2005:204 CIV
M A S T E R ' S T H E S I S
An Autonomous System for Indoor Positioning
in Wireless Networks
Mikael Falk
Luleå University of Technology MSc Programmes in Engineering
Department of Computer Science and Electrical Engineering Division of Media Technology
An Autonomous System For Indoor Positioning in Wireless Networks
Mikael Falk
Supervisor; Johan Kristiansson
Luleå University of Technology
Department of Computer Science and Electrical Engineering 2005-06-06
Preface
This project was conducted as a master’s thesis in computer science at Luleå University of Technology in the spring of 2005.
Abstract
As the presence of wireless mobile devices increases a whole new aspect of services that can be provided to users emerge. Real world contexts such as position play only a marginal role in traditional computing since the user is all but bound to a single fixed location but are very prevalent in mobile computing where the user’s position might change from one moment to the next. The number of services that can be offered to the users of mobile devices based on position is endless, from locating the origin of a distress call to finding friends and relatives.
In this thesis, a system for positioning users in a wireless network using only the existing network infrastructure is presented. The wireless network infrastructure is made up by access points, or base stations, which function as bridges between wireless devices and the regular wired network. By analyzing the degradation of the signal broadcasted from these access points the approximate distance between the access point and the mobile device can be created. The set of this information from all detectable access points at a specific position is called a signal space location and is a unique fingerprint for every real world location. By inserting these signal space locations into an undirected graph the system becomes aware of the surroundings.
Autonomy is a central concept to this method. For a positioning system to be usable in a global context instead of just a local one it needs to be able to independently analyze and adapt to new surroundings as well as detect any changes to the signal space of a building. It also needs to be able to adapt to errors, such as an access point malfunctioning. Basically, autonomy represents the ability to independently handle any situations that might change the system’s view of the surroundings.
Empirical tests of the method yielded discouraging results. Due to the complex nature of indoor radio wave propagation the signal fluctuated, making it difficult for the system to distinguish between signal space locations. The reason for this is the simplified signal
propagation model used in the system, and with a better model many of these problems would be overcome.
Table Of Contents
1. Introduction ... 1
1.1. Research Question... 1
1.2. Background ... 2
1.3. Method ... 2
1.4. Purpose... 3
1.5. Delimitations ... 4
1.6. Wireless Networks ... 4
1.6.1. 802.11 ... 5
1.6.2. Radio Waves ... 6
1.7. Related Work ... 6
1.7.1. Satellite Positioning... 7
1.7.2. RADAR ... 7
1.7.3. Active Badge ... 8
1.7.4. Active Bat... 8
1.7.5. Cricket ... 9
1.7.6. Neural Network Systems... 9
2. Theory ... 10
2.1. Positioning ... 10
2.1.1. Signal Space ... 11
2.1.2. Received Signal Strength Indication ... 12
2.1.3. Discrete Signal Strength... 13
2.2. The Data Structure ... 16
2.2.1. Counting Edges ... 17
2.2.2. Node Interconnectivity ... 17
2.2.3. Concatenation... 19
2.2.4. Naming ... 20
2.3. Autonomy And Fault Detection ... 20
2.4. Application... 22
3. Evaluation... 24
3.1. Implementation ... 24
3.1.1. Manager... 24
3.1.2. Tree... 26
3.1.3. Node ... 27
3.1.4. Positioner... 28
3.1.5. GUI... 28
3.1.6. NetworkClient ... 28
3.2. Tools... 28
3.3. Tests ... 29
3.3.1. Discrete Signal Strength... 30
3.3.2. Concatenation... 34
3.3.3. Positioning... 37
4. Summary And Conclusions... 38
4.1. Method ... 38
4.1.1. Signal Space ... 38
4.1.2. Data Structure... 39
4.1.3. Autonomy... 40
4.2. Tests ... 41
4.3. Problems... 42
4.4. Conclusions ... 42
4.5. Future Work ... 43
5. References ... 44
1. Introduction
In an age where handheld devices and even mobile phones are gradually taking on the properties of regular stationary computers the aspect of mobility comes into focus. As the position of a mobile device can vary from time to time, a new generation of applications that can offer personalized services depending on the current position of the user is becoming increasingly attractive. The amount of services that would benefit from being able to with adequate precision determine the user’s position at any given time is endless, such as for example automatically acquiring the origin of an emergency distress call, locating friends or even offering dating services. Another obviously profitable use would be for direct, targeted marketing, allowing retailers to automatically broadcast special offers or advertisements to nearby users matching certain criteria, thus helping the right consumer find the right product at the right price.
The idea of user positioning is certainly nothing new, and satellite positioning has been available for decades now, remaining the dominant paradigm in accurately determining a device’s position in a global context. Since it was made public, the Global Positioning System (GPS) has been the prevalent implementation, but it is not the only one, the American system having a Russian counterpart in GLONASS and soon being complemented by Galileo, which is a European project currently under work. The only real limitation on satellite positioning is that the device requires line-of-sight to at least three individual satellites. This is because solid objects like the walls of buildings or even dense foliage block the signals from the satellites and consequently the system is unable to position a user for example indoors or in dense rural areas where visibility is limited. It is generally accepted that if satellite positioning was possible even for these locations, all other positioning systems would be rendered obsolete.
