【作者】 于荣欢；

【导师】 吴玲达；

【作者基本信息】 国防科学技术大学 ， 控制科学与工程， 2011， 博士

【摘要】 随着电磁频谱在探测、干扰、通信等武器系统中的广泛使用，以掌控电磁环境为主要内容之一的制信息权已经成为决定战争胜负的关键。如何快速对电磁设备性能进行仿真计算以及对电磁环境的分布、能力进行直观描述已经成为电磁环境领域的一个重要研究方向，其核心是对于电磁辐射问题的数值计算与空间电磁环境的可视化。为解决计算规模问题，目前常用的加速方法包括基于高性能计算机和高性能图形工作站的并行方法、基于GPU通用计算的加速方法和基于PC集群的并行方法，其中基于PC集群的并行方法由于其较低的进入门槛、较高的性价比以及较好的可扩展性，受到众多研究人员的青睐，已逐渐成为当前并行计算与绘制的一个热点方向。当前，对雷达探测能力的描述和对无线通信网络的数据仿真是电磁环境领域研究的两个典型方向。本文针对这两个方向，围绕如何在PC集群上实现电磁环境并行计算与可视化这一问题开展研究，研究内容主要包括电磁环境并行计算、并行等值面提取与绘制、光线投射并行直接体绘制和PC集群环境组合优化等问题。具体而言，本文的主要工作和成果体现在以下几个方面：1)提出了基于三维标量场的电磁环境并行计算方法。首先针对电磁环境的特点，建立了一种基于三维规则标量场的通用电磁环境数据组织形式；并基于此数据组织形式，设计了雷达探测能力分布和通信电磁辐射强度分布的并行计算方法：基于抛物方程的组网雷达联合探测概率并行计算方法和基于ITM模型的通信电磁环境电磁辐射强度分布并行计算方法。2)针对等值面提取与绘制方法与场景内容的关联特点，提出了一种基于Sort-Last模式的并行等值面提取与绘制方法。该方法在任务分配阶段采用基于场景内容的静态任务分配方法；在图像合成阶段设计并实现了一种基于时间统计的自适应动态Direct-Send图像合成算法，算法针对集群节点间存在的性能差异，通过时间统计逐渐调整各节点的负载，以达到图像合成阶段的负载平衡。3)针对光线投射直接体绘制方法面向屏幕空间像素投射计算的特点，提出了一种基于Sort-First模式的并行直接体绘制方法。在并行绘制过程中，设计了一种基于屏幕网格权重计算的任务分配算法，利用投射光线在三维标量场中的穿行距离来进行网格权重的计算，可以实现以较少的负载平衡消耗达到较好的负载平衡效能。4)针对PC集群环境中不同的场景规模所适合的最优集群环境各不相同的特点，提出了一种基于场景复杂度分析的PC集群环境组合优化方法。通过对当前场景时间复杂度进行量化分析，获得当前场景规模的最优集群环境配置。针对节点间性能差异引起的负载不平衡问题，提出了一种基于节点性能测试的负载平衡修正方法。5)在PC集群并行环境中设计并实现了一个并行计算与可视化原型系统-ParaEME。该原型系统集成了本文的各项研究成果，对本文提出的各项成果和算法进行了有效的应用和验证；作为PC集群上的并行计算与绘制的一种有效解决方案，可以为基于PC集群的各种科学计算与可视化提供有效借鉴。更多还原

【Abstract】 With the wide application of electromagnetic in modern weapon systems, theinformation control which is based on electromagnetic has become the key factor of themodern war. So how to describe this environment’s distribution and its ability quicklyand racily has become an important direction of research in this field. The fastnumerical computation and visualization methods for3D space such as complexelectromagnetic environment include parallel computing and rendering method basedon high-performance computer or graphics workstations, general computation base onGPU and parallel method on PC cluster. By reason of the low barrier for entry, the lowcost and the great expansibility, PC cluster has become more and more popular withresearchers and become a research focus of the field in parallel computing andrendering.Currently，how to describe the radar’s detectability and wireless communicationnet’s ability has became two main directions in electromagnetic environment research.Based on the two directions, this thesis focuses on how to realize parallel computationand rendering of complex electromagnetic environment on PC cluster, and the researchwork consists of parallel computing of complex electromagnetic environment, parallelisosurface extracting and rendering of3D scalar field, parallel direct volume renderingand PC cluster environment optimization. In detail, the major work and the maincontributions of this thesis are described as follows:1) Proposed a3D scalar field data structure for common electromagneticenvironment and designed two parallel computing methods for radar net’s probability ofdetection and communication’s distribution of electromagnetic radiation. On the basis ofthese theories, a parallel computing framework for complex electromagneticenvironment is built.2) According to the characteristics of Isosurface Rendering, a Parallel IsosurfaceRendering method based on Sort-Last is proposed. This method adopts static taskallocation in load balance and designs a dynamic time-based adaptive statisticalDirect-Send image synthesis algorithm in the image synthesis stage.3) In allusion to the characteristics of Light projection for direct volume rendering,we propose a parallel direct volume rendering method based on Sort-First model. Thismethod designs a task allocation algorithm based on the weight of the screen area, andcan achieve better load balancing performance in image composing.4) In PC Cluster, different node has different ability, and different sence is fit fordifferent cluster size. According to these characteristics, we proposed an optimizedcombination method for PC cluster environment. In this method, we use the timecomplexity of the current scene to achieve the optimal size of the current cluster environment. A node-based load balancing method is also proposed in order to solve theload imbalance caused by the difference between nodes.5) In the PC cluster parallel environment, we designed and implemented aprototype parallel computing and visualization system-ParaEME, this system inheritsthe results of this thesis and validates the methods in the thesis. As an effective parallelcomputing and rendering solution on PC cluster, ParaEME can provide an impactfulreference for common scientific computing and visualization on PC cluster.更多还原