Content uploaded by Antti Vehkaoja
Author content
All content in this area was uploaded by Antti Vehkaoja on May 06, 2015
Content may be subject to copyright.
Copyright (c) 2009 IEEE. Personal use is permitted. For any other purposes, Permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
TITB-00064-2009
1
Wireless, multi-purpose in-home health
monitoring platform: two case trials
Sakari Junnila, Harri Kailanto, Juho Merilahti, Antti-Matti Vainio, Antti Vehkaoja, Mari Zakrzewski,
and Jari Hyttinen
Abstract— We propose a general purpose home area sensor
network and monitoring platform that is intended for eHealth
applications, ranging from elderly monitoring to early
homecoming after a hospitalization period. Our monitoring
platform is multi-purpose, meaning that the system is easily
configurable for various user needs and is easy to set up. The
system could be temporarily rented from a service company by,
for example, hospitals, elderly service providers, specialized
physiological rehabilitation centers, or individuals. Our system
consists of a chosen set of sensors, a wireless sensor network, a
home client, and a distant server. We evaluated our concept in two
initial trials: one with an elderly woman living in sheltered
housing, and the other with a hip surgery patient during his
rehabilitation phase. The results prove the functionality of the
platform. However, efficient utilization of such platforms requires
further work on the actual eHealth service concepts.
Index Terms— eHealth, health monitoring, smart home,
wireless sensor network
I. INTRODUCTION
hronic diseases, especially in developed countries, have a
significant influence on health care costs and are common
among elderly people. Changes in demographic structure and
lack of health and social care personnel force us to study new
innovations, which could offer a relief to these challenges.
Developments such as increased emphasis on
self-management of diseases, decreased costs of medical
technologies, and new study results have added the potential of
personal health systems to be a solution to the aforementioned
challenges. People with chronic conditions such as
hypertension, diabetes, or sleep apnea, are already using
technologies daily as a part of their treatment, and more
solutions are coming to market all the time.
Manuscript received February 15, 2009, revised July 17, 2009, accepted
November 14, 2009. This work was supported in part by the Finnish Funding
Agency for Technology and Innovation (Tekes) under Grant 40108 and in part
by Tampere Graduate School in Information Science and Engineering.
S. Junnila in with the Department of Signal Processing, Tampere University
of Technology (TUT), P.O.Box 553, FIN-33101 Tampere, Finland (email:
sakari.junnila@tut.fi).
H. Kailanto and J. Hyttinen are with the Department of Biomedical
Engineering, TUT (email: harri.kailanto@tut.fi, jari.hyttinen@tut.fi).
A.-M. Vainio and M. Zakrzewski are with the Department of Electronics,
TUT (email: antti-matti.vainio@tut.fi, mari.zakrzewski@tut.fi).
A. Vehkaoja is with the Department of Automation Science and Engineering
at TUT (phone: +358 40 739 3181, email: antti.vehkaoja@tut.fi).
J. Merilahti is with the VTT Technical Research Centre of Finland, P.O.Box
1300, Tampere, Finland (email: juho.merilahti@vtt.fi).
Various interest groups would benefit from a platform to
which they could connect different devices used to measure
one’s health. This information could be utilized for example in
new health services. This kind of a platform should be easily
modified for different purposes and patients.
Both before and during our study, many new and interesting
technologies for home health monitoring have become
available, and new studies have been published in this area. For
example, in 2008, Intel released Health Guide PHS6000 [1],
which is a home-end unit designed especially for older users.
The unit has a touch screen and can be connected to wireless
and wired vital sign monitoring devices. The Philips Telehealth
solution [2] enables a user to fill in surveys and measure his/her
vital signs, which are then delivered to a health care
professional. The Health Buddy® appliance from Health Hero
includes similar features to Philips’s and Intel’s solutions
(www.healthhero.com).
A large number of research papers have recently been
published on this topic from different perspectives. Quach et al.
[3] investigated the possibility of interfacing different wireless
technologies to form a complete wireless telehealth system.
They used IEEE 802.15.4/ZigBee radios as sensor network
technology and WiFi to transfer the measurement data from the
network coordinator to a database server. Yao and Warren [4]
investigated the application of ISO/IEEE 11073 standards to
wearable health monitoring systems, and demonstrated the
concept using Bluetooth-enabled radios. They concluded that
the aforementioned standards that define the nomenclature and
communication protocol of the medical data, also known as
X73, can be applied to home health monitoring with minor
modifications.
However, there is still a need to study the topic further. For
example, Koch reported the importance of developing better
wireless tools and sensors together with better design for users
like older people [5]. Varshney emphasized that further studies
are needed to develop more reliable wireless networks,
adaptable solutions for diverse users, and security of data
transmission [6]. Scanaill et al. concluded that it is important to
study the possibilities of monitoring old people’s mobility in
order to determine the user’s health status more accurately and
in this way, support independent living [7]. Overall, home
telehealth is shown to decrease health care costs and to improve
the patient’s quality of life [8, 9].
C
Authorized licensed use limited to: Tampereen Teknillinen Korkeakoulu. Downloaded on February 11, 2010 at 16:03 from IEEE Xplore. Restrictions apply.
IEEE owns the copyright for this publication, doi: 10.1109/TITB.2009.2037615
Copyright (c) 2009 IEEE. Personal use is permitted. For any other purposes, Permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
TITB-00064-2009
2
Most of the published studies that focus on developing and
evaluating technology have one specific target group, most
commonly the elderly living at home. What differentiates our
study from the others is that we have focused on the problem of
adaptability, aiming at a solution that would accommodate
different types of end user needs and scenarios. We emphasized
wireless technologies for easy installation, thus enabling future
users to utilize the technology without having to move from
their current residence. In this paper, we propose an
easy-to-install home health monitoring platform that can be
used for various applications, including monitoring dementia
patients, monitoring elderly people’s overall health, and
short-term monitoring of rehabilitation patients after e.g. a
prosthesis surgery. Different commercial measurement/
monitoring products were interfaced to our platform, together
with some custom-made prototype sensors. This paper presents
two case trials, in which the platform has been utilized in real
life. This we consider essential when developing and
evaluating home health technologies. The platform name,
UUTE, is an abbreviation from the Finnish name of our project,
and from now on this abbreviation is used in this document.
