A modified importance sampling scheme for the estimation of detection system performance

For many modern detection systems, analytic expressions for the probability of false alarm and/or probability of error are not available. Monte Carlo simulations have often been used as a direct means of estimating these quantities. A technique known as importance sampling can provide a significant reduction in the variance of such estimates. A number of importance sampling biasing z densities have been shown to be effective for problems of interest. A method for further improving the accuracy of importance sampling estimates is presented. This technique utilises additional information available in the simulation to adapt the standard importance sampling estimate based upon specific realizations of the biased inputs. This new methodology is easily implemented and results in an appreciable improvement in the accuracy of the estimator in all examples considered.<>