expressed in Equation
blood absorbs most of(2),
theislight
the passing
ratio of through
the reflectance 𝑅 of
the tissue, thethe tissue to
intensity the
of the
reflectance
light emerging 𝑅 of the white reflectance standard or background, such
0 from the tissue is inversely proportional to the volume of blood present as a reference sur-
face
in theortissue.
instrument calibrator. principle of PPG can be explained by the modified Beer–
The measurement
Lambert law, expressed in Equations (1) and (3), where 𝑅(λ)I (λ) is the detected light intensity,
Io (λ) is the incident light intensity, εA(λ) (λ) is=the−𝑙𝑜𝑔 10
molar 𝑅0extinction
(λ) coefficient, C is the molar (2)
concentration, l (λ) is the average path length, and G (λ) is an appropriate factor describing
As shown ingeometry
the measurement Figure 1b,[22,23].
the light
Thesource
signaland
thatdetector are separated
records changes in I (λ)such
due that the av-
to pulsatile
erage path of light in the tissue follows a curved trajectory before
cardiac activity is called photoplethysmography (PPG). Here, assuming that the absorbing reaching the detector.
From these
material reflectance
of the arrayslight
initial incident withis different source–detector
a tissue-related componentspacing,
including the penetration
blood or skin,
the amplitude of the AC component of the PPG signal can be expressed as in Equationinto
depth of light becomes approximately one third to one half of the separation distance (4).
the tissue. b and t denote blood and tissue. Tissue does not change the length of the light
Subscripts
travelPhotoplethysmography
path with time, but, inmeasures the case of the effectthe
blood, of arterial bloodchanges
travel path in tissueaccording
on the intensity
to the
of the transmitted or reflected light [11]. As shown in Figure 2, the volume of blood in the
tissue is a function of the arterial pulse, the AC component, with a greater volume present
at the systole and a smaller volume present at the diastole, which is the DC component.
Because the blood absorbs most of the light passing through the tissue, the intensity of the
 𝐼𝑜 (𝜆) is the incident light intensity, ε(λ) is the molar extinction coefficient, C is the molar
concentration, 𝑙(̅ 𝜆) is the average path length, and 𝐺(𝜆) is an appropriate factor describ-
ing the measurement geometry [22,23]. The signal that records changes in 𝐼(𝜆) due to
pulsatile cardiac activity is called photoplethysmography (PPG). Here, assuming that the
absorbing material of the initial incident light is a tissue-related component including
Electronics 2023, 12, 2923 blood or skin, the amplitude of the AC component of the PPG signal can be expressed 5 of 24as
in Equation (4). Subscripts b and t denote blood and tissue. Tissue does not change the
length of the light travel path with time, but, in the case of blood, the travel path changes
accordingand
contraction to the contraction
relaxation of theand relaxation
heart. Therefore,of the heart.
in the caseTherefore, in the
of blood with case ofb,blood
subscript the
with subscript
systolic period is b, the systolic
indicated period
by sys and theis indicated by sys by
diastolic period andsubscript
the diastolic
dia. period by sub-
script dia.
Io (λ)
A(λ) = log = ε(λ)Cl (λ, C ) + G (λ)
𝐼 (𝜆) (3)
( λ )𝑜
A(λ) = Ilog = ε(λ)C𝑙 (̅ 𝜆, 𝐶) + 𝐺(𝜆) (3)
𝐼(𝜆)

h i 𝑡 (𝜆)𝐶𝑡 𝑙𝑡̅ (𝜆)

𝐴𝐶 = 𝐼𝑜 (𝜆)exp[−𝜀 h − 𝜀𝑏 (𝜆)𝐶𝑏 𝑙𝑏−𝑑𝑖𝑎̅ (𝜆)] i
AC = Io (λ)exp −ε t (λ)Ct l t (λ) − ε b (λ)Cb l b−dia (λ) −− )exp −εt (λ(𝜆)𝐶
I0 (𝐼λ(𝜆)exp[−𝜀 λ) −
)Ct l t𝑙(̅ (𝜆) ( ) C l ( )
− 𝜀b (𝜆)𝐶b 𝑙b̅ −sys (𝜆)]
ε λ λ (4)(4)
0 𝑡 𝑡 𝑡 𝑏 𝑏 𝑏−𝑠𝑦𝑠

Figure
Figure 2.2.Absorbed
Absorbedand
and reflected
reflected light in living
light in livingtissue
tissueforforPPG
PPGmodel.
model.The
The amount
amount of absorbed
of absorbed light
light correlates
correlates with
with thethe pulsation
pulsation of arterial
of arterial blood.
blood. In the
In the systolic
systolic phase,
phase, thethe diameter
diameter of the
of the arterial
arterial ves-
vessels
sels isismaximal
maximalandandtherefore
therefore the
the absorbance due to to arterial
arterialhemoglobin
hemoglobinisisalso
alsomaximal
maximalandandthe the
amount
amount ofof detected
detected light
light is is low,
low, which
which is displayed
is displayed as as a peak.
a peak.

According
AccordingtotoEquations
Equations(3)(3)and
and(4),
(4),the
thelight
lightintensity
intensitydetected
detectedbybythe thephotodetector
photodetector
ofofthe
thereflectance-mode
reflectance-modePPG
PPGsensor
sensordepends
dependsononthe theintensity
intensityofofthe
theincident
incidentlight,
light,the
the
blood volume in
blood volume in thethe vascular bed, absorption by the skin, reflection by the bone,
bed, absorption by the skin, reflection by the bone, and tissue and
tissue backscattering.
backscattering.

3.3.Technological
TechnologicalAdvances
Advances
3.1. Sensor
3.1. Sensor
Optical heart rate monitoring sensors have progressively developed to become smaller,
Optical heart rate monitoring sensors have progressively developed to become
more efficient, and more accurate [24–33]. One of the key trends in sensor development
smaller, more efficient, and more accurate [24–33]. One of the key trends in sensor devel-
in the early 2010s, as wearable healthcare devices such as smartwatches began to become
opment in the early 2010s, as wearable healthcare devices such as smartwatches began to
popular, was the integration of the LED and photodiode (PD) into a single package with
anbecome popular,
integrated analogwas theend.
front integration of the LED
An integrated and sensor
LED–PD photodiode
module(PD) into a single
is smaller and
more compact than a conventional optical sensor with separate LED and PD components,
which makes integration into wearable devices or other small-form-factor products more
straightforward. Moreover, the combined module consumes less power than a conventional
optical sensor with separate LED and PD components. Furthermore, this type of LED–PD
optopair system features an analog front-end IC (AFE). An AFE is an electronic circuit
that processes the analog signal generated by the LED–PD sensor, amplifying and filtering
the signal to remove noise and interference. As a result, the trend in optical heart rate
sensor development has been towards smaller, integrated solutions with an AFE, which are
more accurate, reliable, and power-efficient than previous generations of sensors. Table 1
Electronics 2023, 12, 2923 6 of 24

shows representative, commercially available AFEs that feature an integrated LED–PD

optopair system.

Table 1. Analog front-end (AFE) modules for PPG measurement with a built-in sensor system
composed of LED–PD pairs on the market.

Size (mm)
AFE Module Manufacturer LED PD AFE Unit Price (1)
(L × W × H)
Pulse oximetry and
1 Green, 1 Red,
MAX30101 [24] 1 PD 5.6 × 3.3 × 1.55 heart rate monitor USD 11.21
1 IR
module
Maxim ECG, pulse
MAX86150 [25] Integrated 1 Red, 1 IR 1 PD 5.6 × 3.3 × 1.3 oximetry, heart rate USD 7.47
monitor module
Optical sensor
1 Blue, 1 Green,
MAX86916 [26] 1 PD 7.0 × 3.5 × 1.5 module for PPG, USD 8.95 (2)
1 Red, 1 IR
proximity and color
Pulse oximetry and
ADPD144RI
1 Red, 1 IR 4 PD 5.0 × 2.8 × 1.35 heart rate monitor USD 11.88
[27]
Analog Devices module
ADPD188GG Heart rate monitor
2 Green 2 PD 5.0 × 3.8 × 0.9 USD 11.8
[28] module
Heart rate monitor
SFH 7051 [29] 3 Green 1 PD 4.7 × 2.5 × 0.9 N.A (3)
module
Pulse oximetry and
2 Green, 1 Red,
SFH 7072 [30] 2 PD 7.5 × 3.9 × 0.9 heart rate monitor USD 3.47
OSRAM 1 IR
module
Pulse oximetry and
1 Green, 1 Red,
SFH 7050 [31] 1PD 4.7 × 2.5 × 0.9 heart rate monitor N.A (3)
1 IR
module
2.7 × 6.1 × N.A
AS7024 [32] 2 Green, 1 IR 4 PD (4) PPG and ECG N.A (3)
ams AG
2.75 × 6.2 ×
AS7026GG [33] 2 Green 1 IR 4 PD PPG and ECG USD 3.11
N.A (4)
(1)The unit prices mentioned were obtained from digikey.com on 14 June 2023. Please note that prices may vary
over time and are subject to change. (2) The price stated is for each individual piece when a quantity of 2500 pieces
is ordered. (3) The product has been discontinued, and the price is no longer available. (4) The height information
is not provided in the data sheet; only the length and width dimensions are specified.

While integrated LED–PD sensors offer many advantages for PPG monitoring, there
are also some potential disadvantages when compared to traditional sensors that use
separate LEDs and PDs. Integrated LED–PD sensors are designed to be a one-size-fits-all
solution, which means that they may not be optimal for every use. Customizing the sensor
to specific requirements may not be possible, which could limit its usefulness in certain
applications. In addition, due to the compact design of integrated LED–PD sensors, it can
be challenging to optimize the spacing and positioning of the LED and photodiode, which
can lead to a decrease in signal quality and make it more difficult to detect subtle changes
in blood flow and oxygen saturation [34]. This issue is less common with traditional
sensors that use separate LEDs and photodiodes because the spacing and positioning
can be more easily adjusted to optimize the signal quality. As illustrated in Figure 3a,
an integrated LED–PD sensor module typically includes an optical barrier that separates
the light-emitting LED and receiving PD parts with the LEDs and PDs either exposed on
the surface of the sensor or coated with transparent epoxy. Although the light-emitting
and receiving parts are completely separated within the sensor module, a gap is formed
between the module and the device when it is mounted. Additionally, in some cases,
 Electronics 2023, 12, x FOR PEER REVIEW 7 of 24
Electronics 2023, 12, 2923 7 of 24