II. RESEARCH METHODOLOGY
During the study, our objective has been to prototype the
system by creating a concrete experience of the system we are
designing. In the design research field, this methodology is
called experience prototyping, meaning a form of prototyping
that “enables design team members, users, and clients to gain
first-hand appreciation of existing or future conditions through
active engagement with prototypes” [10]. Thus, we attempt to
give both the users, and especially the developers, a concrete
experience of what it would be like to live at home, equipped
with a sensor network. In addition, we hope that with a
hands-on experience, we can gain more insight into what the
good design choices are and what parts need to be discussed
further, while designing such a system for commercial use.
At the beginning of the project, a scenario study to evaluate
different service scenarios was conducted through focus group
discussions. Amongst seven scenarios, the most desired one
(home monitoring) included both indoor and outdoor
positioning of the user for safety and security purposes. In
addition, the system in the scenario senses and alarms abnormal
behavior indoors. Moreover, older adults and health care
professionals valued scenarios including features for
self-management of diseases at home (self-care). Details of the
assessment study are reported in [11].
The scenario work served as a basis for our prototyping
study. The complete system involves several levels of
engineering work. The main emphasis of the study was on the
adaptable wireless sensor network (WSN), along with the
developed sensors. Thus, other important areas, such as the
security of the data transmission, personalization of the
measurement data, final data storage formats, and service
concepts, are only briefly discussed. This means that we are
studying a prototype that does not yet include all the parts
needed for a complete application.
Operational tests, in which members of the research group
spent a day in the test apartment trying to find things to
improve, were conducted prior to the actual trials. During the
installation and operational tests, we collected practical
observations about the setup into a notebook. When studying
home technology, we find it highly important to evaluate the
system in a real home environment, in contrast to laboratory-
like conditions. Thus, both of the case trials were conducted in
the private homes of the test subjects. During the case trials,
acceptability and user experience were evaluated by interviews
with the users and the personnel of health care center.
Our research objective was twofold: we wanted to study the
experience both from the developer’s and user’s point of view.
From the developer’s viewpoint, the research objectives were:
1) to demonstrate the proof-of-concept and validate it in a
real environment,
2) to consider which technical solutions were most
suitable for a platform providing a general solution for
home health monitoring,
3) to determine the problems of such a system and
drawbacks of our technical choices in everyday life, and
4) to determine issues that need to be studied further.
From the user’s viewpoint, the research objectives were how
such a generic platform would suit the diverse needs of
different user groups and environments.
This paper is mainly targeted at developers of such a
measurement system, including researchers and companies in
the fields of home health monitoring, wireless sensor networks,
smart home applications, etc. The paper should be read in a
qualitative manner, as the results were derived from the initial
operational test and two case trials.
III. DEVELOPED MONITORING PLATFORM
An important part of our monitoring platform is the WSN for
sensor data communication. This chapter introduces the
technical details of the network, its components, the home
client, and the sensors we have implemented so far.
A. Network architecture
1) Selection of the network technology
Various wireless technologies for building lightweight
WSNs exist. Our selection was made between Bluetooth and
Zigbee, which are open standards. Bluetooth enables direct
connection with a variety of user interface devices, such as
mobile phones. Its use as a sensor network technology has
limitations in respect to the network configuration and power
consumption. The latter issue has recently been addressed in
the new Bluetooth low-energy specification. The benefits of the
Zigbee technology were low power consumption, relatively
simple structure, network topology management, and large
network size. We chose Zigbee because support for larger
networks was desired and because of its potential for lower
power consumption. We also had some prior expertise with it.
2) Wireless sensor network implementation
Zigbee uses the IEEE 802.15.4 standard to define the
Authorized licensed use limited to: Tampereen Teknillinen Korkeakoulu. Downloaded on February 11, 2010 at 16:03 from IEEE Xplore. Restrictions apply.
Copyright (c) 2009 IEEE. Personal use is permitted. For any other purposes, Permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
TITB-00064-2009
3
physical and data link-layers. The Zigbee standard itself
defines network and higher layer functionality. To design a
Zigbee system, one needs an IEEE 804.15.4-compliant radio
platform on top of which Zigbee functionality is implemented,
usually in software.
At the time the project was started, the Chipcon CC2420 was
one of the most popular IEEE 802.15.4 radios. It implements
the IEEE 802.15.4 physical layer and some data link layer
functionality on the hardware, but requires an accompanying
microcontroller to execute the rest of the software-implemented
data link layer. The data link software or MAC-software
(medium access control), was targeted for Atmel AVR
microcontrollers. We therefore implemented our sensor radio
nodes using ChipCon CC2420 RF-transceivers and Atmel
Atmega 128 microcontrollers.
Fig. 1. Sensor node HW/SW architecture with three main hardware
components: the CC2420 radio chip, the microcontroller unit, and the sensor
hardware. The sensor hardware is usually a separate device or printed circuit
board (PCB), and the microcontroller and radio are located on the same physical
PCB. There are two distinct software stacks, the radio communication stack and
the sensor communication stack, which work under the node’s main
application.
Fig. 2. Sensor node states. When powered on, a sensor node looks for Zigbee
networks with a defined PAN ID, and if it finds one, it connects to it. If the
Zigbee connection is established, the sensor node attempts to move from an
“Undefined” state to a “Configured” state. If the device does not find a Zigbee
network when powered on, or if it notices that the network has gone down
during operation, it deconfigures itself and periodically attempts to reconnect
and reconfigure to the network.
