sensors

Article
A Smart Pillow for Health Sensing System Based on
Temperature and Humidity Sensors
Songsheng Li 1, * and Christopher Chiu 2
1 Department of Computer Engineering, Guangdong College of Business and Technology,
Zhaoqing 526020, China
2 Faculty of Engineering and Information Technology, University of Technologies,
Sydney, Ultimo NSW 2007, Australia; [email protected]
* Correspondence: [email protected]; Tel.: +86-159-1538-6438




Received: 18 September 2018; Accepted: 26 October 2018; Published: 29 October 2018


Abstract: The quality of sleep affects the patient’s health, along with the observation of vital life
signs such as body temperature and sweat in sleep, is essential in the monitoring of sleep as well as
clinical diagnosis. However, traditional methods in recording physiological change amidst sleep is
difficult without being intrusive. The smart pillow is developed to provide a relatively easy way to
observe one’s sleep condition, employing temperature and humidity sensors by implanting them
inside the pillow in strategic positions. With the patient’s head on the pillow, the roles of sensors
are identified as main, auxiliary or environmental temperature, based on the differences of value
from three temperature sensors, thus the pattern of sleep can be extracted by statistical analysis,
and the body temperature is inferred by a specially designed Fuzzy Logic System if the head-on
position is stable for more than 15 min. Night sweat is reported on data from the humidity sensor.
Therefore, a cloud-based health-sensing system is built in the smart pillow to collect and analyze data.
Experiments from various individuals prove that statistical and inferred results reflect normal and
abnormal conditions of sleep accurately. The daily sleeping information of patients from the pillow is
helpful in the decision-making of diagnoses and treatment, and users can change their habits of sleep
gradually by observing the data with their health professional.

Keywords: smart pillow; health sensing system; temperature; humidity; sensor

1. Introduction
With the advent of a globally aging society, the demand for remote health management systems is
gaining acceptance within the medical community. The popularity of sensors, networks and big-data
analytics has established a sound foundation for a networked health system. The problem that
needs improvement is how one can observe physiological signs in a non-invasive and unobtrusive
manner. One third of a person’s life is spent on sleep, therefore the quality and habit of sleep affects
health, and the negative impacts of too much, too little, and poor-quality sleep are taken seriously.
Physiological change in the middle of sleep is a significant indicator of health but is difficult to be
observed using traditional methods. A pillow is used as a head support for comfort when sleeping.
Hence, if data can be collected from a pillow under the no-perception condition, it will be a direct,
effective, non-invasive and unperturbed way to analyze sleep and health. This article concentrates on
the monitoring of body temperature, which is one of the most important physiological signs required
for clinical diagnosis.
This paper provides a straight-forward way to observe sleep conditions by employing temperature
and humidity sensors implanted inside the pillow. Body temperature will be measured in a way
analogous to zero heat flux (ZHF) first, then compensated by a Fuzzy Logic system. Sensing data will

Sensors 2018, 18, 3664; doi:10.3390/s18113664 www.mdpi.com/journal/sensors

Sensors 2018, 18, 3664 2 of 19

be reported to the agent or mobile phone by Bluetooth Low Energy (BLE), then uploaded to cloud
server via the Internet. Data analysis is then processed in the cloud and results presented via mobile
application or web browser. The whole process and parties involved are collectively named the Health
Sensing System. Experiments from various individuals prove that statistical and inferred results reflect
normal and abnormal conditions of sleep accurately.
In the following of the paper, related works such as various measurements of body temperature,
wearable instruments, transmission by BLE or ZigBee, and health monitoring system are presented
first, then from the whole to the parts, details of the proposed health sensing system, smart pillow
and how body temperature being extracted are explained gradually, and result of daily sleeping is
provided and analyzed, which is followed by discussions, conclusions, and the idea of further research.

2. Related Works
In this section, the publications related to measurement of body temperature,
wearable instruments and health monitoring systems are shown.
The importance of body temperature has been agreed by researchers [1–3]; studies on the
relationship between alertness, performance and body temperature prove performance and alertness
is improved if the body temperature is increased, whether or not being synchronized with the internal
biological clock [1]. Research from Lack et al. showed that insomnia in different phases of sleep is
associated with abnormal body temperature rhythms [2] and treatments could adjust the circadian
rhythm by bright light therapy. Coyne et al. found that the mortality from acute stroke was lower,
and thus better outcomes in patients with mild hypothermia on admission, but actually worse if in a
state of hyperthermia [3].
Information of body temperature has become indispensable, that researchers have dedicated their
time to find unobtrusive and accurate measurement approaches. Use of the Kalman Filter is extensively
adopted in engineering tracking problems [4], with a model developed to estimate the time course
of core temperature (CT), employing a series of heart rate (HR) measurements as a main indicator.
The filter is comprised of two models: a time update modeling how CT changes with uncertainty
and an observation model mapping from HR to CT with uncertainty; then the filter is trained and
validated. Then it is examined by the Bland–Altman limits of agreement (LoA) method, with the
results accurate to provide practical indication of CT in the workplace, but it is not a replacement of
direct measurement. Fox and Solman presented a new method for monitoring deep body temperature
from the skin surface [5,6], known as ZHF, by creating a zone of zero heat-flow across the body shell to
bring the deep body temperature to the skin surface, where it can be measured with a simple electronic
thermometer. This method is proven to track CT very well in various ambient conditions [7,8] but
using a heater to stop surface convection is an inherent weakness. An alternative method called
dual-heat-flux (DHF) achieves similar outcomes by creating dual heat flux channels with less energy
consumption and more sensors [9]. Huang et al. proposed improvements for DHF, such as modifying
the probe size, measurement depth beneath the skin, and an aluminum cover to boost measurement
accuracy [10,11], but also suggested the probe height be reduced by half.
A simpler device called a Double Sensor is invented based on skin temperature and heat flow,
but without the use of a heating element [12]. This device employs only two temperature
sensors, with the accuracy and precision being competitive with rectal and distal esophageal
measurements [13,14]. Sim et al. embedded one DHF and two Double Sensors into a neck pillow
to estimate core body temperature in the bed [15]. By this way, measurement can be derived from
3 different sleep positions but sleeping with a neck pillow is inconvenient. A garment made by
Temperature Sensing Fabric (TSF) is manufactured on a knitting machine by laying fine metal wires
into the double-layer knitted structure [16], which employs the principle that the metal wire changes
its electrical resistance with the change in temperature. It is wearable and user-friendly, but difficult to
produce in batches.
Sensors 2018, 18, 3664 3 of 19

Wearable devices are increasingly popular because of the advancement of the electronics industry,
with technologies making low-cost, comfortable, and unobtrusive devices possible [17]. For instance,
high-sensitivity and low-cost printed wearable temperature sensors were developed [18,19], which is
synthesized by poly (3,4-ethlenedioxythiophene) poly (styrene sulfonate) (PEDOT: PSS) and carbon
nanotubes (CNTs). This solves the issue of size, although the next issue to be addressed is the method
of low-power wireless transmission for uploading measurements; hence BLE being the main candidate.
A ring-shaped wrap for the finger was developed by Miah et al. [20], which integrates infrared
Rx and Tx for the heart rate and temperature sensor. The sensing data is transferred to Android
mobile in real time by BLE, then any abnormal signs can trigger an alarm, but the device is still
in its experimental stages. A forehead-worn thermometer that connects to mobile via Bluetooth is
presented [21], which is based on multiple Artificial Neural Networks (ANNs) in which certain hidden
neurons are adaptively selected according to a combination of temperatures, but balancing estimation
accuracy, computational load, and delay time needs to be addressed. Boano et al. designed and built a
non-invasive wearable wireless monitoring prototype for accurate body temperature measurements,
that is accurate to 0.02 ◦ C and provides real-time feedback to the medic [22] based on a Wireless Body
Area Network (WBAN) [23]. The watch has a size of 40 mm × 20 mm whose core component is a
Tyndall 10 mm node and communicates to the sink node by Nordic nRF9e5 RF chipset via the 433 MHz
radio band. Energy consumption issues still need to be solved.
A data-oriented system with reliable transmission and storage is essential for measurement of
vital signs. A WSN-based platform in a homecare monitoring system is presented by Dobrescu et al. in
2008 to share expensive resources at a distance, increase diagnosis accuracy and treatment efficiency
and save time to intervention [24]. They define a mobile node named Patient Wireless Node (PWN)
and fixed node named Home Wireless Node (HWN), with the presented concepts still applicable
today. With the technical limits of the time, an 8-bit microcontroller platform and General Packet Radio
Service (GPRS) are employed in their test, so the performance of the platform is limited.
ZigBee defines a low-data-rate short-range wireless networking based on IEEE 802.15.4 [25,26],
which is widely used in smart home systems for its low-cost implementation and low-power
consumption [27,28]. A remote health monitoring system focusing on body temperature is proposed by
Mansor et al. [29]. They employed the temperature sensor LM35 from Texas Instruments, sending the
reading to Arduino by XBee, with the Arduino board receiving and decoding readings and sending
results to the server by WLAN. Hand temperature measurements were considered in their experiments,
with test results better than a commercial digital thermometer. One of the inconveniences is that they
must hold the sensor in their hand for initial measurements. A more practical system by Kioumars
and Tang adds the physiological signal of the heart rate, in which the sensor is wrapped around the
wrist [30] and the Mean-Body Temperature (MBT) is calculated by Burton’s equation, with the heart
rate measured by pulse oximetry. They found that motion, makeup and light can cause incorrectness
in the measurement. A Long-term Wearable Vital Signs Monitoring System using BSN and ZigBee is
proposed by Guo et al., covering other physiological signs such as ECG, SpO2 (Saturation of Arterial
Oxygen) and systolic blood pressure [31]. The wearable device consists of two parts: a chest band
connects with an ear-probe, as they are working on designing a delicate chest band for comfort and
improving the system performance.
Another device implements an analogous function, but it is worn on the wrist and finger [32].
A mobile gateway for a ubiquitous health care system using Bluetooth to enhance the transmission
capacity of ZigBee is presented by Laine et al. [33], improving the home care system and making
it more flexible and feasible. Doukas and Maglogiannis brings IoT and Cloud Computing towards
pervasive healthcare by reporting data to the cloud server through a lightweight REST-based API [34],
thus completing the information reporting cycle from patient to clinician. Lloret J. et al. proposed
a smart continuous eHealth monitoring system for chronic patients using 5G [35], they employed
wearable devices, mobile phone and an intelligent DataBase for BigData, so heavy traffic needs extra
bandwidth which is the advantage of 5G, but availability and affordability are current issues of 5G.
 Sensors2018,
Sensors 2018,18,
18,3664
x FOR PEER REVIEW 44 of
of1919