In light of the inability for satellite positioning in urban areas or indoor locations there has been research into alternative methods for positioning mobile devices using only the underlying network infrastructure. Some of these methodologies require an additional architecture which results in an increased cost of deployment and lack of scalability while others which require no additional hardware, but in most cases demand extensive input from the system administrator before they can be employed. Additionally, all the solutions to date are sensitive to changes in the underlying infrastructure, such as one of the transmitting base stations being removed or malfunctioning for some reason or other.
1.1 Research Question
Is it possible, given the properties and limitations of the underlying wireless network infrastructure, to create an autonomous system for indoor localization that can be employed anywhere without any additional hardware being deployed and eliminating the need for extensive measurements or awareness of the layout of the building? If positioning of mobile devices on a global rather than localized scale is to be possible such a system seems to be a necessity. Changes in the network infrastructure such as access points being added or
removed and extensive changes in the structural layout of the buildings where such a system is deployed is quite obviously beyond these current systems.
1.2 Background
The underlying network infrastructure is, in this case, the wireless network which is offered through continuously broadcasting access points scattered through an area which offers wireless network interconnectivity. These access points offer some information about their own properties, including a received signal strength indication which can be used to determine the degradation of the signal before it reaches the mobile device. Knowing this degradation of the signal then enables the system to determine the distance between the transmitter and receiver which can be used to calculate the approximate position of the user within the environment. Thus, by compiling these received signal strength indications from several different access points, the system has a good idea of the user’s current location.
An autonomous system is a system that is able to function without any outside intervention from either the user or a system administrator. This means that a positioning system claiming this title should not only be able to configure itself to position users during runtime as it encounters new, previously undiscovered areas, but it should also be able to seamlessly detect faults and update its view of the surroundings when appropriate. Since the access points bridging the gap between wireless mobile devices and the stationary wired network offer very little information about themselves, this requires a data structure that is flexible and by its very nature is able to supply some information about its physical surroundings.
The method proposed in this thesis is different from previous solutions in that it does not require any additional hardware on top of the underlying wireless network infrastructure and is able to function autonomously without input from the user or system administrator. This is central to the purpose of the method, which is to create an independent system that can be employed anywhere there is a wireless network infrastructure and function without outside interference. Localization information is reasonably accurate but presented in a form that is not necessarily immediately useful for the user but instead intended to be used by future applications built on top of this positioning system. In contrast to other systems which tend to present the user’s location on a map this system only presents a user specified name of the current location, limiting the direct usability for the user.
Possible uses of systems employing this method would be video conferencing applications that allow users located in the same geographical area to participate in the same conference, employees at a company being able to notify computer support staff of their current location when encountering problems or friend finder applications notifying the user when an
acquaintance arrives in the vicinity. As can be discerned from these examples the main strength of the system is to associate services with locations in the real world, not offering navigational information or providing the user’s position on a map. When arriving at an airport or shopping mall, for example, a remote server could provide information about available services to interested clients. The use of positional information in these regards is endless.
1.3 Method
The method propose in this thesis offers its own special advantages, which also present their own problems. Basically, the information available to the system about its surroundings are the access points connecting a mobile device to the regular network. The signal broadcasted from these access points are radio waves which have a decidedly complex indoor propagation
scheme which makes calculating the approximate distance between the receiver and
transmitter problematic. Multipath fading, diffraction, free space signal loss and attenuation due to walls and objects in the signals path, all discussed in later chapters, calls for a model which allows the system to efficiently calculate the real distance between the access point and the mobile device. Because an accurate model necessitates precise knowledge about both the placement of the transmitting base station and the layout of the structure in which it is placed, a compromise is essential.
Ideally, the system would be able to calculate the current physical coordinates from the information received from the nearby transmitting access points, but since this information is unavailable without accurate knowledge about the physical environment the received signal can be used to create a virtual coordinate in signal space. This coordinate, a signal space location, then matches against a number of coordinates in the physical universe about which the system assumes no knowledge. These signal space locations can then be described as sets of access points and their received signal strength indications. By storing these in an
appropriate data structure, the system has some degree of knowledge about the surroundings.
The inside of any building can basically be described as an undirected graph where the nodes are rooms and the edges between them doors through which they are connected. This
similarity to the real world environments in which the system is supposed to be used makes an undirected graph a prime candidate for a data structure to describe the physical universe to the system. Consequently, if each node contains the set of access points with received signal strength indications that can be found in exactly one room and an edge is created between each of these rooms, the system potentially has all the information it needs to make qualified, autonomous decisions and react to changes in the surroundings.
