Gpu-based Synthetic Aperture Radar Simulation Technology Research

In recent years, as the technology of Graphics Processing Unit (GPU) in hardware and software developing, people try their best to make use of GPU widely because of the capability of its floating-point calculations and parallel processing. GPU for general purpose computations (GPGPU) become a hot topic for research.This dissertation studies on the airborne Synthetic Aperture Radar (SAR) echo signal simulation. Echo signal simulation is very important in the whole SAR system simulation, and needs to compute all SAR target echo signals in the scene, so the calculations are large and uesed much time on CPU. Because of the high performance of GPU floating-point calculations and parallel processing, this dissertation focuses on the computation of SAR echo signals on GPU, which steps up the performance of the whole SAR system.The main jobs of this dissertation are following:1. The development and characteristics of GPU are described, and the software and hardware patterns of Compute Unified Device Architecture (CUDA) from NVIDIA are showed. The model of airborne SAR echo signal is given, and the SAR imaging principles are analyzed.2. The computation of cosine, matrix multiplication,1D-FFT and 2D-FFT are implemented on GPU and CPU. In order to analyze the result precision, cosine computation is taken for example. The experiment shows that the high performance of GPU floating-point calculations and accuracy of raw data on GPU, and proves that GPGPU can save much time.3. The computation of airborne SAR echo signal simulation is implemented on GPU and CPU. Compared with the calculations of the complex targets in the large scene on CPU, the speed of raw data generation on GPU is 13 times as on CPU. In order to prove the correctness of the raw data on GPU, Range-Doppler arithmetic is used. The results show that the validity of the raw data on GPU.4. In order to speed up the raw echo data generation of large scene, multi-GPU is used. The speed of raw data generation on multi-GPU is 26 times as on CPU single-thread, and 13 times as on CPU multi-thread. Range-Doppler arithmetic is used to prove the correctness of the raw data on multi-GPU. The echo signal calculatied on GPU is used in the whole simulation system, and the performance of the system is improved.