Some schemes based on pillow were proposed recently. An IoT-based smart pillow for sleep
Some
quality schemes based
monitoring on pillow
in Ambient wereLiving
Assisted proposed
(AAL) recently.
environmentsAn IoT-based smart pillow
was presented for
[36], which
sleep quality monitoring in Ambient Assisted Living (AAL) environments
integrates sensors of temperature, humidity, luminosity, sound, and vibration to a node of SparkFun was presented [36],
which
ESP32integrates
Thing, so sensors of temperature,
environmental parameters humidity, luminosity,
can be adjusted sound, andtovibration
automatically meet need toofa qualified
node of
SparkFun
sleeping. ESP32 Thing, so environmental
The comfortability of the cushion parameters
and power canconsumption
be adjusted automatically to meetinneed
of node are trivial of
lab but
qualified sleeping. The comfortability of the cushion and power consumption
have to be addressed in product. An under pillow unperturbed sleep monitoring scheme identifies of node are trivial in lab
but have
sleep to be S1–S4
stages addressed
withinan product.
accuracy Anof under
morepillow
than unperturbed
60% [37], which sleepismonitoring
better thanscheme
existing identifies
scheme
sleep
based on hidden Markov model (HMM). It analyzes the pattern of heart rate variability based
stages S1–S4 with an accuracy of more than 60% [37], which is better than existing scheme (HRV)
on hiddenobtained
through Markov RR model (HMM).
interval It analyzes
signal first, andthe patternensemble
employs of heart rate variability
empirical mode(HRV) through
decomposition
obtained
(EEMD) RR interval signal
to eliminate errorsfirst,
fromand employs differences,
individual ensemble empirical mode decomposition
then combines it with HMM (EEMD) to
to calculate
eliminate errors from individual differences, then combines it with HMM to
corresponding stages, but the accuracy of this noninvasive monitoring system is still far from calculate corresponding
stages, but the accuracy
polysomnography of this noninvasive monitoring system is still far from polysomnography (PSG).
(PSG).
The
Theproposed
proposedhealth
healthsensing
sensingsystem
systemisisbased
basedononaasmart
smartpillow
pillowwith
withtemperature
temperatureand andhumidity
humidity
sensors. Besides keeping the comfortability of pillow in a noninvasive way, we
sensors. Besides keeping the comfortability of pillow in a noninvasive way, we concentrate on concentrate on efficiency
ofefficiency
energy, accuracy
of energy,of accuracy
measurement when designing
of measurement whenand implementing
designing the system. the system.
and implementing

3.3.Health
HealthSensing
SensingSystem
System
This
Thissection
sectionbegins
beginswithwithaabrief
briefintroduction
introductionofofdata dataflow
flowininthethesystem,
system,then
thenarchitecture
architectureand and
parts of the system which followed by details of the smart pillow, and finally
parts of the system which followed by details of the smart pillow, and finally how position and how position and body
temperature are extracted
body temperature by Fuzzy
are extracted by Logic
FuzzyisLogic
presented.
is presented.
The
The system consists of three parts: sensors, transportagent,
system consists of three parts: sensors, transport agent,andandserver.
server.InInorder
ordertotoupload
uploaddata
data
totothe
the server, the Internet function has to be integrated into a host of sensors, such as wires,Wi-Fi,
server, the Internet function has to be integrated into a host of sensors, such as wires, Wi-Fi,
oror4G,
4G,which
which requires
requires an AC
an ACpower supply
power or a rechargeable
supply lithiumlithium
or a rechargeable battery;battery;
althoughalthough
it can enlarge
it can
the host of sensors and in turn make it uncomfortable for the user. In contrast,
enlarge the host of sensors and in turn make it uncomfortable for the user. In contrast, BLE and BLE and ZigBee is
designed with power saving capability, so they are perfect for short distance
ZigBee is designed with power saving capability, so they are perfect for short distance communication [38] and
suitable for a host[38]
communication of sensors. As a result,
and suitable the host
for a host of sensors
of sensors. As ais result,
equippedthe with
host BLE/ZigBee, which is
of sensors is equipped
powered by a cell battery, and a transport agent is necessary in consideration
with BLE/ZigBee, which is powered by a cell battery, and a transport agent is necessary in of power consumption,
which is equipped
consideration of with
power bothconsumption,
BLE/ZigBee and Internet
which capability.with
is equipped The data
bothflow diagram ofand
BLE/ZigBee the Internet
system
iscapability.
shown in Figure
The data 1. flow diagram of the system is shown in Figure 1.

Figure1.1.Data-flow
Figure Data-flowblock
blockdiagram
diagramofofsystem.
system.

Sensing
Sensingdatadatawill
willpass
passthrough
throughthe theagent
agentfromfromaahost
hostofofsensors
sensorstotothe
theserver,
server,which
whichmeansmeansthe
the
data
data is not manipulated in the agent. On one end, the agent receives data from BLE/ZigBee,on
is not manipulated in the agent. On one end, the agent receives data from BLE/ZigBee, onthe
the
other
otherend endthe
theagent
agentsends
sends data to the
data server
to the by Internet.
server In the
by Internet. In host of sensors,
the host whenwhen
of sensors, the main chip gets
the main chip
raw
gets raw data from Input/Output (IO) ports, it assembles the raw data according to an protocol
data from Input/Output (IO) ports, it assembles the raw data according to an exclusive exclusive
first, then first,
protocol the assembled data is encrypted
then the assembled by specialized
data is encrypted algorithmsalgorithms
by specialized that is passed
thattois the server
passed to by
the
the agent. In the server, the reverse process is performed according to the specialized
server by the agent. In the server, the reverse process is performed according to the specialized algorithm and
exclusive
algorithm protocol, then theprotocol,
and exclusive final rawthen
datathe
willfinal
be analyzed
raw datatowillgetbe
results. As BLE
analyzed or ZigBee
to get results.isAsanBLE
openor
standard,
ZigBee isencryption is significant
an open standard, for dataissecurity.
encryption An exclusive
significant protocol An
for data security. is necessary
exclusivetoprotocol
organizeis
the
necessary to organize the sensing data with timestamping into a certain number of bytes toon
sensing data with timestamping into a certain number of bytes to ensure data recovery the
ensure
server. If an encrypted algorithm or assembly protocol changes on one side, the
data recovery on the server. If an encrypted algorithm or assembly protocol changes on one side, other side should be
modified
the otheraccordingly.
side should be modified accordingly.
Sensors 2018, 18, 3664 5 of 19
Sensors 2018, 18, x FOR PEER REVIEW 5 of 19

3.1. Architecture of Current System

3.1. Architecture of Current System
The essential element of the current proposed health sensing system is the smart pillow, whose
The essential element of the current proposed health sensing system is the smart pillow, whose core
core component is a BLE single mode chip, which connects to three temperature sensors and one
component is a BLE single mode chip, which connects to three temperature sensors and one humidity
humidity sensor by input and output ports; communicating to the agent wirelessly by BLE. Sensory
sensor by input and output ports; communicating to the agent wirelessly by BLE. Sensory data
data is collected and pre-processed by the BLE chip and reported to a mobile phone or Wi-Fi agent
is collected and pre-processed by the BLE chip and reported to a mobile phone or Wi-Fi agent by
by Bluetooth, then uploaded to the cloud server by Internet. Data is analyzed and health related
Bluetooth, then uploaded to the cloud server by Internet. Data is analyzed and health related features
features such as sweat, fever, insomnia etc. will be extracted in the server. A daily-based health
such as sweat, fever, insomnia etc. will be extracted in the server. A daily-based health report will be
report will be presented visually by webpage or returned to the mobile phone and visualized in a
presented visually by webpage or returned to the mobile phone and visualized in a specified mobile
specified mobile application. Weekly, monthly, and annual health reports and suggestions are
application. Weekly, monthly, and annual health reports and suggestions are provided gradually in
provided gradually in sequence by analyzing health related features based on known medical
sequence by analyzing health related features based on known medical knowledge, or data is shared
knowledge, or data is shared with professional medical organizations with consent of users, as it
with professional medical organizations with consent of users, as it can provide an extra level of patient
can provide an extra level of patient information and help in diagnosis and treatment. The
information and help in diagnosis and treatment. The architecture of the current system is displayed
architecture of the current system is displayed in Figure 2.
in Figure 2.

Figure 2. Current health system implementation.

Figure 2. Current health system implementation.
The pillow is data source of the system, which communicates via Bluetooth. Every pillow has a
uniqueTheBluetooth
pillow is address
data sourceand ofallthe system,
pillows whichsensing
provide communicates
data by an viaidentical
Bluetooth. and Every pillow
specified has a
service
unique Bluetooth address and all pillows provide sensing data by an identical
defined in BLE. The service makes pillows in the same class and address individualizes them. All the and specified service
defined
data in BLE.
needed is a The service to
timestamp makes
be morepillows
useful,in the
but same
a singleclass
modeandchip
address
onlyindividualizes
maintains a timer them. byAll
an
the data Crystal
internal neededoscillator;
is a timestamp
hence to anbe more useful,
external accurate but a single
time mode chipwhen
is synchronized only maintains
a connection a timer
is made by
anan
to internal
agent.Crystal oscillator; hence an external accurate time is synchronized when a connection is
madeThe to an agent. agent is equipped with both Bluetooth and Wi-Fi capability, the former collects
transport
The transport
data from the pillows, agent
the is equipped
latter reportswithdataboth
to theBluetooth
server, and and Wi-Fi capability,
requests accurate time thefrom
former collects
Internet to
data from
provide to the
the pillows,
pillow. Thethe latter reports
connection data to the
between the pillow
server,andandagent
requests accurate
is flexible, it time
couldfrom Internet
be from one
to provide
agent to onetopillow,
the pillow.
or oneThe connection
agent to many between the pillowagent
pillows. Transport and agent
worksisinflexible,
a polling it could
mode,be from
it scans
one agent to one pillow, or one agent to many pillows. Transport agent works
for pillows by the specified service, setups a polling list, then connects to the pillows one by one, in a polling mode, it
scans fordata
collects pillows by the to
and reports specified
the serverservice,
with setups
a unique a polling
Bluetooth list,address.
then connects to the pillows one by
one, collects
The server datacanand reports to pillows
distinguish the server bywith
theiraunique
uniqueBluetooth
Bluetoothaddress.
address. Results of data analysis
The server can distinguish pillows by their unique Bluetooth
in server cover three sections: statistics, inference and suggestion. Statistics address. Results
is the of data analysis
simple and direct in
server cover
calculation orthree sections:
conclusion statistics,
of original inference
data, such asand suggestion.
duration Statistics
of sleeping, turns,is room
the simple and direct
temperature and
calculation or conclusion of original data, such as duration of sleeping, turns,
humidity, and sleeping position distribution, etc. The second part is challenging, as it includes body room temperature and
humidity, and
temperature sleeping
and sweat position
sensing. distribution,
It is difficult etc. The second
to measure part is
accurate challenging,
body temperature as it using
includes body
a classic
temperature and
thermometer, sweatlogic
so fuzzy sensing. It is difficult
is employed to inferto body
measure accurate based
temperature body temperature using a classic
on head and environmental
thermometer,The
temperature. soother
fuzzy logic element
inferred is employed
is sweat,to inferis abody
which temperature
significant indicator based
for Chineseon head and
traditional
environmental temperature. The other inferred element is sweat, which is a significant indicator for
Chinese traditional medicine. The third section of the result is the recommendation of health, based
on the conclusion from the first two sections and general medical knowledge.
Sensors 2018, 18, 3664 6 of 19

