Research And Implementation Of Subarray Adaptive Beamforming Technology

In order to solve the problems of the limited number of equipment,inflexible deployment,single training environment scenario and excessive resource consumption of training for radar operators using radar training,countries generally use computer simulation technology to design radar simulation training software to solve.However,using the central processing unit(CPU)to realize the radar training system is difficult to solve the real-time problems caused by many complex radar algorithms.In recent years,the continuous development of graphics processing Unit(GPU)and the advent of CUDA(Compute Unified Device Architecture)provide a new direction for real-time radar simulation training systems.Based on the above background,this thesis implements the software of radar simulation training system using the heterogeneous framework of CPU + GPU,and solves the real-time problem of the system.This thesis starts from the actual project,the concrete work is as follows:(1)Based on the software and hardware model of GPU,the whole scheme of radar simulation training software is designed with the characteristics of CPU and GPU,and the system is divided into four sub-modules: scene construction,echo simulation,signal processing and dominant control.In view of the real-time problem of the system,the CUDA-Arrayfire joint programming method is proposed,that is,CUDA is used to complete a large number of cycle calculations of four sub-modules,and arrayfire to complete relevant matrix operation of the big data.With this method,the radar simulation training system does not need to transfer large data with memory frequently.(2)Aiming at the real-time problem of simultaneous simulation of multiple types of clutter based on DEM data,a fast clutter simulation method based on merged CUDA kernel function is proposed.This method uses the GPU to optimize the DEM data process,and merges the kernel functions that generate ground,sea,and weather clutter into one,thereby reducing the memory access overhead and kernel function startup overhead when the clutter simulation program is executed,and improving the execution speed of the clutter module.This method is compared with the CUDA stream parallel method,and the results show that the merger kernel function method is more efficient under the current framework.(3)Using GPU to optimize chaff corridor simulation,short-pulse suppression,pulse compression(PC),sidelobe cancellation(SLC)and sidelobe blanking(SLB)algorithms.Aiming at the large number of unsynchronized threads in the CUDA kernel function of the above algorithm,a method for converting the logic judgment process into calculating flags is proposed.This method optimizes the execution efficiency of the CUDA thread and increases the execution speed of the radar algorithm.(4)Under the architecture of CPU + GPU,Qt,a cross-platform development framework,is used to design and realize the requirements of drawing the track of target or clutter area by mouse and real-time data display.And the radar simulation training software optimized by GPU is tested to verify the correctness and real-time performance of the system.In this thesis,a radar simulation training system software is implementes on the CPU + GPU platform,which has the characteristics of simple operation,beautiful interface and real-time performance.And the software solves the limitations of the installed radar and CPU software and has certain engineering significance.