Radar Forward-looking Imaging Algorithm Based On GPU Platform

Metamaterial aperture radar imaging is a novel method of radar forward-looking imaging.It solves the problems of the complicated design of waveforms in system structure and the harsh application conditions of imaging algorithms in microwave correlation imaging.However,its application platform has high requirements for real-time data processing,and it has a large amount of calculations.The hardware that processes radar signals must have powerful computing capabilities.Traditional radar signal processing process mainly relies on the DSP+FPGA platforms.Although the combination of the two platforms can handle large-scale data,the usability and portability are not high.Therefore,under the background of the ?×××? national defense project,the main work of this thesis is that how to implement the parallel design of the backscattered coefficients estimation algorithm for the metamaterial aperture radar forward-looking scene on the GPU platform.In this thesis,the parallel design based on the GPU platform imaging methods have improved the computational efficiency and shorten the operation time greatly,and solved the technical problems in practical engineering area.The main work in this thesis is as follows:1.After the structure and working principle of the metamaterial aperture radar forward-looking imaging system is introduced,an imaging model is established,the distribution of the backscattered coefficients of the scene targets is estimated by using the sparse reconstruction methods and reconstruction of the scene is realized.It mainly includes the introductions of the OMP algorithm and CG algorithm basic implementation process.Furthermore,it verifies the feasibility of the two algorithms in the forwardlooking imaging of the metamaterial aperture radar using simple scene simulation experiments.2.After introducing the GPU parallel technology and the CUDA programming model,the kernel design of the typical matrix computing units are implemented on the GPU.Combining the design process of the parallel computing units provided some commonly used optimization schemes and provided the basic guarantee for the overall design of the algorithms.In order to reduce the computational complexity,this paper is based on two different reconstruction agorithms implementation features,OMP algorithm uses the matrix block recycling iteration method to replace the matrix inversion process;and in CG algorithm,using the conjugate gradient method to replace the traditional Cholesky matrix decomposition to invert the matrix directly.In order to improve the computational efficiency,the parallel implementation method of the above two metamaterial aperture radar forward-looking imaging algorithms on GPU platform is studied.So this thesis completes the GPU parallel design of the two algorithms according to the optimization methods,mainly including the analysis of the characteristics of each algorithm,GPU processing implementation flow of the algorithm,completing the parallel computing of the two modified matrix inversion processes,and other kernel functions are designed.3.This thesis uses a variety of platforms to compare the imaging effects of the algorithms,requiring operation of multiple source code files frequently,so the experimental process is complex.Inspired by software radar,the demonstration software of radar forward-looking imaging system based on QT platform is designed.On this good human-computer interaction design,the type of the algorithms and the operating platforms can be selected and parameters can be adjusted conveniently.Therefor,it simplifies the operation process.And the results are visualized so the imaging results are easy to be observed.Through the simulation on different platforms,the correctness of the algorithm migration is verified.Comparing to the processing time of the algorithm on different platforms under the same experimental conditions,the stability and high efficiency of the GPU platform is verified.The operation efficiency of imaging algorithms based on GPU platform can be effectively improved,the efficiency of OMP algorithm is improved by 9.3 times compared to the MATLAB platform and the CG algorithm is improved by nearly 17 times compared to the MATLAB platform.Hence,the data processing meets requirements.