The Research And The Realization Of The Gpu Bistatic Sar Imaging Method

Bistatic SAR is a new kind of synthetic aperture radar. Compared with monostatic SAR, bistatic SAR has many advantages, such as it can obtain the target’s non-backscattering feature, has good imperceptibility and high anti-interference ability and survivability. So it has currently been the focus of SAR imaging application. Graphics processing unit (GPU) can be used in Large-scale parallel computing, has been applied on many domains. In this dissertation, research on the technologics of BiSAR imaging algorithm is carried out and parallel back projection algorithm Implementation based on GPU. The main contributions and contents are demonstrated follows:1. Introduce the fundamental theory of BiSAR imaging and GPU. First based on the geometry of BiSAR analyze its slant range history, then introduce the theory and space resolution of SA-BiSAR imaging and its spatial varying characteristic. Second introduce tow kinds of GPU architecture:Tesla architecture and Fermi architecture, then introduce the Programming Tenets of compute unified device architecture (CUDA).2. A fast-factorized back projection algorithm for BiSAR is proposed. First, according to the subimage grids conduction of Cartesian coordinate system and elliptical coordinate system, testify that elliptical coordinate system is more conducive to algorithm Implementation. Second, Analytical expressions for the Nyquist requirements using elliptical coordinate system are derived. The overall discussion includes practical implementation hints and an realistic computational burden estimation. Last, through simulation to testify that this method has good imaging result and sharply reduces computation time compared with direct back projection.3. A fast parallel back projection algorithm based on GPU is proposed. First, for the part of range compress parallelization, the influence of FFT length is discussed and the most reasonable method of FFT length choice is provided. Second, for the part of back projection parallelization, three parallel methods:PRIs parallel, pixels parallel, PRIs and pixels combination parallel are proposed. Then through analysis and simulation, pixels parallel as the best parallel method is determined. 4. For the method of pixels parallel, a shared memory optimal design method is proposed. Through taking advantage of shared memory and the correlation between blocks and slant range history, a shared memory optimal design method is found, then the simulation results show that the parallel performance can be improved on the Tesla architecture.5. The parallel algorithm performance is analyzed for different precision and an improvement method for double precision is proposed. The simulation results show that in the case of double precision, imaging result is well but the performance is poor. Through analysis the characteristics of the data, an optimization method is found and the simulation shows that good imaging result and high performance can be obtained.6. For the shift invariance BiSAR, an improvement fast parallel back projection algorithm based on GPU is proposed. Using the shift invariance character of slant range history, the computational burden of different pixels can be reduced. The simulation results show that this method is extremely effective special for the large scene.