Parallel implementation of physical optics by using GPU with CUDA

A parallel implementation of physical optics (PO) is presented on the graphic card NVIDIA GTX480. Taking the advantage of the characteristic of visible surface detection in PO, an improved Z-buffer detection algorithm for triangle meshes is developed, which can significantly decrease the amount of float point computation. By using reduction operation, we design a fast parallel summing method for the physical optics integrals over illuminated triangles. The numerical examples show that the PO algorithm proposed in this paper has high precision, and the efficiency is much higher (more than an order of magnitude) in comparison with the serial computing program. So it can serve as a good support for radar target characteristics simulation computation and scattering characteristics analysis of complicated target.