Research On Acceleration Algorithm For Pulse-Doppler Radar Signal Processing Based On CUDA Heterogeneous Parallelization

The electromagnetic environment of the informationized battlefield is becoming increasingly complex,and the amount of radar echo data is growing,resulting in the nonnegligible contradiction between the complexity of radar signal processing and the requirement of real-time processing.Software-based radar has multiple advantages of the short development cycle,low R&D cost,and potent portability.Building parallelization algorithms for radar signal processing based on modeling and simulation to realize overall acceleration is one of the hot topics in this field.In recent years,GPU devices based on parallel architecture have performed particularly prominently in real-time computing of large-scale resources.As a general-purpose parallel computing platform,CUDA supports heterogeneous collaboration between CPU and GPU,which fully combines the logic control that CPU is good at and the parallel computing that GPU is good at.Based on the parallel computing platform of Pulse-Doppler Radar and CUDA,this thesis focuses on the heterogeneous parallel radar signal processing algorithm of CPU+GPU.Combined with the actual simulation scene,the result verification and speedup analysis of the serial and parallel algorithms of the key links of radar signal processing are completed.Firstly,the key technologies of pulse Doppler radar signal processing are introduced.The key algorithms such as LFM signal,pulse compression,MTI processing,MTD processing and CFAR detection are studied and simulated.Secondly,the heterogeneous parallel system of CPU-GPU is introduced.On this basis,the thread hierarchy and memory model of CUDA are introduced from the hardware point of view.Then,from the software point of view,the kernel function,software architecture and programming model of CUDA are introduced.Thirdly,the parallelism of key algorithms of radar signal processing under the heterogeneous cooperation of CPU and GPU is studied,and the CUDA parallel acceleration schemes of key operation units,frequency domain pulse compression,MTI processing,MTD processing,two-dimensional CFAR detection are designed.Meanwhile,from the points of view of code optimization,rational allocation of threads,alignment and merging of global memory access,and avoidance of shared memory Bank conflicts,the comprehensive optimization strategies of key algorithms for CUDA parallel programming are proposed.Finally,the GPU and CPU hardware facilities and software development environment needed for experimental testing are configured,and the parallel simulation experiments of radar signal processing algorithms are designed and completed.By analyzing the simulation results of CPU+GPU heterogeneous parallel algorithm,the correctness of radar signal processing parallelization algorithm on CUDA platform is verified.In addition,by comparing the running time of kernel functions of serial and heterogeneous parallel algorithms,the parallel acceleration effects of each key algorithm and the whole algorithm of radar signal processing are tested.In this thesis,the cross platform development tools are preferred in the simulation design of conventional radar signal processing tasks,which improves the reusability and scalability of the algorithm.The final result demonstrates that the running speed of the parallelized algorithm has been significantly improved,which has a certain significance for engineering reference.