Research On Radar Signal Sorting Algorithm Based On CUDA Architecture

The large-scale application of radar equipment has made the electromagnetic threat environment signal density faced by radar reconnaissance equipment reach millions of orders of magnitude,which puts forward higher requirements for the accuracy,reliability,and real-time performance of radar signal sorting.In this paper,the machine learning algorithm is used for learning and training under the multi-parameter feature samples of radar signals,so that the sorting accuracy in complex environments is improved.Considering that software-based radar has the advantages of a short development cycle,easy expansion,and transplantation,and convenient debugging,parallel acceleration of radar signal sorting algorithms in radar reconnaissance equipment has become one of the research hotspots in this field.In recent years,GPU parallel processors have developed rapidly.Compared with traditional CPU processors,their real-time computing performance under large data volumes is more prominent.CUDA,as a general computing development platform,can call CPU and GPU at the same time.It can be used for heterogeneous computing,giving full play to the dual advantages of CPU logic control and GPU large-scale parallel computing.This thesis takes the field of radar signal sorting in dense and complex electromagnetic environments as the background and draws on the SPRINT decision tree and random forest algorithm in machine learning to improve the accuracy of radar signal sorting in complex environments.CUDA is used to shorten the learning time of SPRINT decision tree signal sorting and random forest signal sorting under large data volume,thereby improving the real-time performance of the radar signal sorting algorithm,and verifying the consistency of parallel and serial sorting results.First,parameters such as radar signal PDW are measured to form a training set of radar signal feature vectors.The basic principles of the SPRINT decision tree algorithm and random forest algorithm are explained in-depth,and the CPU+GPU heterogeneous computing platform is discussed from the perspectives of CUDA programming structure,memory management,thread management,and library functions,which lays a theoretical foundation for the parallelized radar signal sorting algorithm.Secondly,the SPRINT decision tree signal sorting algorithm is analyzed and parallelized.The basic mathematical operation units used in the algorithm,including reduction algorithm,prefix sum algorithm,and matrix multiplication,are optimized,and the speedup ratio is verified by simulation.Analyze SPRINT decision tree signal sorting algorithm to create attribute list,find the best split attribute and split value,split attribute list,and carry out the parallel design.In the specific implementation,the overall code is re-optimized from the five aspects of thread warp differentiation,delay hiding,synchronization,program instruction optimization,and avoiding shared memory bank conflicts,shortening the training time of the algorithm model and improving the overall performance.Finally,the parallel simulation software and hardware platform of the radar signal sorting algorithm are built,and the parallel SPRINT decision tree signal sorting algorithm and parallel random forest signal sorting algorithm are simulated and verified.The simulation results show that parallel processing and serial processing are consistent in the accuracy of radar signal sorting,and for the same radar signal eigenvector training set,the random forest signal sorting algorithm based on SPRINT can further improve the sorting accuracy.In terms of acceleration effect,the overall operation time of parallel processing and the operation time of each component is greatly improved compared with serial processing,which has a good acceleration effect,shortens the model training time,and improves the real-time radar signal sorting sexual goals.