crosstalk can
crosstalk can occur
occur due
due toto the
thecover
coverglass,
glass,asasshown
showninin Figure
Figure3b.3b.
As As
a result, unlike
a result, con-
unlike
ventional sensors, the light-receiving part of the optical sensor may
conventional sensors, the light-receiving part of the optical sensor may measure photonsmeasure photons that
are directly coupled to the inside of the sensor or reflected from the skin
that are directly coupled to the inside of the sensor or reflected from the skin surface without surface without
passingthrough
passing throughthe thehuman
humanbody.body.Signals
Signalsthat
thatdo donot
notpass
passthrough
throughbiological
biologicaltissue
tissueareare
considered uninteresting and result in an unnecessary increase in the
considered uninteresting and result in an unnecessary increase in the DC component of the DC component of
the PPG
PPG signal.signal. Manufacturing
Manufacturing a sensing
a sensing unitblocks
unit that that blocks the path
the direct direct paththe
from from
lightthe light
source
source to the light-receiving unit is essential, as an increase in the DC signal
to the light-receiving unit is essential, as an increase in the DC signal can lead to errors in can lead to
errors in measuring oxygen saturation, especially due to the effect of skin
measuring oxygen saturation, especially due to the effect of skin color. The pulse oximeter’s color. The pulse
oximeter’sequation
theoretical theoretical equation
expresses expresses
the the ratio of(AC)
ratio of amplitude amplitude (AC)(DC)
to baseline to baseline
obtained(DC) ob-
from
tained from the photoplethysmographic signal of two wavelengths. This
the photoplethysmographic signal of two wavelengths. This ratio cancels out the difference ratio cancels out
the
in difference
light in light
absorption dueabsorption dueas
to skin color, tothe
skinAC color,
valueas is
the AC value is
normalized tonormalized
the DC value to the
of
DC value
each of each wavelength.
wavelength. Thus, theoretically,
Thus, theoretically, the skin color thedoes
skinnotcolor doesthe
affect notaccuracy
affect theofaccu-
the
racy of saturation
oxygen the oxygenmeasurement.
saturation measurement.
However, ifHowever, if the light-emitting
the light-emitting unit has
unit has a direct patha to
di-
rect path to the light-receiving unit without passing through the body,
the light-receiving unit without passing through the body, the amount of light reflected by the amount of light
reflected
the by the varies
skin surface skin surface varieson
depending depending on theConsequently,
the skin color. skin color. Consequently,
the amount of the amount
crosstalk
of crosstalk
that dependsthat on thedepends on the
skin color skinthe
affects color affects
ratio of ACtheto ratio of AC to
DC, leading toDC, leading
errors to errors
in calculating
in calculating
oxygen saturationoxygen saturation values.
values.

Figure 3. Conceptual diagram of a PPG sensor module with an embedded LED–PD pair. The diagram
Figure 3. Conceptual diagram of a PPG sensor module with an embedded LED–PD pair. The dia-
illustrates two scenarios:
gram illustrates (a) when
two scenarios: (a) the
whenintegrated sensorsensor
the integrated module is in direct
module is in contact with the
direct contact wrist
with the
and (b) when the sensor module is embedded in a device and covered with
wrist and (b) when the sensor module is embedded in a device and covered with a glass surface. a glass surface. In theIn
diagram, the red
the diagram, thesolid line represents
red solid incident
line represents light from
incident lightthe LED
from theentering skin tissue,
LED entering while the
skin tissue, green
while the
solid
greenline indicates
solid light causing
line indicates various
light causing reactions
various insideinside
reactions the tissue. The blue
the tissue. Thesolid line represents
blue solid line repre-
sentsentering
light light entering
the PDthe fromPDskinfromtissue
skin tissue and being
and being detected.
detected. Additionally,
Additionally, thedotted
the red red dotted
line inline
thein
diagram illustrates light reflected inside the sensor module that does not enter the skin, some ofof
the diagram illustrates light reflected inside the sensor module that does not enter the skin, some
whichdirectly
which directlyenters
entersthethe light-receiving
light-receiving unit.unit.
TheTheblueblue dotted
dotted line represents
line represents the that
the light lightisthat is re-
reflected
flected back by the cover glass among the light entering towards the PD.
back by the cover glass among the light entering towards the PD.

Inthe
In theearly
early2010s,
2010s,thethepopularity
popularity ofof wrist-worn
wrist-worn wearables
wearables ledled to the
to the widespread
widespread useuse
of
of sensor modules that included an integrated LED–PD pair and AFE.
sensor modules that included an integrated LED–PD pair and AFE. Figure 4a provides an Figure 4a provides
an expanded
expanded viewview of Samsung
of the the Samsung Galaxy
Galaxy GearGear Fit product,
Fit product, which which was released
was released in 2014.
in 2014. The
The sensor
sensor module module
for thefor the LED–PD
LED–PD optopair
optopair is situated
is situated at the center,
at the center, with a with a case
case and andcover.
glass glass
cover. However,
However, the lacktheoflack of a perfect
a perfect barrierbarrier rib structure
rib structure between
between the LED
the LED and and
PD PD results
results in
in crosstalk.
crosstalk. MoreMore recent
recent wrist-wearable
wrist-wearable devices
devices havehave moved
moved towards
towards usingusing separate
separate sen-
sensors
sors
for for LEDs
LEDs and PDs.and Figure
PDs. Figure
4b shows4b shows the Samsung
the Samsung GalaxyGalaxy
Watch 3, Watch 3, released
released in 2020,in 2020,
which
which has a pulse oximeter function. On the back of the watch, the
has a pulse oximeter function. On the back of the watch, the central multi-wavelength central multi-wave-
length
LED usedLEDto used
measureto measure
the PPGthe PPGissignals
signals is physically
physically separated separated
from thefrom
AFE,the AFE,
with with
several
several surrounding
surrounding PDs. Additionally,
PDs. Additionally, a partitiona partition wall separates
wall separates the LED theandLED
PD. and PD.
Electronics 2023,
Electronics 12, x12,
2023, FOR PEER REVIEW
2923 8 of 824of 24

Figure
Figure 4. Representative
4. Representative examples
examples of PPG
of PPG sensor
sensor systems
systems appliedtotocommercial
applied commercialwrist-worn
wrist-wornwear-
wearable
abledevices:
devices:(a)(a)the Samsung
the Samsung Galaxy
GalaxyGear Fit, Fit,
Gear released in 2014,
released utilizes
in 2014, a sensor
utilizes module
a sensor with anwith
module embedded
an
embedded
LED–PD LED–PD
optopair, optopair,
and (b) and (b) the Samsung
the Samsung Galaxy Galaxy
Watch 3, Watch 3, released
released in 2020,inuses
2020, uses separate
separate LEDs and
LEDsPDsand
as PDs as a sensor.
a sensor. An analogAn front
analog
endfront
(AFE)endis (AFE) is employed
employed in bothtodevices
in both devices measureto the
measure the as
PD output
PD aoutput as a
PPG signal. PPG signal.

To Toobtain
obtaina robust
a robust PPG PPGsignal, various
signal, sensor
various configurations
sensor configurations have been
have attempted.
been attempted.
TheThe
initial approach
initial approach waswasto increase thethe
to increase LED operating
LED operatingcurrent andand
current the the
number
number of LEDs
of LEDs
to enhance
to enhance the the
radiation
radiation power andand
power incident light,
incident as shown
light, as shown in Figure
in Figure5a. 5a.
However,
However, the the
amplitude of the PPG signal may not increase significantly if the active area
amplitude of the PPG signal may not increase significantly if the active area of the PD does of the PD does
notnot
cover thethe
cover radiation
radiation pattern
patternofofthe
thelight
lightsource.
source. Therefore, multiple PDs
Therefore, multiple PDswere
weresometimes
some-
times utilized
utilized to to increase
increase the
the activearea
active areaofofthethePD
PD(Figure
(Figure5b).
5b). With advancements
advancementsin insen-
sensor
sor packaging
packaging technology,
technology, illustrated
illustrated inin Figure 5c, it has become feasible
feasible toto create
createaasensor
sensor of
comparablesize
of comparable size to
to the
the one
one depicted
depicted in Figure 5b usingusing multi-wavelength
multi-wavelengthLEDs, LEDs,instead
insteadof a
of asingle-wavelength
single-wavelength LED.LED.Moreover,
Moreover, as exhibited in Figure
as exhibited 5d, the
in Figure 5d,efficiency can be can
the efficiency improved
be
improved by augmenting the number of PDs. In wearable devices, increasing the number is
by augmenting the number of PDs. In wearable devices, increasing the number of PDs
moreisbeneficial
of PDs than increasing
more beneficial the number
than increasing of LEDs,of
the number asLEDs,
it reduces
as itpower
reducesconsumption
power con-and
heat generation.
sumption and heatUpon analyzing
generation. Uponcommercial
analyzing smartwatches
commercialthat utilize such that
smartwatches designs, it was
utilize
suchfound that the
designs, it wasGear S3 uses
found thattwo
thegreen
GearLEDsS3 usesandtwo
onegreen
photodiode
LEDs and (Figure
one5e), whereas the
photodiode
Galaxy
(Figure 5e),Watch
whereas incorporates
the Galaxyone green
Watch LED and four
incorporates onephotodiodes
green LED and (Figure
four5f). Additionally,
photodiodes
the Garmin Fenix 5 combines three green LEDs with a larger photodiode
(Figure 5f). Additionally, the Garmin Fenix 5 combines three green LEDs with a larger (Figure 5g), and
the Apple Watch employs a pair of green LEDs with two photodiodes
photodiode (Figure 5g), and the Apple Watch employs a pair of green LEDs with two (Figure 5h).
photodiodes (Figure 5h).
3.2. Low-Power-Consumption System
Low power consumption is a critical aspect of wearable technology, as these devices
are designed to be worn continuously and must have a long battery life that can last for
days or even weeks on a single charge. Additionally, wearable devices must be comfortable
to wear since they are meant to be worn on the body for extended periods. To date, most
of the attention has focused on low power consumption in hardware components for
the development of low-power systems. An example of such a microcontroller is the
NXP Semiconductor’s Cortex M0-Based Kinetis KL03 microcontroller, which has ultra-low
power consumption and supports low-power wireless communication protocols such as
Bluetooth Low Energy (BLE) [35]. Researchers are also exploring new battery technologies,
such as solid-state batteries, which offer higher energy density and a longer lifespan than
traditional lithium-ion batteries [36]. In addition to conventional components such as
microprocessors and displays that are used in wearable devices, PPG signal measurement
 Electronics 2023, 12, 2923 9 of 24

chips have also been developed as low-power systems and are already commercially
available [37–39]. These chips exhibit low power consumption in standard operating mode
and even lower power consumption in power save mode. However, it is important to
note that the power consumption of PPG AFE chips can vary significantly depending on
the specific implementation. For instance, incorporating additional sensors or processing
Electronics 2023, 12, x FOR PEER REVIEW algorithms may result in increased power consumption. As a result, research9 efforts
of 24 are
being directed towards minimizing the power consumption through the use of various
embedded software methods.

Figure 5. Examples of PPG sensor configurations that use multiple LEDs and PDs, and their applica-
Figure 5. Examples of PPG sensor configurations that use multiple LEDs and PDs, and their appli-
tion to commercial smartwatches. (a–d) illustrate different PPG sensor structures, while (e) through
cation to commercial smartwatches. (a–d) illustrate different PPG sensor structures, while (e)
(h) present examples of these structures in use: (e) structure (a) applied to the Samsung Galaxy Gear
through (h) present examples of these structures in use: (e) structure (a) applied to the Samsung
Galaxys3; (f) structure
Gear (d) applied
s3; (f) structure to theto
(d) applied Samsung Galaxy
the Samsung Watch;
Galaxy (g) shows
Watch; the sensor
(g) shows structure
the sensor struc- of the
ture of the Garmin Fenix 5 with three LEDs and one PD in the center, and (h) shows the sensor of the
Garmin Fenix 5 with three LEDs and one PD in the center, and (h) shows the sensor structure
Apple
structure Watch
of the Applewith two LEDs
Watch andLEDs
with two two PDs.
and two PDs.

