Evaluation of large deviation probabilities via importance sampling

Error probability is the fundamental performance measure for most communications and detection systems, and many of these can be classified as probabilities of the "large deviation type". Practical examples include very diverse applications, including buffer overflow or cell loss probabilities in queuing systems, near/far bit error rates for a DS-SSMA communications system and ruin probabilities for insurance company investment policy, just to name a few. In this paper we examine the of estimation of very small large deviations probabilities via the Monte Carlo technique commonly known as importance sampling. A new theoretical result on the optimization of the importance sampling technique is presented.<<ETX>>