Decentralized Kalman filtering algorithm with uncertain signal model for heterogeneous sensor networks

This paper investigates the problem of designing decentralized robust Kalman filters for sensor networks observing a physical process with parametric uncertainty. A sensor network consists of distributed collection of nodes, each of which has sensing, communication and computation capabilities. We consider a heterogeneous sensor network consisting of two class of nodes (type A and type B) and central base station. Type A nodes undertake the sensing and make noisy observations of the same physical process while type B nodes play the role of cluster-heads. We derive the information form of robust Kalman filter by using the Krein space approach which proves to be useful to combine the local estimates. We obtain the decentralized robust Kalman filter for each type B node for the state estimation of uncertain physical phenomena of interest by taking into consideration the sensing model of each cluster and the information form of robust Kalman filter. The type B nodes transmit their estimates along with the inverse of error covariance matrix to the central base station which fuses these local estimates to produce the global estimate. Simulation results indicate that the performance of the centralized estimate is comparable to the performance of the global estimate and this suggest that they are identical.