基于深度学习的正交时频空间接收机设计

在第六代(6G) 无线通信系统中，稳定的高移动性通信，如飞机上、高速火车和无人机的移动通信，被认为是必须实现的目标。正交时频空间技术(OTFS) 有望代替传统的正交频分复用技术(OFDM)，并且在高移动通信场景比后者达到更精确和更稳定的性能。然而在以OTFS 调制为基础的通信中，由于OTFS 信道矩阵过大，同时在实际应用当中仍然存在子载波间干扰(ICI) 以及符号之间的干扰(ISI)。
因此，获得准确的信道状态信息(CSI) 以及快速精确地解调OTFS 信号成为一个
巨大的挑战。近年来，深度学习在通信中的应用越来越广泛，并且其相比传统基于模型的方法具有更快的运行速度和泛化能力。本文使用深度神经网络对OTFS信号解调进行优化，使得接收端能够以更低的复杂度和更快的速度去解析信道信息，接着利用正交近似消息传递算法对信号完成解调。具体来说，本文将OTFS 信道估计建模为一种噪声去除问题，同时提出了一种自适应阈值深度残差去噪网络(DRDN)，使其能够以更少的计算损耗以及更快的计算速度处理稀疏的OTFS 信道矩阵。特别地，网络中的阈值是根据注意力机制的获得的。接着本文对DRDN 进行了数学分析和推导，结合统计信号处理理论，给出了基于贝叶斯法则，也就是最小均方误差损失估计(MMSE)。同时，为了完成符号检测，本文采用了一种低复杂度的正交近似消息传递算法(OAMP) 完成符号检测。仿真表明，在使用DRDN 可以比基于模型的方法获得更低的NMSE，即使在复杂场景下，DRDN 表现也优于其他算法从而降低OAMP 系统的误码率(BER)。同时该网络可以实现在一定性能上降低硬件空间实现以及运算复杂度，使得通信系统能够快速地获取到精确的信道信息矩阵。

In the sixth-generation (6G) wireless communication systems, high mobility communication, such as mobile communication on airplanes, high-speed trains, and drones, is considered a crucial issue to address. Orthogonal Time Frequency Space (OTFS) technology is expected to replace traditional Orthogonal Frequency Division Multiplexing (OFDM) technology and to achieve more accurate and stable performance in highmobility communication scenarios than the latter. However, the large size of the OTFS channel matrix, as well as the presence of Inter-Carrier Interference (ICI) and Inter-
Symbol Interference (ISI) in practical applications, make obtaining accurate Channel State
Information (CSI) and quickly and accurately demodulating OTFS signals a huge challenge. In recent years, the application of deep learning in communication has become increasingly widespread, and it has faster-running speed and generalization ability compared to traditional model-based methods. In this paper, we optimize the demodulation of OTFS signals using deep neural networks, enabling the receiver to extract channel information with lower complexity and faster speed. Specifically, we model the OTFS channel estimation as a noise removal problem and propose an attention-based adaptive threshold deep residual denoising network (DRDN) to efficiently process the sparse OTFS channel matrix with minimal computational cost and fast speed. We then mathematically analyze and derive the DRDN using statistical signal processing theory and the Bayesian rule, which leads to the minimum mean square error (MMSE) estimation. For symbol detection, we employ a low-complexity orthogonal approximate message passing algorithm (OAMP). Simulation results demonstrate that the DRDN outperforms other methods, even in complex scenarios, and achieves lower NMSE, thereby reducing the bit error rate (BER) of the OAMP system. Moreover, the network can achieve reduced hardware space implementation and computational complexity with comparable performance.

[1] 徐湛; 李青宇; 巩译; 孟繁轲; 张校旗;. 适用于高多普勒扩展场景的调制技术[J]. 西安邮电大学学报, 2021, 26: 1-14.
[2] CHOWDHURY M Z, SHAHJALAL M, AHMED S, et al. 6G wireless communication systems:Applications, requirements, technologies, challenges, and research directions[J]. IEEE OpenJournal of the Communications Society, 2020, 1: 957-975.
