隨著第五代行動通訊技術(5th-Generation Mobile Communication Technology, 5G)正式進入市場並開始應用，第六代行動通訊技術(6th-Generation Mobile Communication Technology, 6G)對於覆蓋要求、提升通信速率以及更低延遲、更高可靠性等為主要突破技術，其中對於提升通信速率，6G希望能使用兆赫茲頻段做為無線傳輸作為目的。在無線通訊面對提升通信速率勢必會遇上更複雜的通道及系統響應，這時利用機器學習作為解調接收訊號的等化器為一個相當可靠的方案。該應用透過機器學習去嘗試計算受到更多非線性效應(Non-Linear Effect)的兆赫茲無線通道，能夠比起只使用基本的強制歸零等化器(Zero-Forcing Equalizer, ZF Equalizer)計算出更完整的通道響應並進行補償，在提升通信速率的情形下能有更高可靠性。
因此本研究以機器學習等化器為基礎，提出了一個二階段卷積神經網路等化器(2-staged Convolutional Neural Network Equalizer)。以現今多數神經網路等化器的架構，利用深度神經網路(Deep Neural Network, DNN)對頻域上的正交幅度調制符元(QAM symbol, Quadrature Amplitude Modulation Symbol)進行非線性通道估計與補償。本研究提出低複雜度的卷積神經網路(Convolutional Neural Network, CNN)等化器，不只能夠同時處理不同頻域及時域上的QAM符元，更完整的計算整體非線性響應，且透過QAM symbol實部及虛部的分開處理及運算，能夠在犧牲些許改善程度的情況下大幅降低所需的計算複雜度，也同時減少神經網路學習訓練集的時間，這些改善對於未來6G室內無線通訊的情景下，對於提升通信速率以及更可靠的傳輸有著相當大的幫助。
為了進一步驗證該應用的實用性，本研究實際在300十億赫茲(GHz)的次兆赫茲頻段下進行實驗，討論了不使用機器學習、使用深度神經網路以及卷積神經網路對於解調訊號的影響。實驗結果顯示，在低頻寬的情形下，CNN比起DNN有著更低複雜度且更好的提升訊號品質；在高頻寬的情形，CNN雖然需要些許更多的複雜度，但對於提升訊號品質有著比DNN更好的表現。意味著能夠傳輸更有可靠性的訊號，能穩定提升6G無線通訊的品質。

As 5th-Generation Mobile Communication Technology(5G) enters the market, 6th-Generation Mobile Communication Technology (6G) aims to achieve significant advancements in coverage, speed, latency, and reliability. To enhance communication speeds, 6G plans to utilize terahertz frequency bands for wireless transmission. Given the complex channel and system responses, machine learning has emerged as a reliable solution for demodulating received signals. Machine learning can better calculate terahertz wireless channels affected by nonlinear effects, providing more complete channel responses and compensation compared to basic Zero-Forcing equalizers (ZF Equalizer), thus enhancing reliability and speed.
This study proposes a two-staged convolutional neural network (CNN) equalizer based on machine learning. Employing deep neural networks (DNN) for nonlinear channel estimation and compensation of quadrature amplitude modulation (QAM) symbols in the frequency domain, the proposed low-complexity CNN equalizer can handle QAM symbols in both frequency and time domains. By separately processing the real and imaginary parts of QAM symbols, it significantly reduces computational complexity and training time, at the cost of some improvement. These enhancements are particularly beneficial for future 6G indoor wireless communication, boosting speed and transmission reliability.
To validate the practicality of this application, experiments were conducted at a sub-terahertz frequency band of 300 GHz. The study compared the impact of not using machine learning, using DNNs, and using CNNs on signal demodulation. Results show that under low bandwidth conditions, CNNs offer lower complexity and better signal quality improvement than DNNs. In high bandwidth conditions, although CNNs require slightly more complexity, they outperform DNNs in signal quality enhancement. This indicates that CNNs can transmit more reliable signals, steadily improving the quality of 6G wireless communications.

[1] International Telecommunication Union. (2015). IMT Vision – “Framework and overall objectives of the future development of IMT for 2020 and beyond” (Recommendation ITU-R M.2083).
