Performance prediction of hybrid algorithms

This chapter reviews that the hybrid conditional averaging (HYCA) performance prediction technique is applied to many hybrid algorithms, including multiple-model (MM) algorithms. It gives a full consideration to the important scenario dependence of the performance of hybrid algorithms and thus yields good accuracy. This hybrid technique is obtained by a conditional expectation operation through which the randomness of the algorithm performance due to the uncertainties in the continuous-valued state space is averaged out, whereas the dependence of the performance on the scenario is retained. As such, this technique provides a scenario-conditional prediction of the average performance of the algorithm. The MM algorithm which is handled includes the notable Interacting Multiple-Model estimation algorithm, the Generalized Pseudo Bayesian estimation algorithms, and the Viterbi algorithm. The performance of the more sophisticated versions of the interacting multiple model (IMM) algorithm, such as the sojourn-time-dependent IMM and the hybrid IMM is also predicted. It can also be applied to the variable structure MM estimator. As a special case, this performance prediction technique is also applied to single-model schemes because it is valid for algorithms with an arbitrary number of models. Another class of hybrid algorithms whose performance is predicted by this HYCA technique is tracking and data association algorithms. The technique has been used to predict successfully the performance of the well-known PDA filter, the popular nearest neighbor filters, and the simple yet effective strongest neighbor filters.