[3] 李爽; 肖伟; 董志诚; 周维曦;. 高速移动场景下的多用户OFDM 系统功率分配[J]. 长江信息通信, 2021, 34: 46-49.
[4] 王莹; 任军; 史可; 林彬;;. 基于深度学习的广义频分复用系统时频双选择信道估计[J]. 通信学报, 2021, 43: 233-242.
[5] HADANI R, MONK A. OTFS: A new generation of modulation addressing the challenges of5G[A]. 2018.
[6] XIAO L, LI S, QIAN Y, et al. An overview of OTFS for Internet of Things: Concepts, benefits,and challenges[J]. IEEE Internet of Things Journal, 2021, 9(10): 7596-7618.
[7] BELLO P. Characterization of randomly time-variant linear channels[J]. IEEE transactions onCommunications Systems, 1963, 11(4): 360-393.
[8] DANG S, AMIN O, SHIHADA B, et al. What should 6G be?[J]. Nature Electronics, 2020, 3(1): 20-29.
[9] LI S, YUAN J, YUAN W, et al. Performance analysis of coded OTFS systems over high-mobilitychannels[J]. IEEE transactions on wireless communications, 2021, 20(9): 6033-6048.
[10] LIU F, YUAN Z, GUO Q, et al. Message passing-based structured sparse signal recovery forestimation of OTFS channels with fractional Doppler shifts[J]. IEEE transactions on wirelesscommunications, 2021, 20(12): 7773-7785.
[11] TIWARI S, DAS S S, RANGAMGARI V. Low complexity LMMSE receiver for OTFS[J]. IEEEcommunications letters, 2019, 23(12): 2205-2209.
[12] LI L, LIANG Y, FAN P, et al. Low complexity detection algorithms for OTFS under rapidlytime-varying channel[C]//2019 IEEE 89th Vehicular Technology Conference (VTC2019-Spring). IEEE, 2019: 1-5.
[13] AKYILDIZ I F, POMPILI D, MELODIA T. Underwater acoustic sensor networks: researchchallenges[J]. Ad hoc networks, 2005, 3(3): 257-279.
[14] LARSSON E G, EDFORS O, TUFVESSON F, et al. Massive MIMO for next generation wirelesssystems[J]. IEEE communications magazine, 2014, 52(2): 186-195.
[15] YE H, LI G Y, JUANG B H. Power of deep learning for channel estimation and signal detectionin OFDM systems[J]. IEEE Wireless Communications Letters, 2017, 7(1): 114-117.
[16] 廖勇; 李雪; 王幕熙; 杨植景; 周晨虹;. 基于深度学习的信道估计技术研究进展[J]. 大连工业大学学报, 2021: 1-13.
[17] 石佳琪; 金桂月; 金基宇; 樊磊;. 基于深度学习的信道估计[J]. 大连工业大学学报, 2021,40: 367-376.
[18] HADANI R, RAKIB S, MOLISCH A, et al. Orthogonal time frequency space (OTFS) modulationfor millimeter-wave communications systems[C]//2017 IEEE MTT-S International MicrowaveSymposium (IMS). IEEE, 2017: 681-683.
[19] 韩知雨, 董静薇, 闫宁, 等. 大规模MIMO 系统中伪随机序列加扰法的导频分配.[J]. Journalof Harbin University of Science & Technology, 2021, 26(1).
[20] RAVITEJA P, PHAN K T, HONG Y. Embedded pilot-aided channel estimation for OTFS indelay–Doppler channels[J]. IEEE transactions on vehicular technology, 2019, 68(5): 4906-4917.
[21] GAUDIO L, KOBAYASHI M, BISSINGER B, et al. Performance analysis of joint radar andcommunication using OFDM and OTFS[C]//2019 IEEE International Conference on CommunicationsWorkshops (ICC Workshops). IEEE, 2019: 1-6.
[22] ZHAO L, GAO W J, GUO W. Sparse Bayesian learning of delay-Doppler channel for OTFSsystem[J]. IEEE communications letters, 2020, 24(12): 2766-2769.
