本文主要目標為探討神經網路接收器[1]以及經由正交分頻多工(Orthogonal frequency-division multiplexing, OFDM)後的訊號，在多路徑衰減(Multipath Fading)與脈衝雜訊(Impulse Noise)干擾的雙重影響下，是否能透過神經網路的學習，讓接收器有更好的錯誤更正能力。
對於有線通訊或無線通訊，脈衝雜訊的出現無可避免，瞬間的強力干擾不僅會破壞傳送的訊號，更會使系統在資料的解讀上產生大量的錯誤。本文將介紹的雜訊通道模型有三種：第一種是常見的可加性高斯白雜訊(Additive White Gaussian Noise, AWGN)通道模型，第二種是白努利-高斯(Bernoulli-Gaussian)脈衝雜訊通道模型[2]，最後則是本文模擬所採用的固定脈衝雜訊通道模型。
本文將使用全連接神經網路的架構，來模擬訊號在多路徑衰減下，受雜訊干擾的性能。以AWGN所訓練出來的模型作為比較，再介紹微調(Fine-tune)[3]的訓練方法，來訓練脈衝雜訊模型，提升受脈衝干擾的性能。

In this paper, we study the performance of OFDM systems in multipath fading and subject to impulse noise interference, using deep learning whether to improve the error-correcting ability of the neural network receiver[1].
Impulse noise can not avoid in the wired or wireless communication systems. As the impulse noise is different from the general AWGN noise, the energy of impulse noise is often hundreds of times that of the AWGN. Always make the systems have a lot of errors. In this paper, we will introduce three types of the noise channel model: the first type is common AWGN channel model, the second type is Bernoulli-Gaussian(BG) impulse noise channel model[2], and the last type is the fixed impulse noise channel model which we used.
In this paper, we will use the Fully Connected Neural Networks(FCNNs) structure, to simulate the performance of signals in multipath fading and impulse noise interference. And using the fine-tune[3] training method, used to train the impulse noise model to compare with AWGN model.

[1] Hao Ye, Geoffrey Ye Li, and Biing-Hwang Juang, “Power of Deep learning for Channel Estimation and Signal Detection in OFDM Systems,” IEEE Commun. Lett., vol. 7, no. 1, pp. 114-117, Feb. 2018.
[2] 陳韋銘, “極化碼基於無記憶性脈衝通道之效能分析,” 國立臺灣科技大學電機工程研究所碩士論文, 民國一百零七年七月.
[3] Hyeji Kim, Yihan Jiang, Ranvir Rana, Sreeram Kannan, Sewoong Oh, Pramod Viswanath, “Communication Algorithms via Deep Learning,” University of Washington, May 2018.
[4] A. Krizhevsky, I. Sutskever, and G. E. Hinton, “Imagenet classification with deep convolutional neural networks,” in Proc. Adv. Neural Inf. Process. Syst., 2012, pp. 1097–1105.
[5] K. Cho et al. (2014). “Learning Phrase Representations Using RNN Encoder- Decoder for Statistical Machine Translation.” [Online]. Available: http://arxiv.org/abs
/1406.1078
[6] C. Weng, D. Yu, S. Watanabe, and B.-H. F. Juang, “Recurrent deep neural networks for robust speech recognition,” in Proc. ICASSP, Florence, Italy, May 2014, pp. 5532–5536.