Distributed nonlinear state estimation via multi-level quantised innovation and covariance-intersection fusion

Abstract

We study distributed state estimation for nonlinear systems observed by spatially separated sensors that communicate over rate-limited links. Each node transmits a multi-level quantised innovation, and a quantisation-aware approximate minimum mean-square error update is derived in closed form using optimally designed thresholds. The resulting local posteriors are combined by covariance-intersection fusion, and a sufficient condition is obtained that guarantees bounded fused error covariance while clarifying the impact of quantisation resolution. Numerical simulations on a manoeuvring-target tracking example demonstrate that, under comparable communication budgets, the proposed multi-level quantisation–based estimator achieves a favourable accuracy–bandwidth trade-off and exhibits stronger robustness to non-Gaussian noise than a centralized extended Kalman filter (EKF) and several bandwidth-limited quantised EKF baselines.