Study Of Electromagnetic Scattering Property Analysis And BP Imaging Based On GPU

As the development of Synthetic Aperture Radar(SAR),the caculation of the electromagnetic scattering of complex electrically large targets has gained the attractions of scholars of various countries.Physical optics(PO)is one of the high-frequency electromagnetic computing methods,which is simple and accurate.But when dealing whith complex real targets,PO is not efficient enough.The performance bottle-neck of PO is occlusion.Now there are many algorithms dealing with this problem,such as graphical electromagnetic computing?k-d tree? Z-buff and so on.Among these algorithms,Z-buff has low computational complexity,and is easy to implement.But at the same time,it will introduce some occlusion inaccuracy.The RCS data caculated by PO could be validated by SAR imaging algorithm.Packprojection(BP)is one of the time-domain algorithms in SAR imaging.When compared to frequency-domain algorithms,BP is more accurate and is invariant to different imaging modes.But its computational complexity is too high,and need great computation power when being implemented.GPU has great parallel computing power,and it doesn't strop improving its computing ability.Compared with CPU and FPGA,GPU has more computing ability and has better price-quality ratio.The main work and innovations of this paper are:1.The basic theory and computational complexity of PO are analyzed;the theory of Z-buff are introduced,and a method which prevent the occlusion inaccuracy caused by Z-buff from impacting the output of PO is given,providing the theoretical basis for following work.First,the derivation of theory of RCS and PO are presented,and the computational complexity and the performance bottle-neck of PO are analyzed.Second,the theory of Z-buff is presented,and how its occlusion inaccuracy influences the PO output is analyzed,and a method is given to restrain this influence.Last,an occlusion algorithm based on Z-buff is proposed,which has no occlusion inaccuracy,but has high computational complexity.2.Two implemetations of Z-buff based on CUDA is presented.First,the programming of CUDA is introduced.Second,a global lock was implemented based on atomic methods provided by CUDA,and the how to use it correctly is presented,and a Z-buff implementaton based on the global lock is presented.Third,a Z-buff implementation based on no-lock algorithm is presented.Last,numerical example is presented to show the efficiency of these implementations,and an implementation of PO based on CUDA is presented.3.An implementation of BP based on CUDA is presented to validate the output of PO which based on the Z-buff presented above.First,the basic theory of BP is presented,the computational complexity is analyzed.Second,a implementation of BP is presented.Last,numerical example is presented.