Underwater target tracking via the IRWLS filtering approach

Underwater target tracking is treated, and a model is reviewed relating the target states to time delay and bearing measurements as the basis of linearised measurement model. The application of conventional filters to the target-tracking problem is reviewed. Problems associated with fitting models using the least squares procedures are addressed. The success of the procedures depends on the assumption that the distribution of the errors resulting from fitting a model to a set of data is Gaussian. For cases of nonGaussian errors, the least squares performance is far from being optimal. Efforts have been made to improve the performance of the least squares procedures for nonGaussian errors, and to enhance their performance for the Gaussian errors. Robust regression procedures appear to perform much better than the least squares procedures when the errors are nonGaussian and also have improved performances for Gaussian errors. Proposed filters based on the iteratively reweighted least squares method are presented, and computational results are offered to illustrate the performance of the techniques. A comparison with Kalman filters, in terms of prediction accuracy and computational time requirements, shows that the proposed filters are advantageous