Robust and locally-optimum decentralized detection with censoring sensors

In this paper, we examine decentralized detection problems in which a send/no-send rate (censoring rate) on the sensor decisions replaces the communication constraint of D-level sensor decisions in canonical parallel decentralized detection problems. Rago et. al. introduced the censoring problem to accurately address the energy consumption due to infrequent communication, especially when sensor decisions must be accompanied by other side information. We extend the censoring idea by considering the situation where the signal and noise distributions are not perfectly known, which occurs in many practical scenarios. Both a robust formulation of the censoring problem, and a locally-optimum formulation of the censoring problem are considered. We find conditions under which the robust censoring problem can be solved by designing for the least-favorable distributions from the uncertainty classes. For the locally-optimum formulation, we find that the censoring scheme produces a test with the same form as the optimal Neyman-Pearson test.