Dong Guo

May 2009
A New Statistical Localization Framework for Wireless Sensor Networks
Committee Members: Advisor: Ioannis Paschalidis, SE/ECE; Christos Cassandras, SE/ME; Thomas P.C. Little, ECE; Prakash Ishwar, ECE; Appointed Chair: Pirooz Vakili, SE/ME

Abstract: Determining the physical location of the nodes within a wireless sensor network (WSNET) enables a number of innovative services, including asset and personnel tracking, and locating nodes that report a critical event. Radio-Frequency (RF)-based localization techniques have drawn more attention since RF is the common denominator of all WSNET platforms. The low accuracy of deterministic and stochastic triangulation techniques, especially in indoor environments, motivates the work in this dissertation.

First, the localization problem is formulated as a multiple composite hypothesis testing problem. The use of the Generalized Likelihood Ratio Test (GLRT) is proposed and it is established that GLRT has desirable optimality properties (in a Neyman-Pearson sense). The application of the GLRT decision rule involves a number of parameters (thresholds); we develop the necessary theory and optimization methodology to tune these parameters and minimize the maximum probability of error. It is shown that a GLRT-based localization decision can be implemented in a distributed manner by appropriate in-networking processing. Two hypothesis profiling techniques are introduced to associate a family of probability density functions (pdfs) to each hypothetical region: a heuristic method and a landmark-based linear interpolation technique.

Second, a movement detection problem is formulated which can determine whether a sensor node has moved from its most recent position. This problem is again formulated as a composite hypothesis problem for which the Generalized Hoeffding Test (GHT) is optimal. The optimal threshold and a probabilistic performance guarantee are provided.

Combing localization and movement detection, a two-tier tracking system is designed to monitor sensor nodes where the computationally cheaper movement detection tier activates the localization tier. In addition, the problem of optimally placing clusterheads is formulated and an efficient algorithm is proposed.

To validate the proposed approach an experimental testbed platform has been built. The experimental results show that the system’s accuracy is on the same order of magnitude as the mean area of a region.

Finally, a formation detection problem is considered with the objective of identifying the relative position — formation —  of a set of sensor nodes. This problem is treated in a similar composite hypothesis testing framework and has applications in human posture detection and robot swarm formation detection.