Sensors 2018, 18, x FOR PEER REVIEW 6 of 19

medicine. The third section of the result is the recommendation of health, based on the conclusion
from The
the first twophone
mobile sections is and general
a perfect medicalagent
transport knowledge.
as most mobiles are equipped with Bluetooth,
Wi-FiTheandmobile
4G, andphone
it caniseven
a perfect
worktransport
without aagent serverasconnection.
most mobiles Theare equipped
differences with Bluetooth,
between a mobile
Wi-Fi
phoneand and4G, and it cantransport
customized even work without
agent a server
are listed connection.
in Table The differences
1. Compared between transport
to a customized a mobile
phone
agent, and customized of
the advantages transport
mobile agent
phoneare listed
are: it is in
moreTable 1. Compared
flexible as it hastoana customized transport
inherent internet agent,
capability
the
usingadvantages
4G, it canofanalyze
mobile data
phone are: itand
locally is more flexible
present as itresults
visual has animmediately
inherent internet
aftercapability
receiving using 4G,
data, but
it
it can
hasanalyze
to be indata
the locally and present
communication visual
range ofresults immediately
Bluetooth and the after
user receiving
takes the data, but ittohas
initiative to bea
open
in the communication
specified range ofOn
mobile application. Bluetooth and the
the contrary, user takes the
a customized initiative
agent to open
is plugged a specified
into a powermobile
socket
application.
and communicatesOn the contrary, a customized
to the pillow accordingagent to anisautomatic
plugged into a power
schedule, socket
and andreceived
all the communicates
data is
to the pillowtoaccording
transported the servertotransparently
an automatic schedule,
without and all
analysis the received data
or presentation. is transported
A strong advantage to forthe
an
server
agent transparently
is its ability ofwithout analysis or presentation.
multi-connectivity. If more than A strong
one advantage for an agent
pillow is available in is
theitsBluetooth
ability of
multi-connectivity.
connection range ofIfagent, moreitthan
can one
servepillow
them is allavailable
one-by-one. in the Bluetooth connection range of agent,
it can serve them all one-by-one.
Table 1. Comparison of using mobile phone versus customized transport agent.
Table 1. Comparison of using mobile phone versus customized transport agent.
Items/Devices Mobile Phone Customized Agent
Items/Devices
Data process Mobile Phone
Analysis/passthrough Customized Agent
Passthrough
DataInternet
process access 4G/Wi-Fi
Analysis/passthrough Wire/Wi-Fi
Passthrough
Internet access
Presentation 4G/Wi-Fi
App Wire/Wi-Fi
No screen
Presentation App No screen
Time to connect Any time Scheduled
Time to connect Any time Scheduled
Connect
Connect to pillow
to pillow Manual
Manual Auto Auto
Multi-connection
Multi-connection OneOnepillow
pillow Multi-pillows
Multi-pillows
Accurate timer timer
Accurate Inherent
Inherent Timefrom
Time sync sync server
from server
Flexibility
Flexibility Flexible
Flexible Fixed Fixed

3.2.
3.2. Smart
Smart Pillow
Pillow
The
The block
block diagram
diagram of
of the
the host
host of
of sensors
sensors in
in the
the smart
smart pillow
pillow is
is shown
shown in
inFigure
Figure3.3.

Figure
Figure 3. Hardware interface
3. Hardware interface of
of sensors
sensors in
in smart
smart pillow.
pillow.

The host of pillow sensors is the CSR1001, which is a BLE single mode IC from CSR and
The host of pillow sensors is the CSR1001, which is a BLE single mode IC from CSR and
powered by a cell battery CR2032. Two TMP112s (digital temperature sensors from Texas Instruments),
powered by a cell battery CR2032. Two TMP112s (digital temperature sensors from Texas
and SHT20 (an IC combining humidity and temperature sensors from Sensirion) are connected to the
Instruments), and SHT20 (an IC combining humidity and temperature sensors from Sensirion) are
PIO of CSR1001, so only three temperature sensors and one humidity sensor data will be collected
connected to the PIO of CSR1001, so only three temperature sensors and one humidity sensor data
from the pillow.
will be collected from the pillow.
3.2.1. Physical Structure
3.2.1. Physical Structure
Physical structure of smart pillow is presented as Cartesian coordinates in Figure 4. The pillow is
Physical structure of smart pillow is presented as Cartesian coordinates in Figure 4. The pillow
set in a Cartesian coordinate system, one long edge of pillow is x-axis and one short edge is y-axis,
is set in a Cartesian coordinate system, one long edge of pillow is x-axis and one short edge is y-axis,
so the bottom left corner of pillow is origin (0,0), and upper right corner is (6,3). Sensors are installed in
so the bottom left corner of pillow is origin (0,0), and upper right corner is (6,3). Sensors are installed
the surface of pillow, they are shown in Figure 4 in black, so the first temperature sensor T1 should be
in the surface of pillow, they are shown in Figure 4 in black, so the first temperature sensor T1
installed at (1,1) which is marked as a circle, the third temperature sensor T3 at (5,1), the combination of
should be installed at (1,1) which is marked as a circle, the third temperature sensor T3 at (5,1), the
second T2 and humidity sensor H1 in (3,1), which is marked as a square. They are wired to connectors
combination of second T2 and humidity sensor H1 in (3,1), which is marked as a square. They are
in the BLE whose core is CSR1001. Wires and BLE core are hidden inside pillow, they are shown in
wired to connectors in the BLE whose core is CSR1001. Wires and BLE core are hidden inside pillow,
Figure 4 in gray. A cell battery is exposed on the short edge of pillow, which is wired to BLE too.
they are shown in Figure 4 in gray. A cell battery is exposed on the short edge of pillow, which is
wired to BLE too. The long edge of pillow close to sensors, which y = 0, is the area where the head
Sensors 2018, 18, 3664 7 of 19

Sensors 2018, 18, x FOR PEER REVIEW 7 of 19

The long edge of pillow close to sensors, which y = 0, is the area where the head should always lay on.
should
This always
is the lay on. This
recommended is the recommended
position position
of ideal installation, of ideal installation,
proportional relation of proportional relation of
position of components
position of components should be always
should be always maintained in practice. maintained in practice.

Figure 4. Physical structure of smart pillow as Cartesian coordinates.

Figure 4. Physical Structure of Smart Pillow as Cartesian Coordinates.
3.2.2. Operational Principles
3.2.2.All
Operational
the inferences Principles
are based on change of temperatures and differences between temperatures
T1 toAllT3,the
except
inferences are basedH1
that humidity onischange
designed for sweat detection
of temperatures from the between
and differences structuretemperatures
of pillow in
Figure
T1 to T3, except that humidity H1 is designed for sweat detection from the structure ofmarked
4. The positions of the head-on the pillow are defined as 5 areas, their centers are pillow as
in
P1 to P5,
Figure 4. which are P1(1,1,),
The positions of theP2(2,1),
head-on P3(3,1), P4(4,1),
the pillow areP5 (5,1). The
defined as 5sensor
areas, that
theiriscenters
closestare
to head
markedwhenas
sleeping
P1 to P5, iswhich
defined areas main temperature
P1(1,1,), (MT),
P2(2,1), P3(3,1), and the
P4(4,1), P5sensor
(5,1). Thethat sensor
is furthest
thattoisthe headtoishead
closest defined
whenas
environmental
sleeping is defined temperature
as main (ET), the left temperature
temperature (MT), and the sensor
sensor is defined as auxiliary
that is furthest temperature
to the head is defined(AT);
for instance, if head is on P1(1,1), T1 will be MT and T2 is AT, T3 is ET.
as environmental temperature (ET), the left temperature sensor is defined as auxiliary temperature It is not always the case that
the roles of sensors are clear: for example, when head is on P4(4,1), only the
(AT); for instance, if head is on P1(1,1), T1 will be MT and T2 is AT, T3 is ET. It is not always the case fact that T1 is ET is clear,
not
thatT2theand T3,ofsosensors
roles their roles must be
are clear: forjudged
example, by when
the variety
head of is ontemperature
P4(4,1), only values.
the fact that T1 is ET is
Explanation of the mentioned statistical conclusions is
clear, not T2 and T3, so their roles must be judged by the variety of temperature straightforward. Duration
values.of sleeping is
available if timingofofthe
Explanation thementioned
head-on and head-off
statistical pillow areisdetermined.
conclusions straightforward. Before the point
Duration ofof head-on,
sleeping is
values of three temperature sensors are similar and stable, if any of
available if timing of the head-on and head-off pillow are determined. Before the point of head-on, them increases suddenly and
dramatically,
values of three it istemperature
a sign of head-on.
sensorsHead-off
are similarpoint canstable,
and be judged if any in of
a similar but opposite
them increases way, if and
suddenly any
of the temperature
dramatically, it is a values
sign ofdrops
head-on.andHead-off
all of them become
point can beequaljudged gradually,
in a similarit is but
signopposite
of the head-off.
way, if
For
any of the temperature values drops and all of them become equal gradually, it is asign
room temperature and humidity, ET and H1 without head-on can provide related data; turnofin the
the
pillow is defined as change of head position which can be inferred by
head-off. For room temperature and humidity, ET and H1 without head-on can provide related data; dynamics of temperature data,
and
a turnpositions are described
in the pillow as sleeping
is defined as change position
of headdistribution.
position which These indicators
can be inferred can bebysummarized
dynamics as of
the habit of sleeping. The possible application of collected data is extensive,
temperature data, and positions are described as sleeping position distribution. These indicators as only body temperaturecan
and sweat are extracted
be summarized as significant
as the habit medical
of sleeping. The indicators. There are more
possible application possibilities
of collected data if
is large datasets
extensive, as
of certain area of users in a longer period can be analyzed.
only body temperature and sweat are extracted as significant medical indicators. There are more
possibilities if large datasets of certain area of users in a longer period can be analyzed.
3.2.3. Relationship of Position and Body Temperature

