Oba Filter for Online-Adaptive Nonlinear State Estimation With Interference

Abstract

Interference, such as jamming, is widely encountered in practical applications, such as target tracking. Kalman-based nonlinear filters may not be feasible for nonlinear state estimation with nonlinearly transformed interference. This paper presents an online-based adaptive (oba) filter, which we have named the oba filter (OF), for state filtering and prediction of nonlinear discrete dynamic models with estimate constraints, nonlinearly transformed interference and noise. In the OF, noises and interference are approximated by discrete noises and interference; states are quantized; some undesired state estimates, which are not in near vicinities of the actual state values, are detected by a confidence level, and then state quantization levels are updated at the times when undesired estimates occur by using the measurement model; and states are recursively estimated by a suboptimal implementation of multiple composite hypothesis testing. The OF can prevent some undesired state estimates and outperform the unscented Kalman, regularized, sampling importance resampling (SIR), and auxiliary SIR particle filters for some nonlinear models with noise and interference. It is also suitable for state estimation with either missing observations or estimate constraints.