CAJ | 학술논문

随着5G标准化工作进入准备阶段，以Massive MIMO、有源阵列天线等为代表的5G多天线技术的仿真研究也逐步深入。这些技术涉及天线数目多、系统带宽大，极大提升了通信系统仿真工作的总体运算量。为了使仿真总时间不随运算量增加而显著拉长，提出了一种利用Kepler架构GPU中CUBLAS矩阵库和动态并行技术来加速多天线DFT （离散傅里叶变换）处理过程的方法。实验结果表明，相对单核单线程的CPU程序，该方法能够将DFT处理速度最高提升60倍。
수착5G표준화공작진입준비계단，이Massive MIMO、유원진렬천선등위대표적5G다천선기술적방진연구야축보심입。저사기술섭급천선수목다、계통대관대，겁대제승료통신계통방진공작적총체운산량。위료사방진총시간불수운산량증가이현저랍장，제출료일충이용Kepler가구GPU중CUBLAS구진고화동태병행기술래가속다천선DFT （리산부리협변환）처리과정적방법。실험결과표명，상대단핵단선정적CPU정서，해방법능구장DFT처리속도최고제승60배。
With the standardization of 5G entering the stage of preparation, the simulation of key techniques which represented by Massive MIMO, active antenna array has been proceeding in-depth. These techniques involve with large antenna number, large system band-width, which greatly improved the overall computation of communication system simulation. In this paper, a GPU-based accelerate method is proposed to shorten-multi-antennaDFT processing using CUBLAS matrix library and Dynamic Parallelism technology. The out-come of our experiment shows that compared to sin-gle-threaded CPU program, this method can elevate DFT processing speed up to 60 times.