TABLE OF CONTENTS Bhaskar Krishnamachari, Congzhou Zhou and Baharak Shademan
Department of Electrical Engineering-Systems
University of Southern California, Los Angeles, CA, 90089
{bkrishna,congzhoz,shademan}@usc.edu
| Abstract | Wireless sensor networks are expected to be significantly resource-limited in most scenarios, particularly in terms of energy. In recent years, researchers have advocated and studied cross-layer design techniques as the primary methodology to leverage application-specificity for optimizing system performance. We argue that another powerful design principle is to make sensor networks autonomously learn application-specific information through sensor and network observations during the course of their operation, and use these to self-optimize system performance over time. We discuss several example application scenarios where such self-optimization can be used, including localization, data compression and querying. As an in-depth illustration, we then present details of LEQS (Learning-based Efficient Querying for Sensor networks), a novel distributed self-optimizing query mechanism. |
| Keywords: | Sensor networks, self-optimization, self-configuration, node localization. |
[*]This work has has been supported in part by grants from NSF (awards number 0325875, 0347621, and 0435505) and by an education grant from Intel.
Large networks of embedded sensor devices, each capable of a combination of computing, communication, sensing and even limited actuation, are being envisioned to provide an unprecedented fine-grained interface between the physical and virtual worlds. According to a recent National Research Council report, the use of such networks of embedded systems "could well dwarf previous milestones in the information revolution" [1]. The applications of sensor networks that are being investigated and developed range widely, including scientific environmental monitoring, civil...