3.2.3.The first symptoms

Relationship of many
of Position anddiseases start with a fever; hence an accurate body temperature is
Body Temperature
meaningful for the end user of the pillow. The position of the head is defined as 5 possibilities, from P1
The first symptoms of many diseases start with a fever; hence an accurate body temperature is
to P5. From this definition, P1 and T1, P3 and T2, P5 and T3 at the same point respectively and P2 is in
meaningful for the end user of the pillow. The position of the head is defined as 5 possibilities, from
middle of T1 and T2, P4 is between T2 and T3. When extracting temperature, P1, P3 and P5 are in a
P1 to P5. From this definition, P1 and T1, P3 and T2, P5 and T3 at the same point respectively and
similar situation that the head is right on the sensor, and P2, P4 is the other group, where the head is
P2 is in middle of T1 and T2, P4 is between T2 and T3. When extracting temperature, P1, P3 and P5
not exactly on a sensor, it is between two sensors. If the following two conditions are fulfilled, a body
are in a similar situation that the head is right on the sensor, and P2, P4 is the other group, where
temperature is extracted:
the head is not exactly on a sensor, it is between two sensors. If the following two conditions are
fulfilled, a body temperature is extracted:
• Head stays in a position, not less than 15 min; and
Sensors 2018, 18, 3664 8 of 19

• Head stays in a position, not less than 15 min; and

Sensors 2018, 18, x FOR PEER REVIEW 8 of 19
• The main temperature (MT) is available for capture.
• The main temperature (MT) is available for capture.
For the first condition, when the head moves to a new position, the nearby temperature sensor
needsFor
a period ofcondition,
the first time for a stable
when value; 15 moves
the head min is an
to experienced
a new position, value
theasnearby
it relates to the material
temperature sensorof
a pillow
needs case, performance
a period of time for aofstable
temperature sensor,
value; 15 min isand
an main temperature
experienced valueisasthe highesttovalue
it relates that can
the material
be afound
of pillowatcase,
the point. The finalofcore
performance body temperature
temperature sensor, and(CBT)
main is temperature
calculated byisEquation (1).value that
the highest
can be found at the point. The final core body temperature (CBT) is calculated by Equation (1).
CBT = Tmain + Tmaterial + Tadjust (1)
CBT = Tmain + Tmaterial + Tadjust (1)
In which
In which T Tmain is the main temperature; Tmaterialisismaterial
main is the main temperature; Tmaterial
materialrelated
relatedvalue,
value,and
and itit is
is aa verified
verified
value by
value by repeated
repeated test
test on
on the
the specific
specific material
material of of pillow
pillow case,
case, as
as the
the head
head does
does not
not directly
directly meet
meet the
the
sensor. For P1, P3 and P5,
sensor. For P1, P3 and P5, only Tmainonly T + T
main+ Tmaterial as CBT is reasonable, but position P is
as CBT is reasonable, but position P is defined as an
material defined as
an area,
area, thethe
headhead is not
is not exactly
exactly onon point,it itjust
point, justisisininan
anarea
area whose
whose center
center is
is the
the point,
point, thus
thusTTadjust is
adjust is
indispensable, it is a compensatory factor. In the case of P2 and P4, not a sensor is
indispensable, it is a compensatory factor. In the case of P2 and P4, not a sensor is close to head as it close to head as
it is
is ininP1P1ororP3,
P3,P5,
P5,sosoTTadjust shouldbebehigher.
should
adjust higher.As Asaaconclusion,
conclusion,position
positionof of head
head and
and extraction
extraction of of
body temperature
body temperature are are related
related closely.
closely. In
In the
thefollowing
followingdiagram,
diagram,inferring
inferringposition
positionandandgetting
gettingTTadjust
adjust
by Fuzzy Logic are described in
by Fuzzy Logic are described in Figure 5. Figure 5.

Figure 5. Data structure of fuzzy

Fuzzylogic
Logicsystem
Systemfor
forP/T
P/Tadjust.
adjust.

Compared to classical logic in which the conclusion is true or false, Fuzzy Logic accepts partial
true values. A A Fuzzy
Fuzzy Logic
Logic System
System (FLS)
(FLS) isis nonlinear
nonlinear system
system that
that maps
maps an an input
input data
data set
set to a scalar
output [39]. Two Two inputs
inputs areareTemperature
TemperatureDifference
Difference(TD), (TD),TD TD
12 12 and
and TDTD 2323 which
which TD
TD 12
12 == T1 −
T1 − T2 and
TD23 23==T2T2−−T3; T3;they
theyarearecalled
calledLinguistic
LinguisticVariables
Variablesand andtheythey will
will bebe fuzzified
fuzzified by their membership
functions,
functions, then then inferences
inferences will will be
be made
made on on aa set of rules, after that all the inference results will be
mapped to membership functions of P/T P/Tadjust ; finally,
adjust ; finally,a avalue
valueofofP/TP/T
adjust is de-fuzzified
adjust is de-fuzzified as output.
as output.
TD12 12 and
and TDTD2323are
areemployed
employedinsteadinsteadof ofT1,
T1,T2,
T2,T3
T3 because
because thethe judgment
judgment of of position
position is based on
differences of of three
threesensors,
sensors,not nottheir
theirvalues
valuesdirectly.
directly.The The membership
membership functions
functions μ(x)
µ(x) for for inputs
inputs are are
the
the
same same
for TDfor12TD
and and23 ,TD
12 TD 23,shown
it is it is shown
in Figurein Figure 6; the membership
6; the membership functionsfunctions
for output for output
P and P and
Tadjust are
T adjust are different,
different, so they are sodiscussed
they are discussed
afterwards. afterwards.
Linguistic
Linguistic terms
terms for TD is Term (td) (td) == {Less,
{Less, Equal,
Equal, More}, for for TD
TD12 12, Less means T1 < < T2 and
More means T1 >> T2 T2 and
and so so forth.
forth. The
The form
form of of membership
membership function
function isis triangular,
triangular, so so for
for each
each term,
term,
in the
the exact middle,−−10,
exactmiddle, 10, 0,0,and
and10,10, each
each membership
membership function
function reaches
reaches its highest
its highest weight,
weight, 1, but1,parts
but
parts
of theof the functions
functions are overlapped:
are overlapped: for instance,
for instance, td = 1tdbelongs
= 1 belongs
to Moreto Moreand Equaland Equal
at theatsame
the same
time,
time,
so both so membership
both membership functions
functions are effective.
are effective. This This
is theisessence
the essence of Fuzzy
of Fuzzy Logic,
Logic, as tdas=td1 =means
1 meansno
no
largelarge difference
difference of two
of two temperatures,
temperatures, so itsoisitreasonable
is reasonable to suit
to suit More More
andand Equal. Equal.
For output P, Linguistic terms Term (P) = {P1, P2, P3, P4, P5} and the output for P is a value in [0,5].
The membership functions for output P is shown in Figure 7.
 Sensors 2018, 18, x FOR PEER REVIEW 9 of 19

Sensors 2018, 18, 3664 9 of 19

Sensors 2018, 18, x FOR PEER REVIEW 9 of 19

Figure 6. Membership functions μ(x) for input TD12 and TD23. TD: temperature difference.

For output P, Linguistic terms Term (P) = {P1, P2, P3, P4, P5} and the output for P is a value in
[0,5]. TheFigure
membership functions
Figure6.6.Membership
for output
Membershipfunctions
functionsµ(x)
P isinput
μ(x)for
shown
forinput TD
inand
Figure
TD12and
7.
TD23. TD: temperature difference.
TD 12 23.

For output P, Linguistic terms Term (P) = {P1, P2, P3, P4, P5} and the output for P is a value in
[0,5]. The membership functions for output P is shown in Figure 7.

Figure 7.
Figure 7. Membership functions
Functionsfor
foroutput
OutputP.P.

The
Theultimate
ultimatejudgement
judgementof ofposition
positionwill
willdepend
dependonon{P1,
{P1,P2,
P2,P3,
P3,P4,
P4,P5}
P5}== {[0,1),
{[0,1), [1,2),
[1,2), [2,3),
[2,3), [3,4),
[3,4),
[4,5)}
[4,5)}totoavoid
avoidany vagueness.
any vagueness.In In
FLS, it depends
FLS, it on Rules
depends on to map
Rules to inputs
map to output,
inputs to the following
output, the rules
following
Figure 7. Membership Functions for Output P.
are applied
rules for output
are applied P in this
for output FLS:
P in this FLS:

1.1. IfThe ultimate

If (TD12
(TD12 judgement
is Less)
is Less) and(TD23
and (TD23 of position
isisLess)
Less)thenwill (P
then depend P5)on
(PisisP5) {P1, P2, P3, P4, P5} = {[0,1), [1,2), [2,3), [3,4),
(0.5)
(0.5)
[4,5)} to avoid
2.2. IfIf (TD12
(TD12 is any
is Less) vagueness.
Less) and
and(TD23 In FLS,
(TD23isisLess) it
Less)then depends
then(P on
(PisisP4) Rules
(0.5) to map inputs to output, the following
P4)(0.5)
rules are applied for output P in this FLS:
3.3. IfIf (TD12
(TD12 is is Less)
Less) and
and(TD23
(TD23isisEqual)Equal)then then(P (PisisP4)P4)(1)(1)
4. If (TD12 is Less) and (TD23
1. If (TD12 is Less) and (TD23 is More)
4. is More)
Less) then then
then(P (P is
(PisisP5) P3) (1)
P3)(0.5)
(1)
5.5.
2. If
If (TD12
(TD12 is
is Equal)
Less) and
and (TD23
(TD23 is is Less)
Less) thenthen
Equal) and (TD23 is Less) then (P is P5) (1) (P (Pis is
P4)P5) (1)
(0.5)
6.
3. If If (TD12
If (TD12
(TD12 is is Equal)
is Equal)
Less) and and (TD23 is More) then (P is P2) (1)
6. and(TD23
(TD23isisEqual)More)then then(P (PisisP4)P2)(1)(1)
7.
4. If If (TD12
If (TD12
(TD12 is is More)
is More)
Less) andand (TD23 is Equal) then (P is P1) (1)
7. and(TD23
(TD23isisMore) Equal)then then(P(PisisP3) P1)(1)(1)
8. If
5. If (TD12
(TD12 is is Equal)
More) and (TD23
(TD23isis
and(TD23 More)
isMore)
Less) then then (PisisP5) P2) (0.5)
8. If (TD12 is More) and then (P(P is P2) (1) (0.5)
6. If (TD12 is Equal) and (TD23 is More) then (P is P1)
9. If (TD12 is More) and (TD23 is More) then (P is P2) (0.5)
(1)
9. If (TD12 is More) and (TD23 is More) then (P is P1) (0.5)
7. If (TD12 is More) and (TD23 is Equal)
The format of the rule is observable, except the number then (P is P1) (1) after position, it is weight and can be
8. If (TD12
The
interpreted asisconfidence
format More)
of the and
ruleor(TD23 is More) except
ispossibility.
observable, then (Pthe is P2) (0.5) after position, it is weight and can be
number
9. If
interpreted(TD12 as is More)
confidence and or(TD23 is
possibility.
The interpretations of those rules are listed: More) then (P is P1) (0.5)
The interpretations of those rules are listed:
• The format
For rule 1 &of2, the
T1 <rule
T2 <isT3,observable,
position could exceptbethe P4number after position,
or P5, head-on somewhereit is weight
betweenand can be
P4 and P5,
•interpreted
For
so the as confidence
ruleconditions
1 & 2, T1 <are or
T2same, possibility.
< T3, position
but conclusions could beare P4different,
or P5, head-on somewhere
and weight between
is 0.5 for each. P4 and P5,
• so The
For interpretations
therule
conditions
3, T1 < T2 are=of
T3,those
same, rules
but
head aremiddle
listed:are
inconclusions
the of T2different,
and T3,and it isweight
on P4. is 0.5 for each.
••• For Forrule
For rule3,14,T1
rule &T12,<<T1
T2<=>T2
T2 T3,<head
T3, head-on in the
T3, position somewhere
middle
could be ofclose
T2 ortoP5,
P4 and P3.head-on
T3, it is on P4.
somewhere between P4 and P5,
• For
so theruleconditions
4, T1 < T2are > T3,
same,head-on somewhereare
but conclusions close to P3. and weight is 0.5 for each.
different,
• For rule 3, T1 < T2 = T3, head in the middle of T2 and T3, it is on P4.
• For rule 4, T1 < T2 > T3, head-on somewhere close to P3.
Sensors 2018, 18, 3664 10 of 19

