way: seamless positioning using a smart devicerepnin.es/uploads/archivos/ar.bahillo.2016b.pdf ·...
TRANSCRIPT
WAY: Seamless Positioning Using a Smart Device
Alfonso Bahillo1 • Teodoro Aguilera2 • Fernando J. Alvarez2 •
Asier Perallos1
� Springer Science+Business Media New York 2016
Abstract Smart devices are attractive platforms for researchers to collect data coming
from several sensors due to their small size, low cost, and the fact that they are already
carried routinely by most people. The capability of smart devices to be used as the target of
a positioning system has been already demonstrated in previous works. However, most of
them rely on a single technology, or they are specific to the environment or user. In this
paper we tackle these constraints by presenting a novel seamless positioning system which
fuses the sensors information provided by a portable smart device to perform real time
location without interruption and independently of the environment the user is moving. We
have tested the system with a commercial smart device in an uncalibrated three floor
building and its surroundings fusing the GNSS, WiFi and barometer as frequently used
sensors, and the microphone and the proximity contactless technologies as occasionally
used sensors. The obtained positioning accuracy mainly depends on the indoor path-loss
awareness and on the markers density, showing that without using markers but dynami-
cally estimating the path-loss exponents we obtain an error of\2 m for 90 % of cases.
Keywords Sensor fusion � Seamless positioning � Smart devices
& Alfonso [email protected]
Teodoro [email protected]
Fernando J. [email protected]
Asier [email protected]
1 DeustoTech-Fundacion Deusto, Deusto Foundation, Av. Universidades 24, 48007 Bilbao, Spain
2 Electrical Engineering, Electronics and Automation Department, University of Extremadura,06006 Badajoz, Spain
123
Wireless Pers CommunDOI 10.1007/s11277-016-3759-x
1 Introduction
Man has always felt the need to position himself using the sun and the stars as the only
reliable reference systems for many centuries. Today, with the advent of global navigation
satellite systems (GNSS) global positioning in the earth’s surface is a successfully over-
come problem. Nonetheless, local positioning in indoor environments is still a matter of
active research, since GNSS signals get severely degraded in this type of environments due
to multipath and attenuation losses, and thus they cannot be used to track people or objects
with acceptable accuracy [1]. Many local positioning systems (LPS) have been developed
during the past two decades based on different technologies that include ultrasound [2–6],
radio-frequency [7–9], vision cameras [10, 11] and magnetic [12, 13], among others. After
all this research effort, it is becoming clear that none of these technologies clearly out-
performs the others, and we expect now a new tendency to design systems that combine
some of them to benefit from their complementary strengths.
In this work we propose a novel seamless positioning system for the fusion of the
sensors information provided by a portable smart device, intended to perform real time
location without interruption and independently of the environment the user is moving,
thus covering the deficiencies of GNSS indoors and in urban outdoor environments with
poor coverage. The capability of smart devices to be used as the mobile node in privacy-
oriented systems, where this node is in charge of computing its own position, and also in
centralized systems, where the node position is computed by a central processor, has been
already demonstrated in several works. Hence, in [14, 15] two privacy-oriented systems
based on the use of Bluetooth beacons are described, where an Audiovox SMT5600 and a
Nokia N70 devices are respectively positioned with an accuracy of 2–3 m. A centralized
system based on the same technology is proposed in [16], achieving accuracies of 2 m in a
12 9 7 m2 laboratory with a Sony Ericsson P800. In [17] a privacy-oriented system is
proposed based on the measurement of the received VHF signals strength to achieve an
accuracy of 1–3 m. A different approach is followed in the centralized system described in
[18], where a PDA HP5550 is positioned by performing the emission of a pure 4.01 kHz
acoustic tone during 100 ms, with accuracies of about 65 cm. Another centralized system
based on the emission of short ultrasound pulses of 21.5 kHz has been recently proposed in
[19] to position an HTC-G1 smartphone with accuracies of about 10 cm in a restricted
environment of 7 9 7 m2. Also, accuracies below 4 m have been reported in [20] by using
the magnetometer of a HTC Nexus One in several corridors of two buildings. All these
works rely on a single technology, though. Some authors have exploited the benefits of
using additional technologies by incorporating external sensor units whose information is
centralized and managed by the smart device. Hence, in [21] a dual IMU system, one
mounted in the torso and a second one mounted in the foot, are used to collect rich
information that is sent to a OpenMoko phone where a simplified map-matching particle
filter estimates the user position. A more complete system is described in [22], where an
external multi-sensor platform that integrates an enhanced GPS receiver and inertial
sensors is used to retrieve the user position when this information cannot be obtained from
the built-in GPS and WLAN receivers of a Nokia N95 smartphone. Unfortunately, the
main disadvantage of these systems is the loss of portability associated with the use of an
external sensor unit. There are also some recent works that propose the fusion of the
information provided only by the different built-in sensors of a smart device. This is, for
example, the case of the system proposed in [23], that implements a Kalman filter to
combine the information obtained from the WiFi engine and the built-in accelerometer and
A. Bahillo et al.
123
digital compass of a Samsung Nexus S, obtaining accuracies between 2.4 and 4.1 m in a
3000 square meters building level. A similar system is proposed in [24], that uses Hidden
Markov Models to combine the information provided by the WiFi engine, the
accelerometer and the compass of a Nokia N8, to achieve accuracies around 3 m in a real
office environment. In both cases, the system has been designed to operate in a calibrated
environment and it is very sensitive to changes in this environment, since the WiFi
positioning engine is based on the comparison on the measured RSS values with those
previously gathered in a fingerprint database.
To the authors knowledge, the seamless positioning system presented in this paper is the
first system to combine a wide diversity of information provided by different sensors of a
single and commercially available smart device with the aim to perform the seamless
positioning of its user both outdoors and in uncalibrated indoor environments, i.e. envi-
ronments whose sensory signals have not had to be measured and analysed beforehand for
later estimation of the user position during real-time operation. The main improvements of
this system with respect to previous directly related works can be summarized as follows:
• The WiFi positioning engine is not based on fingerprinting, what gives the system the
capability to work in uncalibrated environments.
