波束賦形(Beamforming)是第五代行動通訊(5G)系統中的一項關鍵技術。然而傳統的優化算法通常需要大量且複雜的迭代計算量導致在實時通訊中難以執行。近年來已有些許研究證明在通訊領域的研究中，使用基於深度學習(Deep Learning，簡稱DL)的算法可以有效地快速提高系統效能，同時降低運算時間複雜度。因此本論文提出一種採用非監督式學習(Unsupervised-learning)的DL模型算法，其整體架構的設計概念是受殘餘神經網絡(Residual Network，簡稱ResNet)的工作流程所啟發，我們命名為：受殘餘神經網路啟發式波束賦形(ResNet-Inspired Beamforming，簡稱RI-BF)算法。
在本論文所設計的RI-BF模型中，我們引入了深度展開(Deep Unfolding)技術來設計深度神經網路(Deep Neural Network，簡稱DNN)的架構，使得系統效能能參考基於梯度之波束賦形算法進行優化。該技術的引入使得我們設計的RI-BF能同時繼承迭代優化算法和基於DL的波束賦形設計算法兩者的優勢。
此外，本論文所提出的RI-BF具有三個特徵。首先，與現有基於DL的波束賦形設計算法不同，在訓練RI-BF時不需要在損失函數(Loss function)後加上正則項(Regularization term)或是在模型輸出時對其波束賦形進行功率縮放機制來滿足系統總傳送功率約束。其次，受到ResNet的模型啟發，我們搭建了一個深度展開模塊(Deep unfolding module，簡稱DUM)來模仿ResNet模型的殘差塊(Residual block)，從而進一步提高RI-BF中初始波束賦形的效能。第三，整個RI-BF是採用非監督式學習訓練的方式。因此不需要對訓練資料(Training data)進行額外的標記(Label)工作。最後，模擬結果顯示，本論文所提出之RI-BF其系統效能和運算時間複雜度都明顯優於現有DL的波束賦形設計算法和迭代優化算法。

Beamforming is a key technology in communication systems of the fifth generation and beyond. However, traditional optimization-based algorithms are often computationally prohibited from performing in a real-time manner. On the other hand, the performance of existing deep learning (DL)-based algorithms can be further improved. As an alternative, we propose an unsupervised ResNet-inspired beamforming (RI-BF) algorithm in this paper that inherits the advantages of both pure optimization-based and DL-based beamforming for better efficiency.
In particular, a deep unfolding technique is introduced to implement the optimization process of the gradient ascent beamforming algorithm for the design of our deep neural network (DNN) architecture.
Moreover, the proposed RI-BF has three features. First, unlike the existing DL-based beamforming method, which employs a regularization term for the loss function or an output scaling mechanism to satisfy system power constraints, a novel DNN architecture is introduced in RI-BF to generate initial beamforming with a promising performance. Second, inspired by the success of residual neural network (ResNet)-based DL models, a deep unfolding module is constructed to mimic the residual block of the ResNet-based model, which further improves the performance of RI-BF based on the initial beamforming. Third, the entire RI-BF is trained in an unsupervised manner; as a result, labelling efforts can be avoided. The simulation results demonstrate that the performance and computational complexity of our RI-BF improves significantly compared to the existing DL-based and optimization-based algorithms.

摘要------------------------------i
Abstract-------------------------ii
圖次------------------------------v
表次------------------------------vi
第一章 緒論----------------------1
1.1研究背景與動機------------------1
1.2 論文章節內容安排---------------4
第二章 相關背景及系統模型----------5
2.1 相關背景----------------------5
2.1.1多輸入多輸出技術---------------5
2.1.2波束賦形----------------------5
2.1.3深度神經網路------------------6
2.2 系統模型----------------------7
第三章 系統數學模型---------------10
第四章 本論文所提出RI-BF演算法-----11
4.1 總覽 RI-BF--------------------11
4.2 非監督式粗略估計器模塊----------13
4.3 非監督式深入展開模塊------------16
4.3.1 引入深度展開之原因-----------16
4.3.2 深度展開技術介紹-------------18
4.3.3 非監督式深入展開模塊之模型-----19
第五章 模擬結果與分析--------------25
5.1 模擬環境與比較對象---------------25
5.2 訓練資料產生與設定---------------25
5.3 模擬結果------------------------26
5.3.1 DUM迭代方塊數分析-------------26
5.3.2 RI-BF 傳送速率效能分析--------28
5.3.3 RI-BF 平均運算時間複雜度分析---30
第六章 結論------------------------36
參考文獻-----------------------------37

[1] L. Zheng, "5g mobile communication development trend and several key technologies, " in Matec Web of Conferences, vol. 246, Jan. 2018, p. 3034.
