High-performance Parallel Computing Of Electromagnetic Scattering From Multi-scale Electrically Large Sea Surface

The calculation of the time-evolving electromagnetic(EM)scattering to study the electromagnetic scattering characteristics of sea clutter from time-evolving multi-scale electrically large sea surface is becoming increasing important due to their application in maritime target identification,microwave remote sensing,marine environment monitoring and synthetic aperture radar(SAR).It mainly displays in the following aspects.First,different sea states,radar bands,incident angle and radar beam parameters make sea clutter characteristics by capricious and complexity.Second,due to the high radar resolution at small grazing angle for both spaceborn and airborne radar,the requirements of millions of small facets on the dynamic sea surface and its real-time dynamic complexity(ms)make calculation of the time-evolving electromagnetic scattering field time-consuming and hardly meet real-time requirements.In this paper,based on the GPU-CPU heterogeneous platform with CUDA,the high performance parallel computing of the EM backscattering from the time-evolving multi-scale electrically large sea surface is proposed to analyze the influence of certain parameters,such as wind speed,wind direction,radar parameters,and incident angles,on the EM backscattering characteristics.The main work is as below:1).For different radar bands and incident angles,CUDA and DSM are utilized to establish time-evolving multi-scale electrically large sea surface model based on GPU-CPU heterogeneous platform.The development of a GPU-accelerated massively parallel computation of a time-evolving electrically large sea surface by exploiting GPUs with diverse optimizations is proposed to further increase the speed of the generation of the sea surface.2)Based on the zoning theory,the multi-scale electrically large sea surface is divided into a set of sub-surfaces,and a slope-deterministic composite scattering model(SDCSM)is utilized to calculated the normalized radar cross section of the multi-scale electrically large sea surface.For high sea states,a time-evolving multi-scale electrically large sea surface model with foam coverage was established.The composite scattering model with SDCSM and vector transfer(VRT)is proposed to calculate the EM scattering from the electrically large sea surface with foam coverage.The feasibility of using GPU with four CUDA optimization strategies to improve the calculation efficiency of EM backscattering from electrically large sea surface is verified.3)In order to theoretically analyze the multi-scale rough surface EM scattering by an incident Gaussian beam instead of a plane wave.Based on the plane wave spectrum method,the scattering field of Gaussian beam from multi-scale rough surface is derived and the Kirchhoff approximation,small perturbation approximation and the stationary phase method.Compared with the result of the plane wave,additional beam factor represents the effect of Gaussian beam,which is related to beam waist size,distance from waist to rough surface,the roughness parameters and the incident or scattering angles.4)A facet-based composite scattering model was utilized to calculate the time-evolving multi-scale electrically large sea surface.Two scattering mechanisms,the quasi-specular scattering mechanism and the Bragg scattering mechanism,are taken into account.In addition,the multicore GPU with five optimizations has be exploited to improve the computational performance of the time-evolving EM scattering field.5)Based on the time-evolving EM backscattering filed in the above chapter,the EM backscattering characteristics of sea clutter from time-evolving multi-scale electrically large sea surface are systematically studied for different radar parameters and sea states.