There are various System

3.2. Low-Power-Consumption approaches to extending the battery life of wearable systems other
than developing low-power hardware components. Power management software is a
Low power consumption is a critical aspect of wearable technology, as these devices
representative example [40–42] that can be used to optimize the power consumption of a
are designed to be worn continuously and must have a long battery life that can last for
system by controlling the power states of various components, such as the microcontroller,
days or even weeks
sensors, on a single charge.
and communication Additionally,
modules. By usingwearable devices must
power management be comforta-
software, the system
ble tocan
wear since they are meant to be worn on the body for
dynamically adjust the power consumption of these components based onextended periods. Tothedate,
system’s
most operating
of the attention has focused on low power consumption in hardware
conditions and power requirements, which can help to conserve power. For components
for the development
the microcontroller,of low-power systems. An
there are low-power example
modes, whichof such
can bea microcontroller
activated through is the
software
NXP to Semiconductor’s Cortex M0-Based Kinetis KL03 microcontroller,
reduce power consumption [43,44]. These low-power modes include sleep mode, idle which has ultra-
low power
mode,consumption
and standbyand mode,supports
amonglow-power
others. Usingwireless
thesecommunication
low-power modes, protocols such can
the system
as Bluetooth Low Energy (BLE) [35]. Researchers are also exploring
reduce the power consumption of components when they are not actively performing new battery technol-
ogies,tasks,
such as solid-state
thus conserving batteries,
power.which
Duty offer higher
cycling energy density
is a technique and a longer
that involves turninglifespan
on and off
than traditional lithium-ion
system components in abatteries
controlled[36]. In addition
manner to reduceto conventional components
the average power such For
consumption.
as microprocessors and displays
example, in a wireless sensorthat are used
network, theinsensor
wearable
nodedevices,
may turn PPG
on signal
the radio measure-
periodically
ment tochips have also been developed as low-power systems and are already
transmit data, and then turn it off again to conserve power. With duty cycling, the commercially
available
system[37–39]. These the
can reduce chips exhibit
power low powerofconsumption
consumption the components in standard operating their
without sacrificing
modefunctionality.
and even lower power consumption in power save mode. However, it
For wearable systems, adaptive algorithms can be used to optimize the power is important
to note that the power
consumption consumption
of the of PPG AFE
system by adapting chips cansystem
to changing vary significantly
conditions and depending
adjusting the
on thesystem’s
specificbehavior
implementation.
accordinglyFor [45,46].
instance,Forincorporating
example, adaptiveadditional sensorscan
algorithms or be
pro-
used to
cessing algorithms
adjust may result
the sampling rate in increased
based on thepower
user’sconsumption.
activity level,As a result, research
conserving power when ef- the
forts are
userbeing
is notdirected
moving.towards
In recent minimizing the power systems,
wearable healthcare consumption through
it is not the useto
appropriate ofapply
various embedded software methods.
an adaptive algorithm to the PPG sensor because it provides the function of continuously
There are various
measuring the heartapproaches
rate for 24tohextending the battery
without being aware life of wearable
of the measurement.systems other the
Therefore,
than developing low-power
trend is to lower hardware
the sample components.
rate of PPG signalPower management software is a rep-
measurement.
resentative example [40–42] that can be used to optimize the power consumption of a sys-
tem by controlling the power states of various components, such as the microcontroller,
sensors, and communication modules. By using power management software, the system
can dynamically adjust the power consumption of these components based on the sys-
tem’s operating conditions and power requirements, which can help to conserve power.
Electronics 2023, 12, 2923 10 of 24

Choi and Shin conducted an investigation in which they generated down-sampled

photoplethysmograms (PPGs) at frequencies of 5000, 2500, 1000, 500, 250, 100, 50, 25, 20, 15,
10, and 5 Hz from PPGs sampled at 10 kHz [47]. The researchers then extracted time- and
frequency-domain variables of pulse rate variability (PRV) from the down-sampled PPGs
and statistically compared these variables with the heart rate variability (HRV) derived
from simultaneously recorded 10-kHz-sampled electrocardiograms. The results indicated
that there was no significant difference for any PRV variable when the sampling rate
was greater than 25 Hz, without any interpolation. In another study, Béres and Hejjel
investigated the minimum sampling rate required for PRV analysis [48]. The researchers
abbreviated the PPG signal sampled at 1 kHz with a range of sampling rates (2, 5, 10,
20, 50, 100, 200, and 500 Hz) and derived PRV variables using cubic spline and parabola
approximation methods after interpolation. These variables were then compared with the
PRV variables obtained from the original 1-kHz signal. The results showed that a sampling
rate of 5 Hz was sufficient to calculate the pulse rate without applying the interpolation
method, but a sampling rate of at least 50 Hz or higher was required for variables in
the time and frequency domains of the PRV without applying the interpolation method.
However, when applying any of the interpolation methods, the minimum sampling rate
for PRV increased to 10–20 Hz. Peláez-Coca et al. conducted a similar investigation by
obtaining PRV variables from down-sampled PPG signals (125, 100, 50, and 25 Hz) created
from 1-kHz-sampled PPG signals, and they compared them with HRV variables obtained
from simultaneously recorded electrocardiograms [49]. The researchers used the medium
interpolate point obtained by linear interpolation as a fiducial point to calculate the inter-
beat interval of the PPG signal. Results indicated that a sampling frequency of 50 Hz could
be used without causing a significant change in the time and frequency variables of PRV.
A low sampling rate can reduce the amount of data that needs to be collected and
processed by the microcontroller, which can in turn reduce the power consumption of the
system. By lowering the sampling rate, the system reduces the amount of data transmitted
wirelessly and stored in memory. A lower sampling rate can also reduce the processing
requirements of the microcontroller. By lowering the sampling rate, the system can reduce
the number of calculations and data processing tasks that need to be performed, which
helps to reduce the power consumption of the system. However, it is important to note that
a lower sampling rate may also result in reduced signal quality, which could impact the
accuracy of the biosignal monitoring system. Therefore, when selecting a sampling rate for
a wearable biosignal monitoring system, it is important to achieve a balance between power
consumption and signal quality to ensure that the system meets the desired performance
requirements. To enhance the accuracy of inter-bit interval (IBI) extraction in the context of
low-sample-rate signals, it is necessary to employ interpolation to improve the reference
point of the signal.

3.3. PPG Signal Applications

The most common application of the PPG signal measured from the wrist is for con-
tinuous heart rate measurement. Due to its non-invasive nature, continuous measurement
capability, and portability, the PPG signal can be used to measure the heart rate contin-
uously for up to 24 h while the device is worn on the wrist. In fact, many commercial
wrist-worn wearable devices already include continuous heart rate measurement as a basic
function. In a study conducted by Fuller et al. in 2020, the heart rate measurement accuracy
of commercial wearable devices was investigated by analyzing data from 158 publica-
tions [50]. The study found that in a controlled environment, wearable devices from Apple,
Fitbit, and Garmin reported heart rate measurement errors within ±3%. Furthermore, the
study confirmed that these devices also measured continuous heart rates within ±10%
error, even in free-living environments. It should be noted that the papers analyzed in this
study were published before May 2019, and it is expected that the continuous heart rate
measurement accuracy of the latest wearable devices has improved even further.
Electronics 2023, 12, 2923 11 of 24

Pulse rate variability (PRV) measurement is a commonly used parameter in addition

to pulse rate measurement [51,52]. PRV is considered a valid substitute for heart rate
variability, which is typically obtained through electrocardiogram readings. Similar to
heart rate variability, PRV is a non-invasive marker derived from PPG signals, reflecting
the activity of the sympathetic and parasympathetic branches of the autonomic nervous
system at the sinus node of the heart. Unlike pulse rate measurement, PRV provides
additional information on the regulation of the heart rate (HR) by the ANS, and due to
its clinical significance, PRV is widely used to efficiently monitor the human body and
evaluate the ANS through its sympathetic and parasympathetic components. In recent
years, algorithm research has been actively conducted to derive health parameters related
to the autonomic nervous system using various PRV parameters. The most representative
application example is the development of an algorithm to quantify stress states.
Zubair and Yoon conducted a study to explore the effectiveness of a multilevel stress
detection system in identifying different states of mental stress using ultra-short-term
recordings from PPG sensors. The study involved 14 graduate students who underwent
acute stress-inducing protocols, including the Mental Arithmetic Task (MAT) and Stroop
Color Word Tasks [53]. PPG signals were collected and analyzed to extract various features
related to the beat-to-beat intervals derived from 60-s segments of the PPG signal. These
features encompassed time-domain, frequency-domain, and nonlinear characteristics of
the pulse rate variability (PRV) based on reliability. Additionally, a novel indicator called
the Point Transition Measure (PTM) was introduced by the authors to quantify the time
variation at the point-to-point level of the Poincare plot, which was utilized as a feature in
the analysis. Feature selection was then performed using the Sequential Forward Floating
Selection (SFFS) algorithm to eliminate features with low class correlation and redundant
features with high intercorrelation. The classification model employed was the support
vector machine (SVM), which achieved accuracy of 94.33% in identifying five different
levels of mental stress.
Pugliese et al. developed a rule-based algorithm that detects the onset of drowsiness
by analyzing the change in LF/HF, a frequency-domain parameter of PRV, over time using
PPG [54]. The algorithm’s performance was assessed in two different scenarios: a sleep
testing setting and a driving simulator environment. In both settings, the algorithm’s
determination of the onset of drowsiness was compared with the drowsiness onset point
obtained through sleep behavior analysis performed by a sleep specialist using data from
a polysomnography device. The results demonstrated that the algorithm detected the
onset of drowsiness, on average, approximately 9 min before the onset of actual sleep.
Wulterkens et al. developed and evaluated a long short-term memory (LSTM) recurrent
neural network model for sleep stage classification [55]. The model used PRV parameters
from PPG signals and body movements from acceleration signals as inputs to a classifier
consisting of an input dense layer, three bidirectional LSTM layers, and two dense layers.
The last output layer generated posterior probabilities for each of the four sleep stages
(wake, N1 + N2, N3, and REM), and the final classification was determined by the class with
the highest posterior probability for each epoch. The algorithm was validated against refer-
ence polysomnography in a clinical population of 244 adults and 48 children/adolescents
with various sleep disorders. The LSTM classifier demonstrated substantial agreement
for the four sleep stages, with an average Cohen kappa of 0.62 and accuracy of 76.4%.
For children/adolescents, the agreement was higher, with an average kappa of 0.66 and
accuracy of 77.9%.
Active research is ongoing to develop methods for non-invasive blood pressure mea-
surement using photoplethysmography (PPG) signals acquired from the wrist without
pressurization [56,57]. Various approaches have been investigated in previous studies, with
the most commonly used methods being the pulse transit time (PTT), pulse arrival time
(PAT), pulse wave velocity (PWV), and pulse wave analysis (PWA). These parameters have
a correlation with blood pressure and can be converted into BP values using appropriate
algorithms. However, it is difficult to apply these parameters to a wrist-worn wearable
Electronics 2023, 12, 2923 12 of 24

