Importance sampling simulation techniques applied to estimating false alarm probabilities

An upper bound on the estimation variance and a lower bound on the improvement ratio (IR) are derived for importance sampling (IS) simulations with multidimensional input processes. Not only can the runs needed for some accuracy and the IR be obtained, but also various suboptimum IS parameters. Simulation and numerical results indicate that this bounding technique is tight and applicable to non-Gaussian clutters.<
>