基于卷积网络的OFDM系统信道解码技术  

OFDM system channel decoding technology based on convolutional neural networks

作　　者：李军[1] 辛同亮 李文鑫 何波[2] 高鹤 LI Jun;XIN Tong-liang;LI Wen-xin;HE Bo;GAO He(School of Electrical Engineering and Automation,Qilu University of Technology(Shandong Academy of Science),Jinan 250353,China;School of Information Science and Engineering,Shandong University,Qingdao 266237,China;Shandong Zhengchen Technology Co.Ltd,Jinan 250101,China)

机构地区：[1]齐鲁工业大学(山东省科学院)电气工程与自动化学院,山东济南250353 [2]山东大学信息科学与工程学院,山东青岛266237 [3]山东正晨科技股份有限公司,山东济南250101

出　　处：《齐鲁工业大学学报》2022年第2期39-45,共7页Journal of Qilu University of Technology

基　　金：山东省自然科学基金(ZR2020QF016)。

摘　　要：为了降低信道估计的复杂度,提高信号检测的准确性,同时充分利用当前科技研究的最新成果,提出了一种新的基于卷积神经(convolutional neural networks,CNN)的正交频分复用(orthogonal frequency division multiplexing,OFDM)系统信道解码方法。以最小二乘法(least square,LS)和最小均方误差(minimum mean square error,MMSE)传统算法为例,它们在精确度或复杂度上无法适应当今毫米波通信的要求。利用深度学习工具将信道问题看作自回归问题,将接收信号看作一维数组,采用图像处理流程。利用卷积网络进行特征提取,然后利用全连接(fully-connected,FC)层进行分类,得到软比特(soft bit,SB)信号。仿真结果证明,该方法在复杂性上和LS法相当,性能上优于MMSE算法,在误码率方面有着平均2 dB的性能提升。In order to lessen the complexity of channel estimation,and improve the accuracy of signal detection and absorb the advanced technologies,this paper proposed a way of channel decoding for OFDM system with CNNs.Taking the traditional LS and MMSE methods as examples,they are not capable of meeting the demands of millimeter wave communication in accuracy and complexity today.This paper viewed channel as autoregressive progress and received signal as 1-Dimensional vector,and extracted features with CNNs,and then classified with fully connected layer and got the soft bits finally.Lastly,simulation results show that this method has less complexity and promising performance than traditional methods,and there is an average 2dB performance improvement.

关 键 词：CNN OFDM系统 信道解码 

分 类 号：TN928[电子电信—通信与信息系统]