device because two devices are required to simultaneously measure the parameters at
a certain distance apart. Therefore, the PWA method has emerged as the most preva-
lent approach for cuffless blood pressure measurement using PPG signals acquired from
the wrist. This technique involves extracting various features from the PPG signal and
training machine learning models, such as linear regression, support vector machines,
and neural networks, to estimate blood pressure values. Yao et al. recently presented a
cuffless blood pressure measurement method using the PWA approach [58]. They extracted
12 time-domain features, 53 morphological features, and 3 statistical features from the
entire data set, as well as five demographic features (age, height, weight, body mass index,
and gender). These features were fused to create six composite features, which were then
fed into a two-layer feedforward artificial neural network algorithm for regression. The
proposed method passed the European Society of Hypertension International Protocol
Revision 2010 (ESP-IP2) and demonstrated good accuracy across various populations, with
a mean absolute error of 3.23 ± 4.47 mmHg for SBP and 2.73 ± 3.61 mmHg for DBP. The
reflectance-type sensor used to measure the PPG signal on the wrist typically exhibits
a lower signal-to-noise ratio (SNR) compared to the transmittance-type sensor used on
the finger. Consequently, studies focusing on PWA have often encountered limitations
in performance when utilizing a reflective sensor. Shi et al. proposed a hybrid mathe-
matical model for continuous blood pressure monitoring, specifically investigating the
relationship between transmissive PPG signals obtained from the finger and radial arterial
blood pressure signals [59]. To account for potential attenuation factors and wave propaga-
tion/reflection within the blood circulation, they combined an electrical network model
with a tube load model. By employing a system identification method, they determined
the optimal parameter range for the model, thereby enabling personalized continuous
blood pressure measurement. Experimental results demonstrated that the reconstructed
arterial blood pressure (ABP) waveform, achieved through the hybrid model, exhibited a
strong correlation (average correlation coefficient of 0.96) with the reference waveform. The
mean absolute errors and standard deviations for the estimated SBP and DBP were 3.0/4.4,
2.1/3.0, and 2.1/3.2 mmHg, respectively. These findings confirm that the performance of
the hybrid model proposed by Shi et al. surpasses the results obtained by Yao et al. [58].
Diabetes status has been shown to impact hemodynamics and HRV, leading to grow-
ing interest in the use of spectrum analysis of photoplethysmogram (PPG) signals for
non-invasive blood glucose level prediction [60–62]. Among the methods employed for
PPG signal analysis, the Mel frequency cepstral coefficient (MFCC) function, which is
widely used in speech signal processing and was inspired by human auditory perception,
holds significant relevance [63]. Prabha et al. introduced a novel Mel frequency cepstral
coefficient feature derived from 5-s wristband photoplethysmogram signals, incorporating
physiological parameters such as age, weight, and height. They proposed a non-invasive
blood glucose estimation system utilizing extreme gradient boost regression (XGBR) [64].
To validate their model, they utilized a dataset comprising 217 participants from the hos-
pital, reporting a standard prediction error (SEP) of 9.78 mg/dL. The accuracy of the
predicted glucose values was assessed through standard Clarke error grid analysis and
Bland–Altman analysis. These evaluations revealed that the estimated blood glucose levels
fell within the clinically acceptable region, indicating the clinical accuracy of the majority
of the estimated values.
Boukhechba et al. proposed a novel method for the extraction of meaningful features
from PPG signals to predict human activity [65]. While most technologies that use PPG
signals attempt to remove motion artifacts caused by human activity, their approach
leverages the sensitivity of the PPG signal to motion-induced damage to predict the type
of activity performed by the user. To predict daily activities, such as walking, running,
and jumping, they deconstructed the PPG signal into three components, namely cardiac,
respiratory, and motion artifact signals, and applied convolutional and iterative layers in
an end-to-end deep neural network. Although their experimental results did not reach
the performance level of activity recognition using inertial measurement units, such as
Electronics 2023, 12, 2923 13 of 24

accelerometers and gyroscopes, their approach detected five types of activity (standing,
walking, jogging, jumping, and sitting) at a reasonable level using only the PPG signal. In
summary, the approach proposed by [65] has significance in expanding the application of
PPG signals measured using wearable devices. Their method could potentially be used
to recognize a wider range of activities with improved accuracy, making it an important
contribution to the field of activity recognition.

4. Current Challenges
4.1. Sensor: LED Wavelength
The performance of PPG sensing is influenced by various factors, and one of the most
significant factors is the choice of wavelength for skin illumination. The skin comprises
water, hemoglobin, and melanin, which have distinctive light-absorbing characteristics
based on the wavelength and chemical bonding [66–69]. Water, a primary constituent of the
skin, exhibits high absorption of infrared (IR) light but shows minimal absorption of visible
light. Melanin, which is closely related to skin color, demonstrates increased absorption
as the wavelength decreases. These elements contribute to the DC component of the PPG
signal since their amounts are not easily altered within short time intervals. Similarly, the
absorption properties of hemoglobin decrease with increasing wavelength. However, when
bound with oxygen, hemoglobin exhibits a significant reduction in absorption within the
range of red (570–700 nm), and it becomes more absorbent at longer wavelengths such
as IR compared to non-oxygenated hemoglobin [11,70]. Unlike water or melanin, the
amount of hemoglobin varies with changes in blood flow, corresponding to the heartbeat
at the measurement site. Therefore, hemoglobin’s light absorption contributes to the AC
component, representing the pulsating element of the PPG signal. Scattering phenomena
can occur through two mechanisms: surface effects such as reflection and refraction, or
interactions with compounds in the skin that possess unique optical properties. Research
suggests that around 4–7% of incident light is reflected from the skin surface, irrespective
of the wavelength [69]. Generally, within the skin, the scattering coefficients demonstrate
a decreasing pattern as the wavelength of light increases. During reflectance-mode PPG
sensing, the path followed by photons within the skin is hypothesized to resemble an
arc shape, as illustrated in Figure 1b. This path has a maximum depth referred to as the
penetration depth, which relies on the absorption and scattering coefficients specific to the
skin. As described in the working principle, the penetration depth tends to increase as
the wavelength of light falls within the visible and IR range. It is important to account for
variations in the penetration depth across different measurement sites when considering
reflectance-mode PPG sensing.
Historically, in clinical environments, PPG signals with IR and red wavelengths were
predominantly used to measure oxygen saturation. This choice was based on the significant
difference in light absorption between oxyhemoglobin (HbO2 ) and deoxyhemoglobin (Hb)
at these two wavelengths [22,71,72]. For this reason, early PPG systems primarily utilized
light sources emitting IR or red wavelengths [73–76]. On the other hand, wrist-worn
wearable devices mainly use green light (492–577 nm) for heart rate measurement [35,77].
Hemoglobin is the primary molecule that absorbs light in the blood. Green light (peak
wavelength around 520 nm) is well absorbed by hemoglobin and the absorption is relatively
constant over a wide range of oxygen saturation levels [36,37,78–80]. This characteristic
makes green light a suitable choice for wearable PPG measurements, as it allows for accurate
and consistent monitoring of changes in blood volume. In addition, green light also has an
advantage regarding motion artifacts. Motion artifacts can be a major challenge in PPG
measurements, especially in wrist-worn devices, where there can be significant movement.
Several studies have presented evidence indicating that shorter wavelengths can enhance
the signal-to-noise ratio (SNR) by reducing the susceptibility to motion-induced noise.
This effect can be attributed to the relatively shorter path length and depth of penetration
associated with shorter wavelengths [79,81]. Green light, which has shallower tissue
penetration compared to wavelengths such as red or infrared [38,82], is less susceptible to
Electronics 2023, 12, 2923 14 of 24

motion artifacts. This characteristic makes it a practical choice for wrist-worn wearables.
However, even when utilizing green light with a shorter wavelength, motion-induced noise
contamination can still affect the acquired PPG signal in wearable systems. To mitigate
such artifacts, a commonly employed approach involves the application of adaptive filter
technologies, utilizing the acceleration signal as a motion reference [83–85]. Nevertheless,
recent research has introduced an alternative technique that employs PPG signals within the
infrared wavelength range as a motion reference for adaptive filtering [86,87]. This multi-
wavelength approach takes advantage of the inherent vulnerability of long-wavelength
PPG to motion-induced noise, as previously mentioned.
Shorter wavelengths, such as green light, have limited penetration depths, which
restrict the collection of physiological information from deeper tissue layers. To address
this limitation, there has been an increasing trend towards utilizing light sources of multiple
wavelengths for photoplethysmography (PPG) measurements [88–90]. This approach
capitalizes on the distinct interactions between light and tissue to enhance the physiological
insights derived from PPG signals. Utilizing multiple wavelengths of light enables the
acquisition of more comprehensive information about the cardiovascular system and other
physiological processes. Several studies have conducted comprehensive investigations
into the potential benefits of employing multi-wavelength PPG to estimate a wide range of
health parameters. The specific advantages resulting from these studies will be thoroughly
reviewed in Section 4.3.

4.2. Low-Power-Consumption System: Parabola Approximation

As mentioned in Section 3.2, the strategy of reducing the sampling rate of the PPG
signal is being implemented to minimize power consumption. However, when working
with PPG signals at a low sampling rate, an appropriate interpolation method becomes
crucial in order to accurately extract the inter-bit interval required for PRV analysis. In one
study [91], a simple and minimally computational interpolation method using parabolic
approximation was applied to a PPG signal with a low sampling rate of 20 Hz. This method
uses a generic parabola to estimate the true peak position and height by generating a curve
fit through the three highest samples of the peak. The general vertex form of a quadratic
function can be expressed as
f ( x ) = a ( x − p )2 + b
To explain the parabolic interpolation approach, let us assume that the parabola has its
three highest points at (−1, α), (0, β), and (1, γ). In this case, the interpolated peak position
p and the peak amplitude b can be determined using the quadratic function.
Specifically, the three highest sample points are substituted into the general vertex
form of a quadratic function, resulting in the following expression:

α = ap2 + 2ap + a + b, β = ap2 + b, γ = ap2 − 2ap + a + b

Hence, we can derive that α − γ = 4ap, which implies that p = (α − γ)/4a. From the
expression for β described secondly, the expression b = β − ap2 can also be obtained.
By substituting the expressions for p and b into the first expression for α, we obtain
α = ap2 + α− γ 2
2 + a + β − ap , which implies that a = 0.5( α − 2β + γ ). Then, substituting
the expressions for a into the equation of p yields the following expression for interpolated
peak location:
0.5(α − γ)
p=
α − 2β + γ
Finally, the location of the peak on the x-axis can be determined by adding p to the
x-axis value of the central sample that has the highest amplitude.
The issue of peak detection in PPG signals at low sampling rates is illustrated in
Figure 6. The background graph in gray represents the original 250-Hz PPG signal, while
the round marker graph depicts the down-sampled signal at 30 Hz, 20 Hz, and 10 Hz.
Lowering the sampling rate results in the detection of an incorrect peak position, leading
 Electronics 2023, 12, 2923 15 of 24

to a substantial error in inter-beat interval measurement for PRV extraction. To mitigate

this effect, parabolic approximation with the three largest samples per pulse of each low-
sampling-rate signal can be used, as indicated by the dotted lines in Figure 6b,d,f. The peak
of the parabola significantly reduces the error in peak detection, as evident in the figure,
when compared to the peak obtained from the original 250-Hz PPG signal. Based on the
assumption that the PPG signal is maximally parabolic, parabolic approximation using
the three highest points provides an efficient and accurate method of identifying the true
peak or very close peaks in a single step. Unlike traditional interpolation methods, such as
cubic spline interpolation, this method only requires the location of the maximum point
of the parabola and does not rely on a fully formed quadratic function. As a result, the
Electronics 2023, 12, x FOR PEER REVIEW 16 of 24
computation is much simpler, particularly for low-power wearable applications. Therefore,
the parabolic approximation remains the recommended interpolation method for the
evaluation of trends in HRV measurements.

