Location information and Handover optimization in WLAN/WPAN

Transcription

1 Location information and Handover optimization in WLAN/WPAN Group : 998 Supervisors : Hans Peter Schwefel Istvan Kovacs Group member : Sukesh Reddy Kim Lam Yann Malidor Christophe Martineau Guillaume Monghal Krishna Mohan 9th Semester of Mobile Communication Aalborg University, autumn 2004

2 AALBORG UNIVERSITY - Institute of Electronic Systems Fredrik Bajer Vej Aalborg Mobile communication: 9th semester Location information and Handover optimization in WLAN/WPAN Project Period : 2nd September - 4th January 2005 Group : 998 Supervisors : Hans Peter Schwefel Istvan Kovacs Group member : Kim Lam Yann Malidor Christophe Martineau Krishna Mohan Guillaume Monghal Sukesh Reddy Number of report :7 Number of pages :96 Abstract This project aims to give concepts and provide a simulation of an enhanced handover solution for Bluetooth networks. Handover between Access Points enables the user to maintain a continuous connection while moving from the coverage area of one Access Point to another. Additionally, in this project, location data collection and movement prediction techniques are adequately used to enhance the handover process.

3 Contents 1 Introduction 9 2 Background Bluetooth A recent booming technology Bluetooth technology Connection scheme Data Communication States of Bluetooth Devices Bluetooth packet Location techniques Cell Identification Angle of Arrival Triangulation methods Database Support Discussion on the different location techniques Movement Prediction Utility of movement prediction The predictive techniques Different methods for location prediction Conclusions Theoretical analysis Probability Issues in Location Definitions Algorithm of the Localization Summary of the Localization algorithm Propagation aspects Theoretical Propagation Propagation Movement Prediction Human movement model Parameter Estimation Position Prediction Estimation of the Distance from the Access Point

4 CONTENTS 3.4 Handover Establishing connection in Bluetooth network Inquiry Procedure Paging Procedure The Paging timers Conclusions Simulations Scenario Simulator Principle Functionalities Breakdown Generation of RSSI measurements Location of a fixed Device Location of a moving Device Movement Prediction Handover Conclusions Conclusions 95 Appendix 97 A Detailed calculation of the Distance estimation 98 A.1 Space overview: A.2 Calculation of p fθ,s,d t (d t+ t ): B Paging procedure details 104 C Abbreviation 106 Bibliography 107 4

5 List of Figures 2.1 Connection scheme of Bluetooth Devices Packet exchange between Master-Slave Different states of a Bluetooth Device Intervals in Hold mode Communication slot Packet description Cell Identification location method Angle of Arrival location method Signal strengh (triangulation method) Uplink Time (Difference) of Arrival Location Method Downlink Observe Difference Location Method Database Correlation Location Pattern Matching Principle of movement prediction Predictive techniques Accuracy of the prediction Possible links between APs within a Full Meshed Networks Possible links between APs within an Arbitrary Network Example of probabilities Location Criterion Direction Criterion Circles of probabilities Localization algorithm Free space propagation (distances unit is dm) Curve using our propagation model (distances unit is dm) Construction of the mean curve Description of the stochastic process of the speed PDF of the speed of all the moving bodies Example of PDF of the speed of one precise moving body PDF of the direction of all the moving bodies PDF of the direction for one precise moving body Packet Exchange in the Paging Process Possible configuration of the Room Simulation

6 LIST OF FIGURES 4.3 Summary of all the simulations performed RSSI generation simulator element Delay generation element Trajectory element Block of the RSSI measurement generation along a trajectory Representation of the RSSI measurement on a time axis per AP Room for the first case Room for the second case Room for the third case Room for the forth case Simulator for the localization of a fixed Device PDF of the position estimation in the first case PDF of the position estimation in the second case PDF of the position estimation in the third case Precision of the maximum likelihood estimation Comparison for case Comparison for case Comparison for case Comparison for case Room, trajectory and APs Method for choosing the measurements to localize Updating time scheme One of the possible configurations of the room (view from the top) Possible area transitions performed by the Device in the case of Handover without location Time for Paging in Handover without Location Time for Paging without Location (1) Time for Paging without Location (2) Loss of quality of the signal for Handover without Location Loss of the signal for Handover without Location Example of coverage areas of the Three Access Points Sectorization of the room Principle of Handover with Movement Prediction A.1 Space Overview A.2 Meaning of δp A.3 Calculation of the different angles A.4 Angular area A.5 Example of a distance estimation

7 List of Tables 2.1 Possible use of the locations techniques in an Outdoor with LOS case Possible use of the locations techniques in an Outdoor without LOS case Possible use of the locations techniques in an Indoor with LOS case Possible use of the locations techniques in an Indoor without LOS Advantages and drawbacks of location techniques Summary of the different steps of our Location process Summing-up of the different cases Summing up of the results obtained Table of results Localization with former informations about the location of the Device Results for the case Results for the case Results for the case Results for the case Results of the simulation Results for a moving Device Total time for Paging procedure with and without Self Configurability Comparison between Handover With and Without location information

8 Acknowledgment Our very special thanks are dedicated to our supervisors Hans Peter Schwefel and Istvan Kovacs who provided us useful guidance in formulating the problem, choosing and applying the theories and constructing the structure of the project. We also thank Joao Figueiras for his help at the end of the project. The group members would like to thank each other for good teamwork, devotion and efforts in carrying out this project. We are thankful for this experience, completely contented with our results and we are proud to have managed this project so well. 8

9 Chapter 1 Introduction Nowadays, Bluetooth is subject to considerable developments due to the flexibility of its interface. Moreover, its applications in the currently booming Personal Area Networks (PAN), which enable the equipments of a sole user to be linked, leads Bluetooth to a series of advances. Nonetheless, one of the drawbacks of this technology is the limitation of the mobility of the Bluetooth equipments due to the absence of the handover concept in the Bluetooth Specification. In fact, since Bluetooth has small power requirements, the equipments can be physically small and by this way easily movable. Meanwhile, Bluetooth is a short-range technology: the coverage of the cell is reduced. These two factors lead to the need of a handover system. The main aim of this project is to develop and evaluate a mechanism to obtain handover in a Bluetooth based network. Moreover, it seems that the accurate location of a moving body generally enables to improve resource-saving such as transmitted power, bandwidth and Quality of Services. Consequently, this project will handle the study of a handover based on location information as precise as possible. Finally, if knowing the coordinates of a mobile equipment, tracking it and analyzing its past movements is relevant to improve handover, then why not enhance more location information by predicting the future movements of the body? The 2 nd chapter of this report states the project backgrounds including Bluetooth specifications, location techniques and movement prediction techniques. A theoretical analysis is presented in the 3 rd chapter. Location, movement prediction, propagation model and handover are defined according to a precise scenario. Finally, the 4 th chapter focuses on the simulator implementation and comments on the obtained results. 9

10 CHAPTER 1. INTRODUCTION Motivations As we saw, Handover could be a relevant progress in the Bluetooth technology. In addition, our team is interested in dealing with Bluetooth because it has a significant role in the booming short-range data transfer market, interconnecting all devices of the personal sphere, such as Mobile Phone, PDA, PC, etc. Besides, we are deeply concerned in this project since it allows us to discover an original networking architecture. 10

11 Chapter 2 Background This chapter will give a general description of the specificities and capabilities of the Bluetooth technology regarding interconnections of Devices, communication types and states of Devices. Then, an enumeration and comparison of the available locations techniques will be done in the purpose of using one of them in our project. The last part of this chapter will explore the movement prediction possibilities with the aim of understanding the solution which is chosen in this project. 2.1 Bluetooth Bluetooth is a short range wireless technology and a worldwide open standard which permits Personal Area Networks to be set up instantly among different Devices. It is a low cost, low power technology, originally developed as a cable replacement to connect Devices with each other such as mobile phones, headsets, PDA s, etc. for both voice and data communication with security functionalities A recent booming technology Created in 1994 by the Swedish company Ericsson, this technology was named as Bluetooth in 1998 after the foundation of the Special Interest Group (SIG). At the beginning this industrial organization was composed of Ericsson, IBM, Intel, Nokia and Toshiba. The SIG purpose is to define both Bluetooth specifications and certifications (to verify the compatibility and inter-operability of the products between them). In 2001 appeared the first consumer products for mass market and at the same time specification 1.1 was released. In 2004 the SIG has more than 2500 members. Moreover, the group has launched Bluetooth specification 2.0 +Enhance Data Rate in november Bluetooth technology Bluetooth corresponds to a radio interface between two mobile equipments or between equipments and a transmitter/receiver. The purpose of this interface is to 11

12 CHAPTER 2. BACKGROUND make a network allowing the interconnection of different types of Devices. Bluetooth operates in the unlicensed Industrial-Scientific-Medical (ISM) band at 2.4 GHz ( to ) which is also used by other technologies such as WLAN. Bluetooth devices can operate within two different networking frameworks: - The infrastructure mode, in which devices communicate with each other by first going through an Access Point. - The ad-hoc mode, in which devices or stations communicate directly with each other, without using an Access Point. An Access Point (AP) is a hardware device or a computer s software that acts as a communication hub for users of a wireless device to connect to a wired LAN (WLAN) Connection scheme Bluetooth system can manage a number of low-cost point-to-point (only two Bluetooth units involved) or point-to-multipoint links up to a distance of 100 m (the distance depends on the transmitted power of the Device, which is between 1mW and 100mW).[8] Several connection schemes have been defined in the Bluetooth specification. One of them is called piconet, which can contain up to eight active Devices: one master and its seven active slaves. Any Device can be master or slave. The master is the one that initiates the communication link and the other units are slaves. A master and its slaves belong to the same piconet. Once a piconet has been established, master-slave roles can be exchanged. There is no direct transmission between slaves in a Bluetooth piconet. The Device can only transmit and receive data in one piconet at a time. Furthermore, two or more piconets can be interconnected, forming what is called a scatternet. A Bluetooth unit can simultaneously be a slave member of multiple piconets, but only master in one Data Communication Communication in a piconet is organized so that the master polls each slave according to a polling scheme. A slave is only allowed to transmit after having been polled by the master. The slave will start its transmission in the slave-to-master timeslot after it has received a packet from the master. The master may or may not include data in the packet used to poll a slave Packets on the Physical Links Between master and slave(s), two link types have been defined: Synchronous Connection-Oriented (SCO) 12

13 CHAPTER 2. BACKGROUND Piconet 11 Piconet 9 Piconet 12 Piconet 1 Piconet 10 Piconet 2 Piconet 4 Piconet 8 Piconet 7 Piconet 3 Piconet 5 Bluetooth unit, master Bluetooth unit, slave Bluetooth unit - master in one piconet, slave in another Bluetooth unit - slave in two piconets Bluetooth unit - master in one piconet, slave in two Bluetooth unit - slave in three piconets Piconet 6 Figure 2.1: Connection scheme of Bluetooth Devices [8] Asynchronous Connection-Less (ACL) The SCO link is used for voice transmission. This application is used in real-time two-way communication. A point-to-point link is established between a master and only one slave, and specific time slots at regular intervals are used. The latency time is reduced as much as possible. In this mode, packets are never re-transmitted. The maximum throughput is 64 Kb/s full-duplex. The ACL link is used for non real-time transmission where data integrity is important. The packets are retransmitted until there are no more errors at the reception or if an upper time limit is reached. That is why automatic repeat request is used in this mode. Asynchronous connection can support symmetrical or asymmetrical, packet-switching, point-to-multipoint connections. In asymmetric connection, the maximum bit rate is 723.2Kb/s in one way and 57.6Kb/s in the other way. In symmetrical connection, it is 433.9Kb/s in both ways Time Division Duplexing More precisely, the Bluetooth system provides duplex transmission based on slotted Time-Division Duplex (TDD), where the duration of each slot is 625 µs. The division by slot enables each member of the piconet to participate because TDD 13