[23] LIU Y, ZHANG S, GAO F, et al. Uplink-Aided High Mobility Downlink Channel EstimationOver Massive MIMO-OTFS System[J/OL]. IEEE Journal on Selected Areas in Communications,2020, 38(9): 1994-2009. DOI: 10.1109/JSAC.2020.3000884.
[24] RAVITEJA P, HONG Y, VITERBO E, et al. Practical pulse-shaping waveforms for reducedcyclic-prefix OTFS[J]. IEEE Transactions on Vehicular Technology, 2018, 68(1): 957-961.
[25] LI S, YUAN W, WEI Z, et al. Hybrid MAP and PIC detection for OTFS modulation[J]. IEEEtransactions on vehicular technology, 2021, 70(7): 7193-7198.
[26] RAVITEJA P, PHAN K T, HONG Y, et al. Interference cancellation and iterative detectionfor orthogonal time frequency space modulation[J]. IEEE transactions on wireless communications,2018, 17(10): 6501-6515.
[27] YUAN Z, LIU F, YUAN W, et al. Iterative detection for orthogonal time frequency spacemodulation with unitary approximate message passing[J]. IEEE transactions on wireless communications,2021, 21(2): 714-725.
[28] LIU F, YUAN Z, GUO Q, et al. Multi-block UAMP-based detection for OTFS with rectangularwaveform[J]. IEEE Wireless Communications Letters, 2021, 11(2): 323-327.
[29] ERPEK T, O’SHEA T J, SAGDUYU Y E, et al. Deep learning for wireless communications[J]. Development and Analysis of Deep Learning Architectures, 2020: 223-266.
[30] HU Q, GAO F, ZHANG H, et al. Understanding deep MIMO detection[A]. 2021.
[31] LIU C, LIU X, NG D W K, et al. Deep residual learning for channel estimation in intelligentreflecting surface-assisted multi-user communications[J]. IEEE Transactions on WirelessCommunications, 2021, 21(2): 898-912.
[32] ZHANG X, YUAN W, LIU C. Deep Residual Learning for OTFS Channel Estimation withArbitrary Noise[C]//2022 IEEE/CIC International Conference on Communications in China(ICCC Workshops). IEEE, 2022: 320-324.
[33] SOLTANI M, POURAHMADI V, MIRZAEI A, et al. Deep learning-based channel estimation[J]. IEEE Communications Letters, 2019, 23(4): 652-655.
[34] LIAO Y, HUA Y, DAI X, et al. ChanEstNet: A deep learning based channel estimation for highspeedscenarios[C]//ICC 2019-2019 IEEE international conference on communications (ICC).IEEE, 2019: 1-6.
[35] HE H, WEN C K, JIN S, et al. Model-driven deep learning for MIMO detection[J]. IEEETransactions on Signal Processing, 2020, 68: 1702-1715.
[36] SCHYNOL L, PESAVENTO M. Deep Unfolding in Multicell MU-MIMO[C]//2022 30th EuropeanSignal Processing Conference (EUSIPCO). IEEE, 2022: 1631-1635.
[37] YE H, LIANG L, LI G Y, et al. Deep learning-based end-to-end wireless communication systemswith conditional GANs as unknown channels[J]. IEEE Transactions on Wireless Communications,2020, 19(5): 3133-3143.
[38] YUAN W, WEI Z, YUAN J, et al. A simple variational Bayes detector for orthogonal timefrequency space (OTFS) modulation[J]. IEEE transactions on vehicular technology, 2020, 69(7): 7976-7980.
[39] LI S, YUAN W, WEI Z, et al. Cross domain iterative detection for orthogonal time frequencyspace modulation[J]. IEEE transactions on wireless communications, 2021, 21(4): 2227-2242.
[40] WEI Z, YUAN W, LI S, et al. Off-grid channel estimation with sparse bayesian learning forOTFS systems[J]. IEEE Transactions on Wireless Communications, 2022, 21(9): 7407-7426.
[41] CAO B, XIANG Z, REN P. Low complexity transmitter precoding for MU MIMO-OTFS[J].Digital Signal Processing, 2021, 115: 103083.
