|dc.description.abstract||Alternative indoor localization methods have long been researched due to the unsuitability of GPS and cellular technology in indoor environments. Indoor localization systems find their applications in many areas such as targeted advertising, inventory management and tracking, navigation in museums, arenas, hospitals and in first responder and disaster situations like fire. Wireless localization systems have become very popular in recent years, especially the ones based on Wi-Fi due to its existence in almost any indoor environment today. Our indoor localization system is based on Wi-Fi (IEEE 802.11g) and can be installed on existing wireless devices with a simple software patch. This is very attractive since there are a hundreds of millions of wireless devices already in use and a hardware upgrade would be impractical. Our indoor localization system consists of a target node unaware of its location and several reference peer nodes which are location aware. Target nodes communicate opportunistically with each reference node to measure the Round-trip time (RTT) between the transmission of data frames and reception of acknowledgement frames. This RTT is measured at the driver layer of the OS kernel. These RTT measurements are used to predict the distance between the reference and target nodes by using a Statistical TOA ranging method which compares the measured RTT against stored reference databases collected during an offline stage. The comparison based on a statistical distance measured called Bhattacharyya
coefficient and the results of this new ranging method is fed to a position estimation algorithm based on Linear Matrix Inequality (LMI) to provide the position coordinates of the target node. This thesis describes this new indoor localization system followed by a thorough analysis of RTT measurements made in different indoor environments such as a hallway and multiple rooms. We compare the measured RTT data with the packet timings involved in the IEEE 802.11g standard and introduce our novel idea of Statistical TOA ranging.
We discuss how we can build reference databases effectively by using Bhattacharyya coefficient and finally, we provide some real world experimental results of the ranging method along with the Wireshark view of the DATA and ACK frames involved. The real world experiments have shown that the precision of Statistical TOA ranging is 10 metres, i.e. there is a significant difference between RTT distribution for distances of 10 metres and it is extremely difficult to predict ranges less than 10 metres accurately. Hence, our Statistical TOA ranging predicts if the range is either 10 m, 20 m or 30 m. If the actual range is say 14 m, Statistical TOA ranging result is 10 m since it reduces the overall error. The position estimation stage then constructs LMI equations by incorporating ranging errors. From our real world, real time experiments, we find the average error to be about 3.08 metres and percentage of accurate predictions is about 71%.||en_US