• The information provided by the smart device sensors is combined by means of an
unscented Kalman filter (UKF), a filter that is more accurate than an extended Kalman
filter (EKF) and easier to implement than an EKF or a Gauss second-order filter [25].
• Specifically designed ultrasound markers and the built-in microphone are used to
accurately update the user position in regions of particular interest without the need of
any action from the user side.
This system would be the ideal platform for location based services (LBS), that provide the
user with personalized information depending on his geographic location [26]. The pro-
posed seamless positioning system would easily allow the extension of all those services
that are currently being provided outdoors to indoor environments such as hospitals,
museums, train stations, airports, malls and many others that could undoubtedly benefit
from this type of services. Hence, among the potential applications of WAY are emergency
assistance, indoor routing, car park guidance, personalized advertising, location sensitive
billing and leisure activities such as buddy finder or geocaching. Needless to say, the
involvement of end-users will be necessary to validate the system and to develop these
services. In that sense, the system will ensure confidentiality by not safeguarding sensitive
information, recognizing the participants property over their personal data and their right to
decide on how these data shall be used, share and made public (through informed consent)
and finally, by informing them about how this information will be disposed of.
The rest of the paper is organized as follows. Section 2 presents the general features of
the seamless positioning system, including the management, processing and presentation
blocks. In Sect. 3 a detailed description of the different sensors providing information to
the seamless positioning system is given. Section 4 presents some experimental results
obtained with this system in a real environment composed by a three floor building and its
surroundings. Finally, Sect. 5 outlines the main conclusions that can be drawn from these
results.
WAY: Seamless Positioning Using a Smart Device
123
2 Seamless Positioning System WAY
The seamless positioning system proposed in this paper, called WAY (Where Are You?)
performs a seamless localization estimation of the user’s smart device in real time.
Nowadays, the smart devices already integrate a GNSS receiver, a WiFi adapter, a camera
and a microphone, and shortly, most of them will integrate other sensors such as the
barometer and the proximity contactless technology NFC. The aim of WAY is to develop a
framework that estimates the user position in a continuously and transparent fashion by
fusing the so called signals of opportunity (SoOP) which are transmitted for non-local-
ization purposes, but may be exploited to this end. The SoOP chosen for the evaluation of
WAY are the GNSS, WiFi, ultrasound, NFC, the air pressure values and the QR codes.
Thus, WAY fuses information provided by GNSS, NFC tags, QR codes and ultrasound
markers at the position solution level with the information provided by WiFi at the raw
measurements level. In fact, WAY is thought as a modular system for commercially
available smart devices where SoOP coming from new technologies can be easily inte-
grated. To achieve the WAY’s aim, we will take into account criterions such as available
battery, position estimation accuracy, and system continuity and performance. Thus, WAY
will help to develop LBSs which make easier the user-environment interaction.
In Fig. 1 we describe the flowchart that follows WAY to fuse and smartly prioritize the
different SoOP to be used. In general terms, the flowchart that governs WAY is divided
into three blocks: management, processing and presentation. The management block will
decide what signal will get to the processing block. The processing block will fuse the data
ProcessingBlock
Message: Waitingfor Signal
Accurate PosEnvironment Previous
ScanningGNSS
Scanning WiFi
> # Sat > # APsEnable WiFi
Enable GNSS
GettingLat,Lon,Alt
GettingRSS
NO NO
NOYES
YES YES
Tim
e O
ut
Tim
e O
ut
OutdoorIndoor GNSS
WiFi/None
Start
Disable WiFi Disable GNSS
Management Block
Fig. 1 Flowchart of the management block
A. Bahillo et al.
123
coming from the management block to infer the user position. Finally, the presentation
block will show the estimated position and it can be used as the interface for LBSs.
2.1 Management Block
This block not only decides what SoOP will get to the processing block but it is also
responsible for what smart device sensor will be enabled. Thus, it plays an important role
in the battery consumption. WAY decides what SoOP will be processed taken into account
the available battery and the environment. Thus, if the user is in an open area, where a line-
of-sight between the smart device and the satellites is guaranteed, the barometer is disabled
and the GNSS receiver will be enabled, thus the three-dimensional geographic coordinates:
latitude, longitude and altitude will be used. However, if the user is inside the perimeter of
a building, the barometer and the WiFi adapter will be enabled, thus the received signal
strength (RSS) coming from the WiFi access points will be used. Although, the man-
agement block keeps using the SoOP based on the environment, near any indoor/outdoor
transition zone, such as doors, both GNSS and WiFi, will be enabled.
As shown in Fig. 1 WAY can start from an accurate position. By accurate position we
mean that, at a certain point in time, WAY has received accurate and reliable information
about the user position. This information could arrive from reading a QR code, an ultra-
sound or NFC tag, or even it can be typed by the user itself. In case WAY starts without
information about the initial position, the WiFi adapter starts scanning for beacon signals
coming from WiFi access points. After a threshold time, if the WiFi signals are not found,
the GNSS receiver starts scanning for satellite signals. Both receivers, GNSS and WiFi, are
enabled searching for signals until one of both fulfils the required condition for getting to
the processing block. In the GNSS case, more than three satellites in line-of-sight. In the
WiFi case, more than two access points in range.
Notice that (1) in case WAY does not know the initial position, it first scans for WiFi
signals because it is faster than GNSS scanning; (2) WiFi scanning does not compromise
the data communication and it does not inject data traffic into the network; and (3) both
receivers, GNSS and WiFi, compromise the battery consumption, a reason why just one
receiver is kept enabled collecting signals1 at a time.
