Indoor navigation based on fusion of positioning signals

Indoor navigation based on fusion of positioning signals
Czech Technical University in Prague
Faculty of transportation sciences
Department of transport telematics
Indoor navigation based on fusion
of positioning signals
MASTER’S THESIS
Author: Nikolai Garmaev
Supervisor: Ing. Petr Bures, Ph.D.
Year: 2014
Před svázáním místo téhle stránky vložíte zadání práce s podpisem děkana (bude
to jediný oboustranný list ve Vaší práci) !!!!
Declaration
I have no relevant reason against using this schoolwork in the sense of § 60 of Act
No121/2000 concerning the authorial law.
I declare that I accomplished my final thesis by myself and I named all the sources
I used in accordance with the guideline about the ethical rules during preparation
of University final thesis.
In Prague ....................
........................................
Nikolai Garmaev
Acknowledgements
I would like to express my deepest gratitude to my supervisor Ing. Petr Bures for
his great patience and priceless feedback.
I would like to thank Andrei Popleteev, Ph.D. for his inspirational input and mr.
Richard Brown for sample code.
I would also like to thank my parents and Ms. Olga Lebedeva for their everlasting
belief and Ms. Oyuna Choybsonova for proofreading.
Nikolai Garmaev
Nazev prace:
Indoor navigace zalozena fuzi pozicnich signalu
Autor:
Nikolai Garmaev
Obor:
Druh prace:
Intelligent Transport Systems
Diplomova prace
Vedouci prace:
Ing. Petr Bures, Ph.D.
Department of transport telematics, Faculty of transportation sciences, Czech Technical University in Prague
—
Konzultant:
Abstrakt: Zatimco urcovani polohy venku je diky GPS vyreseno, stejna uloha je v
interieru je slozitejsi. Satelitni technologie, nemohou tento problem spolehlive vyresit, existuji sice ruzne systemy pro zjistovani polohy v interieru, ale i ty se potykaji s
ruznymi problemy. Kombinace techto systemu by ale mohla prinest kyzeny prulom.
Tato prace ma dva hlavni cile, zjisteni soucasneho stavu systemu urcovani polohy v
interieru a vyber 2 vhodnych kandidatu (z hlediska vlastnosti i z hlediska dostupnosti) pro kombinaci technik urceni polohy za pomoci otisku site a RSSI. V praci
provedene experimenty ukazuji, ze fuze ruznych technik muze byt prospesna, prima
kombinace technik muze poskytnout presnost okolo 4 metru.
Klicova slova:
určování pozice v budovách, knn algoritmus, určování pozice dle
otisku wlan, VKV vysílání, rssi
Title:
Indoor navigation based on fusion of positioning signals
Author:
Nikolai Garmaev
Abstract: Indoor positioning has gained a lot of interest during last years and thanks
to GPS, determination of one’s position outdoor is almost solved. However, it’s not
the case indoors, mainly because of multipath, NLOS and interference and so far
no technology from the variety of indoor positioning systems can be considered as a
general solution. In order to manage these problems a substantial effort was made to
combine different technologies together with the idea of emphasizing their strengths
and lessening their drawbacks. This thesis has two major purposes: to investigate
existing indoor technologies and to fuse the most suitable ones together by employing
a Received Signal Strength fingerprinting approach. The experiments in a CVUT
faculty building indicate that fusion of different techniques can be beneficial and
even direct combination of techniques can provide accuracy of 4 m.
Key words:
indoor positioning, knn algorithm, wlan fingerprinting, fm broad-
Contents
Introduction
9
1 Indoor positioning methods
11
2 Wireless technologies
14
2.1
2.2
2.3
2.4
GSM & CDMA . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.1.1
Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.1.2
System example . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.1.3
Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
WLAN (IEEE 802.11) based systems . . . . . . . . . . . . . . . . . . 16
2.2.1
Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.2.2
System example . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.2.3
Enhancements and remarks . . . . . . . . . . . . . . . . . . . 17
2.2.4
Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
Bluetooth (IEEE 802.15) based systems . . . . . . . . . . . . . . . . . 18
2.3.1
Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.3.2
System example . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.3.3
Enhancements and remarks . . . . . . . . . . . . . . . . . . . 19
2.3.4
Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
FM-radio based systems . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.4.1
Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.4.2
System example . . . . . . . . . . . . . . . . . . . . . . . . . . 21
6
2.4.3
2.5
2.6
2.7
2.8
2.9
Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
RF based systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.5.1
Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.5.2
System example . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.5.3
Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
UWB systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.6.1
Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.6.2
System example . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.6.3
Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
Ultrasound positioning systems . . . . . . . . . . . . . . . . . . . . . 26
2.7.1
Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.7.2
System example . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.7.3
Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
Optical indoor positioning . . . . . . . . . . . . . . . . . . . . . . . . 27
2.8.1
Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.8.2
System example . . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.8.3
Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
IR-based . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
2.9.1
Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
2.9.2
System example . . . . . . . . . . . . . . . . . . . . . . . . . . 30
2.9.3
Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
2.10 Other positioning systems . . . . . . . . . . . . . . . . . . . . . . . . 31
3 Fusion proposal and its possible benefits
33
3.1
Grounds for fusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
3.2
Selected method for fusion . . . . . . . . . . . . . . . . . . . . . . . . 34
3.3
Techniques to be fused . . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.3.1
Wi-Fi . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
3.3.2
FM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
7
3.3.3
Conclusion to RSSI properties . . . . . . . . . . . . . . . . . . 39
4 System proposition
40
4.1
General approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
4.2
Positioning approach . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
4.3
Classification approach . . . . . . . . . . . . . . . . . . . . . . . . . . 41
4.4
kNN algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
4.5
Related work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
5 System implementation
43
5.1
Testbed . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
5.2
Data collection setup . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
5.3
Data preprocessing . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
5.4
Data processing software . . . . . . . . . . . . . . . . . . . . . . . . . 50
5.5
Software algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
5.6
Testing campaign . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
Conclusion
60
References
62
List of terms specific to the thesis
66
Appendices
67
.1
Indoor positioning technologies comparison . . . . . . . . . . . . . . . 68
.2
FCC WG-3 trials results . . . . . . . . . . . . . . . . . . . . . . . . . 68
.3
Correlation matrices . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
.4
RSS fluctuations due to user’s orientation
.5
Real readings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
.6
Experiments results . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
8
. . . . . . . . . . . . . . . 70
Introduction
Recently, considerable attention in the area of location-based services has been paid
on research and development of indoor positioning and, subsequently, indoor navigation systems.
It covers a wide variety of situations ranging from communication with individuals
moving in residential or office buildings, hospitals or factories to location detection of
products stored in a warehouse and finding tagged maintenance tools and equipment
scattered all over the area.
Thanks to growing popularity of mobile wireless devices and proliferation of the
GPS/GLONASS, combined with Wi-Fi and cellular networks, the problem of outdoor localization is practically solved. But for the indoor environment GPS is not
applicable, mainly because of impossibility of Line-of-Sight transmission between
satellites and receivers - various obstacles, e.g. walls, equipment, moving people influence the propagation of electromagnetic waves. In addition, due to a phenomenon
known as “multipath fading “, the transmitted signal often reaches the receiver by
more than one path because signal propagation is strongly affected by construction
materials, scattering of radio waves and multiple reflections from structures inside
the building. [38]
Therefore, an optimal solution for indoor navigation hasn’t been proposed yet since
existing IPS are either expensive in terms of infrastructure (UWB, ultrasound),
have limited coverage (Wi-Fi, Bluetooth, RFID) or low accuracy (cellular networks).
Chapter I introduces methods for indoor positioning, while in Chapter II I will review all existing technologies available for indoor positioning together with their
advantages and disadvantages. Chapter III and IV make a proposition of an indoor
positioning system, which is able to estimate user’s location in the particular environment. Results from experiments in a testbed are in Chapter V. This paper is
different from the previous survey papers [14] and [21] in several ways. In the first
paper, authors only describe known IPS categorizing them on a basis of positioning
algorithms as well as the technologies used, while this paper concentrates on comparison of techniques for the purpose of fusing them in a tangible way. The second
9
paper emphasizes security and privacy issues of indoor navigation, which is not in
my scope of research.
It should be noted, that design of a complete navigation system may be quite a
sophisticated task taking into account that it can be difficult to discover orientation
or direction of the object, thus scope of this thesis was limited to detect an object
in a certain known fixed location or report its presence.
10
Chapter 1
Indoor positioning methods
This chapter will review location positioning algorithm, i.e., the method for determining location, making use of various types of measurement of the signal such as
TOF, angle, and signal strength.
1. Proximity based method, as shown in 1.1, determines position of an object
based on its closeness to a reference point in physical space - beacon with
known positions and limited range, so that only one or few beacons are visible
to the mobile unit at any point. The client location is then approximated as
that of the nearest beacon.
Figure 1.1: Proximity-based method
2. Angle of arrival (AoA) is a method for determining the direction of propagation of a radio-frequency wave, which requires only two beacons to estimate
position in 2D (three beacons for 3D localization). (Fig. 1.2)
11
Figure 1.2: Angle of arrival method
3. In time of arrival (ToA), Fig. 1.3, synchronized clocks in the base station
and the client are used to measure the time delay between the two, while the
time difference of arrival (TDoA) uses the difference of time it takes the signal
from the client to reach each of the synchronized beacons.
Figure 1.3: Time of arrival method
Finally, there are two different approaches that use Received Signal Strength Indication (RSSI), namely:
1. Propagation modeling, which attempts to build a model of the signal propagation in the space in order to identify the distance between the user and
beacons, see Fig. 1.4. However, this approach is best suited for line-of-sight
and obstacle-free propagation – conditions which are rarely met indoors.
12
Figure 1.4: Propagation modeling method
2. Fingerprinting, as shown in Fig. 1.5, consists of two phases: calibration and
localization. It relies on a database associating RSSI measurements with corresponding coordinates and then uses statistics and machine learning algorithms
in order to recognize user position among those learned during the training
phase [32].
Figure 1.5: Fingerprinting method
Indeed, to handle with ambiguity of signals any locating service requires at least
three independent measures per target, to which some mathematical algorithm must
be applied subsequently to combine several sensors inputs with the idea to reduce
error accumulation or compensate discrepancies in collected values.
In the following section, existing indoor positioning technologies will be described.
13
Chapter 2
Wireless technologies
2.1
2.1.1
GSM & CDMA
Overview
Cellular networks, such as GSM and CDMA, are well-developed technologies with
more than 7 billion worldwide GSM subscribers1 and more than 500 million CDMA
subscribers in 20132 were not considered for indoor localization for a long time due
to the low accuracy demonstrated in outdoor settings and typically do not show
reasonable potential indoors because the signal strength is too low to penetrate a
building.
