在第五代通信中，毫米波（mmWave）多輸入多輸出（MIMO）系統被應用來滿足高吞吐量需求。其中，大規模 MIMO 系統是提高吞吐量的方法之一。 然而，當天線數量變大時，全數位預編碼的射頻鏈(RF Chain) 在 MIMO 系統中成本很高。因此可以透過混合預編碼器 (Hybrid Precoding) 的設計來達到減少射頻鏈的數量。近年來，深度學習技術在不同領域得到廣泛應用。也有一些研究使用深度學習技術來處理混合預編碼的設計。本論文參考了基於CNN的混合預編碼器（DLHB），提出了線上訓練和離線訓練兩種方案。所提出的方法使用交替最小化算法 (AltMin)將原本以訓練好的模型輸出重新利用迭代演算法做進一步的提升。此外，所提出方法的輸出與通道信息會一起存儲在線訓練數據集中，以達到線上學習的目的。本文考慮了時變信道環境並比較了不同的混合預編碼算法。模擬結果表明，基於深度學習的混合預編碼器與線上訓練方案的頻譜效率得到了提升。此外，本論文分析和比較了不同超參數對於演算法的頻譜效率。由於可以使用 AltMin 算法來提升模型輸出的性能，因此可以減小模型的大小，即便效能會有所下降。與DLHB相比，該模型在非時變環境下可以減少51.54%的可訓練參數，在時變環境下可以減少76.14%的可訓練參數。模型的縮減也可以獲得減少計算時間的好處。所提出的方法比DLHB快1.6 ms，並且具有更好的頻譜效率性能。

In the fifth-generation communication, the millimeter wave (mmWave) multiple input multiple output (MIMO) systems are applied to satisfy the high throughput requirement. In addition, a massive MIMO system is one of the ways to increase the throughput. However, when antenna number becomes large, the RF chains of the fully digital precoding cost a lot in MIMO systems. Therefore, hybrid precoding design is used in the massive MIMO system to reduce the number of RF chains. In recent years, deep learning techniques have been widely used in different fields. There are also some studies using deep learning techniques to deal with the design of hybrid precoding. This thesis refers to the CNN-based hybrid precoding (DLHB) and proposed the online training and offline training schemes. The proposed method uses alternating minimization algorithms (AltMin) to reconstruct the analog precoding from the output of the pre-training model. Beside, the output of the proposed method are stored with channel informations as an online training dataset. This thesis considers the time-varying channel environments and compares different hybrid precoding algorithms. Simulation results show that the spectral efficiency of the deep learning-based hybrid precoding with an online training scheme is improved. In addition, the simulations also analyzes and compares spectral efficiency with different hyperparameters. Because an AltMin algorithm can be used to promote the performance of the model output, the proposed method can reduce the size of the model even though the effect is worse. The model can be reduced 51.54% of the trainable parameters compared with the DLHB in the time-invariant environment and 76.14% of trainable parameters can be reduced in the time-varying environment. The reduction of the model can also obtain the benefit of reducing the computing time. The proposed method is 1.6 ms faster than the DLHB and has better performance of spectrum efficiency.

[1] A. M. Elbir, Cnn-based precoder and combiner design in mmwave mimo systems,"IEEE Communications Letters, vol. 23, no. 7, pp. 1240{1243, 2019.
[2] X. Yu, J.-C. Shen, J. Zhang, and K. B. Letaief, Alternating minimization algo-
rithms for hybrid precoding in millimeter wave mimo systems," IEEE Journal of
Selected Topics in Signal Processing, vol. 10, no. 3, pp. 485-500, 2016.
[3] R. W. Heath, N. Gonz_alez-Prelcic, S. Rangan, W. Roh, and A. M. Sayeed, An
overview of signal processing techniques for millimeter wave mimo systems," IEEE
Journal of Selected Topics in Signal Processing, vol. 10, no. 3, pp. 436-453, 2016.
[4] S. K. Yong and C.-C. Chong, An overview of multigigabit wireless through mil-
limeter wave technology: potentials and technical challenges," EURASIP Journal
on Wireless Communications and Networking, vol. 2007, no. 1, pp. 1-10, 2006.
[5] S. Sun, T. S. Rappaport, R. W. Heath, A. Nix, and S. Rangan, Mimo for
millimeter-wave wireless communications: beamforming, spatial multiplexing, or
both?" IEEE Communications Magazine, vol. 52, no. 12, pp. 110-121, 2014.
[6] H. Huang, Y. Song, J. Yang, G. Gui, and F. Adachi, Deep-learning-based
millimeter-wave massive mimo for hybrid precoding," IEEE Transactions on Ve-
hicular Technology, vol. 68, no. 3, pp. 3027-3032, 2019.
[7] J. Chen, W. Feng, J. Xing, P. Yang, G. E. Sobelman, D. Lin, and S. Li, Hybrid
beamforming/combining for millimeter wave mimo: A machine learning approach,"
IEEE Transactions on Vehicular Technology, vol. 69, no. 10, pp. 11 353-11 368, 2020.
[8] A. M. Elbir and K. V. Mishra, Joint antenna selection and hybrid beamformer de-
sign using unquantized and quantized deep learning networks," IEEE Transactions
on Wireless Communications, vol. 19, no. 3, pp. 1677-1688, 2020.
[9] S.-S. Wong, C.-F. Teng, and A.-Y. Wu, Two-step codebook-assisted alternating
minimization (ca-altmin) for low-complexity hybrid beamforming design," IEEE
Communications Letters, vol. 25, no. 6, pp. 1989-1993, 2021.
[10] A. Saleh and R. Valenzuela, \A statistical model for indoor multipath propagation," IEEE Journal on Selected Areas in Communications, vol. 5, no. 2, pp. 128-137, 1987.
[11] A. Alkhateeb, O. El Ayach, G. Leus, and R. W. Heath, Channel estimation and
hybrid precoding for millimeter wave cellular systems," IEEE Journal of Selected
Topics in Signal Processing, vol. 8, no. 5, pp. 831-846, 2014.
[12] V. Va, H. Vikalo, and R. W. Heath, Beam tracking for mobile millimeter wave
communication systems," 2016 IEEE Global Conference on Signal and Information
Processing (GlobalSIP), pp. 743-747, 2016.
[13] O. El Ayach, S. Rajagopal, S. Abu-Surra, Z. Pi, and R. W. Heath, Spatially sparse precoding in millimeter wave mimo systems," IEEE Transactions on Wireless Communications, vol. 13, no. 3, pp. 1499-1513, 2014.
[14] W.-L. Hung, C.-H. Chen, C.-C. Liao, C.-R. Tsai, and A.-Y. A. Wu, Low-
complexity hybrid precoding algorithm based on orthogonal beamforming code-
book," pp. 10-5, 2015.
[15] N. Boumal, B. Mishra, P. Absil, and R. Sepulchre, Manopt, a matlab toolbox for
optimization on manifolds," CoRR, vol. abs/1308.5200, 2013.
[16] a. S. P.-A.Absil, R. Mahony, Optimization algorithms matrix manifolds."