No free Zigbee implementation was available in mid-2006,
so we implemented our own Zigbee network layer based on the
Zigbee 2004 standard, which was the only standard publicly
available at that time [12]. Our implementation did not include
the optional Zigbee security functions or the higher layer
Zigbee structures. The network-layer stack, as also the
underlying MAC-stack, is interrupt driven. The sensor nodes’
main application code communicates with the Zigbee network
layer directly (Fig. 1).
3) Sensor interface
A so-called “common sensor interface” architecture was
developed [13] to separate the radio technology from the
measurement hardware. The idea behind this was that sensor
hardware could be designed and built to be used with different
radio platforms. This requires a defined physical interface to
connect the sensor hardware and radio part and software
architecture with defined function interfaces. In our software
implementation, each sensor comes with its own sensor driver,
which is controlled by the sensor node application (Fig. 1). The
sensor driver can use functions from a defined library of
interface drivers to talk to the sensor hardware via the
radio/microcontroller part’s I/O ports and peripherals.
We have designed the system architecture so that most types
of new sensors can fairly easily be introduced to the network.
To introduce a new sensor, the sensor developer defines a new
sensor type/class code for the sensor and the data structure that
the sensor uses for transmission. This information is used by
the home client administrator to generate a new sensor model
for parsing the data from this sensor class. The sensor
developer then writes the sensor driver using a framework code
and pre-defined software (SW) interface functions for physical
I/O (SPI, UART, or direct I/O drivers) and network access.
The sensor node’s main application glues the Zigbee
network and the sensory data together. This application
initializes and configures the Zigbee network connection,
initializes the sensor drivers, and relays and processes data
moving between the sensor drivers and Zigbee network (Fig.
1). The network initialization is shown in Fig. 2.
4) Network realization
In our test studies, we used a star type network topology, i.e.
the sensor nodes were directly connected to the network
coordinator. The network range can be extended by adding
router devices. In our small test apartment, the sensor nodes
were able to connect directly to the coordinator from anywhere
within the apartment, so no routers were required. The network
does not support beaconing, as this is not supported in the
licensed MAC-software. This limits the power efficiency,
performance, and optimization possibilities of the network. The
non-beacon network used in our implementation does not
achieve as high throughput as a beacon-enabled network would
under interference. On the other hand, the effect of interference
on the performance is smaller for non-beacon networks [14].
The network master node (network coordinator) is connected to
(and powered by) the UUTE home client.
Because of the limited data bandwidth and potentially large
amount of nodes in the network, the allowed amount of
transmitted data per sensor is restricted. To reduce the amount
of transmitted data, some sensors need to perform data
processing locally before sending (see Table I). The raw
bit-rate of the Zigbee-radio is 250 kbps, but practical
non-optimized implementations using the 2003 version of the
IEEE 802.15.4 MAC have been found to reach 46 kbps
throughput in optimal single hop conditions [15]. In a
non-interfered point-to-point radio link, speeds of up to 125
kbps have been reached, but as has been shown in [16], the
throughput in single channel systems decreases noticeably in
multi-hop networks. We have tested our network
implementations performance in [12], in which issues limiting
Authorized licensed use limited to: Tampereen Teknillinen Korkeakoulu. Downloaded on February 11, 2010 at 16:03 from IEEE Xplore. Restrictions apply.
Copyright (c) 2009 IEEE. Personal use is permitted. For any other purposes, Permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
TITB-00064-2009
4
the throughput are discussed in detail. The total network
capacity and operation performance in larger multi-device
network configurations is yet to be tested.
5) UUTE home client and UUTE server
The UUTE home client is a laptop PC running a Java
application. It decodes the incoming data into a readable
format, according to the device type. Based on the device type,
sensors may also require some additional signal processing
before the data is saved into the hard drive for a backup and
sent forward. Depending on the application, the required
post-processing of the sensor data can also be done at the
UUTE home client. If the system is used for dementia or fall
patient monitoring, the desired acute alarms can be generated.
Or, if the system is used, for example, in the monitoring of
independent rehabilitation, there may be more need to derive
indices related to the long-term progress of the rehabilitation
process.
The UUTE home client provides two levels of limits for
generating alarms; recommendation limits and alarm limits. If a
sensor value would cross the notification limit, a notification
could be sent to for example the user interface. Crossing the
alarm limit would trigger an alarm message for the medical
personnel, for example. These alarms were not implemented in
the case studies, as they were out of scope of the experiment.
The UUTE home client also operates as a gateway to the
Internet and to a UUTE server located elsewhere. The home
client is connected to the server through a TCP socket. The
client is connected to the Internet via a mobile broadband
network or other available means. All the communication with
the server is encrypted with the SSL protocol. After the server
receives the data, it is saved into a Microsoft SQL Express
database. The data is then viewed through dynamically
generated web pages on a Microsoft IIS server or with other
kinds of user interfaces. An example of the Java application
providing such an interface is shown in Fig. 3.
B. Realized test sensors
We have developed for or made compatible with our
monitoring platform different types of sensors. The
self-developed custom sensors include a heart rate (HR) sensor,
an infrared (IR) sensor, and an intelligent pedometer. The
other, commercial sensors that were fitted to operate in our
platform were a blood pressure monitor (LifeSource
UA-767PC, A&D Company Ltd., www.aandd.jp), a weight
scale (KERN DE 300K100N, Kern & Sohn GmbH,
www.kern-sohn.com), and an electromechanical film- (EMFi)
based bed sensor (by Emfit Ltd, www.emfit.com). All of the
fitted commercial sensors had RS-232 interfaces. In addition, a
prototype floor sensor, based on capacitive measurement, was
tested along with the rest of the system.
Most of the sensor data are processed at each sensor node in
order to derive relevant information from the data and to
minimize the amount of the transmitted data.
1) Intelligent pedometer
An intelligent pedometer was specified for measuring the
activity and analyzing the increased fall risk of the user. Our
design uses the signal from a 3D accelerometer for determining
parameters related to the user's activity. The parameters are: the
level of activity, taken steps, average strength of the steps,
RMS amplitude of the acceleration signal, standard deviation
of the acceleration signal, and the largest signal value in the
given time window. The selection of the latter three was based
on a study conducted in the project, which suggests that
features calculated from accelerometer data during walking
could be used to identify people who suffer from degenerated
balance and therefore, an increased risk of falling [17].