Generally speaking, the accuracy is higher in densely covered areas (e.g. urban areas)
and much lower in rural environments.
Indoor positioning based on mobile cellular network is possible if the building is
covered by several base stations or one base station with strong RSS received by
indoor mobile clients.
2.1.2
System example
Otsason et al.[28] presented a GSM-based indoor localization system, which uses
wide signal-strength fingerprints. The wide fingerprint includes the six strongest
GSM cells and readings of up to 29 additional GSM channels, most of which are
strong enough to be detected but too weak to be used for efficient communica1
http://www.itu.int/en/ITU-D/Statistics/Documents/facts/ICTFactsFigures2014-e.pdf
http://www.statista.com/statistics/206604/global-wireless-subscription-growth-bytechnology-since-2010/
2
14
tion. The higher dimensionality introduced by the additional channel dramatically
increases localization accuracy. The results for experiments conducted on signalstrength fingerprints collected from three multifloor buildings using weighted kNN
technique showed that their indoor localization system does a reasonable job differentiating between doors and achieves median accuracy between 2.5 and 5.4 meters.
Speaking about localization using CDMA, CILoS is based on delays finngerprinting,
an empirical localization technique that involves a training or mapping phase in
which a radio map of the environment is constructed by collecting a series of fingerprints in multiple locations. Using a special Condor CDMA scanner, it was able
to evaluate signal delays from nearby stations. Unlike the RSSI, signal delays were
found to be rather stable in time and resilient to cell resizing. Using signal delay
fingerprints, this system reached median localization accuracy between 4.5 and 6.7
m. [35]
2.1.3
Conclusion
Cellular network based indoor positioning systems have three main advantages:
∙ Coverage: unlike Wi-Fi, the GSM/CDMA networks are currently widely available in most countries; the size of large macrocells can reach 30 km.
∙ Low cost: While GSM/CDMA base stations are themselves very expensive
(up to 1 million USD)3 , the costs are covered by the cellular network operator
(and ultimately, the subscribers). Thus, the positioning system can exploit
readily available stations and does not require installation of a dedicated indoor
infrastructure as Wi-Fi does.
∙ Battery life: Although a cellular transceiver module is rather battery consuming even in an idle state, in many scenarios it remains powered in order to
provide the voice or data connectivity. Thus, the overhead introduced by a
positioning system relates only to location estimation and excludes powering
additional wireless module, which is often the case for Wi-Fi.
However, GSM/CDMA positioning has also several shortcomings:
∙ Low accuracy: The presented works [28] [35] rely on the use of wide fingerprints
in order to provide a good accuracy. Acquisition of extended data, however,
requires special hardware (programmable GSM modem and CDMA scanner),
3
Otsason, A.V. Accurate GSM indoor localization, 2005
15
with the narrow fingerprints which could be acquired with conventional hardware, the localization accuracy was rather low.
∙ Low reliability: Given that GSM/CDMA beacons are situated outdoors, the
signal propagation conditions vary due to environmental factors, such as weather
and terrain. In particular, radio signals with frequencies above 1 GHz are affected by rain scatter interference and terrain vegetation; trees in leaf can
cause a 20% higher attenuation than leafless trees. In theory, these factors
can significantly affect the positioning performance; however, no experimental
studies are available yet. [32]
2.2
2.2.1
WLAN (IEEE 802.11) based systems
Overview
This midrange wireless local area network (WLAN) standard, operating in the 2.4GHz Industrial, Scientific and Medical (ISM) band, has become very popular in
public hotspots and enterprise locations during the last few years. With a typical
gross bit rate of 11, 54, or 108 Mbps and a range of 50–100 m, IEEE 802.11 is
currently the dominant local wireless networking standard.
It is, therefore, appealing to reuse an existing WLAN infrastructure for indoor location as well, which lowers the cost of indoor positioning system deployment. [21]
The accuracy of location estimations based on the signal strength of WLAN signals
is affected by various elements in indoor environments such as movement and orientation of human body, the overlapping of Access Points (AP), the nearby tracked
mobile devices, walls, doors, etc. The influence of these sources and their impacts
have been discussed and analyzed in the literature. [14]
2.2.2
System example
One of the pioneering projects in RSSI-based Wi-Fi positioning was RADAR. The
authors applied both propagation modelling and fingerprinting, employing signal
strength and signal-to-noise ratio with the triangulation location technique. The
multiple nearest neighbors in signal space (NNSS) location algorithm was proposed,
which needs a location searching space constructed by a radio propagation model.
The RADAR system can provide 2D absolute position information and thereby
enable location-based applications for users. In the experiments of the RADAR
16
system, three APs measured the signal strength of the RF signals from the target.
Then these measurements were used to calculate a 2-D position of the object. The
median error distance of fingerprinting method is 2.94 meters. Also the error distance
for tracking the moving user is 3.5 meters that is about 19% worse than for a
stationary user, while the radio propagation model with the 50th percentile provides
an error distance of about 4.3 m.
It was also stated that despite the physical proximity between points on adjacent
floors, signal aliasing between a point on a floor and the corresponding point on
an adjacent floor is unlikely because the floor acts as a significant barrier to signal
propagation. Based on measurements, authors conclude that RADAR would work
well in a multi-floor environment. Of course, a radio map of all of the floors, not
just of one floor, would have to be constructed. [5]
2.2.3
Enhancements and remarks
RADAR was improved by the original authors putting Viterbi-like algorithm instead
of NNSS and NNSS-AVG. It significantly improves accuracy, outperforming both of
them, for instance, the median error distance for NNSS (3.59 m) and NNSS-AVG
(3.32 m) are 51% and 40% worse, respectively, compared to Viterbi-like algorithm
(2.37 m). [5]
Brunato and Battiti compared the performance of Wi-Fi fingerprinting localization for several machine learning methods, such as multi-layer perceptron (MLP),
support vector machine (SVM) and k-nearest neighbor (kNN), both weighted and
unweighted. The SVM approach demonstrated the best median accuracy (2.75 m).
Notably, the median performance of a simple unweighted kNN classifier was only
0.16 m less, while 95th percentile errors were almost the same (6.09 m for SVM and
6.10 m for kNN). [6]
Chen et al. investigated the dependence of the Wi-Fi positioning accuracy on such
environmental factors as humidity, doors, and people presence. Door states (all open
or all closed) and people presence in receiver’s vicinity were found to have a significant impact on positioning error (236% and 86% increase, respectively), while the
humidity had smaller effect (43% increase). While such degradation of performance
is typical for fingerprinting based systems, the impact of each component varies with
signal frequency: when the obstacles are small in comparison to wavelength, their
interaction with the wave is negligible. [12]
17
2.2.4
Conclusion
After all, Wi-Fi based positioning systems have several advantages, such as:
∙ Leveraging the already widely deployed infrastructure,
∙ Wide availability in mobile devices,
∙ Good accuracy.
However, there are certain limitations:
∙ Limited coverage. Despite the popularity, the coverage of Wi-Fi networks are
mostly concentrated in office buildings and dense urban areas. Wi-Fi networks
are rare in less populated cities and developing countries
∙ Interference. The 2.4 GHz industrial, scientific and medical (ISM) band used
by Wi-Fi is shared by many other electronic devices, such as cordless phones
and microwave ovens, which may interfere with Wi-Fi signals and affect the
positioning accuracy.
∙ Power consumption. Another factor is power efficiency of the positioning system, especially on the battery powered mobile devices. Wi-Fi modules have
a substantial power consumption about 300 mW in idle power-saving mode4 ,
which shortens the battery life of the mobile device.
2.3
2.3.1
Bluetooth (IEEE 802.15) based systems
Overview
Bluetooth, the IEEE 802.15.1 standard, operates in the 2.4-GHz ISM band. Bluetooth enables a range of 100 m (Bluetooth 2.0 standard) communication and it’s
highly ubiquitous, being implanted in various types of devices such as mobile phones,
laptops, desktop PC’s, etc.
Bluetooth chipsets are small size transceivers of low cost: the high expected production volumes (hundreds of millions annually) lead to less than 5 USD per chip,
which results in low price tracked tags used in the positioning systems.
4
Anand, M. et al. Self-tuning wireless network power management
18
Moreover, Bluetooth hardware and communication protocol have been designed with
a focus on low power consumption. All of this makes Bluetooth an interesting technology for indoor positioning, and there are several works [42], [17] dedicated to
Bluetooth based localization systems.
However, the coverage of such systems is very limited due to the short range of
Bluetooth modules, and, more importantly, the lack of stationary Bluetooth devices. Another drawback is that each location acquisition runs the device discovery
procedure, which significantly increases both the localization latency (10–30 s) and
power consumption. [32]
2.3.2
System example
The Topaz location system is a local area positioning software and hardware system
that calculates local position of Bluetooth tags and other devices (e.g. mobile phones,
PDAs, etc.).
By using Bluetooth technology, Topaz can only provide 2-D location information
with an error range of around 2 m, which is not sufficient to provide room level
accuracy in a multi-obstacle indoor environment. Thus the Topaz system combines
the Tadlys’ Bluetooth-based positioning infrastructure with IR-based positioning
technique, where IR location technology is suitable for this goal.
This modular positioning solution consists of positioning server, IR-enabled wireless
access points, and wireless tags as well as software parts for local positioning of
Bluetooth tags.
The system’s performance makes it suitable for tracking humans and assets. A score
of objects can be tracked simultaneously. This system provides roomwise accuracy
(or, alternatively, 2 m. spatial accuracy), with 95% reliability. The positioning delay
is 15–30 s. And the tags using batteries need to be charged once per week, which is
a short period compared with tags used in other positioning systems. [42]
2.3.3
Enhancements and remarks
Another Bluetooth-based system example presented by [17] consists of several fixed
stations and a mobile station and then applies the trilateration method to three
to five distance measurements. One of the fixed stations is connected to PC, which
serves as a position calculation server. Each fixed station and the mobile station
are composed of a Bluetooth module and a microcomputer. Density of stations was
0.02 st./m2 (6 stations in 15*20 m area). To overcome the attenuation of a human
19
body an additional mobile station was attached on a subject’s back resulting in an
improvement of standard deviation from 2.8 m to 1.9 m and with higher density
of fixed stations obtained accuracy is 1.2 m, which is satisfactory for positioning in
typical workshops or roomwise positioning in cluttered environment.
