Process Noise Selection Based on Mismatched Filter Design
Abstract
This report derives a explicit recursions for the MSE and bias of both predicted and posterior estimates of a constant gain discrete-time Kalman filter when run on an arbitrary mismatched model. For constant maneuvers, asymptotic solutions are also provided. These solutions are then used to optimize over process noise parameters for Kalman filtering to either minimize the MSE of a mismatched model or such that the maximum error of a mismatched model equals the measurement MSE. As compared to similar solutions in previous work, this formulation of the problem allows for explicit solutions to be obtained in certain scenarios and also to optimize over more difficult trajectory types than are traditionally considered. A number of examples, including finding the optimal covariance scale factor for an evasive weaving target in 2D, are considered.
Document Details
- Document Type
- Technical Report
- Publication Date
- Mar 02, 2023
- Accession Number
- AD1194686
Entities
People
- David F. Crouse
Organizations
- United States Naval Research Laboratory