Fig. 3. User interface for viewing the data stored at the UUTE server. The view
shows the latest measurements, graphs for one week of weight, blood pressure,
and sleep time measurements, and limit values for notifications sent to
healthcare professionals. These measurements were recorded by the patient
during the second trial presented in this paper.
The parameters are calculated from every 10-second period
of the acceleration signal and then sent once per minute to the
UUTE home client in a predefined package. As the intelligent
pedometer is also designed to be used outside the apartment, it
is equipped with an external memory, where the parameters are
saved even when the network is not available. When the device
is brought back into the range of home network, the stored data
is sent to the UUTE home client.
2) Heart rate sensor
Our heart rate sensor uses a one-channel ECG, measured
with commercial electrodes, to calculate the HR. We are using
a very simple, but fairly robust, QRS-recognition algorithm in
which QRS-complexes are recognized by a simple descending
slope detector, which adapts to the measured ECG signal
amplitude. Five R-R intervals are gathered to the memory and
transmitted in one packet to save energy.
3) Bed sensor
The EMFi bed sensor measures basic physiology such as
HR, movements, and respiration rate during the bed time. The
bed sensor is installed below the user’s mattress. The sensor
calculates different minute-to-minute features, which are sent
to the UUTE home client. During the tests, time spend in bed,
average HR, average breathing frequency, and average activity
were calculated for each night by the home client and sent to the
health database. The time spent in bed was calculated from a
minute-to-minute bed occupancy data between 8 pm and 10
Authorized licensed use limited to: Tampereen Teknillinen Korkeakoulu. Downloaded on February 11, 2010 at 16:03 from IEEE Xplore. Restrictions apply.
Copyright (c) 2009 IEEE. Personal use is permitted. For any other purposes, Permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
TITB-00064-2009
5
Fi
g
4. Senso
r
setu
p
s for the two trials
,
and the information
p
rovided b
y
the sensors.
am.
4) Infrared sensor
An infrared (IR) sensor was developed to detect the presence
of a person or use of heat-producing appliances, such as a stove
or coffee maker. The sensor array consists of seven sensors
made of Polyvinylidene Fluoride (PVDF) and can be directed
towards the area of interest. The incoming IR radiation is
interrupted by a slowly rotating, perforated disc. The amplitude
of the generated signal is used to calculate the temperature
difference between the incoming IR radiation and the sensor.
5) The capacitive floor sensor
The sensor consists of a thin, laminated mat installed under
the floor mat and sensor electronics developed by
UPM-Kymmene Corp. (www.upmpresence.com). The sensor
electronics were hidden within baseboards. The measurement
is based on the change of stray capacitance between two
electrode tiles (size 305 x 445 mm) caused by movements of a
conductive material near the laminate mat. Since the human
body is fairly conductive, the technology measures the
presence of the person itself. The advantage of this method is
that one does not need to carry any kind of tag to be positioned.
On the other hand, this makes the identification of a person and
the tracking of two or more closely situated persons extremely
challenging. However, in single-person households, such as in
our test cases, this multiple person separation is not a major
problem.
The data from the sensor elements is collected with an eBox
mini PC (www.gadgetcomputer.com). As the signal processing
methods for the accurate positioning were still under
development, they were not implemented into the eBox
computer. Instead, all the measurement data were sent to the
UUTE home client and signal processing was done afterwards.
Because of the huge amount of raw data to be sent, the WSN
was not used. Instead, the data were transferred via wired
ethernet connection to the UUTE home client.
IV. TRIALS WITH THE DEVELOPED CONCEPT
We evaluated the developed platform in two real-life trials.
The first trial was done in a recently constructed, sheltered
housing apartment. This housing solution is designed for older
people and it offers various nursing and assisted living services.
The second trial targeted a hip surgery patient during his
rehabilitation period and was carried out in co-operation with a
health care center. The second trial study was approved by the
Ethics Committee of Pirkanmaa Hospital District.
To emphasize the simple installation of the system to
differently equipped apartments, the Internet connection was
achieved using a mobile broadband network. Yet, because the
power consumption of the system was not fully optimized, we
chose to use mains current as the energy source in order to
prevent the test person from changing or recharging the
batteries during the usage period, even though the sensors were
enabled for battery use. Table I summarizes the sensors that
were included in each trial.
A. The first trial: Elderly monitoring at home
In the first trial, we constructed a system somewhat similar to
the desired scenarios (home monitoring and self-care) of the
assessment study of [11]. However, the outdoor positioning
functionality was not realized due to the need for some kind of
a tag (such as a GPS-enabled mobile phone) carried by the
participant.
As a test environment, we used a two-room apartment with a
toilet and a kitchen. The system installation (Fig. 4) included
the weight scale, the blood pressure sensor, the bed sensor, the
IR sensor, and the capacitive floor sensor. The placement of the
sensors and the coverage of the floor sensor are shown in Fig 5.
We affixed the IR sensor to the kitchen ceiling where it is able
to monitor, for example, when the stove or the coffee machine
is turned on. The UUTE home client was placed in the footage
of the closet in the bedroom so that it would not disrupt the test
subject. The test environment was installed and tested in the
apartment during the summer and fall of 2008.
The test person was a 70-year-old female living alone. Two
test phases were conducted, lasting six weeks in total. The first
one included all the sensors and lasted four weeks. After the
first phase, we were not satisfied with the outcome of the floor
sensor; thus, a second two-week test phase was conducted,
including only the floor sensor. After the tests, the test person
continued living in the apartment.
The test person was instructed to perform her normal daily
routines during the study period. In addition, the person got a
light exercise plan including calisthenics at home.