2.3.4
Conclusion
To conclude, advantages of Bluetooth-based positioning systems are:
∙ Deployment of devices, already equipped with Bluetooth technology,
∙ Low-cost solution,
∙ Low power consumption,
While the disadvantages of Bluetooth-based positioning system are:
∙ Accuracy only from 1.5 m to 3 m with the delay of about 20 s.,
∙ Susceptibility to interference in ISM band. [14]
Therefore, Bluetooth is commonly agreed to be unsuitable for localization systems,
unless future Bluetooth specification decides to make Received Power (RX) level
available not only through Received Signal Strength Indication (RSSI), currently
defined very loosely, but directly. [19]
2.4
2.4.1
FM-radio based systems
Overview
Potential of FM-based positioning as-of-yet is not so well-investigated comparing it
to Wi-Fi positioning technique, which is prevailing now on the indoor positioning
market.
FM radio employs the frequency-division multiple access (FDMA) approach which
splits the band into a number of separate frequency channels that are used by
stations. FM band ranges and channel separation distances vary in different regions,
as shown in Table 2.1
Hereafter, under “FM” we generally imply radio waves of the corresponding frequencies rather than to modulation type.
20
Table 2.1: FM broadcast frequencies and channel spacing in different region
Region
Europe
US
Japan
Frequency range Channel spacing
87.5 – 108.0 mHz
100 kHz
87.7 – 108.0 mHz
200 kHz
76.0 – 90 mHz
100 kHz
Table taken from [32]
The major difference of FM radio signals from other technologies, such as Wi-Fi,
GSM or DECT, is defined by the significantly (9 to 50 times) lower operational
frequencies. The low frequency provides the FM localization with a number of advantages:
∙ FM signals are less affected by weather conditions,
∙ Low frequency radio waves are less sensitive to the terrain conditions,
∙ The attenuation of radio waves by building materials increases with frequency
and thus FM signals penetrate walls more easily than Wi-Fi or GSM.
One of the problems related to FM is the so-called capture effect, the phenomenon
in which only the station with the strongest signal will be demodulated and reach
the receiver’s output, while the other will be attenuated.
The most crucial problem of using an FM signal, however, is that they do not carry
any timing information, which is a critical factor in range calculation. Measurements
that can be taken from FM signals for navigation purposes are based on: Time of
Arrival (TOA), Time Difference of Arrival (TDOA), Angle of Arrival (AOA), and
Received Signal Strength (RSS). For the first three methods, the lack of timing information in FM signals is critical. Hence, the most appropriate choice is localization
based on RSS and signal propagation modeling. [25]
2.4.2
System example
In [25], authors chose a fingerprinting technique of an area 11*23 m consisting of 7
rooms with the corridor, which is a typical indoor office environment and filled it with
150 Reference Points (RP), 28 Test Points, that people are most likely to require and
took 17 FM channels from 88 to 108 MHz. Then, applying the K-nearest neighbor
and K-nearest weighted neighbor algorithms, they have acquired results around 3
m.
21
Another example of positioning system based on FM signals so called FINDR [30]
also uses short-range FM transmitters as wireless beacons and measures Received
Signal Strength (RSS) by a fingerprinting approach.
Their results were strongly correlated to previous works, meaning the results of the
system evaluation have shown a median accuracy of about 1.0 m and 5.0 m at 95%
confidence level, which was close to Wi-Fi characteristics in chosen conditions.[30]
Further evaluation of FINDR by the same group has shown some improvement by
using KNN algorithm and Gaussian Process (GP) regression. The median estimation
error (50th percentile) of the system was 0.97 m for GP and 0.93 m for kNN while
95th percentile error was 2.65 m for GP and 3.88 m for kNN. [23]
2.4.3
Conclusion
FM technology has some advantages:
∙ Availability in majority of stationary and mobile devices,
∙ Power effectiveness: on average Wi-Fi consumes around 300 mW, while FM
receivers consume around 15 mW. [43], [44]
∙ Safety in particular environments, e.g. medical facilities, where Wi-Fi cannot
be used because of interference with many other electronic devices,
∙ Cost-effectiveness: an FM transmitter is up to 10 times cheaper than a Wi-Fi
access point while also widely available off-the-shelf.
Still, there are drawbacks of using FM technology for indoor positioning, mainly
∙ Multipath,
∙ NLOS signals,
∙ Capture effect,
∙ No timing information.
22
2.5
2.5.1
RF based systems
Overview
The radio frequency identification (RFID) is a means of storing and retrieving data
through electromagnetic transmission to an RF compatible integrated circuit.
It is widely used for asset tracking, shop security systems and complex indoor environments such as office, hospital, etc. Due to the short communication range (dozens
of centimeters), it provides a good localization accuracy. The short reading distance,
however, also significantly limits its possible application areas.
There are two kinds of RFID technologies, passive RFID and active RFID, as shown
in Table 2.2
With passive RFID, a tracked tag is a receiver, thus the tags with passive RFID are
small and inexpensive, but the coverage range of tags is short.
Active RFID tags are transceivers, which actively transmit their identification and
other information, thus their cost is higher, on the other hand, the coverage area of
active tags is larger. [21]
Table 2.2: Active and passive RFID comparison
Passive
Active
Up to 40ft (fixed reader) and
Read range
up to 300ft
up to 20 ft (handheld reader)
Power
No power source
Battery-powered
Up to 10 years
Tag life
3-8 years depending on a tag
depending upon the environment
Tag costs
from 10cents to 4 USD
from 15 to 50 USD
Assets inventorying,
Real-time asset
Perfect use
assets tracking
monitoring
Readers
Typically higher cost
Typically lower cost
Source: www.inlogic.com/rfid/passive_vs_active.aspx
2.5.2
System example
From the variety of IPS based on RFID the one that can be regarded as a showcase
is LANDMARC. Its prototype uses the RFID reader’s operating frequency with
308 MHz. In order to increase accuracy without placing more readers, the system
23
employs the idea of having extra fixed location reference tags to help location calibration.
These reference tags serve as reference points in the system. The LANDMARC
approach requires signal strength information from each tag to readers, if it’s within
the detectable range. Then, the kNN method is adopted to calculate the location of
the RFID tags. It is reported that the 50 percentile has an error distance of around
1 m while the maximum error distances are less than 2m for LANDMARC system.
[27]
2.5.3
Conclusion
RFID-based IPS have plenty of valuable advantages for positioning, such as:
∙ Size, weight and cost of tags which could be tracked or embedded in a given
location,
∙ Possibility of unique identification of multiple objects,
∙ Remarkable accuracy compared to other technologies.
However, while RFID based systems can accurately detect proximity and determine
absolute position, its drawbacks are:
∙ Dense infrastructure is required to fully cover big working area,
∙ Sporadic location updates,
∙ Rather short battery life.
Based on the above, RFID based systems are considered unsuitable for generalpurpose indoor localization. [32]
2.6
2.6.1
UWB systems
Overview
UWB is based on sending ultrashort pulses (typically <1 ns), with a low duty cycle
(typically 1:1000).
24
Unlike conventional RFID systems, which operate on single bands of the radio spectrum, UWB transmits a series of signals in the time domain, which in turn spreads
information over multiple bands of frequencies simultaneously, from 3.1 to 10.6 GHz.
UWB ranging systems often measure the time of arrival (TOA) of signals travelling
between a target node and a number of reference nodes. The transmitter is either a
mobile unit placed on the pedestrian, or an “access point” mounted on a known location inside the building. Three TOA are necessary to estimate the mobile position,
which requires the receiver and the transmitter clocks to be precisely synchronized.
This difficulty can be avoided by using time difference of arrival (TDOA), because
UWB systems can also measure the angle of arrival (AOA) of radio signals in order
to determine positions.
Two different angles are measured for each AOA; one of them is measured in a
vertical plane, and the second is measured in the horizontal plane. Two measures of
AOA from, at least, two different access points are necessary to compute the target
location. [29]
2.6.2
System example
One of the systems, which demonstrated very good localization accuracy, is Ubisense
- commercially available indoor localization system, which employs TDOA and AOA
methods for UWB radio signals. It consists of a central computer equipped with
the Ubisense software platform connected with all access points, which computes
3D positions of the mobile unit and controls the pulses emission frequency to the
mobile. Ubisense is capable of achieving 15-30 cm accuracy in three dimensions.
However, the system has a very high cost (An active research package costs about
16875 USD) which severely impacts wide adoption. [21]
2.6.3
Conclusion
The main advantages of utilizing UWB for positioning purposes are:
∙ Extreme accuracy because of a very large signal bandwidth and short pulse
duration,
∙ Less power consumation than conventional RF tags and ability to operate
across a broad area of the radio spectrum,
∙ Reduced interference to other RF signals and systems because of the absence
of carrier frequency and the low power spectral density,
25
∙ No multi-path distortion: UWB short duration pulses are easy to filter in order
to determine which signals are correct and which are generated from multipath,
which is essential in localization applications.
∙ No LOS requirement and high penetration ability resulting in a possibility of
a UWB signal to easily pass through walls, equipment and clothing. [21]
Despite promising technical characteristics, the disadvantages of UWB positioning
are:
∙ Strong signal interference with metallic and liquid materials highly affects
UWB positioning performance,
∙ Communication distance of UWB is only 10 m, well lower than Wi-Fi, RFID
and other IPSs,
∙ High infrastructure cost. [29]
2.7
2.7.1
Ultrasound positioning systems
Overview
Making use of ultrasound technology for positioning is rather simple: inexpensive
nodes (badges/tags) attached to the surface of persons, objects and devices, which
then transmit an ultrasound signal to communicate their locations to microphone
sensors.
Because ultrasound signal wavelengths have short reach, they are confined to lesser
distant locations than with wireless transmissions with higher susceptibility to multiple reflection, multipath and through-the-wall multiple room responses. Hence
ultrasound-based RTLS is considered a more robust alternative to passive radiofrequency identification (pRFID) and even to active radio-frequency identification
(aRFID) in complex indoor environments (such as hospitals), where radio waves get
multiply transmitted and reflected, thereby compromising the positioning accuracy.
2.7.2
System example
Cricket is a location system with the aim of offering user privacy, efficient performance and low cost. The cricket system uses TDOA measuring method and triangulation location technique to locate a target.
26
The cricket system includes ultrasound emitters as infrastructure attached on the
walls or ceilings at known positions, and a receiver mounted on each object to be
located. This approach provides privacy for the user by performing all the position triangulation calculation locally in the located object, which allows the located
object to decide how and where to publish its location information.
