| Wireless sensor networks monitor the environment by measuring various physical phenomena. These measurements, and consequently the quantities computed from them, frequently contain a certain amount of error. One of the basic tasks that handles inaccurate measurements is location discovery, a process where nodes in the network estimate their actual locations. We have focused on uncertainties in location discovery as the primary target of our study because many sensor network applications are dependent on location information, and therefore uncertainties in location information can affect performances of such applications.;The technical highlight of our work is a statistically validated parameterized model of location errors that can be used to evaluate the impact of a location discovery algorithm on subsequent tasks. We propose estimators of the unknown parameters of the location error, derived from inconsistencies between the final location estimates and the initial distance measurements. We demonstrate that the numerical results in applications that require locations from the majority of nodes are more impacted by the location error, while the impact of location error on the results is more predictable. Finally, we describe how applications can organize resource consumption to achieve optimal results in the presence of the estimated magnitude of error.;The second major topic we focus on in this work is the extent to which malicious beacons can reduce the accuracy of the location estimates. We examine the optimal strategies for malicious beacons under varying level of information that the malicious beacons have about distribution of the distance measurement error, deployed outlier detection algorithms in the network, and the number and the distribution of beacons in the network. Then, we propose an outlier generation algorithm that exploits flip ambiguities ("bad geometries") when appropriate, or moves the location estimate away from the original estimate along the direction of the least resistant path. We also show that outlier detection algorithms based on linear models do not perform well in the presence of flip ambiguities ("bad geometries"), and we propose an improved adjustable algorithm that detects flip ambiguities.;Finally, we show the importance of error models in validating performances of location-dependent algorithms in sensor networks, and we derive an error model based on real acoustic data measurements. |
