WiGuide: Indoor System for LBS

Transcription

1 Faculty of Media Engineering and Technology German University in Cairo WiGuide: Indoor System for LBS Bachelor Thesis Author: Supervisor: Ahmed Ali Sabbour Prof. Dr. Amal Elnahas July, 7

2 WiGuide: Indoor System for LBS Declaration I declare that this thesis comprises only my original work toward the Bachelor Degree. Information derived from the published and unpublished work of others has been acknowledged in the text and a list of references is given. Ahmed Ali Sabbour July 11, 7 ii

3 WiGuide: Indoor System for LBS Abstract The rapid advances in wireless technology as well as in the manufacturing of portable devices caused a growing interest in location aware services. A location sensing system responsible for locating a mobile user, is a crucial factor for the success of such services. With the prevalence of wireless hotspots and wireless area networks, the use of a wireless network infrastructure as basis for an indoor positioning system becomes a viable option. In this work, we propose an indoor guide system that provides information about points of interest and objects within the vicinity of the user. Our system can be deployed on a university campus or inside a museum equipped with a wireless network. The system uses the fingerprinting technique to associate position dependent information such as the strength of the received signal with a location. A range based matching algorithm for matching the physical signal strength with the fingerprints in the database is used and its performance is compared to an Euclidean distance based matching algorithm. The properties of the wireless signals and their distribution under several controlled experiments are analyzed. iii

4 Contents Abstract Acknowledgments iii vi List of Figures vii 1 Introduction Motivation Objectives and challenges Thesis organization Background Localization Indoor localization Trilateration Angulation Scene analysis and Fingerprinting Proximity Outdoor localization Global Positioning System Cellular Networks Technologies and related applications Examples of indoor localization systems ActiveBadge (infrared) Cricket (ultrasonic) EasyLiving (scene analysis) SpotON (radio frequency) Indoor tour guide systems Cyberguide Marble Museum IrReal Hippie iv

5 WiGuide design and implementation Localization approach WiGuide design Database Wireless signal acquisition Message passing protocols Fingerprinting algorithm Matching algorithm Range based matching Euclidean distance matching Location presentation Conclusion and Future Work Appendix A Screenshots A.1 Client interface A. Fingerprinter interface Appendix B Installation instructions B.1 Client B.1.1 Installation of the NDIS driver B.1. Configuration B. Server B..1 Installation of Abyss web server and configuration of PHP.. 7 B.. Installation of Oracle 1g XE and running the Database... 7 B.. Configuration B. Running v

6 Acknowledgments I would like to thank all of the people who made it possible for me to finish this thesis. I would like to express my gratitude to my supervisor Prof. Dr. Amal Elnahas for her continuous support and guidance throughout the duration of the project. I would also like to thank my friends and colleagues, specially Raed Rizk, for being there and providing me with moral support. Finally, I would like to thank my family, for helping me getting good education and for being supportive and patient. vi

7 List of Figures.1 Location Based Services applications Location Based Service components Triangulation Classification of outdoor localization technologies Effect of sampling period on the distribution of the signals from access point xa7c Effect of the number of access points on probability of correct location estimation Effect of user presence. AP: xa7c Coverage of an Omnidirectional antenna Experiment location Experiment setup of orientation. AP: xa7c Effect of user orientation while parallel to antenna. AP: xa7c Effect of user orientation while perpendicular to antenna. AP: xa7c8.9 Relation between the average signal strength and the signal distribution 7.1 Client-Server architecture of the system Database ERD The fingerprinting algorithm Determining the group leader Calculating the unfiltered range Calculating the filtered range The client side of the matching algorithm A.1 Location in progress A. Services associated with the room A. Location of the room on the map A. Adding/Removing buildings A. Adding/Removing rooms A. Adding/Removing services A.7 Setting location of a room on the map vii

8 A.8 Fingerprinting of rooms viii

9 Chapter 1 Introduction This thesis presents an indoor location based system that provides users with location sensitive information. The motivation, objectives and the challenges addressed in order to achieve the desired results are outlined in the following sections. 1.1 Motivation With the rapid advances in wireless telecommunication and portable devices technologies, the need for smart applications that could offer personalized services to the mobile users has attracted a lot of research interest in the past few years. This interest has led to the development of a range of services called Location Based Services. Location based services are aimed at making use of geospatial location information as well as user context in order to provide the end-user with useful personalized information. Several usage trends have been observed in location based services among which emergency services, information services, tracking services and entertainment services where the most general. Nevertheless, most of the previous implementations of location based services focused on outdoor localization and outdoor services due to the widespread use of GPS systems. Only a few, discussed in Chapter, addressed indoor localization. Those systems that could locate users indoors relied on installing sensor networks and other options that increased the cost of system deployment. With the prevalence of wireless LAN infrastructure as well as the ever decreasing cost of wireless access points, the use of wireless access points as a basis for location determination becomes an attractive option. 1

10 CHAPTER 1. INTRODUCTION 1. Objectives and challenges The primary aim of this project is to design a system that would be able to locate mobile users in an indoor environment as well as provide them with location sensitive information in the most efficient and cost effective way. The proposed system could be deployed in multiple applications. Inside a museum, for example, tourists carrying a PDA and moving inside the museum sections, would be able to know what kind of art is present in the section they are standing in. Another example would be new students visiting a campus. Students moving with their PDAs or laptops would be able to determine where the nearest food outlets or copy centers are. We based our localization system on a fingerprinting approach, where a fingerprint database is generated for the selected location. This fingerprint database consists of the measured signal strength at different locations. A user s signal strength is measured and compared against the database in order to find the location. To be able to realize that aim, the project is subdivided into several goals: ˆ Creating an application that provides administrators with means to add new buildings, rooms and services through an intuitive interface. Developing an algorithm to generate the fingerprints of a location and providing the administrators with a tool to allow rapid fingerprinting of locations. ˆ Creating a server that could handle multiple clients simultaneously. ˆ Streamlining the client application for easy deployment on devices with limited processing power. ˆ Developing an algorithm to match a location to a fingerprint. The measurement and collection of the wireless signals in different locations imposes many challenges such as the instability of the wireless signal and its lognormal Gaussian distribution, the granularity of the system and its effect on the accuracy and the effect of user orientation on the signals collected. These challenges are analyzed in detail in Chapter. 1. Thesis organization The thesis is organised as follows. In chapter, location based services and their components are discussed. A survey of the current location matching technologies with the focus on indoor positioning techniques is presented. In addition, different applications proposed for indoor tour guide systems are viewed.