The Cricket system uses emitters fixed on the ceiling, while the target object receives
and processes the ultrasound signals to locate itself, which allows the system is
scalable for large area deployment inside a building, and the object receiver is cheap
(about 10 USD), so the cost of the whole system is low.
Moreover, the Cricket system can provide a position estimation accuracy of 10 cm
and an orientation accuracy of 3∘ . However, the located receivers in the system
perform location estimations and receive both ultrasound and RF signal at the
same time.
Thus a receiver in the cricket system consumes more power, and its power supply
needs to be designed in an efficient way to bring convenience to the users instead of
frequently changing batteries in the receiver [8]
2.7.3
Conclusion
Benefits of ultrasound positioning systems are:
∙ Low price compared to UWB and RFID-based systems,
∙ Basically low coverage area.
And as drawbacks of such systems can be stated:
∙ They usually have to be combined with RF signals, which perform synchronization and coordination in the system,
∙ Incredible sensibility even to small obstacles, reflected ultrasound signals and
other noise sources such as hanging metal objects, crisp packets, etc. [21]
2.8
2.8.1
Optical indoor positioning
Overview
Optical indoor positioning systems can be categorized into ego-motion systems where
a mobile sensor (i.e. the camera) is to be located and static sensors that locate
27
moving objects in the images. All camera-based system architectures measure image
coordinates that represent only angular information and exclusively built on the
Angle of Arrival (AoA) technique. There are different systems approaches in optical
positioning classified by reference:
∙ Reference from 3D building models:
This class of positioning methods relies on the detection of objects in the images and matching those objects with a building data base (such as CityGML)
that contains position information of the building interior. The key advantage
of these methods is that there is no requirement for the installation of local
infrastructure such as the deployment of sensor beacons.
∙ Reference from images:
The so-called view-based approach relies on sequences of images taken beforehand by a camera along certain routes in the building. Thereby, the current
view of a mobile camera is compared with these previously captured view
sequences. The main challenge of this approach is to achieve real-time capability. For the identification of image correspondences the computational load
is particularly high since operability is assumed without deployed passive or
active optical targets. Nevertheless, all systems require an independent reference source from time to time in order to control the accumulated error.
∙ Reference from deployed coded targets:
Optical positioning systems that rely entirely on natural features in the images
lack of robustness, in particular under conditions with varying illumination.
In order to increase robustness and improve accuracy of reference points, dedicated coded markers are used for systems with demanding requirements for
positioning. Common types of targets include concentric rings, barcodes or patterns consisting of colored dots. There are retro-reflective and non-reflective
versions.
∙ Reference from projected targets:
The projection of reference points or patterns spares the physical deployment
of targets in the environment, making this method economical. For some applications the mounting of reference markers is undesirable or not feasible. In
contrast to systems relying only on natural image features, the detection of
projected patterns is facilitated due to their distinct color, shape and brightness.
28
∙ Systems without reference:
The purpose of systems in this class is to observe position changes of objects
directly and therefore do not require external reference. The common approach
is to track mobile objects with high frame rates in real-time by a single or
multiple static cameras. [24]
2.8.2
System example
Kohler et al. have built a model called TrackSense consisting of a projector and
a simple webcam. Then grid pattern is projected onto plain walls in the camera’s
field of view. Using an edge detection algorithm and triangulation, the distance and
orientation to each point relative to the camera is computed. The evaluation of
TrackSense indicates that such a system can deliver up to 4cm accuracy with 3cm
precision. [18]
2.8.3
Conclusion
Main benefits in using optical PS are:
∙ Low-cost camera can cover a large area,
∙ Users don’t need to carry any additional device and can be tracked only by
camera.
But this approach has significant drawbacks:
∙ The privacy of people is not provided by such kind of a system,
∙ Vulnerability to many interference sources (light, weather, etc.),
∙ Less robust in a dynamic changing environment,
∙ Tracking multiple objects simultaneously can be a hard task even for a smart
camera. [14]
29
2.9
IR-based
2.9.1
Overview
IR-positioning systems are very common positioning systems because IR-technology
is available in various wired and wireless devices such as TV, printer, mobile phones,
etc.
An IR-based positioning system, which offers absolute position estimations, every
node emits IR impulses, which are received by stationary catchspot (receiver) and
location then is computed by the TOF. It requires line-of-sight communication between transmitters and receivers without interference from strong light sources. Thus
the coverage range per infrastructure device is limited within a room.
2.9.2
System example
OPTOTRAK PROseries was designed by Northern Digital Inc. for congested shops
and workspaces. It uses system of three cameras as a linear array to track 3D positions of numerous markers on an object, which can cover a volume of 20 cbm and
a maximum distance between tracked targets and the tracker is about 6.0 m. The
system is a type of active system, where markers mounted on different parts of a
tracked object emits IR light that is detected by the camera to estimate the location
of them. The triangulation technique is used in the positioning process to calculate
the positions of IR light emitters in the space. The system can offer a high accuracy
of 0.1 mm to 0.5 mm with 95% success probability [31]
2.9.3
Conclusion
∙ Very accurate positioning estimations (mm.),
∙ IR emitters are small, light-weight and easy to be carried by a person,
∙ The system architecture is simple and does not need time-consuming installation and maintenance.
However, certain disadvantages of IR-based systems prevent it from general usage:
∙ Interference from fluorescent light and sunlight,
30
∙ Expensive system hardware requirements. Although the IR emitters are cheap,
the whole system using camera array and connected via wires is expensive
comparing to the coverage area,
∙ The system fails to work, when an IR device is taken by a person covered by
his/her clothes since the IR wave cannot penetrate opaque materials. [21]
2.10
Other positioning systems
MEMS result from the integration of mechanical and electrostatic elements on a
common substrate. Sensors based on this technology are essentially accelerometers,
gyroscopes and magnetometers. Inertial data from these systems are used for dead
reckoning navigation where the current position is estimated by accumulating movements determined using onboard measurements. The advantages of inertial measurements are their regularity and their independence from any existing infrastructure.
MEMS hardware is also compact and relatively cheap compared to other high-end
inertial systems.
The magnetic positioning systems offer high accuracy and do not suffer from the
line-of-sight problems, where the positions are measured in the case of an obstacle between the transmitters and receivers. For example, MotionStar Wireless is a motion
tracking system that uses pulsed direct current magnetic fields to simultaneously locate up to 120 sensors within 3 m coverage area in real time. The systems consist of
a transmitter and controller, a base station, mounted sensors and RF transmitters.
The transmitter and controller send magnetic pulses to the body mounted sensors,
which The sensors are connected through wires to the RF transmitter, which is carried by the tracked person. Then RF transmitter transmits the measured data to
the base station. Finally, the base station calculates the position and orientation of
sensors and transfers the measured data to the user’s computer. The error range of
the static position estimating is about 1 cm. The update rate of the position measurements is up to 120 measurements per second. However, the disadvantage of the
Motion Star system is that the magnetic trackers are quite expensive. The battery
life time for continuous motion tracking is around 1 hour or 2 hours, which is a short
period for daily position estimations and the performance of the Motion Star system
is influenced by the presence of metal elements in the positioning estimating area.
In addition, the coverage range of each transmitter is limited within 3 m, which is
not scalable for large indoor public applications and services. [26]
Locata Corporation has invented a positioning technology called Locata, for precision positioning both indoors and outside. Part of the “Locata technology” consists
31
of a time-synchronized pseudolite transceiver called a LocataLite. A network of
LocataLites forms a LocataNet, which transmits GPS-like signals that allow singlepoint positioning using carrier-phase measurements for a mobile device. Indoor industrial machine tracking showed subcentimeter precision: crane moved to 9 known
points and max position error was 1.8 cm, while max. absolute error in orientation
test was 1.2∘ but multipath still caused problems. [22]
32
Chapter 3
Fusion proposal and its possible
benefits
During last years the main problems in indoor positioning were identified and plenty
of efforts were made to mitigate NLOS, multipath and achieve high-accuracy ranging. Compared with satellite channel of GNSS, indoor positioning faces terrestrial
channel which is more complex. The high-accuracy ranging information based on
time delay and Received Signal Strength (RSS) is the key information for positioning. The phenomenon like multipath and fast fading is much more serious in
terrestrial channel, especially in urban indoor environment. [13]
3.1
Grounds for fusion
As it follows from the overview of technologies, at the present moment there is no
technology for indoor positioning, which will satisfy various potential users, because
neither of them is able to provide an accurate positioning for adequate amount of
money.
For instance, Wi-Fi is not suitable for positioning in rural areas and prone to interference from 2.4 GHz devices; RFID and UWB while assuring cm-precision, require
investments in a subject area.
The main idea behind this thesis is that by combining two or more techniques these
bottlenecks could be avoided or at least diminished, it is well-known and has shown
decent results. [9], [11]
Fusion of techniques, which complement each other, also can provide better accuracy
or area coverage, but the selection must be done in an appropriate way in order to
33
emphasize strength of both techniques.
My hypothesis is to use fingerprinting approach based on RSSI values from transmitters of both types (FM & Wi-Fi).
3.2
Selected method for fusion
Positioning methods, stated in introduction, are divided into geometry-based and
RSS-based. Geometric positioning technique is widely applied in cellular, UWB,
pseudolite, lasers and ultrasound positioning systems. This technology is easy to
popularize, but the error increases while NLOS exists.
On the other hand, fingerprint positioning technology was firstly designed to be used
with Wi-Fi RSS values and can mitigate NLOS error effectively, but it is limited by
the heavy workload of fingerprint acquisition and the large amount of fingerprint
database.
A number of factors that may cause fluctuations of fingerprints for a system using
local beacons (Wi-Fi or local beacons), such as:
∙ Furniture layout in the room of interest,
∙ Furniture layout in nearby rooms,
∙ Air temperature and humidity,
∙ Temperature of the beacons’ components (Wi-Fi access points may warm up
under a heavy load),
∙ Presence of people.
Systems employing external beacons, such as broadcasting FM stations, have additional sources of uncertainty:
∙ Buildings and other large structures (especially RF-reflective),
∙ Weather conditions (rain, clouds, thunderstorms),
∙ Vegetation, season of the year.
Depending upon the type of the router, transmission power and the antenna (if
present) orientation of the router, the RSS by the same receiver and at the same
34
distance may vary. But, from the observations, whatever WLAN routers we use and
whichever emitter topology, the statistical RSS distribution models remain rather
similar and they depend only on the building structure (e.g., wall and floors materials, number of floors, room and window layout, etc). [37]
That is to say, signal strengths are consistent in time: the signal strength from
a given source at a given location is likely to be similar tomorrow and next week,
while also elimination the timing problem of FM signals. Also, it reduces the effect of
multipath compared to other methods based on distance measurements. To conclude,
this means that there is a radio profile that is feature-rich in space and reasonably
consistent in time.