Examples
Figure6.6.Examples
Figure ofofthe
thelow
lowsampling
samplingrate rateproblem
problemininpulse
pulserate
ratevariability
variabilityanalysis
analysisusing
usingPPG
PPG
signals.The
signals. The gray
gray background
background line
line represents
represents the
the PPG
PPG signal
signal atat 250
250 Hz,while
Hz, while thecircular
the circularmarker
marker
line
line representsthe
represents thePPG
PPGsignal
signalatata alower
lowersampling
samplingrate.
rate.The
The dotted
dotted line
line showsthe
shows the parabolicapprox-
parabolic approxi-
imation,
mation,which
whichwas
was estimated
estimated using the three three largest
largestsamples
samplesinineach
eachpulse
pulseofofthe
the low-sampling-
low-sampling-rate
signal.
rate Specifically,
signal. thethe
Specifically, 250-Hz
250-Hz PPG PPG signal was
signal down-sampled
was down-sampled to to
(a)(a)
30 30
Hz, (c)(c)
Hz, 2020Hz, and
Hz, and(e)10 Hz,
(e)10
respectively. (b,d,f) are enlarged view of the second peak area of
Hz, respectively. (b,d,f) are enlarged view of the second peak area of (a,c,e). (a,c,e).

4.3.Multi-Wavelength
4.3. Multi-Wavelength PPG
PPG Signal
Signal Applications
Applications
Multi-wavelength PPG sensing technology
Multi-wavelength PPG sensing technology is iswidely
widelyrecognized
recognizedasasa arobust
robustmethod
method
for the measurement of PPG signals [92,93]. Depending on an individual’s skin color,
for the measurement of PPG signals [92,93]. Depending on an individual’s skin color,
PPG-sensing light sources with various wavelengths can be selected to acquire the most
PPG-sensing light sources with various wavelengths can be selected to acquire the most
robust PPG signal [94]. Furthermore, studies have indicated that the use of an optimal
robust PPG signal [94]. Furthermore, studies have indicated that the use of an optimal
wavelength can enhance the signal-to-noise ratio (SNR), even in scenarios where reduced
wavelength can enhance the signal-to-noise ratio (SNR), even in scenarios where reduced
blood perfusion at low temperatures leads to a decrease in the SNR [95,96]. Blood pressure
blood perfusion at low temperatures leads to a decrease in the SNR [95,96]. Blood pressure
measurement is a notable application field for multi-wavelength PPG. As discussed in
Section 4.1, multi-wavelength PPG signals provide blood pulsation information from var-
ious blood vessels located at different depths within the skin. Therefore, the time differ-
ence observed between PPG signals of different wavelengths acquired from the same body
Electronics 2023, 12, 2923 16 of 24

measurement is a notable application field for multi-wavelength PPG. As discussed in Sec-

tion 4.1, multi-wavelength PPG signals provide blood pulsation information from various
blood vessels located at different depths within the skin. Therefore, the time difference
observed between PPG signals of different wavelengths acquired from the same body site
can be considered as a localized pulse transit time (PTT), reflecting the characteristics of a
specific segment of blood vessels beneath the skin. The PTT can serve as an indicator of sys-
temic vascular resistance (SVR). As SVR directly influences the mean blood pressure (MBP),
Liu et al. propose a PTT-based approach for continuous and cuffless BP measurement, and
this method utilizes a single-site multi-wavelength PPG (blue (470 nm), green (570 nm),
yellow (591 nm), and red (635 nm)) [97]. In their study, the artery is conceptualized as
a bilateral elastic tube filled with blood. The MBP of this elastic tube depends on two
factors: the cardiac output, which represents the flow rate at the inlet, and the SVR, which
represents the resistance to blood flow at the outlet of the elastic tube. To establish their
approach, the researchers employed physiological models that considered the relationship
between the blood pressure, heart rate, stroke volume, and SVR. They also incorporated the
electrical representation of the two-element arterial Windkessel model. The experimental
findings demonstrated a strong correlation coefficient (r = 0.91) and a mean absolute differ-
ence (MAD) of 2.72 mm Hg between the predicted MBP and the invasively measured MBP.
They also presented an enhanced algorithm to estimate blood pressure by deciphering the
combined multi-wavelength PPG signals, which include diverse physiological informa-
tion and motion artifacts. The algorithm utilizes principal component analysis (PCA) to
extract the motion artifact component from the multi-wavelength PPG signal, allowing
for the identification and exclusion of segments affected by motion. The results demon-
strate the effectiveness of the PCA-based approach in improving the accuracy of blood
pressure estimation, yielding errors of 1.44 ± 6.89 mm Hg for systolic blood pressure and
−1.00 ± 6.71 mm Hg for diastolic blood pressure [98]. Multi-wavelength PPG is widely em-
ployed for the quantitative assessment of various blood parameters. In a study conducted
by Hossain et al., PPG signals at three specific wavelengths (465 nm, 525 nm, and 615 nm)
were utilized to estimate the levels of glycated hemoglobin in vivo [99]. The researchers
proposed that blood consists of oxygenated hemoglobin (HbO), deoxyhemoglobin (HHb),
and glycated hemoglobin (HbA1c). They further hypothesized that the HbA1c component
remains consistent in the blood, comprising a mixture of 98% oxygenated HbA1c and 2%
deoxygenated HbA1c. By applying the Beer–Lambert law to the PPG signal, they were
able to estimate the quantity of glycated hemoglobin present in the blood. The study
findings indicated that the estimation errors for HbA1c were not influenced by the oxygen
saturation levels (SpO2) and were found to be within the magnitude of 10−14 %. Rachim
et al. conducted a study to estimate blood glucose levels through the measurement of multi-
wavelength PPG signals obtained from the wrist. The wavelengths used for measurement
were 950, 850, 660, and 530 nm. The researchers employed an ensemble averaging algorithm
based on the PPG peak to obtain the average signal. Subsequently, they extracted a total
of 24 features from the average signal: three features were obtained from the difference in
optical density between the amplitudes of the pulsatile and non-pulsatile components, and
three features were obtained from the Teager–Kaiser energy operator for each wavelength.
The extracted features were then used in combination with the partial least squares (PLS)
algorithm to establish a relationship between changes in feature data and variations in
baseline glucose concentrations obtained from blood samples of the subjects. For the ex-
perimental validation, 12 volunteers consumed a modified carbohydrate-rich meal for 2 h,
and the estimated blood glucose concentrations were compared with the baseline values.
The results demonstrated a mean correlation coefficient (R) of 0.86 between the estimated
and baseline blood glucose concentrations, with a standard error of prediction (SPE) of
6.16 mg/dL [100].
Electronics 2023, 12, 2923 17 of 24

5. Future Directions
Since the previous literature review on wearable photoplethysmography (PPG) signals
published in 2014 in the journal Electronics [16], wearable PPG technology has been rapidly
developed. The principle of signal measurement is widely recognized and it has become
a commonly used sensor technology in the latest wearable devices, such as wristbands
and smartwatches. However, acquiring high-quality PPG signals during daily life remains
challenging due to external light interference, which is often caused by subjects not wearing
the wrist device sufficiently tightly. Motion noise caused by hand movements is also a
factor. Most studies have not evaluated the robustness of PPG signals and the accuracy
of health and wellness index monitoring in a 24-h free-living situation. While continuous
heart rate measurements have been reported to be accurate in a free-living environment,
the same cannot be said for other PPG-derived health and wellness parameters [101,102].
Considering the unavoidable presence of motion noise inherent in such devices, it is
imperative to explore and develop new parameters that can be extracted from wearable
PPG systems. Despite the utilization of advanced signal processing techniques, accurately
restoring the phase and amplitude of a signal contaminated by motion artifacts remains
challenging. Consequently, the implementation of many of the application fields listed
in Table 2 for continuous 24-h monitoring becomes difficult. For instance, cuffless blood
pressure measurement technology based on the pulse wave analysis (PWA) method still
faces technical limitations in operating continuously for 24-h measurements. Similarly,
techniques such as stress measurement using heart rate variability (HRV) may result in
phase changes in the PPG waveform due to motion artifacts, making it impossible to
accurately measure the inter-beat interval and leading to errors in deriving HRV parame-
ters. In addressing this issue, one study [103] introduced a novel index that incorporated
heart rate variability (HRV) information. The algorithm employed an adaptive notch-
filter-based frequency tracking algorithm, optimized to better reflect the instantaneous
heart rate variability by adjusting the filter’s bandwidth and the forgetting factor of the
adaptation mechanism. The standard deviation of the extracted heart rate was introduced
as a novel HRV index calculated from wearable PPG signals. The study reported that
the proposed method enabled the monitoring of statistically significant changes in HRV
during different sleep stages. Furthermore, when subjects were exposed to stress during
a socially evaluated speaking task, the proposed method also demonstrated a significant
decrease in HRV. These findings highlight the potential of the proposed method to monitor
health and wellness-related indicators, even in the presence of motion noise. To achieve the
continuous monitoring of the health and wellness-related indicators presented in Table 2
using PPG signals obtained during daily activities, it is crucial to conduct research focused
on developing novel analysis parameters that move beyond traditional methods. Such en-
deavors will contribute to enhancing the accuracy and reliability of wearable PPG systems
in real-world scenarios.
Electronics 2023, 12, 2923 18 of 24

Table 2. Summary of recent PPG signal applications for mobile healthcare.

Ref. Application Subjects Data Input Algorithm Results

Selected features among
PPG signal collected using pulse
time-domain, (5 levels of stress)
sensor amped on the index finger
74 participants frequency-domain, nonlinear Support vector machine Accuracy: 94.33%
[53] Stress of the left-hand under mental
(72 M, 2 F) and point transition measure (SVM) model Sensitivity: 86.01%
arithmetic tasks and Stroop
(PTM) parameters of PRV Specificity: 96.46%
experimental design.
extracted from PPG.
(At home) EEG, EOG, EMG, ECG,
(At home) respiration and PPG signals (Drowsiness prediction)
17 participants recorded continuously for 12 h, LF/HF ratio extracted PPG 6 m 39 s prior to the actual
Sleep (7 M, 10 F) including nighttime sleep, using signal acquired with sampling Rule-based heuristic onset of sleep for at-home
[54]
(Drowsiness) Mean age: 44.3 polysomnographic device. rate of 50 Hz for the duration algorithm data, and 9 m 9 s prior to the
(For driving simulator) (For driving simulator) of one 2048 sample window. actual onset of sleep for
9 participants Polysomnographic device during driving simulator data.
pre-defined driving mission.
PPG and accelerometry signal
PPG-derived PRV features
292 participants measured from non-dominant
combined with a measure of Long short-term (4 stages of sleep)
Sleep (244 adults and 48 wrist during sleep (SOMNIA
[55] gross body movement for each memory (LSTM) Accuracy: 76.4 ± 7.3
(Sleep Stage) children) dataset).
30-s epoch based on recurrent neural network Kappa: 0.62 ± 0.12
Mean age: 42.3 ± 19.7 Reference sleep measurement:
accelerometer signal.
clinical polysomnography.
PPG signal collected using
wrist-wearable device of their Feature sets derived by feature
own design under the protocol in fusion with time-domain, (SBP/DBP)
33 participants A two-layer
IEEE Standard for Wearable morphological, statistical, and ME = −0.07/0.00 mmHg
[58] Blood Pressure (21 M, 12 F) feed-forward artificial
Cuffless BP. demographic features (gender, MAE = 3.23/2.73 mmHg
Mean age: 41.1 ± 16.8 neural network
Reference blood pressure age, height, weight, and body SDE = 4.47/3.61 mmHg
measurements: cuff electronic mass index (BMI)).
sphygmomanometer.
PPG signal collected using
Empatica E4 wristband. MFCC features of wristband R2 = 0.995
217 participants
Blood Glucose Features extracted from 5-s PPG signal and physiological Extreme gradient boost MAE = 1.76 mg/dL
[63] (90 M, 127 F)
Level segments. parameters (age, weight, and regression (XGBR) SEP = 5.53 mg/dL
Mean age: 49 ± 11
Reference blood glucose level height) (total of 5 features) Accuracy = 98.24%
measurements: laboratory test.
Electronics 2023, 12, 2923 19 of 24

Table 2. Cont.

