软件化雷达任务模块自动部署系统的设计与实现

【作者】 李硕；

【导师】 徐国伟； 于涛；

【作者基本信息】 天津工业大学 ， 控制工程（专业学位）， 2022， 硕士

【摘要】 随着集成电路和数字化技术的迅速发展,基于定制硬件板卡的传统雷达系统逐渐显现出诸多弊端,如软硬件成本高、无法升级、灵活性低、开发效率低等。在此趋势下,传统雷达正朝着高灵活性、可模块化、可扩展化和高性能的方向发展,即软件化雷达。传统雷达任务的分配需要人工操作,这种分配方式不仅要求设计者具备较高的雷达知识水平和丰富的任务分配经验,且分配效率低下,无法满足当前软件化雷达的设计标准。因此,本文针对软件化雷达的任务分配问题进行了深入研究。本文首先分析了“软件化雷达”的设计准则。针对硬件资源有限条件下,传统分配方式存在的任务调度困难、资源分配不均、配置效率低、受限操作者水平等问题,提出一种利用二次分配问题(Quadratic Assignment Problem,QAP)解决雷达任务自动部署的设计方案。该方案通过封装雷达处理流程参数,实现流程任务的模块化,进而在系统吞吐时间的约束下,利用构建的分配算法,生成任务模块与处理器核的逻辑关系映射表,实现任务模块的自动部署。在任务模块的自动部署过程中,针对任务模块数量大于处理器核数量的实际情况,本文制定了任务模块的划分规则,编写划分算法。该算法首先找到任务流图的关键路径,对其进行无环划分和封装,使任务模块不仅被规整为与处理器核数目相同的任务集合,缩短模块间的通信时间,实现“1对1”的关系映射,而且任务集合之间没有循环连接,避免因划分而导致的额外调度开销。随着雷达任务流图中任务数量的增加,可行分配的数目呈指数级增长,导致QAP的算法在分配任务时宕机。为解决这一问题,提出采用蚁群算法(Ant Colony Optimization,ACO)辅助QAP分配,并对蚁群算法进行优化和改进,将改进后的蚁群算法与QAP相结合,编写任务分配算法。在任务模块部署至处理器核时,算法利用底层硬件通信带宽参数的差异,搜索通信时间最短,系统延迟最小的分配策略。仿真实验表明,本文方案有效划分了雷达任务流图,实现了任务模块与处理器核之间“多对少”的自动部署。通过多组实验结果的比较分析,划分算法的使用和蚁群算法的改进不仅减少了硬件板卡的使用,且将系统延迟时间降低了近7%。最后,通过与人工调试结果的对比分析,本文分配方案可在较短的时间内搜索到同等质量的分配策略。更多还原

【Abstract】 With the rapid development of integrated circuit and digital technology,the traditional radar system based on customized hardware board shows many disadvantages,such as high cost of software and hardware,unable to upgrade,low flexibility and low development efficiency.In this trend,traditional radar is developing towards the direction of software radar with high flexibility,modularity,scalability and high performance.The allocation of traditional radar tasks requires manual operation.This allocation method not only required designers to have extensive radar knowledge and experience in task assignment,but also had low assignment efficiency and could not meet the current design standards of software radar.Therefore,this dissertation delves into the task allocation of software radar.This dissertation analyzes the design criteria of "software radar".Under the condition of limited hardware resources,the traditional allocation methods has some problems,such as difficult task scheduling,uneven resource allocation,low allocation efficiency,limited operator experience and so on.This dissertation proposes a design scheme to solve the automatic deployment mapping of radar tasks by using quadratic assignment problem(QAP).The scheme realizes the modularization of process tasks by encapsulating radar process data parameters.Under the constraint of system specific throughput time,the scheme generates the logical relationship mapping table between task module and multiprocessor core,and realizes the automatic deployment of task module.In view of the problem that the number of task modules is larger than the number of processor cores.The task module partition algorithm of radar flow graph is compiled in this dissertation.The algorithm first finds out the critical path of the task flow graph,and then divides and encapsulates it without loops.The processed task flow graph not only arranges the task modules into task sets with the same number of processor cores,but also realizes the "1-to-1" relationship mapping.Since there is no cyclic connection between task sets,the additional scheduling overhead caused by partitioning is avoided.As the increase of the number of tasks in the radar flow graph,the number of allocation strategies increases exponentially,resulting in the collapse of QAP algorithm when allocating task sets.In this scheme,Ant Colony Optimization(ACO)assisted QAP allocation is used to solve this problem.And the scheme optimizes the ant colony algorithm,combines the improved ant colony algorithm with QAP.and write the task allocation algorithm code.The task assignment algorithm can search the mapping strategy with the shortest communication time and minimum system delay by using the difference of the communication bandwidth parameters.The simulation results show that the radar task allocation scheme can effectively divide the radar task flow diagram,and realize the "many to few" automatic deployment between the task module and the processor core.Through the comparative analysis of multiple groups of experimental results.The use of partition algorithm and the improvement of ant colony algorithm not only reduce the use of hardware board,but also reduce the system delay time by nearly 7%.Finally,by comparing with the manual debugging,the allocation scheme in this dissertation can search for high-quality distribution strategy in a short time.更多还原