正交频分复用（Orthogonal Frequency Division Multiplexing, OFDM）因其抗多径衰落能力和高频谱效率已被广泛应用于水声通信，但传统接收机在多衰减快时变的复杂水声信道下会出现性能下降的问题。深度学习（Deep Learning, DL）良好的特征提取和学习拟合能力为水声OFDM接收机的优化设计提供了新思路，因此，本文主要研究内容如下：

（1）针对数据驱动水声OFDM接收机可解释性差的问题，提出了DL联合水声OFDM专家知识的接收机。所提出的接收机设计了基于跳跃连接（Skip Connection, SC）卷积神经网络（Convolutional Neural Network, CNN）的信道估计子网和基于注意力机制（Attention Mechanism, AM）增强双向长短期记忆（Bi-directional Long Short-Term Memory, BiLSTM）网络的信号检测子网的级联网络（简称SCABNet），其中SC-CNN利用图像超分辨率的思想根据导频子载波重建数据子载波的信道频率响应，AM-BiLSTM提取序列数据相关性，并关注接收信号的有效信息以训练最佳网络权重，提高数据恢复的准确性。采用实测水声信道仿真，结果表明，与传统经典估计和均衡方法、数据驱动全连接深度神经网络和模型驱动ComNet相比，所提出的SCABNet接收机具有更低的误码率（Bit Error Rate, BER），在不同水声信道和调制阶数、导频数减少、去除循环前缀、存在载波频率偏移和符号时间偏移等多种条件下，SCABNet接收机具有鲁棒的BER性能。

（2）针对水声信道失配导致接收机性能损失泛化能力差的问题，提出了基于深度迁移学习的SCABNet接收机并设计了预训练-微调策略。首先，建立预训练SCABNet接收机并在源域水声信道下离线训练以提取水声信道共同特征；随后，采用预训练网络中CNN的特征提取层将网络模型在源域水声信道中所学知识应用于目标水声信道的数据恢复中，利用预训练网络初始化深度迁移网络并在目标域中训练微调；最后，将深度迁移网路模型用于在线测试。实测水声信道仿真结果表明，采用深度迁移学习的SCABNet接收机可以改善水声信道失配对接收机性能的影响，提高泛化能力并减少训练时间。

Orthogonal Frequency Division Multiplexing (OFDM) has been widely used in underwater acoustic (UWA) communications due to its resistance to multipath fading and high spectral efficiency, but conventional receivers suffer from performance degradation in complex UWA channels with multiple fading and fast time variation. The good feature extraction and learning fitting capability of deep learning (DL) provides a new idea for the optimal design of UWA OFDM receivers, therefore, the main research of this thesis is as follows:

(1) To address the issue of poor interpretability of data-driven UWA OFDM receivers, a DL receiver with joint UWA OFDM expert knowledge is proposed. The proposed receiver is designed with a channel estimation sub-network based on a skip connection (SC) convolutional neural network (CNN) and a cascaded network based on an attention mechanism (AM) enhanced bi-directional long short-term memory (BiLSTM) network for signal detection (Abbreviated SCABNet). The SC-CNN uses the idea of image super-resolution to reconstruct the channel frequency response of the data sub-carrier based on the pilot sub-carrier, and the AM-BiLSTM extracts the sequence data correlation and focuses on the effective information of the received signal to train the optimal network weights to improve the accuracy of data recovery. Using simulations of the measured UWA channel, the results show that the proposed SCABNet receiver has a lower bit error rate (BER) than the conventional classical estimation and equalization methods, data-driven fully-connected deep neural networks and model-driven ComNet, with different UWA channels and modulation orders, reduced number of pilots, removal of cyclic prefixes, presence of carrier frequency The proposed SCABNet receiver has a robust BER performance under various conditions such as different UWA channels and modulation orders, reduced number of pilots, removal of cyclic prefixes, presence of carrier frequency offset and symbol time offset.

(2) To address the problem of poor generalization ability due to the loss of receiver performance caused by UWA channel mismatch, a SCABNet receiver based on deep transfer learning is proposed and a pre-training-trimming strategy is designed. Firstly, the pre-trained SCABNet receiver is built and trained offline under the source domain UWA channel to extract common features of the UWA channel. Subsequently, the feature extraction layer of the CNN in the pre-trained network is used to apply the knowledge learned by the network model in the source domain UWA channel to the data recovery of the target UWA channel, and the pre-trained network is used to initialize the deep transfer network and train fine-tuning in the target domain.  Finally, the deep transfer network model is used for online testing. Simulation results of the measured UWA channel show that the SCABNet receiver with deep transfer learning can improve the impact of UWA channel mismatch on receiver performance, improve generalization and reduce training time.

[1]李加林, 沈满洪, 马仁锋, 等. 海洋生态文明建设背景下的海洋资源经济与海洋战略[J]. 自然资源学报, 2022, 37(4): 829-849.

[2]万敏, 扬帆. 海洋强国新航程[J]. 红旗文稿, 2023, 483(3): 4-8+1.