11 CHAPTER 1. INTRODUCTION The system implementation including different algorithms used are discussed in chapter. Finally, in chapter, the conclusion and suggestions for possible future enhancements are presented.

12 Chapter Background Several usage trends have been observed in location based services among which emergency services, information services, tracking services and entertainment services where the most general. Emergency services usage include providing the user with recent security alerts and public safety announcements such as earthquake and hurricane warnings. Information services usage include providing the user with the latest news, sports results, weather forecasts, stock quotes as well as routing assistance. Tracking services involve asset, fleet and logistic monitoring in addition to individual person tracking [1]. The anatomy of different location based services applications presented in [], are shown in Figure.1. Figure.1: Location Based Services applications. A location based service system involves four enabling technologies as Figure. depicts []. The user device is the component by which the user requests information

13 CHAPTER. BACKGROUND and views the resulting services. It acts as a method to interface between the user of the service and the service itself. It could be any mobile device such as a smartphone, a PDA, a laptop or a tablet PC. Depending on the portability, any of those devices could perform this function. For example, a PDA or a mobile phone may prove to be the most portable device but this could result in sacrificing some of the information that could be presented due to the limited battery and processing power these devices might have. On the other hand, a laptop could be very un portable (in case of performing an on-foot tour, for example), but can make up for this with the ability to display large maps. The content provider is an entity which provides services and information based on the given location of a user. This provider might be implemented specifically for the purpose of the system, such as in the case of a campus or museum guide. In other cases, it can be a more generic one such as Yellow Pages, providing users with information added through third party content providers. The network component is a crucial part of the system because it is the backbone of information delivery and data transmission between the user device and the content provider. The communication network through which the user is connected to the service provider affects the amount of data that could be transferred to the user. If the user is using a WiFi Wireless network, then it would be safe to assume that the user is on a high bandwidth network and probably a low cost one. On the other hand, if the user is connected through a cellular network like GPRS, then the cost of data transfer should be taken into account as well as the limited bandwidth availability. Every location based service system involves a positioning component of some sort. The positioning component, also known as localization component, is responsible for locating the user at any time.depending on the intended usage of the system, the system could resort to positioning using several technologies including using the mobile network, GPS, WLAN, radio beacons and active badges as will be discussed in the following sections..1 Localization Localization process involves several issues and characteristics that need to be defined before we can start discussing the several approaches. Each approach maybe suitable in one application over the other. According to the following comparison criteria, a choice regarding which location-sensing system could be chosen would be made.

14 CHAPTER. BACKGROUND Figure.: Location Based Service components Physical Position and Symbolic Location Physical position is the position obtained from the system using physical co-ordinates such as those obtained from a GPS device. The GPS device usually provides the user with co-ordinates in the form of lateral and longitudinal degrees as well as altitude with respect to the sea. On the other hand, a symbolic location is a location that has a meaning in its context. For example, a symbolic location system could inform users that they are in the living room instead of giving them co-ordinates. Usually, a physical positioning system can provide the co-ordinates to another interface that would query a database of symbolic locations in order to emulate a symbolic location system []. Absolute vs. Relative In an absolute positioning system, if two mobile devices are located at the same spot, their position will be the same (for example, in a GPS system). On the other hand, if a relative positioning system is used, the device will report its location with respect to a specific reference point or in case of locating other devices, with respect to itself. As the case with physical position and symbolic location, either systems; absolute or relative, can be augmented to simulate the other, though with limited success []. In case of an absolute positioning system, the system could emulate relative positioning by searching for a reference point and modifying the reported location to be with respect to that reference point. In case of relative positioning, one could use the knowledge of the position of other reference points in order to determine the absolute location. Location Computation The location computation parameters defines whether the client is the one that calculates the location using a specific algorithm or that it is a centralized server that manages this calculation. This choice would depend upon the size as well as the designated power consumption of

15 CHAPTER. BACKGROUND 7 the client. If the client computes its location (for example using triangulation from GPS signals), it would need to have sufficient computational power in order to calculate its position in adequate time. This computational power would translate as more electronic power consumption. On the other hand, a device that relies on an external server to locate it by broadcasting a beacon signal or receiving the locations directly from the server would not need much computational power but would mean that the server would have to undertake more responsibilities which might affect the number of clients that could be served simultaneously. On another note, having the server compute the clients locations and possibly saving them raises privacy concerns []. Accuracy and Precision Accuracy represents how much is the location deviated from the actual position. Precision is the percentage of time the system is able to determine the location and achieve the specified accuracy. Both accuracy and precision are important when determining the effectiveness of a specific positioning system. Depending on the intended application, a m accurate system might be sufficient, for example in an outdoor city tour guide. In contrast, the same system might not be effective at all if the intended application was indoor positioning since this m window of error is simply too large that it might span the whole building. New research is directed towards the use of information from several sources by aggregating them into a statistical model in effort to increase both the accuracy and precision of the system. Responsiveness and Latency For the application to be productive, it has to be responsive which means that the location of the user would have to be calculated in real-time. This can sometimes pose a challenge because if the user device was to rely on the server for location calculation, there might be some latency when obtaining the information due to the effects of network propagation delays. The other option of letting the device calculate its position might also prove challenging since the device might not have enough processing power in order to calculate the information in real time []. In general, localization techniques can be broadly classified into main categories: indoor and outdoor..1.1 Indoor localization Several interesting location-aware applications for indoor environments are emerging in the research fields []. Some of the examples are as follows. Asset tracking and monitoring of objects such as projectors for ease of finding as well as theft protection.

