A new method of generating multivariate Weibull distributed data

Abstract

In order to fully test detector frameworks, it is important to have representative simulated clutter data readily available. While measured clutter data has often been fit to the Weibull distribution, generation of simulated complex multivariate Weibull data with prescribed covariance structure has been a challenging problem. As the multivariate Weibull distribution is admissible as a spherically invariant random vector for a specific range of shape parameter values, it can be decomposed as the product of a modulating random variable and a complex Gaussian random vector. Here we use this representation to compare the traditional method of generating multivariate Weibull data using the Rejection Method to a new approximation of the modulating random variable that lends itself to efficient computer generation.