Optimized Data Selection for Nonlinear Filtering

Nonlinear estimation is ubiquitous in control and signal processing. It aims to estimate the most probable state of a system and its range of uncertainty by fusing data from multiple sensors over time using a filter. Although many efficient filters exist in the literature, their computational cost may increase when the set of data to process is significant. Besides, some data can be redundant or bring little information to the estimation of the state. In that case, their processing is costly and does not contribute to the filter's estimation. This paper introduces a semi-heuristic method to select a relevant subset of observations from a more extensive set of available data, using a cost function based on the approximation of the Cramer-Rao Lower Bound. This approach is adopted on an angles-only optimal navigation scenario where an extensive signal set is available. The results show that the filter achieves close-to-optimal accuracy for a lower computational cost.