B. The second trial: A hip surgery rehabilitation patient
The hip surgery patient participated into the study during his
first rehabilitation period two weeks after the hip surgery. This
period is a normal part of the rehabilitation process of a hip
surgery patient in the health care. The system (Fig. 4) included
a diary (item relevant to the rehabilitation process), health
monitoring sensors (blood pressure monitor, weight scale, and
Authorized licensed use limited to: Tampereen Teknillinen Korkeakoulu. Downloaded on February 11, 2010 at 16:03 from IEEE Xplore. Restrictions apply.
Copyright (c) 2009 IEEE. Personal use is permitted. For any other purposes, Permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
TITB-00064-2009
6
bed sensor), and a video call environment (ArctiCare
Technologies, Kemijarvi, Finland, www.arcticare.com). The
diary was used for self-assessment of sleep length and quality,
daily activities, rehabilitation program activity, and pain
severity on a visual analogue scale (VAS, from 0 to 10).
Moreover, medication intake and open comments were
instructed to be written down.
The test person sent the diary markings to a researcher
weekly via regular mail. Morning blood pressure, morning
weight, and bed sensor information were collected during the
study using the developed platform. The subject used a
computer with a touch screen display once a week to make a
video call to a physiotherapist to follow up on the progress of
his rehabilitation. The physiotherapist initiated the call.
Normally the patient and the physiotherapist are not necessary
in contact during this period. A summary from the
measurements and the diary markings were sent weekly to the
patient and to the physiotherapist, who observed the data for
possible problems in the rehabilitation.
The test person was a 72-year-old man living with his wife.
The study phase lasted for one month.
V. RESULTS
A. The radio network
Fig. 5. The location of the inhabitant during a five-minute time window. The
filled dot represents the latest location. The area covered with the floor sensor is
shown in light gray. The location is measured using the floor localization
sensor.
The first version of the Zigbee-based sensor network,
presented in [13], worked well as long as the network
coordinator was not rebooted. However, rebooting caused
problems: the Zigbee network stack in the coordinator notices
the sensor nodes when they try to send something, and
automatically sets up the Zigbee connection. However, the
coordinator, which starts from a “blank” network state, gives
the first connecting node the first available Zigbee
short-address. This address can be, and usually is, previously
assigned to some other sensor node. After this, we end up with
two configured devices in the Zigbee network with the same
short-address, which leads to confusion in the coordinator
when sensor data is received. In our case, it leads to a
never-ending reconfiguration loop, as the coordinator
application was receiving two types (sensor type) of messages
from the same short-address, and after noticing this, asked the
node to reconfigure. The reconfiguration request is received by
both of the two nodes, and thus a never-ending loop is formed.
To solve this problem, we modified the sensor state diagram;
the sensor will leave the configured state if it loses network
connection (Fig. 2). The network was deemed lost if a packet
transmission failed after a certain amount of retransmissions
where a “NO_NWK” return status was received from the
Zigbee stack. When leaving the configured state, the sensor
node sends a leave-message to the network and disassociates
from the network by dropping the given short-address. After
this modification, the network was able to recover from any
device’s reboot, including the coordinator.
B. Case trial findings from a technology point of view
In total, the system performed without any major problems
during the trials. The data were successfully transmitted from
the sensors to the home client. The latest measurements from
the second trial, as well as graphs of measured data, were
reviewed through the user interface presented in Fig. 2.
However, breaks in the wireless Internet connection during
the first trial caused the server transmission to fail occasionally.
TABLE I
SUMMARY OF THE TRIALS AND SENSORS USED
The 1st trial: elderly monitoring
The 1st phase The 2nd phase
The 2nd trial: hip surgery
patient
Sensors Data
processing Used
sensors User’s compliance Used sensors Used
sensors
User’s
compliance
Heart rate sensor L, S, P - - - - -
Infrared sensor L, S, P x - - - -
Intelligent pedometer L, S, P - - - - -
Blood pressure monitor S, P x 67 % (20/30) - x 96 % (27/28)
Weight scale S, P x 83 % (25/30) - x 83 % (25/30)
Bed sensor L, S, P x 25 %
(removed after 7 days) - x 100 %
Floor sensor L, S x - x - -
Video call to a physiotherapist P - - - x 67 % (2/3)
Abbreviations: L = data processed locally in a sensor node, S = data stored locally in the UUTE client, and P = data passed on to the UUTE
server. None of the sensor nodes included memory to store the data in a sensor node. x means the sensor was used in the corresponding trial,
- meaning that it was not used. The user’s compliance is calculated from the sensors that require some active participation of the user. The
compliance percentage was compared to what was instructed to the user in the beginning of the trial.
Authorized licensed use limited to: Tampereen Teknillinen Korkeakoulu. Downloaded on February 11, 2010 at 16:03 from IEEE Xplore. Restrictions apply.
Copyright (c) 2009 IEEE. Personal use is permitted. For any other purposes, Permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
TITB-00064-2009
7
As the data was stored locally in the UUTE home client, it was
not lost. Similar problems did not occur during the second trial
because at this time, we used a mobile broadband modem with
a reconnection feature. The video call feature was used weekly
between the patient and the physiotherapist. In addition, we had
some problems with the bandwidth of the mobile network. The
voice is not prioritized in the system, which caused some
problems, particularly in regards to the voice quality. Both the
patient and the physiotherapist reported that the voice quality
was more important than the video.
The network-compatible blood pressure monitor
malfunctioned at the beginning of the second trial, so it was
replaced with a regular one (Omron 705IT). The user wrote
down the blood pressure monitor results in the diary. He
continued this routine after the new network-compatible blood
pressure monitor was installed in the system, which enabled us
to review the reliability of the network.
Fig. 6. Self-reported and bed sensor sleep times during the second trial (one
outlier from the bed sensor is visible in the graph).
As for the capacitive floor sensor, part of the wiring
apparently broke during the installation, so not all the sensor
elements could be used. The floor sensor system that was a
non-production version also had some reliability issues, which
caused breaks in the recorded data. This complicated the
analysis of the test subject’s location.