16 CHAPTER. BACKGROUND 8 People tracking inside a facility, like doctors in a hospital to quickly route the nearest doctor to emergencies. Shopping assistance in shopping malls by giving the users the ability to find the locations of specific items in stores as well as displaying information about those items such as the price and where is the cheapest place to get them. Museum guide systems that inform the visitors of a museum with the location of the various attractions as well as providing specific information about those attractions once the user is near them. The common factor in the above applications is that they are mostly indoor applications thus, they need an efficient indoor localization algorithm in order to function properly. Several indoor localization techniques have been developed such as [ 1]. The following section presents one of the possible classifications of the different techniques that can be used in localization as well as their implementations Trilateration Trilateration falls under the category of Triangulation where the geometric properties of triangles are used to compute the objects locations. The lateration technique works by measuring the distance to three distinct non-collinear reference points from the object of interest X and then applying a triangulation algorithm to obtain the position of the object. The triangulation algorithm is achieved by drawing a circle of radius d which centered at the reference point, where d is the measured distance from the access point to the object of interest at X. Using three access points, three circles could be drawn which would intersect at a point X. This point would indicate the location of the user. [11]. In order to measure such distance, three general approaches exist. a) Direct This method involves physically measuring the distance between the mobile object and three different reference points which is infeasible and inapplicable to the applications of indoor positioning. It generally involves robots that measure the distances through extension of probes. b) Time-of-flight Time-of-flight works by recording the time it takes for a signal to propagate from a transmitter to a receiver at a known velocity. Since this method involves time measurement, a specific clock resolution must be attainable by the

17 CHAPTER. BACKGROUND 9 Figure.: Triangulation system according to the timing technology used. Ultrasound based systems [1][8], do not require a very good resolution as the waves travel at the speed of sound so the differences to be measured are usually in the milliseconds range. On the other hand, systems using RF Signals [1] such as WLAN and Bluetooth would need to have a much higher clock resolution since radio waves travel at the speed of light so the time difference is usually in the nanosecond range which is times less in the order of magnitude than the ultrasound based technologies []. According to the timing technology, two systems for measuring the time-offlight exist. In the first one, which could be used with ultrasound, the devices must have highly accurate synchronized clocks so that the difference in the time of transmission of the signal and the time of its arrival at the other end can be measured and converted into a distance. For the other system that can be used with WLAN, a method based on round trip time is used where the sender sends a signal and expects a reply. The time it takes for receiving the reply implies the distance between the two nodes [1]. This system overcomes the need for an accurately synchronized clock because the sender s clock would only need to be accurate enough for it to measure the round trip time. A challenge that has to be faced when using the time-of-flight method is how the receiver could distinguish between the direct pulses and pulses arriving after reflecting off objects in the environment, specially that they look identical to the receiver. Such reflecting pulses would have traveled a longer path and

18 CHAPTER. BACKGROUND 1 hence have a longer time-of-flight which would imply to the receiver that the sender is further away than the actual distance resulting in a reduced accuracy. Some implementations such as [1] make use of statistical analysis and analysis of the reflective properties of the environment to prune some of the pulses which are reflected, resulting in improved accuracy []. Some of the systems that are based on the time-of-flight method include GPS[1], ActiveBat[1], Cricket[8], Bluesoft[1] and PulsON[1]. c) Attenuation The intensity of the emitted signal is inversely proportional to the square of the distance from the source. Using this relation and given the signal strength at the source, the measured signal strength could be used to estimate the distance of the object from the source. Measuring the distance using attenuation is usually less accurate than using time-of-flight specially in environments with many obstructions which cause a lot of signal reflection [1] Angulation In angulation, angles instead of distances are used to determine the location of an object. To compute the position, two angle measurements in addition to one length measurement between the two reference points are required. Using an array of multiple phased antennas with a known distance between them, by measuring the phase shift angle between the signal arriving at the first antenna and the signal arriving at the second antenna and given the differences in arrival time and the geometry of the array, the system can compute the angle from which the transmission was emitted [] Scene analysis and Fingerprinting The scene analysis technique depends on analyzing the scene to obtain features that are easily compared and represented. A scene can be described either by images for the location or by the signal strength at that location. In static scene analysis, the observed features are looked up in the database that maps objects to locations. In differential scene analysis, the difference between successive scenes is tracked []. Those differences will imply that the user has moved. If features are known to be at specific locations, the observer can compute its location relative to them. The advantage of such system is that it can act independently from an external server by storing the database locally which means that less power is required for data transmission and the privacy of the user could be maintained. The disadvantages

19 CHAPTER. BACKGROUND 11 include the need for a pre-built lookup database of the features that might need to be rebuilt if the features of the scene change. Fingerprinting is a type of scene analysis. Usually, the fingerprinting approach consists of two stages. ˆ The first stage is the calibration stage. At specific points called reference points within the environment, the signal strength, along with other parameters such as the signal to noise ratio at each point is recorded and saved in a database. For the fingerprinting database to be accurate, a significant number of reference points should be made available which increases the time needed for the calibration phase [11]. ˆ The second phase which is the actual positioning phase where the object measures the signal strength at its current position and then, using a matching algorithm, the measured signal is matched with vales stored in the database in order to obtain an approximation on the user s current position. The matching could be done on the client side which requires that the database of fingerprints is stored locally, thus increasing the storage and memory requirements on the client. The other option is matching on the server side which increases the load on the server and possibly the load on the network. The algorithm by which the location of a user could be determined could either be deterministic [][1] or probabilistic [17][18]. The deterministic method is a straightforward method where the average signal strength received from each access point is measured at the location to be identified and the result is stored as a fingerprint which is a vector F = (f 1, f,..., f N ) consisting of N readings f where each reading consists of an access point identifier and a minimum and maximum Received Signal Strength Indicator (RSSI) value. When matching, the Euclidean distance distance = n (r i f i ) between the actual reading vector R and each fingerprint vector F stored in the database is calculated. The fingerprint that generates the minimum Euclidean distance is considered the best match [19]. i=1

20 CHAPTER. BACKGROUND 1 On the other hand, the probabilistic approach using Bayesian probability works by defining the fingerprint as a likelihood function where the probability of a being at a location observing a fingerprint is given by the equation is calculated [17][]. p(l f) = p(f l)p(l) p(f) Using the probabilistic approach requires choosing a likelihood function p(f l) as well as having pre knowledge of the wireless signal propagation model of the environment. The probabilistic approach yields better results in terms of accuracy [11][1] Proximity The proximity technique is useful in knowing whether an object is near a specific location rather than knowing the exact location. One type of proximity detection relies on the presence of physical proximity detectors such as pressure pads. Another type of proximity detection relies on the usage of automatic ID systems such as ATM machines or point-of-sale credit card machines where accessing an ATM at a certain time gives an indicator of the place of that user at that time. In a university for example, the use of computer login history can help in estimating the possible user location. The accuracy of the proximity detection system is usually lower than other methods so it cannot be used alone and could be used together with another method..1. Outdoor localization Outdoor localization involves locating users in a relatively open environment with no obstacles. The applications of outdoor localization include mapping services as well as providing location sensitive information where the exact location of the user is not necessarily known. Current technologies for outdoor localization can be classified into main classes as in []: ˆ Handset Based ˆ Cellular Network Based ˆ Hybrid Approaches.1..1 Global Positioning System The Global Positioning System (GPS), which is the most widely used outdoor positioning system, relies on the active satellites as well as the redundant satellites deployed by the United States and Russian military. A device that is capable of

