Data-Centric Resource Allocation for Machine-Type Communications With Lossy Links Based on Compressive Sensing

Compressive sensing (CS) has been applied as an effective technique to reconstruct collected data from a subset of sensors in a wireless sensor network (WSN). Although many algorithms have been proposed to exploit data correlation among sensors, a majority of these research endeavors focus on sensor selection under ideal communication links. As a result, reconstruction quality would be severely affected when packet losses occur during transmissions. In this paper, we propose to optimize allocation of communication resources to sensors for allowing <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">data retransmissions</i> so the impact of packet losses can be effectively mitigated. Unlike previous work that either does not allow retransmissions or determines the number of retransmissions for each sensor purely based on its packet delivery ratio while disregarding data correlation with other sensors, we further exploit data correlation among sensors for optimization of resource allocation. To proceed, we apply CS with Bayesian estimation to derive the analytical lower bound of data reconstruction error in the target scenario and incorporate retransmission opportunities into the problem formulation. Based on the principle of <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">rank-one adjustment for matrix inversion,</i> we efficiently compute matrix inversion of the data-loss matrix required to minimize reconstruction error. We then propose a greedy yet effective algorithm to determine resource allocation strategy and apply a meta-heuristic search algorithm to further validate its solution. Our proposed algorithms are evaluated using both simulated environment and data from real-world sensor deployment under various resource constraints and data correlation settings. Evaluation results show that the proposed algorithm can outperform existing algorithms in terms of reconstruction error and computation efficiency under link losses.