Sensor networks gain their power from the cooperative action of the nodes in data gathering, processing and communication. In this way a wide variety of physical phenomena can be studied at spatial scales that were previously impractical. In deploying a large-scale network questions of optimization of resources arise, and in particular to limit the number of nodes and their duty cycle of use. Questions naturally arise concerning what collections of nodes should cooperate so that some user-defined quality of service requirements are met, how they should cooperate, and what fundamental limits apply to this cooperative activity. Such questions fall in the domain of network information theory. Unfortunately, the usual information theoretic criteria of optimality lead to the conclusion that the entire network should be engaged in some form of cooperation for every event, and results in bounds that are weak. A challenge for information theorists will thus be to frame research questions that reflect problem constraints in such a fashion that there is sufficient realism for the theory to provide guidance in systems design but yet at least result in tractable optimizations or computation of performance bounds. The talk is organized as follows. First, the conventional set of constraints that frame a large majority of the information theoretic studies now being undertaken is presented. Next it is argued that practical experimental deployments indicate that this constraint set is not representative of a large number of applications, and consequently a richer set of questions can be pursued by the community.
Audio (MP3 File, Podcast Ready)
Copyright. All Rights Reserved.