邏輯閘在數位訊號處理中扮演重要的角色。在本研究中，我們使用光通訊軟體(OptSim)來設計以回聲狀態網路(Echo State Network, ESN)為類神經網路數學模型的光學架構。光源為一CW雷射，神經元中使用到的在光學元件包含：摻鉺光纖放大器、耦合器、光纖。輸入電信號經調制器調制光源CW雷射信號進入神經元。在本論文中，我們使用馬赫-曾德爾強度調制器與相位調制器，來調制輸入信號。在邏輯閘的輸出端，我們使用相位調制器來優化輸出結果，邏輯閘輸出信號的光強度由光偵測器接收，轉換為電信號。
邏輯閘的神經元中的耦合器提供了光干涉的效果，光干涉的輸出與輸入的為非線性的關係。我們利用此一關係來提供在ESN中所需的非線性函數效果。輸入信號偽亂數二進序列(Pseudo-Random Binary Sequence, PRBS)訊號作為訓練資料。神經元中的光纖提供了光延遲的效果，使得每一bit信號皆能與前一bit信號重疊。我們分別透過監督式與非監督式的學習方式來訓練類神經網路，最後透過誤碼率(Bit Error Rate, BER)及歸一化均方根誤差(Normalized Root Mean Square Error, NRMSE)來評估系統的表現。
在使用相位調制器調制2048位元PRBS信號的結構中，我們可得到XOR的邏輯閘(監督式學習系統誤碼率為0，NRMSE為0.20；非監督式學習系統誤碼率為0，NRMSE為0.16)。在使用強度調制器調制2048位元PRBS信號的結構中，OR邏輯閘亦呈現較低的誤碼率(誤碼率為0.0654，NRMSE為0.51)。
相位調制器與強度調制器比較起來，相位調制器的優點為: 成本較低、速度較快、元件較小、光耗損較少。以本研究結果來看，誤碼率也展現較優越的性能。因此選擇相位調制器用於光神經網路系統的輸入信號調制是比較好的選擇。

Logic gates play an important role in digital signal processing. In this work, we use the software (OptSim) for optical fiber communication system to investigate an optical system based on the echo-state networks which is a kind of the neural networks.
A continuous wave laser is used as the light source of the system. The Optical neurons consists of erbium-doped fiber amplifiers, directional couplers and optical fibers. The continuous wave laser is modulated by the modulators with the input electrical signals. In this work, we use Mach-Zehnder intensity modulators and phase modulators to modulate the input signals. The performance of the optical logic gate is optimized by a phase modulator. The optical output signal received by a photodiode to convert to electrical signals.
The directional couplers in the optical neurons provide the effect of optical interference. The input and output relation in the effect of the optical interference is non-linear. This relation provides the necessary non-linearity for the neuron networks. We use the Pseudo-Random Binary Sequence (PRBS)s the input signals. The optical fibers offer a delay to combine each bit with previous bit. The bit error rate (BER) and the normalized root mean square error (NRMSE) is analyzed to study the performance of the system.
In the system, we use the phase modulator to modulate the 2048 PRBS input signals to test the performance of the logic gates. XOR logic gate can be obtained with BER to be 0, NRMSE to be 0.20 for supervised learning, and BER to be 0, NRMSE to be 0.16 for unsupervised learning. In the system using Mach-Zehnder modulator to modulate the 2048 PRBS input signals, OR logic gate can be obtained with BER to be 0.0654, NRMSE to be 0.51.
For the commercial modulators, the phase modulators are often more compact and cost-effective than intensity modulators. The operation speed of the phase modulators can be higher and the optical loss can be lower than those of the intensity modulators. From the results obtained in this work, we can also obtain lower BER for the systems using phase modulators. Therefore, the phase modulators are more suitable to modulate the input signals for the optical neural network system.

參考文獻
[1] C. Luo, L. Jin, and Z. Sun, "MORAN: A Multi-Object Rectified Attention Network for scene text recognition," Pattern Recognition 90, 109-118 (2019).
[2] A. Czyzewski, B. Kostek, P. Bratoszewski, J. Kotus, and M. Szykulski, "An audio-visual corpus for multimodal automatic speech recognition," Journal of Intelligent Information Systems 49, 167-192 (2017).
[3] G. S. Meltzner, J. T. Heaton, Y. Deng, G. De Luca, S. H. Roy, and J. C. Kline, "Silent speech recognition as an alternative communication device for persons with laryngectomy," IEEE/ACM transactions on audio, speech, and language processing 25, 2386-2398 (2017).
[4] A. Graves, M. Liwicki, S. Fernández, R. Bertolami, H. Bunke, and J. Schmidhuber, "A novel connectionist system for unconstrained handwriting recognition," IEEE transactions on pattern analysis and machine intelligence 31, 855-868 (2008).
[5] V. Badrinarayanan, A. Kendall, and R. Cipolla, "Segnet: A deep convolutional encoder-decoder architecture for image segmentation," IEEE transactions on pattern analysis and machine intelligence 39, 2481-2495 (2017).
[6] G. E. Hinton, and R. R. Salakhutdinov, "Reducing the dimensionality of data with neural networks," science 313, 504-507 (2006).
[7] A. Graves, A.-r. Mohamed, and G. Hinton, "Speech recognition with deep recurrent neural networks," in 2013 IEEE international conference on acoustics, speech and signal processing(IEEE2013), pp. 6645-6649.
[8] Y. Chauvin, and D. E. Rumelhart, Backpropagation: theory, architectures, and applications (Psychology press, 2013).
