非正交多工存取(Non-Orthogonal Multiple Access, NOMA)的概念已被決定當作2020年以後5G的無線接取技術。將多個正交頻分多工(Orthogonal Frequency-Division Multiplexing, OFDM)分配不同功率並疊加形成NOMA傳遞給多個用戶，將可提升系統的頻譜效益。接收者再採用連續干擾消除法(Successive Interference Cancellation, SIC)做解調，然而離發送者較近的近處基地台做解調要依靠來自遠處基地台的訊息，這樣的方法造成不同的基地台有不同的複雜度問題，於是我們利用機器學習(Machine Learning, ML)取代SIC做解調改善系統複雜度不公平的問題。
本篇論文模擬光纖無線整合(Radio over Fiber, RoF)網路架構，於實驗中建構了NOMA-OFDM的光載中頻訊號並提出兩種類型的ML演算法，分別是K-means分群法(K-means clustering, K-means)與人工神經網路(Artificial Neural Network, ANN)兩種演算法，並與SIC做比較。實驗中量測兩個用戶與三個基地台的情況，前者使用6dB的功率比來分配功率，接著個別傳送B2B與25公里的光纖來模擬不同基地台與中央機房的短距離傳輸，後者則使用1:4:16的功率比來分配功率，接著個別傳送0.1公里、4.2公里以及14.7公里的光纖來模擬不同基地台與中央機房的短距離傳輸，透過比較利用SIC解調所得到的位元錯誤率(Bit Error Rate, RER)以及利用不同的ML演算法解調所得到的BER做比較，驗證了使用ML比SIC更能得到較佳的BER，且ANN相較於K-means更能得到較佳的BER。

The concept of Non-Orthogonal Multiple Access (NOMA) will become one of the wireless access technologies for 5G after 2020. By allocating multiple Orthogonal Frequency-Division Multiplexing (OFDM) with different powers and superimposing them to form a NOMA which is delivered to multiple base stations, the spectrum efficiency of the system will be largely improved. The receiver used to apply the Successive Interference Cancellation (SIC) for NOMA demodulation, but the base station closest to the transmitter must conduct the demodulation by relying on messages from other base stations. Such an approach leads to problems with the fairness of the system.
We use Machine Learning (ML) to replace SIC for demodulation to improve system security. This paper simulates a Radio over Fiber (RoF) network architecture. In the experiment, the optical IF signal of NOMA-OFDM is employed. Two types of ML algorithms are proposed, namely K-means clustering (K-means) and artificial neural network (Artificial Neural Network, ANN), and compared with SIC. In the experiment, the scenarios of two base stations and three base stations are measured. The former uses a power ratio of 6 dB to distribute powers between NOMA signals respectively for a B2B and 25-km fiber transmissions, the short-distance transmission between different base stations (BS) and central office (CO). The latter uses a power ratio of 1:4:16, and then the NOMA signal individually transmits via 0.1 km, 4.2 km, and 14.7 km of fiber. By comparing the bit error rates (RERs) obtained by SIC demodulation and the BER obtained by demodulating with different ML algorithms, the better BER can be obtained by using ML than using SIC, while NN can get better BER than K-means.

中文摘要 I
ABSTRACT II
致謝 III
目錄 IV
圖目錄 VI
表目錄 IX
第一章 緒論 1
1.1前言 1
1.2研究目的與動機 5
1.3 論文架構 6
第二章 訊號介紹與實驗元件原理 7
2.1 數位訊號NOMA-OFDM的產生與接收 7
2.2 電光轉換器-馬赫詹德調變器 11
第三章 K-MEANS分群法 14
3.1 疊加編碼調製 14
3.2 機器學習 18
3.3 K-MEANS分群法 19
第四章 人工神經網路 22
4.1 人工神經網路原理介紹 22
4.2 偏壓的意義 29
4.3向前傳播法 30
4.4梯度下降法 34
4.5反向傳播法 40
4.6權重的初始設定 47
第五章 實驗架構設置與結果 49
5.1 無線光纖整合網路系統 49
5.2 兩個用戶端的實驗架構與結果 51
5.3 三個用戶端的實驗架構與結果 58
第六章 結論 62
參考文獻 63

