On optimum and suboptimum biasing procedures for importance sampling in communication simulation

Importance sampling can significantly reduce simulation run time in estimating the bit error probability of digital systems if a suitable procedure is used. Various aspects of optimum and suboptimum biasing procedures are considered. It is shown that the strict optimum does not lead to a realistic procedure. An examination is made of the suboptimal variations that can be implemented reasonably; in one variation, it is found that for a linear system, the biasing scheme should have a time-varying mean proportional to the impulse response. Aspects of implementation are also discussed.>