SBL-Based Adaptive Sensing Framework for WSN-Assisted IoT Applications

Wireless sensor networks (WSNs) often have limited battery capacity-sensor nodes (SNs), which severely limit continuous monitoring-based Internet-of-Things applications. To substantially increase the lifetime of a densely deployed WSN for monitoring spatio-temporally varying signal, this paper presents a novel sparse Bayesian learning-based adaptive sensor selection framework. The developed strategy selects an active sensor set and turns off the remaining SNs by jointly optimizing two conflicting performance measures: 1) sensing quality and 2) energy efficiency, while considering prevailing energy parameters of the network. To achieve this, a multiobjective optimization problem is formulated. Further, a joint principal component analysis-sparse Bayesian learning (PCA-SBL) scheme is presented which uses PCA-based estimated transformation matrix to sparsify the data sensed by sensors, and subsequently uses approximate overcomplete dictionary-based SBL scheme to estimate it. Employing PCA-SBL-based signal estimate, a closed loop adaptive mechanism is developed which estimates variations of the monitored signal to predict the number of active sensors for next measurement cycle such that the sensing error remains within an acceptable range. This predicted value is then used in sensor selection problem to dynamically select the active set. The sensor selection, signal recovery, and feedback loop use spatial and temporal correlation inherent in the monitored phenomenon. Extensive simulation studies validate the energy efficiency and stable sensing performance of the proposed framework using both synthetic and real data of a WSN.