CUDA加速城市宏小区下的光线追踪算法

39 浏览量更新于2024-08-26 收藏 1.43MB PDF 举报

"加速光线跟踪算法在城市宏蜂窝环境下的应用" 在《Urbanmacro单元下的加速光线跟踪算法》这篇研究论文中，作者Z.-Y. Liu, L.-X. Guo, 和 X.-W. Guan来自西安电子科技大学物理与光电子工程学院，探讨了在复杂城市宏蜂窝环境中如何提高基于光线跟踪的传播预测模型的效率。光线跟踪算法因其高效率和可靠的预测精度而被广泛采用。然而，当应用于大范围覆盖区预测时，计算时间会随着预测点总数的增加呈线性增长，这在处理复杂地理环境的城市宏蜂窝场景时会导致计算时间过长，甚至成为难以承受的瓶颈。为了解决这一问题，论文提出利用Compute Unified Device Architecture (CUDA)技术来加速计算过程。CUDA是NVIDIA开发的一种并行计算平台和编程模型，它提供了细粒度的数据并行性和线程并行性，能够充分利用GPU（图形处理器）中的数千个线程。通过CUDA，作者将覆盖预测问题进行分解，使得计算任务能够在多个线程之间并行执行，从而显著提升计算速度。在光线跟踪算法的基础上，论文还引入了几种加速技术。这些技术可能包括但不限于以下几点： 1. 空间分区：将预测区域划分为多个子区域，每个子区域独立进行光线跟踪计算，然后合并结果。这可以减少不必要的计算，提高计算效率。 2. 早期剔除策略：在光线追踪过程中，通过判断光线与物体的相对位置，提前排除不会发生交互的光线，避免无效计算。 3. 光线批处理：将多条光线打包成一批进行计算，利用GPU的并行性同时处理大量光线，提高计算吞吐量。 4. 缓存优化：通过合理组织数据存储和访问，减少内存访问延迟，提高计算速度。 5. 动态调度：根据计算负载动态调整线程的分配，确保GPU资源得到充分使用。通过这些优化策略的结合，论文中的加速光线跟踪算法能够在保持预测精度的同时，大大缩短了在城市宏蜂窝场景下进行无线传播预测的时间，为大规模无线通信网络的设计和优化提供了更高效的方法。这篇论文对城市宏蜂窝环境下无线传播预测的计算效率问题进行了深入研究，提出了基于CUDA的光线跟踪算法优化方案，对于未来智能城市的通信网络规划与性能提升具有重要的理论与实践意义。

Accelerated ray tracing algorithm under urban macro cell

Z.-Y. Liu*, L.-X. Guo, X.-W. Guan

School of Physics and Optoelectronic Engineering,Xidian University, Xi’an, Shaanxi, China 710071

ABSTRACT

In this study, an ray tracing propagation prediction model, which is based on creating a virtual source tree, is used

because of their high efficiency and reliable prediction accuracy. In addition, several acceleration techniques are also

adopted to improve the efficiency of ray-tracing-based prediction over large areas. However, in the process of employing

the ray tracing method for coverage zone prediction, runtime is linearly proportional to the total number of prediction

points, leading to large and sometimes prohibitive computation time requirements under complex geographical urban

macrocell environments. In order to overcome this bottleneck, the compute unified device architecture (CUDA), which

provides fine-grained data parallelism and thread parallelism, is implemented to accelerate the calculation. Taking full

advantage of tens of thousands of threads in CUDA program, the decomposition of the coverage prediction problem is

firstly conducted by partitioning the image tree and the visible prediction points to different sources. Then, we make

every thread calculate the electromagnetic field of one propagation path and then collect these results. Comparing this

parallel algorithm with the traditional sequential algorithm, it can be found that computational efficiency has been

improved.

Keywords: Electromagnetic propagation, ray tracing, urban macro cell, CUDA

1. INTRODUCTION

The rising demand for mobile communications, particularly in dense urban areas, has led to the adoption of

macrocellular systems to accommodate an influx of users despite limited frequency resources [1], [2]. The successful

implementation of these systems requires a fast and accurate propagation prediction model for system deployment. Ray

tracing techniques as a site-specific prediction model can offer significant advantages in terms of the accurate and

comprehensive prediction of radio channel characterization (such as the amplitude, delay, and direction of arrival of

multipath echoes created by the propagation environment). Using ray tracing models as tool, many scholars have

investigated the effects of different ray permutations, wall characteristics, antenna position offsets and database

inaccuracies on predicted received power [3], and the delay characteristic of signal reception in the coverage area [4].

Unfortunately, such techniques, although accurate, turn out computationally intensive when a great number of

intersection tests and electromagnetic field calculations may be required because of the increase in interested prediction

points [5]. The result is disadvantageous, as the run time becomes linearly proportional to the total number of prediction

points. Inorder to overcome this problem, in [6], the discretization and geometric preprocessing of the environment was

proposed. O’Brien et al. [7] adopted an approach of a transmitter to a multiple receiver technique, resulting in greatly

reduced computational times.

The current study is focused on improving the prediction efficiency of macrocellular ray tracing algorithm based on

graphics processing unit (GPU). This paper is organized as follows. The proposed macrocellular ray tracing model [8],

which takes into account the radio channel characterization between base stations and users in complex macrocellular

envirenment, is introduced in Section 2. The main focus is described in Section 3, where the parallel computing platform

and programming model CUDA are discussed. The results computed by parallel algorithm and the prediction results

used the traditional sequential algorithm are compared in Section 4. Finally, the conclusions are drawn in Section 5.

2. RAY TRACING MODEL

The ray-tracing-based prediction model proposed in [8] is used to predict the propagation in complex macrocellular

envirenment. The process of performing the model can be divided into five main steps. In Figure 1, a simplified flow

diagram of implementing the macrocellular three-dimensional (3D) ray tracing model is presented. The process of

performing the model can be divided into five main steps.

High-Performance Computing in Remote Sensing V, edited by Bormin Huang, Sebastián López, Zhensen Wu,

Jose M. Nascimento, Boris A. Alpatov, Jordi Portell de Mora, Proc. of SPIE Vol. 9646, 96460U

Proc. of SPIE Vol. 9646 96460U-1

Downloaded From: http://spiedigitallibrary.org/ on 10/22/2015 Terms of Use: http://spiedigitallibrary.org/ss/TermsOfUse.aspx

下载后可阅读完整内容，剩余5页未读，立即下载

weixin_38570854

粉丝: 5
资源: 931

CUDA加速城市宏小区下的光线追踪算法

MFC+C++：详解李海广的光线跟踪算法实现与OpenGL应用

多光源室内渲染效果的光线跟踪C++源码

光线追踪算法详解与优化

光线跟踪算法技术

基于粒子滤波的视频目标跟踪算法matlab源程序

openGL和C++实现光线跟踪作业

基于Meanshift的单目标跟踪算法【matlab+c两个版本】

GPU加速下的Moller-Trumbore光线三角形交叉算法

高性能计算驱动的光线追踪模拟加速算法

多目标跟踪算法优化：MATLAB实现多假设跟踪

最新资源