Optimal decision rules for decentralized detection in resource-constrained wireless sensor networks

Wireless sensor networks (WSN) are envisioned to encompass wide applications ranging from battlefield surveillance, health care and telemedicine, to environmental and habitat monitoring and control. Technical challenges, however, must be addressed to make the vision come true. For instance, stringent resource constraints and unreliable wireless transmission medium make the traditional signal processing techniques not applicable to the emerging WSN domain.;In this dissertation, we investigate several important problems in the area of signal processing for resource constrained WSN systems. Specifically, we study optimal decentralized detection schemes under non-ideal communication channels and energy and bandwidth constraints. We first develop optimal fusion strategies for decentralized detection in multi-hop WSN. We begin with formulating the multi-hop detection and fusion structure for Rayleigh fading channels. Given the fusion structure, we then propose the likelihood ratio (LR) based optimal fusion rule and several alternative suboptimal fusion rules that are easier to implement with less complexity.;Furthermore, we address the problem of robust local quantizer design for decentralized detection under transmission losses due to channel outage or sensor failure. Considering binary local decisions, we propose two design schemes where the first does not require explicit channel information while the latter approach is essentially a channel-aware design. We show that for both cases the optimal local decision rules lead to a likelihood ratio test (LRT) at local sensors. The proposed designs protect the system detection performance from severe degradation in events of channel breakdowns or sensor node failures.;Finally, to accommodate the bandwidth constraint of the WSN, we study the decision rules for decentralized detection over multiple access channels. In particular, we establish that the optimal local decision rules at local sensors again amount to the LRT for both synchronized and asynchronous cases. For asynchronous sensor transmissions, we propose a fusion structure that consists of a RAKE receiver with square-law detector. This receiver structure enables the implementation of the optimal local decision rule and the fusion rule and compensates for the unknown delays and channel fading coefficients.