[2] E. Bj ̈ornson, E. G. Larsson, and M. Debbah, "Massive mimo formaximal spectral efficiency: How many users and pilots should be allocated?" IEEE Transactions on Wireless Communications, vol. 15,no. 2, pp. 1293–1308, Feb. 2016.
[3] A. F. Molisch, V. V. Ratnam, S. Han, Z. Li, S. L. H. Nguyen, L. Li, and K. Haneda, "Hybrid beamforming for massive mimo: A survey," IEEE Communications Magazine, vol. 55, no. 9, pp. 134–141, 2017.
[4] S. S. Christensen, R. Agarwal, E. D. Carvalho, and J. M. Cioffi, "Weighted sum-rate maximization using weighted mmse for mimo-bc beamforming design," IEEE Transactions on Wireless Communications, vol. 7, no. 12, pp. 4792–4799, Dec. 2008.
[5] Q. Shi, M. Razaviyayn, Z. Luo and C. He, "An Iteratively Weighted MMSE Approach to Distributed Sum-Utility Maximization for a MIMO Interfering Broadcast Channel," in IEEE Transactions on Signal Processing, vol. 59, no. 9, pp. 4331-4340, Sept. 2011.
[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 Vehicular Technology, vol. 68, no. 3, pp. 3027–3032, Mar. 2019.
[7] A. Klautau, N. Gonz ́alez-Prelcic, and R. W. Heath, "Lidar data for deep learning-based mmwave beam-selection," IEEE Wireless Communications Letters, vol. 8, no. 3, pp. 909–912, 2019.
[8] C. Wen, W. Shih, and S. Jin, "Deep learning for massive mimo csi feed back," IEEE Wireless Communications Letters, vol. 7, no. 5, pp.748–751, Oct. 2018.
[9] G. Gui, H. Huang, Y. Song, and H. Sari, "Deep Learning for an Effective Nonorthogonal Multiple Access Scheme," IEEE Transactions on Vehicular Technology, vol. 67, no. 9, pp. 8440-8450, Sep. 2018.
[10] G. Qian, Z. Li, C. He, X. Li, and X. Ding, "Power Allocation Schemes Based on Deep Learning for Distributed Antenna Systems." IEEE Access no. 8, pp. 31245-31253, 2020.
[11] J. C. C. Z. J. Tao, J. Xing, and S. Fu, "Deep neural hybrid beamforming for multi-user mmwave massive mimo system," in IEEE Global Conference on Signal and Information Processing, Nov. 2019, pp. 1–5.
[12] E. Balevi, and J. G. Andrews, "One-Bit OFDM Receivers via Deep Learning," in IEEE Transactions on Communications, vol. 67, no. 6, pp. 4326-4336, Jun. 2019.
[13] W. Xu, Z. Wu, Y. Ueng, X. You, and C. Zhang, "Improved polar decoder based on deep learning," IEEE International Workshop on Signal Processing Systems, pp. 1-6, 2017,.
[14] T. Lin, and Y. Zhu, "Beamforming design for large-scale antenna arrays using deep learning, " IEEE Wireless Communications Letters, vol. 9, no. 1, pp. 103–107, Jan. 2020.
[15] H. Huang, W. Xia, J. Xiong, J. Yang, G. Zheng, and X. Zhu, "Unsupervised learning-based fast beamforming design for downlink mimo," IEEE Access, vol. 7, pp. 7599–7605, 2019.
[16] S. Lin, and H. Narasimhan, "Towards software-defined massive mimo for 5G&B spectral-efficient networks, " in IEEE International Conference on Communications, May 2018, pp. 1–6.