Ref. Application Subjects Data Input Algorithm Results

PPG signal collected using
Huawei Watch 2 under 5 activities
Cardiac, respiratory, and Combination of
(standing, walking, jogging,
12 participants motion artifact signals convolutional and (5 types of activity)
[65] Activity jumping, sitting). Each activity
(6 M, 6 F) extracted from PPG signal recurrent neural F1 score = 0.78
lasted around 90 s. with around
using bandpass filter. networks
1-min gap between two
consecutive activities.
Electronics 2023, 12, 2923 20 of 24

Table 2 presents findings demonstrating the significant monitoring of various health

indicators through photoplethysmography (PPG) measurements using wrist-wearable
devices. While these studies have provided valuable insights, a significant limitation exists
in the current literature. Most of these investigations have primarily focused on evaluating
the accuracy of the algorithm at the time of its development and creation. However, it is
crucial to acknowledge that human physiology is subject to change over time. Wearable
devices are designed to be used in diverse real-world environments rather than controlled
laboratory conditions. As a result, there is a need to investigate the algorithm’s accuracy
and performance in such daily scenarios. Evaluating the sustained accuracy of these
algorithms becomes imperative to ensure their reliability and effectiveness in practical
applications. One aspect that warrants attention is the short-term accuracy assessment,
involving monitoring periods as short as 24 h. This evaluation provides insights into the
algorithm’s stability over a brief duration. Within this timeframe, factors such as daily
activities, physiological variations, and external influences may impact the algorithm’s
performance. By examining the accuracy over a 24-h period, potential deviations or
inconsistencies that may arise during typical daily routines can be identified. Furthermore,
it is also essential to investigate the long-term accuracy of these algorithms over several
months. Human physiology undergoes gradual changes over time due to factors such
as aging, lifestyle modifications, or underlying health conditions. Additionally, wearable
devices are subject to environmental factors such as temperature, humidity, and motion
artifacts that can affect the measurement accuracy. Evaluating the algorithm’s performance
over an extended period provides insights into its robustness and ability to maintain
accuracy amidst changing physiological and environmental conditions. Assessing the
sustained accuracy of these algorithms is crucial for their practical implementation in
healthcare settings and ensures that they remain reliable and effective beyond their initial
development stages. By evaluating their accuracy in real-world scenarios over extended
periods, healthcare professionals can have confidence in utilizing these algorithms for
continuous health monitoring and timely interventions.
While there have been instances where users have worn photoplethysmography (PPG)
systems for extended periods without experiencing any skin burns or allergic reactions,
it is important to highlight that reports of burns or skin irritation in areas where the
system is worn have been documented within the internet community [104–106]. PPG
sensors typically use LEDs to emit light into the skin. LEDs emit a certain amount of
heat as a byproduct of their operation. If the LEDs are not properly calibrated or they
generate excessive heat, localized heating of the skin can occur, increasing the risk of
burns or discomfort. In addition, prolonged contact and repetitive movement during
physical activity can create friction between the device and the skin. Friction generates
heat, which, when trapped between the device and the skin, can cause thermal injuries
or burns. This heating is especially relevant if the device fits tightly or if the user engages
in intense physical exercise that leads to excessive sweating and reduced heat dissipation.
Nevertheless, there is currently a lack of comprehensive research addressing the occurrence
of skin burns, specifically in relatively low-temperature environments, resulting from the
prolonged use of PPG sensors. Therefore, it is crucial to investigate which users may be
more susceptible to low-temperature burns and to explore strategies aimed at developing
sensors and systems that effectively prevent such occurrences.

Author Contributions: Conceptualization, K.B.K. and H.J.B.; methodology and formal analysis,
K.B.K. and H.J.B.; investigation, K.B.K. and H.J.B.; writing—original draft preparation, K.B.K.;
writing—review and editing, H.J.B.; visualization, K.B.K. and H.J.B.; supervision, H.J.B. All authors
have read and agreed to the published version of the manuscript.
Funding: This article was supported by the “Regional Innovation Strategy (RIS)” administered
through the National Research Foundation of Korea (NRF), funded by the Ministry of Educa-
tion (MOE) (2021RIS-004). This work was also supported by the Soonchunhyang University Re-
search Fund.
Electronics 2023, 12, 2923 21 of 24

Data Availability Statement: Not applicable.

Conflicts of Interest: The authors declare no conflict of interest.

References
1. Xie, Y.; Lu, L.; Gao, F.; He, S.-J.; Zhao, H.-J.; Fang, Y.; Yang, J.-M.; An, Y.; Ye, Z.-W.; Dong, Z. Integration of artificial intelligence,
blockchain, and wearable technology for chronic disease management: A new paradigm in smart healthcare. Curr. Med. Sci. 2021,
41, 1123–1133. [CrossRef] [PubMed]
2. Chu, E.C.-P.; Mok, S.T.K.; Chow, I.S.W.; Chin, W.L. The opportunity to unlock the architecture of healthcare model: Chiropractic
care-at-home. J. Contemp. Chiropr. 2022, 5, 44–49.
3. Lim, Y.G.; Hong, K.H.; Kim, K.K.; Shin, J.H.; Lee, S.M.; Chung, G.S.; Baek, H.J.; Jeong, D.-U.; Park, K.S. Monitoring physiological
signals using nonintrusive sensors installed in daily life equipment. Biomed. Eng. Lett. 2011, 1, 11–20. [CrossRef]
4. Lee, S.M.; Lee, D. Healthcare wearable devices: An analysis of key factors for continuous use intention. Serv. Bus. 2020, 14,
503–531. [CrossRef]
5. Ahmed, A.; Aziz, S.; Abd-alrazaq, A.; Farooq, F.; Sheikh, J. Overview of Artificial Intelligence–Driven Wearable Devices for
Diabetes: Scoping Review. J. Med. Internet Res. 2022, 24, e36010. [CrossRef]
6. Miller, D.J.; Sargent, C.; Roach, G.D. A Validation of Six Wearable Devices for Estimating Sleep, Heart Rate and Heart Rate
Variability in Healthy Adults. Sensors 2022, 22, 6317. [CrossRef]
7. Cho, J. Current status and prospects of health-related sensing technology in wearable devices. J. Healthc. Eng. 2019, 2019, 3924508.
[CrossRef]
8. Steven Eyobu, O.; Han, D.S. Feature Representation and Data Augmentation for Human Activity Classification Based on Wearable
IMU Sensor Data Using a Deep LSTM Neural Network. Sensors 2018, 18, 2892. [CrossRef]
9. Fotouhi-Ghazvini, F.; Abbaspour, S. Wearable Wireless Sensors for Measuring Calorie Consumption. J. Med. Signals Sens. 2020, 10,
19–34. [CrossRef]
10. Pham, T.T.; Duong, H.T.; Suh, Y.S. Opportunistic Calibration Method for Walking Distance Estimation Using a Waist-Mounted
Inertial Sensor. IEEE Trans. Instrum. Meas. 2020, 69, 7906–7913. [CrossRef]
11. Allen, J. Photoplethysmography and its application in clinical physiological measurement. Physiol. Meas. 2007, 28, R1–R39.
[CrossRef]
12. Castaneda, D.; Esparza, A.; Ghamari, M.; Soltanpur, C.; Nazeran, H. A review on wearable photoplethysmography sensors and
their potential future applications in health care. Int. J. Biosens. Bioelectron. 2018, 4, 195–202.
13. Loh, H.W.; Xu, S.; Faust, O.; Ooi, C.P.; Barua, P.D.; Chakraborty, S.; Tan, R.-S.; Molinari, F.; Acharya, U.R. Application of
photoplethysmography signals for healthcare systems: An in-depth review. Comput. Methods Programs Biomed. 2022, 216, 106677.
[CrossRef]
14. Biswas, D.; Simões-Capela, N.; Van Hoof, C.; Van Helleputte, N. Heart Rate Estimation from Wrist-Worn Photoplethysmography:
A Review. IEEE Sens. J. 2019, 19, 6560–6570. [CrossRef]
15. Maity, A.K.; Veeraraghavan, A.; Sabharwal, A. PPGMotion: Model-based detection of motion artifacts in photoplethysmography
signals. Biomed. Signal Process. Control 2022, 75, 103632. [CrossRef]
16. Tamura, T.; Maeda, Y.; Sekine, M.; Yoshida, M. Wearable Photoplethysmographic Sensors—Past and Present. Electronics 2014, 3,
282–302. [CrossRef]
17. Schmitt, J.M. Simple photon diffusion analysis of the effects of multiple scattering on pulse oximetry. IEEE Trans. Biomed. Eng.
1991, 38, 1194–1203. [CrossRef]
18. Rolfe, P. In Vivo Near-Infrared Spectroscopy. Annu. Rev. Biomed. Eng. 2000, 2, 715–754. [CrossRef]
19. Kumar, G.; Schmitt, J.M. Optimal probe geometry for near-infrared spectroscopy of biological tissue. Appl. Opt. 1997, 36,
2286–2293. [CrossRef]
20. McCully, K.; Hamaoka, T. Near-infrared spectroscopy: What can it tell us about oxygen saturation in skeletal muscle. Exerc. Sport
Sci. Rev. 2000, 28, 123–127.
21. Maikala, R.V. Modified Beer’s Law—Historical perspectives and relevance in near-infrared monitoring of optical properties of
human tissue. Int. J. Ind. Ergon. 2010, 40, 125–134. [CrossRef]
22. Wukitsch, M.W.; Petterson, M.T.; Tobler, D.R.; Pologe, J.A. Pulse oximetry: Analysis of theory, technology, and practice. J. Clin.
Monit. Comput. 1998, 4, 290–301. [CrossRef] [PubMed]
23. Brown, W.E.L.; Hill, A.V. The oxygen-dissociation curve of blood, and its thermodynamical basis. Proc. R. Soc. Lond. Ser. B 1923,
94, 297–334.
24. Maxim Integrated. MAX30101 Data Sheet. 2020. Available online: https://www.analog.com/media/en/technical-
documentation/data-sheets/MAX30101.pdf (accessed on 18 May 2023).
25. Maxim Integrated. MAX86150 Data Sheet. 2018. Available online: https://www.analog.com/media/en/technical-
documentation/data-sheets/MAX86150.pdf (accessed on 18 May 2023).
26. Maxim Integrated. MAX86916 Data Sheet. 2019. Available online: https://www.analog.com/media/en/technical-
documentation/data-sheets/MAX86916.pdf (accessed on 18 May 2023).
Electronics 2023, 12, 2923 22 of 24