14 CHAPTER 2. BACKGROUND uses the same channel and continuously alternates between sending and receiving. SCO ACL SCO ACL SCO Master Slave1 Slave2 Figure 2.2: Packet exchange between Master-Slave Frequency Hop Spread Spectrum Bluetooth uses Frequency Hop Spread Spectrum (FHSS) as an interference avoidance technique. The binary data in the baseband level of Bluetooth is modulated by using Gaussian Frequency Shift Keying (GSFK). Then, they are transmitted using a carrier determined by a frequency synthesizer. Instead of producing only a single carrier frequency, the synthesizer is controlled by a hop code generator that causes it to change carrier frequency at a nominal rate of 1,600 hops per second. One Bluetooth data packet is sent per hop. A Device uses one frequency in one timeslot. Then, by a frequency hop, it will change of frequency in the next timeslot and so on. Thus, for two Devices to communicate using FHSS, they must be properly synchronized in order to hop together from channel to channel. This means that the Devices must: Use the same channel set Use the same hopping sequence within that channel set Be time-synchronized within the hopping sequence Ensure that one transmits while the other receives, and vice versa (TDD principle) All of these synchronization parameters are determined by the piconet master. The master passes the FHSS synchronization parameters to a slave during the Page process. When an external Device wants to enter the piconet, it has to acknowledge this continuation of frequency hoping to be able to follow it States of Bluetooth Devices Figure 2.3 shows all possible states of a Bluetooth Device. There are two main states in the Bluetooth link controller: standby and connected. 14

15 CHAPTER 2. BACKGROUND - The standby state is the default state in the Bluetooth unit. In this state, the Bluetooth unit is in a low-power mode where the energy consumption of the Device is highly reduced. - The connected state means that the Device participates in a piconet. Stanby Page scan Page Inquiry Inquiry scan Slave Response Master Response Inquiry response Connect Active Hold Sniff Park The others sub-states are: Figure 2.3: Different states of a Bluetooth Device - Inquiry: The master will search which units are in range, and what their Device addresses and clocks are to initialize the communication. This request will be repeated as long as a unit has not been found. - Inquiry scan: used by a slave to listen to an Inquiry. - Inquiry response: the state of the Device switches from the Inquiry scan substate to the Inquiry response substate when it answers to the master Inquiry by sending its address and its clock state. After receiving the Inquiry response, a connection is established for the Paging procedure. A more detailed part about Inquiry and Paging is done in Section Paging: used by a master to establish a piconet with a particular slave whose Bluetooth Device address is known. - Page scan: used by a slave to listen to its page. - Slave response: state of the Device after receiving the message from the master for a connection. Then the slave will send its Access Code to the master (explained in detail in ). 15

16 CHAPTER 2. BACKGROUND - Master response: after the reception of the slave response, the master will send a packet called Frequency Hopping Synchronization (FHS) which will permit the slave to be synchronized with the master clock. - Connected: the connection has been established and packets can be sent back and forth. The channel (master) Access Code and the master Bluetooth clock are used to determine the sequence of Frequency Hopping used in this piconet. - Active: the Bluetooth unit actively participates on the channel. The master schedules the transmission based on traffic demands to and from the different slaves. Regular transmissions are made by the master to keep the slaves synchronized to the channel. Once connected, the unit is able to transmit and receive data. To save battery power, three low power modes are available: Sniff, Hold, and Park (in decreasing order of power efficiency). These modes are useful for: - enabling more than seven slaves to be in a piconet - giving the master time to bring other slaves into its piconets - conserving energy The main goal of these modes is to reduce the time for a Device receiver to remain on. It allows the Devices to adjust the power depending on the range of communication. The lower power level covers a distance of about 10 meters, while the higher power level can cover about 100 meters. [8] In this part, only Hold mode will be treated as it is the one that interests us in the Handover part. The Hold mode is a one-time exit from the obligation of a piconet and it can be used when no data needs to be transmitted for long time intervals (up to 41s without re-synchronization)[4]. An internal timer determines when the unit will be reactivated. In this mode, a slave does not receive any asynchronous packets (ACL packets are suspended) and only listens to determine if it should become active again. It does not affect SCO traffic. In the Hold mode ( hold timeout ) the slave can do other things like scanning, paging, inquiring, or attending another piconet. During this mode the Device is still considered an active member of the piconet. Thus, it remains synchronized with the master (Figure 2.4). Hold mode cannot begin until 6 T poll intervals after the hold request packet has been sent (9 T poll if the parameters must be negotiated). T poll is a poll interval that is negotiated between the master and the slave. T poll =40slots [30]. 16

17 CHAPTER 2. BACKGROUND 6*T(poll) or 9*T(poll) Hold time Figure 2.4: Intervals in Hold mode Bluetooth packet Bluetooth Baseband packet format Information is exchanged through packets which are transmitted on a different hop frequency (usually sent packet by packet). Moreover, it is possible to send packets that cover either three or five slots long (Figure 2.5). 625 us F(k) F(k+1) F(k+2) F(k+3) F(k+4) F(k+5) F(k) F(k+3) F(k+4) F(k+5) F(k) F(k+5) Figure 2.5: Size of the Communication slot[4] A baseband packet is composed of three parts: the Access Code, the Header and the Payload. Packets can be constructed either: Access code only Access code and Header Access code, header and Payload Access Code In almost every wireless packet communication system, the packet itself begins with a special pattern of bits: the Access Code. It provides bit and word synchronization. In general the Access Code: 17

18 CHAPTER 2. BACKGROUND 68 (72) bits access code packet header payload 4 bits (4) preambule sync. (trailer) AM adress type flow ARON SECN HEC bits Figure 2.6: Packet description Can be used by a slave to resynchronize its clock to the clock of the piconet Provides bit and word synchronization Includes basic piconet identification information The Access Code is derived from the 24 first least significant bits of the BD_ADDR. Thus, different Access Code are required depending on the context: Channel Access Code (CAC) Device Access Code (DAC) General Inquiry Access Code (GIAC) Dedicated Inquiry Access Code (DIAC) Particularly the DAC is used by the master for Paging a specific Bluetooth Device for entry into its piconet. The master knows the paged Device s BD_ADRR via an Inquiry process and can assemble the correct DAC from this address Using the Access Code in Short Hopping Sequences During the Inquiry and Paging processes, a prospective master tries to find (inquire) or connect (page) with a prospective slave. The time for a successful Inquiry or Page can be reduced significantly if the usual 79-channel frequencies are reduced to 32-channel frequencies. This is possible because the Access Code used for Inquiry and Paging transmissions meets the Federal Communications Commission (FCC) rules for a hybrid Spread Spectrum system. This 32-channel frequencies can be further divided into two parts.[8] 18

19 CHAPTER 2. BACKGROUND 2.2 Location techniques This section sums up different location techniques used in wireless technologies. They can be divided into different groups according to the used technology and the accuracy: Cell-Identification, Angle of Arrival, Triangulation, Time Difference of Arrival and Database Correlation. Discussion of their advantages and drawbacks in different environments and uses will be made at the end of this section Cell Identification This technique just returns that the Device is in the coverage area of the AP which it is bound to, as presented in Figure 2.7. The accuracy depends directly from the coverage area of the AP. Thus, this is not an accurate mean of location. However, no calculation is needed. So it is easily implementable. Device x,y Base Station Figure 2.7: Cell Identification location method Angle of Arrival Two APs measure the arrival angle of the signal which is transmitted by a Device. The intersection area of the two lines determines the position of the Device.This is shown in Figure 2.8. The main drawback of this technique is that a line of sight is required. Furthermore, directional antennas are needed. Possible location of the device α 1 α 2 Beacon 1 Beacon 2 Figure 2.8: Angle of Arrival location method 19

20 CHAPTER 2. BACKGROUND Triangulation methods The following methods use triangulation methods (need at least 3 APs) Signal strength The Device measures the strength of the AP signal and sends it back to the AP: indeed, in Bluetooth frame there is a possibility to get the Ratio Signal Strength Indicator (RSSI) calculated by the Device. Thus, because the power received is proportional to the distance between two Devices, when three APs get this information, three circles can be obtained and the intersection of them defines the probable location of the Device.This is presented in Figure 2.9. The main drawbacks are the need of at least three APs and preferably an environment with LoS, otherwise a variation of db can appear in the measures. d 3 Confidence Ellipse d 2 d 1 Figure 2.9: Signal strengh (triangulation method) Uplink Time of Arrival ToA (Time of Arrival): APs measure the time for the signal to arrive from the Device. Because this measurement is directly related to the distance between the two stations, triangulation method can be used. Hyperbolas are obtained and their intersections give the location of the Device, as illustrated in Figure TDoA (Timing Differences technique): If stations are not synchronised, TDoA is used to determine the (relative) time of arrival between the 2 stations. The main drawback is that this method needs 3 APs Downlink Observed Difference Downlink Observed Difference is a Timing Differences technique. Measurements are made by the Device, which measures the time difference of the signals from 20

21 CHAPTER 2. BACKGROUND Difference 1-3 Difference 2-3 BTS1 Clock Time 1 Clock Time 2 Clock Time 3 BTS3 BTS1 Difference 2-3 Figure 2.10: Uplink Time (Difference) of Arrival Location Method several APs. Synchronization is needed between the Device and the APs. The accuracy also depends on LoS, multipath, etc. (Figure 2.11). Difference 1-3 Difference 2-3 BTS1 Clock Time 1 Clock Time 2 Clock Time 3 BTS3 BTS1 Difference 2-3 Figure 2.11: Downlink Observe Difference Location Method Database Support The two following techniques use information on location stored in a database Database Correlation Information samples, called fingerprints, are taken from the areas covered by the APs: it can be signal strength or time delay. When the Device measures one of these parameters, it sends the measurements to a database server which compares this data with the data stored in a database (Figure 2.12). 21

22 CHAPTER 2. BACKGROUND BTS Application servers Received signal fingerprint Location estimate Location estimate Mobile terminal Location server - system information - Calibration data - Digital maps GSM/GPRS/UMTS Network + Internet Location estimation by using signal fingerprint and database Figure 2.12: Database Correlation Location Pattern Matching This technique compares the results obtained by measurements taken from a Device with some pre-trained sequences, simulated with some algorithms, on a server. These two techniques cannot be used in dynamic networks.(figure 2.13) 2 Carriers antennas receive signal and forward it to the carriers switch 3 Sophisticated gear analyses the acoustic characteristics of the signal, compares it to previously acquired patterns, and determines the callers location 2 4 Switch forwards voice call and location data to a server 1 handset or vehicle 2 Figure 2.13: Location Pattern Matching Discussion on the different location techniques For each case, different number of Devices and Access Point have been chosen. Indeed some techniques cannot be used if the number of Devices increases or if the number of AP is not sufficient. 22

23 CHAPTER 2. BACKGROUND 1. Outdoor with LOS The table 2.1 shows the cases where the different techniques can be used depending on the number of Access Points, Devices and on two factors of usage: outdoor and with line of sight (LoS). 1 Access Point 2 Access Points 3 Access Points 1 Device - Cell-ID - Cell-ID - Cell-ID - Database Correlation - Database Correlation - Database Correlation - Location Pattern Matching - Location Pattern Matching - Location Pattern Matching - AoA - AoA -ToA - RSSI 2 Devices - Cell-ID - Cell-ID - Cell-ID - Database Correlation - Database Correlation - Database Correlation - Location Pattern Matching - Location Pattern Matching - Location Pattern Matching - AoA - AoA -ToA - RSSI 7 Devices - Cell-ID - Cell-ID - Cell-ID - Database Correlation - Database Correlation - Database Correlation - Location Pattern Matching - Location Pattern Matching - Location Pattern Matching - AoA - AoA - ToA UL (but problems) - ToA DL - RSSI Table 2.1: Possible use of the locations techniques in an Outdoor with LOS case 2. Outdoor without LOS The table 2.2 shows the cases where the different techniques can be used depending on the number of Access Points, Devices and on two factors of usage: outdoor and without line of sight (NoLoS). 3. Indoor with LOS The table 2.3 shows the cases where the different techniques can be used depending on the number of Access Points, Devices and on two factors of usage: indoor and with line of sight (LoS). 4. Indoor without LOS The table 2.4 shows the cases where the different techniques can be used depending on the number of Access Points, Devices and on two factors of usage: indoor and without line of sight (NoLoS): 5. Advantages and drawbacks of location techniques (Table 2.5) 23