21 CHAPTER. BACKGROUND 1 Figure.: Classification of outdoor localization technologies

22 CHAPTER. BACKGROUND 1 locating itself using GPS is able to know its latitude, longitude, time as well as height at anytime during the day and night. For the positioning to work, the user device has to calculate its distance from four or more satellites. This distance can be calculated by measuring the round trip time difference [1]. Generally, GPS cannot work indoors or in places with heavy forestations due to the fact that it needs at least a signal level of -1Bm in order to work. For the GPS to work indoors, retransmitters called pseudolites have to be installed. These pseudolites work by relaying the GPS signal from outside the building inside. The drawback of using this method is the reduction in accuracy []..1.. Cellular Networks Cellular networks such as GSM (Global Standard for Mobile Communications) and UMTS (Universal Mobile Telecommunication System) are network systems used to provide mobile telecommunication services. Locating a user on one of these networks can be done differently as follows. The first method is the cell-of-origin method, the coverage area is divided into cells. At any given time, the mobile user, using the cellular network, has to register with the cell he is currently in, which enables the system to locate him at any given time. The radius of the range that a cell could cover is in the order of a few kilometers []. Positioning accuracy of up to 1km can be achieved using GSM. The other two methods that can be used with cellular networks are somehow similar to the methods used indoors. The first one is based on a trilateration timeof-flight approach as described before. The second method is based on the fingerprinting approach where the signal characteristics at several areas are measured and stored in a database. Both of these methods have a better accuracy than the cell-of-origin method averaging between 1m to m [].. Technologies and related applications..1 Examples of indoor localization systems Several research and commercial localization systems have been developed. In the following section, some of those systems are examined ActiveBadge (infrared) The ActiveBadge system was developed by the Olivetti Research Laboratory [9]. It uses infrared transmitters that are attached to users badges. Infrared sensors are attached through out the building. A central server collects the unique tags

23 CHAPTER. BACKGROUND 1 that are transmitted by the users badges and associates them with the infrared sensor that is fixed inside the room. ActiveBadge localization system is room based localization system, and it provides a programmable application interface that allows applications to be developed on top of the system...1. Cricket (ultrasonic) The Cricket localization system [8] uses ultrasound technology. Emitters are distributed within the environment and the receivers are embedded within the objects to be located. The devices that are to be located perform triangulation and timeof-flight calculations in order to locate themselves. The Cricket system can identify and ignore ultrasound signals that result from reflection, thus increasing the accuracy. Embedding the receivers in the devices and forcing them to computer their own location increases the power and processing requirements of such devices...1. EasyLiving (scene analysis) The EasyLiving system is a system based on image analysis []. It uses a Digiclops D camera to record the features of a location and compare it to a database of scenes. EasyLiving requires high-performance cameras as well as systems that have adequate processing power in order to analyze the images captured and match them with the database...1. SpotON (radio frequency) SpotON is an ad hoc location-sensing system [1]. It uses low cost radio tags as well as radio receiving base stations as a basis for the localization system. Each radio tag transmits a unique identification number. Each base station then, sends a central server the unique number along with the signal strength that it is receiving from the tag. The central server then uses the trilateration technique to compute the location of the tag... Indoor tour guide systems There have been several previous implementation of indoor tour guide system. In the following section, some of those systems are described....1 Cyberguide Cyberguide is a tour guide system developed in Georgia Institute of Technology in order to guide users during exhibits held in the GVU (Graphic, Visualization and Usability) center []. The system was implemented on an Apple MessagePad 1

24 CHAPTER. BACKGROUND 1 with the Newton 1. operating system as well as pen-based PCs running Windows for Pen Computing 1.. The design of the system is highly modular with three main components namely positioning, mapping and information components, working together in order to provide the services. The first component is the navigator which is concerned with positioning. It is responsible for determining the user s location as well as orientation on the map. The indoor version of Cyberguide used an active beacon system in order to locate the users. Using an array of infrared remote controls, each remote control transmits a unique beam pattern. Using an infrared receiver that is connected to the Apple MessagePad using a serial port, these unique patters are used to identify which cell the user is currently in, hence the location of the user could be updated on the map. The range of each infrared cell is small, so it makes the process of wide deployment of infrared remote controls a rather expensive process. The mapping component, called cartographer, provides the maps of the environment that the system is used in. The maps are interchangeable and can be based on bitmap graphics or vector graphics. Bitmap maps are easily obtained through scanning existing hard copy maps but were difficult to handle during scaling and rotation as well as representing the points of interest. The vector map proved to be of higher quality and was easier to rotate and scale. Drawing a path representing a tour for example was easier on the vector based map but expensive in terms of computation, specially on small hand held devices with limited processing power. The third component, the librarian is the information component. that represents the database of information attached to each point of interest. Each point of interest is marked on the map with a star. When the user selects a star, its name is displayed and the information attached to it is fetched from the information repository to be displayed on the screen. The system also provides means for the user to search the repository.... Marble Museum This tour guide application was developed to support the human tour guides in the Marble Museum in Italy. The system was implemented on Compaq IPAQ PDAs running the Windows CE operating system with an additional 1Gb memory card that stores the various multimedia such as photos and audio files needed for the tour. The usage of an internal memory reduced the need for a high bandwidth network in order to support the different users requesting the files. The location determination technology used in this application is infrared beacons. Each entrance to a section in the museum is fitted with an infrared transmitter on the ceiling. The transmitter can cover up to 9 and the user is assumed to hold the PDA almost vertically when entering a new section. The transmitters transmit

