Neural Extended/Unscented Kalman Filter for Submarine Passive Target Tracking

In order to apply undersea target tracking using passive bearings only, this work aims to investigate the performance capabilities of the neural-based extended/unscented Kalman (EKF/UKF) filter. In this paper, the assumption is that a submarine is tracking in underwater another submarine or ship. Application of UKF for target tracking utilizing passive bearings-only measurements is a standard and well-known technique and published in the literature. Here a case study is taken up with the objective of improving results further adding neural network (NN) to EKF/UKF. This technique uses NN nonlinear state-space model approximation and NN's weights are trained on-line by the EKF/UKF. The Monte-Carlo simulations is utilized, and the outcomes are compared to the EKF/UKF procedure, since it is stochastic. It is discovered that the results with EKF/UKF are better than that of the corresponding EKF/UKF with NN. It is understood that NN is to be used when the process uncertainties cannot be modeled exactly or where there is unobservability. In passive target tracking, unobservability exists and this problem is solved by observer maneuver. However, when NEKF/NUKF is used, the observer maneuver is not able to solve the problem. Hence, NN need not be utilized when the analytical solution is feasible, and the solution is tractable.