A hybrid conditional averaging technique for performance prediction of algorithms with continuous and discrete uncertainties

Increasing attention has been given to hybrid algorithms those that involve both continuous-valued and discrete-valued uncertainties. The performance of these algorithms are, however, difficult to evaluate without recourse to costly and time-consuming Monte Carlo simulations. In this paper, a general and accurate technique for nonsimulation performance evaluation of hybrid algorithms is presented. This technique gives full consideration to the important scenario dependence of the performance by using a scenario-conditional expectation of the performance. The system mode sequence is adopted as the essential description of the scenario. Two versions of the technique are given: mode-sequence-conditional, and current-mode-conditional. The first one is applied to the notable interacting multiple model algorithm and the second one to the popular probabilistic data association filter for tracking in clutter. The accuracy of the technique is demonstrated via examples.