Distributed classification under statistical dependence with application to automatic modulation classification

In this paper, we consider the distributed classification of discrete random signals in wireless sensor networks (WSNs). Observing the same random signal makes sensors' observations conditionally dependent which complicates the design of distributed classification systems. In the literature, this dependence has been ignored for simplicity although this may significantly affect the performance of the classification system. We derive the necessary conditions for the optimal decision rules at the sensors and the fusion center (FC) by introducing a “hidden” random variable. Furthermore, we introduce an iterative algorithm to search for the optimal decision rules. The proposed scheme is applied to a distributed Automatic Modulation Classification (AMC) problem. It is shown to attain superior performance in comparison with other approaches which disregard the inter-sensor dependence.