[3]Stojanovic M, Preisig J. Underwater acoustic communication channels: Propagation models and statistical characterization[J]. IEEE Communications Magazine, 2009, 47(1): 84-89.

[4]Domingo M C. Overview of channel models for underwater wireless communication networks[J]. Physical Communication, 2008, 1(3): 163-182.

[5]Qarabaqi P, Stojanovic M. Statistical Characterization and Computationally Efficient Modeling of a Class of Underwater Acoustic Communication Channels[J]. IEEE Journal of Oceanic Engineering, 2013, 38(4): 701-717.

[6]Diamant R, Lampe L. Low Probability of Detection for Underwater Acoustic Communication: A Review[J]. IEEE Access, 2018, 6: 19099-19112.

[7]Wang Z, Huang J, Zhou S, et al. Iterative Receiver Processing for OFDM Modulated Physical-Layer Network Coding in Underwater Acoustic Channels[J]. IEEE Transactions on Communications, 2013, 61(2): 541-553.

[8]Radosevic A, Ahmed R, Duman T M. Adaptive OFDM Modulation for Underwater Acoustic Communications: Design Considerations and Experimental Results[J]. IEEE Journal of Oceanic Engineering, 2013, 39(2): 357-370.

[9]Khan M R, Das B, Pati B B. Channel estimation strategies for underwater acoustic (UWA) communication: An overview[J]. Journal of the Franklin Institute, 2020, 357(11): 7229-7265.

[10]Barhumi I, Moonen M. MLSE and MAP Equalization for Transmission Over Doubly Selective Channels[J]. IEEE Transactions on Vehicular Technology, 2009, 58(8): 4120-4128.

[11]LeCun Y, Bengio Y, Hinton G, Deep Learning[J]. Nature, 2015, 521: 436–444.

[12]Pan S, Yang Q. A Survey on Transfer Learning. IEEE Transactions on Knowledge and Data Engineering[J]. 2010, 22(10): 1345-1359.

[13]Lam W K, Ormondroyd R F. A novel broadband COFDM modulation scheme for robust communication over the underwater acoustic channel[C] //IEEE Military Communications Conference. Proceedings. MILCOM 98, Boston, MA, USA, 1998: 128-133.

[14]Kim B -C, Lu I -T. Parameter study of OFDM underwater communications system[C] //OCEANS 2000 MTS/IEEE Conference and Exhibition. Conference Proceedings, Providence, RI, USA, 2000: 1251-1255.

[15]Miyoshi K. Preliminary design of OFDM and CDMA acoustic communication system[C] //MTS/IEEE Oceans 2001. An Ocean Odyssey. Conference Proceedings, Honolulu, HI, USA, 2001: 2216-2219.

[16]Kim M -S, Cho Y -H, Im T -H, et al. A Study on Performance of Underwater Acoustic OFMD System in Shallow Water Environment[C] //OCEANS 2018 MTS/IEEE Charleston, Charleston, SC, 2018: 1-4.

[17]Li B, Zhou S, Stojanovic M, et al. Multicarrier communication over underwater acoustic channels with nonuniform Doppler shifts[J]. IEEE Journal of Oceanic Engineering, 2008, 33(2): 198-209.

[18]王明华, 桑恩方, 乔钢. 高速水声通信中正交频分复用技术试验研究[J]. 海洋工程, 2006, 24(4): 6.

[19]申晓红, 黄建国, 周倩, 等. OFDM水声通信基带传输及试验研究[J]. 西北工业大学学报, 2008, 26(3): 5.

[20]周德富, 刘云涛, 蔡惠智, 等. STBC-OFDM技术在水声通信中的试验研究[J]. 声学技术, 2012, 31(5): 466-470.

[21]殷敬伟, 王驰, 潘争荣, 等. 基于MDAPSK的OFDM水声通信研究[J].华中科技大学学报（自然科学版）, 2013, 41(3): 20-24.

[22]周跃海, 江伟华, 陈磊, 等. 采用时反和时频差分OFDM的水声语音通信方法[J]. 应用声学, 2015, 34(4): 283-290.

[23]张舒然, 武岩波, 朱敏, 等. 一种鲁棒性强的OFDM水声通信系统[J]. 应用声学, 2019, 38(4): 635-644.

[24]郭庆, 李昂迪, 吴景, 等. 基于ZoomFFT的水声OFDM信号解调算法[J]. 系统仿真学报, 2022, 34(2): 314-322.

[25]Dorner S, Cammerer S, Hoydis J, et al. Deep Learning-Based Communication Over the Air[J]. IEEE Journal of Selected Topics in Signal Processing, 2018, 12(1): 132-143.

[26]Wang T, Wen C -K, Wang H, et al. Deep learning for wireless physical layer: Opportunities and challenges[J]. China Communications, 2017, 14(11): 92-111.

[27]Ye H, Li Y G, Juang B. Power of deep learning for channel estimation and signal detection in OFDM systems[J]. IEEE Wireless Communications Letters, 2017, 7(1): 114-117.

[28]Cheng X, Liu D, Wang C, et al. Deep Learning-Based Channel Estimation and Equalization Scheme for FBMC/OQAM Systems[J]. IEEE Wireless Communications Letters, 2019, 8(3): 881-884.