25 CHAPTER. BACKGROUND 17 8 characters per second. From those 8 characters, are used to identify the section and more to validate the correctness of the signal. When the PDA receives the 8 character signal, the application will translate it into a section, play a sound to confirm and then automatically present the user with a map of the current section. As soon as the user is in a section, an audio commentary regarding that section is played which minimizes the users interaction with the PDA and letting them concentrate on enjoying their visit. Currently, the system cannot adapt the map to the user orientation and offers limited support of path finding, where a user requests to see the route to a specific artwork [].... IrReal IrReal is a system that was originally implemented in the CeBIT computer fair that was held in Hannover. It is currently being used in the Computer Science building in the University of Saarbrücken in Germany to provide users with information such as the current food menu as well as the bus schedule of the bus that passes infront of the computer science building [7]. It is based on strong infrared transmitters and PalmPilot PDAs but can as well run on any IrDA [8] enabled PDA. The system is based on an adapted idea of the European video text system. The idea behind this project was to use an infrastructure of strong infrared transmitters that have a range of about meters to transmit location sensitive information without the need of an additional database stored on the client or an access to a wireless network to obtain information. Each infrared transmitter is connected via a serial port to the nearest computer workstation. Using the localized information on this computer, or information obtained through a wired network, the transmitter sends formatted text and graphics. Each mobile device is loaded with a small application called BrowsIR that is capable of displaying the information that is currently being transmitted by the nearby infrared transmitter. To control the localization of the broadcasted information, each infrared transmitter is tuned to a specified range which also increases the accuracy as infrared beams do not cross walls. In addition, the infrared receivers on the PDAs are sensitive to orientation. This orientation sensitivity has allowed the system to be deployed in such a way that, depending on the user s orientation, different information will be received.

26 CHAPTER. BACKGROUND Hippie Hippie is an Internet-based nomadic information system which allows users to access information independent of the device that they are currently using. It has been developed for cultural environments providing various information about art exhibitions [9]. The system allows pre planning of the visits at home through an Internet browser. A web based interface provides several visitor activities such as browsing the exhibit database, printing summary information, checking of the opening hours as well as ticket pricing and making virtual tours. The location of users in the exhibit is detected by means of an infrared infrastructure which detects the position of the user as well as an electronic compass which detects the orientation. The infrared transmitters are fixed next to the various art items where each transmitter sends out a unique ID identifying which art item is currently being looked at. An infrared receiver that is either attached to the clothing of the visitor or is embedded in a hand held device receives the ID. Information related to the received ID is downloaded through wireless LAN and is presented to the user in the form of text, pictures as well as audio clips. Hippie allows the visitors of the exhibition to search the different sections, leave comments on the art work and contact other visitors by leaving them notes. Finally, Hippie takes into consideration the user preferences by adapting the results to display artwork that the user is interested in as well as providing tips about similar artwork.

27 Chapter WiGuide design and implementation This project has been designed to serve as a tourist guide in a museum. However, it has been tested on the campus of the German University in Cairo. It is a guide application where a student moving with his laptop would be able to locate himself within the university buildings. Upon successful location, the application would display his location on the map and then provide him with the services and points of interest that are in his vicinity. In this chapter, the approach chosen for localization as well as the challenges faced, the matching algorithm are described in details..1 Localization approach Fingerprinting is chosen as the localization approach as fingerprinting has a higher accuracy then other approaches such as time-of-flight specially in indoor environments with multi path effects. Also using methods such as the time-of-flight requires accurate synchronization of the clocks of both the mobile device and the access points because an error small error in time measurement can result in an relatively large error in the detected position []. As the fingerprinting approach relies heavily on the signal strength, many factors needed to be studied before implementing our system such as the sampling period, grid granularity and the length of the fingerprint vector. Sampling period Building fingerprint database requires selecting calibration points where samples of the singal strength are measured and saved in the database to represent this point. As the signal can be considered as a random property and has a lognormal Gaussian distribution [19], taking more than one reading is necessary. In addition, the duration of the sampling period as well as the sampling rate are important factors to be determined. In order to decide upon 19

28 CHAPTER. WIGUIDE DESIGN AND IMPLEMENTATION the proper values, a testing experiment is done at a specific location where the number of samples taken were chosen to be 1,, and 1 readings. The sampling rate was chosen to be one sample per second. The results as shown in Figure.1, show the strength of the signal received from the nearest access point which was xa7c8 as opposed to the frequency of appearance. We define a leader signal to be the one with the highest frequency. It is worth noticing that the RSSI varies in all cases between 17 and. In order to decrease the variation, a filtered range, explained later in more detail, which is a range excluding the readings that are below a certain threshold, was calculated. This threshold was chosen to be % of the number of readings of the leader. The effect of choosing different sampling periods for measuring the signal strength at a specific location on the range and the filtered range is summarized in Table.1. The sampling period chosen is seconds. It was chosen because 1 seconds did not always yield a representative range, specially in cases where the signal was unstable. Choosing a period that is more than seconds would impose a large delay on the user of the system and would reduce the user experience. FREQ FREQ Readings 1 RSSI Readings RSSI FREQ FREQ Readings RSSI 1 Readings 1 RSSI Figure.1: Effect of sampling period on the distribution of the signals from access point xa7c8 Grid granularity The distance between the different reference points to take will affect the granularity of our system. As shown in Table., making the spacing between the grid points small, and increasing the number of reference points

29 CHAPTER. WIGUIDE DESIGN AND IMPLEMENTATION 1 will increase the accuracy of the system but at the same time, it will increase the time it takes to generate the database dramatically [11]. It will also increase the search time while trying to obtain a match. It is also worth saying that the increasing the number of reference points beyond a certain limit will have little to no effect on the accuracy, as the rate of increase of accuracy decreases because there is simply not enough variations in the signal strength [19]. Length of fingerprint vector The number of access points to consider when the fingerprint database is being built as well as when the client is trying to match his location is important. In a study made in [1], Figure. shows that a higher number of access points does increase the precision, specially when the standard deviation of the samples is large, though the rate of increase of the precision decreases when the number of access points is greater than five. Unfortunately, the number of access points that are observed by a client might be less than five most of the times. This has led to the choice of relying on three access points. Figure.: Effect of the number of access points on probability of correct location estimation samples leader sd f ilteredsd range f ilteredrange to to to 7 to to 7 1 to to 78 9 to 77 Table.1: Effect of sampling period on signals from access point xa7c8

