1 Location Estimation in ZigBee Network Based on Fingerprinting Department of Computer Science and Information Engineering National Cheng Kung University, Taiwan R.O.C. Authors ： Qingming Yao, Fei-Yue Wang, Hui Gao, Kunfeng Wang and Hongxia Zhao Publisher ： Vehicular Electronics and Safety, ICVES Present ： Yu-Tso Chen Date ： November, 5, 2008

2 Outline 1. Introduction 2. Research Methodology 3. Implement and Results 4. Conclusions and Future Work

3 Introduction Transform ordinary environments into intelligent spaces Context refers to the physical (position, time, weather) and social situation (work or leisure place) As location becomes one of the most import contexts, location-aware computing is a recent interesting research area

4 WSN Location Estimation System Authors implement the location estimation system adopting ZigBee based network Advantage ： short range, low data rate, low power consumption and low cost network technology easily constructing ad-hoc, mesh networks

5 ZigBee Network RSSI (Received Signal Strength Indication) is the basic function we use to form fingerprinting and measure data only the Local Location cluster is used

6 System Methodology (1/2) Beacons are fixed in several points evenly to make sure that mobile station (MS) can receive n (3 to 5 as usual) points’ radio signal at each location MS records and processes the RSS vector and then searches the fingerprinting database to find some fingerprinting which makes the algorithm criterion maximum

7 System Methodology (2/2) Oxy =(o 1 xy, o 2 xy,..., o n xy ) T is the observed RSS vector from beacons at location L xy RSSI = -( 10n d + A) 10 log Fij =(f 1 ij, f 2 ij,..., f n ij ) T is average RSS of location Fingerprinting database is constructed by a process of offline training

8 Build the Histogram Beacon k at location Lij is L =[l k 0 ij,..., l k M−1 ij ] L enables computing the histogram h k ij of signal strengths for each beacon indexed k

9 Location Estimation Algorithm Practical environment, the radio channel is of noisy characteristics The observation Oxy deviates significantly from Fij Map the online observed data Oxy to some physical Lxy We applied a probabilistic approach using Bayesian inference

10 Estimation algorithm’s target The estimation algorithm’s target is to find a location Lij that makes the probability P(Lij |Oxy) maximized

11 Estimation algorithm’s target (cont.) Conditional probability P(Oxy|Lij) is the likelihood of Oxy occurring in the training phase of Lij P(Lij) is the prior probability of location Lij being the correct position (uniformly distributed) CSMA/CA mechanism ensures the signal from different beacons independent from each other joint probability distribution => marginal probability distributions,where

12 IMPLEMENT AND RESULTS ZigBee module TI’s single-chip 2.4 GHz IEEE compliant RF transceiver CC2420 Fxed on ceiling usually (Beacon)

13 Layout of Experimentation Office room dimensions of 7.2m×9m×2.6m All the beacons are fixed on the ceiling Calibration points where the mobile stations’ signal strength was collected are denoted by the gray square

14

15 Signal Statistical Character Multi-path fading and people’s activities lead the RSSI fluctuating

16 Two error distances to evaluate the accuracy Physical space’s Euclidian distance ： Singal space’s Euclidian distance between Oxy and Lxy ：

17 Short-term Measurement Experimentation was designed as collecting RSSI of beacon3 at location L1 and L2

18 Long-term Measurement

19 Two Clusters with Frequency of RSSI with Two Beacons Investigate how the pattern of fingerprinting at different locations effects the location separation frequency of occurrence of each sample pattern

20 Mean RSSI in Calibration Points Fluctuate frequently

21 Conclusion System can triangulate the location within 70% accuracy with the tolerance of 0.5 meters that is quite encouraging.

22 Thanks

23 Existed Location Systems AT&T’s Active Badge diffuse infrared technology Disadvantage - difficulty with fluorescent lighting or direct sunlight Active Bats Ultrasound time-of-flight lateration technique Disadvantage - requires large scale deployment and high cost

24 Existed Location Systems (cont.) Cricket MIT complemented the Active Bats by using a radio frequency control signal Disadvantage - centralized management, and mobile receivers have heavy computational and power burden All the three systems is that they only provide light-of-sight (LOS) location estimation

25 Offline Training Phase The location fingerprinting is collected at each point of the 30 calibration points The probabilistic distributions of four directions are obtained by (1)

26 Online Estimation Phase measure and average the RSSI from beacons The average RSSI forms the observation tuple Oxy = (o 1 xy,o 2 xy,..., o n xy ) T and be applied in (4) (5) (6) to triangulate location Lij Search the fingerprinting database which stores the prior probability h k ij (ζ) to find the (i, j) which makes P(Lij |Oxy) maximized

27 Conclusion System can triangulate the location within 70% accuracy with the tolerance of 0.5 meters that is quite encouraging.