C. Case trial findings from an application point of view
The user’s compliance with the measurements was also
rather high. The results are summarized in Table I. The blood
pressure monitor was absent during the first two days of the
second trial, which explains the two-day difference between the
number of instructed weight and blood pressure measurements.
The location of the user during a five-minute period is
plotted on the floor plan in Fig. 5. This kind of information
could be used by a relative to look the status of an elderly
inhabitant, or to check whether the inhabitant has already gone
to sleep and if it is too late to call him/her. Another scenario
where this kind of information would be useful is the case
where an automatic alarm has been raised and the inhabitant
cannot be reached. In such a case, an emergency or health care
unit heading to the apartment could check what had happened
in the apartment to set off the alarm and prepare for what they
might be facing.
Self-reported sleep time and bed sensor sleep time
correspond reasonably well; collected during the second trial.
One sensor-measured value was removed as an outlier (the
difference was over five hours), as seen in Fig. 6. Compared to
the numbers reported by the user, the bed sensor overestimates
sleep time on average by four minutes (SD = 21 minutes),
which is logical because the sensor detects bed occupancy time
rather than the actual sleep time.
However, in the first trial, the user reported that she was
uncomfortable with the bed sensor, as it was placed at such a
close proximity to her body for relatively long times each day.
Due to this anxiety, the user wanted the bed sensor to be
removed after using it for seven days.
VI. DISCUSSION AND FUTURE WORK
A. The radio network
Texas Instruments (TI) bought ChipCon, the manufacturer
of CC2420 radio chip. This ended the support for the IEEE
802.15.4 MAC-software (CCMAC) targeted for Atmel
microcontrollers. Some technical issues with the old MAC
were never resolved, including the lack of support for
beaconing. The newer MAC (TIMAC) versions are designed
for TI’s MSP430 microcontrollers, and the MAC has been
completely rewritten. It also now supports beaconing, but only
with CC2430, not with CC2420. TI also now provides a free
Zigbee implementation that supports the newest Zigbee
standard version, which is not compatible with the older
versions. This new software cannot be used with the sensor
radio boards built for the project, as they have the Atmel
Atmega 128 microcontroller, not TI’s MSP430. Moreover, as
the old Zigbee 2004 standard is not compatible with the newer
versions, it is not possible to mix our sensors and newer Zigbee
radios based on CC2420 and TI’s newer stacks. Future versions
of the developed system should therefore have redesigned
sensor radio boards using the microcontroller TI MSP430 to
benefit from the fully functional and compliant Zigbee stack. In
addition, if two-way radio communication and high-power
efficiency in the sensor nodes is desired, the radio chip should
be changed to CC2430, as the MAC and Zigbee software
provided for the latter support beacon-enabled networks.
The operation of the network under interference has not been
extensively tested or studied. We have used it in a normal office
environment with wireless local-area networks (WLANs)
present and in use. As a narrow-band single channel system, the
Zigbee radio-link is more susceptible to interference than, for
example, the frequency hopping Bluetooth. Interference from
WLANs can be greatly reduced by using only those Zigbee
radio channels that do not overlap with the WLAN.
Interference caused by Bluetooth is lesser than WLANs
because of Bluetooth’s frequency hopping. As noted in [14],
the non-beaconing Zigbee network performance is altered less
by Bluetooth interference than the beaconing networks. Our
network implementation tolerates short breaks in the
communication, as sent data is buffered and delayed
retransmissions applied. The need for preserving the unsent
sensor data depends on the sensor type, and the cost of
preserving it depends on the amount of data the sensor
produces. For some sensors, like the blood pressure sensor, a
Authorized licensed use limited to: Tampereen Teknillinen Korkeakoulu. Downloaded on February 11, 2010 at 16:03 from IEEE Xplore. Restrictions apply.
Copyright (c) 2009 IEEE. Personal use is permitted. For any other purposes, Permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
TITB-00064-2009
8
permanent, non-volatile, memory-based buffer for unsent data
could be considered since the recordings are rarely done and
produce a small amount of data requiring little memory, i.e. the
measurement value versus cost of preserving factor is high.
Some sensors, like the IR sensor, produce continuously vast
amounts of data with small diagnostic value, and thus the value
versus cost of non-volatile buffering factor is poor.
B. Issues discovered during the case trials
Table II lists observation made during the operational test
and during the case trials. Some of our practical findings may
seem rather simple and not novel from a technical point of
view, but they are all issues that need to be taken into account
during development of the functional home eHealth platform.
One of the objectives of the study was to combine both
custom and commercial sensors in the system. However, the
lack of common standardized communication interfaces is still
a huge problem for WSN developers. In their paper [18],
Edwards and Grinter present a list of challenges in smart home
development. One of the challenges they name is impromptu
interoperability, which means “not just the simple ability to
interconnect between devices, but the ability to do so with little
or no advance planning or implementation.” We are still far
from solving this problem. It was hard to find a blood pressure
monitor or a weight scale with at least some kind of digital
interface for data extraction. Even after finding and purchasing
those sensors, the communication specifications were not
available to us. Currently a non-profit, open industry coalition
called Continua Health Alliance is tackling the interoperability
problem in the area of personal healthcare.
In the first trial, an acceptability problem arose regarding the
bed sensor. The sensor was excluded in the middle of the trial
due to the anxiety caused by the technology. These kinds of
fears may be common while introducing new technologies.
However, during the second study phase, similar problems did
not occur, which indicates a different willingness and need to
use the system. These differences need to be taken into account
by service providers of such systems. The tailoring of the
system should be based on the user’s needs, as stated in [19].
More generally, the acceptance and usefulness of these types of
systems is strongly based on the quality of the related services.
C. Future work
There were a few challenges in the floor sensor’s location
detection that occasionally made the detection with the used
algorithm unreliable. For instance, the carpets on the floor
reduced the sensitivity of the floor sensor, and a person seated
on furniture creates a signal too small to be detected. Further
studies should be done to improve the detection methods.