The environment being described as an undirected graph offers the system a lot of the
knowledge it requires about its surroundings to function autonomously. This setup allows for a continuous awareness of where in the world the user is currently located and to which neighboring locations it should be possible to move. In essence, this is all the information that is needed to detect new access points inserted into the system, faulty access points that
suddenly stop transmitting and major reconstructions which change the way in which locations are perceived by the system. By reorganizing the undirected graph, adding and deleting signal space locations from different nodes as well as creating and deleting edges between nodes the system is able to flexibly adjust to any and all events that might occur.
1.4 Purpose
While most – if not all – user localization systems attempts to exactly determine the user’s position in physical space, this thesis takes a different approach, focusing on the user’s relative position in signal space and offering only the most superficial correlation between signal space and a physical location. This makes the system presented in this thesis unable to for example pinpoint a user’s position on a map but instead presents it in a context that enables other applications to use that position to offer the user various services. Also, other systems limit use of the application to certain areas – Active Bat requires hardware to be installed where the system is to be used and RADAR requires a radio map to be built of a certain area before the user can be positioned – but this system takes an approach that can be employed anywhere there’s an access point and a connection to a wireless network.
The requirements for this system are,
• no additional hardware
• autonomous
• no external setup
• usable anywhere
• fault tolerant
The system is dependant on the underlying wireless network infrastructure, requiring no additional hardware to position the user. It is autonomous and requires no external setup, automatically adjusting to the surroundings and fault tolerant, able to detect when a new access point has been added or one that was previously present has been moved or stopped transmitting. It is usable anywhere, meaning that it is not limited to any pre-measured locations or otherwise augmented areas.
1.5 Delimitations
The system implemented and evaluated in this thesis examines an approach that offers little in the way of results that are immediately beneficial to the user. The precision of the results is not nearly sufficient to pinpoint the user’s position on a map or in any other to the user intuitively recognizable form. Instead the results – the system producing a position in an abstract signal space – offer application developers an easy method of using these results to offer the user services based on the mobile device’s position.
As such the system makes a few distinct limitations. It offers and presents only an abstract correlation to physical space to the user which might at a glance appear to make it appear to be of less practical use than systems like RADAR [1][2][6], but does so at the gained ability to be applicable at any given location instead of previously mapped areas.
The second limitation imposed on the system is that it does not concern itself with any services using the position of the users. Presented in this thesis is a user localization system, not a platform of services intended to use that information for any purpose.
1.6 Wireless Networks
Wireless coverage is offered by a series of wireless transmitters called access points or base stations that are distributed across a location. Each of these wireless access points has a certain omni directional signal range which creates a circular coverage area surrounding them, thus representing the area in which the access point is active. The task of distributing the access points needs to be given some attention to maximize their coverage, care not only given to the size of the location but the layout of the structure so that coverage can be maximized, offering the best possible quality of service to the users. As a user carrying a mobile device equipped with a wireless network card moves through the location it detects transmitting access points, connecting to the one with the strongest signal at the current location.
Wireless applications quite naturally require a wireless network infrastructure to build on. The currently dominating paradigm is the IEEE 802.11 standards, which offer a good and solid foundation upon which to offer wireless networking. Unfortunately this standard offers no support for positioning such as angle of arrival or time difference of arrival which is used in other methodologies for locating devices.
1.6.1 802.11
The currently favored wireless networking infrastructure is the 802.11 standards which were developed by the Institute of Electrical and Electronics Engineers (IEEE) and promotes an industry standard. This holds obvious advantages for the ends user as it allows for a seamless combination of products from a plethora of developers instead of getting locked into a single- vendor solution.
802.11 is actually a veritable alphabet soup of standards, where 802.11b is the one most commonly used allowing for network transmissions up to 11Mbps, thus making it on par with 10BaseT wired Ethernet connections. It has a signal range that varies from roughly 30 to a few hundred meters, depending on obstructions in the environment. The signal emitted by 802.11b base stations are radio signals and as such are affected by terrain features and objects between the transmitter and receiver, meaning that the signal travels further in an open, outdoor environment than inside a cluttered office space. Water also absorbs the radio signals, and since the human body consists mainly of water the amount of people between the receiver and transmitter will affect the signal.
The 802.11b standard operates in the unlicensed 2.4Ghz radio spectrum. This means that it is free, not requiring any licensing or fees to be used, but also that there is no guarantee that other applications will not interfere with the wireless transmission. For example, in some cases it has been noted that microwaves, baby monitors and laboratory equipment interfere with the radio waves from wireless networks [10], which is an unfortunate and undesirable byproduct of using this standard. The same effect is encountered when different wireless networks overlap or even when individual overlapping wireless access points in the same network use the same part (channel) of the frequency band.
This interference when overlapping wireless access points share the same part of the
frequency band prompts thoughtful deployment and configuration before use. In practice the 802.11 standard uses one of two signal schemes to facilitate operations, namely Frequency Hopping Spread Spectrum (FHSS) and Direct Sequence Spread Spectrum (DSSS). FHSS works by transmitting over a different carrier frequency at different times. If transmits a small amount of data over a channel for a certain amount of time before seamlessly changing to a different predetermined channel within the same frequency band. DSSS sends several low powered redundant signals over the breadth of the spectrum of which only one needs to received and assembled correctly for the transmission to have been a success.