2.2 Processing Block
This block represents the framework which estimates the user’s position fusing the SoOP
gathered by the smart device sensors. Bayesian filters, which use a probabilistic framework
to perform reasoning, are a theoretically sound way to combine multiple and different
SoOP. Bayes filters probabilistically estimate a dynamic system’s state from noisy
observations. They represent the state at time k by random variables xk. At each point in
time, a probability distribution over xk, called belief, represents the uncertainty. Bayes
filters aim to sequentially estimate such beliefs over the state space conditioned on all
information contained in the observations [27]. In this paper, the state is the user’s location,
xk ¼ ½xk; yk; zk�T , while the SoOP provide observations about the state. Among the different
Bayes filters, Kalman filters are the most widely used. They are optimal estimators,
assuming the initial uncertainty is Gaussian, the observation model and system dynamics
are linear functions of the state, and the measurement and process noise distributions are
1 If the battery consumption will not be a constraint, both GNSS and WiFi will succeed to the processingblock.
WAY: Seamless Positioning Using a Smart Device
123
Gaussian. However, the lack of linearity in the models that relates most of the SoOP to the
user’s location implies the usage of a suboptimal solution, where the most common is to
use the EKF. Nevertheless, we selected the unscented Kalman filter (UKF) since it better
captures the higher order moments caused by the non-linear transformation and avoids the
computation of Jacobian and Hessian matrices [25]. Furthermore, the overall number of
computations performed by the UKF are the same order as the EKF, and much lower than
solutions such as the particle filter which better represents the belief but needs a number of
computations unacceptable for most smart devices. By using an UKF as the framework,
WAY finds the trade-offs among accuracy, real time and battery consumption.
Consider the following non-linear system, described by the dynamic and measurement
models with additive noise:
xk ¼ fðxk�1Þ þ wk�1 ð1Þ
zk ¼ hðxkÞ þ vk�1 ð2Þ
where wk and vk are the process and observation noise which are both assumed to be zero
mean multivariate Gaussian noise with covariance Qk and Rk, respectively. The function
f can be used to compute the predicted state from the previous estimate and similarly the
function h can be used to compute the predicted measurement from the predicted state.
On the one hand, the dynamics of the system can be represented as
xk ¼ xk�1 þ _xk�1Dt þ wk�1 ð3Þ
where Dt ¼ ðtk � tk�1Þ is the time step and _xk is the first derivative of the state, in this case,the user’s velocity. Finally, wk is assumed to be a zero-mean Gaussian variable with
covariance matrix Qk. The values of Qk depend on the dynamic of the target, in this paper
a walking person. In practice, Qk is a diagonal matrix whose in-diagonal elements rep-
resent the variance of the user’s position and velocity [28].
On the other hand, the function h depends on the SoOP. In the WiFi case the mea-
surement model, called hw, can be represented as
zwk ¼ a� 10n log 10ðjjxk � APjjÞ þ vwk�1 ð4Þ
where zwk is the RSS measured value; a is a parameter that remains constant in those
scenarios where the antennas gain and the power transmitted from the access points are
also constant, a situation typically found in most WiFi WLANs, (in practice, this value can
be known beforehand from experimental measurements taken in a generic environment
similar to that where the location system is going to operate) [29]; AP is the position of the
WiFi access point; and n is the path-loss exponent corresponding to the actual propagation
environment. In free space n ¼ 2, however in practice, depending on the environment the
path-loss exponents ranging from 1.5 to 4.5 [29, 30]. In this paper we analyze the per-
formance of WAY using the path-loss exponent values under two different situations: (1)
setting fixed values for the path-loss exponents sampling values ranging from 1.5 to 4.5
according to this rule: the stronger the RSS measured value (assuming a closer position to
the antenna) the smaller the path-loss exponent value (assuming lower reflections and
diffractions), and (2) dynamically estimating the path-loss exponent values following the
algorithm proposed in [8]. By using this algorithm we found the path-loss exponents that
best fit the propagation environment between the receiver (user’s smart device) and each
WiFi access point in range at each time interval. vwk represents the RSS noise and it can be
assumed to be zero-mean Gaussian variable with covariance matrix Rwk . In practice, we
A. Bahillo et al.
123
have hwi with i ¼ 1; 2; . . .Mk functions, where Mk is the number of WiFi access points in
range at each time step k. Accordingly, Rwk is a diagonal matrix whose in-diagonal ele-
ments represent the variance of the measurements comming from each WiFi access point.
In the GNSS case, the measurement model, called hg, can be represented as
zgk ¼ xk þ vgk�1 ð5Þ
where zgk is the GNSS position estimation. As the GNSS reports the position in geographic
coordinates, and WAY works in UTM (Universal Transverse Mercator) coordinates which
uses a 2-dimensional cartesian coordinate system to give locations on the surface of the
Earth, we previously perform the corresponding transformation from geographic to UTM
coordinates. vgk represents the GNSS noise which can be assumed to be zero-mean
Gaussian variable with covariance matrix Rgk . In practice, Rg
k is a diagonal matrix whose
in-diagonal elements represent the variance of the measurements comming from the
satellites, and its value depends on the number of satellites in line-of-sight. The more
satellites with good GDOP (Geometric Dilution of Precision), the more reliable the GNSS
data.
In the barometer case, the measurement model, called hb, can be represented as
zbk ¼ b�1ðxk½3�Þ þ vbk�1 ð6Þ
where zbk is the air pressure, xk½3� is the third element of the state, the altitude, and b�1 is
the inverse of the model used to convert the air pressure into altitude. This model, b, will be
explained in the next section. vbk represents the air pressure noise which can be assumed to
be zero-mean Gaussian variable with covariance matrix Rbk .
Finally, in the contactless technologies cases using ultrasound signals, NFC tags and QR
codes, the measurement model, calle hc, can be represented as
zck ¼ xk þ vck�1 ð7Þ
where zck is the position of the ultrasound or NFC tag, or QR code. vck represents the
observation noise which can be assumed to be zero-mean Gaussian variable with covari-
ance matrix Rck. As the SoOP gather by the contactless technologies have to be read at few
centimeters from the tag or code, in practice, Rck is a diagonal matrix whose in-diagonal
elements are lower than 1 meter.