[2] Saad, W., Bennis, M. and Chen, M. (2019). “A Vision of 6G Wireless Systems Applications Trends Technologies and Open Research Problems”
[3] Wei Jiang, B. Han, Mohammad Asif Habibi, H. Schotten. “The road towards 6g: A comprehensive survey.” IEEE Open Journal of the Communications Society, 2:334–366, 2021
[4] M. Chen, U. Challita, W. Saad, C. Yin, and M. Debbah, “Artificial neural networks-based machine learning for wireless networks: A tutorial,” IEEE Communications Surveys and Tutorials, to appear, 2019
[5] Torkaman, Pouya & Latifi, Seyed Mostafa & Feng, Kai-Ming & Yang, Shang-Hua. “Nonlinearity-Robust IM/DD THz Communication System via Two-Stage Deep Learning Equalizer.” IEEE Communications Letters, PP, 1-1, 2024
[6] Lukasz Lopacinski, Miroslav Marinkovic, Goran Panic, Mohamed Hussein Eissa, Alireza Hasani, Karthik Krishnegowda, and Rolf Kraemer. “Data link 103 layer processor for 100 gbps terahertz wireless communications in 28 nm cmos technology”. IEEE Access, 7:44489–44502, 2019
[7] Robert A Minasian, EHW Chan, and Xiaoke Yi. “Microwave photonic signal processing.” Optics Express, 21(19):22918–22936, 2013
[8] Xianbin Yu, Rameez Asif, Molly Piels, Darko Zibar, Michael Galili, Toshio Morioka, Peter U. Jepsen, and Leif K. Oxenløwe. “60 gbit/s 400 ghz wireless transmission.” pages 4–6, 2015.
[9] Xianbin Yu, S Jia, Hao Hu, Michael Galili, Toshio Morioka, Peter Uhd Jepsen, and Leif Katsuo Oxenløwe. “160 gbit/s photonics wireless transmission in the 300-500 ghz band”. Apl Photonics, 1(8), 2016
[10] Yuan Zhang, Miao Song, and Haitao Zhang. “Experimental Demonstration of 128QAM-OFDM Encoded Terahertz Signals over 20-km SMF.” pages 64–67, 2021
[11] S Nellen, T Ishibashi, A Deninger, RB Kohlhaas, L Liebermeister, M Schell, and B Globisch. “Experimental comparison of utc-and pin-photodiodes for continuous-wave terahertz generation.” Journal of Infrared, Millimeter, and Terahertz Waves, 41:343–354, 2020.
[12] Tadao Nagatsuma, Masato Sonoda, Taiki Higashimoto, Ryo Kimura, Li Yi, and Hiroshi Ito. “300-ghz-band wireless communication using fermi-level managed barrier diode receiver.” pages 762–765, 2019
[13] A. M. Jaradat, J. M. Hamamreh and H. Arslan, "OFDM with subcarrier number modulation", IEEE Wirel. Commun. Lett., vol. 7, no. 6, pp. 914-917, 2018
[14] Jürgen Schmidhuber “Deep Learning in Neural Network An Overview” Neural Networks Volume 61, January 2015, Pages 85-117
[15] Ryan Dahl, Mohammad Norouzi, Jonathon Shlens, “Pixel Recursive Super Resolution” 2017
[16] Chen Wang; Kaihui Wang; Yuxuan Tan; Feng Wang; Bohan Sang; Weiping Li; Wen Zhou; Jianjun Yu, ”High-Speed Terahertz Band Radio-Over-Fiber System Using Hybrid Time-Frequency Domain Equalization” V 34, IEEE Photonics Technology Letters, 2022 [15] Ryan Dahl, Mohammad Norouzi, Jonathon Shlens, “Pixel Recursive Super Resolution” 2017
[17] Chen Wang; Kaihui Wang; Yuxuan Tan; Feng Wang; Bohan Sang; Weiping Li; Wen Zhou; Jianjun Yu, ”High-Speed Terahertz Band Radio-Over-Fiber System Using Hybrid Time-Frequency Domain Equalization” V 34, IEEE Photonics Technology Letters, 2022