Surveillance Radar Signal Processing's Software Design Based On GPU

With the improvement of the technology in the aspect of surveillance radar signal processing,the design pattern of surveillance radar will gradually develop to the software.This pattern is "software technology as the center,for the actual needs",which is based on GPU.In recent years,GPU(graphic process unit)has a powerful floating-point computing ability and high storage bandwidth,and has the advantages of low development cost,short period,portability and so on.Therefore,the thesis will achieve the parallel research of the surveillance radar signal processing by the software development pattern based on GPU.These will achieve the platform transplant and the acceleration of algorithm,so we can solve the problem of real-time and function expansion to some extent.The research content of this thesis has a practical role in promoting the research and establishment of software radar system in laboratory.Firstly,the thesis will discuss the background and significance of the research.The research status and development trend of the GPU general technology and its application in radar signal processing are discussed.Then GPU general technology and radar signal processing algorithm are studied,and the parallel decomposition process of radar signal processing algorithm is described in detail.Next we can realize the parallel optimization and acceleration of the whole algorithm and its parallel unit through CUDA programming based on CUDA model,and the maximum acceleration of algorithm is achieved to meet the real-time requirement of signal processing.Finally,we will complete the parallel processing research and software design.The main works is summarized as follows:(1)Analyzing the CUDA programming model,storage model,software system and programming language,and then we research the principle and key technology points of signal processing algorithms,such as PC,MTI,MTD and CFAR.Through the parallel analysis of the algorithm,the processing units which can be used for parallel operation are found out,which provides theoretical basis and technical basis for the parallel analysis of the signal processing algorithms in the following paragraphs.(2)According to the CUDA software model,the CUDA parallel scheme of signal processing is determined and the CPU+GPU cooperation mode is used to complete the parallel software design of PC,MTI,MTD and CFAR algorithm through coarse-grained and fine-grained parallel mode.Finally,the running time of the algorithm on GPU and CPU platform is obtained through experiments,which is to verify the acceleration effect of the parallel computing.(3)Using the optimization strategy of CUDA program,this thesis analyzes the factors that affect the acceleration efficiency based on CUDA platform,and then optimizes the signal processing algorithm on CUDA in the aspects of data communication,memory access and instruction flow.Then the overall performance analysis of the CUDA program is completed by increasing the amount of data,and the experiment shows that the signal processing algorithm is improved by 10~30 times based on GPU.