27. Analog Devices. ADPD144RI Data Sheet. 2019. Available online: https://www.analog.com/media/en/technical-documentation/
data-sheets/ADPD144RI.pdf (accessed on 18 May 2023).
28. Analog Devices. ADPD188GG Data Sheet. 2018. Available online: https://www.analog.com/media/en/technical-
documentation/data-sheets/adpd188gg.pdf (accessed on 18 May 2023).
29. Osram Opto Semiconductors. SFH 7051 Data Sheet. 2016. Available online: https://look.ams-osram.com/m/1bc51f319852c566
/original/SFH-7051.pdf (accessed on 18 May 2023).
30. Osram Opto Semiconductors. SFH 7072 Data Sheet. 2022. Available online: https://look.ams-osram.com/m/682b32d8c8dd371
3/original/SFH-7072.pdf (accessed on 18 May 2023).
31. Osram Opto Semiconductors. SFH 7050 Data Sheet. 2016. Available online: https://look.ams-osram.com/m/470e18924403347a/
original/SFH-7050.pdf (accessed on 18 May 2023).
32. AMS AG. AS7024 Data Sheet. 2018. Available online: https://ams.com/documents/20143/36005/AS7024_DS000469_4-00.pdf
(accessed on 18 May 2023).
33. AMS AG. AS7-26GG Data Sheet. 2021. Available online: https://look.ams-osram.com/m/6615cb0b344731f0/original/AS702
6GG-DS000622.pdf (accessed on 18 May 2023).
34. Baek, H.J.; Shin, J.; Cho, J. The Effect of Optical Crosstalk on Accuracy of Reflectance-Type Pulse Oximeter for Mobile Healthcare.
J. Healthc. Eng. 2018, 2018, 3521738. [CrossRef]
35. NXP Semiconductors. Kinetis KL03 Data Sheet. 2022. Available online: https://www.nxp.com/docs/en/data-sheet/KL03P2
4M48SF0.pdf (accessed on 18 May 2017).
36. Zhang, D.; Liu, Z.; Wu, Y.; Ji, S.; Yuan, Z.; Liu, J.; Zhu, M. In Situ Construction a Stable Protective Layer in Polymer Electrolyte for
Ultralong Lifespan Solid-State Lithium Metal Batteries. Adv. Sci. 2022, 9, 2104277. [CrossRef]
37. Ownby, N.B.; Flynn, K.A.; Calhoun, B.H. Modeling Energy Aware Photoplethysmography for Personalized Healthcare Applica-
tions. IEEE Trans. Biomed. Circuits Syst. 2022, 16, 570–579. [CrossRef]
38. Lee, E.; Lee, C. PPG-Based Smart Wearable Device with Energy-Efficient Computing for Mobile Health-Care Applications. IEEE
Sens. J. 2021, 21, 13564–13573. [CrossRef]
39. Lin, B.; Ma, Z.; Atef, M.; Ying, L.; Wang, G. Low-Power High-Sensitivity Photoplethysmography Sensor for Wearable Health
Monitoring System. IEEE Sens. J. 2021, 21, 16141–16151. [CrossRef]
40. Ortegón-Aguilar, J.; Castillo-Atoche, A.; Becerra-Nuñez, G.; Estrada-López, J.J.; Osorio-de-la-Rosa, E.; Carrasco-Alvarez, R.; Datta,
A.; Vázquez-Castillo, J. Multimodal Power Management Based on Decision Tree for Internet of Wearable Things Systems. Appl.
Sci. 2023, 13, 4351. [CrossRef]
41. Ghasemzadeh, H.; Amini, N.; Saeedi, R.; Sarrafzadeh, M. Power-Aware Computing in Wearable Sensor Networks: An Optimal
Feature Selection. IEEE Trans. Mob. Comput. 2015, 14, 800–812. [CrossRef]
42. Kalantarian, H.; Sideris, C.; Mortazavi, B.; Alshurafa, N.; Sarrafzadeh, M. Dynamic Computation Offloading for Low-Power
Wearable Health Monitoring Systems. IEEE Trans. Biomed. Eng. 2017, 64, 621–628. [CrossRef]
43. Liang, J.M.; Chen, J.J.; Cheng, H.H.; Tseng, Y.C. An energy-efficient sleep scheduling with QoS consideration in 3GPP LTE-
advanced networks for Internet of things. IEEE J. Emerg. Sel. Topics Circuits Syst. 2013, 3, 13–22. [CrossRef]
44. Wu, H.; Chen, C.; Weng, K. An Energy-Efficient Strategy for Microcontrollers. Appl. Sci. 2021, 11, 2581. [CrossRef]
45. Rezaie, H.; Ghassemian, M. An Adaptive Algorithm to Improve Energy Efficiency in Wearable Activity Recognition Systems.
IEEE Sens. J. 2017, 17, 5315–5323. [CrossRef]
46. Covi, E.; Donati, E.; Liang, X.; Kappel, D.; Heidari, H.; Payvand, M.; Wang, W. Adaptive Extreme Edge Computing for Wearable
Devices. Front. Neurosci. 2021, 15, 611300. [CrossRef]
47. Choi, A.; Shin, H. Photoplethysmography sampling frequency: Pilot assessment of how low can we go to analyze pulse rate
variability with reliability? Physiol. Meas. 2017, 38, 586. [CrossRef]
48. Béres, S.; Hejjel, L. The minimal sampling frequency of the photoplethysmogram for accurate pulse rate variability parameters in
healthy valunteers. Biomed. Signal Process. Control 2021, 68, 102589. [CrossRef]
49. Peláez-Coca, M.D.; Hernando, A.; Lázaro, J.; Gil, E. Impact of the PPG Sampling Rate in the Pulse Rate Variability Indices
Evaluating Several Fiducial Points in Different Pulse Waveforms. IEEE J. Biomed. Health Inform. 2022, 26, 539–549. [CrossRef]
50. Fuller, D.; Colwell, E.; Low, J.; Orychock, K.; Tobin, M.A.; Simango, B.; Buote, R.; Van Heerden, D.; Luan, H.; Cullen, K.; et al.
Reliability and Validity of Commercially Available Wearable Devices for Measuring Steps, Energy Expenditure, and Heart Rate:
Systematic Review. JMIR mHealth uHealth 2020, 8, e18694. [CrossRef]
51. Schäfer, A.; Vagedes, J. How accurate is pulse rate variability as an estimate of heart rate variability? A review on studies mhealth
comparing photoplethysmographic technology with an electrocardiogram. Int. J. Cardiol. 2013, 166, 15–29. [CrossRef]
52. Gil, E.; Orini, M.; Bailón, R.; Vergara, J.M.; Mainardi, L.; Laguna, R. Photoplethysmography pulse rate variability as a surrogate
measurement of heart rate variability during non-stationary conditions. Physiol. Meas. 2010, 31, 1271. [CrossRef] [PubMed]
53. Zubair, M.; Yoon, C. Multilevel mental stress detection using ultra-short pulse rate variability series. Biomed. Signal Process.
Control 2020, 57, 101736. [CrossRef]
54. Pugliese, L.; Violante, M.; Groppo, S. A Novel Algorithm for Detecting the Drowsiness Onset in Real-Time. IEEE Access 2022, 10,
42601–42606. [CrossRef]
Electronics 2023, 12, 2923 23 of 24

