In the last decade there has been an increase in application areas for wireless sensor networks (WSNs), which can be attributed to the advances in the enabling sensor technology. These advances include integrated circuit miniaturization and mass-production of highly-reliable hardware for sensing, processing, and data storage at a lower cost. In many emerging applications, massive amounts of data are acquired by a large number of low-cost sensing devices. The design of signal processing algorithms for these WSNs, unlike in wireless networks designed for communications, face a different set of challenges due to resource constraints sensor nodes must adhere to. These include: (i) limited on-board memory for storage; (ii) limited energy source, typically based on irreplaceable battery cells; (iii) radios with limited transmission range; and (iv) stringent data rates either due to a need to save energy or due to limited radio-frequency bandwidth allocated to sensor networks.

This work addresses distributed data-reduction at sensor nodes using a combination of measurement-censoring and measurement quantization. The WSN is envisioned for decentralized estimation of either a vector of unknown parameters in a maximum likelihood framework, or, for decentralized estimation of a random signal using Bayesian optimality criteria. Early research effort in data-reduction methods involved using a centralized computation platform directing selection of the most informative data and focusing computational and communication resources toward the selected data only. Robustness against failure of the central computation unit, as well as the need for iterative data-selection and data-gathering in some applications (e.g., real-time navigation systems), motivates a rethinking of the centralized data-selection approach. Recently, research focus has been on collaborative signal processing in sensor neighborhoods for the data-reduction step. It is in this spirit that investigation of methods for sensor node-based data reduction is pursued, where each sensor decides whether and what to communicate.

The scope of this dissertation encompasses distributed algorithms for the measurement-reduction step and algorithms for estimation which are amenable to either in-network or fusion center-based implementation using a WSN. Clearly-defined optimality criteria are used as the foundation for development of algorithms for data reduction and estimation. Performance analysis is provided and corroborated using simulated and real-world test cases illustrating the potential of the novel methods.