LDPC码并行译码算法的研究及其基于CUDA的实现

发布时间：2019-06-06 17:50

【摘要】：低密度校验码(Low-Density Parity-Check,LDPC Codes)最早由Gallager博士在1962年提出。随着计算机仿真水平的提高和现代编码理论的兴起,近年来LDPC码获得了信道编码理论研究者的广泛关注。作为一种具有接近Shannon限性能的好码,LDPC码不仅设计简单,而且在纠错能力、译码性能和硬件实现复杂度方面具有较大优势,因此在许多通信系统中具有广阔的应用前景。近年来,LDPC码的高效编译码器实现逐渐成为研究工作的一个热点,而且科研人员对编译码器的数据传输速率的要求越来越高。通常,LDPC码的编译码器设计可以通过不同的计算平台来实现,大致上包括CPU,FPGA,GPU等等。近年来,利用图像处理器(Graphic Processing Unit,GPU)的并行计算功能来进行算法加速成为一项新的研究热门领域。针对基于CPU平台的LDPC码译码可能带来的延迟问题,本文提出了一种基于CUDA架构的LDPC码的并行译码方案。CUDA是一种全面支持GPU通用计算的并行架构。文章首先研究了CUDA架构下的硬件系统和软件体系,介绍了LDPC码的基本原理并且重点介绍了它的归一化最小和译码算法,然后结合两者以及本文使用的软硬件环境,实现了CUDA并行译码实现时的具体编译码方案,最后分析了影响系统译码速度的因素。并与CPU平台上LDPC码的串行译码的仿真性能进行了比较,验证了采用CUDA并行译码在保证译码性能的同时,实现了对译码过程的加速,有效减少了译码时延,提高了系统吞吐量。
[Abstract]:Low density check code (Low-Density Parity-Check,LDPC Codes) was first proposed by Dr. Gallager in 1962. With the improvement of computer simulation level and the rise of modern coding theory, LDPC codes have been widely concerned by channel coding theory researchers in recent years. As a good code with close to Shannon limit performance, LDPC code is not only simple in design, but also has great advantages in error correction ability, decoding performance and hardware implementation complexity, so it has a broad application prospect in many communication systems. In recent years, the implementation of efficient codecs of LDPC codes has gradually become a hot topic of research work, and researchers have higher and higher requirements for the data transmission rate of codecs. In general, the encoder and decoder design of LDPC codes can be realized by different computing platforms, including CPU,FPGA,GPU and so on. In recent years, using the parallel computing function of image processor (Graphic Processing Unit,GPU) to accelerate the algorithm has become a new research hot field. In order to solve the problem of delay caused by LDPC code decoding based on CPU platform, a parallel decoding scheme of LDPC code based on CUDA architecture is proposed in this paper. Cuda is a parallel architecture that supports GPU general computing in an all-round way. This paper first studies the hardware system and software system under CUDA architecture, introduces the basic principle of LDPC code and focuses on its normalization minimum and decoding algorithm, and then combines the two and the software and hardware environment used in this paper. The encoding and decoding scheme of CUDA parallel decoding is implemented, and the factors that affect the decoding speed of the system are analyzed. The simulation performance of serial decoding of LDPC codes on CPU platform is compared with that of LDPC codes on LDPC platform. It is verified that CUDA parallel decoding not only ensures the decoding performance, but also accelerates the decoding process, effectively reduces the decoding delay and improves the throughput of the system.
【学位授予单位】：西安电子科技大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TN911.22

【相似文献】