DBCS: A Decomposition Based Compressive Sensing for Event Oriented Wireless Sensor Networks

Abstract

Demarcating distributed event region is a key issue in various application domains of wireless sensor networks. In this paper, the problem of energy conservation in event region demarcation is studied. Two major challenges of event region demarcation problem are; accurate estimation of homogeneous regions in presence of noisy observations and continuous monitoring for detecting the boundary of the region. A Markov random field (MRF) structure model is proposed for decomposition of area of the network into different homogeneous areas using efficient belief propagation based in-network inference. To achieve the homogeneity in each distinguished homogeneous areas, sensor node updates its local estimate based on the neighborhood information and its local observation. Considering the communication constraints in such continuous monitoring systems, a Decomposition based compressed sensing (DBCS) approach is integrated with the proposed MRF model for globally estimating the state of target area. DBCS provides an energy efficient solution compared to other similar data collection techniques. Simulation results proves our model’s strength in terms of accuracy of the critical region detection, and is capable of achieving significant 90% reduction over transmissions required for approximate reconstruction. Moreover, the proposed DBCS allows early reconstruction which reduces the average energy consumption up to 15% in the network as compared to existing multi-hop compress sensing using random walk (M-CSR) approach.