As shown in Fig. 2, and it is well described in [25], once the dynamic and measurement
models, and their noise covariace matrices are described, the UKF is straightforward to
implement. Notice that the initial values of the state covariance matrix, P0, depends on the
initial state confidence. UKF main advantage is their computational efficiency (same as
EKF and lower than particle filters), better linearization than EKF (accurate in the third-
order Taylor series expansion), and derivative-free (no Jacobian and Hessian matrices are
needed) [25].
2.3 Presentation Block
There are many ways of representing the user position and its covariance. On its basic
form, WAY shows an ellipse on a map of the smart device screen, where its centre is the
user position estimate and its major and minor axes represent the uncertainty on each
direction. Furthermore, the user position and its covariance can be sent to a server and
show it remotely, or it can be saved for subsequent evaluation. The format of the position
WAY: Seamless Positioning Using a Smart Device
123
coordinates can be chosen and even the attitude of the user can be inferred from con-
secutive positions. This block will be the interface with LBSs which make easier the user-
environment interaction.
3 Sensors Information
To develop a seamless positioning solution on a smart device we have to find those
technologies which can provide location information, how accurate it could be and where
they could work. In this section we briefly describe the sensors information to be fused,
their characteristics (availability, inputs, outputs, functionality, etc.) and their main role
into WAY. These sensors have been classified into three gruops, namely, the GNSS
receiver, the signals of opportunity, and the ultrasound markers and microphone. It is
important to remark that, although WAY has been designed to interpret and fuse the
information provided by all these sensors, this does not mean that all of them will be
simultaneously active during a regular operation. On the contrary, in most real scenarios
only part of this sensory information will be available, that might be different depending on
the particular features of the environment. The large variety of sensors managed by WAY
confers high versatility to this application, that is always trying to exploit the benefits of
each sensor to achieve a more accurate location in any scenario. These benefits are
described next.
ManagementBlock
MeasurementModel
MeasurementUpdate
TimeUpdate
DynamicModel
PresentationBlock
Lat,Lon,Alt(GNSS)
RSS(WiFi) BarometerContactless
zk, Rk, hk
xk, Pkx0, P0
Qk, Fk Pk|k-1, xk|k-1
Pk|k, xk|k
Processing Block
Fig. 2 Flowchart of the processing block
A. Bahillo et al.
123
3.1 GNSS Receiver
The GNSS receiver which is integrated in most smart devices is extensively used to
estimate the user position in open areas. WAY uses the GNSS signals to infer the user
location where at least three or more satellites were visible. Therefore, WAY beneficts
form the GNSS data only in open areas where it accurately reports the user’s position in
terms of its latitude, longitude and altitude. Usually, GNSS data are already Kalman
filtered inside the GNSS receiver, thus a cascade of two filters take place: internal GNSS
KF and the WAY UKF.
3.2 Signals of Opportunity
As stated in the introduction, WAY uses a new set of SoOP that includes the WiFi adapter
and the barometer as frequently used sensors, and the proximity contactless technologies as
ocassionally used sensors.
3.2.1 Wireless WiFi Adapter
Not only data communication through the wireless ethernet but also position estimation
can be achieved by using the wireless WiFi adapter which is integrated in most smart
devices. By scanning the environment, the smart device can get the RSS of WiFi signals
coming from each WiFi access point in range identified through its physical address. The
benefit of using this SoOP is that WiFi signals predominate inside buildings where GNSS
signals are blocked. Assuming the WiFi access points position is previously known, we can
perform the procedure shown in Sect. 2, which dynamically reports an estimation of the
user position by only using the RSS. WAY frequently uses the WiFi adapter to estimate the
position of the user inside buildings, where the WiFi signals predominate.
3.2.2 The Barometer
Many personal navigation systems will benefit from a barometer because it reports
accurate altitude information inside buildings where GNSS data is blocked. Shortly, it will
be easy to find a barometer integrated in most smart devices which accurately measures the
air pressure. As it is commonly known, pressure measured by barometers can be converted
into altitude information. Altitude can be modelled accurately by the International Stan-
dard Atmosphere (ISA) [31]. Other used models to convert the air pressure into altitude
such as U.S. Standard Atmosphere and WMO Standard Atmosphere are equivalent to the
ISA model up to 32 km over the sea level [32]. The ISA equation for altitude xk½3� is
xk½3� ¼ bðzbkÞ ¼T0
k1� zbk
p0
� �kRairg
0@
1A ð8Þ
where T0 ¼ 288; 15 K is the temperature at sea level, k ¼ �0:0065 K/m is the temperature
gradient, p0 ¼ 1013:25 mbar is the pressure at sea level, Rair ¼ 287 m2(s2 K)-1 is the
atmosphere gas constant, and g ¼ 9:8 m/s2 is the earth gravity. This model is not exact
because it assumes some variables as constants, and it does not take into account other
factors such as humidity, weather, or the presence of air conditioning systems. However,
WAY does not use the absolute but the relative altitude. When available, the absolute
WAY: Seamless Positioning Using a Smart Device
123
altitude is given by the GNSS. The estimated relative altitude is used to infer any building
floor change. Furthermore, it can also be used to distinguish if the user got to a floor
through the lift or the stairs. In the lift case, the observed altitude profile has a constant and
higher slope than in the stairs case. Figure 3 shows the estimated altitude profile corre-
sponding to the path described in Fig. 6 when the target is inside the building. The path
followed has four floor changes: two by using the stairs (going up from ground floor to 1st
floor, and from 1st to 2nd floor), and two by using the lift (going down from 2nd to 1st
floor, and from 1st to ground floor). The observed slope when going up (or down) by lift is
always constant and higher than 0.75 m/s, but when going up (or down) by stairs the slope
is more noisy and it is always lower than 0.5 m/s. Therefore, thanks to the barometer data,
WAY can easily identify when the user is on the stairs or in the lift which also can be used
to update the user position.