Sensors 2018, 18, x FOR PEER REVIEW 10 of 19

• For rule 5, T1 = T2 < T3, head-on P5.
•• Forrule
For rule6,5,T1
T1==T2>T3,
T2 < T3,head-on
head-onP2.
P5.
•• For rule 6, T1 = T2>T3, head-on P2.
For rule 7, T1 > T2 = T3, head-on P1.
•• For rule 7, T1 > T2 = T3, head-on P1.
For rule 8 and 9, T1 > T2 > T3, position could be P1 or P2, head-on somewhere between P1 and P2,
• For rule 8 and 9, T1 > T2 > T3, position could be P1 or P2, head-on somewhere between P1 and
so the conditions are same, but conclusions are different, and weight is just 0.5 for each.
P2, so the conditions are same, but conclusions are different, and weight is just 0.5 for each.
There are two other possible conditions in theory, but not in the rules:
There are two other possible conditions in theory, but not in the rules:
•• T1
T1==T2T2== T3,
T3, usually
usually this
this is
is condition
condition ofof empty
empty pillow,
pillow,no nohead-on
head-onso sovalues
valuesof ofall
allsensors
sensorsareare
equal,
equal, but from practical experience, if temperature of room is high enough to close to
but from practical experience, if temperature of room is high enough to close to body
body
temperature,
temperature,itithappens
happensevenevenhead-on
head-onpillow,
pillow,thus
thusextreme
extremeweather
weatherisisnot
notconsidered.
considered.
•• T1
T1 > T2 < T3 is not in the rules because it cannot happen usually, only possibleififuser
> T2 < T3 is not in the rules because it cannot happen usually, only possible userjust
justturns
turns
around from one side of pillow to the other side, but the data from pillow is based
around from one side of pillow to the other side, but the data from pillow is based on interval on interval of
5ofmin, so momentary changes do not appear, and CBT is only extracted when
5 min, so momentary changes do not appear, and CBT is only extracted when head position head position is
stable for no less than 15 min, so the option is improbable.
is stable for no less than 15 min, so the option is improbable.
For
Foroutput
outputTadjust, Linguistic terms
Tadjust, Linguistic terms Term
Term (Tadjust)
(Tadjust) = ={Lower,
{Lower,Proper,
Proper, Higher}
Higher} and the
the adjustable
adjustable
range for Tadjust is [–2,2]. The terms
range for Tadjust is [–2,2]. The terms areare self-explanatory, for Lower the output
Lower the output of T of T adjust will
adjust will be
be
negative,
negative,and
andpositive
positiveforforHigher.
Higher.Membership
Membershipfunction
functionfor foroutput
outputTTadjust
adjust is shown in Figure 8.

Figure8.8.Membership
Figure Membershipfunctions
Functionsfor
foroutput
OutputTaTdjust.
adjust.

Rules
Rulesdesigned
designedfor
foroutput
outputTTadjust in this
adjust in this FLS are as stated:

1.1. If (TD12
If (TD12 isis Less)
Less) and
and (TD23
(TD23 is is Less)
Less) then
then (adjust
(adjustisisHigher)
Higher)(1)(1)
2.2. If (TD12 is More) and (TD23 is Equal) then (adjust
If (TD12 is More) and (TD23 is Equal) then (adjust is Proper) (1) is Proper) (1)
3.3. If (TD12
If (TD12 isis Equal)
Equal) and
and (TD23
(TD23 is is More)
More) then
then (adjust
(adjustisisProper)
Proper)(1)
(1)
4.4. If (TD12 is Less) and (TD23 is Equal) then (adjust
If (TD12 is Less) and (TD23 is Equal) then (adjust is Proper) (1)is Proper) (1)
5. If (TD12 is Less) and (TD23 is More) then (adjust is Higher) (1)
5. If (TD12 is Less) and (TD23 is More) then (adjust is Higher) (1)
6. If (TD12 is Equal) and (TD23 is Less) then (adjust is Proper) (1)
6. If (TD12 is Equal) and (TD23 is Less) then (adjust is Proper) (1)
7. If (TD12 is More) and (TD23 is More) then (adjust is Higher) (1)
7. If (TD12 is More) and (TD23 is More) then (adjust is Higher) (1)
8. If (TD12 is Equal) and (TD23 is Equal) then (adjust is Lower) (1)
8. If (TD12 is Equal) and (TD23 is Equal) then (adjust is Lower) (1)
The principle for the rules is that if the head is on exactly defined positions, the CBT is proper
The principle
so adjustment is for
notthe rules is that
necessary, if the head
otherwise, is on exactly
adjustment defined
should be positions, the CBTthat
positive, except is proper so
all three
adjustment is not necessary, otherwise, adjustment should be positive, except that
values of temperature are similar, which is mentioned before, heat cannot disappear in air as the all three values of
temperature
environmental are temperature
similar, which is is mentioned
close to CBT,before, heat cannot should
so the adjustment disappear in air as the environmental
be lower.
temperature is close to CBT, so the adjustment should be lower.
The rules are self-explanatory when presented in Table 2, in which all positions are observed in
The rules are self-explanatory when presented in Table 2, in which all positions are observed
parenthesizes.
in parenthesizes.
 Table 2. Matrix of Rules for Fuzzy Logic System.

TD12/TD23 Less Equal More

Less Higher (P4/P5) Proper (P4) Higher (P3)
Equal Proper (P5) Lower Proper (P2)
Sensors 2018, 18, 3664 More None Proper (P1) Higher (P1/P2) 11 of 19

In all rules stated above, “and” is used to combine TD12 and TD23, which still means an
Table
intersection in traditional logic, 2. Matrix
except thatofitrules for fuzzy logic
is calculated system. (x), μ
by Min{μ TD12 TD23(x)}, which means

the smaller valueTD12/TD23

is employed. Less Equal More
The last step ofLess FLS is de-fuzzification,
Higher (P4/P5) which is finding
Proper (P4) a value for P/T(P3)
Higher adjust in this case. By

input rules, the appropriate

Equal membership
Proper function
(P5) and the Lowercorresponding smaller
Proper (P2)weight are found.
Then they are mapped Moreto membership None
function of output, Proper (P1)
cut parts Higher (P1/P2) function and
of membership
form shapes, and the output value will be the center of gravity of the accumulated shape, it can be
calculated
In all by Equation
rules stated (2).above, “and” is used to combine TD12 and TD23, which still means an
intersection in traditional logic, except that it is calculated by Min{µTD12 (x), µTD23 (x)}, which means
P/Tadjust = tμ(t)dt μ(t)dt (2)
the smaller value is employed.
In which
The μ()of
last step is FLS
the is
membership function
de-fuzzification, which of output
is finding after accumulation,
a value for P/Tadjust it is
incomputable
this case. Byas it is
input
combination of some triangles or trapezoids; t is value of P or T , for
rules, the appropriate membership function and the corresponding smaller weight are found. Then they
adjust P min is 0 and max is 5, and
minmapped
are is −2 and to max is 2 for Tfunction
membership adjust in this case. The
of output, cutfollowing Figure 9 is afunction
parts of membership visual process
and form of the FLS
shapes,
when
and theinput
outputof [TD12, TD23]
value will be is
the[−2.37,
center 3.18] for Tadjust
of gravity of. the accumulated shape, it can be calculated by
Rule(2).
Equation 1, 6 are not met in this case as TD23 is not Less, neither rule 2, 7 as TD12 is not More, but
Z max Z max
both TD12 and TD23 are EqualP/T as well according tµto
(t)their
dt/ membership µ(t)dt functions, so rule 3, 4, 5, 8 are
adjust = (2)
all met. For rule 3, μTD12(−2.37) = 0.474 and minμ
TD23 (3.18) = 0.318,
min so 0.318 is transferred to output
In which making
membership, µ() is thethe membership function of
proper membership outputμafter
function accumulation, it is computable as it is
Tadjust(Proper) = 0.318. Explanations of other
combination of some triangles or trapezoids; t is value
rules are similar, all the left shapes of rules accumulate to the final of P or Tadjust
shape,, forthen
P min is 0isand
Tadjust max isby
calculated 5,
and min is − 2 and max is 2 for
Equation (2), which is the gravitationalT adjust in this case. The following Figure 9 is a visual
center of the accumulated shape and marked with the red process of the FLS
when input of
line in Figure 9. [TD12, TD23] is [ − 2.37, 3.18] for Tadjust .

Figure9.9.Visual
Figure Visualprocess
processofofinput
input[−
[−2.37, 3.18]in
2.37, 3.18] inFLS
FLSfor
foroutput
outputTTadjust.
adjust.

