Comparison of Batch and Kalman Filtering for Radar Tracking

Download or read book Comparison of Batch and Kalman Filtering for Radar Tracking written by Haywood Satz and published by . This book was released on 2001 with total page 7 pages. Available in PDF, EPUB and Kindle. Book excerpt: Radar tracking performance was compared among two choices of statistical filtering algorithms for the noisy measurements of exo-atmospheric objects in ballistic motion. Such motion is characteristic of satellites and missiles. Object position and velocity were governed by the nonlinear dynamics of body motion in a central force field, and measurements were modeled as nonlinear observations of those object motions in Cartesian coordinates. The two choices of statistical filtering algorithms were distinguished by their method of handling a sequence of noisy measurements. The first processed measurements, one-at-a-time, in a sequential recursive estimation using the Extended Kalman Filter (EKF), and the second processed that same sequence of measurements, simultaneously, in a batch-least-squares (BLS) estimation algorithm. Both algorithms used first-variation approximations of the nonlinear observations and error dynamics of object motion. Taylor series expansions were centered about the current best estimates of the state vector. The EKF used those approximations to implement the often referenced linear-minimum-variance (Kalman) estimation formulas. The BLS processed those same measurements simultaneously in a least-squares search over object trajectories spanning the tracking interval, and initial state estimation was based on convergence to the best object path. Results were obtained for both algorithms performing in a desktop program with a reasonable degree of radar systems modeling, and in a comprehensive mission simulator where radar system errors were represented in greater detail. Those included radar-cross-section fluctuations, scan angle loss, antenna gain patterns, radar signal-to-noise sensitivity, monopulse measurement errors, and front-end noise. The BLS algorithm was seen to converge faster, and predict more accurate 1-sigma values, than the EKF in all comparisons.