24 CHAPTER 2. BACKGROUND 1 Access Point 2 Access Points 3 Access Points 1 Device - Cell-ID - Cell-ID - Cell-ID - Database Correlation - Database Correlation - Database Correlation - Location Pattern Matching - Location Pattern Matching - Location Pattern Matching -ToA - RSSI (but with less accuracy) 2 Devices - Cell-ID - Cell-ID - Cell-ID - Database Correlation - Database Correlation - Database Correlation - Location Pattern Matching - Location Pattern Matching - Location Pattern Matching - ToA (but with less accuracy) - RSSI 7 Devices - Cell-ID - Cell-ID - Cell-ID - Database Correlation - Database Correlation - Database Correlation - Location Pattern Matching - Location Pattern Matching - Location Pattern Matching - ToA UL (but problems) - ToA DL - RSSI (but with less accuracy) Table 2.2: Possible use of the locations techniques in an Outdoor without LOS case 1 Access Point 2 Access Points 3 Access Points 1 Device - Cell-ID - Cell-ID - Cell-ID - Database Correlation - Database Correlation - Database Correlation - Location Pattern Matching - Location Pattern Matching - Location Pattern Matching - AoA - AoA -ToA - RSSI 2 Devices - Cell-ID - Cell-ID - Cell-ID - Database Correlation - Database Correlation - Database Correlation - Location Pattern Matching - Location Pattern Matching - Location Pattern Matching - AoA - AoA -ToA - RSSI 7 Devices - Cell-ID - Cell-ID - Cell-ID - Database Correlation - Database Correlation - Database Correlation - Location Pattern Matching - Location Pattern Matching - Location Pattern Matching - GPS - AoA - AoA - ToA UL (but problems) - ToA DL - RSSI Table 2.3: Possible use of the locations techniques in an Indoor with LOS case 24

25 CHAPTER 2. BACKGROUND 1 Access Point 2 Access Points 3 Access Points 1 Device - Cell-ID - Cell-ID - Cell-ID - Database Correlation - Database Correlation - Database Correlation - Location Pattern Matching - Location Pattern Matching - Location Pattern Matching - ToA UL (but problems) - ToA DL - RSSI (but with less accuracy) 2 Devices - Cell-ID - Cell-ID - Cell-ID - Database Correlation - Database Correlation - Database Correlation - Location Pattern Matching - Location Pattern Matching - Location Pattern Matching - ToA UL (but problems) - ToA DL - RSSI (but with less accuracy) 7 Devices - Cell-ID - Cell-ID - Cell-ID - Database Correlation - Database Correlation - Database Correlation - Location Pattern Matching - Location Pattern Matching - Location Pattern Matching - ToA UL (but problems) - ToA DL - RSSI (but with less accuracy) Table 2.4: Possible use of the locations techniques in an Indoor without LOS 25

26 CHAPTER 2. BACKGROUND Techniques Cell Identification Angle of Arrival Signal Strength Uplink Time (Difference) of Arrival +Low-cost +Only 2 APs +Low-cost +Relative accuracy Downlink Observed Difference +Relative accuracy Database Correlation +Good accuracy in obstructed areas +Easy to implement Advantages +No computation needed +Good accuracy under good propagation conditions +Easy to implement because no add-software +Good accuracy under good propagation conditions +Low data transfer to get location information +No delays -Low accuracy -LoS needed -High number of APs -Narrowing of the angle Drawbacks -Large number of users -Expensive (directional antennas) -Accuracy depends on the propagation channel +No changes on Devices -High number of APs -Large number of users -No location while Device is idle +Possible when Device is idle +Mobile-based implementation network possible -High number of APs -Software changes in Device -LoS needed if good accuracy wanted -Creation and maintenance -Search in realtime -Updates on dynamic networks Table 2.5: Advantages and drawbacks of location techniques Location Pattern Matching +Optimal accuracy in obstructed areas +Possible with one AP -Updates 26

27 CHAPTER 2. BACKGROUND Summary Pros and cons of each technique have been studied. They are summarized below: Cell Identification: this technique is not enough accurate to enable a good implementation of handover Angle of Arrival: the main drawback of this technique is the need of Line of Sight which is not possible in real conditions Database support: this technique needs too much maintenance and updates to be used with Bluetooth technology Triangulation methods: this technique seems to be the more appropriable one. Among the three location methods, the Signal Strength technique appears to be more easy-implementable, adaptable and accurate in most cases than the two others (Uplink time of Arrival and Downlink Observed Difference). Moreover, RSSI measurements can be directly obtained from Bluetooth packets Thus, Signal Strength seems to be a good choice for location in Bluetooth. Besides, it is a well-documented technique [5] which has been already implemented to offer new ways of communicating by enabling end-users to obtain corporate information wherever they are. 27

28 CHAPTER 2. BACKGROUND 2.3 Movement Prediction Utility of movement prediction Movement prediction allows to predict the future location of the users with a certain accuracy. Then it will be easier to anticipate handover situation and consequently save time and/or power resources. By tracking the Device, it is surely possible to Access Point 3 T+1? Predicted point Access Point 1 T-2 + T-3 T T + T-1 + Access Point 2 Periodical positions of the device Figure 2.14: Principle of movement prediction predict in which cell it is going to move and warn only the concerned Access Point. Moreover, according to its velocity, it is possible to predict when it will enter the cell (Figure 2.14). Thus, the handover procedure which is quite long could be timereduced. It is assumed that: Users move with a certain degree of regular (random) movements. Users carry with themselves movement history from both past and present. It is possible to gather and use this movement history (recent and past) to predict future movement The predictive techniques To determine the movement of a user, the techniques of location prediction are mainly based on a mobility model (Figure 2.15). 28

29 CHAPTER 2. BACKGROUND mobility model random model Arbitrary movements without constraints deterministic model Predefined movement path or real mobility trace hybrid model Movement bounded by environmental constraints Figure 2.15: Predictive techniques Most simulations are based on random mobility models, but these ones are insufficient to reflect the environmental constraints. Moreover deterministic models are too complex and real user traces are hard to obtain. Thus, hybrid mobility models combine both of the two and make tradeoffs between simplicity and reality. In our case, we will use the hybrid model [35]. There are three main techniques for movement prediction: storage of the historical movement pattern in a database and comparison between the recent states and the movement tracks comparison with a table of possible locations and the probability that the user is located in there in a determined period of time Different methods for location prediction Different methods for location prediction are based on historical data. A method to track the Device in a cell is to periodically locate it in the cell (Figure 2.14). Then, with at least one precedent position it is possible to predict where the next point could be. Obviously, the more previous positions taken into account, the better the prediction. Therefore this information has to be collected over a sufficiently long period of time to reflect user behavior (Figure 2.16) Movement prediction using Location criterion and Direction criterion With a Full Meshed Networks In a full meshed network, each Access Point is directly linked with all the neighbors, even if the handover between two of them is not possible (because of walls, machines,etc.) (Figure 2.17). In this case the link is useless. With an Arbitrary Network In this example (Figure 2.18), the Device cannot move from AP1 to AP3 or to 29

30 CHAPTER 2. BACKGROUND + T-3 + T-2 + T-1 T + By taking into account at least two previous positions it is possible to know that the device is turning. By taking into account only the positions T and T-1 you cannot. + T+1 + T+1 Figure 2.16: Accuracy of the prediction AP AP AP AP AP AP AP AP AP AP AP AP N = 2 N = 12 N = 30 N: Number of possible handovers AP Possible moves Figure 2.17: Possible links between APs within a Full Meshed Networks 30

31 CHAPTER 2. BACKGROUND AP4 because of the wall. These links are consequently useless. AP 1 AP 2 AP 4 AP 3 Figure 2.18: Possible links between APs within an Arbitrary Network Moreover, the probability that the Device goes from AP2 to AP4 (and conversely) is less important than the probability it moves from AP2 to AP3 or from AP3 to AP4 (and conversely). These factors could be taken into account to optimize the handover process. Example with the previous case: It is now possible to develop criterions (Figure 2.20 and Figure 2.21). p=1 p=0.3 AP 1 AP 2 p=0.2 p=0 p=0.5 p=0 p=0 AP 4 p=0.7 p=0.3 p=0.4 p=0 p=0.6 AP 3 Figure 2.19: Example of probabilities that a Handover could occur using an Arbitrary Network In the location criterion, probabilities are calculated without using the precedent location, whereas in the direction criterion the previous location is used to establish the probabilities of the most probable future locations. 31

32 CHAPTER 2. BACKGROUND AP2 30% 50% AP1 AP3 Direction of travel AP4 AP2 15% 80% AP1 AP3 Present Location 20% AP4 5% AP4 Figure 2.20: Location Criterion[15] Figure 2.21: Direction Criterion[15] 2.4 Conclusions For the purpose of our project, a Bluetooth Device must be located and forecast in its motion in the most accurate way. Within this context, we have been through the possible techniques and have chosen the most appropriate one according to some assumed operational constraints. Thus, we have chosen the triangulation method as the location technique, and particularly Signal Strength based on RSSI measurements. As motion prediction, the hybrid one has been chosen. Our method for location and movement prediction will be inspired on historical data. The following chapters will now deal with the analysis and the implementation of those methods into details. 32

33 Chapter 3 Theoretical analysis This chapter gives a detailed explanation right from the location process of the Device to the Handover procedure encountered by the mobile Device. Using the concepts of RSSI values and probabilities, location of the mobile Device is estimated. Future movement of the Device is predicted using this location information properly. Finally this will help the handover procedure to be implemented. 3.1 Probability Issues in Location Definitions Triangulation and Probabilities Our choice of location technique is the triangulation using Ratio Signal Strength Indicator (RSSI) measurements. As seen, it is the most adapted to Bluetooth. The principle of the triangulation with power measurements looks very simple. A moving Device is receiving a signal from three APs. Power of the signal is evaluated and sent back to the APs. From this power W, each AP deduces the distance d of the device based on a propagation model. By drawing circles of radius d, itis possible to find the position M(x, y) of a device at the intersection point of the circles. W 1,2,3 d 1,d 2,d 3 M(x, y) (3.1) Practically it is not simple. Indeed, each power measurement will not correspond to only one distance but to a probability density function (PDF) of the distance: f D (d RSSI = W ). In fact by measurements at a fixed distance, it is noticed that the power measurement is not constant [7]: it moves around a center value. Then, with three AP, the PDF of the location f X,Y (x, y RSSI = W ) is deduced with (X,Y) as the coordinates of the Device. W f D (d RSSI = W ) f X,Y (x, y RSSI = W )) (3.2) The following sections focuses on the parameters and the functions that have to be defined to solve the problem properly and rigorously. 33

34 CHAPTER 3. THEORETICAL ANALYSIS Definition of The Parameters D: this is the random variable of the distance between the considered AP and the moving Device. Its unit is the meter. Ω(D): this is the set of all the possible distances that can be observed. If the APs are in a 20-meter large and 20-meter long square room, the distance is limited by the walls. In the project, the used distances are equal to the length of the projection on the floor. So in this case: Ω(D) =[0,d Max ] and d Max =20 2 RSSI: this is the random variable of RSSI measurements. Ω(RSSI): this is the set of all the possible RSSI measurements that can be observed. The Device has a lower limit under which it is not possible to collect any value. It also has an upper limit. Here Ω(RSSI) =[ 85, 0[ in db. t: this variable represents the elapsed time during the movement of the Device. It is a relevant factor because the aim is to track a moving Device. It means that different probabilities have to be considered at different times. t can be put as an index on each random variable Probability Density Functions In this part, all the PDF that will be used in the further steps of this work are defined: Location, Movement Prediction and Handover. Also, their relationship with the problem and the meaning of each variable will be given. The order in which these PDFs are presented is the order that has to be used to solve the problem. At a fixed distance d f RSSI (W D = d) (3.3) This is the PDF of the RSSI measurement given that the distance is equal to d. It is the only PDF that can be built by measurements. Indeed, by making thousands of measurements at the distance d, a distribution of RSSI measurements at this distance is got. Then this distribution is put into a PDF. This has been done in [7]. Yet the parameter t is not taken into account in the measurement. Indeed by measurement, a function that is correct at any time is built. So: At fixed RSSI measurement f RSSIt (W t D t = d t )=f RSSI (W D = d) (3.4) f D (d RSSI = W ) (3.5) This is the PDF of the distance knowing the RSSI measurement. This PDF will directly be used to locate the Device. In fact, each time an AP get a RSSI measurement, this function expresses the probability of the distance between the AP and 34

