Scalable Acceleration of Characteristic Mode Analysis Using Big Data Techniques

In this paper, we present an approach to scale Characteristic Mode Analysis (CMA) using big data techniques over a wide range of Method of Moments (MoM) impedance matrices sizes. CMA is a computational toolbox that is used to decompose the electromagnetic response of a structure into a set of fundamental modes which provides physical insight into a target's electromagnetic response. CMA can potentially be time-consuming especially for large MoM matrices as well as when it is applied to a wide range of frequencies. Therefore, we propose a parallel processing approach to accelerate CMA using Apache Spark and Apache Hadoop, which are the most popular big data frameworks that support large-scale data processing in a commodity cluster. This paper describes our proposed approach and the scalability improvements that we achieved using Spark and Hadoop. Our approach improved the CMA processing time by 12X for 100 matrices of size
$\boldsymbol{2\mathrm{K}\times 2\mathrm{K}}$
) and by 10X for 51 matrices of size
$\boldsymbol{14\mathrm{K}\times 14\mathrm{K}}$
. Furthermore, our approach that builds atop big data techniques produced the same computational results as traditional tools for CMA.