3.2.3 Proximity Contactless Technologies
The benefit of using the proximity contactless technologies is that we can set the position
of the user to the position of the marker when reading the marker because the user has to be
placed at few centimeters form the mearker to this purpose. Shortly, these proximity
contactless technologies will be predominant therefore, assuming as previously known the
position of the markers, once on a while WAY will be able to update accurately the
position of the user. Additionally, depending on the context we can configure more actions
on the smart device when the tag is read such as displaying local information, configuring
the wireless connectivities, enabling or disabling sounds, setting calendar issues, etc. [33].
These proximity contactless technologies work anywhere and they can update the location
algorithm with accurate position information.
3.3 Ultrasound Markers and the Microphone
Similarly to the proximity contactless technologies, but without the need of any action
from the user side, the microphone can be used to accurately update the user position in
Fig. 3 Altitude profile of the path followed at the ETSIT (detailed in Fig. 6)
A. Bahillo et al.
123
locations of particular interest. This property derives from the capability of smart devices
microphones to acquire ultrasound for accurate acoustic code identification. The fre-
quencies of the generated emissions are between 18 and 22 kHz, which are high enough to
be inaudible for almost every people but low enough to be generated by standard sound
hardware.
Previous works had shown that it is possible to emit a 21 kHz acoustic signal, just above
the human hearing range, from a smart device speaker and successfully receive it with a
conventional microphone [1, 19, 34]. WAY explores the reversed circumstance, i.e. low
frequency ultrasound signals are generated by specifically designed ultrasound markers
based on the Kingstate KSSG1708 transducer [35], and their reception is carried out by the
smart device, without the need of any additional microphone, acquisition or processing
hardware. To this respect, it is important to note that although the maximum theoretical
value for the frequency response of the built-in device microphone has been specified at
20 kHz by an external analysis [36], we have successfully performed acquisitions at fre-
quencies above 22 kHz. Figure 4 shows the frequency response of the emitter/receiver pair
experimentally obtained in our lab when the microphone is placed at one meter distance
over the acoustic axis of the ultrasound marker.
The ultrasound markers emit 63-bit Kasami sequences BPSK modulated with a symbol
of four cycles of a 20 kHz carrier. The good autocorrelation and crosscorrelation properties
of these codes allows the identification of different emissions in very noisy environments.
However, to avoid hearing these emissions, some additional processing is necessary. All
codes are band-pass filtered with lower and upper passband frequencies of 18 and 22 kHz
respectively, and then Hamming-windowed to attenuate the transient effects. Figure 5
shows the temporal and spectral features of Code 1 before (a) and after (b) this processing.
The acquisition of audio signals with an Android device requires the implementation
and configuration of the class which manages the audio resources for Java applications.
The system sampling rate was set to 48 kHz and the buffer capacity was chosen accord-
ingly to assure the acquisition of at least one complete code. Once acquired, the samples
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23−40
−35
−30
−25
−20
−15
−10
−5
0
Frequency (kHz)
Nor
mal
ized
Sou
nd P
ress
ure
Leve
l (dB
)
Fig. 4 Experimental frequency response of the ultrasound marker/smart device microphone pair
WAY: Seamless Positioning Using a Smart Device
123
are processed making use of a library of basic DSP algorithms that allows the identification
of the emitted signals by matched filtering and thresholding.
The microphone works anywhere, but WAY only uses it when the estimated position is
in the proximity of a ultrasound marker, thus alleviating the smart device processor
computational load and reducing the power consumption.
4 Experimental Results
WAY has been developed on Android, a mobile operating platform which nowadays
controls most of smart devices. As a way to verify its behaviour under a real environment it
has been tested using the GNSS, WiFi and barometer as frequently used sensors, and the
microphone and the proximity contactless technologies as occasionally used sensors.
4.1 Experimental Setup
For the experimental evaluation we conducted real trials at the building of the Higher
Technical School of Telecommunications (ETSIT) in the city of Valladolid, whose floor
map is shown in Fig. 6. This three floor building is 125 m long and 75 m wide. In the
experiments the target is a person who carries the Galaxy Nexus GT-i9250 smartphone as
smart device. This smartphone incorporates the broadcom BCM4330 single chip device
providing with integrated IEEE 802.11 a/b/g among other handheld wireless systems, the
Bosch BMP180 digital barometric pressure sensor as barometer, the CSR GSD4t as
internal satellite signal tracking engine, the GH59-11319A module as microphone, and the
NXP PN533 NFC controller.
0 200 400 600
−1
−0.5
0
0.5
1
Samples
Nor
mal
ized
Am
plitu
de
0 5 10 15 20 250
0.5
1
Frequency (kHz)
Nor
mal
ized
Am
plitu
de
0 200 400 600
−1
−0.5
0
0.5
1
Samples
Nor
mal
ized
Am
plitu
de
0 5 10 15 20 250
0.5
1
Frequency (kHz)
Nor
mal
ized
Am
plitu
de
(a)
(b)
Fig. 5 Temporal and spectral features of Code 1 before (a) and after (b) processing
A. Bahillo et al.
123
As WiFi access points we used the Trapezee MP372 already deployed at the ETSIT
building which send a beacon frame each 100 ms at constant power on frequency channels
1, 6 and 11 (around 2.412, 2.437 and 2.462 GHz, respectively). The building contains 8
WiFi access points at the ground floor, 10 at the first floor, and 11 at the second floor. Thus,
we have approximately one WiFi access point each 900 m2. As QR codes we printed three
matrix barcodes which encode an URL direction pointing to a XML file. The XML file
contains information about the actual position of the QR code among other contextual
information. As NFC tag we used the NXP MF1 S70 smart card with a writable 4096 bytes
EEPROM. Equally, the NFC tag contains an URL direction which points to a XML file
that contains contextual information and its actual position. The ultrasound markers used in
the experimental setup have been specifically designed for this work. They are based on the
Kingstate KSSG1708 low power transducer, the LM386 low power audio amplifier and the
LPC1768 microcontroller, which stores the 2520 samples of the BPSK modulated Kasami
code and dispatch them at a 200 kS/s rate every 20 ms.