35 CHAPTER 3. THEORETICAL ANALYSIS the Device. However, it is not possible to build it because it is impossible to fix the parameter RSSI and observe the distance. So this function has to be calculated through other parameters using especially f RSSI (W D = d). PDF of both RSSI measurement and distance f D,RSSI (d, W ) (3.6) This PDF has two variables. In practice one of the two parameters has to be fixed to observe the other one. PDF of the distance f D (d) (3.7) This is the PDF of the distance considering all the possible RSSI measurements. According to the theory of probabilities [33], its definition is: f D (d) = f D,RSSI (d, W )dw (3.8) PDF of the RSSI measurement Ω(RSSI) f RSSI (W ) (3.9) This is the PDF of the RSSI measurement that can be obtained considering all the possible distances. From the point of view of one AP, it corresponds to the probability distribution of the RSSI measurements that can be obtained. The definition is : f RSSI (W )= f D,RSSI (d, W )dd (3.10) Ω(D) Algorithm of the Localization The localization is performed in two steps. First, it consists in deducing the PDF of the distance between the Device and the AP for each AP. Then Triangulation must be performed with these PDF of the distance. The result is a PDF of the location PDF of the distance between the Device and the AP As presented previously, only f D (d RSSI = W ) is interesting for the location. Indeed, what is got from the AP is the RSSI measurement which is used to know the probability of the distance. Then with three Access Points it is possible to perform triangulation. Now the question is: how can f D (d RSSI = W ) be computed? The theory of the probabilities states that for a problem with two random variables, the formula is: f D,RSSI (d, W )=f D (d RSSI = W ) f RSSI (W ) (3.11) 35

36 CHAPTER 3. THEORETICAL ANALYSIS Thereafter: So it is deduced: f D,RSSI (d, W )=f RSSI (W, D = d) f D (d) (3.12) f D (d RSSI = W )= f RSSI(W, D = d) f D (d) f RSSI (W ) (3.13) In this equation, f RSSI (W D = d) is already known because it was determined with the initial measurements. The question now is: what about f RSSI and f D?Is it necessarily to know everything about them? First, it is noticed that f D (d RSSI = W ) is a PDF of the distance d. So because it is a PDF it is stated that: f D (d RSSI = W )dd =1 (3.14) Ω(D) As f RSSI is not a function of d : Ω(D) (f RSSI(W, D = d) f D (d)) d d So, eq (3.13) + eq (3.15) give: f D (d RSSI = W )= f RSSI (W ) =1 (3.15) f RSSI (W,D=d) f D (d) RΩ(D) (f RSSI(W,D=d) f D (d)) d d (3.16) This means that there is no need to find f RSSI to calculate f D (d RSSI = W ). Nevertheless, f D has to be known. Two possibilities could be considered : There is a priori no knowledge about the distance. That is why it is possible to consider either that all the distances have the same probability or that all the coordinates of the Device have the same probability (and it is not the same at all). In this case: f D (d) = 1 D Max (3.17) There is a priori knowledge about the distance. In this case, there is no uniform probability distribution: there are only areas where the Device has more chances to be. It corresponds to a case in which time is taken into account, and where f D can be built from f DT t.it means that the function F t is such as f Dt = F t (f Dt t (d t t W t t )) (3.18) This function F could be built after having mentioned the Human Movement model. 36

37 CHAPTER 3. THEORETICAL ANALYSIS Triangulation: build a PDF of the location As the Triangulation method is used in a location process, three RSSI measurements have to be obtained. According to the mechanism described above, three PDF of the distance f D1,D2,D3 (. W ) are deduced from these measurements. With these three PDFs, the PDF of the location could be built. The definition is: M(X, Y ) is a location. (X, Y ) are the coordinates in the Cartesian coordinates system (0,x,y) Ω(M) is the set of all the possible locations f M (X, Y RSSI = W 1,W 2,W 3 ) is the PDF of the location at a time t knowing the three RSSI measurements of the three APs: W 1, W 2 and W 3 build f M (X, Y RSSI = W 1,W 2,W 3 ). The explanation of how this function will be built are described right after. Simple example considering only two APs : AP1 and AP2. >P 1 (d) is the probability that the distance between the AP1 and the Device is d >P 2 (d) is the probability that the distance between the AP2 and the Device is d The distance between AP1 and AP2 is supposed to be 10 meters and the probability distribution is: >P 1 (1) = 0.8 and P 1 (8) = 0.2 >P 2 (3) = 0.4 and P 2 (5) = 0.6 The circles of probabilities are drawn in Figure 3.1. P1=0,2 A P1=0,6 B P1=0,4 P1=0,8 d=1m AP2 C AP2 d=3m d=8m D d=5m Figure 3.1: Circles of probabilities 37

38 CHAPTER 3. THEORETICAL ANALYSIS There are four intersection points. To find the probability of each intersection point, the probabilities of two crossing circles have to be multiplied and then normalized with P A + P B + P C + P D =1. Consequently: >P A = P D = =0.3 >P B = P C = =0.2 The division by 0.4 is used for the normalization. The PDF of the location is built in the same way. The first calculated function is: M(X, Y ) Ω(M) g M (X, Y RSSI = W 1,W 2,W 3 )= 3 f D (d i RSSI = W i ) (3.19) i=1 d i = (X i X) 2 +(Y i Y ) 2 (3.20) where d i is the distance between the APi and the Device. After normalization it gives: f M (X, Y RSSI = W 1,W 2,W 3 )= RR g M (X,Y RSSI=W 1,W 2,W 3 ) Ω(M) g M (X,Y RSSI=W 1,W 2,W 3 )dx dy (3.21) Summary of the Localization algorithm Static Localization The bloc Power to distance uses the formula 3.16 with either the assumptions of the formula 3.17 or of the formula Dynamic Localization If a notion of time is introduced in the localization process, this one becomes more complex. The table 3.1 summarizes its different steps. In each column (at each measurement time), is measured or calculated the function or the parameter of each cell. The different steps that are calculated at the time t are stated in this order: 1. The RSSI is measured 2. The PDF of the distance knowing the RSSI measurement is calculated. This function is the most relevant concerning the distance 3. With the three PDFs of the distance knowing the RSSI measurement, the PDF of the location knowing the RSSI measurements is found 38

39 CHAPTER 3. THEORETICAL ANALYSIS W 1 W 2 W 3 Power to distance Power to distance Power to distance f D1 f D2 f D3 f D1 (. RSSI 1 ) f D2 (. RSSI 2 ) f D3 (. RSSI 3 ) Triangulation AP 1 (X 1,Y 1 ) AP 2 (X 2,Y 2 ) AP 3 (X 3,Y 3 ) f M (. RSSI 1,RSSI 2,RSSI 3 ) Figure 3.2: Localization algorithm 39

40 CHAPTER 3. THEORETICAL ANALYSIS It is noticeable that for f Dt0 (d RSSI = W i ), there is no previous f D (d RSSI = W i ) where T < t 0. Thus it is not possible to compute any f Dt0. A function that does not depend on time f D as to be used. It can be chosen at the point where all the distances have the same probabilities because there is a priori no knowledge about the position. f Mt0 (d t0 RSSI t0 = Time AP i PDF of d PDF of d knowing W i PDF of M knowing W i t 0 W t0 f D f Dt0 (d t0 RSSI t0 = W t0,i W 1,2,3,t0 ) t 1 = t 0 + t 1 W t1,i f Dt1 = F t1 (f Dt0 (d t0 RSSI t0 = W t0,i )) f Dt1 (d t1 RSSI t1 = f Mt1 (d RSSI t0 = W i ) W 1,2,3,t1 ) t 2 = t 1 + t 2 W t2,i f Dt2 = F t2 (f Dt0 (d t1 RSSI t1 = W t1,i )) f Dt2 (d t2 RSSI t2 = f Mt2 (d RSSI t0 = W i ) W 1,2,3,t2 ) Table 3.1: Summary of the different steps of our Location process 40

41 CHAPTER 3. THEORETICAL ANALYSIS 3.2 Propagation aspects Theoretical Propagation The following assumptions are made in order to simplify the problem 1. APs are placed on the ceiling 2. The Device is moving in a specific trajectory in the xy-plane at the tableheight (approx. 1m) 3. Reflections from walls and ground are considered. They are neglected from the ceiling. 4. Only single perfect reflections are considered 5. Antennas are Omni directional Propagation Free Space Propagation In free space propagation, the electric field in the xy-plane is considered only for direct rays. For the three APs in the room, the general expression of the electric field in free space propagation is where k = wave number = 2π λ λ = wavelength : c f ω =2πf: angular frequency r: distance between AP and Device t: time f = 2400MHz: Bluetooth frequency E d (r, t) =E m (r)e i(kx+ωt) (3.22) In equation (3.22), E m (r) represents the free space path loss and the exponential term stands for the sinusoidal dependency with distance and time. E m (r) is given by: E m (r) = 1 Z0 P T G T (3.23) r 2π where G T is the gain of the transmitter, Z 0 is the impedance in the air and P T is the power transmitted by the antenna. The relation between the electric field and the power in function of the distance is given by P (x) = E(r, t) 2 2Z 0 A eff (3.24) 41

42 CHAPTER 3. THEORETICAL ANALYSIS where A eff = effective aperture of Device antenna A eff = λ2 4π G R (3.25) where G R is the gain of the receiving antenna. Thus, in free space propagation, the final expression for the received power received in db is given by P db (r) =10log(P T G T G R )+20log( λ 4πx ) (3.26) This equation does not include any expression for reflections. In fact it is only for direct ray Propagation with AP1, AP2, AP3 The three APs (AP1,AP2,AP3) are placed against the ceiling of a room. From their locations, reflections from floor (r F ) and from the opposite walls (r W ) are considered. This assumption leads to the following expression of the electric field: E(x, t) =E d (r, t)+[e F (r F )+E W (r W )] (3.27) where E d (r, t) = electric field for direct ray E F = electric field after reflections from floor E W = electric field after reflection from walls Reflection coefficients: The reflections are considered as perfect. It means that the reflections coefficients Γ F and Γ W are equal to e jπ. E(x, t) =E m 1 r ej(kr wt) α=f,w Simulation of the propagation model E m 1 r α e j(krα ωt) (3.28) There is a comparison between a low passed curve of the propagation model that takes into account all the reflections against the walls (Figure 3.4) in a room of 20m by 20m and the free space model (Figure 3.3). The coordinates of the AP for this example are ( X=18.0, Y=2.0 ). Explanation of the low passed curve: As there is a lot of fast fading (Figure 3.3) we take a mean curve that is more readable as in the Figure 3.4. This mean curve is built by taking for each point the mean of the neighboring values as presented in the Figure 3.5. The mean value is constructed by taking a square which is 1.1 meters length. The wavelength in Bluetooth is equal to about meter. Thus, the square is about 10 times longer than the wavelength. This fulfills the conditions exposed in [34]. 42

43 CHAPTER 3. THEORETICAL ANALYSIS Figure 3.3: Free space propagation (distances unit is dm) Figure 3.4: Curve using our propagation model (distances unit is dm) 43

44 CHAPTER 3. THEORETICAL ANALYSIS 0,5 meter 0,5 meter The value of this point is equal to the sum of all the values of the neighbouring points that are distant from the considered point from less than half a meter, divided by the number of these points. 3.3 Movement Prediction Figure 3.5: Construction of the mean curve The previous parts only focus on static aspects. From this point, movement is introduced. Because the project mainly deals with handover, location is performed on a moving Device. Movement is a capital factor. By estimating how a Device has moved, it could be possible to improve the performed location. By predicting the future movement, it could also be possible to make the Handover easier. First, a simple intuitive Human movement model is built. This model is based on fundamental parameters that describe the movement. In this part is explained how to get these fundamental parameters from the previous movement. Then, our method to estimate or to predict the position and the distance from an AP with these parameters is explained Human movement model Parameters to model the movement The movement can be modeled with only two parameters. These two parameters are chosen as: the instant speed s = s the instant direction Θ=arg( s) The aim now is to model these two parameters. The first assumption is that there is no correlation between them. An independent model can be built for these two parameters Speed Parameter Assumption When an individual is mobile, he is mainly moving with a characteristic speed. Our intuition is that when somebody is walking in a place without 44

45 CHAPTER 3. THEORETICAL ANALYSIS any obstacle, he moves with a certain speed: his characteristic speed. Description of the model Our model of speed is a stochastic process that is stationary, but not ergodic. It means that the statistic of the values collected across the ensemble of speed at any instant is different from the statistic of the values collected on one sample. To be clear, this assumption needs further explanations. At a fixed time, is considered an infinity of people moving according to the model. g S is the function of the statistic representing all the speeds s. g S (s) is the frequency of apparition of the speed s. For a moving body i, we consider all the speeds it can have during an infinitely long period of time. f Si is the function of the statistic representing all the speeds s it has during this period. f Si (s) is the frequency of apparition of the speed s. The stochastic process speed is not ergodic because f Si and g S are not necessarily equals. We dwell on this fact since it is important to consider that each person moves according to personal movement settings. In addition, the whole stochastic process will be mathematically described by these two functions f Si and g S. The principle is illustrated in figure 3.6. samples(i) statistic across t on one sample s 3 (t) s 2 (t) s 1 (t) time(t) statistic across the samples Figure 3.6: Description of the stochastic process of the speed Mathematical description of the model: 45

