OFDM定时同步并行算法研究与实现

发布时间：2018-10-10 20:25

【摘要】：在无线通信领域中,正交频分复用(Orthogonal Frequency Division Multiplexing, OFDM)技术由于其抗衰能力强、抗符号间干扰能力强、高频谱利用率等优点而得到广泛的应用。但是OFDM技术也存在着固有的局限性,主要表现在OFDM系统对同步误差十分敏感。没有准确的同步算法,就无法保障可靠的数据传输,同步技术直接关系到通信系统的性能,而OFDM定时同步又是其它所有同步的前提。所以如何实现定时同步也早已成为该领域中的重要问题。随着数据的存储、传输、处理的工作量越来越大,如何提高定时同步算法的运行速度正逐渐成为当今无线通信系统中的热点问题之一。图形处理器(Graphics Processing Unit, GPU)的快速发展为这个问题带来了一个新的解决方案。由于GPU的计算能力以及它的存储器带宽已经达到了同时期中央处理器(Central Processing Unit, CPU)的百倍,甚至更多,因此,GPU为高性能计算提供了一个高效的平台。但是由于GPU硬件结构的限制,编程人员很难有效地利用GPU的资源进行通用计算,而英伟达(NVIDIA)公司推出的统一计算设备架构(Compute Unified Device Architecture, CUDA)则改变了这一现状。CUDA可以利用GPU强大的计算能力进行通用计算,并且以C语言为基础,使编程人员能够非常容易地利用它在GPU上进行程序的开发。本文探讨了利用CUDA架构如何实现定时同步样算法的并行化,并提高该算法的运行速度。主要工作包括：(1)在CPU上实现了基于循环前缀的定时同步串行算法。(2)将串行算法中可以并行的部分进行归纳及改写,然后在GPU上执行,实现对算法的并行化。(3)利用Matlab编写定时同步程序,对串行算法和并行算法的仿真结果进行正确性验证,并将串行算法、并行算法执行时所耗的时间进行对比实验。结果表明,基于GPU的并行算法所消耗的时间要远远低于基于CPU的串行算法。(4)最后,根据结果,指出在GPU上实现并行计算的瓶颈,并给出一些对应的优化策略。
[Abstract]:In the field of wireless communication, orthogonal Frequency Division Multiplexing (Orthogonal Frequency Division Multiplexing, OFDM) technology has been widely used because of its strong anti-fading ability, strong anti-symbol interference ability, high spectral efficiency and so on. However, OFDM technology also has inherent limitations, mainly because OFDM system is sensitive to synchronization error. Without accurate synchronization algorithm, reliable data transmission can not be guaranteed. Synchronization technology is directly related to the performance of communication system, and OFDM timing synchronization is the premise of all other synchronization. So how to realize timing synchronization has already become an important problem in this field. With the increasing workload of data storage, transmission and processing, how to improve the speed of timing synchronization algorithm is becoming one of the hot issues in wireless communication systems. The rapid development of graphics processor (Graphics Processing Unit, GPU) has brought a new solution to this problem. Because the computing power and memory bandwidth of GPU has reached 100 times or even more than that of CPU (Central Processing Unit, CPU) in the same period, GPU provides an efficient platform for high performance computing. However, due to the limitation of GPU hardware structure, it is very difficult for programmers to utilize the resources of GPU effectively for general computation. Nvidia (NVIDIA) 's unified computing device architecture, (Compute Unified Device Architecture, CUDA), has changed the situation. CUDA can use GPU's powerful computing power to do general computing, and it is based on C language. Make it very easy for programmers to develop programs on GPU. This paper discusses how to realize the parallelization of timing synchronization algorithm using CUDA architecture and improve the speed of the algorithm. The main works are as follows: (1) the timing synchronous serial algorithm based on cyclic prefix is implemented on CPU. (2) the parallel parts of the serial algorithm are summarized and rewritten, and then executed on GPU. (3) using Matlab to write timing synchronization program to verify the correctness of the simulation results of serial algorithm and parallel algorithm, and compare the time of serial algorithm and parallel algorithm execution. The results show that the parallel algorithm based on GPU consumes much less time than the serial algorithm based on CPU. (4) finally, according to the results, the bottleneck of parallel computing on GPU is pointed out, and some corresponding optimization strategies are given.
【学位授予单位】：东北大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TN929.53

【相似文献】