Quickest distributed detection via running consensus

Abstract

Running consensus is a recently proposed distributed strategy for fostering agreement among sensors of fully flat networks, by interleaving the two stages of measurements and node-to-node communications. Quickest detection is a well-established technique for discovering abrupt changes (if any) in the statistical distribution of the observed data. In this paper we tailor the running consensus idea to the quickest detection problem, to address change-detection issues in distributed inference systems with random and time-varying sensors' connections, in architectures without fusion center. Performance benchmarks are expressed in terms of detection delay and false alarm rate, for which closed form approximations are derived, yielding a simple analytical expression of the operational characteristic of the detector. The proposed system is tested on typical signal processing problems by means of numerical simulations that validate the theoretical analysis.