In April 2013 the Federal Communications Commission (FCC) Working Group 3
(WG-3) released results of intensive indoor location trials of various technology solutions. The tests trialed thousands of attempted location fixes in four representative
morphologies (dense urban, urban, suburban, rural) and various building types.
The technologies used were: Qualcomm’s hybrid AGPS/AFLT solution, NextNav’s
beacon transmitters deployed across an area and Polaris Wireless’ RF fingerprinting. Results have shown, that the yield from Polaris was the best (96.9% in rural
buildings), while QualComm failed in reliability of getting fix position, obtaining
only 85.8% of test calls in all dense urban buildings. Then, from the overall location
errors table can be concluded 1 , that NextNav came out on top while having in
mind that beacons solution is the most expensive, requires beacons infrastructure
and specific receiver configured to decode NextNav readings. [46]
Based on the above, I decided to implement for my studies pattern matching approach (fingerprinting) for RSS values to compare signal strength of Wi-Fi and FM
waves in order to locate myself in a testbed.
The biggest advantage of this approach is low cost while delivering high network
accuracy performance and without need in equipping given building with beacons.
3.3
Techniques to be fused
In the following chapter the techniques, that I’m going to fuse, will be specified.
From all possible techniques used indoor, Wi-Fi-based positioning is the de-facto
standard for indoor localization because of its abundance and well-elaborated characteristics.
1
Overall trials results are in Appendix C
35
It has certain advantages such as possibility to use an existing infrastructure, decent
accuracy and FM radio signals are less affected by weather conditions, such as rain
or fog, in comparison to Wi-Fi or GSM. Low-frequency radio waves are less sensitive
to terrain conditions, such as woodland and tree foliage. Amount of attenuation of
radio waves, caused by building materials is directly proportional to the operating
frequency therefore, FM signals penetrate walls more easily in comparison to Wi-Fi
or GSM. The FM wavelength of around 3 m (from 2.78 m to 3.43 m in Europe and
US) interacts differently with most indoor objects in comparison to the wavelength
of 0.12 m of Wi-Fi waves. At low frequencies, when the obstacles are small compared
to the wavelength, they do not interact significantly with the electromagnetic fields
of the wave. The described considerations suggest that FM based indoor positioning
has a number of theoretical advantages over the current high-frequency systems. [33]
Description of each selected technique starts with its basic SWOT analysis followed
by properties, directly used for positioning.
3.3.1
Wi-Fi
Table 3.1: Wi-Fi SWOT chart
Strengths
Weaknesses
Limited coverage in
Well-explored
rural areas
Good in urban environment
Interference with other devices
Low-cost infrastructure
Relatively high power consumption
Opportunities
Threats
WiMax
Security
Rich R&D
Country-dependent features
Leading technology
Because my approach is to collect fingerprints in order to construct the so called
open radio map of a building or a specific area inside it, my main interest in a Wi-Fi
signal lies in a received signal strength, penetration and attenuation of a signal.
In an IEEE 802.11 system, RSSI is the relative received signal strength in a wireless
environment, actually an indication of the power level being received by the antenna.
It is usually measured in dB and ranges from 0 to -100 and the higher the RSSI
number, the stronger the signal. The 802.11 standard does not define any relationship
between RSSI value and power level in mW or dBm.
Wi-Fi range is based on power of signal, for example for transmitting power 800
36
mW range is 30m.
Wi-Fi attenuation varies for different obstacles, for instance, interior office door
worsen RSSI for 4 dB, 3.5’ brick for 6 dB, interior office window for 3 dB. [34]
I will examine stability of a Wi-Fi signal in detail.
∙ Human body presence impact
According to [15], the impact of human body blocking LOS is very small.
[16] also reviews user’s body influence on RSS distribution and states that by
spreading the range of RSS values the standard deviation increased from 0.68
to 3 dBm where user was present and mean changed from -70.4 dBm to -71.6
dBm. I consider these values as insignificant for my system and therefore the
presense of people were not taken into account later in experiments.
∙ Human body orientation impact
During the offline phase the laptop was rotated by different angles in randomly
chosen reference points and RSS of desired APs were measured. It is seen from
the results2 that the impact of human body orientation is small and can be
neglected.
∙ Time of day fluctuations
Of course, the 2.4 GHz range is well occupied by a lot of appliances from cordless phones to microwave ovens and Bluetooth devices and is highly affected
by human activities, door openings and AP status changes (e.g. from active to
non-active). [15] In order to get rid of such effects, the measurements in both
offline and online phases were taken at a time frame from 4 PM to 6 PM.
∙ Number of APs impact
Number of APs was picked intentionally, as works [32] show that more APs
give better results but only up to some limit, after which system performance
remains at the same level or even degrades, so there is little benefit in going
beyond 3 APs in case of RF.
∙ Number of samples impact
While it may be reasonable to construct the data set with a large number
of samples, there may be constraints on the number of samples that can be
obtained in real-time to determine a user’s location. So investigation in [5]
showed that only a small number of real-time samples is needed to approach
2
The acquired values are in Appendix E
37
the accuracy obtained using all of the samples. With two samples it’s only
about 11% worse than using all samples (4 per second at each AP). Therefore,
the number of samples was limited to 2 per each Reference Point.
3.3.2
FM
Table 3.2: FM SWOT chart
Strengths
Weaknesses
Availability
Low power consumption
Capture effect
Stable against weather and terrain conditions
No timing information
Better penetration through walls
Prone to multipath and NLOS
No interference
Opportunities
Threats
DAB and LPFM
Future obsolescense
For the sake of my work, I examine only FM broadcasting stations, which occupy
frequencies from 88 to 108 MHz and send VHF (Very high frequency) signals.
VHF signals is less affected by atmospheric noise and interference from electrical
equipment, it is less affected by buildings.
Average RX power of a broadcasting station is 40 mW, while range can be up to 40
miles LOS. From the characteristics of FM broadcast three features could be used
for positioning purposes with a fingerprinting method: RSSI, SNR (Signal-to-ratio)
and SCS (Stereo channel separation).
It was shown in [30], that SCS is suitable only for shorter distances between transmitter and receiver and the stereo-signal must be known and SNR demonstrates
worse accuracy than RSSI, thus, only RSSI was used as a definitive feature of FM
broadcast. It is usually measured in dB and ranges from 0 to 100 and the higher the
RSSI number, the stronger the signal.
In the following I will examine conditions, which may or may not affect RSSI of FM
broadcasting stations.
∙ FM beacon selection impact
In order to detect the list of active FM channels the FM Pira receiver ran
all the frequencies twice and those with the level of RSSI above the threshold
of 25 were chosen. It should be noted, that not only stations with highest
38
RSSIs were chosen, but rather stations distributed in space between, because
works show that indoor stronger stations have no advantage over weaker ones
in positional sense, because FM signal RSSI varies mainly due to walls and
other obstacles, which equally affect all beacons transmitting from the same
direction despite their signal strength.
∙ Number of beacons impact
Well-known that as the number of beacons increases, the accuracy of fingerprinting approach improves positioning accuracy, but only to some limit and
further increase of beacons doesn’t affect the accuracy, possibly due to external
interference. With the 7 stations total accuracy of the system is only slightly
inferior than using all 76 beacons (only 0.4m worse than full system using 10%
of beacons).
∙ Human body presence impact
It should be noted, that in the presence of people FM signals generally are
not affected. The same work of showed, that for 80% of stations the shift in
crowded and in empty environment was within 10%. However, the FM signal
fluctuations increase manyfold in a crowded room, probably because that radio
waves of FM band (about 100 MHz) are scattered by human bodies and not
absorbed as Wi-Fi waves. [30]
3.3.3
Conclusion to RSSI properties
Firstly, I should note that modern Wi-Fi Access Points are able to adjust their
power according to user needs, for example, via web-interface, but for our purpose
it was not taken into account. Secondly, while it’s a subject of research, the weather
conditions (rain, sun, blocked line-of-sight outside, snow) were not taken into account
as well.
Thirdly, no additional antenna was used in both offline and online phases for Wi-Fi
receiver and only stock antenna for FM Pira receiver in both phases as well.
Depending on the properties of Wi-Fi and FM signals, I expect that RSS of both
signals will be consistent in time, that fusion of techniques could make sense because
of the similar nature of measurements and my assumptions will not impede the
experiment much.
39
Chapter 4
System proposition
4.1
General approach
My proposition of localization system is based on the existing infrastructure of
broadcasting FM stations and Wi-Fi Access Points as signal sources and embedded
FM and Wi-Fi radio modules on client devices. This kind of system does not require
any additional infrastructure, which can be a significant advantage over other indoor
positioning systems.
The common method of finding active broadcasting stations during seek tuning,
employed by virtually all FM receivers, is RSSI thresholding, where the receiver
registers a broadcasting station at a specific channel if its RSSI level is above the
predefined threshold.
As a fingerprint matching technique k-Nearest Neighbor (KNN) algorithm was employed. To examine the feasibility of fusing FM & Wi-Fi positioning signals to obtain
one’s position within defined area the experiments were performed in our faculty
building.
4.2
Positioning approach
My approach follows the general fingerprinting method and consists of two phases:
1. Offline training (surveying) phase, which collects RSS samples at reference
positions and builds a training database,
2. Online determination phase which calculates the location of a mobile user by
40
comparing the measured RSS values with the training database and subsequently uses kNN algorithm to determine the location of a user.
It should be noted that locations of APs weren’t determined, because in fingerprinting approach it’s not needed.
4.3
Classification approach
In a fingerprinting-based positioning system there is a task to associate acquired
fingerprints with locations using the data collected during offline phase.
The classification approach considers vector with values at each reference location
as a discrete class. Given a fingerprint, a classifier returns the class to which this
fingerprint most likely belongs. This method considers each location independently
and almost immediately returns the closest class. As an output format this approach
produces a class label only from those, that were present in the training data. Thus,
the positioning accuracy of classification approach is limited to granularity of the
calibration data.
4.4
kNN algorithm
The kNN algorithm is a simple yet powerful classification method. It determines the
K most likely locations of a mobile user. Among these locations usually the one with
the lowest difference from stored value is selected.
The algorithm works as follows:
Given a fingerprint to classify, it evaluates the distances in signal space from this
fingerprint to the fingerprints in the training set. Then a specific distance metric has
to be used and this thesis utilizes commonly used Euclidean distance metric,
𝐷=
⎯
⎸ 𝑛
⎸∑︁
⎷
𝑦2
𝑖
𝑖
where yi is a difference between each element of stored vector of measurements and
currently recorded vector.