[42] 邢旺; 唐晓刚; 周一青; 张冲; 潘振岗. 面向OTFS 的时延-多普勒域信道估计方法综述[J].通信学报, 2022, 43: 188-201.
[43] LECUN Y, BENGIO Y, HINTON G. Deep learning[J]. nature, 2015, 521(7553): 436-444.
[44] 钱蓉蓉; 吕孝明; 任文平;. 基于深度学习的高维信号信道估计算法[J]. 电讯技术, 2022, 62:1554-1559.
[45] 郭晟; 余乐; 朱立东;. 星地场景下基于CNN 的OTFS 系统信道估计方法[J]. 天地一体化信息网络, 2022, 3: 37-45.
[46] CHOI D, SHALLUE C J, NADO Z, et al. On empirical comparisons of optimizers for deeplearning[A]. 2019.
[47] CUNNINGHAM P, CORD M, DELANY S J. Supervised learning[J]. Machine learning techniquesfor multimedia: case studies on organization and retrieval, 2008: 21-49.
[48] 卢梅. 基于深度学习的低频SKA 点扩展函数效应消除研究[D]. 贵州大学, 2022.
[49] 甘亚斌; 张代青; 于国荣; 徐丽华; 邹进. 滇中受水区用水结构时空演变态势分析[C]//水力发电. 昆明理工大学电力工程学院, 2022.
[50] ZOU F, SHEN L, JIE Z, et al. A sufficient condition for convergences of adam and rmsprop[C]//Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. 2019:11127-11135.
[51] ALI M N Y, SAROWAR M G, RAHMAN M L, et al. Adam deep learning with SOM forhuman sentiment classification[J]. International Journal of Ambient Computing and Intelligence(IJACI), 2019, 10(3): 92-116.
[52] 涂岳. 基于深度学习的OTFS 信号解调技术研究[D]. 北京邮电大学, 2021.
[53] KATTENBORN T, LEITLOFF J, SCHIEFER F, et al. Review on Convolutional Neural Networks(CNN) in vegetation remote sensing[J]. ISPRS journal of photogrammetry and remotesensing, 2021, 173: 24-49.
[54] IOFFE S, SZEGEDY C. Batch normalization: Accelerating deep network training by reducinginternal covariate shift[C]//International conference on machine learning. pmlr, 2015: 448-456.
[55] SAFARI M S, POURAHMADI V, SODAGARI S. Deep UL2DL: Data-driven channel knowledgetransfer from uplink to downlink[J]. IEEE Open Journal of Vehicular Technology, 2019,1: 29-44.
[56] MOLISCH A F. Statistical description of the wireless channel[M]. Wiley-IEEE Press, 2011.
[57] 林珠, 刘勃, 归琳. C 频段空地链路无线信道大尺度衰落模型[J]. 电讯技术, 2010, 50(12):112-115.
[58] 蔺伟, 沈京川, 李辉, 等. 高速铁路电波小尺度衰落特性分析[J]. 中国铁道科学, 2004, 25(3): 97-102.
[59] SCHNITER P. Low-complexity equalization of OFDM in doubly selective channels[J]. IEEETransactions on Signal processing, 2004, 52(4): 1002-1011.
[60] MURALI K, CHOCKALINGAM A. On OTFS modulation for high-Doppler fading channels[C]//2018 Information Theory and Applications Workshop (ITA). IEEE, 2018: 1-10.
[61] DENG L, YU D, et al. Deep learning: methods and applications[J]. Foundations and trends®in signal processing, 2014, 7(3–4): 197-387.
[62] LIU Y, AL-NAHHAL I, DOBRE O A, et al. Deep-Learning-Based Channel Estimation forIRS-Assisted ISAC System[C]//GLOBECOM 2022-2022 IEEE Global Communications Conference.IEEE, 2022: 4220-4225.
[63] MA J, PING L. Orthogonal amp[J]. IEEE Access, 2017, 5: 2020-2033.
[64] MA J, PING L. Orthogonal AMP[J/OL]. IEEE Access, 2017, 5: 2020-2033. DOI: 10.1109/ACCESS.2017.2653119.