802.11a is another standard under the IEEE 802.11 umbrella which – quite paradoxically when considering the name – was released in 2001, after the more commonly user 802.11b standard. 802.11a offers powerful wireless networking with transmission speeds up to 54Mbs, making it a viable alternative for demanding networking tasks like video conferencing. Unlike the 802.11b standard it operates in the 5Ghz spectrum band which is also unlicensed but less crowded, which unfortunately makes the two technologies incompatible. There is some dual- band equipment that enables wireless trafficking over both of these standards, but to this point the current technology only allows the two technologies to coexist, not interoperate.
802.11g is a third standard which tries to combine the best features of 802.11a and 802.11b, providing 56Mbs wireless networking over the 2.4Ghz spectrum band. It is more costly than
802.11b but offers backwards compatibility, thus being able to work with network adapters designed for the older technology.
1.6.2 Radio Waves
The transmission between wireless access point and the network cards attached to portable computers is conducted using radio waves. The inherent nature of the radio waves causes them to behave in a seemingly erratic way in a typical indoor, cluttered environment,
numerous factors affecting the path the radio waves end up taking between the transmitter and the receiver as well as the time it takes them to get there.
Multipath fading is a term used to describe the phenomenon where radio waves reach a receiver through multiple paths. When the radio signal reaches the receiver through multiple paths one of two things can happen; either the two waves are received in phase, causing them to reinforce each other and strengthen the signal or they are received out of phase, producing a weak or fading signal. Alterations in the transmission path might change the phase
relationship between received signals and cause periodic fading.
Diffraction is the innate ability of a radio wave to bend around a sharp corners or obstacles and is dependant on the relationship between the wavelength of the signal and the diameter of the object. Signals with low frequencies are able to turn sharper corners than high frequency signals, the latter which may create zones of radio shadow behind large objects.
Radio waves are partially reflected when they hit solid objects in the environment such as walls or ceilings, bouncing the signal around until it eventually fades. It is only partially reflected, however, part of the signal penetrating the object with the degree of penetration depending on the material and thickness of the solid.
The one easily calculated factor when modeling the behavior of radio waves is the free space path loss incurred as the distance between the transmitter and the receiver increases. This is a function of the signal’s frequency as well as the distance and measured in decibel. As the distance between the transmitter and the receiver increases the attenuation of the signal is increased by -6dB because of decibel being a logarithmic scale, not a linear one. This aspect of radio wave propagation modeling specially impacts on the design of the system since it dictates the basic relationship between the received signal strength indication and the
approximate distance from the access point. This is important since the signal range is divided into a number of discrete intervals and if the signal strength indication is used instead of the approximate distance the intervals will vary greatly in size. If -40dB to -46dB was one
discrete interval and -46dB to -52dB was another, the second interval would be twice the size of the first.
1.7 Related Work
During the years there has been quite some work done in designing systems for positioning users in the environment, the precision in most cases ranging from good to acceptable. Most of the systems to date, however, are bound to using additional hardware on top of the underlying network infrastructure or demand extensive measurements to be made at the location where the system is to be used. As was noted in the previous chapter the behavior of radio signals in an indoor environment is a non trivial problem which is why additional hardware is required in the most accurate systems, or extensive measurements are made to
avoid having to mathematically model the radio wave propagation. Furthermore, most of these systems are sensitive to changes in the environment – such as a wireless access point malfunctioning in a system built directly on top of the wireless network infrastructure – as they lack self awareness and the ability to adjust to these changes.
1.7.1 Satellite Positioning
In outdoor environments with unlimited line-of-sight the most efficient way to determine a device’s position is through satellite positioning [11] which is achieved by triangulating the distance of three or more satellites in high orbit with precisely known positions and a receiver, measuring the time-of-flight of the signals and making allowances for known degradation of the transmission. Three satellites will give an accurate two dimensional measurement while a fourth is necessary for determining position in a three dimensional context. Since the radio waves transmitted from the orbiting satellites is not strong enough to penetrate the walls of buildings, dense foliage or solid terrain features satellite navigation is as of yet not available for indoor use.
GPS is generally synonymous with satellite navigation, but it is by no means the only system there is. GLONASS is a Russian satellite navigation system also approved for use by civilians and Galileo is an array of satellites under the authority of the European Union.
1.7.2 RADAR
Developed by a Microsoft Research Group, RADAR offers a solution for indoor positioning that is built directly upon the existing IEEE 802.11 wireless network, meaning that it requires no additional hardware to be set up. RADAR maps a set of locations in signal space –
combinations of base stations and their received signal strengths at a certain position – to physical locations as marked on a two dimensional map and then uses a nearest neighbor in signal space (NNSS) algorithm to approximate the actual position between these.
