基于GPU的超宽带SAR实时成像技术研究

发布时间：2018-03-10 13:56

本文选题：超宽带合成孔径雷达　切入点：图形处理器　出处：《国防科学技术大学》2014年硕士论文　论文类型：学位论文

【摘要】：本文针对常规后向投影(Back Projection,BP)算法运算量大的问题,综合硬件多核并行以及孔径分级优化两类技术,实现对高分辨率超宽带合成孔径雷达(Ultra Wideband Synthetic Aperture Radar,UWB-SAR)实时成像,具有重要的理论意义和工程实用价值,论文主要工作如下:建立了步进频率信号条带式SAR的回波模型,分析了BP算法距离压缩和方位压缩过程。详细介绍了图形处理器(Graphic Processing Unit,GPU)和统一计算设备架构(Compute Unified Device Architecture,CUDA)编程知识,得出多核并行化、存储资源分配及访问影响最终算法效率的基本结论,是后续并行化方法优化的基础。提出了一种基于GPU的并行优化网格BP算法。重点对方位压缩过程设计了三种优化方法,利用网格BP算法结构减少了多线程访问原始回波数据的次数;利用纹理存储器加速了插值计算效率;合理设计了子图像存储方式,避免共享存储器访问冲突。从而有效利用GPU的多核架构提高了成像效率。提出了一种基于GPU的并行化因式分解快速BP(Factorized Fast BP,FFBP)算法。设计了FFBP算法的实现流程,分析了FFBP在算法层面的加速比,最后利用GPU实现了并行加速。与基于GPU的BP算法相比,成像效率获得进一步显著提升。为了有效处理实测UWB-SAR回波数据,设计了上述两种算法的分块实现途径,解决输入数据量大但GPU存储资源不足的问题,获得了聚焦良好的高分辨率图像,与CPU单线程实现的BP算法相比,加速比分别为70倍和232倍左右,验证了论文所提方法的有效性。
[Abstract]:In this paper, aiming at the problem of large computation of conventional back projection back projection (BP) algorithm, combining hardware multi-core parallelism and aperture grading optimization techniques, the real-time imaging of Ultra Wideband Synthetic Aperture Radarar (UWB-SAR) for high resolution ultra-wideband synthetic aperture radar (UWB-SAR) is realized. It has important theoretical significance and practical engineering value. The main work of this paper is as follows: the echo model of striped SAR with step frequency signal is established. The process of BP algorithm distance compression and azimuth compression is analyzed, and the programming knowledge of graphic processor graphic Processing Unit (GPU) and Unified Computing equipment Architecture (Compute Unified Device Architecture) are introduced in detail, and the multi-core parallelization is obtained. The basic conclusion that storage resource allocation and access affect the efficiency of the final algorithm is the basis of the optimization of the subsequent parallelization method. A parallel optimization mesh BP algorithm based on GPU is proposed, and three optimization methods are designed for the azimuth compression process. The mesh BP algorithm structure is used to reduce the number of multithreading access to the original echo data; the texture memory is used to accelerate the interpolation calculation efficiency; the sub-image storage mode is designed reasonably. In order to avoid the access conflict of shared memory, the multi-core architecture of GPU is used effectively to improve the imaging efficiency. A parallel factorization algorithm based on GPU is proposed for fast BP(Factorized Fast BP- FFBPs, and the implementation flow of FFBP algorithm is designed. The speedup ratio of FFBP at the algorithm level is analyzed. At last, parallel acceleration is realized by using GPU. Compared with BP algorithm based on GPU, the imaging efficiency is further improved. In order to solve the problem that the input data is large but the GPU storage resource is insufficient, the high resolution image with good focus is obtained, which is compared with the BP algorithm implemented by CPU single thread. The speedup ratio is about 70 times and 232 times respectively, which verifies the effectiveness of the proposed method.
【学位授予单位】：国防科学技术大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TN957.52

【相似文献】