在本論文中，我們針對無線通訊系統提出一個基於深度學習(DL)的通道估測方法，其中通道的時間/頻率響應被視為符合3GPP進階長期演進技術(LTE-A)規範的二維圖像資料；所提出之DL通道估測方法被稱為AttenChNet，乃由一個具有超分辨率之卷積神經網路(CNN)以及一個包含一個自注意力層和幾個全連接層的神經網路所組成。我們首先依據LTE-A規範之參考訊號安排方式，並利用傳統的最小平方(LS)演算法和適當的內插方法產生初步通道估測結果；接著，產生足夠大量的初步通道估測結果以組成一個資料集，再據以對AttenChNet進行離線訓練，然後將訓練好的模型應用至線上階段。基於各種通道模型之電腦模擬結果顯示，與傳統的LS通道估測演算法以及一個被稱為ChannelNet (由一個具有超分辨率的CNN與一個去噪CNN所組成)的相關作法相比，所提出之AttenCNet可提供明顯較佳的通道估測均方誤差效能。

In this thesis, we propose a channel estimation scheme based on deep learning (DL) for wireless communication systems. It treats the time/frequency response of a channel as two-dimensional image data that conforms to the 3GPP Long Term Evolution-Advanced (LTE-A) specifications. The DL-based scheme is referred to as AttenChNet that is composed of a super-resolution convolutional neural network (CNN) and a neural network with a self-attention layer and several fully-connected layers. According to the LTE-A reference signals’ arrangement, preliminary channel estimation results are generated by using the conventional least squares (LS) algorithm and an appropriate interpolation method. The AttenChNet is trained offline by a sufficiently large dataset with each element formed by such preliminary channel estimates, and the well-trained model is then applied at the online stage. Simulation results based on various channel models demonstrate that the proposed AttenCNet offers much better channel estimation performance in terms of the mean-squared error, as compared with the conventional LS algorithm and a previous related ChannelNet formed by a super-resolution CNN and a denosing CNN.

Abstract
Contents
List of Figures
List of Tables
I. Introduction----------------------------------------1
II. System Model---------------------------------------4
A. Transmitted Signal Architecture-------------------4
B. Downlink FDD-LTE System---------------------------5
III. Proposed Method-----------------------------------7
A. Structure of Convolutional Neural Network (CNN)---7
B. AttenChNet----------------------------------------8
C. Time Complexity----------------------------------13
IV. Simulations---------------------------------------15
A. Simulation Settings------------------------------15
B. Simulation Results-------------------------------17
V. Conclusion-----------------------------------------21
References--------------------------------------------22

[1]J.-J. van de Beek; O. Edfors; M. Sandell; S.K. Wilson; P.O. Borjesson, "On channel estimation in OFDM systems," in Proc. IEEE 45th Veh. Technol. Conf., Chicago, USA, Jul, 1995, pp. 815–819.
[2]H. Ye, G. Y. Li and B. -H. Juang, "Power of deep learning for channel estimation and signal detection in OFDM systems," IEEE Wireless Commun. Lett., vol. 7, no. 1, pp. 114–117, Feb. 2018.
[3]Y. Yang, F. Gao, X. Ma and S. Zhang, "Deep learning-based channel estimation for doubly selective fading channels," IEEE Access, vol. 7, pp. 36579–36589, Mar. 2019.
[4]Q. Bai, J. Wang, Y. Zhang and J. Song, "Deep learning-based channel estimation algorithm over time selective fading channels," IEEE Trans. Cogn. Commun. Netw., vol. 6, no. 1, pp. 125–134, Mar. 2020.
[5]J. Gu, C. Shan, X. Chen, H. Yin and W. Wang, "A novel pilot-aided channel estimation scheme based on RNN for FDD-LTE systems," in Proc. IEEE Int. Conf. Wireless Commun. Signal Process. (WCSP), Hangzhou, China, Dec. 2018, pp. 1–5.
[6]Y. Liao, Y. Hua, X. Dai, H. Yao and X. Yang, "ChanEstNet: A deep learning based channel estimation for high-speed scenarios," in Proc. IEEE Int. Conf. Commun. (ICC), Shanghai, China, Jul. 2019, pp. 1–6.
[7]M. Soltani, V. Pourahmadi, A. Mirzaei and H. Sheikhzadeh, "Deep learning-based channel estimation," IEEE Commun. Lett., vol. 23, no. 4, pp. 652–655, Apr. 2019.
[8]C. Dong, C. C. Loy, K. He and X. Tang, "Image super-resolution using deep convolutional networks," IEEE Trans. Pattern Anal. Mach. Intell., vol. 38, no. 2, pp. 295–307, Feb. 2016.
[9]K. Zhang, W. Zuo, Y. Chen, D. Meng and L. Zhang, "Beyond a Gaussian denoiser: residual learning of deep CNN for image denoising," IEEE Trans. Image Process., vol. 26, no. 7, pp. 3142–3155, Jul. 2017.
[10]V Mnih, N Heess, A Graves, "Recurrent models of visual attention," in Proc. Neural Inf. Process. Syst. (NIPS), Montreal, Canada, Dec. 2014, pp. 2204–2212.
[11]D. Bahdanau, K. H. Cho, Y. Bengio, "Neural machine translation by jointly learning to align and translate," in Proc. Int. Conf. Learn. Represent. (ICLR), San Diego, USA, May. 2015, pp.1–15.
[12]A. Vaswani, N. Shazeer, N. Parmar, and J. Uszkoreit, "Attention is all you need," in Proc. Neural Inf. Process. Syst. (NIPS), Long Beach, USA, Dec. 2017, pp. 6000–6010.
[13]Evolved Universal Terrestrial Radio Access (E-UTRA); Physical Channels and Modulation, 3GPP TS 36.211 V 16.5.0, Mar. 2021.
[14]Evolved Universal Terrestrial Radio Access (E-UTRA); Base Station (BS) Radio Transmission and Reception, 3GPP TS 36.104 V 17.1.0, Mar. 2021.
[15]D. Kingma and J. Ba, "Adam: A method for stochastic optimization," in Proc. Int. Conf. Learn. Represent. (ICLR), San Diego, USA, May. 2015, pp. 1–15.