In this case, however, the K most likely locations instead of 1 location were selected
and then resulting one was evaluated because experiments show that sometimes
actual location may not be the location with the lowest Euclidean distance.
41
The advantages of kNN algorithm are:
∙ fast training phase, which comprises only storing training data,
∙ often the best positioning performance, [20], [36], [41]
∙ superior performance in obstructed areas (indoors), [10]
Thus, the kNN algorithm was applied as the main classification approach in this
thesis.
4.5
Related work
One of the most elaborated works based on FM signals is called FINDR and also
uses short-range FM transmitters as wireless beacons and measures Received Signal
Strength (RSS) by a fingerprinting approach.
Their results had strong correlation to what was done before, meaning the results of
the system evaluation have shown a median accuracy of about 1.0 m and 5.0 m at
95% confidence level, which was close to Wi-Fi characteristics in chosen conditions.
[30] Further evaluation of FINDR [23] by the same group has shown some improvement by using KNN algorithm and Gaussian Process (GP) regression. The median
estimation error (50th percentile) of the system was 0.97 m for GP and 0.93 m for
kNN while 95th percentile error was 2.65 m for GP and 3.88 m for kNN.
Also worth mentioning the work of [11], who used the same principle of RSS and
getting database of fingerprints by combining FM & Wi-Fi, which gives notable
result that combination of WiFi and FM signals into a single signature provides up
to 83% higher localization accuracy compared to WiFi only RSSI fingerprinting. In
addition to this, paper discovered that to achieve the maximum localization accuracy
(i.e., accuracy when all radio stations or access points are used), 30 FM radio stations
and approximately 50 Wi-Fi access points are required.
Altintas et. al. present a short term memory scheme using previous WLAN RSS
observations to smooth error distance during the online determination phase. The
shorter the distance to the prior position, the higher probability of the current
position. [2]
42
Chapter 5
System implementation
5.1
Testbed
Two rooms in the Konvitska faculty building were used as a testbed, located right
above each other.
K305 (Prednaskovy Sal), as shown in Fig. 5.1 and K404 classroom, Fig. 5.2 The
former is approx. 17*9 meters, which was transformed into 2D 3*9 grid, while the
latter is 9*7, which resulted in 5*3 grid.
In both locations dimensions of each cell are approx. 2m*1.5m.
It should be noted, that K305 is almost 2 times bigger than K404, so only rows from
1 to 5 in K305 are directly adjacent to K404.
43
Figure 5.1: K305 lecture hall
Figure 5.2: K404 classroom
5.2
Data collection setup
First phase in the fingerprinting approach was “training phase”, where two mobile devices were moved through the testbed recording the strength of signals. On
one side, Lenovo laptop running under Windows 7 with external D-Link DWA-121
44
Wireless Adapter was collecting RSSIs from 7 Wi-Fi Access Points with the help of
c Wi-Fi inSSIDer application, see Fig. 5.3.
MetaGeek○
Figure 5.3: Experimental setup
First step in the surveying phase is to pick only channels that don’t overlap and
have a significant distinction in frequencies:
1. eduroam 802.11g, freq: 2427 GHz, channel 4
2. eduroam 802.11n, freq: 2472, channel 13
3. eduroam 802.11n, freq: 2457, channel 10
4. Bagr 802.11n, freq: 2412, channel 1
5. Julka 802.11n, freq: 2472, channel 13
6. Elissei.com 802.11n, freq: 2467, channel 12
7. K401 802.11n, freq: 2437, channel 6
inSSIDer application, as shown in Fig. 5.4 allows to calculate average value, when
the signal varies, e.g. when it’s oscillates from 82 to 74, the average was taken as 78
dBm.
45
Figure 5.4: inSSIDer application window
Another option to retrieve RSSI values is using “netsh” command in Windows systems and then calculate RSSI values from quality in % to dB with the Signal-to-Noise
formula, but it wasn’t taken into account.
It should be stated, that values were collected twice within the time interval of 10
minutes and average value was calculated from them.
In total, 294 fingerprints from 7 Wi-Fi APs in 42 cells were acquired.
On the other side, I engaged Pira FM Analyzer and its software “FM scope”. FM
Scope has a very useful feature “BandScan”, which scans all FM range and gives
stations, their strengths and frequencies as a result. Thanks to it, we’re able to
choose 7 broadcasting stations with most diverse signals, while at the same time
export RSSI values to a text file for post-processing:
1. frequency 90.3
2. frequency 91.9
3. frequency 92.6
4. frequency 96.2
5. frequency 98.1
6. frequency 99.7
46
7. frequency 103.6
C6 in K305 is located right under C1 in K404, so these cells’ bandscans (Fig. 5.5
and Fig. 5.6) were picked for visibility reasons. While the pattern is basically the
same, the chosen frequencies have the biggest differences in these adjacent cells.
Figure 5.5: Bandscan of C6 in K305
Figure 5.6: Bandscan of C1 in K404
Finally, the database associating RSSI measurements with the corresponding cell on
a grid was created.
As I mentioned before, actual locations of APs weren’t determined. However, coarse
locations of APs, obtained from the Ekahau HeatMapper software are seen in Fig.
5.7.
47
Figure 5.7: Ekahau HeatMapper
In order not to affect results, measurements were collected always on a same height
of approx. 1 m. and embedded antenna of Pira were always facing the window and
tests were performed between 4 PM and 6 PM.
It should be stated that for the sake of experiment a couple of assumption were
made.
Firstly, that user that performs measurement in an online phase, already knows
which of the broadcasting stations and access points should be measured.
Another assumption is that APs and BSs are consistent in time, meaning that no
one would change the name of AP, its frequency or real position in the building,
intentionally or not.
The missing data handling is not an issue, because at the moment measured values
are put into software manually and user puts -100dBm for Wi-Fi no signal or signal
lost and 0 for no FM signal.
5.3
Data preprocessing
As a preprocessing step, a heatmap was made in Matlab to demonstrate strengths
of each signal transmitter in every cell of K305 on a predefined scale, as shown in
Figures 5.8 and 5.9.
48
Figure 5.8: Heatmap of Wi-Fi APs in K305
Figure 5.9: Heatmap of FM BSs in K305
49
To check for statistical dependency between APs a Pearson product-moment correlation coefficient was calculated for Wi-Fi APs and FM broadcasting station.A
correlation of 1 indicates that one of APs is redundant and brings no additional
information, while 0 indicates that they share no information.
In general, Pearson coefficient showed very weak correlation (from 0.01 to 0.3) meaning that there is a weak linear dependence between each AP and each BS. Sometimes
it was moderate (up to -0.63), but either tendency was opposite (higher values of
one AP lead to lower values of second AP) or values had remarkable difference (e.g.
AP1 ranging from 45 to 69, whilst AP5 takes values from 72 to 100), so no action
has been called1 .
5.4
Data processing software
My main contributor is the MatLab software and a GUI, an experimental positioning
application for reference locations, namely “cells” on a grid, as indicated in a Fig.
5.10. It estimates the user location using the currently obtained fingerprint of RSSIs,
which, up to this moment, have to be put into GUI manually.
For the time being only the k-nearest neighbor (kNN) positioning method has been
utilized.
The object (Sample) is being assigned to the closest class amongst its k nearest
neighbors from the Training set.
1
The correlation matrices between APs and BSs are placed in the Appendix D.
50
Figure 5.10: Screenshot of a GUI
Designed software offers options to choose from: Radius (currently deactivated),
Number of nearest neighbors and selection of techniques to calculate (works on a
SelectionChange basis).
Two uitables resemble K305 and K404 and actual position is shown as a colored
number in a grid.
Due to the fact, that my positioning method is cell-oriented and not distanceoriented, I can evaluate two types of positioning errors.
First is a wrong number of cell, while the second is a RSSI difference between
measured and stored values.
The software uses support function “nearestneighbor” by Richard Brown, because it
offers variety of functions not available in the standard matlab knnsearch function
like distance or option to choose multiple neighbors.
51
I’d say that my software is rather positioning than actual navigating, but I did
incorporate here a history of movement. It works in such a way:
When result is achieved with sufficient distance level, one can click on a “Store
position button” and it will be saved into array “Position history”. This procedure
has to be done for each new position.
52
5.5
Software algorithm
Figure 5.11: Location estimation algorithm
53
The algorithm, illustrated in a Fig. 5.11, works as follows:
The data is stored in a Training set, then vector or values, that was collected during
the online phase was compared to stored data to find desired number of nearest
neighbors by using the closest Euclidean distance as metrics.
Result gives N most probable cells with the minimal distances between stored value
and readings. Then, this cell number is highlighted in the uitable and distance is
a difference between readings from actual position and readings from estimated
position.
The possibility of including certain confidence level was investigated, but transition
from distance to probabilities with the current metrics doesn’t satisfy requirements
for a probability.
For instance, it’s common, when first nearest neighbor differs from second only by
distance equaled to 1, which leads to two close cells both with high probabilities.
Another case is when the algorithm assigns biggest confidence to closest neighbor
regardless of large distance between estimated and actual cell.
Therefore, current Euclidean metrics was kept.
Regarding the fusion of two techniques, first and easiest way was to combine fingerprints from APs and BSs into one large fingerprint with 14 values, then apply kNN
to it.
Algorithm combined in that way was expected to have worse result than each technique alone, because it simply concatenates two sample vectors, thus increasing total
distance and results showed it.
Thus, an algorithm utilizing arithmetic average of N nearest neighbors was implemented. After providing N nearest neighbors an arithmetic average of N cell numbers
is calculated to improve cell accuracy.
5.6
Testing campaign
In testing campaign three Wi-Fi adapters have been used:
1. External D-Link DWA-121, which was used for both the training and the
online phase,
2. Internal QualComm Atheros,
3. Broadcom BCM432, embedded into HTC Desire 500 smartphone.
54
Unfortunately, it wasn’t possible to find an FM signal analyzer apart from the Pira,
which would be able to give the signal strength for defined frequencies, so in both
training and measuring phases Pira FM Analyzer was used.
During all the testing campaign, both techniques were used to estimate location and
a route and best result (technique, which provided minimal distance) was chosen.