The path shown in Fig. 6 tries to cover all the situations that a user can face up when
walking inside a building with an uncalibrated WiFi network. The path contains outdoor to
indoor and indoor to outdoor transitions, floor changes by stairs and lift, and wide and
narrow corridors. The user was following the path walking through the corridors and
offices of the three floors where at all times there were people walking around. In this case
the initial position was established by the GNSS measurements because the path starts in
an open area. Figure 6a–c show the actual path in red slash, the estimated positions in blue
diamonds, and the position of the WiFi access points, QR codes, NFC and ultrasound
markers deployed in the ETSIT ground, first, and second floors.
Fig. 6 Path followed carrying the smartphone at the ETSIT building. In red slash the actual path, in bluediamonds the estimated positions. a Second floor, b first floor and c ground floor. (Color figure online)
WAY: Seamless Positioning Using a Smart Device
123
4.2 Results
In Fig. 6 we observe that at first the user is outside in an open area where the GNSS signals
are of good quality. Thus, when outside WAY is only using the GNSS observations, i.e. hg
and Rgk . Prior to enter the building we have a QR code with information about the building
services. When reading the QR code WAY incorporates the accurate position information
hidden in the QR code, i.e. uses hc and Rck values. Once the user is inside the building, the
GNSS signals are weaker enough to enable the WiFi adapter. In few seconds the GNSS is
disabled and the WiFi signals and pressure observations start to be used, i.e. hwi and Rwk;i.
The WiFi signals are strong enough to be used during the time the user is inside the
building. Thus, while inside the building, the observed positions are performed by the
WAY algorithm with the WiFi RSS observations. Once in a while, an accurate position
coming from the barometer, the QR codes, NFC or ultrasound markers update the WAY
algorithm, i.e. uses hc and Rck values. The markers density is sufficiently low to avoid
masking the accuracy of the whole system.
The barometer sensor controls the relative altitude profile. When it detects a change
higher than 1.5 m the user position is updated to the nearest stairs or lift location (we
assume the stairs and lift locations to be previously known). After a few seconds the slope
of the altitude profile differentiates between transitions by stairs or lift. Finally, when the
altitude profile detects a change equal or higher than the floor height the planimetry is
automatically updated to the current floor on the smart device screen.
Other source of valuable information comes from the markers. When reading QR codes
or NFC tags the WAY scheme is updated with the corresponding accurate position of the
marker. Similarly to these proximity contactless technologies, but without the need of the
user action, works the microphone sensor. This property derives from the capability of
smart devices’ microphones to acquire ultrasound for accurate acoustic code identification.
Each ultrasound marker emits an ultrasound code which identifies it unambiguously. Equal
to the barometer sensor, the advantage of ultrasound technology is that the smart device
automatically detects the measurements without the need of any user action.
For the evaluation of the error distribution, we compared the obtained positions with a
ground truth measured using marks in the walking path. The cumulative distribution
function (CDF) of the positioning errors is presented in Fig. 7. The proposed WAY is able
to obtain positions errors of \8 m for 90 % of the samples when using fix path-loss
exponents and a few markers such as the QR codes, NFC tags or ultrasound beacons which
once in a while update the estimated position. This error increases up to 12 m for 90 % of
samples without using the markers. We compared these two CDFs to the one obtained by
dynamically estimate the path-loss exponents that best fits the propagation environment in
real time without the need of any calibration stage, but only using the actual RSS mea-
surements, following the algorithm proposed by Mazuelas et al. [8], in both cases with and
without markers. When the path-loss exponents are dynamically estimated the positions
errors are lower than 2 m for 90 % of samples. These results show the importance of the
path-loss exponents values when working with RSS measurements.
Finally, with the aim of analyzing the WAY battery consumption, in Fig. 8 we show the
battery charge comparison of three Samsung smartphones, Note 2, Nexus 3 and Nexus 5
when running the WAY application. After 40 minutes, we observe that the smartphones
Note 2, Nexus 3 and Nexus 5 reduce the battery charge 3, 6 and 4 %, respectively, when
the WAY application is running with respect to the situation where only a white screen is
shown. The differences observed among the different smartphones are mainly because
A. Bahillo et al.
123
their batteries have been used for different periods of time along their lives. However, we
have observed that on average, the WAY application uses around 87mAh of the battery
when it is running with all the sensors enabled.
5 Conclusions
In this paper we have presented an UKF scheme for the real time fusion of the SoOP
provided by a portable smart device sensors to guarantee a seamless positioning system.
The flowchart that follows WAY smartly combines and prioritizes the different SoOP to be
fused. The GNSS signal provides a good three-dimensional position estimation in open
areas, while the WiFi RSS provides a good two-dimensional estimation in indoor envi-
ronments. The barometer provides a good estimation of the relative altitude profile indoors,
thus, this sensor extends the indoor positioning to 2.5 dimensions. Additionally, the
markers such as the QR codes, NFC and ultrasound can be used to accurately, and once in a
while, update the estimated position. The QR and NFC markers need to enable the camera
or the NFC technology, respectively, while the ultrasound markers do not need any action
from the user side. WAY is thought as a modular system, therefore, new SoOP which can
be related to the user position can be easily integrated, such as map information, user
acceleration, rate of turn or magnetic north. In fact, the next step is to integrate the inertial
sensors of the smart device in the UKF to better implement the dynamic model.
The proposed seamless positioning system called WAY has been tested with a com-
mercially available smart device in an uncalibrated real environment. It performs real time
location without interruption and independently of the environment. WAY is able to obtain
positions errors of \8 m for 90 % of the samples when using a few markers which
occasionally update the estimated position. This error increases up to 12 m for 90 % of
samples without using the markers. However, when the path-loss exponents are dynami-
cally estimated, in both cases with and without markers, the positions errors improve taken
values lower than 2 m for 90 % of samples. Therefore, indoors the obtained positioning
0 5 10 15 20 25 300
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
[m]
Per
cent
age
WAY using known tags and dynamically estimating the path−loss exponentsWAY dynamically estimating the path−loss exponentsWAY using fixed path−loss exponents and tagsWAY using fixed path−loss exponents
Fig. 7 Cumulative distribution function of the position error for WAY using known and fixed path-lossexponents with and without markers
WAY: Seamless Positioning Using a Smart Device
123
accuracy mainly depends on the indoor path-loss awareness. Even the markers could be
dispensable if we use the path-loss exponents that best fit the propagation path between the
target and each WiFi access point in range.
