深度学习辅助水下光通信信道估计和信号检测  

Deep Learning Aided Channel Estimation and Signal Detection for Underwater Optical Communication

作　　者：石佳 黄爱萍[1] 陶林伟[1] Shi Jia;Huang Aiping;Tao Linwei(School of Marine Science and Technology,Northwestern Polytechnical University,Xi'an 710072,Shaanxi,China)

机构地区：[1]西北工业大学航海学院,陕西西安710072

出　　处：《中国激光》2022年第17期95-102,共8页Chinese Journal of Lasers

基　　金：水声对抗重点实验室基金(D5120190034);中央高校基本科研业务费重点布局类基金(D5000210974)。

摘　　要：近年来,水下无线光通信(UWOC)技术因其高速数据传输能力而成为了研究热点,但水波的吸收和散射等因素使得UWOC信道变得十分复杂。对复杂的信道做出准确的信道估计(CE)和信号检测(SD)是目前高速UWOC面临的主要问题之一。针对这一问题,提出了一种在光正交频分复用系统中利用深度学习以端对端的方式对UWOC信道进行估计并直接检测的方案。首先根据在不同水域类型的UWOC信道下模拟生成的数据离线训练深度神经网络(DNN),然后使用DNN直接对信号进行补偿,该方案可以隐式地估计出信道状态信息并直接恢复传输数据。仿真结果表明,提出的信道估计和信号检测方案在复杂的UWOC信道环境中具有优越的性能,特别是在导频数量较少以及去除循环前缀时,深度学习方案比传统方案鲁棒性更好。Objective At present,underwater wireless optical communication(UWOC)is widely concerned in underwater communication because of its high transmission efficiency and excellent transmission capacity.For underwater acoustic communication,the transmission delay is large because of the limited bandwidth of sound wave in kilohertz frequency region.UWOC technology can achieve the data transmission rate of Gbit/s,while maintaining low transmission delay.In addition,UWOC can carry more data because of the shorter wavelength of light.However,light wave propagation in UWOC channel is affected by absorption,scattering and other factors.The absorption effect is irreversible,and the light energy is converted into other forms of energy,causing the signal to decay.In scattering,the direction in which each photon is emitted varies randomly,so that the energy captured by the receiver is reduced.In order to accurately evaluate the complex UWOC channel information,many scholars have studied the absorption,scattering and turbulence effects in different water areas and characterized these effects.This greatly improves the accuracy of UWOC channel modeling and channel estimation.However,due to the complexity of underwater environment,the channel state information estimated by traditional methods is usually not accurate enough and the recovered signals have high bit error rate(BER).Based on the above research,this paper designs a scheme of underwater optical communication channel estimation(CE)and signal detection(SD)aided by deep learning method.Methods This paper presents an end-to-end solution to the challenging CE and SD problems in UWOC systems using a deep neural network(DNN).Firstly,the scheme uses optical orthogonal frequency division multiplexing(OOFDM)system as the system model,and classical UWOC channel as the channel model.Then the DNN model is built according to the channel characteristics of UWOC and the DNN is trained off-line using the simulated data under different UWOC channels.The scheme combines CE and SD in UWOC syst

关 键 词：光通信 水下无线光通信 信道估计 信号检测 深度学习 光正交频分复用 

分 类 号：TN929.1[电子电信—通信与信息系统]