Rule 1, 6 are not met in this case as TD23 is not Less, neither rule 2, 7 as TD12 is not More, but both
TD12 and TD23 are Equal as well according to their membership functions, so rule 3, 4, 5, 8 are
all met. For rule 3, µTD12(−2.37) = 0.474 and µTD23(3.18) = 0.318, so 0.318 is transferred to output
membership, making the proper membership function µTadjust (Proper) = 0.318. Explanations of other
rules are similar, all the left shapes of rules accumulate to the final shape, then Tadjust is calculated by
Equation (2), which is the gravitational center of the accumulated shape and marked with the red line
in Figure 9.
 x FOR PEER REVIEW
Sensors 2018, 18, 3664 12
12 of
of 19

Every time if a [T1, T2, T3] meets the standard of CBT, a [TD12, TD23] is calculated and fed to
Every time if P/T
the FLS to output a [T1, T2, T3] meets the standard of CBT, a [TD12, TD23] is calculated and fed to the
adjust, then the real CBT is calculated by Equation (1).
FLS to output P/Tadjust , then the real CBT is calculated by Equation (1).
4. Results
4. Results
In this section, prototype of experimental trial pillow is presented first, then a daily sample is
In this
provided andsection, prototype
interpreted, of experimental
the sleeping patterns trial
and pillow
reportsisarepresented
describedfirst,
in thethen a daily sample is
end.
provided and interpreted,
The prototype of thethe sleeping
trial pillow patterns and reports
is displayed in Figureare 10,
described
which in is the
a latexend.pillow with the
The prototype of the trial pillow is displayed in Figure 10, which
advantages of comfort, support and easy maintenance. The master chip of BLE is CSR1001, is a latex pillow with
whichthe is
advantages
a Bluetooth smart device from CSR, as it is hidden inside the middle of pillow and connected to a3
of comfort, support and easy maintenance. The master chip of BLE is CSR1001, which is
Bluetooth
sensors and smart device
a cell from CSR,
battery. Threeasred
it is circles
hidden at inside the middle
bottom of pillow
of Figure and connected
10 indicate to 3 sensors
where sensors are
and
located. The left and right are TMP112s, which are digital temperature sensors from TI. The The
a cell battery. Three red circles at bottom of Figure 10 indicate where sensors are located. left
sensors
and right to
are fixed aretheTMP112s,
surface which
of the are digital
pillow andtemperature
their surfacesensors from
is coated TI. The
with a whitesensors are fixedglue
waterproof to the
to
surface
prevent sweat from affecting their performance. The middle one is SHT20, which is a humidityfrom
of the pillow and their surface is coated with a white waterproof glue to prevent sweat and
affecting
temperature theirsensor
performance. The middle
from Sensirion. one is is
Its surface SHT20,
coatedwhich
with is a humidity
white waterproofand temperature
glue, except forsensor
the
from Sensirion. Its surface is coated with white waterproof glue, except for the
sensing window of humidity sensor, hence a small black point is visible from the figure. Inside the sensing window of
humidity sensor, hence a small black point is visible from the figure. Inside the left red
left red circle is a battery case for cell battery CR2032. Users are required to sleep on the side with the circle is a battery
case for cell battery
pre-positioned CR2032. Users are required to sleep on the side with the pre-positioned sensors.
sensors.

Figure 10. Prototype of experimental trial pillow.

To maintain
maintain the accuracy of the CBT, CBT, extensive
extensive experiments
experiments were carried out during the
development
development stages.stages.TheTheexperimental
experimental scenario applies
scenario 2 sets of
applies 2 sensing
sets of kits whichkits
sensing is thewhich
combination
is the
of
combination of a CSR1001 connected with 3 sensors and a cell battery, shown in Figure 3. Oneinside
a CSR1001 connected with 3 sensors and a cell battery, shown in Figure 3. One kit is installed kit is
the pillow,inside
installed shown in pillow,
the Figure 10, whileinthe
shown other 10,
Figure is used
while to the
measure
otherthe real body
is used temperature
to measure by fixing
the real body
one of the temperature
temperature by fixing one sensors
of theintemperature
tester’s armpit. While
sensors testing, armpit.
in tester’s the tester’s
Whilehead is onthe
testing, the tester’s
pillow
with the fixed temperature sensor under the armpit simultaneously; he/she
head is on the pillow with the fixed temperature sensor under the armpit simultaneously; he/she then sleeps on a position
steadily
then sleepsfor on
15 amin, then steadily
position moves tofor a different
15 min, then position
moves ontothe pillow. After
a different testing,
position on the data from After
pillow. both
kits are extracted,
testing, data fromCBT bothinferred
kits are by extracted,
the data from CBTthe pillow is
inferred bythen
the compared
data fromtothe the pillow
direct result
is thenTr
from the armpit − Trto ◦
≤ 0.2 the
compared to thesensor. Then the
direct result algorithm
Tr from is adjusted
the armpit sensor. to Then
meet the standard
algorithmofis|CBT
adjusted | meet C.
Experiments
standard of |CBT were performed
− Tr| ≤ 0.2under an ambient environmental
°C. Experiments were performed temperature
under an of 15 ◦ C to 28 ◦ C.
aroundenvironmental
ambient
The measurement
temperature of around 15 of °C to body
real 28 °C.temperature is shown in Figure 11. T3 is employed to detect
the body temperature, of as real
its value ◦ C as soon as it is kept under the armpit but
The measurement body jumps to nearis 36
temperature shown in Figure 11. T3 is employed to detect the
needs another 10–20asmin
body temperature, to be close
its value jumps totothenear
real36 body
°C as temperature.
soon as it isInkeptour under
experiments,
the armpitthe maximum
but needs
temperature value will be used as a reference for the pillow’s algorithm.
another 10–20 min to be close to the real body temperature. In our experiments, the maximum In this way, the reference
body temperature
temperature value is measured
will be used at as the same time,
a reference by the
for the samealgorithm.
pillow’s type of device and
In this way,under the same
the reference
environmental
body temperature conditions.
is measuredHowever,at theit same
is noted thatbythe
time, the measurements
same type ofare takenand
device in different
under the parts
sameof
the body, of which
environmental it will beHowever,
conditions. discusseditfurther
is noted in the
thatconclusion.
the measurements are taken in different parts
of the body, of which it will be discussed further in the conclusion.
Sensors 2018, 18, x FOR PEER REVIEW 13 of 19
Sensors 2018, 18, 3664 13 of 19
Sensors 2018, 18, x FOR PEER REVIEW 13 of 19

Figure 11. Measurement of Real Body Temperature.

Figure 11. Measurement of real body temperature.
Figure 11. Measurement of Real Body Temperature.
4.1. Interpretation
4.1. of Whole
Interpretation of Whole Night
Night Data
Data
4.1. Interpretation of Whole Night Data
Data of
Data of aa typical
typical night
night sleep
sleep is
is presented
presented in
in Figure
Figure 12.
12.
Data of a typical night sleep is presented in Figure 12.

12. Sample
Figure 12. Sampledatadataofof sleep
sleep with
with (a) (a) humidity
humidity H ofHa night,
of a night,
(b) T1,(b)
T2,T1,
andT2,
T3and T3 of a(c)night,
of a night, (c)
head-on
Figure 12.
head-on
positions ofSample
a night.ofdata
positions of sleep with (a) humidity H of a night, (b) T1, T2, and T3 of a night, (c)
a night.
head-on positions of a night.
Horizontal axis of the whole Figure 12 shows the time range (from 23:00 to 8:00 of next day),
and the vertical axis of (a)
Horizontal the shows
whole humidity
Figure 12 in percentage
shows the time (%), at the
range beginning
(from 23:00 toit 8:00
is stable at about
of next day),
about
70%, from
and the about axis
vertical 00:08,ofit(a)
increases
shows sharply
increases sharply up
humidity upinto 94%.
94%. It means
topercentage (%),head-on pillow from
at the beginning it isthat time,
stable atthen
aboutit
changes a lot,
70%, from up 00:08,
about and down, as it implies
it increases sharplythat
upuser rollsIton
to 94%. the pillow,
means pillow,
head-on when head
pillow
when from
head is on ortime,
that closethen
to the
it
changes a lot, up and down, as it implies that user rolls on the pillow, when head is on or close to the
Sensors 2018, 18,
Sensors 2018, 18, 3664
x FOR PEER REVIEW 14of
14 of 19
19

humidity sensor, it is up, otherwise down. At the end of the figure, it gradually returns to 65% in the
humidity sensor, it is up, otherwise down. At the end of the figure, it gradually returns to 65% in the
morning after the user wakes up and leaves from the pillow.
morning after the user wakes up and leaves from the pillow.
Vertical axis of (b) is temperatures in Celsius◦ (°C) of three temperature sensors. First, the blue
Vertical axis of (b) is temperatures in Celsius ( C) of three temperature sensors. First, the blue line,
line, which is T2, it changes almost synchronously in time with H in (a) although trends are not
which is T2, it changes almost synchronously in time with H in (a) although trends are not uniform all
uniform all the time, because they are integrated in the same chip SHT20. For convenience of
the time, because they are integrated in the same chip SHT20. For convenience of comparison, T1 is
comparison, T1 is drawn in green and T3 in red. Like H at the beginning, all three values are 28 °C
drawn in green and T3 in red. Like H at the beginning, all three values are 28 ◦ C before head-on pillow,
before head-on pillow, it is the room temperature, then values develop respectively when the user’s
it is the room temperature, then values develop respectively when the user’s head is on and rolls on
head is on and rolls on the pillow. Roles of temperature alternate with changes of head ◦position.
the pillow. Roles of temperature alternate with changes of head position. Values return to 28 C in the
Values return to 28 °C in the morning after the user wakes up.
morning after the user wakes up.
Vertical axis of (c) is head-on position, which is from 1 to 5, representing P1 to P5. According to
Vertical axis of (c) is head-on position, which is from 1 to 5, representing P1 to P5. According to
the designed FIS, position of the head depends on differences of three temperature sensors: at the
the designed FIS, position of the head depends on differences of three temperature sensors: at the
beginning when T2 increases sharply and faster than other sensors, it proves head-on T2/P3 which is
beginning when T2 increases sharply and faster than other sensors, it proves head-on T2/P3 which is
in the middle of pillow. As P3 is the middle of pillow, position which is higher than 3 means head-on
in the middle of pillow. As P3 is the middle of pillow, position which is higher than 3 means head-on
right side of pillow, lower means left. From (c) of the figure, the user is used to sleeping on the right
right side of pillow, lower means left. From (c) of the figure, the user is used to sleeping on the right
side of pillow as positions of P4 and P5 is much more than P2. Number of turns can be counted from
side of pillow as positions of P4 and P5 is much more than P2. Number of turns can be counted from
(c) if each change of position is treated as one turn.
(c) if each change of position is treated as one turn.
The standard for extraction of body temperature is that main temperature lasts no less than 15
The standard for extraction of body temperature is that main temperature lasts no less than 15 min.
min. From (b) of the figure, at least three points are available, they are T3 at about 01:40, T3 at 03:10,
From (b) of the figure, at least three points are available, they are T3 at about 01:40, T3 at 03:10, and T2
and T2 at 05:15, they all meet the standard. Details will be described in the next figure.
at 05:15, they all meet the standard. Details will be described in the next figure.
4.2.
4.2. Details
Details of
of Typical Sleep Patterns
Typical Sleep Patterns
To describe more
To describe more details
detailsof
ofsleep,
sleep,which
whichincludes
includesextraction
extractionofofbody
bodytemperature,
temperature,two
twohours
hoursofofa
anight
night
is is presented
presented inin Figure
Figure 13.13.

Figure 13. Extraction of body temperature data.