[9] T. Panayiotou, S. P. Chatzis, and G. Ellinas, "Performance analysis of a data-driven quality-of-transmission decision approach on a dynamic multicast-capable metro optical network," Journal of Optical Communications and Networking 9, 98-108 (2017).
[10] H. Jaeger, "The “echo state” approach to analysing and training recurrent neural networks-with an erratum note," Bonn, Germany: German National Research Center for Information Technology GMD Technical Report 148, 13 (2001).
[11] F. Duport, A. Smerieri, A. Akrout, M. Haelterman, and S. Massar, "Fully analogue photonic reservoir computer," Scientific reports 6, 22381 (2016).
[12] A. Dejonckheere, F. Duport, A. Smerieri, L. Fang, J.-L. Oudar, M. Haelterman, and S. Massar, "All-optical reservoir computer based on saturation of absorption," Optics express 22, 10868-10881 (2014).
[13] H. Jaeger, and H. Haas, "Harnessing nonlinearity: Predicting chaotic systems and saving energy in wireless communication," science 304, 78-80 (2004).
[14] L. Appeltant, M. C. Soriano, G. Van der Sande, J. Danckaert, S. Massar, J. Dambre, B. Schrauwen, C. R. Mirasso, and I. Fischer, "Information processing using a single dynamical node as complex system," Nature communications 2, 1-6 (2011).
[15] L. Larger, M. C. Soriano, D. Brunner, L. Appeltant, J. M. Gutiérrez, L. Pesquera, C. R. Mirasso, and I. Fischer, "Photonic information processing beyond Turing: an optoelectronic implementation of reservoir computing," Optics express 20, 3241-3249 (2012).
[16] F. Duport, B. Schneider, A. Smerieri, M. Haelterman, and S. Massar, "All-optical reservoir computing," Optics express 20, 22783-22795 (2012).
[17] D. Marković, N. Leroux, M. Riou, F. Abreu Araujo, J. Torrejon, D. Querlioz, A. Fukushima, S. Yuasa, J. Trastoy, and P. Bortolotti, "Reservoir computing with the frequency, phase, and amplitude of spin-torque nano-oscillators," Applied Physics Letters 114, 012409 (2019).
[18] J. Dong, M. Rafayelyan, F. Krzakala, and S. Gigan, "Optical reservoir computing using multiple light scattering for chaotic systems prediction," IEEE Journal of Selected Topics in Quantum Electronics 26, 1-12 (2019).
[19] U. Paudel, M. Luengo-Kovac, J. Pilawa, T. J. Shaw, and G. C. Valley, "Classification of time-domain waveforms using a speckle-based optical reservoir computer," Optics Express 28, 1225-1237 (2020).
[20] T. Houbavlis, K. Zoiros, A. Hatziefremidis, H. Avramopoulos, L. Occhi, G. Guekos, S. Hansmann, H. Burkhard, and R. Dall'Ara, "10 Gbit/s all-optical Boolean XOR with SOA fibre Sagnac gate," Electronics Letters 35, 1650-1652 (1999).
[21] R. P. Webb, R. J. Manning, G. D. Maxwell, and A. J. Poustie, "40 Gbit/s all-optical XOR gate based on hybrid-integrated Mach-Zehnder interferometer," Electronics Letters 39, 79-81 (2003).
[22] A. Kotb, K. E. Zoiros, and C. Guo, "All-optical XOR, NOR, and NAND logic functions with parallel semiconductor optical amplifier-based Mach-Zehnder interferometer modules," Optics & Laser Technology 108, 426-433 (2018).
[23] A. Kotb, and K. E. Zoiros, "Performance analysis of all-optical XOR gate with photonic crystal semiconductor optical amplifier-assisted Mach–Zehnder interferometer at 160 Gb/s," Optics Communications 402, 511-517 (2017).
[24] I. Kang, M. Rasras, L. Buhl, M. Dinu, S. Cabot, M. Cappuzzo, L. Gomez, Y. Chen, S. Patel, and N. Dutta, "All-optical XOR and XNOR operations at 86.4 Gb/s using a pair of semiconductor optical amplifier Mach-Zehnder interferometers," Optics Express 17, 19062-19066 (2009).
[25] D. Rani, R. S. Kaler, and B. Painam, "All-optical NOR gate based on cross structures in 2D photonic crystal using logic NOT and OR gates," Journal of Optical Technology 84, 851-857 (2017).
[26] 劉慶權, "人工智慧光學電腦," (國立中央大學, 2018).
[27] T.-Y. Cheng, D.-Y. Chou, C.-C. Liu, Y.-J. Chang, and C.-C. Chen, "Optical neural networks based on optical fiber-communication system," Neurocomputing 364, 239-244 (2019).
[28] M. Lukoševičius, "A practical guide to applying echo state networks," in Neural networks: Tricks of the trade(Springer, 2012), pp. 659-686.
[29] A. Ben-Israel, and T. N. Greville, Generalized inverses: theory and applications (Springer Science & Business Media, 2003).
[30] G. Keiser, "Optical fiber communications," Wiley Encyclopedia of Telecommunications (2003).
[31] A. B. Carlson, Communication system (Tata McGraw-Hill Education, 2010).
[32] P. Horowitz, and W. Hill, "Digital-noise generation," in The art of electronics(Cambridge Univ. Press, 1989), pp. 974-975.
[33] O. S. Group, "Block Mode," in OptSim Models Reference(Synopsys, Inc.).