Empirical Study on the Variability of Monte Carlo Simulation

Abstract

In a Monte Carlo simulation there are variations in the results as a function of the priming numbers, or seeds, of the random number generators. Some of this variation can be interpreted as arising from finite-time observation, but some of it may be due to inherent non-ideal properties of the number generators themselves. These properties cannot generally be deduced from standard statistical tests. Hence the result of the experiment itself must serve as the final measure of variability. An extensive empirical study was conducted in which two nonlinear satellite links using MSK modulation were simulated. The results should provide guidance as to the variability that can result, and the causes thereof, in Monte Carlo simulation of communication systems.