46 CHAPTER 3. THEORETICAL ANALYSIS The statistic g S of the values collected across the ensemble of speed at any instant gives the distribution assumed on the figure 3.7: a gaussian centered on the speed 1m/s with a variance equal to 0,5m/s : for s<0: for s>0: g S (s) = g S (s) =0 (1 s) 2 e( ) (3.29) (1 s)2 ( 0 e ) probability speed (m/s) Figure 3.7: PDF of the speed of all the moving bodies The statistic f Si of a given sample function S i is equal to a gaussian centered in the characteristic speed s char i of the moving body with a standard deviation of 0.1m/s: (s 1 char i s) 2 f Si (s) = 2π 0.5 e( ) (3.30) An example is proposed with the Figure 3.8 where s char =1.2m/s: Direction Parameter Assumption It is considered in our model that humans go straight to their goal. If it is assumed that there is no obstacle, it means that the instant direction will not 46

47 CHAPTER 3. THEORETICAL ANALYSIS probability Schar= speed (m/s) Figure 3.8: Example of PDF of the speed of one precise moving body change a lot and will move around a centered value which is called the characteristic direction. Description of the model Our model of the direction is similar to the model of speed. It is also a stochastic process that is stationary but not ergodic. Mathematical description of the stochastic process: Statistics g Θ of the values collected across the ensemble of directions at any instant gives a PDF with a uniform distribution for all the directions as presented in figure 3.9. for θ< 180 and θ>180: for 180 <θ<180 g Θ (θ) =0 g Θ (θ) = (3.31) The statistic of a given sample function f θi is equal to a gaussian centered in the general direction of the moving body θi char and with a standard deviation of 30: (θ 1 i char θ) 2 f Θi (θ) = 2π 30 e( ) (3.32) An example is proposed with the Figure 3.10 where θ char =45. 47

48 CHAPTER 3. THEORETICAL ANALYSIS probability 1/360 1/ direction (degrees) Figure 3.9: PDF of the direction of all the moving bodies probability direction (degrees) Figure 3.10: PDF of the direction for one precise moving body 48

49 CHAPTER 3. THEORETICAL ANALYSIS Parameter Estimation Our location system enables to get PDFs of the location at the present location time t and also at the last measurement time t t p. The aim is to predict a PDF of the location at time t + t from the two last performed locations and using our model of human movement. The t parameter can be chosen. It means that it is possible to choose when a prediction must be perform. Information from the past: The two last locations are considered. The Device is at the point M t tp at time t t p and at the point M t at time t. Since there are different locations at different times, the speed is not known continuously. Velocity vectors representing the mean of the speed between the two last locations are the only available factors. The speed and the direction of the mobile can be deduced from these vectors. Statistic estimators: The prediction uses statistic estimators. It is considered that the speed of the mobile between t t p and t is close to the mean of the instant speed i.e. the characteristic speed. So the estimation of the characteristic speed is: M t tp M t s char = (3.33) t p It is the same process for the estimation of the direction: θ char =arg( M t tp M t ) (3.34) Position Prediction A Device i is considered and now the aim is to determine a PDF of its location at time t + t. The PDF of the speed f S and the PDF of the direction f Θ are useful for this work. There are two cases: less than two localization have been previously performed. The parameters Θ char and S char cannot have been computed. So we take: f Θ = g Θ (3.35) f S = g S (3.36) more than two localizations have been previously performed. Consequently, the parameters Θ char and S char can be deduced from the two last localizations: f Θ = f Θi (3.37) f S = f Si (3.38) 49

50 CHAPTER 3. THEORETICAL ANALYSIS The prediction of the PDF of the location at t+ t is performed using the following process: M t+ t (X t+ t,y t+ t ) Ω(M) f Mt+ t (X t+ t,y t+ t )= M t Ω(M) f M t (X t,y t ) RR f Θ (σ) f S (υ) Ω(M) f Θ(σ) f S (υ) dx dy where σ =arg( M t M t+ t ) and υ = MtM t+ t t (3.39) Estimation of the Distance from the Access Point Since the location has already been predicted, it is useless to estimate the distance. However, it has been shown in the section 3.1 that f Dt+ t(d t+ t ) is estimated from f Dt (d t RSSI). This prediction is useful at this point of the location algorithm. Our procedure is explained in this section. The estimated PDF of the speed of the Device f S, the estimated PDF of the direction of the Device f Θ at time t and f Dt (. RSSI) are used to build this function. The general formula is: where: P(d t+t )= s,d t Ω(D) p fθ,s,d t (d t+ t )f S (s)f Dt (d t RSSI t )δs.δd t (3.40) P(d t+t ) is a discrete probability distribution function of the estimated distance between the Device and the considered AP at time t + t. It is a discrete Probability because a finite distance method is used for computing it. Indeed, there is no way to compute it with analytic methods. p fθ,s,d t (d t+ t ) is the probability considering a fixed speed s and a fixed distance from the AP at time t: d t. The index f Θ means that the calculation of this parameter includes the use of the knowledge about the direction. Anyway, since the computation of this parameter, considering the angle, needs too much computation power, we computed it without taking into account the knowledge about the direction. f S (s) is the PDF of the speed that is used for this estimation. It can be either an estimation from a characteristic speed or the general distribution of the speed. It depends on the history of the movement. We will further explain which one we chose. δs is the space between two considered speed values. f Dt (d t RSSI) is the PDF of the distance at the time t knowing the RSSI measurement. 50

51 CHAPTER 3. THEORETICAL ANALYSIS δd t is the space between two considered distance values. In the formula 3.40, the product δd t.δs corresponds to a little integration surface of the integration of the functions f S (s) and f Dt (d t RSSI t ) weighted by the factor p fθ,s,d t. The details of the calculation of the factor p fθ,s,d t are given in Appendix A. 51

52 CHAPTER 3. THEORETICAL ANALYSIS 3.4 Handover Handover is a process of seamless connectivity in which a Device moves from one Access Point coverage area to another. At first, the Inquiry and Paging process are detailed in order to have a better understanding of our handover Establishing connection in Bluetooth network During the establishment of a connection in Bluetooth network, the Access Point and the Device can take different states [1]. Inquiry Used by an AP to discover the Bluetooth Device Address (BD_ADDR), the clock (CLK) and other information of the Device in range Inquiry scan Used by a Device to listen for an Inquiry Page Used by AP to establish a connection with a particular Device whose BD_ADDR is known Page scan Used by a Device to listen for its page The connection between an AP and a Device can be established either by Inquiry and Paging when the Device details are unknown, or only by Paging when the Bluetooth Device Address is already known by the AP Inquiry Procedure This process occurs when the piconet has not been established. The AP is defined as the prospective Master and the Device as the prospective Slave. The following steps sum up the Inquiry procedure [1]: 1. To discover a new Device in its range, the AP launches an Inquiry procedure. It transmits two Inquiry Access Code (IACs) on two consecutive hop frequencies from the 32-hop Inquiry sequence during an even-numbered time slot. The AP listens on the two corresponding Inquiry response hop frequencies in sequence during the next odd-numbered slot. 2. Meanwhile, the Device listens for the IAC on one of the Inquiry hop frequencies for about 11 ms each time it enters INQUIRY SCAN state. 3. Upon hearing the IAC, the Device delays a random time. Then it listens for the IAC on the same frequency again. After having received the second IAC packet, the Device transmits its FHS packet 625 µs later on the corresponding Inquiry response hop frequency. 4. The AP does not respond to the FHS packet, but stores it for future use if needed. 52

53 CHAPTER 3. THEORETICAL ANALYSIS Thus, after the Inquiry, the AP knows the following information about the Device thanks to the FHS packet: the Device address (BD_ADDR) which is unique for each Bluetooth Device. the Scan Repetition which is the time interval between two successive Page Scan windows the Scan Period which is the time period in which the mandatory Page Scan mode will be used by the Device after it responds to an Inquiry. the clock of the Device (CLKN). As noticed, the FHS packet contains the sender s entire BD_ADRR and enough bits of CLKN that can help the receiver to program a hop generator to follow the sender s hopping from channel to channel. Thus, this information is sent to a Handover Agent (HA) which will transmit it to all the APs in our scenario: this process could speed up the page process as explained in the next section Paging Procedure After the Inquiry process, the next step to the connection of the Device to a piconet is the Paging process. The Paging procedure is used by the AP to establish a connection with the Device. For Paging, the AP has to know the BD_ADDR of the Device (it has been done in the Inquiry process). In this process, the Device must be in PAGE state and the AP in PAGE SCAN state. The Paging procedure follows different steps as explained below [1]. 1. The AP transmits two Devices DACs on two consecutive hop frequencies, f(k) and f(k+1) from the 32-hop page sequence during an even-numbered time slot. 2. The AP listens on the two corresponding hop frequencies in sequence during the next odd-numbered time slot. 3. Meanwhile, the Device listens for its DAC on one of the page hop frequencies for about 11 ms each time it enters the PAGE SCAN state. 4. Upon hearing its DAC, the Device then transmits the same DAC 625µs later on the corresponding page response hop frequency. 5. If the AP hears the DAC returned on the corresponding page response frequency then it sends its FHS packet to the Device. 6. Now the AP and the Device are connected as Master and Slave. 53

54 CHAPTER 3. THEORETICAL ANALYSIS In our scenario, the Device information is known by all the APs. Thus, the Page and Page response hop sequences are both derived from the Device s BD_ADDR. However the hop phase has not yet been coordinated precisely, except for the AP s estimate of the clock of the Device. It will be done with the FHS packet response of the AP (step 5). Two cases may happen in the Paging process: either the Device f(k) f(k+1) f (k) f(k+1) f (k+1) g(m) g(m+1) Tx AP DAC Page FHS POLL t Rx t Tx Device DAC DAC NULL t Rx 625us Page and Page response hoping sequence Channel Hoping sequence t Figure 3.11: Packet Exchange in the Paging Process receives the DAC packet in the first half of the even-numbered time slot or in the second half. The p-salve cannot know that. So, it considers the worst case (the second half of the time slot). Thus, after µs, the Device is ready to listen to the frequency f(k+1). Overall, the Paging process, until a complete connection, lasts 4 timeslots (2.5 ms) in the best case. If all the 32 frequencies have to be scanned by the AP, the page can lasts 36 timeslots (22.5 ms) The Paging timers Several timers are defined to avoid infinite loops when a response is expected by the AP or by the Device from the other one [1]. Page timeout (pageto): it is the number of time slots that the PAGE state can last before exiting if there is no response to the page. By default pageto is set to 5,12s. However it can be set to another value by the user. Page response (pagerespto): for the Device it is the time between the acknowledgment of the Page from 54

55 CHAPTER 3. THEORETICAL ANALYSIS the Device and the FHS packet from the AP. It is set to 8 slots. for the AP it is the number of time slots it will wait for the Device to respond to the FHS packet. This Paging timer is also set to 8 slots. New connection timeout (newconnectionto): it is number of timeslots the AP has to wait for the Device to respond to the POLL packet. After expiration of newconnectionto, the AP and the Device return respectively to PAGE and PAGE SCAN states. 3.5 Conclusions To estimate the future location of the Device and its movement, previous location, speed and direction are taken into account. Probabilities calculation are computed with these parameters to determine the most probable location of the Device. Besides, a boundary will be defined for each AP. They limit the area covered by each AP. When a boundary is crossed by the Device, the handover procedure is performed. 55

56 Chapter 4 Simulations This chapter will cover the implementation details of the methods previously discussed in the Theoretical Analysis. In order to achieve a realistic simulation, we need to set up some near real-life conditions through a given scenario. Given these assumptions, the simulation will perform the actual sequential steps of our model. In other words, how to get data from the Device, forecast its movements and use both real and forecast information to achieve handover. At last but not least, the different results obtained with the simulation and a comparative study will be given. 4.1 Scenario The proposed scenario consists in 3 APs in a single room and in a mobile device moving along a specified trajectory in the xy plan. For the sake of the scenario, some parameters have been defined. The room is 20 meters large and 20 meters long. The height of the room is 2.3 meters. The Access Points are placed on the ceiling of this room. They are linked by wire to a Handover Agent whose role will be defined in the Handover section (4.3.5). The Device is moving in a plane parallel to the ground at a height of 1m. The figure 4.1 represents the room in 3-D. However all calculations and movements will be considered in 2-D (in the plane of the Device). The place of the APs (in the figure they are placed in the corners) are arbitrary : this is just an example of a possible configuration of the room. They can be put in other places on the ceiling. 4.2 Simulator Principle Our simulations are split into five parts. The goal is to simulate the whole scenario by taking into account fundamental parameters. Here are the different followed steps: + Generation of RSSI measurements + Location of the Device 56

