Parallel computation methods for enhanced MOM and MLFMM performance

Abstract

The success of present and future intelligence, surveillance and reconnaissance (ISR) systems, in an increasingly electromagnetically complex world, is going to depend directly upon the speed and efficiency of our computational systems. These systems are used for advanced electromagnetic computations such as antenna cosite coupling, intermodulation, and radar cross section (RCS) analyses, among many more applications. Such computations require the use of first principle electromagnetic codes, such as method of moments (MoM) and Multilevel Fast Multipole Method (MLFMM), to perform full wave analyses. Unfortunately, these methods are very time consuming and memory prohibitive due to the inherent complexity of our ISR systems. At present, the models currently being used for analysis of EM computations could take days or even weeks to formulate a solution. Many times, it takes hours to simply determine if there is an error in the problem or if it is unsolvable. Since real-time computation analysis is so important to the defense industry, Northrop Grumman has been working extensively to discover ways in which to make these necessary calculations faster and more efficient. Graphics Processing Unit (GPU) computation offers a unique opportunity for electromagnetic simulation acceleration. GPU technology has been advancing faster than CPU technology due to a consumer fueled gaming industry. GPUs use a unique pixel based system that can not be simulated in an ordinary CPU and therefore allows for unique benefits when running computations. Northrop Grumman has been collaborating with Stony Brook University to explore their research in GPU computation. Northrop Grumman has its own, functioning, 6 node GPU cluster that we hope to use, in parallel with compressive sensing. Our GPU cluster will be able to parallelize the complex computations across the six nodes of the system, which will again decrease computation time. GPU computation has many applications besides electromagnetic modeling and RCS analysis. These modern adaptations for complex computing can be applied to virtually any large, complex and time-consuming problem. With these modifications, we hope to be able to increase the ability of our systems to handle computations that are more difficult because the complexity of our world will only continue to increase.