Efficient Change-Point Detection Over Fully Decentralized Wireless Networks With Low Communication Rate

Change-point detection (CPD), also known as quickest detection, aims to detect an abrupt change in the environment according to its observations as quickly as possible. Its applications in networks, such as wireless sensor networks and vehicular networks, motivate the study of distributed CPD, also known as multi-sensor CPD. In this context, multiple sensors monitor the environment via observations, but only an unknown subset of sensors is affected by the abrupt change. In this paper, we propose a novel distributed CPD algorithm, which simultaneously enjoys the advantages of being free of a fusion center, having low energy cost, being robust to the number of affected sensors, and suitable for a general data model. In particular, the proposed algorithm incorporates the thresholding technique into the a decentralized variant of cumulative sum (CUSUM) procedure. Sensors exchange information only when a designed local statistic exceeds a threshold, thereby reducing the number of communications and the energy cost. In addition, the proposed algorithm approximates the sum-CUSUM procedure and achieves robustness to the number of affected sensors. Theoretical analysis guarantees that the proposed algorithm achieves an exponential relationship between two detection performance criteria, the average running length (ARL), which measures the false alarm rate, and the expected detection delay (EDD). This result holds regardless of the number of the affected sensors. Furthermore, the thresholding operation only has a mild influence on the detection performance. Finally, we conduct simulations to verify the efficiency and effectiveness of the proposed algorithm. We also apply it to distributed source detection task in multiple-antenna wireless networks as a potential application.