57 CHAPTER 4. SIMULATIONS AP2 AP1 AP3 Device 2.3m z y 1m 20m x 0 20m Figure 4.1: Possible configuration of the Room + Location of a moving Device + Movement Prediction + Handover The principle of all these parts is explained in the section 4.3. They have been computed in the same order because each part needs the previous ones to be achieved. The Figures 4.2 and 4.3 summarizes all the performed simulations. Different possible settings Simulations Figure 4.2: Simulation 4.3 Functionalities Breakdown Generation of RSSI measurements Location is based on RSSI measurements get from the three APs. These measurements have to be simulated properly in order to measure the performances of the Location, the Movement Prediction and the Handover processes. Two main parameters have to be generated. First, the RSSI measurement of a Device at a given position. Then, the delay between two RSSI measurements that corresponds exactly to the Inquiry process delay Generation of one RSSI measurement It is the first step. The considerations in this part are limited to one AP and the Device. The parameters taken into account to generate measurements are the re- 57

58 CHAPTER 4. SIMULATIONS General Setings Room of 20*20m Localization in the horizontal plan 3 AP Generation of measurements R S S I D e l a y measurement Free space measurements (distance dependant) measurements taking into account the propagation model with reflections (position dependent) between two measurements o f R S S I Delay corresponding to the inquiry time (no dependance) Device Fixed in one coordinate (X,Y) Location o f a fixed point Location Performances in free space Location performances in the propagation model with reflections Location with adding information about position Location of a moving device Performances with basic location Performances taking into account the movement model Moving along a defined trajectory [X(t),Y(t)] Movement Prediction Reliability of the prediction H a n d o v e r Figure 4.3: Summary of all the simulations performed 58

59 CHAPTER 4. SIMULATIONS spective positions of the AP and the Device. There are two different ways to generate measurements: With a free space propagation model. In this case, the received power depends only on the distance between the AP and the Device. With a propagation model considering reflections (see Section 3.2). In this case, the received power depends on both places of the Device and the AP. These two models give us the mean of the power the AP should receive from the Device at different places. According to the Master Thesis of Joao Figueiras [7], the power received by the AP when the Device is at a given place follows a Gaussian centered in the mean power with a standard deviation of σ = Anyway, we can represent the RSSI measure generation with the box in the Figure 4.4. AP(X,Y) RSSI Generation device (x,y) ω Figure 4.4: RSSI generation simulator element Generation of a Delay between two RSSI measurements The RSSI measurements are not instantaneous. As the considered RSSI measurements are the measurements given after an Inquiry, there is a delay corresponding to this Inquiry time defined in Bluetooth Specifications [4]. This delay is generated on the basis of Joao Figueira s master thesis [7]. It means the delay is generated randomly, weighted by the PDF of the Inquiry time calculated for 1 AP considered in [7]. The box in the Figure 4.5 represents the Delay generation Generation of measurements for a trajectory The scenario is a Device following a precise trajectory in a room with three APs. So measurements have to be generated for the three APs during the whole trajectory. As there are three APs inquiring at the same time, we are making the following assumption: there is no influence between each AP for the Inquiry time. Because only one packet is sent at the end of the Inquiry, it is reasonable to make this 59

60 CHAPTER 4. SIMULATIONS Delay t Figure 4.5: Delay generation element assumption. Thus, each AP can be considered the same way and a set of measurements can be done independently for each AP. In order to explain the complete procedure of the generation of measurements, it is better to explain what a trajectory in our programs is. A trajectory is considered as a set of points. Each point corresponds to a precise point in time. The Figure 4.6 represents a trajectory. t Trajectory x(t),y(t) Figure 4.6: Trajectory element The Figure 4.7 explains how the measurements are generated for the whole trajectory. The RSSI measurements can be represented on the three APs with the Figure 4.8. Finally, it is noticeable that all APs can get measurements at really different times. One of the difficulties will be to locate a moving Device with RSSI measurements which do not correspond to the same location of the Device. Anyway, before going further into these problems, we have to be able to localize a fixed Device Location of a fixed Device The first step, after generating measurements, is to locate the Device at a fixed point. The first test appraises the validity of the location system built in this project (section 3.1). It also gives a visual impression of the mean behavior of the location by averaging the obtained PDF of a great number of localizations. The other tests analyse the performances for this location system by changing different parameters. The labels of the figures of the rooms are in meter. 60

61 CHAPTER 4. SIMULATIONS t Stop the loop Enter the loop At t=0 AP(X,Y) Delay New delay generated once the RSSI is generated t>t max t<t max Trajectory RSSI Generation (x(t),y(t)) w t Figure 4.7: Block of the RSSI measurement generation along a trajectory AP1 t AP2 t AP3 t = RSSI measurement Figure 4.8: Representation of the RSSI measurement on a time axis per AP 61

62 CHAPTER 4. SIMULATIONS Different Location of the Access Points The table 4.1 gives the characteristic of proximity in our cases. Device far from the AP Case 1 Case 2: worst case Device closer to the AP Case 3: best case Case 4 Table 4.1: Summing-up of the different cases Case 1 The AP and the Device are at the following coordinates as in Figure 4.9: AP1 (2.0,18.0) AP2 (2.0,2.0) AP3 (18.0,18.0) Device (10.0,10.0) 25 Room AP1 AP3 Y axis 10 Device 5 0 AP X axis Figure 4.9: Room for the first case Case 2 The AP and the Device are at the following coordinates as in Figure 4.10: AP1 (2.0,18.0) AP2 (2.0,2.0) AP3 (18.0,18.0) Device (18.0,2.0) Case 3 The AP and the Device are at the following coordinates as in Figure 4.11: AP1 (7.5,12.5) AP2 (7.5,7.5) AP3 (12.5,7.5) Device (10.0,10.0) 62

63 CHAPTER 4. SIMULATIONS 25 Room AP1 AP3 Y axis AP2 Device X axis Figure 4.10: Room for the second case 25 Room Y axis 10 AP1 Device 5 AP2 AP X axis Figure 4.11: Room for the third case 63

64 CHAPTER 4. SIMULATIONS Case 4 The AP and the Device are at the following coordinates as in Figure 4.12: AP1 (12.5,7.5) AP2 (12.5,2.5) AP3 (17.5,2.5) Device (18.0,7.5) 25 Room Y axis 10 5 AP1 Device 0 AP2 AP X axis Figure 4.12: Room for the forth case General behavior of the program The Figure 4.13 shows how the different parts of the simulator are put together to perform the localization of a fixed Device. Device (x,y) AP1 RSSI AP2 RSSI AP3 Generation Generation RSSI Generation W 1 W 2 W 3 f D1 f D2 f D3 Localization AP1 AP2 AP3 f M (. RSSI 1,2,3 ) Figure 4.13: Simulator for the localization of a fixed Device It consists in computing the mean of the n PDFs of the location calculated for 64

65 CHAPTER 4. SIMULATIONS n iterations. It uses the formula (3.16) to compute the PDF of the distance for each AP and then, equation (3.21) to compute the PDF of the location. As it is a static process, we do not have any former information about the location of the Device. Thus, the assumption of the equation (3.17) is made: the Device can be anywhere in the room. As the aim is to see the reliability of the program, the first set of tests are made by generating measurements with the free space model. As indicators, we compute the following parameters on the obtained curves: + The coordinates of the maximum likelihood estimator in the final PDF of the location. + The distance between the maximum likelihood estimator and the real location: this parameter gives us obviously the reliability of the estimation of the location by the highest point of the density. The greater the distance the worse this estimation. + The MeanRadius: it is the radius of an area we assimilate to a circle. On the curves of density, we calculate the area A where there are values greater than the half or the fifth of the maximum value. If we assimilate A to a circle, the A radius is π. It is called the MeanRadius because it expresses the mean of the distance between the point with the greatest value and the borders of the considered area. Then, this parameter helps us to have an idea on how far from the good location the result obtained could be. + The Half-beam area: it is the area where the values of the density are greater than the half of the maximum value. + The Probability Ratio of the highest point: it is the ratio of the value of the real position density to the maximum of the density. We have different cases to interpret this value: - If it is very close to 1, the estimation of the location by the highest point is a good option. - If it is more than 0.5, it means the half-beam area is a correct estimation of the location. Results: Table 4.2 Case 1 The coordinates of the point calculated tend to be at a close distance from the original ones: in a radius of 0.84 meter. However the mean radius (half beam) computed is very large, around 4.3 meters: that means that the shape of the final PDF is very spread (Figure 4.14). It implies that the location program cannot localize very precisely with such a configuration. 65

66 CHAPTER 4. SIMULATIONS case Number of measures Coordinate Max distance(m) Meanradius(m) probability ratio X Y Table 4.2: Summing up of the results obtained Figure 4.14: PDF of the position estimation in the first case 66

67 CHAPTER 4. SIMULATIONS Case 2 In this case, we can observe that the coordinates of the point calculated tend to be quite far from the original ones: in a radius of 1.7 meters. This result means that this case for location is a very bad one. This observation is confirmed by the fact that the mean radius (half power) computed is very large, around 4.5 meters: the shape of the final PDF, like in the first case, is very spread. Thus, this case is the worst case for location: indeed, the Access Points are very far from the Device and this latter is near from the borders. Besides, the shape is as spread as in the first case: this means the borders have just a little effect on the calculated PDFs (Figure 4.15). Figure 4.15: PDF of the position estimation in the second case Case 3 This case shows best results: indeed, the Access Points are near from the Device which is far from the borders. The distance from the calculated point and the original one is small, only 2.8 meters. The location is fairly good here. Moreover, the Mean Radius (half power) is the lowest of all the four tests: 1.44 meters (Figure 4.16). Case 4 Same conclusions can be made in this case. Indeed, the configuration is the same as in the case 3. As we used a free space model to generate the measurements, there is no effect from the proximity to the walls Conclusion For the accuracy of our location, we can say that a good percentage of location is obtained when a Mean Radius is taken for a fifth of the power. Indeed, the percentage is improved significantly compared to the case when a mean radius is taken for the half power. However, in the two first cases, the Mean Radius (fifth power) is large: it means that the location is not very accurate. On the contrary, 67

68 CHAPTER 4. SIMULATIONS Figure 4.16: PDF of the position estimation in the third case when the APs and the Device are close to each other, the accuracy of location is increased Efficiency of the localization These tests consist in computing the following parameters for 300 localizations: the distance between the maximum point and the real location, the half beam, the fifth beam area and their MeanRadius. We also compute the percentage of the times when the real location is in the half beam and the fifth beam area. By this way we can determine the accuracy and the efficiency of the location. The results are displayed in the table 4.3. Results: Conclusion Probability and distance are relevant of the nature of the localization. In the two first cases, the APs are further than 10 meters from the Device. It is a large distance for Bluetooth. The Device is located only nearly 20% of the cases at less than 2m from the maximum likelihood estimation. In the cases 3 and 4, where the device is closer to the AP (nearly 5 meters), the device is located in more than 80% of the cases at less than 2m from the maximum likelihood estimation. The simple conclusion that can be made there is that the localization is more precise when the APs are close to the device as seen on the Figure

69 CHAPTER 4. SIMULATIONS Case 1 Case 2 Case 3 Case 4 Mean of the Distance(m) 3,43 3,62 1,33 1,04 Standard Deviation (m) 2,05 2,49 0,94 0,57 Confidence Interval (m) (95% of probability) + or 0,23 0,28 0,11 0,06 - Percentage below 1m ,3 54 Percentage below 2m 26,3 16,3 85,3 94,3 Percentage below 3m Percentage below 4m 68,3 69, Mean of the MeanRadius in the area greater than 3,01 2,72 1,08 1,04 50% of the maximum likelihood (m) Standard Deviation of the MeanRadius (m) 0,36 0,81 0,15 0,28 Confidence Interval (95% of probability) + or - 0,04 0,09 0,02 0,03 Percentage of presence of the Device in this area 49,6 59,6 46,3 55,3 Confidence Interval of the percentage (95% of probability) + or - Mean of the MeanRadius in the area greater than 4,65 3,91 1,74 1,56 20% of the maximum likelihood (m) Standart Deviation of the MeanRadius (m) 0,5 0,97 0,2 0,39 Confidence Interval (95% of probability) + or - 0,06 0,11 0,03 0,05 Percentage of presence of the Device in this area 7 9,3 86,3 77,6 89 Confidence Interval of the percentage (95% of probability) + or Table 4.3: Table of results 100 percentage of cases where the distance is inferior Case 4 Case 3 Case 1 Case Distance between the maximum likelihood estimate and the real location (m) Figure 4.17: Precision of the maximum likelihood estimation 69

