基于深度学习的空间脉冲位置调制多分类检测器  

Spatial pulse position modulation multi-classification detector based on deep learning

作　　者：王惠琴[1] 侯文斌 黄瑞[1] 陈丹[2] WANG Hui-qin;HOU Wen-bin;HUANG Rui;CHEN Dan(School of Computer and Communication,Lanzhou University of Technology,Lanzhou 730050,China;School of Automation and Information Engineering,Xi'an University of Technology,Xi’an 710048,China)

机构地区：[1]兰州理工大学计算机与通信学院,甘肃兰州730050 [2]西安理工大学自动化与信息工程学院,陕西西安710048

出　　处：《中国光学（中英文）》2023年第2期415-424,共10页Chinese Optics

基　　金：国家自然科学基金资助项目(No.61861026,No.61875080);甘肃省自然科学基金资助项目(No.20JR5RA472);陕西省科技计划产业研究项目(No.2020GY-036);西安科技局项目(No.GXYD14.21)。

摘　　要：为有效避免最大似然(ML)检测复杂的计算过程,根据空间脉冲位置调制(SPPM)信号的特点,将深度神经网络(DNN)与分步检测相结合,提出了一种基于深度学习的SPPM多分类检测器。在该检测器中,利用DNN建立接收信号与PPM符号间的非线性关系,并以此为准则完成在线接收PPM符号的检测,从而有效避免了对PPM符号的穷搜索检测过程。结果表明,采用本文检测器后,SPPM系统在大幅降低检测复杂度的前提下,取得了近似最优的误比特性能,同时还克服了K均值聚类(KMC)分步分类检测所出现的错误平台效应。当PPM阶数为64时,本文方法较ML检测和线性均衡DNN检测器的计算复杂度分别降低了约95.45%、33.54%。In order to effectively avoid high computational complexity when using Maximum Likelihood(ML)detection,a deep learning-based Spatial Pulse Position Modulation(SPPM)multi-classification detector is proposed by combining a Deep Neural Network(DNN)and step detection.In the detector,the DNN is used to establish a non-linear relationship between the received signal and the PPM symbols.Thereafter,the subsequent received PPM symbols are detected according to this relationship,so as to avoid the exhaustive search process of PPM symbol detection.The simulation results show that with the proposed detector,the SPPM system approximately achieves optimal bit error performance on the premise of greatly reducing detection complexity.Meanwhile,it overcomes the error platform effect caused by K-Means Clustering(KMC)step classification detection.When the PPM order is 64,the computational complexity of the proposal is about 95.45%and 33.54%lower than that of ML detectors and linear equalization DNN detectors,respect-ively.

关 键 词：无线光通信 空间脉冲位置调制 深度学习 多分类检测器 

分 类 号：TN929.12[电子电信—通信与信息系统]