The Information Theory of Sleeping Sensors and Idle Servers

While information theory has had a tremendous impact on the design of point-to-point, unidirectional communication systems, it has not yet similarly impacted the design of communication networks. This is primarily because (1) many simple network generalizations of the point-to-point coding problem are unsolved, and (2) information theory has generally ignored delay constraints, which are often crucial in networks. I will present recent work on each of these two problems.

I will first present a new bound on the fundamental limits of lossy data compression for multiple, separated encoders. This bound improves upon the best-known bound in the literature, in some cases strictly, and can be used to unify most of the work done on this problem during the last 30 years. I will demonstrate the efficacy of the bound by showing how it yields a new, conclusive result for an instance of the problem motivated by sensor networks in which the sensors sometimes "sleep" to conserve power.

I will then present the solution to an open problem of Arikan on the tradeoff between delay and error probability in channel coding for a single-server queue in the low-rate regime. The keys to the solution are identifying a metric for queues that parallels Hamming/Euclidean distance for conventional channels and observing that one need only consider "light-traffic" codewords, which keep the server idle most of the time.

As a UM undergraduate student Aaron Wagner won numerous awards including, the UM William J. Branstrom Freshman Prize, UM EECS Dept. Junior Scholar, Honorable Mention in the Student VLSI Design Contest, UM EECS Dept. Senior Scholar, UM College of Engineering Distinguished Achievement Award, UM Eight-Semester James B. Angell Scholar, and the UM College of Engineering Andrew A. Kucher Award for Undergraduate Research. He is just finishing his Ph.D. at the University of California, Berkeley working in communications networks.