30 CHAPTER. WIGUIDE DESIGN AND IMPLEMENTATION points error(m) Table.: Effect of number of reference points on the distance error User body and orientation In most modern laptops with built in wireless cards, the wireless antenna is embedded in the LCD screen. The human body is mainly composed of water. Since the resonance frequency of water is. GHz and the frequency at which the wireless signals propagate is. GHz, the human body acts as a barrier and absorbs a significant amount of the signal that is propagating [19]. Due to this, the received signal strength is affected by the presence of the user. Due to the previous observation, the orientation of the user affects the strength of the signal received, depending on where the user is standing with respect to the access point. The effect of the presence of the user is also studied in Figure.. Although it is insignificant to our application because a user will always be holding the device, nevertheless, as the graphs in Figure. show, the difference between the signal distribution in presence and absence of the user is negligible. The antennas on most wireless access points are omnidirectional antennas. As Figure. shows, omnidirectional antennas transmit their signal uniformly in one plane only. Hence, the positioning of the antennas on the access points and the position of the user relative to the antenna affect the signal strength. An experiment like the setup in Figures. and. has been done. The experiment was done by measuring the signal strength for a period of seconds at the locations marked parallel and perpendicular. Those locations represent measurements that are parallel to the axis of the antenna and perpendicular to the axis of the antenna, respectively. In Figures.7 and.8, the effect of both, the user orientation as well as the position of the user with respect to the antennas is shown. The graphs show that there is indeed a difference between the readings taken parallel to the antenna axis and the readings taken perpendicular to the antenna access. The results taken perpendicular to the antenna access are more stable (having the same leader), than their parallel counterpart. As for the orientation, it is obvious that the orientation affects the distribution of the signals, specially when the readings are taken parallel to the antenna. When the readings are taken perpendicular to the antenna, the

31 CHAPTER. WIGUIDE DESIGN AND IMPLEMENTATION With User Without user Figure.: Effect of user presence. AP: xa7c8 effect of the user orientation affects the frequency of the leader. The readings taken in the South and West orientations have a leader frequency which is greater than 1, while the readings taken in the North and East orientations have a leader frequency which is less than 1. Signal sensitivity The mapping between the actual radio signal strength and the signal strength reported by the wireless card in dbm is non linear. It is in fact logarithmic [19]. This leads to the signal strength not being different enough to distinguish between different locations close to the access point which leads to ambiguity with respect to certain locations. In Figure.9, it is obvious that as the average signal strength increases when the user is near the access point, the more the distribution becomes left skewed. This means that as we move away from the access point, the probability of an error in location detection increases. At the same time, as we move closer to the access point, it becomes more difficult to identify the location, because the values of the RSSI reported by the wireless card would be very similar to each other. [19]. Who senses the location The wireless signals that are used to locate the user can be measured at both, the client device as well as the access point that is transmitting the signal. If the signal measurement is going to be done on the access point, access to the API of each access point must be provided in

32 CHAPTER. WIGUIDE DESIGN AND IMPLEMENTATION Figure.: Coverage of an Omnidirectional antenna order to program it to provide a list of clients that are currently associated with it. This approach was unfavorable due to the complexity of the different access points APIs and due to the fact that it would require a different implementation for each access point manufacturer. On the other hand, the second approach which relies on the client device to read the signals received is more appropriate for future expansion because if more access points are added, no further reprogramming of the access points is required. Who calculates the position Calculating the position of the user requires access to the database of fingerprints and locations. Implementing the location algorithm on the client side would require storing the database on the client s device which could pose a problem on small devices such as PDAs because of the large memory requirements that a database management system needs. Location computation also needs a device with adequate processing power so that the localization could be done in a short time. For these reasons, calculating the position on the server is considered a better solution although it puts a greater load on the server as well as the network due to the network traffic needed to transmit location information back and forth between the server and the client.

33 CHAPTER. WIGUIDE DESIGN AND IMPLEMENTATION Figure.: Experiment location Figure.: Experiment setup of orientation. AP: xa7c8

34 CHAPTER. WIGUIDE DESIGN AND IMPLEMENTATION North South East West Figure.7: Effect of user orientation while parallel to antenna. AP: xa7c North South East West Figure.8: Effect of user orientation while perpendicular to antenna. AP: xa7c8

35 CHAPTER. WIGUIDE DESIGN AND IMPLEMENTATION 7 Figure.9: Relation between the average signal strength and the signal distribution

36 CHAPTER. WIGUIDE DESIGN AND IMPLEMENTATION 8. WiGuide design Our WiGuide system has been designed according to the client-server architecture. The system model is shown in Figure.1. The client is a thin client with only an interface to the application. This design would allow the system to be easily implemented on devices with limited processing power and limited memory availability such as PDAs and mobile phones. The client collects wireless signals through the PlaceLab library. Figure.1: Client-Server architecture of the system The server consists of a database as a backend to store the various data entities which are described in detail in the following sections. A background service is always running at the server listening for connections from clients and processing the information. Finally, a webserver is installed on the server as a frontend for providing the results to the client. The server and the client interact together using protocols that use XML as the message format. Several external libraries and applications have been used in order to facilitate the development of the application.

37 CHAPTER. WIGUIDE DESIGN AND IMPLEMENTATION 9 PlaceLab NDIS Wrapper to add functionality to Java to enable the collection and use of raw data acquired from the wireless card. Hibernate a powerful high performance object/relational persistence and query library to map database entities into classes that conform to the Enterprise Java Beans (EJB) standards. JDOM v1. a Java-based solution for accessing, manipulating, and outputting XML data from Java code. It is used in implementing the protocol of passing messages between the server and the client. OpenChart v.1.1 an open source Java library and toolkit for creating different kinds of charts and embedding them into web applications or Swing applications. It is used in the statistical analysis application for drawing bar charts of signal distributions. The Jakarta Mathematics Library is a library of lightweight, self-contained mathematics and statistics components. It is used in the fingerprinting algorithm as well as the location algorithm. It is also used in the statistical analysis application. Oracle 1g XE as the back end database for storing the different data required by the application. Abyss Web Server v. as the gateway to provide the users with the services and content in the form of HTML web pages. PHP as the programming language to interface with the database and render the content with the help of the Abyss Web Server...1 Database The database consists of tables that are pictured in the Entity Relationship Diagram (ERD) in Figure.11, and are described below. AccessPoint This table represents access point entries. An access point is identified by its MAC address. Building This table contains the different buildings that could be present within the environment. A building can be given a name, for example, B. Room A room is the main area where a client is located. A room belongs to a building and has a name, a number and two co-ordinates that represent the location of the room on the map. These co-ordinates are pixel co-ordinates and not actual physical co-ordinates.

