Information Flow Over Wireless Networks: A Deterministic Approach

SPEAKER: Salman Avestimehr, Ph.D. Candidate, Department of EECS, UC-BerkeleyABSTRACT How does information flow over wireless networks? Answer to this basic question is one of the most challenging problems in the field of wireless network information theory. From a practical point of view, the answer to this question will have a significant impact on the architectural design of future wireless systems. So far, the majority of research done in this area have been based on the classical additive Gaussian noise model for wireless channels. However, due to the complexity of this model, except for the simplest networks, the analysis of most other networks has been an open problem for many years.To make further progress, we develop a deterministic channel model which is analytically simpler than the Gaussian model, but at the same time captures the essential physical layer properties of the wireless medium: signal strength, superposition and broadcast. We will demonstrate how this model can be an effective tool to help visualize the flow of information and obtain intuitive insights in many challenging network scenarios. Furthermore, somewhat surprisingly, these deterministic results translate to good approximation for the Gaussian case. As an example, we apply this approach to cooperative wireless relay networks (with a single source node and a single sink node and arbitrary number of relay nodes to help with the communication), whose capacity even in the simplest case with only one relay is unsolved for more than 30 years. We first determine the capacity of deterministic relay networks. This result is a generalization of the max-flow min-cut theorem for wireline networks. Next we use the connections between the deterministic model and the Gaussian model to approximate the capacity of Gaussian relay networks within a constant number of bits, independent of the channel parameters.In addition, the deterministic model can also be used to replace other simplistic models, such as collision model, to capture some abstraction of the physical layer at higher layers of network design. This is an important step towards developing new networking algorithms that exploit the available degrees of freedom at the physical layer.BIO: Salman Avestimehr is presently a Ph.D. candidate advised by Prof. David Tse at UC Berkeley. He was a Vodafone fellow at UC Berkeley during 2003-2005. He received his B.Sc. degree (with honors) from Sharif University of Technology in 2003, and his M.Sc. degree from UC Berkeley in 2005, both in Electrical Engineering and Computer Sciences. His research interests include wireless communications and networks, information theory and signal processing.HOST: Professor Michael J. Neely, mjneely@usc.edu

Location: Hughes Aircraft Electrical Engineering Center (EEB) - 248

Audiences: Everyone Is Invited

Contact: Mayumi Thrasher