| Decentralized information processing systems, such as wireless sensor networks, facilitate the acquisition, processing and dissemination of information. With the phenomenal growth in the development of digital hardware and wireless information technologies, the efficient handling of data in such systems has become of prime concern. This thesis deals with several inter-related problems associated with the processing of information in wireless peer-to-peer networks. It presents a collection of analysis, algorithmic techniques and results that are designed to optimize performance in such decentralized systems.; Sensor network architectures are analyzed as the starting point of the study. Current frameworks for unified, probabilistic, information processing approaches to data fusion networks for heterogeneous, multi-sensor networks are reviewed. It is observed that a likelihood metric aggregation architecture, based on the Bayesian approach, is a convenient unifying technique for decentralized organization and interpretation of information. The need for fundamental performance limits for such decentralized networks is then discussed, and some bounds are derived. In particular, the rate distortion region for a sensor fusion network with n-helper nodes in a Gaussian setting is described. The asymptotic delay order for data packets in a random wireless network is also derived.; Next, some practical issues dealing with the administration of wireless resources in heterogeneous, multi-tiered ad hoc networks are discussed. A hybrid, gateway-based architecture for enabling range extended connectivity is proposed, and gateway trajectory control and network optimization algorithms for such an architecture are devised. Finally, the dependability of such networks is analyzed, and a distributed reliability algorithm is formulated that can be applied to optimize the dependability of decentralized, dynamic peer-to-peer networks. Specific fault tolerance algorithms for the gateway-based architecture are also proposed and investigated.; The overall goal is thus to develop a set of cross layer techniques that can be applied to quantify some of the basic performance limits for distributed information processing systems. |