RADAR offers two different methods for building the database of pre-measured signal space locations; an empirical method and a radio propagation model. The empirical method means that the administrator manually measures the base stations’ signal strengths at different location and associates these with the physical location where they were measured. This is a time consuming process since precision requires a lot of measurements – since the human body absorbs radio waves on the 2.4GHz band, for instance, it is usually required to make up to four different measurements at each location of interest, facing four different directions where the user himself affects the measured signal differently – but offers good precision.
The radio propagation model approximates the degradation of the signal between the base station and the physical location and is certainly a non-trivial problem. An indoor
environment subjects the radio waves to scattering, diffraction and reflections not to mention that different materials dampen the signal strength differently. The phenomenon known as multipath fading means that the transmitted signal arrives to the receiver along different paths, causing fluctuations in the signal as well as distortion. The algorithm employed in RADAR is based on the Floor Attenuation Factor presented in [7].
C nW
C nW WAF
C
WAF nW
d n d dBm d
P dBm d
P ≥
<
⎩⎨
−⎧
⎟⎟⎠
⎜⎜ ⎞
⎝
− ⎛
= *
log * 10 ] )[
( ] )[
(
0 0
P is the signal strength as measured at different distances from the transmitter, with d being the absolute distance between the transmitter and receiver and being a reference distance used for determining free space loss of signal strength. C is the maximum number of walls the signal can pass through before the free space loss becomes greater than the increased loss incurred by further obstacles, nW the number of obstacles between the transmitter and
receiver and WAF the constant – dependant on materials and thickness – describing the signal loss incurred by obstacles. This method offers less accuracy than the empirical method but does not require the extensive measuring that the latter is subject to.
d0
The problem with the method presented with RADAR – which rightfully has won much acclaim and has made the radio map approach the dominant method for indoor positioning – is that it’s extremely sensitive to disturbances in the environment. Trivial activities such as moving a file cabinet, closing a door or unavoidable factors like the amount of people
between the base station and the location to be measured will cause significant fluctuations in the system’s precision. These problems are, however, unavoidable and unfortunate
shortcomings in the underlying technology and extremely difficult to affect.
1.7.3 Active Badge
Active Badge [5][6] was one of the first – if not the first – localization systems and is quite typical for “badge” systems in that it requires additional hardware to be built upon the present infrastructure. The user carries a lightweight, handheld unit that acts as a beacon, at regular intervals transmitting an infrared signal to receivers placed at intervals in the ceiling. The user’s location can now be calculated by a central unit with reasonable accuracy depending on the amount of receivers spread out through the building. Obvious weaknesses with this design is the fact that it does not take advantage of the underlying infrastructure – making it costly and cumbersome to install – and that infrared signals are hampered by florescent lighting and even daylight [8]. Figure 1.2 depicts the design of both the Active Badge and the base stations used in the system’s infrastructure.
Figure 1.2 An Active Badge (right) and a base station (left) used in the system’s infrastructure.
1.7.4 Active Bat
Active Bat is very similar to Active Badge in that it requires the user to be located to carry a lightweight, handheld device which sends a beacon that’s registered by ceiling-mounted sensors but only does so when queried by a controller over short-ranged radio. The main difference between the two systems is that instead of infrared light Active Bat uses ultrasound and measures the time-of-flight to calculate the user’s position. This system boasts an
incredible precision [6] when compared to other systems discussed in this thesis but does so at the expense of poor scalability, cumbersome deployment and expensive hardware.
1.7.5 Cricket
Like Active Bat, the Cricket Location Support System [4][6] uses ultrasound time-of-flight to determine a device’s position but instead embeds the receiver in the mobile device which places the computational burden on the mobile unit but in doing so also addresses the privacy issue. Since the receiver is mounted in the mobile device there is no centralized unit that is aware of the device’s location unless it chooses to divulge it. This approach offers some improvements with decentralized scalability but still suffers from many of the same problems as both the Active Badge and Active Bat.
1.7.6 Neural Network Systems
Another approach to determining the user’s position within the environment is presented in [9]. The described system is in many ways similar to RADAR in that it is first necessary to build a radio map covering the area in which the system should be able to position the mobile user, but instead of a nearest neighbor search algorithm to determine the best approximation of the user’s position it uses a MLP artificial neural network. The artificial neural network (ANN) needs to go through a training phase where it learns how to interpret the received input to represent the correct location. Because of the complex dependencies between the received signal and the actual position of the user, however, this method offers pretty much the same accuracy as RADAR at the cost of a much more complex implementation.
2. Theory
A wireless network infrastructure is created by a number of wireless access points spread out across a location creating a coverage map over the area. Each of these wireless access points has an omni directional signal range which in effect creates a roughly circular area with the access point as the center. The wireless network card attached to a mobile device can measure the degradation of this received signal’s strength when it arrives at the device and thus a relative distance from the access point can be determined. By dividing the signal range of the access points into a number of discrete intervals and calculating into which of these discrete intervals the received signal strength indications of all perceivable access points falls into a signal space location is created. Thus, each signal space location is a set of base stations and the discrete interval into which their received signal strength indication falls into at a certain physical location.
