Fusing Censored Dependent Data for Distributed Detection

In this paper, we consider a distributed detection problem for a censoring sensor network where each sensor's communication rate is significantly reduced by transmitting only “informative” observations to the Fusion Center (FC), and censoring those deemed “uninformative”. While the independence of data from censoring sensors is often assumed in previous research, we explore spatial dependence among observations. Our focus is on designing the fusion rule under the Neyman-Pearson (NP) framework that takes into account the spatial dependence among observations. Two transmission scenarios are considered: one where uncensored observations are transmitted directly to the FC, and second where they are first quantized and then transmitted to further improve transmission efficiency. Copula-based Generalized Likelihood Ratio Test (GLRT) for censored data is proposed with both continuous and discrete messages received at the FC corresponding to different transmission strategies. We address the computational issues of the copula-based GLRTs involving multidimensional integrals by presenting more efficient fusion rules, based on the key idea of injecting controlled noise at the FC before fusion. Although, the signal-to-noise ratio (SNR) is reduced by introducing controlled noise at the receiver, simulation results demonstrate that the resulting noise-aided fusion approach based on adding artificial noise performs very closely to the exact copula-based GLRTs. Copula-based GLRTs and their noise-aided counterparts by exploiting the spatial dependence greatly improve detection performance compared with the fusion rule under independence assumption.