[29]Zheng S, Chen S, Yang X. DeepReceiver: A Deep Learning-Based Intelligent Receiver for Wireless Communications in the Physical Layer[J]. IEEE Transactions on Cognitive Communications and Networking. 2021, 7(1): 5-20.

[30]O’Shea T, Hoydis J. An Introduction to Deep Learning for the Physical Layer[J]. IEEE Transactions on Cognitive Communications and Networking. 2017, 3(4): 563-575.

[31]Gao X, Jin S, Wen C K, et al. ComNet: Combination of Deep Learning and Expert Knowledge in OFDM Receivers[J]. IEEE Communications Letters, 2018, 22(12): 2627-2630.

[32]Liu C, Arslan T. RecNet: Deep Learning-Based OFDM Receiver with Semi-Blind Channel Estimation[C] //2020 IEEE International Symposium on Circuits and Systems (ISCAS), Seville, Spain, 2020: 1-4.

[33]Jiang R, Wang X, Cao S, et al. Deep Neural Networks for Channel Estimation in Underwater Acoustic OFDM Systems[J]. IEEE Access, 2019, 7: 23579-23594.

[34]Zhang Y, Wang H, Li C, et al. On the Performance of Deep Neural Network Aided Channel Estimation for Underwater Acoustic OFDM Communication[J], Ocean Engineering, 2022, 259.

[35]Zhang Y, Li J, Zakharov Y, et al. Deep learning based underwater acoustic OFDM communications[J]. Applied Acoustics, 2019, 154: 53-58.

[36]Zhang J, Cao Y, Han G, et al. Deep Neural Network based Underwater OFDM Receiver[J], IET Communications, 2019, 13(13): 1998-2002.

[37]Zhao H, Ji F, Wen M, et al. Multi-Task Learning Based Underwater Acoustic OFDM Communications[C] //2021 IEEE International Conference on Signal Processing, Communications and Computing (ICSPCC), Xi'an, China, 2021: 1-5.

[38]Zhang Y, Li C, Wang H, et al. Deep learning aided OFDM receiver for underwater acoustic communications[J]. Applied Acoustics, 2022, 187.

[39]付晓梅, 王思宁, 胡雅琳. 基于被动时间反转-卷积神经网络的OFDM水声通信系统研究[J]. 湖南大学学报（自然科学版）, 2022, 49(8): 169-178.

[40]Liu C, Wei Z, Ng D W K, et al. Deep Transfer Learning for Signal Detection in Ambient Backscatter Communications[J]. IEEE Transactions on Wireless Communications, 2021, 20(3): 1624-1638.

[41]Yang Y, Gao F, Zhong Z, et al. Deep Transfer Learning-Based Downlink Channel Prediction for FDD Massive MIMO Systems[J]. IEEE Transactions on Communications, 2020, 68(12): 7485-7497.

[42]Alves W, Correa I, González-Prelcic N, et al. Deep Transfer Learning for Site-Specific Channel Estimation in Low-Resolution mmWave MIMO[J]. IEEE Wireless Communications Letters, 2021, 10(7): 1424-1428.

[43]Qing C, Dong L, Wang L, et al. Transfer learning-based channel estimation in orthogonal frequency division multiplexing systems using data-nulling superimposed pilots[J]. PLoS ONE, 2022, 17(5): e0268952.

[44]Zeng J, Sun J, Gui G, et al. Downlink CSI Feedback Algorithm With Deep Transfer Learning for FDD Massive MIMO Systems[J]. IEEE Transactions on Cognitive Communications and Networking, 2021, 7(4): 1253-1265.

[45]Zeng Z, Fu S, Zhang H, et al. A Survey of Underwater Optical Wireless Communications[J]. IEEE Communications Surveys & Tutorials, 2017, 19(1): 204-238.

[46]Huang G, Liu Z, Maaten L V D, et al. Densely Connected Convolutional Networks[C] //2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Honolulu, HI, USA, 2017: 2261-2269.

[47]Yu J, Gao X, Tao D, et al. A Unified Learning Framework for Single Image Super-Resolution[J]. IEEE Transactions on Neural Networks and Learning Systems, 2014, 25(4): 780-792.

[48]Greff K, Srivastava R K, Koutník J, et al. LSTM: A Search Space Odyssey[J]. IEEE Transactions on Neural Networks and Learning Systems, 2017, 28(10): 2222-2232.

[49]Hu J, Shen L, Albanie S, et al. Squeeze-and-Excitation Networks[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2020, 42(8): 2011-2023.

[50]Gao J, Hu M, Zhong C, et al. An Attention-Aided Deep Learning Framework for Massive MIMO Channel Estimation[J]. IEEE Transactions on Wireless Communications, 2022, 21(3): 1823-1835.

[51]van Walree P A, Socheleau F -X, Otnes R, et al. The Watermark Benchmark for Underwater Acoustic Modulation Schemes[J]. IEEE Journal of Oceanic Engineering, 2017, 42(4): 1007-1018.