查询结果:   陈敏华,李杨,张武雄.基于卷积神经网络的信道均衡算法[J].计算机应用与软件,2017,34(9):257 - 261.
中文标题
基于卷积神经网络的信道均衡算法
发表栏目
算法
摘要点击数
666
英文标题
CHANNEL EQUALIZATION ALGORITHM BASED ON CONVOLUTIONAL NEURAL NETWORK
作 者
陈敏华 李杨 张武雄 Chen Minhua Li Yang Zhang Wuxiong
作者单位
中国科学院上海微系统与信息技术研究所 上海 201210     
英文单位
Shanghai Institute of Microsystem and Information Technology,Chinese Academy of Sciences,Shanghai 201210,China     
关键词
信道均衡 卷积神经网络 无线通信 深度学习
Keywords
Channel equalization Convolutional neural network Wireless communication Deep learning
基金项目
国家自然科学基金项目(61471346);国家科技重大专项(2014ZX03005001);上海市自然科学基金项目(14ZR1439700)
作者资料
陈敏华,硕士生,主研领域:数据挖掘。李杨,博士生。张武雄,博士。 。
文章摘要
在现代无线通信系统中,为了克服由传输信道的非线性以及多径效应引起的符号间干扰,解决传统信道均衡算法难以适应的时变信号均衡问题,提出一种基于卷积神经网络的信道均衡算法。通过采集实际通信系统中发送端的相位偏移调制QPSK(quadrature phase shift keying)发送符号序列及接收端的接收符号序列,并将其分割为训练集和测试集来训练及测试卷积神经网络均衡器。实验结果验证了在相同信噪比条件下,基于卷积神经网络的信道均衡算法对QPSK恢复的误符号率相比RLS算法和MLP算法分别降低了20%和5%。
Abstract
In modern wireless communication system, in order to reduce the influences of InterSymbol Interference (ISI) introduced by nonlinear channel and multipath effect and solve the problem that the traditional channel equalizer cannot adapt to time-varying signals, a channel equalization algorithm based on convolutional neural network (CNN) is proposed. The quadrature phase shift keying (QPSK) dataset collected from the transmitting side and the receiving side of real wireless communication system is split into training set and test set to train and test the CNN algorithm. We thus conclude that the SER performance for QPSK systems with the CNN equalizer outperforms that of recursive least square (RLS) and multilayer perceptron (MLP) by in average 20% and 5% at low signal to noise ratio.
下载PDF全文