38 CHAPTER. WIGUIDE DESIGN AND IMPLEMENTATION (1,1) (1,) Figure.11: Database ERD

39 CHAPTER. WIGUIDE DESIGN AND IMPLEMENTATION 1 Section A section is a point within a room where a reading is taken. The room is divided into 1 sections that are meters apart. In the case that one side of the room is longer than the other or if the room is not symmetric, then the sections are divided arbitrarily so that they could cover the whole room. Reading A reading consists of a signal strength received from the access point and a MAC address identifying this reading with the access point that it came from. Service When a client is located, he is presented with a list of services that are associated with the room. A service could represent actual entities that are physically present within the room such as a computer or a printer, or it could represent points of interest that are near that room... Wireless signal acquisition Java does not provide a way to access the wireless card and obtain readings from it directly. Fortunately, Java provides a feature called Java Native Interface (JNI) that allows a Java application to interface with libraries that are not written in Java. In Windows, the driver that is responsible for the wireless cards is called NDIS driver, which stands for Network Driver Interface Specification and is written in C++. A possible solution to access the wireless driver is to write an interface to the NDIS C++ driver using JNI. The other solution relies on a library that has been developed in Java and already wraps the NDIS C++ driver. This library is called PlaceLab []. PlaceLab has been developed by Intel Research in Seattle and has been released to the open-source community. Using the PlaceLab library to read the signals from the wireless card involves initializing a WiFiSpotter and then invoking a method in the PlaceLab library to get the array of readings. The above is repeated according to the sampling period that is chosen with a delay of one second between each sample and the other... Message passing protocols Messages are passed between the client and the server in situations. ˆ When the client passes the observed signal vector to the server. ˆ When the server passes the location information back to the client. The protocol needed to transfer information between the server and the client, a protocol needed to be implemented. The chosen method was to implement the

40 CHAPTER. WIGUIDE DESIGN AND IMPLEMENTATION protocol using Extensible Markup Language (XML) which is a standard for data sharing and data exchange between different systems. The Java Document Object Model (JDOM) library is an open-source Java library that is used to create, parse and interpret XML documents []. JDOM is used in order to construct the XML protocol messages that are sent between the client and the server and is also used in parsing those messages to obtain useful information. The protocol implemented is composed of main components as follows: ˆ Client-Server protocol defines the message that is passed from the client to the server. It includes information about the observed signal vector that is supposed to be matched. The signal vector consists of a series of access points where the Medium Access Control (MAC) address of each access point is recorded along with the Received Signal Strength Indicator (RSSI) value. <?xml version="1." encoding="utf-8"?> <message> <accesspoint apid=""> <mac>18dac</mac> <signal>7</signal> </accesspoint> <accesspoint apid="1"> <mac>17f71f8</mac> <signal>1</signal> </accesspoint> </message> ˆ Server-Client protocol defines the message that is passed back to the client either indicating a successful location or indicating that an error has occurred. It consists of a flag that indicates whether there were results returned from the database or not. If results have indeed been returned from the database and the client could be located, the server sends the database ID of the room where the client has been located, along with the exact row and column of the reading. The latter row and column are only used in determining the precision of the system and they make no difference in the result of positioning. An accuracy level indicator is also returned to specify the confidence level of such detection. No results found <?xml version="1." encoding="utf-8"?> <content valid="" />

41 CHAPTER. WIGUIDE DESIGN AND IMPLEMENTATION Successful location <?xml version="1." encoding="utf-8"?> <content valid="1"> <room roomid="7" accuracy="" nr="" nc="" mapx="9" mapy=""> <name>photocopier room</name> <num></num> <building>b</building> </room> </content>.. Fingerprinting algorithm The fingerprinting algorithm is the algorithm that is used to build the database of fingerprints. It is used in the initial stages of deployment. For a user to be located within a certain room, it has to be fingerprinted. The room to be fingerprinted is divided into 1 sections that are roughly meters apart. If the room is not symmetric, the 1 sections are divided so that they cover the area of the whole room. The fingerprinting algorithm starts by collecting the reading samples then, processing those readings because of the variations of the signal strength between samples, as shown in figure.1. Processing the readings implies selecting a leader which represents the reading that has the maximum frequency of occurrence, then calculating a filtered range of signal strengths that can be represented by that point. The calculations of the frequency distribution are done using the Jakarta Mathematics Library. The details of each of those steps is explained below. Collecting samples The process of fingerprinting first starts by standing in the section to be fingerprinted and then initiating the fingerprinting algorithm. For seconds, which is the time that has been decided as the sampling period, the system administrator should be standing still while the computer is recording the wireless signals received. Grouping signals After the seconds have passed, the algorithm will start aggregating the signals received by the MAC address of the access point that transmitted it. After the process of aggregation, we will end up with N groups representing N access points that were observed during the second sampling period. Usually the number of access points observed varied from one point to the other, depending one how many access points are seen by that point.

42 CHAPTER. WIGUIDE DESIGN AND IMPLEMENTATION times Yes No Figure.1: The fingerprinting algorithm Determining the leader of each group The leader is the signal whose received signal strength indicator (RSSI) is the most within a group of signals belonging to the same access point. The process of determining the leader works by first sorting the group by the frequency of repetitions of each RSSI. Normally, the first reading in the group should now be the leader. But, in the case of several readings having the same number of repetitions, the leader in this case is the mean of those readings. The flowchart for the algorithm of determining the leader is presented in Figure.1. Determining the unfiltered range and the filtered range The unfiltered range is simply the minimum and maximum readings to be considered in the group to be used in calculating the filtered range explained below. In the case of having only one leader, those minimum and maximum values are both equal to the RSSI of the leader. Otherwise, the minimum is the minimum RSSI

