Airborne SAR Imaging Algorithm Design And Implementation Based On GPU

With the development of synthetic aperture radar(SAR), it has been widely used in the national defence and civil industry. Since SAR is capable of observing earth surface day and night, all-weather, long range, high resolution, wide area, higher resolution and wider swath are required in updating SAR system. Wide azimuth beam high resolution airborne SAR is a way to achieve high resolution and wide swath. However, in Wide azimuth beam high resolution SAR, the synthetic aperture time is very long and the amount of data and computation is very large, data processing on a traditional CPU-based workstation or server is rather time-consuming, making real-time processing of SAR data impossible. With the emergence of GPGPU, GPGPU provides a feasible approach to the real-time processing of high resolution SAR data. GPGPU is designed for accelerating computing-intensive applications. It has now developed into a highly parallel, multi-threaded, multi-core and large-bandwidth platform with hundreds of high-performance processing unit. First this thesis introduces the basic principles of SAR algorithms and the hardware architecture of GPGPU, then based on Compute Unified Device Architecture(CUDA) technology, it implements the wide azimuth beam high resolution airborne SAR algorithm with GPU and optimize the algorithm for characteristics of GPU. The main content of this thesis is summarized as follows:1. As the low spectrum utilization of traditional RMA algorithm, a new plan for an improved RMA algorithm operated on a GPU is proposed. The new proposal makes it possible for the data processing procedure and the CPU/GPU data exchange to execute concurrently, especially when the size of SAR data exceeds the total GPU global memory size. In order to exploit the total computational resources of GPU, the GPU technology of shared memory and CUDA stream is used to optimize the plan. It has been shown by an experiment on an NVIDIA K40 c that the proposed plan accelerates SAR data processing by about 13 times.2. For a wide azimuth beam high resolution airborne SAR system, the non-ideal motion of radar platform velocities result in serious blurring and geometric distortion. The motion compensation need to be performed. The traditional method based on full aperture data can not meet real-time processing requirement. In this thesis, a new method based on subaperture operated on a GPU is proposed. Parallel reduce algorithm and consolidation memory access technology is performed to improve the efficiency of the algorithm. Then use MATLAB and CUDA C mixed programming algorithm to implement the various time-consuming modules and packaged into MEX file, so that MATLAB can be called directly. It has been shown by an experiment on an NVIDIA K40 c that the proposed solution accelerates SAR data processing by about 9 times.