70 CHAPTER 4. SIMULATIONS Another tested parameter is the area with more than a certain threshold of probability. In previous analysis we have tested the areas with a probability superior to 20% and 50% of the probability of the maximum likelihood estimation. It is there noticeable that for the same threshold, the probabilities of presence of the device in the area are comparable. Indeed it is nearly 50% for a threshold of 50% and 80% for a threshold of 20%. Nevertheless, the size of the area changes a lot according to the different cases. In the cases 1 and 2, the area with a threshold of 20% has a MeanRadius of nearly 4m. In the cases 3 and 4, the MeanRadius is nearly equal to 1.7m. There is also a relevant difference between the precisions in both cases, when the APs are far and when they are close to the Device. Choice These conclusions lead us to an orientation to determine the best position for the APs. They clearly have to be as close as possible to the Device. It means they do not have to be close to the walls. A configuration similar to the case 3 seems to be one of the best. Indeed, anywhere in the room, the device would never be further than 15m from any AP Location with former information about the location of the Device This test consists in adding information about the location of the Device. Indeed, the goal of the project is to find the location of a moving Device. The locations before t give information about the locations at t. The goal in this test is to see how information about the location can improve the localization. We are not making the assumption (3.17) anymore now. In this part, only the case 3 is treated because one case is sufficient to have and idea of the improvement. Instead of (3.17): Results: d [3.5, 8.5] f D (d) = 1 5 else f D (d) =0 Former information MeanDistance(m) MeanRadius Half- Power(m) Percentage of good location in the halfbeam MeanRadius Fifth- Power(m) No Yes Percentage of good location in the fifthbeam Table 4.4: Localization with former informations about the location of the Device Conclusion The results are quite impressive compared to the original results. The main significance is that all information that can be obtained on previous locations are really relevant for the accuracy of the computed location of the Device. 70

71 CHAPTER 4. SIMULATIONS Localization by generating measures with the room propagation model The goal of these tests is to see the influence of the reflections on the walls on the precision of the localization. For each case, the test of efficiency has been redone. The following tables contain results without reflection and with reflections (Tables 4.5, 4.6, 4.7 and 4.8): Case 1 No reflection With reflection Mean of the Distance(m) Standard Deviation (m) Confidence Interval (m) (95% of probability) + or Percentage below 1m Percentage below 2m Percentage below 3m Percentage below 4m Mean of the MeanRadius in the area greater than 50% of the maximum likelihood (m) Standard Deviation of the MeanRadius (m) Confidence Interval (95% of probability) + or Percentage of presence of the Device in this area Confidence Interval of the percentage (95% of probability) or - Mean of the MeanRadius in the area greater than 20% of the maximum likelihood (m) Standard Deviation of the MeanRadius (m) Confidence Interval (95% of probability) + or Percentage of presence of the Device in this area Confidence Interval of the percentage (95% of probability) + or Table 4.5: Results for the case 1 71

72 CHAPTER 4. SIMULATIONS Case 2 No reflection With reflection Mean of the Distance(m) Standard Deviation (m) Confidence Interval (m) (95% of probability) + or Percentage below 1m Percentage below 2m Percentage below 3m Percentage below 4m Mean of the MeanRadius in the area greater than 50% of the maximum likelihood (m) Standard Deviation of the MeanRadius (m) Confidence Interval (95% of probability) + or Percentage of presence of the Device in this area Confidence Interval of the percentage (95% of probability) or - Mean of the MeanRadius in the area greater than 20% of the maximum likelihood (m) Standard Deviation of the MeanRadius (m) Confidence Interval (95% of probability) + or Percentage of presence of the Device in this area Confidence Interval of the percentage (95% of probability) + or Table 4.6: Results for the case 2 72

73 CHAPTER 4. SIMULATIONS Case 3 No reflection With reflection Mean of the Distance(m) Standart Deviation (m) Confidence Interval (m) (95% of probability) + or Percentage below 1m Percentage below 2m Percentage below 3m Percentage below 4m Mean of the MeanRadius in the area greater than 50% of the maximum likelihood (m) Standard Deviation of the MeanRadius (m) Confidence Interval (95% of probability) + or Percentage of presence of the Device in this area Confidence Interval of the percentage (95% of probability) or - Mean of the MeanRadius in the area greater than 20% of the maximum likelihood (m) Standard Deviation of the MeanRadius (m) Confidence Interval (95% of probability) + or Percentage of presence of the Device in this area Confidence Interval of the percentage (95% of probability) + or Table 4.7: Results for the case 3 73

74 CHAPTER 4. SIMULATIONS Case 4 No reflection With reflection Mean of the Distance(m) Standart Deviation (m) Confidence Interval (m) (95% of probability) + or Percentage below 1m Percentage below 2m Percentage below 3m Percentage below 4m Mean of the MeanRadius in the area greater than 50% of the maximum likelihood (m) Standard Deviation of the MeanRadius (m) Confidence Interval (95% of probability) + or Percentage of presence of the Device in this area Confidence Interval of the percentage (95% of probability) or - Mean of the MeanRadius in the area greater than 20% of the maximum likelihood (m) Standard Deviation of the MeanRadius (m) Confidence Interval (95% of probability) + or Percentage of presence of the Device in this area Confidence Interval of the percentage (95% of probability) + or Table 4.8: Results for the case 4 74

75 CHAPTER 4. SIMULATIONS Conclusion: The results are globally worse when the reflections are taken into account while generating the measurements than without reflection. Indeed, as seen in the part 3.3, reflections bring variations to the smooth curve of the free space propagation model. These variations bring less correlation between the distance and the signal power. Thus, there is less precision in the localization. Cases 1 and 2 give worse localization results than without considering reflections. However the difference is not so great, especially when considering the distance between the maximum likelihood estimation (red curve) and the real location (green curve)(see Figures 4.18 and 4.19). Indeed, at distances over 10m, the power does not vary too much. It remains between -40 and -45dB. It can be inferred that the variations brought by the consideration of the reflections hardly affect the precision of the localization. Indeed the power is always in the same narrow area. Given the powers found, it is normal that large areas are deduced. We can see on the results that the most reliable percentage of presence in an area is the percentage of presence in the area greater than 20% of the maximum likelihood estimation. Finally, only bad localizations can be expected when the Device is far from the APs. 100 percentage of cases where the distance is inferior Distance between the maximum likelihood estimate and the real location (m) Figure 4.18: Comparison for case 1 Cases 3 and 4 show greater differences. There the APs are close to the device. The variations brought by the reflections have more influence on the results. Indeed, when the Device is close to the APs (below 10m), the power varies a lot with respect to the distance: it varies between -25dB and -40dB. This phenomenon can also been observed through the mean of the distance between the good location and the maximum likelihood estimated point. In case 3, this distance increases by 0.7m and in case 4 by 2.1m. The 2-meter area around the maximum likelihood estimation is not a reliable area anymore for the localization. Indeed, in case 3 only 75

76 CHAPTER 4. SIMULATIONS 100 percentage of cases where the distance is inferior Distance between the maximum likelihood estimate and the real location (m) Figure 4.19: Comparison for case 2 65,6% of cases are good compared to 85.3% and in case 4 only 35% compared to 94.3%. We can also state that the results for the case 4 become really worse than those for the case 3 after considering the reflections. We can clearly see it through the distance between the maximum likelihood estimation and the real location (see Figures 4.20 and 4.21). The percentage of good localization in the area of probability greater than 20% of the maximum likelihood estimation goes from 77.6% to 65.6% in case 3. It sags from 89% to 50% in the case 4. The latter deterioration is due to the proximity of the APs from the wall. When an AP is close to a wall, the reflected signal from the closest wall has almost the same power than the direct signal. It means that the phase difference affects more the resulting signal. Anyway, it does not mean that the localization is better in cases 1 and 2 than in cases 3 and 4. The areas found in the two first cases are too large. To improve the localization percentage in cases 3 and 4, it is sufficient to increase the size of the considered areas. The different results of the percentage in the cases where the distance between real location and maximum likelihood estimation is below 1, 2, 3 and 4m shows that in all cases a good localization is achieved below 4m. Definitive Choice: All in all, configuration of case 3 seems to be the best one. The APs are as close as possible to the probable Device and the APs are not close to the walls. This kind of configuration will be used in our future experiments with a moving Device. 76

77 CHAPTER 4. SIMULATIONS 100 percentage of cases where the distance is inferior Distance between the maximum likelihood estimate and the real location (m) Figure 4.20: Comparison for case percentage of cases where the distance is inferior Distance between the maximum likelihood estimate and the real location (m) Figure 4.21: Comparison for case 4 77

78 CHAPTER 4. SIMULATIONS Location of a moving Device In this section, time and motion are two more parameters taken into account. The coming problem is due to the fact that RSSI measurements of each AP will not be taken at the same time. It also means that the device will not be at the same place at each measurement Scenario settings Position of the Access Points: the positions are chosen according to the previous tests. They are centered in the room. AP1 (10.0,15.0) AP2 (5.0,5.0) AP3 (15.0,5.0) Trajectory: The Device is moving from the right of the room to the left of the room at the speed of 2m/s. It means that the carrier of the Device is a fast walker if we refer to the considered movement model. More precisely it starts from the point of coordinate (20,10) and it reaches the point (0,10) in 10 seconds. 20 Room AP1 Y axis Device 4 2 AP2 AP X axis Figure 4.22: Room, trajectory and APs Basic location of a moving body The basic idea to locate a moving device is to perform localization as soon as we have one RSSI measurement for each AP and then wait for three other RSSI measurements to localize again. The obvious problem is that the time interval between the three considered RSSI measurements used for each location is not optimized. It can be very long. It seems to be difficult to optimize it and to find sets of three RSSI measurements that have the smallest time interval. Indeed, it 78

79 CHAPTER 4. SIMULATIONS is not sure that there is always a set with a small time interval. Furthermore it adds some problems with the frequency of localization. There is also the option of localizing by triangulation after using each received RSSI measurement from one AP with the last RSSI measurements of the two other APs. This method which is our option is explained on the Figure nd Set 3 rd Set AP1 t AP2 t AP3 t 1 st Set 4 th Set RSSI Measurements Figure 4.23: Method for choosing the measurements to localize Simulation The localization PDF is computed each time an RSSI measurement is received as explained above. The PDF is computed with the formulas (3.16) and (3.21). Moreover the assumption (3.17) is made. This is made exactly like in the location of a fixed Device. Finally, the main error factor that appears, compared to a fixed location, is the delay between the different measurements. The trajectory is run 20 times and we calculate the mean of the evaluation parameters on all the locations performed. The results are displayed in the table 4.9. Number Of MeanDistance MeanRadius Percentage of MeanRadius Percentage of Locations (m) Halfbeam (m) good location in the halfbeam Fifthbeam (m) good location in the fifthbeam Table 4.9: Results of the simulation Conclusion The results are really bad there. It is impossible to rely on these localizations. Another system has to be found. 79

80 CHAPTER 4. SIMULATIONS Localization with human movement model The solution found is to take into account the human movement model. Localization is also performed there each time an RSSI measurement is received. The difference with the basic location of a moving body is that the localization is not performed directly on the three last RSSI measurements. The following calculations are made at time t where one of the APs receives an RSSI measurement: For the AP that receives the measurement, the corresponding PDF of the distance is calculated directly with (3.16). There are two solutions for f D :. If an RSSI measurement has been performed before on this AP, f D is the updated PDF of the distance with the last RSSI measurement with Eq: This PDF represents information about the location. The human movement model says that the Device can be at certain positions with more probability than other positions. Combined with the present RSSI measurement, it is possible to compute a more precise PDF of the distance.. If there has not been any previous RSSI measurement, the assumption (3.17) is made. For the two other APs, the PDF of the distance is updated with Eq: 3.40 by taking for t the distance between the time t and the last RSSI measurement of the considered AP. This way, there is less chance that the PDF used for the triangulation might be false. With the three resulting PDFs of the distance, triangulation is performed with the formula Eq: The Figure 4.24 summarizes what is done with the RSSI measurements. AP1 AP2 AP3 Updating Updating Updating t t t Localization Figure 4.24: Updating time scheme Simulation: To simulate this process, we should deduce a characteristic speed and a characteristic direction s char and θ char after two localizations with Eq: 3.33and 80