By this methodology the environment is divided into a number of fixed and unique cells or signal space locations which each correlate to a certain location in the physical universe. Each of these signal space locations is stored as a node in an undirected graph with edges
connecting nodes containing neighboring signal space locations, giving the system a more comprehensive overview of the layout of the environment. By matching the current set of received signal strength indications to the undirected graph the system is able to determine the user’s current position. Because of the way in which the nodes in the graph are connected, the system gains additional information which can be used to maintain the system’s autonomy when unexpected events occur, such as a wireless access point malfunctioning or major reconstructions of a building drastically changing the layout.
Autonomy in the context of this system not only pertains to fault detection, tolerance and correction but also the system’s ability to generate, store and maintain a correct representation of the surrounding environment. This independence from user or administrator input is one of the most important aspects of this system as it enables it to function automatically when introduced in a new environment of which it has no prior knowledge.
2.1 Positioning
The predominant approach to accurately determining a user’s position in a wireless network is the method presented by RADAR [1][2] where a radio map is constructed through measuring the received signal strength indications from all the transmitting base stations in a certain area at different points. These signal strength indications are then mapped against a physical location. During runtime the user is positioned through an algorithm approximating the user’s current position by comparing the current received signal strength to the database, producing an adequately accurate value.
The other approach would be to mathematically model the radio waves transmitted by all the different base stations and through these values receive an approximation of the user’s position. This approach presents its own set of problems, however, as the modeling of radio waves in a typical indoor environment – due to things like clutter and multipath fading – is a
far from trivial matter. An accurate simulation such as tracing each individual ray from the base station to the receiver, for example, would require a disproportional amount of
computational power and certainly be impractical for any current handheld device. A
simplified model, such as the one employed by RADAR, can present adequate results, but is still subjected to problems and requires extensive knowledge of the structural composition of the area it is supposed to work in.
The main reason that neither of these two approaches can be incorporated into the system presented in this thesis, however, is that even if it was possible to properly model the signals transmitted by the base stations it would be impossible to determine the user’s position in any sort of context without knowing the locations of the individual base stations. Generally, as this information is not available an alternate approach is needed for a system that is supposed to be able to position users in any given location without requiring additional information about the placement of the base stations.
2.1.1 Signal Space
All systems built upon the existing wireless network infrastructure first measure the user’s position in signal space – meaning the set of received signal strength indications from nearby transmitting base stations – and attempt to correlate this position to exact physical coordinates, usually displaying the result on a map. The system presented in this thesis similarly measures the received, degraded signal strength indication but instead of matching it to a database that maps these indications directly to a physical location these measurements are stored as a signal space location in a node in an undirected graph. What this means is that each position in signal space is a unique set of wireless access points and their received signal strength indications at a certain physical location. Because every user will receive the same
measurement at the same place this works as a positional fingerprint for the physical location, allowing for relative positioning of users. This is more beneficial for offering various services to users than providing them with positional information.
Access Point A Access Point B Access Point C
*
*
*
A
B
C A; B
A; C A; B; C B; C
Figure 2.1. An example of a unique position in signal space where the signal breadth of three access points intersect.
The precision that can be offered by this method increases with the amount of wireless access points covering a certain area, since more access points means that the physical location is divided into more signal space locations. Figure 2.1 shows an example where the signal ranges of three access points – A, B and C – intersect, creating a common signal space location denoted A; B; C.
n
n A
A A
A L
x∉ = 1∩ 2 ∩...∩ −1∩ Equation 2.1
Mathematically this particular signal space location is defined as the intersection of the three individual wireless access points A, B and C. Generally, if x is a position in the physical universe within the signal ranges of n access points, A1 to An, the signal space location L containing x can be written as equation 2.1.
2.1.2 Received Signal Strength Indication
There are three pieces of information that can be deciphered from the information broadcasted from an individual access point in the wireless network; MAC address, Service Set Identifier (SSID) and Received Signal Strength Indication (RSSI). The MAC address is a unique string that is used to identify an individual base station, SSID is the name of the wireless network of which the access point is a part and RSSI is the strength of the degraded radio signal when received by the mobile device. This is a value usually between -100dB and 0dB and can be used to determine the approximate distance between the transmitter and receiver since the radio signal degrades as it travels due to free space signal loss. These three values are what is used to identify an access point at a certain physical location, and the set of these three values for all detectable base stations is what creates a signal space location, giving each physical location a unique fingerprint.
Intervall 1 Intervall 2 Intervall 3
Figure 2.2. An access point’s signal strength divided into discrete intervals.
