Applications of information theory in filtering and sensor management

“A classical sensor tasking methodology is analyzed in the context of generating sensor schedules for monitoring resident space objects (RSOs). This approach, namely maximizing the expected Kullback-Leibler divergence in a measurement update, is evaluated from a probabilistic perspective to determine the accuracy of the conventional approach. In this investigation, a newdivergence-based approach is proposed to circumvent themyopic nature of the measure, forecasting the potential information contribution to a time of interest and leveraging the system dynamics and measurement model to do so. The forecasted objective exploits properties of a batch measurement update to frequently exhibit faster optimization times when compared to an accumulation of the conventional myopic employment. The forecasting approach additionally affords the ability to emphasize tracking performance at the point in time to which the information is mapped. The forecasted divergence is lifted into the multitarget domain and combined with a collision entropy objective. The addition of the collision consideration assists the tasking policy in avoiding scenarios in which determining the origin of a measurement is difficult, ameliorating issues when executing the sensor schedule. The properties of the divergencebased and collision entropy-based objectives are explored to determine appropriate optimization schemes that can enable their use in real-time application. It is demonstrated through a single-target tasking simulation that the forecasted measure successfully outperforms traditional approaches with regard to tracking performance at the forecasted time. This simulation is followed by a multitarget tasking scenario in which different optimization strategies are analyzed, illustrating the feasibility of the proposed tasking policy and evaluating the solution from both schedule quality and runtime perspectives”--Abstract, page iii