Noise covariance matrix estimation in navigation and tracking: Impact of linearisation error
Oliver Kost, Jindřich Duník, Ondřej Straka · 2017
The paper deals with the estimation of the noise covariance matrices of nonlinear state-space models, which are linearised for the purpose of the estimation. A special attention is focused on an analysis of the linearisation error effect on the quality of the covariance matrix estimates. For the analysis, typical representatives of four fundamental approaches to noise covariance matrix estimation, i.e., a correlation method, a maximum-likelihood method, a covariance matching method, and a Bayesian method are selected and briefly introduced. The analysis is performed using a navigation example with an additional assessment of the direct and indirect impact of the linearisation error on the state estimate.