GPU-Accelerated SAR Computation with Arbitrary Averaging Shapes

Andre Kuehne¹, Frank Seifert¹, Bernd Ittermann¹

Manual SAR averaging from simulated EM field data is necessary for validating the output of commercial solvers, overcoming limitations of standardized SAR algorithms and gaining flexibility and speed. We developed a GPU-accelerated SAR averaging algorithm based on FFT convolutions that is easy to implement without knowledge of GPU computing paradigms using free software for the GPU programming. It is able to utilize arbitrary averaging shapes, such as cubical or spherical volumes, and yields consistent results even at low resolution (5 mm) of the input data by employing a sub-voxel growing scheme for the averaging kernel.