水声通信深度学习增强最小二乘信道估计算法  

Deep learning enhanced least squares channel estimation algorithm for underwater acoustic communication

作　　者：刘进 孟昊 林臻 刘思成 王斌[1] 侯长波 LIU Jin;MENG Hao;LIN Zhen;LIU Sicheng;WANG Bin;HOU Changbo(School of Information and Communication Engineering,Harbin Engineering University,Harbin 150001,China;Information Application and Network Measurement and Control System Room,Beijing Institute of Aerospace Systems Engineering,Beijing 100076,China)

机构地区：[1]哈尔滨工程大学息与通信工程学院,黑龙江哈尔滨150001 [2]北京宇航系统工程研究所信息应用与网络测控系统室,北京100076

出　　处：《应用科技》2024年第5期39-46,共8页Applied Science and Technology

摘　　要：针对传统水声通信单载波频域均衡(single-carrier frequency domain equalization,SC-FDE)最小二乘(least squares,LS)信道估计算法受信道噪声影响大、估计性能差的问题,提出一种基于深度学习的增强LS信道估计算法,利用深度学习模型提取数据特征的能力,弥补LS算法受信道噪声影响的缺点。首先介绍水声通信中传统LS和最小均方误差(minimum mean square error,MMSE)信道估计算法的原理和优缺点,然后针对LS算法估计性能差的缺陷,利用深度学习模型学习LS信道估计的误差以此来纠正信道的频率响应。本文研究了深度神经网络(deep neural network,DNN)和卷积神经网络(convolutional neural network,CNN)2种不同网络结构的Enhanced-LSNet性能。仿真结果表明,本文提出的信道估计算法在低信噪比(signal to noise ratio,SNR)下较MMSE算法提升约4.1 dB,在高SNR下提升约为7 dB。In order to solve the problem that the traditional single-carrier frequency domain equalization(SC-FDE)least squares(LS)channel estimation algorithm is greatly affected by channel noise and has poor estimation performance,this paper proposes an enhanced LS channel estimation algorithm based on deep learning,which utilizes the ability of deep learning model to extract data features,making up for the shortcomings of the LS algorithm affected by channel noise.Firstly,the principle,advantages and disadvantages of the traditional LS and minimum mean square error(MMSE)channel estimation algorithm in underwater acoustic communications are introduced.And then the deep learning model is used to learn the error of LS channel estimation to correct the frequency response of the channel in view of the defect of poor estimation performance of LS algorithm.The enhanced-LSNet performance of two different network structures—deep neural network(DNN)and convolutional neural network(CNN),is studied.The simulation results show that the proposed channel estimation algorithm is about 4.1dB higher than the MMSE algorithm at low signal-to-noise ratio(SNR),and about 7dB higher at high SNR.

关 键 词：水声通信 单载波频域均衡 最小二乘 信道估计 深度学习 最小均方误差 特征提取 卷积神经网络 

分 类 号：TN929.3[电子电信—通信与信息系统]