Collaborative Signal and Information Processing: An Information-Directed Approach

This chapter focuses on the collaborative signal and information processing (CSIP) issues in designing and analyzing sensor network applications. The complexity of the applications, the collaborative nature of the algorithms, and the plurality and diversity of resource constraints demand novel ways to construct, configure, test, and debug the system, especially the software. A decentralized sensing system is inherently more robust against individual sensor node or link failures, because of redundancy in the network. In a typical sensor network, each sensor node operates untethered and has a microprocessor and limited amount of memory for signal processing and task scheduling. A sensor network is designed to perform a set of high-level information processing tasks, such as detection, tracking, or classification. Directed-diffusion routes sensor data in a network to minimize communication distance between data sources and data sinks. Tracking is an essential capability in many sensor network applications, and is an excellent vehicle to study information organization problems in CSIP.