Associating the measurement data with the correct person is
an issue that needs to be considered if more than one person can
use the same measurement devices. In future tests, we are
planning to investigate different methods for solving this issue.
If accurate position information is available, for example from
the floor sensor, it could be used in the association. Another
more reliable, but also more obtrusive way, is to use an RFID
tag, a personal identification number, or a fingerprint to
associate the measurement with a person. Furthermore, the
measurement result itself often gives an implication of the
person. The most reasonable identification method always
depends on the nature of the measurement; therefore, it might
be useful to implement more than one method.
Another important area in the implementation of an
end-to-end system is the final storage format of the data and
connections for the service provider, be that a hospital or a
health care center. As our study in this phase focused on the
measurement devices and on the WSN, with less work done for
the service provider end, we ended up using self-defined data
formats in our case trials. However, a lot of effort has been put
into standardizing this area, such as the work of HL7 [21],
which should be considered for the wider implementation of
our system. In addition, the network performance in larger
configurations and under interference should be tested more
extensively.
VII. CONCLUSIONS
We propose a health monitoring platform consisting of a
chosen set of sensors, a Zigbee network, a home client, and a
remote server. The platform is easily adaptable for various user
needs with the aid of a common sensor interface. Compared to
similar existing systems, we used both medical sensors and
sensors measuring the activity of the user in our realized
system. The concept was evaluated in two distinct trials: with
an elderly woman living in a sheltered housing and with a hip
surgery patient during his rehabilitation phase. The trials were
conducted in test subjects’ real homes, in contrast to commonly
reported laboratory settings.
TABLE II
LESSONS LEARNED FROM THE TRIALS
Estimating
sleep time
The bed sensor accurately estimated the patient’s sleep time,
but this is not the case if a person stays in the bed longer than
the sleep time. However, there are also indications that with
the bed sensor, the sleep time can be estimated reasonably
reliably, and the sleeping log could be left out [20].
Hiding of
the
electronics
The footage of the closet offered a convenient hiding place
for the UUTE home client and other electronics, as it is not
desired that the inhabitant should worry about or see the
technical details of the system. The disadvantage, however,
was inadequate cooling. Therefore, attention should be paid
to the proper air circulation at the places where any of the
power-consuming parts of the system are located.
Data
storage in
sensors
For the case of network absence, sensors should have a small
memory in which to store the readings until the network is
functional again and the readings can be moved to a server.
Local data
storage
For test purposes, all possible data should be gathered also
locally to ensure that data are not lost if the connection to the
remote server is disconnected.
The capacitive floor sensor that was used shows interesting
promise for activity monitoring. However, further studies
should be done to develop the research prototype version
algorithms more reliable for commercial use. The biggest
challenges we faced with the system ranged from the lack of the
device interoperability with commercial sensors, to the
problem of allocating measured values for the identified user,
and to the user acceptance of these kinds of systems.
Authorized licensed use limited to: Tampereen Teknillinen Korkeakoulu. Downloaded on February 11, 2010 at 16:03 from IEEE Xplore. Restrictions apply.
Copyright (c) 2009 IEEE. Personal use is permitted. For any other purposes, Permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
TITB-00064-2009
9
ACKNOWLEDGEMENT
The authors would like to acknowledge study subjects for
participating in the trials, Lempäälä Health Care Center, and
Leena Hankela (physiotherapist) who helped us during the
second trial.
REFERENCES
[1] Intel® Health Guide, “PHS6000: Connecting patients and healthcare
professionals for personalized care,” Product brief, Intel, 2008. Available:
www.intel.com/healthcare/telehealth
[2] Philips Remote Patient Monitoring, Sept. 2007, Available:
http://www.medical.philips.com
[3] B. Quach, M. Balakrishnan, D. Benhaddou, and X. Yuan,
“Implementation of integrated wireless health monitoring network,” in
Proc. 1st ACM int. workshop on Medical-grade wireless networks, New
Orleans, USA 2009, pp. 63–68.
[4] J. Yao and S. Warren, "Applying the ISO/IEEE 11073 standards to
wearable home health monitoring systems," J. Clin. Mon. and Comp., vol.
19, pp. 427–436, 2005.
[5] S. Koch, “Home telehealth — Current state and future trends.” J. of
Medical Informatics, vol. 75:8, pp. 565–576, Aug. 2006.
[6] U. Varshney, “Pervasive Health care and Wireless Health Monitoring,”
Mobile Netw. Appl, vol. 12(2–3), pp. 113–127, Mar. 2007.
[7] C. N. Scanaill, S. Carew, P. Barralon, N. Noury, D. Lyons, and G.M.
Lyons, “A review of approaches to mobility telemonitoring of the elderly
in their living environment,” Annals of Biomedical Engineering, vol. 34
(4), pp. 547–563, Dec. 2005.
[8] J. Cleland, A. Louis, A. Rigby, U. Janssens, and A. Balk, “Noninvasive
Home Telemonitoring for Patients with Heart Failure at High Risk of
Recurrent Admission and Death: The Trans-European
Network-Home-Care Management System (TEN-HMS) study,” J of Am
College of Cardiology, vol. 45, pp. 1654–1664, May 2005.
[9] A. Darkins, P. Ryan, R. Kobb, L. Foster, E. Edmonson, B. Wakefield, and
A. E. Lancaster, “Care coordination/home telehealth: The systematic
implementation of health informatics, home telehealth, and disease
management to support the care of veteran patients with chronic
conditions,” Telemedicine and e-Health, vol. 14(10), pp. 1118–1126,
Dec. 2008.
[10] M. Buchenau, J. F. Suri, “Experience prototyping,” in Proc. 3rd Conf. on
Designing Interactive Systems: Processes, Practices, Methods, and
Techniques, 2000, New York City, USA, pp. 424–433.
[11] O. Kenttä, J. Merilahti, T. Petäkoski-Hult, V. Ikonen, and I. Korhonen,
“Evaluation of Technology-Based Service Scenarios for Supporting
Independent Living,” in Proc. 29th Annu. Int. Conf. of the IEEE EMBS,
Lyon, France, 2007, pp. 4041–4044.