As the
As the interval
interval ofof data
data from
from pillow
pillow is
is 55 min,
min, time
time of
of data
data in
in the
the figure
figure is
is from
from 00:03
00:03 to
to 01:53,
01:53,
every 5
every min. Three
5 min. Threeparts
partsofoffigure
figure are
are the
the same
same asas Figure
Figure 12,
12, except
except more
more details
details in
in time.
time. AtAt the
the
beginning of the figure, before 00:08, values of all three temperatures are similar and
beginning of the figure, before 00:08, values of all three temperatures are similar and humidityhumidity is stable.
is
It means no head is on the pillow, then suddenly T2 and H increase sharply and T1,
stable. It means no head is on the pillow, then suddenly T2 and H increase sharply and T1, T3 in T3 in different
degrees of
different growth,
degrees of because user goes
growth, because to goes
user bed and head-on
to bed somewhere
and head-on between
somewhere P3 andP3P4and
between as P4
T3 as
is
T3 is faster than T1 in growth, then user moves slightly from somewhere between P3 and P4 to
Sensors 2018, 18, 3664 15 of 19

faster than T1 in growth, then user moves slightly from somewhere between P3 and P4 to somewhere
between P2 and P3 as T1 on the rise and T3 drops from 00:13. Then at 00:23 it is confirmed that head-on
P2 as T2 drops but T1 and T3 are stable similarly. There is some confusion of the next data as the
changes of values are not obvious, they are all close. But 5 min later at 00:33, T1 and T2 are up similarly
and T3 drops which reclaims head-on P2. In the following 20 min, trend keeps stable, but it does not
reach P1 as T1 and T2 keeps stable. After that the user turns, from left to right, so position changes
from P2 to P3, P4 and finally P5 as T3 becomes leader and T1 and T2 drops synchronously. This lasts
30 min, then user turns slightly back to left, in P4 with T2 on the rise and T3 drops to meet T2 at 01:43.

4.3. Daily Report

From Figure 13, T1 at 00:53 and T3 at 01:38 meet the standard of body temperature,
the 01:38 lasts longer, its values [T1, T2, T3] = [27.8, 28.0, 34.4], so [TD12, TD23] = [−0.2, −6.4],
they are fed to designed FIS as input, output Tadjust = 0.0826, so CBT = 34.4 + 2 + 0.08 = 36.48 based on
Equation (1). This is a typical position in P5, where T1 = T2 < T3, so the adjustment is small.
The other obvious body temperature point is found at 05:10 of Figure 12, it is a typical P2,
T1 = T2 > T3, [T1, T2, T3] = [33.7, 34.5.0, 27.9], [TD12, TD23] = [−0.8, 6.6], Tadjust = 0.258
from FIS, it is greater than the last P5 adjustment, which is in accordance with theoretical analysis,
so CBT = 34.5 + 2 + 0.258 = 36.76. Those two values are reasonable body temperatures.
Only one humidity sensor is installed in the middle of pillow, when the head is near P3,
including P2, P3, and P4, H reflects the sweat of the head, otherwise it is just humidity of room.
The observation of (a) and (c) proves this, when head is on P3, the humidity is high, if the head is far
from P3, humidity drops to a local minimum that is the room humidity. At the beginning of Figure 12,
humidity is up to 94%, which implies that user goes to bed without drying his/her hair, which Chinese
medicine opposes strongly, but half an hour later, humidity is stable at about 84%, which means the
user is in a condition of slight sweat. This will be included in the daily report.
As a conclusion, Table 3 is derived from the daily data as a report.

Table 3. Daily report for sample data.

Item Value
Bedtime 00:08
Get up 07:08
Sleep time 7h
Body Temperature 36.5/36.8
Sweat 94%
Turns 37
Left/Middle/Right 14/30/40

Bedtime is obvious from Figure 12, as all the values increase suddenly from flat as head-on pillow.
Time of getting up is vague from Figure 12, usually the time that all values decrease or stabilizes and
gradually gets close to room temperature is selected. In Figure 12, it is 07:08, so the total sleep time
is available too. All the qualified body temperatures will be listed in report, in this case, the highest
humidity is 94%, it will be reported that user sleeps without drying hair, and slightly sweats at night.
Turns in the report is counted on position changes, it is not real turns of sleeping as data from
the pillow with an interval of 5 min. It means actions inside 5 min are ignored, but under the same
standard, so the data is still meaningful for comparison.
The area of positions demonstrates which part of pillow user clings to during sleep. This can
provide valuable reference as concerns arise on the relation of quality and sleep position.
Sensors 2018, 18, 3664 16 of 19

5. Conclusions
The paper presents a smart pillow based on 3 temperature sensors and a humidity sensor, which is
the core element of the Health Sensing System. Specifically, a report is provided daily by analyzing time
series data from the pillow, which reflects the sleep status of a user faithfully like a camera, but it offers
extra information such as body temperature and sweat. The daily summary is presented to user visually
in a website or mobile application, numbers of one day are not of much significance, but comparison of
daily data is more of value. From the weekly or monthly sleep data, the user can observe the dynamic
status of sleep day-by-day: improvement, worse, fever, sweat, then take appropriate actions initially to
concentrate on his/her sleep quality, such as sleeping earlier, or consulting professional help. This is
just the comparison of an individual—if statistical data of all users can be analyzed, users can find
their sleep levels in overall. Once pertinent questions can be asked by an individual when comparing
with a larger dataset, this will ultimately help the patient improve their sleep quality and health.
An obvious issue of the pillow is that data is only reported every 5 min. All the details inside
the 5 min are lost. First, this is a compromise for the energy efficiency of BLE. As described before,
the core of pillow, which is a CSR1001 BLE chip, is powered by the cell battery CR2032. Compared
to ubiquitous lithium battery, it has lower energy capacity and is non-rechargeable, but it is safer,
smaller, and easier to be accepted by the user of pillow. The other question is how long the appropriate
interval is. From the viewpoint of provided report, accuracy of turns is affected seriously by interval,
all other items are less dependent. For Bedtime and Getting Up, the largest possible error is 5 min,
which is trivial compared with total sleep time. A larger interval means less data and lower power
consumption, so therefore longer battery life. In our extensive experiments, a cell battery can maintain
normal function for 3 months of general use. It is acceptable and reasonable for the user to change the
battery about once a season.
To the best of my knowledge, our smart pillow is the only mass-produced and temperature-related
smart pillow on the market, the other two temperature-related pillows are neck pillow [15] and
cushion [36], they are just tested in labs. Comparison of them are presented in Table 4.

Table 4. Comparison of three temperature-related pillows. CBT: core body temperature; ET:
environmental temperature.

Name Accuracy Comfortability Energy Efficiency Cost Purpose

Neck pillow High Low Low High CBT
Cushion High Mild Low High ET
Smart pillow Mild High High Low CBT

Accuracy of neck pillow and cushion is higher as the former employing DHF and Double Sensors,
the latter installing sensor on the outside for just ET not CBT, the cost of both are higher as they use
more sensors, as a result their energy consumption are higher too, and sleeping on a cushion is more
comfortable than with neck pillow.
From massive experiments of our pillow, the inferred normal CBT is very closed to normal adult
temperature based on the proposed Fuzzy Logic system. It is not compared with body temperature
measured by clinical thermometer as body temperature is deeply related to time, position, way and
instrument of measurement. So the accuracy of CBT is less weighted as change of CBT. As a commercial
product, comfortability and energy efficiency is our advantages.

6. Further Work
Additionally, the smart pillow could employ energy scavenging techniques typically applied for
wireless sensor networks and the Internet of Things. While it would make the pillow more complex to
manufacture, the advantage is that the pillow would have more energy reserve to support additional
sensors, or last longer between battery replacements. Pressure-based or heat-differential energy capture
Sensors 2018, 18, 3664 17 of 19

methods can be employed to add to the total energy reserve of the pillow. Therefore, a balance can
be established to make a logical trade-off between utility and comfort, so that a more comprehensive
pillow with a longer energy life can be made to support more sensors for specific patients in mind, or a
simpler constructed pillow for general use.
There are various ways to measure body temperature clinically, and accuracy is different. Accuracy
of body temperature obtained from the smart pillow is closely related to environment temperature
and the material of the pillowcase according to our experiments. The essential problem is that heat
dissipation is limited by them. If environment temperature is high enough to be close to body
temperature, it will be difficult for heat to escape from the pillow, and as a result, the value of
temperature sensor changes slowly, which is harmful to our algorithm as it deeply depends on
difference of temperature. Some materials of pillowcase also cause obstacles to heat dissipation,
so usually the smart pillow is provided with a case and other cases are not recommended, then the
value of Tmaterial is accurate in our algorithm, but with massive data, an appropriate Tmaterial + Tadjust
could be learned based on normal body temperature.
Further work needs to be done to obtain core body temperature from the pillow and a reference
point that is as close as possible to both source points. The current experiments collected the body
temperature from a different source point to the pillow, due to uncomfortable nature of placing a
thermometer near the user’s head while it is also on the smart pillow. Special techniques will need
to be developed to place the thermometer in such a way to reduce the incidence of user discomfort,
otherwise the results will be affected by excessive head turns or other unanticipated body movements.
As a health sensing system, body temperature and sweat measurement are only the start of
development; if some other clinical indicators such as blood pressure and heart rate can be integrated
into the pillow or system it will make the system more useful. Theoretically, it is possible to consider a
photo-plethysmogram (PPG) [40,41], as it is extensively applied in smart devices for measurements,
but it is an optical technique that requires a camera and light source such as an LED flash to
operate effectively. Integrating more components will make the pillow more complex to manufacture,
and worsen comfort and increase battery consumption, so adding a smart device with blood pressure
and heart rate detector could be an optimal decision.
The prospect of big-data on sleep and health is promising evolution to the path of connected
home medicine. Not only can it discover hidden disease in advance in a non-intrusive, natural,
and unconscious way, it can open new developments to determine medical trends among patients for
better medical diagnosis—hence the necessity for the design of a smart pillow.

Author Contributions: S.L. designed and developed the concept, implemented the simulation parts, analyzed the
data, and documented the results. C.C. provided suggestions and critical comments during the development of
this paper. Both authors made substantial contributions in the writing and revising of the paper.
Funding: This research received no external funding. The APC was funded by Guangdong College of Business
and Technology.
Conflicts of Interest: The authors declare no conflict of interest.