First route (Table. 5.1) in K305 was performed with adapter 1:
B4 –> A4 –> A5 –> A6 –> A7 –> B7 –> C7 –> C6;
Table 5.1: Real route 1
A K305 B K305 C K305
Row1
Row2
Row3
Row4
Row5
Row6
Row7
Row8
Row9
2
3
4
5
1
6
8
7
Table 5.2: Estimated route 1
A K305 B K305 C K305
Row1
Row2
Row3
Row4
1
Row5
2
Row6 5+6
Row7
3
Row8 8
4
Row9 7
55
Figure 5.12: Error distance in K305, points
According to Table 5.2, from 8 checkpoints of a route1 only one was estimated
correctly. However, in 50% of measurements an adjacent cell was estimated, thus
giving cell-circle accuracy.
Taking into account cell dimensions of 1.5m*2m, it gives an error from 2.25 to 4m.
Also, in two cases correct cell was the 2nd neighbor with distances only 1 and 3 from
the estimated cell’s distance.
Most of the time it was Wi-Fi that provided best results2 , while fused performance
was somewhat poor because of the chosen mechanism of fusion, which accumulates
errors rather then fixes them, as shown in a Fig. 5.12
Then, the experiments with number of nearest neighbors were performed under
selected cells.
In my software k affects only fusion of techniques and the effect of increased number
k of nearest neighbors was investigated with the help of adapter 1.
The main objective was to determine if small integer number k=1 to 4 is sufficient
and that error distance fails to improve afterwards.
2
Complete results are in Appendix G
56
Figure 5.13: Results of number of NN tests
It’s visible in a Fig. 5.13, that increasing the number of NN positively affects a
distance error, for example, with k=4 and 5 it was possible to obtain zero error,
which is a great result.
Besides, overall distance error seems to decrease with more neighbors and in the
unknown environment a k=1 would be the most sensible choice, even if it not uses
all the advantages of the fusion. Unfortunately, the obtained information wasn’t
used for the first route; however, it is believed that with the increased number of
neighbors result would be better.
Then experiments were performed in the upper room with the smartphone receiver
and k was set to 1.
Second route in K404, as indicated in a Table 5.3, was performed with adapter 3
and estimated results are shown in a Table 5.4.
A2 -> A3 -> A4 -> B4 -> B5 -> C5
57
Table 5.3: Real route 2
A K404 B K404 C K404
Row1
Row2
Row3
Row4
Row5
1
2
3
4
5
6
Table 5.4: Estimated route 2
A K404 B K404 C K404
Row1 2+3
Row2 1
Row3
Row4
Row5
6
Figure 5.14: Error distance in K404
Wi-Fi showed best results while the combination became a bit farther from other
techniques in comparison with first route, which is seen in a Fig.5.14.
The software was able to correctly determine position in two cases, both close to
the inside wall.
Checkpoint 5 was wrongly estimated on a different floor and overall performance
was unsatisfactory.
58
Figure 5.15: Empirical cumulative distribution function
Note: Red line marks – FM, Green – Wi-Fi, Yellow – Combination.
Cumulative distribution function of distances (Fig. 5.15) confirms my hypothesis
that Wi-Fi provides the best accuracy, while fusion demonstrates worst results because of concatenation.
59
Conclusion and future work
I presented the indoor positioning system based on a fusion of two positioning signals. Simulation software proved, that it’s able to roughly estimate position of the
user, achieving floor-level and cell-square accuracy.
During work on the thesis, the comprehensive elaboration of all existing possibilities
in indoor navigation were performed and the Matlab application was constructed,
which proved the feasibility of indoor positioning by means of Wi-Fi and FM signals. Experiments have shown that room-wise accuracy can be achieved in any given
building without additional infrastructure only with the need of radio map construction.
Most of the time software was able to differentiate between floors mainly because of
Wi-Fi eduroam APs, which signal strength degrades significantly on the 4th floor,
the reason of it could be that Konviktska building is built from the material, that
effectively blocks Wi-Fi signal propagation.
No positional delay is a great advantage of a system, because as soon as fingerprints
are obtained, the calculations are done practically immediately.
However, when it comes to cell-determination, software shows only modest results,
often failing to follow the movement and correctly estimate position on a grid.
Apparently, the fusion of selected techniques using their direct combination makes
no sense, often only worsening results. Possible reasons of it are: capture effect, good
penetration of FM signal and random fluctuation of Wi-Fi even when the user is
static.
Related work of [32] showed better results because of huge number of FM broadcasting beacons, which equaled 76 and 17 Wi-Fi APs. Also it should be said that
positioning accuracy is highly dependent on a device, which was used for both the
training and the measure. It was evident, that QualComm and Broadcom adapters
deliver weaker RSSIs than D-Link adapter at the same time and the same conditions.
Indeed, a number of future research directions remains to be investigated.
60
If RSSI values are to be obtained automatically by some low-level application, then
some method of filtration, for instance, Kalman filter could be implemented. It
operates recursively on streams of noisy input data (location estimates in my case)
to produce a statistically optimal estimate of the underlying system state. Even
without knowing the nature of measurements it usually significantly improves results
by filtering erroneous results.
A modification of kNN algorithm, for example a weighted kNN algorithm could be
incorporated in order to assign weights to closest neighbors thus achieving better
results.
While continuing with kNN algorithm, a different metric can be included, for example Chebyshev, Manhattan or Hamming distance metrics.
Different learning algorithm can also be deployed, starting from algorithms utilizing
Bayesian rule and Support Vector Machine and then neural networks, or some rulebased systems.
Better mechanism of fusion of two techniques, for instance the multiplication of
two Gaussian distributions obtained by each technique alone should provide better
results.
Finally a completely different combination of techniques could be used, e.g. Wi-Fi
+ MEMS.
61
References
[1] Al Nuaimi, K., Kamel H. A Survey of Indoor Positioning Systems and Algorithms. 2011, International Conference on Innovations in Information Technology
[2] Altintas, B., Serif, T. Indoor location detection with a RSS-based short term
memory technique (KNN-STM) 2012, Proceedings of IEEE International Conference on Pervasive Computing and Communications Workshop
[3] Anand, M., Nightingale, E. B., Flinn, J. Self-tuning wireless network power
management. 2005, Wireless Networks, vol. 11
[4] Bahl, P., Padmanabhan V.N. RADAR: An In-Building RF-based User Location
and Tracking System. 2000, In INFOCOM.
[5] Bahl, P., Padmanabhan V.N. Enhancements to the RADAR User Location and
Tracking System. 2001, Microsoft research
[6] Battiti R., Brunato, M. Statistical learning theory for location fingerprinting in
wireless LANs. 2005, Computer Networks.
[7] Benevente-Peces, C., Puente, M., Dominguez-Garcia, A. Global System for Localization and Guidance of dependant people. 2009, Ambient Assistive Health
and Wellness Management in the Heart of the City. Proc. Seventh Int. Conf.
on Smart Homes and Health Telematics 2009
[8] Bodhi Priyantha, N. The Cricket Indoor Location System, 2005, Massachusets
Institute of Technology
[9] Chai et al. Continuous Indoor Localization and Navigation based on low-cost
INS/WIFI integration. 2011, International Conference of Positioning and Indoor Navigation 2011
[10] Chen M. et al. Practical metropolitan-scale positioning for GSM phones. 2006,
Proceedings of the 8th International Conference on Ubiquitous Computing
62
[11] Chen, Y. et al. FM-based Indoor Localization. 2009, MobiSys’12
[12] Chen, Y.C. et al. Sensor-assisted Wi-Fi indoor location system for adapting
to environmental dynamics. 2005, Proceedings of the 8th ACM international
symposium on MSWiM.
[13] Deng Z. et al. Situation and development tendency of indoor positioning. 2013,
China communications.
[14] Gu, Y., Lo, A., Niemegeers, I. A Survey of Indoor Positioning Systems for
Wireless personal networks. 2009, IEEE Communications surveys & tutorials,
vol. 11
[15] Jiang, L. A WLAN fingerprinting based indoor localization technique. 2012,
Thesis in University of Nebraska.
[16] Kaemarungsi, K., Krishnamurthy, P. Properties of Indoor Received Signal
Strength for WLAN Location Fingerprinting. 2004, Mobile and Ubiquitous Systems: Networking and Services
[17] Kawakubo, S. et al. Wireless network system for indoor human positioning.
2006, Wireless Pervasive Computing
[18] Kohler M. et al. TrackSense: Infrastructure Free Precise Indoor Positioning
Using Projected Patterns. 2007, Pervasive Computing
[19] Kotanen, A., Hannikainen, M., Leppakoski, H. Experiments on Local Positioning with Bluetooth. 2003, Proceedings of the International conference on Information technology
[20] Lin, T-N., Lin, P-C. Performance comparison of indoor positioning techniques
based on location fingerprinting in wireless networks. 2005, In Proceedings of
International Conference on Wireless Networks, Communications and Mobile
Computing
[21] Liu, H. et al. Survey on Wireless Indoor positioning techniques and systems.
2007, IEE Transaction on systems, man and cybernetics.
[22] Locata
Corporation
Locata
technology
brief
Retrieved
from
http://www.locata.com/wp-content/uploads/2014/07/Locata-TechnologyBrief-v8-July-2014-Final1.pdf
[23] Matic, A. et al. Tuning to Your Position: FM radio based Indoor Localization
with Spontaneous Recalibration. 2009, Pervasive Computing and Communications (PerCom)
63
[24] Mautz, R., Tilch, S. Optical Indoor Positioning Systems. 2011, International
Conference of Positioning and Indoor Navigation
[25] Moghtadaiee, V., Dempster, A. G., Lim, S. Indoor Localization Using FM Radio
Signals: A Fingerprinting Approach. 2011, International Conference of Positioning and Indoor Navigation
[26] Ascension tech (2008) MotionStar real-time motion capture Retrieved from
http://www.irit.fr/PREVI/docs/MotionStarWireless.pdf
[27] Ni, L.M. et al. LANDMARC: indoor location sensing using active RFID. 2003,
Pervasive Computing and Communications, 2003.
[28] Otsason, A. V. Accurate GSM indoor localization. 2005, Proceedings of the 7th
International Conference on Ubiquitous Computing
[29] Pittet, S. et al. UWB and MEMS Based Indoor Navigation. 2008, The journal
of navigation
[30] Papliatseyeu, A., Osmani, V. Indoor Positioning Using FM Radio. 2011, Multimedia Interaction and Smart Environments
[31] Pappas, E., States, R. A. Precision and repeatability of the Optotrak 3020 motion measurement system. 2006, Medical Engineering and Technology
[32] Popleteev A. Indoor positioning using FM radio signals. 2011, PhD dissertation
in University of Trento.
[33] Popleteev, A., Osmani, V. Investigation of indoor localization with ambient FM
radio signals. 2012, Pervasive Computing and Communications (PerCom)
[34] Rackley, S. Wireless Networking Technology. 2007, Newnes Publishing.
