Multi-Sensor Information Filtering With Information-Based Sensor Selection and Outlier Rejection

Nowadays, multi-sensor networks are evolving into large scale networks with limited bandwidth and energy reservoirs. Hence, reducing the number of information exchanges among the sensors is an efficient approach to meet the stringent requirements of bandwidth and energy in the context of multi-sensor state estimation. In this paper, a surprisal-based multi-sensor information filtering is proposed to allow on average a desired number of sensors with most informative measurements to participate in the information exchange, while discarding the sensors with non-informative measurements and outliers. The concept of surprisal is used to facilitate the sensors in the network to independently classify their measurements as informative, non-informative and outliers by setting upper and lower thresholds. Simulation results show that the proposed scheme ensures that on average only a desired number of sensors with the most informative measurements participate in the information exchange and achieves greater estimation accuracy compared to the method, where the sensors are selected randomly for the same number of transmissions.