Acknowledgments This work has been supported by the Spanish Ministry of Economy and Competi-tiveness under the ESPHIA project (ref. TIN2014-56042-JIN) and TARSIUS project (ref. TIN2015-71564-
0 5 10 15 20 25 30 35 4090
92
94
96
98
100
Bat
tery
cha
rge
perc
enta
ge
Minutes
White ScreenWAY
0 5 10 15 20 25 30 35 4085
90
95
100
Bat
tery
cha
rge
perc
enta
ge
Minutes
White ScreenWAY
0 5 10 15 20 25 30 35 4092
94
96
98
100
Bat
tery
cha
rge
perc
enta
ge
Minutes
White ScreenWAY
(a)
(b)
(c)
Fig. 8 Battery charge percentage comparison of Samsung. a Note 2, b Nexus 3 and c Nexus 5
A. Bahillo et al.
123
C4-4-R). The authors would like to thank Dr. Jose M. Villadangos for helping with the hardware design ofthe ultrasound marker.
References
1. Zandbergen, P. (2009). Accuracy of iPhone locations: A comparison of assisted GPS, Wifi and cellularpositioning. Transactions in GIS, 13(s1), 5–25.
2. Addlesee, M., Curwen, R., Hodges, S., Newmann, J., Steggles, P., Ward, A., et al. (2001). Implementinga sentient computing system. IEEE Computer, 34(8), 50–56.
3. Randell, C., & Muller, H. (2001). Low cost indoor positioning system. In Proceedings of the 3rdinternational conference on ubiquitous computing (pp. 42–48).
4. Hazas, M., & Ward, A. (2002). A novel broadband ultrasonic location system. In Proceedings of the 4thinternational conference on ubiquitous computing (pp. 264–280).
5. Urenna, J., Hernandez, A., Villadangos, J. M., Mazo, M., Garcıa, J. C., Garcıa, J. J., et al. (2007).Advanced sensorial system for an acoustic LPS. Microprocessors and Microsystems, 31, 393–401.
6. Gonzalez, J. R., & Bleakley, C. J. (2009). High-precision robust broadband ultrasonic location andorientation estimation. IEEE Journal of Selected Topics in Signal Processing, 3(5), 832–844.
7. Werb, J., & Lanzl, C. (1998). Designing a positioning system for finding things and people indoors.IEEE Spectrum, 35(9), 71–78.
8. Mazuelas, S., Bahillo, A., Lorenzo, R. M., Fernandez, P., Lago, F., Garcıa, E., et al. (2009). Robustindoor positioning provided by real-time RSSI values in unmodified WLAN networks. IEEE Journal ofSelected Topics in Signal Processing, 3(5), 821–831.
9. Bahillo, A., Mazuelas, S., Lorenzo, R. M., Fernandez, P., Prieto, J., Duran, R. J., et al. (2010). HybridRSS-RTT localization scheme for indoor wireless networks. EURASIP Journal on Advances in SignalProcessing, 2010, 17:1–17:12. doi:10.1155/2010/126082.
10. Tilch, S., & Mautz, R. (2010). Current investigations at the ETH Zurich in optical indoor positioning. InProceedings of the 7th workshop on positioning navigation and communication (Vol. 7, pp. 174–178).
11. Willert, V. (2010). Optical indoor positioning using a camera phone. In Proceedings of the 2010international conference on indoor positioning and indoor navigation.
12. Blankenbach, J., Norrdine, A., & Hellmers, H. (2011). Adaptive signal processing for a magnetic indoorpositioning system. In Proceedings of the 2011 international conference on indoor positioning andindoor navigation.
13. Chung, J., Donahoe, M., Schmandt, C., Kim, I.-J., Razavai, P., & Wiseman, M. (2011). Indoor locationsensing using geo-magnetism. In Proceedings of the 9th international conference on mobile systems,applications and services (MobiSys’11) (pp. 141–154).
14. Cheung, K. C., Intille, S. S., & Larson, K. (2006). An inexpensive bluetooth-based indoor positioninghack. In Proceedings of the 8th international conference on ubiquitous computing.
15. Subramanian, S. P., Sommer, J., Schmitt, S., & Rosenstiel, W. (2007). SBIL: Scalable indoor local-ization and navigation service. In Proceedings of the 3rd international conference on wireless com-munication and sensor networks (pp. 27–30).
16. Wendlandt, K., Berhig, M., & Robertson, P. (2005). Indoor localization with probability densityfunctions based on bluetooth. In IEEE 16th International symposium on personal, indoor and mobileradio communications, PIMRC 2005 (Vol. 3, pp. 2040–2044).
17. Inoue, Y., Sashima, A., & Kurumatani, K. (2009). Indoor positioning system using beacon devices forpractical pedestrian navigation on mobile phone. In UIC 2009, LNCS, (Vol. 5585, pp. 251–265).
18. Mandal, A., Lopes, C. V., Givargis, T., Haghighat, A., Jurdak, A., & Baldi, P. (2005). Beep: 3D indoorpositioning using audible sound. In Proceedings of the 2nd IEEE consumer communications and net-working conference (pp. 348–353).
19. Filonenko, V., Cullen, C., & Carswell, J. D. (2013). Indoor positioning for smartphones using asyn-chronous ultrasound trilateration. ISPRS International Journal of Geo-Information, 2(3), 598–620.
20. Subbu, K. P., Gozick, B., & Dantu, R. (2013). Locateme: Magnetic-fields-based indoor localizationusing smartphones. ACM Transactions on Intelligent Systems and Technology (TIST), 4(4), 73:1–73:27.
