Bias-Constrained Optimal Fusion Filtering for Decentralized WSN with Correlated Noise Sources

We propose and investigate a robust fusion optimal unbiased finite impulse response (FOUFIR) algorithm, which is composed of robust UFIR filters implemented in smart sensors and a bias-constrained optimal fusion filter (FF) implemented in a local node of a decentralized wireless sensor network (WSN). The algorithm is designed in discrete time-invariant state space for correlated sensor and node noise sources. To convert the model with correlated noise sources to that without the correlation, the method of indefinite coefficient is used. Assuming that the estimates are delivered to a node with no delay and missing data, a bias-constrained optimal FF is designed employing the linear unbiased minimum variance approach. Simulations are provided for a node operating with 12 smart sensors. Experimental verification is given for a WSN composed with 8 sensors of temperature and humidity. It is shown that the FOUFIR algorithm has better robustness against errors in the noise statistics and model uncertainties than the existing fusion and decentralized Kalman filtering algorithms.