Efficient and Fault-Tolerant Feature Extraction in Wireless Sensor Networks

Abstract

Abstract. We consider a canonical task in wireless sensor networksthe extraction of information about environmental features -and propose a multi-step solution that is fault-tolerant, self-organizing and energyefficient. We explicitly take into account the possibility of sensor measurement faults and study a distributed algorithm for detecting and correcting such faults, showing through theoretical analysis and simulation results that 85-95% of faults can be corrected using this algorithm even when as many as 10% of the nodes are faulty. We present a self-organizing algorithm which combines shortest-path routing mechanisms with leaderelection to permit nodes within each feature region to self-organize into routing clusters. These clusters are used in data aggregation schemes that we propose for feature extraction. We show that the best such aggregation scheme can result in an order-of-magnitude improvement in energy savings.

Effect of redundancy on the mean time to failure of wireless sensor networks

“…For long-haul networks, cluster heads may relay data cluster-by-cluster to reach a processing center (e.g., where a user query is issued). Moreover, for applications concerned with sensor readings such as the minimum/maximum/average of sensor data, cluster heads can also perform in-network data aggregation and compression functions to reduce energy consumption [6]. A cluster can be reconfigured globally by the system for optimization purposes or in a distributed manner on a persensor-node basis [5].…”

“…On the other hand, redundancy should be used only as needed so as not to quickly deplete the energy of the system. Current research work on fault tolerance mechanisms to cope with sensor faults in WSNs can be classified into hardware redundancy, time redundancy and information redundancy [1,2,6,7,8]. This paper concerns with hardware redundancy which utilizes extra hardware for fault detection or masking.…”

“…This paper concerns with hardware redundancy which utilizes extra hardware for fault detection or masking. A sensor can be made to disambiguate a sensor measurement fault from a true event by using a distributed Bayesian algorithm [6,7] after comparing readings obtained from its neighbor sensors of the same type. At the processing center (i.e., user query) end, more than one sensor reading responding to a query can be propagated back to the processing center via multiple paths, from which a voting can be performed using the first three readings.…”

Supporting Topographic Queries in a Class of Networked Sensor Systems

“…A lot of work has been done on query processing in sensor networks [6,10]. However, most research efforts have focused upon supporting spatial and/or aggregate queries.…”

“…The end user might be interested in periodically obtaining the total area spanned by a set of feature regions in the network or estimating the boundary of a particular region. Other useful topographic queries of a smaller scope include enumeration and/or description of regions with sensor readings in a specific range [10]. Boundary estimation and counting regions of interest can be used in acoustic monitoring and tracking applications [13].…”