Collaborative estimation in networks

Consider a network of nodes that are deployed to monitor a common phenomenon. In many cases, the network nodes need to estimate the common phenomenon from \noisy" observations they make. Although each node can independently obtain the estimate from its own observations, it can obtain a better estimate by communicating with the other nodes and by exploiting the interdependence between the observations at di erent nodes in estimation. In this dissertation, we study such a collaborative estimation problem in which a network of nodes, each indirectly observing an underlying source through \noisy" measurements, communicate with each other in order to form better estimates of the underlying source.Two primary constraints, complexity and communication, should be taken into account in the design of collaborative estimation algorithms. Although it is referableto have both low complexity and low communication, the limits on the estimation error performance can be studied by relaxing either or both of these constraints. In particular, signal processing and machine learning based approaches tend to focus on developing low complexity collaborative estimation algorithms with less strict attention to communication, while information theory focuses on characterizing the fundamental tradeo between communication rate and estimate performance, with less attention to complexity. Reconciling complexity with communications, modern practical coding theory can aid the design of collaborative estimation algorithms with low-complexity and low-communication.In this dissertation, we apply tools from aforementioned areas of study to develop collaborative estimation algorithms with low-complexity, low-communication and lowcomplexity/low-communication, and evaluate the estimation error performances.