Efficient Wireless Location Estimation Through Simultaneous Localization and Mapping

Location estimation for IEEE 802.11 (aka Wi-Fi) networks typically relies on radio fingerprinting which, in turn, requires a radio map, usually acquired through empirical measurements via a process known as a site survey (the less used alternative being a simulated radio model of the environment). The site survey is a lengthy, tedious process, involving a lot of measurements over extended periods of time. This may pose as a barrier to the adoption of location estimation technologies and current research does not adequately reduce this required effort. This research aims to reduce the overall effort for deploying location estimation systems by eliminating the site survey and, to that end, presents a method to perform location estimation without requiring an initial radio map. It was theorized that, given some basic information about the environment (accessible areas, locations of access points (APs)), the initial radio map is not necessary. Two approaches were attempted, one based on a deterministic algorithm and another on a probabilistic one. The initial deterministic algorithm modeled the user and radio measurements close to each AP to obtain small maps in the vicinity of each AP. These small maps were then stitched together to form a complete map of the environment. This approach eventually proved too error-prone and limiting as it still required extensive calibration and site-specific measurements. However, much was learnt from it and that led to the development of the next algorithm. The subsequent probabilistic approach uses a novel reformulation of classic robotic simultaneous localization and mapping (SLAM) algorithms to map the radio environment while providing location estimates. SLAM algorithms are typically used in autonomous robots and have never been applied to IEEE 802.11 location estimation, especially with human users. This probabilistic approach proved more successful and was able to provide reasonably accurate estimates with minimal initial information about the environment and no radio map. The author's earlier research [35] also highlighted how different radio characteristics of consumer hardware affected the radio measurements used for radio fingerprinting. This dissertation proposes an alternative way of viewing the radio map in terms of a height field. This technique enables the SLAM algorithm to operate without requiring information about the radio characteristics of the APs and mobile device. The main contributions of this research are: (1) The identification of the problem with existing IEEE 802.11 location estimation techniques that utilize radio fingerprinting, namely the time-consuming site survey. (2) The use and adaptation of classic SLAM algorithms from robotics to the context of IEEE 802.11 location estimation, eliminating the need for the site survey. (3) An alternate view of the radio map, obviating the need for the radio characterization of the transmitters and receivers. This is also key to the functioning of the Wi-Fi SLAM algorithm.