[35] Rehman, W., de Lara, E., Saroiu, S. CILoS: A CDMA Indoor Localization
System. 2008, Proceedings of the 10th International Conference on Ubiquitous
Computing UbiComp 2008, Seoul.
[36] Schwaighofer, A. et al. GPPS: A Gaussian Process Positioning System for Cellular Networks. 2004, Advances in Neural Information Processing Systems 16
[37] Shrestha, S., Talvitie, J., Lohan, E.S. On the fingerprints dynamics in WLAN
indoor localization. 2013, International Conference on ITS telecommunications.
[38] Tam, W.K., Tran, V.N. Propagation modelling for indoor wireless communication. 1995, Electronics & Communication engineering journal
64
[39] Want, R. et al. The Active Badge Location System. 1992, AT&T Laboratories
Cambridge.
[40] Ward, A., Jones A., Hopper A. A New Location Technique for the Active Office
1997 IEEE Personal Communications, Vol. 4.
[41] Wassi, G.I. et al. Indoor location using received signal strength of IEEE 802.11b
access point. 2005, In Proccedings of the Canadian Conference on Electrical
and Computer Engineering.
[42] Weissman, Z. Indoor location 2004, White paper, Tadlys LTD.
[43] Broadcom
(2006)
Broadcom
BCM
4326
Data
Sheet
Retrieved
from
http://pdf1.alldatasheet.com/datasheetpdf/view/175096/BOARDCOM/BCM4326.html
[44] Philips (1996, May 14) FM receiver circuit for battery data sheet Retrieved from
http://pdf.datasheetcatalog.com/datasheet/philips/TDA7088.pdf
[45] Technology tracking department (2003) Wi-Fi Technology Retrieved from
http://www.tra.gov.eg/uploads/technical material/Wi-Fi report.pdf
[46] Working Group3, Indoor Location Test Bed Report 2013, The Communications
Security, Reliability and Interoperability Council III.
65
List of terms specific to the thesis
AoA - Angle of arrival
AP - Wi-Fi Access Point
BS - FM broadcasting station
CDMA - Code Division Multiple Access, a technology used in mobile phone communications
DAB - Digital Audio Broadcasting
Fingerprint - a set of RSSI values from specific location
FM - broadcasting radio waves of corresponding frequency
GSM - Global System for Mobile communication, an ETSI cellular networks technology
IPS - Indoor positioning system
KNN - k-nearest neighbour machine learning algorithm
LPFM - Low power FM broadcast stations
MEMS - microelectromechanical systems
NLOS - Non-Line-of-Sight
RFID - Radio-frequency identification
RSSI - Received Signal Strength Indication
RTLS - Real-time locating system
TDoA - Time difference of arrival
ToA - Time of arrival
UWB - Ultra-wideband radio technology
Wi-Fi (WLAN) - an IEEE 802.11 technology
66
Appendices
67
.1
Indoor positioning technologies comparison
Following Table 5 is a short comparison of existing indoor technologies, which reviews the most important parameters for an IPS.
Table 5: Indoor positioning technologies comparison
Technology
Accuracy
Coverage Power consumption
medium
Wi-Fi
low
high
(10-20m)
low
Cellular
high
high
(50-300 m)
Bluetooth
medium
low
high
RFID
high
low
low/high
UWB
high
low
low
Ultrasound
medium
medium/low
low
Optical
medium
medium/low
medium
Infra-red
medium/high
low
low
FM
low
high
low
.2
Infrastructure cost
low/medium
low
high
low/high
high
high
medium
medium/high
low
FCC WG-3 trials results
Table 6 shows the results of trials, performed by a Working Group 3 of Federal
Communication Commission on indoor positioning in different areas, utilizing Qualcomm’s hybrid AGPS/AFLT solution, NextNav’s beacon transmitters deployed across
an area and Polaris Wireless RF fingerprinting.
68
Table 6: FCC WG-3 trials results
Horizontal error (m)
Total N
Building ID
67%
95%
of calls
NextNav_All dense urban build.
4859
57.1
154.0
NextNav_All urban buildings
4238
62.8
196.1
NextNav_All suburban buildings
3581
28.6
62.2
NextNav_All rural buildings
820
28.4
60.3
Polaris_All dense urban build.
5372
116.7 569.3
Polaris_All urban buildings
3874
198.4 729.9
Polaris_All suburban buildings
3489
232.1 571.4
Polaris_All rural buildings
726
575.7 3072.3
Qualcomm_All dense urban build.
5145
155.8 328.1
Qualcomm_All urban buildings
4338
226.8 507.1
Qualcomm_All suburban build.
3716
75.1
295.7
Qualcomm_All rural buildings
709
48.5
312.3
.3
Avg.
error
57.5
69.5
27.2
70.3
150.3
203
215.1
845.6
136.4
233.9
92
639.9
Stand.
dev.
64.9
99.9
99.7
1231.5
193.3
225.9
161.9
961.3
94.7
547.7
173.6
2999.2
Max
error
1059.2
4367.2
5854.2
35255.9
1656.1
3131.9
1089.1
5809.2
722.5
18236.7
4639.4
27782.4
Correlation matrices
As shown in the following tables 7 and 8, the statistical independency within Wi-Fi
Access Points and within FM Broadcasting stations was proved with the help of
Pearson coefficient.
AP1
AP2
AP3
AP4
AP5
AP6
AP7
AP1
1
-0.3436
-0.6362
-0.3951
0.5956
0.5127
0.3298
Table 7:
AP2
-0.3436
1
0.3682
0.053
-0.1651
-0.2916
0.1813
Wi-Fi APs correlation matrix
AP3
AP4
AP5
AP6
-0.6362 -0.3951 0.5956 0.5127
0.3682
0.053 -0.1651 -0.2916
1
0.4791 0.3159 -0.3502
0.4791
1
-0.2324 -0.1551
0.3159 -0.2324
1
0.4128
-0.3502 -0.1551 0.4128
1
-0.1688 -0.2031 -0.0234 0.3165
69
AP7
0.3298
0.1813
-0.1688
-0.2031
-0.0234
0.3165
1
Min
error
0.6
2.1
0.4
1.5
2.2
0.4
8.4
66.2
0.5
1.6
0.2
1.0
BS1
BS2
BS3
BS4
BS5
BS6
BS7
.4
BS1
1
0.3531
0.2588
0.2594
0.2578
0.2416
0.0832
Table 8: FM BSs
BS2
BS3
0.3531 0.2588
1
-0.1858
-0.1858 1
0.0227 0.0943
0.4974 0.0757
0.3651 -0.0975
-0.017
-0.0169
correlation matrix
BS4
BS5
BS6
0.2594 0.2578 0.2416
0.0227 0.4974 0.3651
0.0943 0.0757 -0.0975
1
0.3947 0.2025
0.3947 1
0.507
0.2025 0.507
1
0.1459 0.0991 0.2606
BS7
0.0832
-0.017
-0.0169
0.1459
0.0991
0.2606
1
RSS fluctuations due to user’s orientation
The additional experiment was performed to examine the change in RSSI values
when facing different directions, see Table 9
Table 9: RSS fluctuations
Rotational angle
AP1 RSS in RL1
AP2 RSS in RL5
AP3 RSS in RL9
AP4 RSS in RL11
AP5 RSS in RL16
AP6 RSS in RL21
AP7 RSS in RL26
.5
due to user’s orientation
90∘ 180∘ 270∘
44
45
44
53
56
53
53
52
52
38
39
38
47
48
47
22
22
26
32
33
33
Real readings
In the Tables 10 and 11 readings of performed routes are listed.
70
Table 10: Route 1 in K305
AP
B4
45
71
68
73
73
73
74
BS
B4
32
38
46
33
34
37
29
AP
A4
45
66.5
63
72.5
73
78
82
BS
A4
29
27
54
51
35
30
49
AP
A5
47
65.5
60
74
72
77.5
90
BS
A5
51
29
51
30
36
29
36
AP
A6
51
71
58
72
80
73
82
BS
A6
51
30
59
30
51
*
31
AP
A7
47
73
57
69
74
73.5
82
BS
A7
54
39
50
49
38
30
30
AP
B7
53.5
74
59
74
74
77.5
80
BS
B7
36
34
38
51
30
36
30
AP
C7
49
73
65.5
72
82
77
91
BS
C7
37
27
45
30
36
26
30
AP
C6
48
73
63
74
82
71
74
Table 11: Route2 in K404
AP
B1
34
68
75
83
85
79.5
87
BS
B1
30
31
56
27
33
30
29
AP
B2
57
74
78
83
81
83
88
BS
B2
59
51
59
35
44
38
30
AP
B3
47
74
75
88.5
85.5
82.5
87.5
BS
B3
50
46
44
29
39
37
39
AP
C3
55
86
72
90
87
84
86.5
BS
C3
42
30
38
35
29
*
35
AP
C4
56
75
80.5
85
84.5
87
91.5
BS
C4
47
42
44
30
38
34
26
AP
C5
49
73.5
76
80.5
83
86
87.5
BS
C5
52
38
39
38
39
39
39
AP
B5
57.5
78
77
88
82
85
90
BS
B5
56
40
39
30
40
34
39
* . . . no signal
.6
Experiments results
Table 12 demonstrates final results of an estimation, while 13 is a result of experiments on a different number of nearest neighbors.
71
BS
C6
30
34
37
33
29
38
30
Table 12: Results of route1 estimation
Actual cell
Predicted cell
Proximity
B4
A4
A5
A6
A7
B7
C7
C6
B4
B5
C7
B8
A6
A6
A9
A8
adjacent cell
adjacent cell
2nd neighbour
adjacent cell
adjacent cell
2nd neighbour
-
Number
of step
1
2
3
4
5
6
7
8
FM
error
34
43
19
28
30
40
24
53
WiFi
error
0
7
22
12
22
20
17
10
Fusion
error
33
39
29
38
52
48
41
48
Table 13: Results of experiments on a number of nearest neighbors
Number of NN
1
2
3
4
5
6
7
8
9
10
Error distance B1
46
28
56
52
53
51
50
33
50
39
B2 K305
37
38
42
57
34
53
33
33
53
56
72
B3 K305
36
49
49
45
57
21
21
36
21
57
C3 K305
33
50
44
0
34
27
34
34
35
27
C4 K305
47
46
34
46
0
29
46
47
47
39
Was this manual useful for you? yes no
Thank you for your participation!

* Your assessment is very important for improving the work of artificial intelligence, which forms the content of this project

Download PDF

advertising