Distributed detection and data fusion in resource constrained wireless sensor networks

Abstract

Wireless sensor networks have received immense attention in recent years due to their possible applications in various fields like battery-field surveillance, disaster recovery etc. Since these networks are mostly resource-constrained there is a need for efficient algorithms in maximizing the network resources. In this thesis, energy and bandwidth-efficient detection and fusion algorithms for such resource constrained wireless sensor systems are developed. A Sequential Probability Ratio Test (SPRT) based detection algorithms for an energy-constrained sensor network is proposed. Performance is evaluated in terms of number of nodes required to achieve a given probability of detection. Simulation results show that a network implementing the SPRT based model outperforms a network having a parallel fusion detector. To implement distributed detection and fusion in energy and bandwidth constrained networks, non-orthogonal communication is considered to be one of the possible solutions. An optimal Bayesian data fusion receiver for a DS-CDMA based distributed wireless sensor network having a parallel architecture is proposed. It is shown that the optimal Bayesian receiver outperforms the partitioned receivers in terms of probability of error. But the complexity of this optimal receiver is exponential in the number of nodes. In order to reduce the complexity, partitioned receivers that perform detection and fusion in two stages are proposed. Several well-known multi-user detectors namely, JML, matched filter, Decorrelator and linear MMSE detectors are considered for the first stage detection and performance is evaluated in terms of probability of error at the fusion center. Conventional detector based fusion receiver has a performance close to that of optimal fusion receiver with quite less complexity under specific channel conditions. Performance and complexity trade-offs should be considered while designing the network.