Throughput Scaling and Data Gathering in Wireless Networks.

This dissertation investigates three problems associated with wireless networks. First, throughput scaling in extended random access communication networks is examined. In these networks, the number of nodes and the network area increase in such a way that the density of nodes remains constant. Franceschetti et al. (2008) have shown an improvement in the per-node attainable throughput from the scaling demonstrated in the seminal paper of Gupta-Kumar (2000). This dissertation explores the dependence of this result on the new features introduced by Franceschetti: a capacity based link-rate model, hierarchical routing, and percolation based route construction. By creating a non-hierarchical system attaining the same improvement, it is concluded that the improved scaling is due to percolation-based routing, which enables shorter hops and, consequently, less interference. Next, the reliability-efficiency tradeoff in wireless sensor networks (WSNs) is examined. Distributed lossless source coding in WSNs offers the potential for sizable reductions in coding rates. However, joint decoding makes these schemes highly sensitive to encoder failures. To improve reliability, this dissertation considers schemes where each source is decoded using only a subset of the encoders. In comparison to previous schemes, called rigid, a new class of schemes, called flexible, is introduced. To analyze performance of these schemes, the Slepian-Wolf lossless source-coding theorem is extended to the case where encoders may fail. For an underlying field that is one-dimensional Gauss-Markov, numerical results show flexible schemes achieving significant performance gains over rigid. Finally, sensor placement and real time data gathering in WSNs is investigated. Here, each sensor observes an underlying random process at its location and communicates its observations to the collector, which then estimates the process over the entire network region. For a one-dimensional Markov process with exponential autocorrelation, it is shown that uniform placement of sensors is optimal in the mean square error sense. Next, in order to demonstrate effects of communication constraints, a simple algorithm for data gathering in a one-dimensional network is proposed. For a stationary process with a separable correlation model, the optimal density of sensors over a fixed area is shown to increase with increasing temporal correlation and/or decreasing spatial correlation.Ph.D.Electrical Engineering: SystemsUniversity of Michigan, Horace H. Rackham School of Graduate Studieshttp://deepblue.lib.umich.edu/bitstream/2027.42/86467/1/awlok_1.pd