Only using the received signal strength indication would be an impractical solution to devising signal space locations. There are two major reasons to this. To start with, this
approach would generate a large number of individual signal space measurements and storing all of these values in any sort of data structure would result in a system that did not scale as well as it could. Secondly inconsistencies in the received signal due to variable factors such as weather, amount of people in the room and multipath fading leads to the measured signal strength varying even when the user remains in the exact same place. This would lead to the system consistently positioning the user incorrectly and – which is even worse – interpret a user standing in the exact same spot as moving about haphazardly as the received signal strength indication fluctuates. The logical solution employed in the system presented in this thesis is to divide the signal range of the individual access points into smaller discrete intervals as shown in figure 2.2, sacrificing some precision for robustness and scalability.
The signal strength indication provided by the wireless network infrastructure for the system is actually the attenuation of the signal measured in decibel. This attenuation is the collaborate
sum of the various multipath fading and diffraction that the signal is subjected to as well as the attenuation caused by objects in the signal’s path and free space path loss incurred as the distance between the transmitter (access point) and receiver (wireless network card) increases.
As the attenuation of the signal is measured in decibel which is a logarithmic scale and not a linear one it is evident that directly using the information received from the access point as is would cause the system to behave oddly. For this purpose the received signal strength
indication, the attenuation of the transmitters signal when it reaches the receiver, only serves as a means for finding the approximate distance in meters between them. Frii’s equation [12]
describes this expected signal propagation in open areas, ignoring reflections, multipath fading and antenna gain.
A better model for indoor radio wave propagation is the Wall Attenuation Factor used in RADAR [1] which gives the signal strength P at a distance d as a function of the distance between the transmitter and receiver, a reference distance d0 and signal strength P at the reference distance. A variable n describes the general attenuation of the system. If the part of the equation that regulates signal strength loss for known obstacles is omitted since the system has no such prior knowledge the result is presented in equation 2.2.
) log(
*
* 10 ) ( ) (
0
0 d
n d d
P d
P = − Equation 2.2
This equation can then be manipulated to a more desirable form where the distance between the receiver and transmitter is a function of the received signal strength, as shown in equation 2.3.
n
d P d P
d 10
0) ( ) (
10
−
= Equation 2.3
The variable n, representing a general approximation of signal attenuation due to factors such as multipath fading and antenna gain, must be derived empirically. As has been previously noted simply dividing the signal strength indication into a number of discrete intervals would result of intervals of varying size as it is measured in decibel, a scale that is logarithmic instead of linear. By using these equations the correlation between the signal strength and the distance between the receiver and transmitter can be approximated. Since this scale is linear it can then easily be divided into a series of intervals of approximately the same size.
2.1.3 Discrete Signal Strength
As previously mentioned, the signal strength indication received from the nearby transmitting wireless access points are split into a number of discrete intervals to facilitate robustness and scalability without entirely sacrificing the system’s precision. This division causes the imagined circle created by the omni directional radio signals from the access point to be similarly divided into smaller circular areas centered on the actual access point. When two or more access points are placed close enough that their signal ranges intersect a cell or signal space location is created and this intersection will be denoted as a set of the different access points, as seen in figure 2.1, and their discrete signal strength indication value as noted in figure 2.3.
Access Point A Access Point B
Figure 2.3. Two intersecting access points each with five discrete intervals.
With the addition of discrete intervals, equation 2.1 has to be somewhat modified to accurately describe a set of access points denoting a signal space location. Instead of the intersection of the whole access points signal range the intersection between the intervals within which x, the position in the physical universe, is found has to be calculated.
As the number of discrete intervals that the signal range is divided into increases, so does the precision with which the user can be positioned within the environment. This precision does not come for free, however, because as has been previously stated an increased amount of discrete signal intervals results in a system that requires more computational power as the undirected graph used for storing the different signal space locations will become larger. The important question now is which level of precision is actually beneficial to the system,
especially as the design is geared towards positioning the user in a physically connected space – for example a room – and will reorganize the data structure towards this goal, making an excess of cells superfluous.
The amount of possible cells created by a set of wireless access points can be determined by a simple mathematical formula. If the number of access points (ap) and the desired of discrete intervals (di) into which their signal range is to be divided is known, the maximum amount of cells that can be generated is determined by the number of discrete intervals plus one raised to the power of the number of access points, as shown in equation 2.4.
Equation 2.4
di 1)ap
( +
This is because by multiplying all the possible values of all the base stations with each other presents the maximum amount of cells that can be generated ideally. The actual amount of cells will never be as high as this value, however, as some of the combinations are impossible when the signal range of several access points intersects in certain ways. To some extent the degree of coverage between the access points also affect the actual number of cells that are created.
For example, the above formula would state the maximum amount of possible cells to be in figure 2.3 while a manual count would make the actual number 31. This is because some of the cells are never created, such as the central discrete interval marking a signal space location by itself, and that the degree of intersection affects the final number of cells. When implemented some of the discrete intervals might denote areas inside walls of impassable terrain and the actual size and shape of the regions might differ because of objects disturbing the signal, but this gives an approximation to the maximum number of different signal space locations that may be created by intersecting access points.
