A Novel Information Theoretic Measure Based Sensor Network Design Approach for Steady State Linear Data Reconciliation

Abstract

The current work proposes a novel information theoretic based sensor network design (SND) approach for data reconciliation in a steady state linear process. The proposed approach is based on Kullback-Leibler divergence (KLD), which measures the difference of a density function from a reference density function. In particular, the optimal design is the one that leads to the smallest KLD value of the designed density function of the estimates from a reference density function. This reference density function can be provided by the end-user, and the approach thus enables explicit incorporation of the end-user’s preference in the SND procedure. Additionally, the approach does not assume specific forms for the density functions of the estimates and is thus also applicable for cases when the estimates have non-Gaussian density. The significance of the approach is illustrated on a small example. To demonstrate its utility in obtaining optimal sensor networks, it is also applied to a popular case study from SND literature and results are compared with existing approaches.