81 CHAPTER 4. SIMULATIONS Eq: To simplify that, we suppose that we already know s char. Moreover, we do not consider any calculation about θ char because it takes too much computation power and we do not have enough power to compute easily statistical values. By taking a value of 2m/s for s char we obtain the following results: Number Of MeanDistance(m)MeanRadius Percentage of MeanRadius Percentage of Locations Halfbeam(m) good location in the halfbeam Fifthbeam(m) good location in the fifthbeam Table 4.10: Results for a moving Device Conclusion By taking into account the movement model, the result can really be improved. In this simulation part, a really reasonable localization with almost 75% of success in a radius of 3m has been found Movement Prediction By lack of computer-resources, no test has been performed on the movement prediction yet. However, the basics will be summed up to show how globally our program works for this step and the philosophy of our reasoning. A moving device is considered in this part. The parameters used are those defined in Section 3.3: the characteristic speed the characteristic direction To compute both of them, locations at time t and t tp are needed. This is possible because the location step has already been performed. Indeed, with two locations, it is possible to calculate the characteristic direction and the characteristic speed of a Device with equations 3.33 and With the characteristic speed and the characteristic direction, two PDFs can be calculated: the PDF of the speed with equation (3.30) the PDF of the direction with equation (3.32) Then, these two elements enable to compute the PDF of the movement prediction. The movement prediction step is mainly based on previous locations of a Device. Thus, if an error occurs in the localization step, it will pass it to the movement prediction computation. Although the accuracy and the reliability of our program 81

82 CHAPTER 4. SIMULATIONS have not been tested yet, this part cannot be eluded. Indeed, as we will notice later, movement prediction is really relevant for the implementation of the handover procedure Handover General Scenario In this chapter we will first focus on Handover without Location information and then on Handover with Location information and Movement Prediction. This will enable us to compare both cases. As seen in the section 4.1 of this chapter, we already have defined a scenario. Three Access Points (APs) are placed on the ceiling of a room. We have just seen that it was preferable to put them in the center of the room. However, for more commodities, they have been put in the corners in this chapter (Figure 4.25): this has no impact on the handover study. Each AP is connected by wire to a Handover Agent (HA), a server whose principal functions are: Without location To transmit the clock and the BD_ADRR of the Device to all the APs To concentrate the RSSI measurements of the APs With location and movement prediction To transmit the clock and the BD_ADRR of the Device to all APs To receive the RSSI measurements of the APs and analyse it to predict next location To maintain the database of the room (location of the APs, boundary regions, location of the Device,etc.) Timers defined during Paging procedure To evaluate the time for the Paging procedure in both cases (with and without location), several timers have been defined in the section A value has been assigned for each one: Page timeout (pageto) = 1.28s, which is the average time of Paging process [8] and equal to the Time Between Page Scans (see Appendix B). Page response (pagerespto) = 8 slots for the Device and for the APs. New connection timeout (newconnectionto) = 32 timeslots. Tpoll = 40 slots 82

83 CHAPTER 4. SIMULATIONS Handover Agent Wired network AP2 AP1 Device AP3 Figure 4.25: One of the possible configurations of the room (view from the top) Handover without location information In this case the accurate location of the Device is not known. In fact we only have Cell ID location (see Section 2.2.1). Handover is carried on the basis of Received Signal Strength Indicator (RSSI). Assumptions The Device has already been connected to the AP1 with Inquiry and Paging procedures. The first Inquiry for connecting the Device has given to each AP the FHS packet of the Device through the HA. A Threshold is defined to set the area boundary of each AP (Figure 4.26). Information transmission between the APs and the HA are immediate The Device cannot be in STANDBY state. It will only be either in CON- NECT state or PAGE SCAN state. The HA initiates the AP for the Paging procedure in an increasing order (first AP2 and then AP3). The origin of the time for Paging is taken as soon as the Device is in Hold mode (150ms after the Hold Request). The Paging timers are taken into account. The Paging procedure always succeeds before the Page timeout if the Device is in the range of the considered AP. 83

84 CHAPTER 4. SIMULATIONS Handover Agent Wired network AP2 Transit 1-2 Transit 2-1 AP1 Transit 3-2 Device Transit 2-3 AP3 Transit 1-3 Transit 3-1 Figure 4.26: Possible area transitions performed by the Device in the case of Handover without location Handover procedure The principle of this handover method is that, when the Device needs to change of AP, the APs try to page the Device in an increasing order because the Device cannot receive two page packets (DAC packet) in the same time slot. 1. First the Device is bound to the AP1. This AP1 is known as the Active Access Point and is called A(AP). Two cases have to be considered: (a) Movement from AP1 to AP2 i. The Device is moving from AP1 region to the AP2 region. When the RSSI value is equal or below to the defined threshold, the AP1 forces the Device into PAGE SCAN state. At the same time, the AP1 notifies the HA for a handover procedure. ii. HA notifies the AP2 for a Paging procedure: the AP2 enters the PAGE state. iii. As the Device is in the coverage area of the AP2, the Paging process can happen. When the Device is connected to the AP2, it deletes the link with the AP1. (b) Movement from AP1 to AP3 i. The Device is moving from AP1 region to the AP3 region. When the RSSI value is equal or below to the defined threshold, the AP1 forces the Device into PAGE SCAN state. At the same time, the AP1 notifies the HA for a handover procedure. 84

85 CHAPTER 4. SIMULATIONS ii. HA notifies the AP2 for a Paging procedure: the AP2 enters the PAGE state. iii. As the Device is not in the coverage area of the AP2, AP2 will not receive any response. After the Page timeout, the AP2 informs the HA that the Device is not in its area. Then, the HA notifies the AP3 for a Paging procedure: the AP3 enters the PAGE state. iv. As the Device is in the coverage area of the AP3, the Paging process can happen. When the Device is connected to the AP3, it deletes the link with theap1. Time for Paging There is a difference during the procedure between the two cases: (a) Movement 1-2. The Device is in the coverage area of the AP2. The time for the Paging procedure is between 4 Time Slot(TS)= 2.5 ms and Page timeout = 2048 TS = 1280 ms (that means 64 scans)(figure 4.27). f(k) f(k+1) f (k) f(k+1) f (k+1) Tx AP3 DAC Page FHS t Rx t Tx Device DAC DAC t Rx 625us 0 2.5ms Figure 4.27: Time for Paging in Handover without Location t (b) Movement 1-3. The Device is in the coverage area of the AP3, but the Handover Agent notifies the AP2 to enter in PAGE state first.(figure 4.28) The Device is not in the coverage area of the AP2. Thus, the HA notifies the AP3 for entering PAGE state 1.28s later. In this case, the time for the Paging procedure is between: (2048+4) TS (=1282,5 ms) and 2*Page timeout = 4096 TS = 2560 ms. Remark: The more neighboring APs you have, the longer can be the Paging duration. Self configurability Because the choice of the increasing order for the choice of the AP by the HA is 85

86 CHAPTER 4. SIMULATIONS f(k) f(k+1) f (k) f(k+1) f(k+3) f(k+31)f(k+32) Tx AP2 DAC Page DAC Page DAC Page t Rx t Tx t Device Rx 0 625us 1280ms t Figure 4.28: Time for Paging without Location (1) f(k) f(k+1) f (k) f(k+1) f (k+1) Tx AP3 DAC Page FHS t Rx t Tx Device DAC DAC t Rx 625us 1280ms ms Figure 4.29: Time for Paging without Location (2) t 86

87 CHAPTER 4. SIMULATIONS not very pertinent, self-configurability has been developed. It will try to enhance the Paging duration. This method consists in defining an order of priority which allows to launch the Inquiry procedure from the most probable AP where the Device has more probability to be located (see section probability tree ). To do that, transitions are defined. Each time that the Device crosses a transit boundary, this one is loaded into memory which name is order table. Movement 1-2 The order table gives the following results: Transit 1-2 happens more than transit 1-3: time for Paging= [2.5; 1280]. Transit 1-3 happens more than transit 1-2: time for Paging= [1282.5; 2560]. Movement 1-3 The order table gives the following results: Transit 1-2 happens more than transit 1-3: time for Paging= [1282.5;2560]. Transit 1-3 happens more than transit 1-2: time for Paging= [2.5;1280]. Summary of the total time for Paging procedure: Movement 1-2 Movement 1-3 Paging Min Time Max Time Min Time Max Time Without self configurability 2.5ms 1280 ms ms 2560 ms Transit 1-2 > Transit ms 1280 ms ms 2560 ms Transit 1-3 > Transit ms 2560 ms 2.5ms 1280 ms Table 4.11: Total time for Paging procedure with and without Self Configurability When the probability of the movement from the AP1 to the AP where the Device has moved is bigger than the other one, time taken by Paging procedure is either equal or diminished. For the less probable case, the time is equal or increased. Thus, the self configurability has been set to diminish the time for Paging if the movement studied has the highest probability. Drawback of the Handover without Location PRSSI measurements are done each time the Device receives an inquiry from one of the APs. This procedure takes generally between 1 and 4 seconds. That could be a disadvantage in some case as demonstrated in Figure 4.30 and Figure Following parameters are defined: 87

88 CHAPTER 4. SIMULATIONS AP a td+tp T0 AP b tl Figure 4.30: Loss of quality of the signal for Handover without Location T 0 is the moment of the first inquiry of APa. t D is the time between the first and the second Inquiry of the APa. t P is the duration between the moment the APa warns the Handover Agent and the moment the APb finds the Device in range. t L is the total time during which the quality of the connection decreased. Explanation: At time T 0, APa launches an Inquiry to know the RSSI that the Device calculates. Thus, it knows that the Device is still in its range. After the delay t D, APa launches a new Inquiry, trying to know the RSSI of the Device and more precisely if this RSSI is under the defined Threshold. In fact the Device has moved out of its coverage area. Thus, APa warns the Handover Agent and asks the other APs to alternatively search for the Device (according to the order defined in the order table ). Right from the moment that the APb launches the Paging, it will find the Device. But during the time t L the connection state between the Device and the APa has not been as good as it could have been if the APb had recovered the Device. In the worse case, this time t L can reach 4 seconds (considering as maximum time of the Inquiry procedure + Paging procedure of several Access Points). In the case of the Figure 4.30 this phenomena is not very relevant because the connection is kept by APa during t L. But the case of the Figure 4.31 is more problematic since during the time t L, the connection with APa is totally lost. Conclusion for Handover without Location We notice that the duration of the Paging procedure depends on the digit (order) 88

89 CHAPTER 4. SIMULATIONS AP b AP a td+tp T0 tl Figure 4.31: Loss of the signal for Handover without Location affected to the AP. A system of self-configurability has been implemented tempting to optimize this digit allocation. However there always will be cases where the Paging duration will be higher than 1280 ms (if the Device moves in one direction which is not the most probable one). In our scenario, the problems of connection, where there is a loss of the connection, are not relevant. But, as we have seen, they could be in the case of a room change. Thus, Handover without Location has some shortcomings Handover with location and movement prediction In this case the accurate location of the Device is known and the next location is estimated by the HA. As the next location is estimated, there is no need to consider two different moves: they are symmetrical considering Paging duration. Handover is carried on the basis of Received Signal Strength Indicator (RSSI). We consider only the movement from the AP1 to the AP2. Handover is carried on the basis of Received Signal Strength Indicator (RSSI). Assumptions The Device is already connected to the AP1 with Inquiry and Paging procedures. The first Inquiry for connecting the Device has given to the APs the FHS packet of the Device through the HA. Each point of the boundaries (Figure 4.32) is defined as being at equal distance from the 2 closest Access Points. So the quality of the connection on each side of the boundary is the best we can have. Information transmission between the APs and the HA are immediate. The Device cannot be in STANBY state. It will only be either in CONNECT state or PAGE SCAN state. The HA initiates the AP for the Paging procedure in an increasing order (first AP2 and then AP3). 89

90 CHAPTER 4. SIMULATIONS Figure 4.32: Example of coverage areas of the Three Access Points Handover Agent Wired network AP2 AP1 Area 2 Device Area 1 Area 3 AP3 Figure 4.33: Sectorization of the room 90