Decentralized Detection in Clustered Sensor Networks.

We investigate decentralized detection in clustered sensor networks with hierarchical multi-level fusion. We focus on simple majority-like fusion strategies, leading to closed-form analytical performance evaluation. The sensor nodes observe a binary phenomenon and transmit their own data to an access point (AP), possibly through intermediate fusion centers (FCs). We investigate the impact of uniform and nonuniform clustering on the system performance, evaluated in terms of probability of decision error on the phenomenon status at the AP. Our results show that, under a majority-like fusion rule, uniform clustering leads to the minimum performance degradation, which depends only on the number of decision levels rather than on the specific clustered topology. We then extend our approach, taking into account the impact of spatial variations of the phenomenon, noisy communication links, and weighed fusion rules. Finally the proposed distributed detection schemes are characterized with simulation and experimental results (relative to IEEE 802.15.4-based networks), which confirm the analytical predictions. [ABSTRACT FROM PUBLISHER]