Accurate Reduced Dimensional Polynomial Chaos for Efficient Uncertainty Quantification of Microwave/RF Networks

This paper presents a polynomial chaos (PC) formulation based on the concept of dimension reduction for the efficient uncertainty analysis of microwave and RF networks. This formulation exploits a high-dimensional model representation for quantifying the relative effect of each random dimension on the network responses surface. This information acts as problem-dependent sensitivity indices guiding the intelligent identification and subsequent pruning of the statistically unimportant random dimensions from the original parametric space. Performing a PC expansion in the resultant low-dimensional random subspace leads to the recovery of a sparser set of coefficients than that obtained from the full-dimensional random space with negligible loss in accuracy. Novel methodologies to reuse the preliminary PC bases and SPICE simulations required to estimate the sensitivity indices are presented, thereby making the proposed approach more efficient and accurate than standard sparse PC approaches. The validity of the proposed approach is demonstrated using three distributed network examples.