21. Ascher, C., Kessler, C., Wankerl, M., & Trommer, G. (2010). Dual imu indoor navigation with particlefilter based map-matching on a smartphone. In 2010 International conference on indoor positioning andindoor navigation (IPIN) (pp. 1–5).
22. Pei, L., Chen, R., Chen, Y., Leppakoski, H., & Perttula, A. (2009). Indoor/outdoor seamless positioningtechnologies integrated on smart phone. In First International Conference on advances in satellite andspace communications, 2009. SPACOMM 2009 (pp. 141–145).
WAY: Seamless Positioning Using a Smart Device
123
23. Gallagher, T., Wise, E., Li, B., Dempster, A., Rizos, C., & Ramsey-Stewart, E. (2012). Indoor posi-tioning system based on sensor fusion for the blind and visually impaired. In Proceedings of the 2012international conference on indoor positioning and indoor navigation (IPIN’12) (pp. 1–9).
24. Liu, J., Chen, R., Pei, L., Guinness, R., & Kuusniemi, H. (2012). A hybrid smartphone indoor posi-tioning solution for mobile LBS. Sensors, 12(12), 17208–17233.
25. Julier, S., Uhlmann, J. K., & Durrant-Whyte, H. F. (2000). A new method for the nonlinear transfor-mation of means and covariances in filters and estimators. IEEE Transactions on automatic control,45(3), 477–482. http://dblp.uni-trier.de/db/journals/tac/tac45.html#JulierUD00.
26. Steiniger, S., Neun, M., & Alistair, E. (2006). Foundations of location based services. In CartouCHe1-lecture notes on LBS (Vol. 1).
27. Fox, D., Hightower, J., Liao, L., Schulz, D., & Borriello, G. (2003). Bayesian filters for locationestimation. IEEE Pervasive Computing, 2(3), 24–33.
28. Wakim, C. F., Capperon, S., & Oksman, J. (2004). A Markovian model of pedestrian behavior. In SMC(4). IEEE (pp. 4028–4033). http://dblp.uni-trier.de/db/conf/smc/smc2004-4.html#WakimCO04.
29. Li, X. (2006). Rss-based location estimation with unknown pathloss model. IEEE Transactions onWireless Communications, 5(12), 3626–3633. http://dblp.uni-trier.de/db/journals/twc/twc5.html#Li06.
30. Pahlavan, K., & Levesque, A. H. (1995). Wireless information networks. New York: Wiley.31. Davies, M. (2003). The standard handbook for aeronautical and astronautical engineers. New York:
McGraw-Hill.32. Guide to Reference and Standard Atmosphere Models, American National Standard ANSI ed.,
American Institute of Aeronautics and Astronautics, 2004.33. Csapodi, M., & Nagy, A. (2007). New applications for NFC devices. In Proceedings of the 16th IST on
mobile and wireless communications summit (pp. 1–5).34. Borriello, G., Liu, A., Offer, T., Palistrant, C., & Sharp, R. (2005). Walrus: Wireless acoustic location
with room-level resolution using ultrasound. In Proceedings of Mobisys 05 (pp. 191–203).35. Kingstate Electronics Corp., Website, (2014). http://www.kingstate.com.tw.36. http://www.gsmarena.com/samsung_galaxy_nexus-review-699p5.php. (2014).
Alfonso Bahillo got his Telecommunications Engineering and PhDdegrees at the University of Valladolid, Spain, in 2006 and 2010,respectively. He got the PMP certification at the PMI in 2014. From2006to 2010 he joined CEDETEL working as research engineering. From2006 to 2011 he has beenworking as assistant professor at the Universityof Valladolid. From 2010 to 2012 he joined LUCE Innovative Tech-nologies as product owner. Currently he is working as postdoc at theUniversity of Deusto and project manager at DeustoTech in Bilbaowhere he trains PhD students and collaborates in several national andinternational research projects. He has worked (leading some of them) inmore than 20 regional, national and international research projects andcontracts. He is co-author of 17 research manuscripts published ininternational journals, more than 30 communications in internationalconferences, and 3 national patents.His interests include local and globalpositioning techniques, ambient assisted living, intelligent transportsystems, wireless networking, and smart cities.
A. Bahillo et al.
123
Teodoro Aguilera obtained his Physics degree from the University ofExtremadura (Spain) in 2009. He is currently an Assistant Professor ofAutomation in the Department of Electrical Engineering Electronicsand Automation at the University of Extremadura, Spain. He is also amember of the Sensory Systems Group of this University, where he ispursuing his Ph.D. degree on the topic of ultrasonic local positioningsystems
Fernando J. Alvarez received his Physics degree from the Universityof Sevilla (Spain), his Electronics Engineering degree from theUniversity of Extremadura (Spain) and his Ph.D. degree in Electronicsfrom the University of Alcala (Spain) in 2006. He is currently anAssociate Professor of Digital Electronics in the Department of Elec-trical Engineering Electronics and Automation at the University ofExtremadura, Spain, where he is also the head of the Sensory SystemsGroup. His research areas of interest include local positioning systems,ultrasonic signal processing and embedded computing
Asier Perallos hold a Ph.D., M.Sc. and B.Sc. in Computer Engineeringfrom the University of Deusto. Since 1999 he has been working as alecturer in the Faculty of Engineering at the University of Deusto. Histeaching focuses on software design and distributed systems, havingtaught several B.Sc., M.Sc. and Ph.D. courses. He is currently the Headof the Computer Engineering Department and Director of M.Sc. inSoftware Engineering at the University of Deusto. He is also HeadResearcher at the DeustoTech Mobility Unit at the Deusto Institute ofTechnology, where he coordinates the research activities of around 25researchers. This research unit promotes the application of ICT toaddress smarter transport and mobility. In particular, Perallos’ researchbackground is focused on telernatic systems, vehicular communicationmiddleware and intelligent transportation systems
WAY: Seamless Positioning Using a Smart Device
123