Data reduction and gps-free node localization in wireless sensor networks

This thesis addresses the topics of data reduction via sampling in both central database environments, and wireless sensor networks, and GPS-free node localization in wireless sensor networks. The first contribution of this thesis is a deterministic sampling algorithm for sampling count data, which is common in data mining applications. We show that our algorithm creates more accurate and higher quality samples compared to previous work, and the samples it generates can be used as a surrogate for the original high volume data. Our second contribution is a deterministic weighted sampling algorithm that can be used as a new data aggregation method for wireless sensor network data. The aggregation algorithm shares similar ideas with our previous sampling algorithm. In order to adapt to the sensor network environment, however, we designed our algorithm to perform weighted sampling in a distributed manner. The weighted sampling design allows the algorithm to work with any arbitrary network topology, while the distributed design divides the sampling work equally on all the sensor nodes in the network and prevents any node from being a bottleneck (both with regards to CPU consumption, and communication). We show that our aggregation algorithm generates samples of better quality than previous algorithms, using far less energy. Our last contribution is two GPS-free node localization algorithms, termed GPS-free Directed Localization (GDL), and GPS & Compass-free Directed Localization (GCDL). The importance of localization is apparent in mobile wireless sensor networks, where the neighborhood changes frequently and knowledge about the neighbor positions is essential for performing additional tasks such as aggregation or coherent movement. Our algorithms perform localization without the need of Global Positioning System (GPS) or any other infrastructure (e.g., anchor points). These algorithms work with only local knowledge without using historical data, and exploit mobility to perform localization. The memoryless aspect of our algorithms avoids the accumulation error over time, which is essential in mobility scenarios where coherent movement of a swarm of nodes is required. We show that our algorithms do work even at high environmental noise levels, and keep a nice semi-rigid network formation in mobility scenarios.