Wireless sensor networks (WSNs) have the potential to significantly impact many aspects of our everyday life and are anticipated to be critically important for environmental monitoring, health monitoring, surveillance, military, and homeland security applications. Distributed parameter estimation serves as one of the main challenges in WSNs, where spatially distributed battery-powered sensors are deployed over a sensing field to monitor physical or environmental conditions. This research will investigate distributed estimation of an unknown vector with a general observation model, when the transmission strategy is digital.
The following three problems will tackled: (1) Bayesian Cramer-Rao bound (CRB) and linear estimators for hard-decoding and soft-decoding link layer designs. The Bayesian CRB will be derived and linear estimators will be developed. In hard-decoding design, the fusion center (FC) uses the recovered transmitted discrete messages for estimation, whereas in soft-decoding design, the FC uses directly the continuous-valued channel outputs for estimation. The impacts of the observation model, fading and communication channel noises and choice of modulation on the optimization solutions will be studied, and the existing tradeoffs between transmit power, rate, and estimation accuracy will be explored. Contrasting the power-rate-distortion regions will enable the investigator to quantify the performance improvement provided by soft-decoding (compared with hard-decoding) link layer design; (2) channel estimation problems in distributed vector estimation. The influence of wireless channel estimation errors on the distributed signal processing designs will be studied. Also, the investigator will study how the combined effects of channel estimation errors and energy cost of transmitting training symbols further limit the estimation accuracy; and, (3) distributed vector estimation in a cluster-based hierarchical network architecture with digital transmission. The existing tradeoffs between transmit power allocation among the clusters and the estimation accuracy will be explored.
