Large Scale Parallel Computation of Radiation from Complex Arrays

 

Cinzia Zuffada, Tom Cwik, Daniel S. Katz*, Vahraz Jamnejad

Jet Propulsion Laboratory

California Institute of Technology,

Pasadena, CA 91109

 

*Cray Research, Inc.

El Segundo, CA 90245

 

ABSTRACT

 

A technique combining finite elements modeling with (a) an integral equation of the surface (chosen to be a surface of revolution) outwardly truncating the computational domain and (b) waveguide mode matching representing the source, is used to model radiation from antennas fed by waveguides and coaxial cables. This method allows the representation of fields in highly inhomogeneous, penetrable, single and multiple feed radiators. Because of the electrical sizes involved, large scale parallel computation is an enabling resource to investigate realistic systems. The complete software package is implemented on the Cray T3D massively parallel processor located at JPL, using both Cray Adaptive FORTRAN (CRAFT) compiler constructs to simplify portions of the code that operate on irregularly distributed data, and optimized message passing constructs on portions of the code that operate on regularly distributed data and require optimum machine performance. The unstructured mesh associated with finite element simulation of electromagnetic fields inside and around complex objets is handled here directly by the host Cray Y-MP, without using traditional mesh decomposition algorithms. An optimized parallel iterative solver, developed to solve large sparse systems, is employeed here to operate on the large and mainly sparse component of the system resulting from the combination of the finite elements and the mode matching technique, together with a parallel dense matrix solver that operates on a much smaller, reduced component of the system of equations, related to the integral equation on the truncating surface. Results of the complete simulation are presented for problems of varying size and geometry.