Research On Time-domain Image Formation For Bistatic SAR

Bistatic SAR plays an important role in Earth science and remote sensingapplications. It has become a hot research field of the radar. The graphics processor hasbeen used more and more widely in the field of scientific computing due to itspowerful parallel processing capability. This thesis will research the time-domainalgorithm for bistatic SAR and the parallel implementation of these algorithms on GPU,the main contents are as follows:Establish the bistatic SAR’s geometry model and echo model on random flighttrajectory; Analyze the slant range history of bistatic SAR; Analyze the groundresolution and Doppler resolution of bistatic SAR by the method of Gradient;Introduce some theoretical knowledge about the GPU and describe the programmingmodel based on the CUDA architecture.The BP algorithm based on the NUFFT is proposed. Analyze the procedure of BPalgorithm and Solve the problem of calculating the specific data relate to fast-time bynon-uniform fast Fourier transform to build the BP algorithm based on the NUFFT, wecompare the proposed method in terms of accuracy and computation time with thetraditional BP algorithm which is based on the sinc interpolation by some simulationradar data. The advantages of the proposed algorithm are verified.A parallel BP algorithm based on the NUFFT is proposed. We employ the parallelstrategy based on pixel parallel which require fewer compute memory and own a highparallelization degree; We employ the image segmentation method to apply the sharememory for a fast accessing of inputdata; Use an extra phase table to accelerate thephase calculaton in the kernel; we prove these proposed method is beneficial by somesimulation test on GPU.A factorization BP algorithm base on the Cartesian coordinate image gridding isproposed. Research the factorization BP algorithm base on the polar coordinate imagegridding; Analyze the relation between the maximum phase error and the factorizeparameters for a reasonable selection of some parameters which is necessary; improvethe original method of merging data to make the algorithm available for a Cartesian coordinate image gridding which is more simple and clear; the improved method iscompared with BP algorithm in terms of accuracy and time and is verified moreefficiency by some simulation data.A parallel FFBP algorithm based is proposed. Employ the parallel strategy basedon pixel parallel to build the procedure; the validity of this parallel method is provedby the experiment bistatic SAR data. Use an extra range table to optimize the parallelmethod for a better efficiency for the translate invariant bistatic SAR, the benefit ofthis optimizing is verified by some simulation test on GPU.