[17] Y. Zhang, Y. Tian, Y. Kong, B. Zhong, and Y. Fu, "Residual dense network for image super-resolution," in IEEE Conference on Computer Vision and Pattern Recognition, Jun. 2018.
[18] E. Wang, M. Jiang, C. Qian, S. Yang, C. Li, H. Zhang, X. Wang, and X. Tang, "Residual attention network for image classification," in IEEE Conference on Computer Vision and Pattern Recognition, Jul. 2017.
[19] A. Balatsoukas-Stimming, and C. Studer, "Deep unfolding for communications systems: A survey and some new directions," IEEE International Workshop on Signal Processing Systems, pp. 266–271, 2019.
[20] T. G. Dietterich, "Ensemble methods in machine learning," Lecture Notes in Computer Science, vol. 1857, pp. 1–15, 2000.
[21] H. Sung, K. Lee, S. Park, and I. Lee, "An iterative precoder optimization method for k-user interference channel systems," in IEEE Global Telecommunications Conference, Nov. 2009, pp. 1–5.
[22] R. J. G. van Sloun, R. Cohen, and Y. C. Eldar, "Deep learning in ultrasound imaging," Proceedings of the IEEE, vol. 108, no. 1, p. 11–29, Jan. 2020.
[23] S. Khobahi, N. Naimipour, M. Soltanalian, and Y. C. Eldar, "Deep signal recovery with one-bit quantization," in IEEE International Conference on Acoustics, Speech and Signal Processing, 2019, pp. 2987–2991
[24] K. He, X. Zhang, S. Ren, and J. Sun, "Deep residual learning for image recognition," in IEEE Conference on Computer Vision and Pattern Recognition, 2016, pp. 770–778
[25] H. Yao, F. Dai, S. Zhang, Y. Zhang, Q. Tian, and C. Xu, "Dr2-net: Deep residual reconstruction network for image compressive sensing," Neuro computing, vol. 359, pp. 483 – 493, 2019.
[26] J. Yosinski, J. Clune, Y. Bengio, and H. Lipson, “How transferable are features in deep neural networks?" Advances in Neural Information Processing Systems, pp. 3320–3328, 2014.
[27] A. L. Maas, A. Y. Hannun, and A. Y. Ng, "Rectifier nonlinearities improve neural network acoustic models," in Proceedings of the 30th International Conference on Machine Learning, 2013.
[28] S. Ioffe, and C. Szegedy, "Batch normalization: Accelerating deep network training by reducing internal covariate shift," 2015.
[29] D. Kingma, and J. Ba. "Adam: A method for stochastic optimization, =" in IEEE International Conference on Acoustics, 2015.
[30] H. He, S. Jin, C. Wen, F. Gao, G. Y. Li, and Z. Xu, "Model-Driven Deep Learning for Physical Layer Communications," in IEEE Wireless Communications, vol. 26, no. 5, pp. 77-83, Oct. 2019.
[31] J. R. Hershey, J. Le Roux, and F. Weninger, "Deep unfolding: Model-based inspiration of novel deep architectures," Nov. 2014.
[32] H. He, C. Wen, S. Jin, and G. Y. Li, "A Model-Driven Deep Learning Network for MIMO Detection," in IEEE Global Conference on Signal and Information Processing, 2018, pp. 584-588.
[33] N. Samuel, T. Diskin, and A. Wiesel, "Learning to Detect," IEEE Transactions on Signal Processing, vol. 67, no. 10, pp. 2554-2564, May 2019.
[34] S. Wisdom, J. Hershey, J. Le Roux, and S. Watanabe, "Deep unfolding for multichannel source separation," in IEEE International Conference on Acoustics, Speech and Signal Processing, 2016, pp. 121-125.
[35] K. Gregor, and Y. LeCun, "Learning fast approximations of sparse coding," Proceedings of the 27th International Conference on Machine Learning, pp. 399-406, Jun. 2010.
[36] M. Borgerding, and P. Schniter, "Onsager-corrected deep learning for sparse linear inverse problems," in IEEE Global Conference on Signal and Information Processing, 2016, pp. 227-231.
[37] C. Lin, Y. Lee, W. Chung, S. Lin, and T. Lee, "Unsupervised ResNet-Inspired Beamforming Design Using Deep Unfolding Technique," submitted to IEEE Global Communications 2020.