[12] M. Armholt, S. Junnila, and I. Defee, “A Non-beaconing ZigBee Network
Implementation and Performance Study,” in Proc. IEEE Int. Conf. on
Communications, Glasgow, UK, 2007. pp. 3232–3236.
[13] S. Junnila, I. Defee, M. Zakrzewski, A.-M. Vainio, and J. Vanhala,
”UUTE Home Network for Wireless Health Monitoring,” in Proc. of the
Int. Conf. on Biocomputation, Bioinformatics, and Biomedical
Technologies, Bucharest, Romania, 2008, pp. 125–130.
[14] M.M. Herrera, A. Bonastre, and J.V. Capella, “Performance Study of
Non-beaconed and Beacon-Enabled Modes in IEEE 802.15.4 under
Bluetooth Interference, in Proc 2nd Int- Conf. on Mobile Ubiquitous
Computing, Systems, Services and Technologies, Sept. 29–Oct. 4, 2008,
pp. 144–149.
[15] T. R. Burchfield, S. Venkatesan, D. Weiner, “Maximizing Throughput in
ZigBee Wireless Networks through Analysis, Simulations and
Implementations,” in Proc of 1st Int. Workshop on Localized Algorithms
and Protocols for Wireless Sensor Networks), Santa Fe, USA, 18–20
June, 2007.
[16] F. Österlind, and A. Dunkels, “Approaching the maximum 802.15.4
multi-hop throughput,” in Proc. 5th ACM Workshop on Embedded
Networked Sensors, 2–3 June 2008, Charlottesville, Virginia, USA.
[17] J. Merilahti, T. Petäkoski-Hult, M. Ermes, H. Lahti, A. Ylinen, L. Autio,
E. Hyvärinen, J. Hyttinen, and M. van Gils, “Evaluation of new concept
for balance and gait analysis: patients with neurological disease, elderly
people and young adults,” 6th Conf. of the Int. Society for
Gerontechnology, 2008, Tuscany, Italy, pp. 164–168.
[18] K. Edwards and R. E. Grinter, “At home with ubiquitous computing:
Seven challenges,” in Proc. 3rd Int. Conf. on Ubiquitous Computing,
Atlanta, GE, 2001, pp. 256–272.
[19] I. Korhonen, J. Parkka, and M van Gils, “Health monitoring in the home
of the future,” Engineering in Medicine and Biology Magazine, IEEE,
Vol. 22(3), pp. 66–73, 2003.
[20] D.C. Mack, J.T. Patrie, P.M. Suratt, R.A. Felder, and M. Alwan,
“Development and Preliminary Validation of Heart Rate and Breathing
Rate Detection Using a Passive, Ballistocardiography-Based Sleep
Monitoring System,” IEEE Trans. Inf. Tech. Biom., vol. 13, pp. 111–120,
Jan. 2009.
[21] Health Level Seven. www.hl7.org
Sakari Junnila received his M.Sc. (with honors) and Dr.Tech. degrees in
Information Technology (Digital and Computer Engineering) and Signal
Processing from Tampere University of Technology (TUT), Tampere, Finland,
in 1999 and 2009, respectively. Since 1998 he has held various researcher
positions in TUT, where he works currently as a research scientist in the
Department of Signal Processing. His current research interests include digital
interface technologies, sensors for personal health monitoring, medical
monitoring and data-acquisition systems, and medical device standardization.
Harri Kailanto received his M.Sc. degree in electrical engineering (biomedical
engineering) from TUT, Tampere, Finland in 2006. Since that, he has been
working as a research scientist in TUT, Department of Biomedical Engineering,
where he is also carrying out his post graduate studies. His main research
interests include the processing and analysis of biomedical signals in mobile
environments, and mobile phone applications assisting personal healthcare and
welfare.
Juho Merilahti received the M.Sc. (Tech) degree in information technology
(software engineering) from TUT, Finland, in 2006. He is currently a Research
Scientist at Technical Research Centre of Finland (VTT), Tampere. His current
research interests include biomedical signal processing and personal health
systems.
Antti-Matti Vainio is a researcher at the TUT. He is doing research in the
Personal Electronics Group at the Institute of Electronics. His main research
interests include software systems for smart environments, user interfaces for
these environments, machine learning and adaptive control systems in smart
spaces. He received his Master of Science in 2006 from TUT, majoring in
Software Engineering. He has been working in the academia for 6 years and is
working on his postgraduate studies in this research area.
Antti Vehkaoja received his M.Sc. degree in electrical engineering from TUT,
Tampere, Finland, in 2004. Since then, he has been working towards the PhD
degree in the Sensor Technology research group at the Department of
Automation Science and Engineering of TUT. His current research interests
include devices for physiological monitoring and processing of physiological
signals.
Mari Zakrzewski received her M.Sc. degree (with honors) in electrical
engineering from TUT, Tampere, Finland, in 2005. Since then, she has been a
research scientist in the Department of Electronics in the Personal Electronics
Group and is working towards the PhD degree. Her research interests include
microwave radar monitoring of heart, ubiquitous measurement technologies,
and smart home applications.
Jari Hyttinen received his M.Sc. and Ph.D. degrees from TUT, Tampere,
Finland in 1986 and 1994, respectively. Currently he is Director of the
Department of Biomedical Engineering, TUT. He is an author or co-author in
more than 150 scientific papers as well as inventor or co-inventor in some
patents. He is former chairman and current member of the board of the Finnish
Society of Medical Physics (www.lfty.fi) and Biomedical Engineering (affiliate
of the IFMBE). He is member of the general council of the European Alliance
for Medical and Biological Engineering & Sciences (www.eambes.org).
Authorized licensed use limited to: Tampereen Teknillinen Korkeakoulu. Downloaded on February 11, 2010 at 16:03 from IEEE Xplore. Restrictions apply.
