The Research On Parallel Radar Signal Processing Based On Multi-core And Many Core Architecture

The programmers of traditional radar signal processing system based on FPGA + DSP require a lot of professional knowledge about hardware,and its development cycle is very long.Besides,the increasingly strong function of radar and more and more complexity of radar signal processing algorithm make the development cost increase several times over.Software defined radar is put forward for the development of radar signal processing system provides a solution,but most serial software radar signal processing system is difficult to meet the requirement of real-time radar signal processing system.Nowadays,With the development of multi-core and many core devices technology and parallel computing,it provides the possibility to meet the requirement of real-time software define radar,and parallel radar signal processing has become the focus of radar signal processing system.In multi-core CPU platform,parallel strategies are diverse,including data parallel,task parallel and pipelined parallel.Because parallel efficiency and application scenarios of different parallel strategies are different,how to choose the parallel strategy for software defined radar becomes a realistic problem urgently to be solved.At the same time,with the development of hardware technology,there are more and more general architectures of computing devices,such as CPU,GPU,FPGA,MIC,etc.Designing a cross-platform parallel algorithm for radar signal processing becomes a possible direction of radar signal processing,which provides portability of the radar signal processing system and is of great significance in practice.In order to study the advantages and disadvantages of the radar signal processing system of different parallel strategies,a parallel radar signal processing system based on multi-core CPU is proposed in this thesis.Serial algorithm of radar signal processing system is analyzed.Three kinds of the radar signal processing algorithms based on three different parallel strategies are proposed in this thesis,and theory analysis on different parallel strategies are made,and experiments are designed on multi-core CPU platform.NO 1:a radar signal processing algorithm based on the data parallel is proposed,in which the pulse groups are parallel processed,different pulse group process are mapped to different threads.NO2: a radar signal processing algorithm based on pipeline parallel methods is proposed,according to the principle of load balance,all the processes are divided into multiple task groups which have similar load,and different task groups are mapped to different threads.NO3:a radar signal processing algorithm based on task queue is proposed,task is partitioned and put into the task queue;according to which the tasks are scheduled on different threads.Finally,this thesisanalyzes the experimental results and compares the advantages and disadvantages of different parallel strategies.In order to study the cross-platform radar signal processing algorithm,a parallel CFAR algorithm based on OpenCL is proposed.Parallelism of CFAR detection algorithm is analyzed,and based on the independence of the calculation procedure of each data block,parallel CFAR detection algorithm is designed,in which calculation procedure of each data block are mapped to different work-item of OpenCL.The CFAR detection algorithm is implemented on both the GPU platform and FPGA platform,an optimization method combining setting calculation on each thread,calculation process optimization,local storage optimization and merge kernel function is designed to optimize the parallel program and further explore the parallel efficiency;on FPGA platform,in addition to using the GPU optimization method,kernel replication methods are used in order to make full use of FPGA and further improve the speedup.The experiment results show that the CFAR detection algorithm obtained 10.3008 times speedup on FPGA platform and 193.7886 times speedup on GPU platform.