55. Wulterkens, B.M.; Fonseca, P.; Hermans, L.W.; Ross, M.; Cerny, A.; Anderer, P.; Long, X.; van Dijk, J.P.; Vandenbussche, N.; Pillen,
S.; et al. It is all in the wrist: Wearable sleep staging in a clinical population versus reference polysomnography. Nat. Sci. Sleep
2021, 13, 885–897. [CrossRef]
56. Mukkamala, R.; Stergiou, G.S.; Avolio, A.P. Cuffless Blood Pressure Measurement. Annu. Rev. Biomed. Eng. 2022, 24, 203–230.
[CrossRef]
57. Islam, S.M.S.; Chow, C.K.; Daryabeygikhotbehsara, R.; Subedi, N.; Rawstorn, J.; Tegegne, T.; Karmakar, C.; Siddiqui, M.U.;
Lambert, G.; Maddison, R. Wearable cuffless blood pressure monitoring devices: A systematic review and meta-analysis. Eur.
Heart J. Digit. Health 2022, 3, 323–337. [CrossRef]
58. Yao, P.; Xue, N.; Yin, S.; You, C.; Guo, Y.; Shi, Y. Multi-Dimensional Feature Combination Method for Continuous Blood Pressure
Measurement Based on Wrist PPG Sensor. IEEE J. Biomed. Health Inform. 2022, 26, 3708–3719. [CrossRef]
59. Shi, W.; Zhou, C.; Zhang, Y.; Li, K.; Ren, X.; Liu, H.; Ye, X. Hybrid modeling on reconstitution of continuous arterial blood
pressure using finger photoplethysmography. Biomed. Signal Process. Control 2023, 85, 104972. [CrossRef]
60. Monte-Moreno, E. Non-invasive estimation of blood glucose and blood pressure from a photoplethysmography by means of
machine learning techniques. Artif. Intell. Med. 2011, 53, 127–128. [CrossRef]
61. Tsai, C.; Li, C.; Lam, R.; Li, C.; Ho, S. Diabetes Care in Motion: Blood Glucose Estimation Using Wearable Devices. IEEE Consum.
Electron. Mag. 2020, 8, 30–34. [CrossRef]
62. Islam, T.; Ahmed, M.; Hassanuzzaman, M.; Amir, S.B.; Rahman, T. Blood Glucose Level Regression for Smartphone PPG Signals
Using Machine Learning. Appl. Sci. 2021, 11, 618. [CrossRef]
63. Prabha, A.; Yadav, J.; Rani, A.; Singh, V. Design of intelligent diabetes mellitus detection system using hybrid feature selection
based on XGBoost classifier. Comput. Biol. Med. 2021, 136, 104664. [CrossRef]
64. Prabha, A.; Yadav, J.; Rani, A.; Singh, V. Intelligent estimation of blood glucose level using wristband PPG signal and physiological
parameters. Biomed. Signal Process. Control 2022, 78, 103876. [CrossRef]
65. Boukhechba, M.; Cai, L.; Wu, C.; Barnes, L.E. ActiPPG: Using deep neural networks for activity recognition from wrist-worn
photoplethysmography (PPG) sensors. Smart Health 2019, 14, 100082. [CrossRef]
66. Zonios, G.; Dimou, A.; Bassukas, I.; Galaris, D.; Tsolakidis, A.; Kaxiras, E. Melanin absorption spectroscopy: New method for
noninvasive skin investigation and melanoma detection. J. Biomed. Opt. 2008, 13, 014017. [CrossRef]
67. Bashkatov, A.N.; Genina, E.A.; Kochubey, V.I.; Tuchin, V.V. Optical properties of human skin subcutaneous and mucous tissues in
the wavelength range from 400 to 2000 nm. J. Phys. D Appl. Phys. 2005, 38, 2543–2555. [CrossRef]
68. Lemay, M.; Bertschi, M.; Sola, J.; Renevey, P.; Parak, J.; Korhonen, I. Application of optical heart rate monitoring. In Wearable
Sensors: Fundamentals Implementation and Applications; Academic Press: Cambridge, MA, USA, 2014; pp. 105–129.
69. Lister, T.; Wright, P.A.; Chappell, P.H. Optical properties of human skin. J. Biomed. Opt. 2012, 17, 909011. [CrossRef]
70. Taroni, P.; Pifferi, A.; Torricelli, A.; Comelli, D.; Cubeddu, R. In vivo absorption and scattering spectroscopy of biological tissues.
Photochem. Photobiol. Sci. 2003, 2, 124–129. [CrossRef]
71. Mannheimer, P.D.; Cascini, J.R.; Fein, M.E.; Nierlich, S.L. Wavelength selection for low-saturation pulse oximetry. IEEE Trans.
Biomed. Eng. 1997, 44, 148–158. [CrossRef]
72. Sinex, J.E. Pulse oximetry: Principle and limitations. AM J. Emerg. Med. 1999, 17, 59–66. [CrossRef]
73. Baek, H.J.; Chung, G.S.; Kim, K.K.; Park, K.S. A smart health monitoring chair for nonintrusive measurement of biological signals.
IEEE Trans. Inform. Technol. Biomed. 2012, 16, 150–158. [CrossRef] [PubMed]
74. Naschitz, J.E.; Bezobchuk, S.; Mussafia-Priselac, R.; Sundick, S.; Dreyfuss, D.; Khorshidi, I.; Karidis, A.; Manor, H.; Nagar, M.;
Peck, E.R.; et al. Pulse Transit Time by R-Wave-Gated Infrared Photoplethysmography: Review of the Literature and Personal
Experience. J. Clin. Monit. Comput. 2004, 18, 333–342. [CrossRef] [PubMed]
75. Stojanovic, R.; Karadaglic, D. A LED-LED-based photoplethysmography sensor. Physiol. Meas. 2007, 28, N19. [CrossRef]
[PubMed]
76. Teng, X.F.; Zhang, Y.T. The effect of contacting force on photoplethysmographic signals. Physiol. Meas. 2004, 25, 1323. [CrossRef]
77. Maeda, Y.; Sekine, M.; Tamura, T. The Advantages of Wearable Green Reflected Photoplethysmography. J. Med. Syst. 2011, 35,
829–834. [CrossRef]
78. Mainster, M.A. Wavelength Selection in Macular Photocoagulation: Tissue Optics, Thermal Effects, and Laser Systems. Ophthal-
mology 1986, 93, 952–958. [CrossRef]
79. Matsumura, K.; Rolfe, P.; Lee, J.; Yamakoshi, T. iPhone 4s photoplethysmography: Which light color yields the most accurate
heart rate and normalized pulse volume using the iPhysioMeter application in the presence of motion artifact? PLoS ONE 2014,
9, e91205. [CrossRef]
80. Shao, D.; Liu, C.; Tsow, F.; Yang, Y.; Du, Z.; Iriya, R.; Yu, H.; Tao, N. Noncontact Monitoring of Blood Oxygen Saturation Using
Camera and Dual-Wavelength Imaging System. IEEE Trans. Biomed. Eng. 2016, 63, 1091–1098. [CrossRef]
81. Wijshoff, R.W.C.G.R.; Veen, J.; Van der Lee, A.M.; Mulder, L.; Stijnen, M.; Van Tuijl, S.; Aarts, R.M. PPG motion artifact handling
using a self-mixing interferometric sensor. Proc. SPIE 2011, 7894, 78940F.
82. Lee, J.; Matsumura, K.; Yamakoshi, K.I.; Rolfe, P.; Tanaka, S.; Yamakoshi, T. Comparison between red, green and blue light
reflection photoplethysmography for heart rate monitoring during motion. Conf. Proc. IEEE Eng. Med. Biol. Soc. 2013, 2013,
1724–1727.
Electronics 2023, 12, 2923 24 of 24

83. Lee, B.; Han, J.; Baek, H.J.; Shin, J.H.; Park, K.S.; Yi, W.J. Improved elimination of motion artifacts from a photoplethysmographic
signal using a Kalman smoother with simultaneous accelerometry. Physiol. Meas. 2010, 31, 1585. [CrossRef]
84. Ram, M.R.; Madhav, K.V.; Krishna, E.H.; Komalla, N.R.; Reddy, K.A. A Novel Approach for Motion Artifact Reduction in PPG
Signals Based on AS-LMS Adaptive Filter. IEEE Trans. Instrum. Meas. 2012, 61, 1445–1457. [CrossRef]
85. Chowdhury, S.S.; Hyder, R.; Hafiz, M.S.B.; Haque, M.A. Real-Time Robust Heart Rate Estimation from Wrist-Type PPG Signals
Using Multiple Reference Adaptive Noise Cancellation. IEEE J. Biomed. Health Inform. 2018, 22, 450–459. [CrossRef]
86. Zhang, Y.; Song, S.; Vullings, R.; Biswas, D.; Simões-Capela, N.; van Helleputte, N.; van Hoof, C.; Groenendaal, W. Motion Artifact
Reduction for Wrist-Worn Photoplethysmograph Sensors Based on Different Wavelengths. Sensors 2019, 19, 673. [CrossRef]
87. Wang, C.Y.; Tang, K.T. Active noise cancellation of motion artifacts in pulse oximetry using isobestic wavelength light source.
In Proceedings of the 2011 IEEE International Symposium of Circuits and Systems, Rio de Janeiro, Brazil, 15–18 May 2011;
pp. 1029–1032.
88. Ray, D.; Collins, T.; Wolley, S.I.; Ponnapalli, P.V.S. A Review of Wearable Multi-Wavelength Photoplethysmography. IEEE Rev.
Biomed. Eng. 2023, 16, 136–158. [CrossRef]
89. Liu, J.; Yan, B.P.; Dai, W.-X.; Ding, X.-R.; Zhang, Y.-T.; Zhao, N. Multi-wavelength photoplethysmography method for skin arterial
pulse extraction. Biomed. Opt. Express 2016, 7, 4313–4326. [CrossRef]
90. Lee, J.; Kim, M.; Park, H.-K.; Kim, I.Y. Motion Artifact Reduction in Wearable Photoplethysmography Based on Multi-Channel
Sensors with Multiple Wavelengths. Sensors 2020, 20, 1493. [CrossRef]
91. Baek, H.J.; Shin, J.; Jin, G.; Cho, J. Reliability of the Parabola Approximation Method in Heart Rate Variability Analysis Using
Low-Sampling-Rate Photoplethysmography. J. Med. Syst. 2017, 41, 189. [CrossRef]
92. Asare, L.; Kviesis-Kipge, E.; Grabovskis, A.; Rubins, U.; Erts, R.; Spigulis, J. Multi-spectral photoplethysmography biosensor. Proc.
SPIE 2011, 8073, 374–379.
93. Kumar, M.; Veeraraghavan, A.; Sabharwal, A. DistancePPG: Robust non-contact vital signs monitoring using a camera. Biomed.
Opt. Express 2015, 6, 1565–1588. [CrossRef]
94. Yan, L.; Hu, S.; Alzahrani, A.; Alharbi, S.; Blanos, P. A multi-wavelength opto-electronic patch sensor to effectively detect
physiological changes against human skin types. Biosensors 2017, 7, 22. [CrossRef] [PubMed]
95. Jeong, C.; Yoon, H.; Kang, H.; Yeom, H. Effects of skin surface temperature on photoplethysmography. J. Healthc. Eng. 2014, 5,
429–438. [CrossRef] [PubMed]
96. Spigulis, J.; Gailite, L.; Lihachev, A.; Erts, R. Simultaneous recording of skin blood pulsations at different vascular depths by
multiwavelength photoplethysmography. Appl. Opt. 2007, 46, 1754–1759. [CrossRef] [PubMed]
97. Liu, J.; Yan, B.P.; Zhang, Y.T.; Ding, X.R.; Su, P.; Zhao, N. Multi-Wavelength Photoplethysmography Enabling Continuous Blood
Pressure Measurement with Compact Wearable Electronics. IEEE Trans. Biomed. Eng. 2019, 66, 1514–1525. [CrossRef]
98. Liu, J.; Qiu, S.; Luo, N.; Lau, S.K.; Yu, H.; Kwok, T.; Zhang, Y.T.; Zhao, N. PCA-Based Multi-Wavelength Photoplethysmography
Algorithm for Cuffless Blood Pressure Measurement on Elderly Subjects. IEEE J. Biomed. Health Inform. 2021, 25, 663–673.
[CrossRef]
99. Hossain, S.; Haque, C.A.; Kim, K.-D. Quantitative Analysis of Different Multi-Wavelength PPG Devices and Methods for
Noninvasive In-Vivo Estimation of Glycated Hemoglobin. Appl. Sci. 2021, 11, 6867. [CrossRef]
100. Rachim, V.P.; Chung, W.-Y. Wearable-band type visible-near infrared optical biosensor for non-invasive blood glucose monitoring.
Sens. Actuators B Chem. 2019, 286, 173–180. [CrossRef]
101. Gorny, A.W.; Liew, S.J.; Tan, C.S.; Müller-Riemenschneider, F. Fitbit Charge HR Wireless Heart Rate Monitor: Validation Study
Conducted Under Free-Living Conditions. JMIR mHealth uHealth 2017, 5, e157. [CrossRef]
102. Nelson, B.W.; Allen, N.B. Accuracy of Consumer Wearable Heart Rate Measurement During an Ecologically Valid 24-Hour Period:
Intraindividual Validation Study. JMIR mHealth uHealth 2019, 7, e10828. [CrossRef]
103. Baek, H.J.; Cho, J. Novel heart rate variability index for wrist-worn wearable devices subject to motion artifacts that complicate
measurement of the continuous pulse inverval. Physiol. Meas. 2019, 40, 105010.
104. Daily Mail. Samsung Galaxy Watch User Claims Wearable Burned Their Wrist When Left on While They Were Sleeping.
September 2022. Available online: https://www.dailymail.co.uk/sciencetech/article-11232191/Samsung-Galaxy-Watch-user-
claims-wearable-BURNED-wrist-left-sleeping.html (accessed on 22 May 2023).
105. 7 News Miami. Fort Lauderdale Woman Says Apple Watch Left Burn on Her Arm. April 2022. Available online: https:
//wsvn.com/news/local/fort-lauderdale-woman-says-apple-watch-left-burn-on-her-arm/ (accessed on 22 May 2023).
106. ABC 7 News. Fitbit Smartwatch Recalled for Burn Risk 8 Years after Previous Tracker Recall. March 2022. Available online:
https://abc7news.com/fitbit-ionic-recall-smartwatch-tracker-burn/11615270/ (accessed on 22 May 2023).

Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual
author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to
people or property resulting from any ideas, methods, instructions or products referred to in the content.