An Algorithm Research And Software Implementation Of Video SAR Imaging Based On GPU Parallel Architecture

SAR(synthetic aperture radar),with the strengths of long detection distance,being little affected by climate and light,and so on,plays an important part in national defense and military field and in people's life.Due to the large amount of data it needs to process and the complex operations it needs to proceed,it is hard to achieve high frame rate ViSar(Video SAR).In order to meet the real-time processing requirement of the monitoring systems and achieve the desired effect of fast imaging and continuous display,the optimization of the imaging algorithm and the construction of the most suitable hardware and software environment are two necessary aspects.The SAR imaging algorithms develop rapidly and become much more mature than before.The key to improve the execution efficiency of the imaging algorithms is to find the most suitable hardware and software environment.The thesis is to solve the issue of algorithm research and software implementation of video SAR imaging based on parallel architecture,and it mainly discusses the implementation and optimization process of SAR imaging algorithm on GPU parallel computing platform—CUDA,and the efficiency of the imaging is boosted by about 24 times in the end.The main contents of the thesis are divided into the following four parts.1.This part mainly studies the comparison and selection of the imaging algorithm.Starting from the general principles of SAR imaging,the thesis compares three kinds of high resolution SAR imaging algorithms(RD algorithm,CS algorithm and ?K algorithm),mainly based on their characteristics,processing flows and the insufficiencies,then the thesis chose ?K algorithm as the suitable video SAR imaging scheme.2.This part mainly studies the simulation of ?K algorithm and the hardware mapping to achieve parallel computing.Simulate and implement the ?K algorithm on the MATLAB platform and pass the imaging test.In order to take advantage of multi-core parallel computing ability of the GPU parallel computing platform—CUDA,the hardware mapping of the selected imaging algorithm is necessary.According to the concrete operations involved in the imaging algorithm,abstract out interfaces of matrix and complex,and implement them on CUDA platform while considering both the extensibility and the reusability of the program.Then use the two interfaces to map the imaging algorithm to CUDA platform.3.This part mainly studies the optimization of the execution efficiency of the imaging algorithm.In order to maximize GPU execution efficiency,the thesis analyzes the hardware resource allocation and the execution mode of GPU,and according to the differences of execution mode between GPU and CPU,the thesis analyzes two aspects that restrict GPU's operation efficiency.On that basis,it discusses the optimization strategy from three aspects: wrap branch,resource allocation and time delay hiding,and it gives the optimal configuration rule of the kernel function.Then it improves the accuracy of matrix multiplication,which is used most frequently in the algorithm,and it further optimizes the speed and times of video memory access.By changing the operation structure of matrix multiplication and partitioning the matrix,it further reduces the times of video memory access,and the execution speed is improved.Finally,with the help of the performance analysis tools of Visual Studio,the execution time of the high-density computation part of the imaging algorithm is reduced,and by analyzing video memory access,the overall implementation of the algorithm is optimized.4.This part mainly studies the modular design of the imaging software.According to the required function,the software can be divided into three parts—radar interaction part,imaging algorithm processing part,image and waveform display part.The radar interaction part sets waveform parameters and radar hardware parameters by serial port communication,and controls radar to transmit and receive waveform signals,and controls upper computer(the host which is running video SAR imaging algorithm)to receive return signals which are processed by radar by UDP communication.The imaging algorithm processing part calls and runs the imaging algorithm on the upper computer based on the input imaging parameters and the selected imaging mode.The image and waveform display part displays the image and waveform outputted by the imaging algorithm with the function of setting display mode and controlling display process.Finally,by analysing the imaging quality,it is proved that the video SAR imaging algorithm based on CUDA platform can obviously improve imaging speed and achieve real-time imaging on the premise of not sacrificing imaging quality.