WorksMathematics
Nonlinear Joint State and Noise Estimation under Packet Dropouts and Additive Measurement Corruption
Abstract
This paper investigates joint state and noise covariance estimation for nonlinear systems subject to dual reliability impairments: random packet dropouts and additive measurement corruption. This scenario poses challenging nonconjugate inference, as clean and corrupted measurements share an unknown baseline covariance with fixed known corruption covariance . Packet reception is observable, whereas the authenticity of received measurements remains latent and must be inferred. To resolve this intractable estimation problem, we cast joint posterior inference as a variational optimization problem and develop a conjugate variational inference adaptive Kalman filter with dual reliability (CVIAKF-DR). The proposed method retains analytical updates for the prediction covariance and two reliability probabilities, while adopting CVI to estimate the state and baseline measurement covariance. The covariance update combines the analytical contribution from clean measurements with Monte Carlo gradient estimates for corrupted measurements. Numerical results on a re-entry tracking benchmark demonstrate the superior accuracy of the proposed method in both state and noise covariance estimation. Remarkably, even built upon the linear Kalman filtering (KF) framework that is originally designed for linear systems, our method achieves more accurate estimation performance than dedicated nonlinear filtering algorithms based on the cubature Kalman filter (CKF).