| We address several well-known and new canonical location discovery (localization) problems in wireless networks, including engineering change-based node addition, and design and operation of location discovery infrastructure for mobile users. All problems are solved using coordinated development of data-driven statistical models, and combinatorial and continuous optimization techniques. The models are built in such a way that they are amenable for consequent optimization (e.g. linear, or convex, or isotonic). The optimization procedures are conducted such that they are robust against outliers and statistical uncertainties.;From modeling point of view, our goal is to identify factors that predominantly influence the accuracy of location discovery and to establish relationships between different properties of the network and the corresponding location error. For this task, we use a combination of nonparametric statistical methods and combinatorial optimization techniques to construct and validate three types of models: measurement error models, terrain models and localization error models.;From optimization point of view, we formulate the location discovery problem as an instance of (integer) linear or nonlinear program due to its computational complexity. However, if a related subproblem can be solved optimally, then we linearize or approximate the system of equations and use linear or convex programming techniques to solve them. Our techniques are realized both as centralized and localized algorithms; and our models and algorithms are built and evaluated using actual data collected from deployed networks. |