36 62 =
The precision of this method is dependant on the amount of access points covering any given location with their signal range and the range of the radio waves. An Apple AirPort Base Station has an active range specified by the manufacturer to be 45 meters but which through use and empirical testing has been shown to be shorter in real-life situations. If the range is considered to be what the manufacturer specified, however, what precision does what amount of covering base stations offer? In other words, how many cells will be created in a certain space?
2 discrete intervals 27
33 = cells
27/45 = 0.6 cells/m 2
4 discrete intervals 125
53 = cells
125/36 = 2.7 cells/m 2
8 discrete intervals 729
93 = cells
729/45 = 16.2 cells/m 2
Figure 2.4. Precision when dividing the access points’ signal range into varying number of discrete intervals.
Figure 2.4 is based on the assumption that a normal position in signal space is covered by the signal ranges of three individual access points and that they cover an area of roughly 45 meters (as specified by the manufacturer) and gives an idea of the precision offered by the system in certain conditions. It is obvious that when the number of discrete steps increases the number of cells rises towards an amount that is both impractical and unmanageable. There is no practical reason for having a precision of 16.2 cells per meter as this system would be both unstable – fluctuations in the received signal strength constantly leading to the system
reevaluating the current position in the environment – and scale poorly, the huge number of signal space locations causing the managing data structure to grow out of control.
Since the system is designed towards all adjacent signal space locations in a room
representing the same physical location the ideal solution would be if each room was covered by exactly one cell, thus being represented by a single signal space location. Since not all rooms have the same size and the division of the cells is handled independently of the layout of the environment this is obviously impossible. A good reference when deciding upon the number of discrete steps is that an average office is somewhere in the range of 2-3 meters squared in size, so the maximum cell-size obviously has to be smaller than this. From the calculations above it can be assumed that dividing the signal range into 3 or 4 discrete steps – receiving a precision of about 1 meter – should be adequate and offer a manageable amount of cells inserted into the data structure.
2.2 The Data Structure
A unique set of access points and their measured discrete signal strength indication describes a specific location within the environment, but a data structure is needed to store all of these signal space locations to give the system a comprehensive view of the physical universe. This data structure needs to be able to display each individual location and furthermore needs to have the ability to discern which of these locations are actually physically connected to each other for the proposed design of the system to function correctly. The solution is to store each signal space location as a node, N, in an undirected graph, G, where physically neighboring nodes are connected by an edge, E. Elementary graph theory then presents equation 2.5.
Equation 2.5
) , (N E G =
The system knows that two nodes are physically connected and creates an edge between them whenever the user moves between two different nodes, gradually expanding the undirected graph until it correctly describes the physical universe. Since it is very likely that the user is unable to move through walls or other impassable terrain it seems that the margin of error for this approach is negligible. Figure 2.5 displays an ideal undirected graph where each node describes an enclosed physical space – for example a room – and each edge describe a door or walkway through which they are connected.
Door Room
Figure 2.5. An example of a small undirected graph.
As previously noted the situation described in figure 2.5 is an idealized one where each signal space location can be exactly matched to one physically connected location. In reality this is seldom if ever the case since the size of a physically connected area varies from as small as an office cubicle to as large as a cafeteria or lecture hall. The system should preferably be able to analyze the data structure and decide which of the connected nodes in the undirected graph are actually contained within the same physical location and present them as such. Two methods were considered for solving this problem.
2.2.1 Counting Edges
The first of the two methods operates by counting the number of edges separating a central node from the peripheral nodes in the system, considering all nodes separated from this central node by up to n edges to be different representations of the same physical location. A subgraph of all the nodes included in all paths of length n centered upon this central node would be generated and assigned to represent the same general physical area. Figure 2.6 illustrates this with four paths leading out from the central node of the graph.
This solution rather poorly reflects reality, however. The original problem that lead to the formulation of this method was that physical locations could come in a variety of shapes and sizes and the system had no way of determining which beforehand, thus making it impossible to specify a general value for n covering most or all situations. Moreover the density of access points covering an area would affect the number of signal space locations being generated in this area and create another variable that must be considered. Besides being unaware of the shape and size of this physical location it is not impossible that the system might have problems selecting the central node in a room, a selection on the fringe of the location resulting in increasingly faulty calculations.
Central node Subgrah node
Figure 2.6. A subgraph containing all nodes separated by
≤n edges.
Figure 2.6 demonstrates this first method by marking all of the nodes that would be members of the subgraph representing the same physical location if n was set to 1.
2.2.2 Node Interconnectivity
The second of the two proposed methods is to study a cluster consisting of a central node c and its neighboring nodes pi, and if the cluster displays sufficient interconnectivity the nodes would be considered to be different representations of the same physical location. For
sufficient connectivity, meaning n nodes which all have edges created between one another, to be able to exist the central node c and its neighboring nodes pi needs to be able to create a spanning subgraph S of the undirected graph G which contains the complete collection of signal space locations stored in the system. This relationship is described in equation 2.6.
n i G
S p
c, i ∈ ⊆ , 1≤ < Equation 2.6