References
1. Wright, K.P.; Hull, J.T.; Czeisler, C.A. Relationship between alertness, performance, and body temperature in
humans. Am. J. Physiol. Regul. Integr. Comp. Physiol. 2002, 283, R1370–R1377. [CrossRef] [PubMed]
2. Lack, L.C.; Gradisar, M.; Someren, E.J.V.; Wright, H.R.; Lushington, K. The relationship between insomnia
and body temperatures. Sleep Med. Rev. 2008, 12, 307–317. [CrossRef] [PubMed]
3. Coyne, M.D.; Kesick, C.M.; Doherty, T.J.; Kolka, M.A.; Stephenson, L.A. Circadian rhythm changes in core
temperature over the menstrual cycle: Method for noninvasive monitoring. Am. J. Physiol. Regul. Integr.
Comp. Physiol. 2000, 279, R1316–R1320. [CrossRef] [PubMed]
4. Buller, M.J.; Tharion, W.J.; Cheuvront, S.N.; Montain, S.J.; Kenefick, R.W.; Castellani, J.; Latzka, W.A.;
Roberts, W.S.; Richter, M.; Jenkins, O.C.; et al. Estimation of human core temperature from sequential heart
rate observations. Physiol. Meas. 2013, 34, 781–798. [CrossRef] [PubMed]
Sensors 2018, 18, 3664 18 of 19

5. Fox, R.H.; Solman, A.J. A new technique for monitoring the deep body temperature in man from the intact
skin surface. J. Physiol. 1971, 212, 8–10.
6. Fox, R.H.; Solman, A.J.; Isaacs, R.; Fry, A.J.; MacDonald, I.C. A new method for monitoring deep body
temperature from the skin surface. Clin. Sci. 1973, 44, 81–86. [CrossRef] [PubMed]
7. Teunissen, L.P.J.; Klewer, J.; Haan, A.; Koning, J.J.; Daanen, H.A.M. Non-invasive continuous core
temperature measurement by zero heat flux. Physiol. Meas. 2011, 32, 559. [CrossRef] [PubMed]
8. Zeiner, A.; Klewer, J.; Sterz, F.; Haugk, M.; Krizanac, D.; Testori, C.; Losert, H.; Ayati, S.; Holzer, M.
Non-invasive continuous cerebral temperature monitoring in patients treated with mild therapeutic
hypothermia: An observational pilot study. Resuscitation 2010, 81, 861–866. [CrossRef] [PubMed]
9. Kitamura, K.; Zhu, X.; Chen, W.; Nemoto, T. Development of a new method for the noninvasive measurement.
Med. Eng. Phys. 2010, 32, 1–6. [CrossRef] [PubMed]
10. Huang, M.; Chen, W.; Kitamura, K.I.; Nemoto, T.; Tamura, T. Improvement of the dual-heat-flux method for
deep body temperature measurement based on a finite element model. IEEE EMBC 2013, 13, 1202–1205.
11. Huang, M.; Tamura, T.; Tang, Z.; Chen, W.; Kanaya, S. Structural optimization of a wearable deep body
thermometer: From theoretical simulation to experimental verification. J. Sens. 2016, 2016, 4828093.
[CrossRef]
12. Gunga, H.C.; Sandsund, M.; Reinertsen, R.E.; Sattler, F.; Koch, J. A noninvasive device to continuously
determine heat strain in humans. J. Therm. Biol. 2008, 33, 297–307. [CrossRef]
13. Gunga, H.C.; Werner, A.; Stahn, A. The Double Sensor—A non-invasive device to continuously monitor
core temperature in humans on earth and in space. Respir. Physiol. Neurobiol. 2009, 169, S63–S68. [CrossRef]
[PubMed]
14. Kimberger, O.; Thell, R.; Schuh, M.; Koch, J.; Sessler, D.I.; Kurz, A. Accuracy and precision of a novel
non-invasive core thermometer. Br. J. Anaesth. 2009, 103, 226–231. [CrossRef] [PubMed]
15. Sim, S.Y.; Lee, W.K.; Baek, H.J.; Park, K.S. A nonintrusive temperature measuring system for estimating
deep body temperature in bed. In Proceedings of the 2012 Annual International Conference of the
IEEE Engineering in Medicine and Biology Society, San Diego, CA, USA, 28 August–1 September 2012;
pp. 3460–3463.
16. Husain, M.D.; Kennon, R. Preliminary Investigations into the Development of Textile Based Temperature
Sensor for Healthcare Applications. Fibers 2013, 1, 2–10. [CrossRef]
17. Majumder, S.; Mondal, T.; Deen, M.J. Wearable Sensors for Remote Health Monitoring. Sensors 2017, 17, 130.
[CrossRef] [PubMed]
18. Honda, W.; Harada, S.; Arie, T.; Akita, S.; Takei, K. Printed wearable temperature sensor for health monitoring.
In SENSORS; IEEE: Piscataway, NJ, USA, 2014; pp. 2227–2229.
19. Takei, K.; Honda, W.; Harada, S.; Arie, T.; Akita, S. Toward Flexible and Wearable Human-Interactive
Health-Monitoring Devices. Adv. Healthc. Mater. 2015, 4, 487–500. [CrossRef] [PubMed]
20. Miah, M.A.; Kabir, M.H.; Tanveer, M.S.R.; Akhand, M.A.H. Continuous heart rate and body temperature
monitoring system using Arduino UNO and android device. In Proceedings of the International Conference
on Electrical Information and Communication Technologies, Khulna, Bangladesh, 10–12 December 2015.
21. Song, C.; Zeng, P.; Wang, Z.; Zhao, H.; Yu, H. Wearable Continuous Body Temperature Measurement Using
Multiple Artificial Neural Networks. In IEEE Transactions on Industrial Informatics; IEEE: Piscataway, NJ,
USA, 2018. [CrossRef]
22. Boano, C.A.; Lasagni, M.; Romer, K.; Lange, T. Accurate Temperature Measurements for Medical
Research Using Body Sensor Networks. In Proceedings of the 2011 14th IEEE International
Symposium on Object/Component/Service-Oriented Real-Time Distributed Computing Workshops,
Newport Beach, CA, USA, 28–31 March 2011; pp. 189–198.
23. Ullah, S.; Higgins, H.; Braem, B.; Latre, B.; Blondia, C.; Moerman, I.; Saleem, S.; Rahman, Z.; Kwak, K.
A Comprehensive Survey of Wireless Body Area Networks. J. Med. Syst. 2012, 36, 1065–1094. [CrossRef]
[PubMed]
24. Dobrescu, R.; Popescu, D.; Dobrescu, M.; Nicolae, M. Intergration of WSN-based platform in a homecare
monitoring system. In Latest Trends on Communications and Information Technology; WSEAS Press: Greece,
2008; pp. 165–170.
25. Farahani, S. ZigBee Wireless Networks and Transceivers; Newnes: Burlington, VT, USA, 2008; pp. 1–24.
ISBN 978-0-7506-8393-7.
Sensors 2018, 18, 3664 19 of 19

26. Salman, N.; Rasool, I.; Kemp, A.H. Overview of the IEEE 802.15.4 standards family for Low Rate Wireless
Personal Area Networks. In Proceedings of the 7th International Symposium in Wireless Communication
Systems (ISWCS), York, UK, 12–22 September 2010.
27. Arnil, J.; Punsawad, Y.; Wongsawat, Y. Wireless sensor network-based smart room system for healthcare
monitoring. In Proceedings of the 2011 IEEE International Conference on Robotics and Biomimetics,
Phuket Island, Thailand, 7–11 December 2011; pp. 2073–2076.
28. Wenqi, G.; Healy, W.M.; Mengchu, Z. An experimental study of interference impacts on ZigBee-based
wireless communication inside buildings. In Proceedings of the International Conference Mechatronics and
Automation (ICMA), Xi’an, China, 4–7 August 2010.
29. Mansor, H.; Shukor, M.H.A.; Meskam, S.S.; Rusli, N.Q.A.M.; Zamery, N.S. Body temperature measurement
for remote health monitoring system. In Proceedings of the 2013 IEEE International Conference on Smart
Instrumentation, Measurement and Applications (ICSIMA), Kuala Lumpur, Malaysia, 26–27 November 2013;
pp. 1–5.
30. Kiourmars, A.H.; Tang, L. Wireless Network for Haealth Monitoring Heart Rate and Temperature Sensor.
In Proceedings of the Fifth International Conference on Sensing Technology, Palmerston North, New Zealand,
28 November–1 December 2011; pp. 362–368.
31. Guo, D.G.; Tay, F.E.; Xu, L.; Yu, L.M.; Nyan, M.N.; Chong, F.W.; Yap, K.L.; Xu, B. A Long-term Wearable Vital
Signs Monitoring System using BSN. In Proceedings of the 2008 11th EUROMICRO Conference on Digital
System Design Architectures, Methods and Tools, Parma, Italy, 3–5 September 2008; pp. 825–830. [CrossRef]
32. Malhi, K.; Mukhopadhyay, S.C.; Schnepper, J.; Haefke, M.; Ewald, H. A Zigbee-Based Wearable Physiological
Parameters Monitoring System. IEEE Sens. J. 2012, 12, 423–430. [CrossRef]
33. Laine, T.H.; Lee, C.; Suk, H. Mobile Gateway for Ubiquitous Health Care System Using ZigBee and Bluetooth.
In Proceedings of the 2014 Eighth International Conference on Innovative Mobile and Internet Services in
Ubiquitous Computing, Birmingham, UK, 2–4 July 2014; pp. 139–145. [CrossRef]
34. Doukas, C.; Maglogiannis, I. Bringing IoT and Cloud Computing towards Pervasive Healthcare.
In Proceedings of the 2012 Sixth International Conference on Innovative Mobile and Internet Services
in Ubiquitous Computing, Palermo, Italy, 4–6 July 2012; pp. 922–926. [CrossRef]
35. Lloret, J.; Parra, L.; Taha, M.; Tomas, J. An architecture and protocol for smart continuous eHealth monitoring
using 5G. Comput. Netw. 2017, 129, 340–351. [CrossRef]
36. Veiga, A.; Garcia, L.; Parra, L.; Lloret, J.; Augele, V. An IoT-based smart pillow for sleep quality monitoring
in AAL environments. In Proceedings of the Third International Conference on Fog and Mobile Edge
Computing (FMEC), Barcelona, Spain, 23–26 April 2018; pp. 175–180. [CrossRef]
37. Li, X.; Liu, Y.; Chen, P.; Wu, J.; Zhang, H. Novel type of unperturbed sleep monitoring scheme under pillow
based on hidden Markov model. J. Biomed. Eng. 2018, 35, 280–289. [CrossRef]
38. Narendra, P.; Duquennoy, S.; Voigt, T. BLE and IEEE 802.15.4 in the IoT: Evaluation and Interoperability
Considerations. In Internet of Things. IoT Infrastructures. IoT360 2015. Lecture Notes of the Institute for Computer
Sciences, Social Informatics and Telecommunications Engineering; Mandler, B., Ed.; Springer: Cham, Switzerland;
Volume 170.
39. Dernoncourt, F. Introduction to Fuzzy Logic; MIT: Cambridge, MA, USA, 2013.
40. McKay, N.; Griffiths, B.; Di Maria, C.; Hedley, S.; Murray, A.; Allen, J. Novel photoplethysmography
cardiovascular assessments in patients with Raynaud’s phenomenon and systemic sclerosis: A pilot study.
Rheumatology (Oxford). 2014, 53, 1855–1863. [CrossRef] [PubMed]
41. Li, K.; Warren, S. A Wireless Reflectance Pulse Oximeter with Digital Baseline Control for Unfiltered
Photoplethysmograms. IEEE Trans. Biomed. Circuits Syst. 2012, 6, 269–278. [CrossRef] [PubMed]

© 2018 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC BY) license (http://creativecommons.org/licenses/by/4.0/).