Entropy-based Distributed Fault-tolerant Event Boundary Detection Algorithm in Wireless Sensor Networks

• Main Authors: Yu-Ting Liu, 劉昱庭
• Other Authors: Wen, Ouyang
• Format: Others
• Language: zh-TW
• Published: 2009
• Online Access: http://ndltd.ncl.edu.tw/handle/20367876122430890397

碩士 === 中華大學 === 資訊工程學系(所) === 97 === Wireless sensor networks(WSN), comprising of huge number of sensors to monitor environments, observe the occurrences of interesting events such as forest fire, chemical spills, etc. Then, the sensors collaboratively compute and report the results to the back-end system via wireless communication. Since the occurrences of event may cover a substantially large area, it consumes less energy if only those sensors near the event boundary are allowed to alert the event back to the base station. Our goal is to successfully identify the sensors closed to the event boundary when faulty sensors exist in wireless sensor networks. Some previous works proposed solutions based on 0/1 decision predicate, others presented their methods using scalar sensing values. The formers use only 0 or 1 as the sensing values which may ignore local variations. Others rely on centralized mechanism to retrieve the fault probability related threshold or it may not be applied to dynamic situation efficiently. Therefore, we propose an entropy-based distributed fault-tolerant event boundary detection algorithm (EDFEBD). To the best of our knowledge, we are the first one that proposes system entropy concept from thermodynamics and information theory to solve this problem. Our algorithm accepts scalar values as inputs and consists of two phases: distributed faulty sensor detection and recovery (DFDR) phase and distributed event boundary detection (DEBD) phase. It’s distributed and localized and thus are scalable. Simulation results show that we have enhanced 15% performance on degree of fitting than previous approach even in the WSN with 0/1 decision predicate.