[1] X. Wang, "OFDM and its application to 4G," in Wireless and Optical Communications, 2005. 14th Annual WOCC 2005. International Conference on, 2005, p. 69: IEEE.
[2] J. J. J. o. l. t. Armstrong, "OFDM for optical communications," vol. 27, no. 3, pp. 189-204, 2009.
[3] D. Bandyopadhyay and J. J. W. P. C. Sen, "Internet of things: Applications and challenges in technology and standardization," vol. 58, no. 1, pp. 49-69, 2011.
[4] E. Brown. (2018). Introduction on Optical Communication and Its Advantages. Available: https://medium.com/@echobrown1314/introduction-on-optical-communication-and-its-advantages-b74e01d6e181
[5] Z. Pi and F. J. I. c. m. Khan, "An introduction to millimeter-wave mobile broadband systems," vol. 49, no. 6, 2011.
[6] V. Jungnickel et al., "The role of small cells, coordinated multipoint, and massive MIMO in 5G," vol. 52, no. 5, pp. 44-51, 2014.
[7] Y. Huang, S. Li, Y. T. Hou, and W. Lou, "GPF: A GPU-based Design to Achieve," in Proceedings of the 24th Annual International Conference on Mobile Computing and Networking, 2018, pp. 207-222: ACM.
[8] J. I. Choi, M. Jain, K. Srinivasan, P. Levis, and S. Katti, "Achieving single channel, full duplex wireless communication," in Proceedings of the sixteenth annual international conference on Mobile computing and networking, 2010, pp. 1-12: ACM.
[9] S. R. Islam, N. Avazov, O. A. Dobre, K.-S. J. I. C. S. Kwak, and Tutorials, "Power-domain non-orthogonal multiple access (NOMA) in 5G systems: Potentials and challenges," vol. 19, no. 2, pp. 721-742, 2017.
[10] L. Dai, B. Wang, Y. Yuan, S. Han, I. Chih-Lin, and Z. J. I. C. M. Wang, "Non-orthogonal multiple access for 5G: solutions, challenges, opportunities, and future research trends," vol. 53, no. 9, pp. 74-81, 2015.
[11] R. S. Michalski, J. G. Carbonell, and T. M. Mitchell, Machine learning: An artificial intelligence approach. Springer Science & Business Media, 2013.
[12] S. McCartney, ENIAC: The triumphs and tragedies of the world's first computer. Walker & Company, 1999.
[13] A. Hodges, Alan Turing: The Enigma: The Enigma. Random House, 2012.
[14] D. J. A. I. Marr, "Artificial intelligence—a personal view," vol. 9, no. 1, pp. 37-48, 1977.
[15] wangyunfei12345. (2018). Introduction to Machine Learning. Available: https://www.smwenku.com/a/5b8145e72b71772165abcc8a
[16] M. T. Hagan, H. B. Demuth, M. H. Beale, and O. De Jesús, Neural network design. Pws Pub. Boston, 1996.
[17] D. H. Hubel and T. N. J. T. J. o. p. Wiesel, "Receptive fields, binocular interaction and functional architecture in the cat's visual cortex," vol. 160, no. 1, pp. 106-154, 1962.
[18] S. J. I. J. o. U. Hochreiter, Fuzziness and K.-B. Systems, "The vanishing gradient problem during learning recurrent neural nets and problem solutions," vol. 6, no. 02, pp. 107-116, 1998.
[19] Y. LeCun, Y. Bengio, and G. J. n. Hinton, "Deep learning," vol. 521, no. 7553, p. 436, 2015.
[20] K. Krishna, M. N. J. I. T. o. S. Murty, Man,, and P. B. Cybernetics, "Genetic K-means algorithm," vol. 29, no. 3, pp. 433-439, 1999.
[21] L. PATEL. (2018). Machine Learning, Deep Learning & the Wisdom of the Crowd. Available: https://www.jumio.com/deep-learning-online-identity-verification/
[22] proximacentauri360. (2012). OFDM. Available: https://proximacentauri360.wordpress.com/ofdm/
[23] L. Lei, "From Orthogonal to Non-orthogonal Multiple Access: Energy-and Spectrum-Efficient Resource Allocation," Linköping University Electronic Press, 2016.
[24] T. Ratnarajah. (2018). Non-Orthogonal Multiple Access (NOMA). Available: http://www.profratnarajah.org/noma-communication.html
[25] Y. Zhao and S.-G. J. I. T. o. C. Haggman, "Intercarrier interference self-cancellation scheme for OFDM mobile communication systems," vol. 49, no. 7, pp. 1185-1191, 2001.
[26] I. c. s. I. am. (2018). Gradient Descent. Available: https://ithelp.ithome.com.tw/articles/10198147
[27] H. Al-Raweshidy and S. Komaki, Radio over fiber technologies for mobile communications networks. Artech House, 2002.
[28] M. Bakaul. (2018). Millimeter-Wave Fiber-Radio Systems for Broadband Wireless Access Applications. Available: https://people.eng.unimelb.edu.au/bakaulm/research.html
[29] H. J. E. L. Schmuck, "Comparison of optical millimetre-wave system concepts with regard to chromatic dispersion," vol. 31, no. 21, pp. 1848-1848, 1995.