多輸入多輸出技術可以顯著提高通訊系統的頻譜使用效率，在行動裝置上實做時，亦常常搭配天線選擇技術一同使用，使用天線選擇技術的優點包括可以完整發揮多輸入多輸出系統的效能提升優勢與降低傳送端傳送資料流的平均功耗。但是，綜觀現有天線選擇技術之相關技術，可以發現大部分的天線選擇技術都因為複雜度過高或者不能實時完成天線選擇任務而無法在行動裝置上使用，業界被迫採用具備單天線選擇思維的最大化選定通道係數之 -norm 方法來進行天線選擇任務。

本論文提出利用有限狀態機的天線選擇方法，本方法可以套用在任何具備固定天線配置的行動裝置天線選擇問題上，經過空口測試證明，本方法可以在行動裝置可負荷的計算複雜度下得到相較於最大化選定通道係數之 -norm 方法更好的效能結果，並且解決了最大化選定通道係數之 -norm 方法可能發生的效能驟降問題。

本論文亦提出了基於深度學習的天線選擇網路，在訓練過程中，不同於傳統方法透過增加資料量與其他數學手段來減緩深度學習會遇到的過擬合問題，我們使用了隨機資料位移層近乎完美的解決了過擬合問題，最後，我們也利用了具備集成學習的天線選擇網路來進一步提升本方法之效能，在變化的模擬環境當中，本方法在複雜度僅僅倍於隨機天線集合選擇方法的前提下，皆能展現近乎窮舉法的天線選擇效能結果，是一種非常具有商業化潛力的天線選擇解決方案。

MIMO technology can improve the spectral efficiency of the communication system obviously. And we usually employ antenna selection technology when we implement MIMO on the mobile device. The benefit of antenna selection are that we can make the most of the advantage of MIMO technology and lower the total power of the transmitter. But most of the antenna selection technology are too complex to be implemented on the mobile device. Industry are forced to use the Maximum norm method which is based on the concept of choosing the single antenna.

In this paper, we purpose the finite state machine antenna selection method. This method can solve any antenna selection problem with fix antenna setting. In the over the air test, the purpose method can surpass the Maximum norm method in terms of performance under affordable computing complexity. And the performance of this method will not drop immediately which is common in Maximum norm method.

We also purpose the antenna selection network based on deep learning. Instead of the traditional method like increasing the number of the training samples or using some mathematical methods, we employ the random channel shifter to solve the overfitting problem in the training process perfectly. Finally, we use the antenna selection with the ensemble learning to enhance the ability of the antenna selection network. In various testing environment, this method can achieve the performance close to brute force method when the complexity is only many times of random selection.

論文審定書 i
致謝 ii
摘要 iii
Abstract iv
目錄 v
圖次 vii
表次 viii
第一章 緒論 1
1.1 研究背景與動機 1
1.2 論文安排 4
第二章 相關背景 5
2.1 多輸入多輸出技術 5
2.2 系統模型 5
2.3 優化驅動決策天線選擇算法 6
2.4 數據驅動預測天線選擇算法 7
2.4.1 SVM算法 7
2.4.2 K-NN算法 10
第三章 利用有限狀態機的天線選擇 12
3.1 利用有限狀態機的天線選擇機制 12
3.2 有限狀態機的天線選擇機制實例 13
3.2.1 初始方法 16
3.2.2 穩態狀態 18
3.2.3 暫態狀態 19
第四章 利用深度學習的天線選擇 22
4.1 深度學習的天線選擇機制 22
4.2 天線選擇模型網路結構 (ASNet) 24
4.2.1 隨機資料位移層 24
4.2.2 類神經網路層 26
4.2.3 天線選擇模型訓練方式 27
4.2.4 具備集成學習的天線選擇網路 28
第五章 模擬結果與分析 32
5.1 Capacity based FSMAS 在 OTA 的效能結果 32
5.2 ASNet 資料生成 35
5.3 ASNet 模擬結果 36
第六章 結論 47
參考文獻 48

[1] F. Rusek, D. Persson, B. Lau, E. Larsson, T. Marzetta, O. Edfors, and F. Tufvesson, “Scaling up MIMO: Opportunities and challenges with very large arrays,” IEEE Signal Process. Mag., vol. 30, no. 1, pp. 40–60, 2013.
[2] E. Nachmani et al., “Learning to decode linear codes using deep learning,” in “Proc. Allerton Conf.,” (Illinois, USA, 2016)
[3] C. K. Wen, W. T. Shih, and S. Jin, “Deep learning for massive MIMO CSI feedback,” IEEE Wireless Commun. Lett., 2018. DOI: 10.1109/LWC.2018.2818160
[4] P. Domingos, “A few useful things to know about machine learning”. Commun. ACM, vol. 55, no. 10, pp. 78-87, 2012.
[5] R. W. Heath, Jr., S. Sandhu, and A. J. Paulraj, “Antenna selection for spatial multiplexing systems with linear receivers,” IEEE Commun. Lett., vol. 5, no. 4, pp. 142–144, Apr. 2001.
[6] S. Sanayei and A. Nosratinia, “Antenna selection in MIMO systems,” IEEE Commun. Mag., vol. 42, no. 10, pp. 68–73, Oct. 2004.
[7] D. A. Gore, R. U. Nabar, and A. J. Paulraj, “Selecting an optimal set of transmit antennas for a low rank matrix channel,” ” in Proc. IEEE Int. Conf. Acoustics, Speech and Signal Processing, Jun. 2000.
[8] J. Joung, “Machine learning-based antenna selection in wireless communications,” IEEE Commun. Lett., vol. 20, no. 11, pp. 2241–2244, Nov. 2016
[9] J. Milgram, M Cheriet, and R. Sabourin, “One against one or one against one all: Which one is better for handwriting recognition with SVMs?”. in Tenth International Workshop on Frontiers in Handwriting Recognition, Oct. 2006
[10] N. Srivastava, G. Hinton, A. Krizhevsky, I. Sutskever, and R. Salakhutdinov, “Dropout: A simple way to prevent neural networks from overfitting,” The Journal of Machine Learning Research, 15(1): 1929–1958, 2014.
[11] K. He, X. Zhang, S. Ren, and J. Sun. “Delving deep into rectifiers: Surpassing human-level performance on imagenet classification,” in IEEE Int. Conf. Comput. Vis pp 1026–1034, 2015.
[12] I. J. Goodfellow, D. Warde-Farley, M. Mirza, A. Courville, and Y. Bengio, “Maxout networks,” arXiv:1302.4389, 2013.
[13] S. Ioffe, and C. Szegedy, “Batch normalization: Accelerating deep network training by reducing internal covariate shift,” arXiv:1502.03167, 2015.
[14] T. G. Dietterich, “Ensemble methods in machine learning,” Lecture Notes in Computer Science, vol. 1857, pp. 1–15, 2000.