43 CHAPTER. WIGUIDE DESIGN AND IMPLEMENTATION Store as add to Yes, add to Store as No Figure.1: Determining the group leader that has the same number of repetitions as the leader of that group, and the maximum is the RSSI value of the leader of that group. The flowchart of calculating the unfiltered range is presented in Figure.1. Because sometimes the variations of the signal within a group are so large, they often include signals with very low RSSI values and having very low frequencies as well. Those signals hugely affect the value of the standard deviation of the group which in turn, affects the accuracy of the system. As a solution for this problem, a threshold defined as the minimum percentage of frequency of occurrence of a reading for it to be considered in calculations, is used in order to control which signals take part in calculating the standard deviation of a group as shown in Figure.1. It is used to filter out readings that are very infrequent but evenly spaced within the range so that they don t affect the standard deviation of the readings. The threshold frequency is calculated by multiplying the threshold by the frequency of the group leader. After that, when calculating the filtered standard deviation, any reading whose frequency is less than the threshold frequency is disregarded. The filtered range is determined by just adding the filtered standard deviation to the maximum value of the unfiltered range that has been calculated before,

44 CHAPTER. WIGUIDE DESIGN AND IMPLEMENTATION Yes No Figure.1: Calculating the unfiltered range and subtracting the filtered standard deviation from the minimum value of the unfiltered range. Finally, the strongest and most frequent access points along with their RSSI and the minimum and maximum values of the filtered range are stored in the database, associated with the current section. This process repeats until no more sections for that room are left to be fingerprinted... Matching algorithm When a client wants to get located, the client application starts taking readings for seconds, determines the group leader just like in the fingerprinting algorithm, and constructs a signal vector, containing the signal strength of the top three groups. The top three groups are determined by multiplying each group leader by its frequency, sorting the groups using this number descendingly and then selecting the first three groups. The signal vector is then sent to the server through an XML message as in Figure.1. The server upon receiving the client s signal vector, runs a matching algorithm to match this vector against its database entry, then sends the room ID and the location on the map to the client. In WiGuide, two algorithms have been developed in order to match a user: a range based algorithm, and an Euclidean distance based algorithm, presented in [].

45 CHAPTER. WIGUIDE DESIGN AND IMPLEMENTATION 7 Yes Yes, add to No No Figure.1: Calculating the filtered range According to the selected matching algorithm, the server would then perform the matching and determine a best location match of the user. The server would then send the room ID of the matched room to the client for the next step....1 Range based matching The range based algorithm is based on the idea that the received signals from the client are being matched with the ranges that are stored in the database for each access point. If the server gets an XML message with observations A,B and C, it tries to find a reading, where an access point with MAC address A.MAC exists and the received signal strength A.RSSI lies within the range of that reading. The server tries this with each observation received from the client in order to filter down the result set.

46 CHAPTER. WIGUIDE DESIGN AND IMPLEMENTATION 8 times Figure.1: The client side of the matching algorithm The results obtained are then grouped by the rooms that are associated with the readings. The results are then sorted by the number of occurrences that a room appears in the result set. Sorting by the room occurrences will ensure that if a reading is repeated several times within a room, this means that this room has a higher probability of being the correct location. Finally, the server returns the ID of the first room in the results as well as the x and y coordinates for the map that are associated with that room.... Euclidean distance matching The Euclidean distance between two vectors, R = (r 1, r,..., r n ) and F = (f 1, f,..., f n ), can be represented by distance = n (r i f i ) i=1

47 CHAPTER. WIGUIDE DESIGN AND IMPLEMENTATION 9 where n represents the fingerprint size and the observation size. The fingerprint size is the number of access points that has been saved in the database for a particular reading vector. The observation size is the number of access points that have been seen in this observation. The value of n is equal to the length of the vectors R and F, where R = F. The length of the vectors be equal to, or 1. At the beginning, the server issues an SQL query to return all the fingerprints that have sizes equal to the size of the observation vector. The Euclidean distance based algorithm works by calculating the Euclidean distance between the observed readings vector R and each fingerprinting vector F returned from the database. Sometimes, only a subset of the access points in an observation can be matched with a fingerprint vector, in this case, a penalty which is equal to the square of the RSSI value of the reading that was not found in the fingerprint vector is added to the value of the distance. Finally, after calculating the Euclidean distances between the observation vector and all the fingerprints that were returned, the fingerprint vector that yields the minimum distance is considered the best match. The room ID associated with this fingerprint is returned to the client, as well as the x and y coordinates for the map that are associated with that room... Location presentation An XML reply from is sent from the server to the client as explained above. The client, then, parses the room ID part and includes it in an HTML request that is sent to a page written in PHP stored on the server. This room ID is used by PHP to query the database for services associated with this room and then output it in an HTML form. The client then just displays this HTML page and presents the user with the services located in his/her current location. In addition, the x and y coordinates sent by the server are used to plot this point on the map as seen in Figure A..

48 Chapter Conclusion and Future Work Designing location-aware systems that provide location sensitive information involves several components that cooperate in a seamless way. In this project, a WiGuide system is proposed that provides a push-based locationbased service. The system is based on the client/server architecture and consists of main components: a localization component and a data related component. The localization component uses the finterprinting algorithm along with a matching algorithm to locate a user. The matching algorithm is either using the Euclidean distance based or a range based algorithm. The system is able to locate a user correctly with an accuracy of 7% for the Euclidean distance based algorithm and % for the range based algorithm with a mean error of meters which is acceptable for room based localization. In the future, the system s accuracy could be enhanced by using a probabilistic approach in fingerprinting instead of the deterministic approach that is currently being used. Instead of the currently used Euclidean distance algorithm, a more accurate algorithm such as taking the k-nearest neighbors into account may be used. Deploying more wireless access points and distributing them over the area to be localized will enhance the accuracy as well. Taking the orientation of the user into account, by using an electronic compass will enable the application to provide orientation sensitive information which is important in case of deployment in a museum. For the services, more interaction could be built, which should be easy to extend because the current services frontend is based on HTML instead of being hard coded into the application. Providing the user with audio and video based services for museum deployments, or providing the user with the schedule of the current room that he is standing in would be an interesting development. The current prototype is implemented on a laptop. For a truly mobile application, this project could be implemented using JME, so that it work on any Java enabled mobile device.

49 Appendix A Screenshots A.1 Client interface Figure A.1: Location in progress 1

50 APPENDIX A. SCREENSHOTS Figure A.: Services associated with the room Figure A.: Location of the room on the map

51 APPENDIX A. SCREENSHOTS A. Fingerprinter interface Figure A.: Adding/Removing buildings

52 APPENDIX A. SCREENSHOTS Figure A.: Adding/Removing rooms Figure A.: Adding/Removing services

53 APPENDIX A. SCREENSHOTS Figure A.7: Setting location of a room on the map Figure A.8: Fingerprinting of rooms