Acceleration of shooting and bouncing rays method and physical theory of diffraction solvers used for radar cross section calculation of complex and electrical large targets on CPU and GPU hardware through parallelization

There are different methods and assumptions available for obtaining the radar cross-section (RCS) of Computer Aided Design (CAD) models. Most of the solutions that calculate the high-frequency RCS rely on the Geometric Optics (GO) technique and utilize the "Shooting and Bouncing Ray" (SBR) method. Calculating the scattered electromagnetic field from 3D CAD models can take days depending on the surface count of the CAD model, the required bandwidth, and the angle range. This duration is getting considerably longer with expectations of improved resolution quality. High precision is also required for floating-point calculations during the calculations to achieve maximum proximity to the electromagnetic field. Parallelization methods and Graphics Processing Unit (GPU) implementation methods have been investigated and applied to accelerate the SBR method. Within the scope of this thesis, a RCS calculation software was coded in the first stage using the SBR method, which can run on a single CPU. After ensuring the consistency of comparisons made on benchmark targets, the code was parallelized in the second stage. The code, parallelized with the worker thread structure, provided significant performance gains. By coding the Physical Theory of Diffraction (PTD) on the SBR, simulation results reached a point nearly equivalent to measurement results. Especially in structures with wedges, more accurate results were obtained using the PTD. During the PTD coding process, the need arose to identify the edges of the CAD model, and a multi-threaded edge detection method was developed. Later, the PTD code was also parallelized. To further increase the SBR calculation performance, the GPU technology was utilized. For the GPU implementation, NVIDIA CUDA was used, and two separate SBR codes, one using single-precision floating-point arithmetic and the other using double-precision floating-point arithmetic, were developed. The newly coded GPU CUDA-based SBR codes achieved dramatic performance improvements. There was a numerically acceptable level of numerical noise between the CPU-based code and GPU code outputs, and there were no calculation errors in the results.