物聯網裝置的正在迅速發展與廣布。因此我們希望提出一套物聯網裝置的合作模式，讓各個物聯網裝置在傳輸過程中可以整合彼此的運算能力，來完成複雜度較高的任務。基於這樣的理念，本篇論文將模型化簡為一個傳送端、一個接收端，以及一個中繼站，並以圖片分類作為探討的任務。
在傳統通訊中，傳送端在傳送圖片前必須先對圖片進行訊源編碼(壓縮)以降低傳輸量；而接收端則必須經過解碼、解壓縮等步驟獲取原圖資料，才能進行進一步圖片處理(例如：圖片辨識)。
在我們所提出的方法中，傳送端、接收端以及中繼站皆由類神經網路(Neural Network)實現。我們將三段類神經網路連接為一個深度學習網路(Deep neural network)進行共同訓練，並以不可訓練的雜訊層模擬傳輸通道。在所提出的網路中，傳輸端的前若干層以及接收端的最後若干層，皆是依據典型的圖片辨識類神經網路所設計；而在傳輸端以及中繼站的最後若干層的設計，則是用以滿足通訊上傳送能量與符碼長度限制。透過訓練此深度網路，我們達到較傳統「傳輸後辨識」之方法更高的辨識準確率，且在低訊雜比情境下優勢更加顯著。

Considering the development and deployment of Internet-of-Things (IoT) devices, we aim to propose a scheme where machine learning tasks can be collaboratively accomplished during the transmission among devices. Based on this vision, this work simplifies the model to a single hopping relay channel, and chooses the task as image classification.
To transmit an image through a wireless communication channel in the conventional system, the source first compress the image to reduce the amount of data transmission. The destination need to decode and decompress the image for further tasks such as classification.
In our proposed method, all functions of source, relay, and destination are performed by neural networks. By concatenating them together, with non-trainable layers representing the noisy channel placed in between, we get a deep neural network.
The first few layers of the source, as well as the last few layers of the destination, are designed according to typical image classification neural network model. The last few layers at the destination and the layers at the relay are designed to mimic the conventional signal transmission under the constraint of average power and symbol length.
By training the network at the source, the relay, and the destination jointly, we yield a higher classification accuracy than conventional methods where compression and transmission are done separately. The superiority is even more obvious when the signal-to-noise ratio is low.

Abstract i
Contents ii
1 Introduction 1
2 Background and Related Works 6
3 System Model 10
4 Conventional Method 13
5 Deep Learning Based End-to-End Communications 18
6 Experimental Results 22
6.1 Conventional method 22
6.2 Deep learning based source-channel coding 25
7 Conclusion 39
Bibliography 40

[1] Y. LeCun, L. Bottou, Y. Bengio, and P. Haffner, “Gradient-based learning applied to document recognition,” Proceedings of the IEEE, vol. 86, pp. 2278–2324, November 1998.
[2] A. Krizhevsky, I. Sutskever, and G. E. Hinton, “Imagenet classification with deep con- volutional neural networks,” in Advances in Neural Information Processing Systems 25 (F. Pereira, C. J. C. Burges, L. Bottou, and K. Q. Weinberger, eds.), pp. 1097–1105, Curran Associates, Inc., 2012.
[3] K. Simonyan and A. Zisserman, “Very deep convolutional networks for large-scale image recognition,” in 3rd International Conference on Learning Representations, ICLR 2015, San Diego, CA, USA, May 7-9, 2015, Conference Track Proceedings, 2015.
[4] C. Szegedy, W. Liu, Y. Jia, P. Sermanet, S. Reed, D. Anguelov, D. Erhan, V. Vanhoucke, and A. Rabinovich, “Going deeper with convolutions,” in Computer Vision and Pattern Recognition (CVPR), 2015.
[5] L. Wan, M. Zeiler, S. Zhang, Y. L. Cun, and R. Fergus, “Regularization of neural networks using dropconnect,” in Proceedings of the 30th International Conference on Machine Learning (S. Dasgupta and D. McAllester, eds.), vol. 28 of Proceedings of Machine Learning Research, (Atlanta, Georgia, USA), pp. 1058–1066, PMLR, 17–19 Jun 2013.
[6] G. E. Hinton, N. Srivastava, A. Krizhevsky, I. Sutskever, and R. Salakhutdinov, “Im- proving neural networks by preventing co-adaptation of feature detectors,” CoRR, vol. abs/1207.0580, 2012.
[7] P. G. Sherwood and K. Zeger, “Progressive image coding for noisy channels,” IEEE Signal Processing Letters, vol. 4, pp. 189–191, July 1997.
[8] Jujian Zhang, Chunli Chen, and Chaoyuan Lv, “A robust image transmission strategy over wireless channels,” in 2008 IEEE International Conference on Service Operations and Logistics, and Informatics, vol. 1, pp. 606–609, Oct 2008.
[9] Jianfei Cai and Chang Wen Chen, “Robust joint source-channel coding for image trans- mission over wireless channels,” IEEE Transactions on Circuits and Systems for Video Technology, vol. 10, pp. 962–966, Sep. 2000.
[10] S. Drner, S. Cammerer, J. Hoydis, and S. t. Brink, “Deep learning based communication over the air,” IEEE Journal of Selected Topics in Signal Processing, vol. 12, pp. 132–143, Feb 2018.
[11] X. Jin and H. Kim, “Deep learning detection in mimo decode-forward relay channels,” IEEE Access, vol. 7, pp. 99481–99495, 2019.
[12] C. Wen, W. Shih, and S. Jin, “Deep learning for massive mimo csi feedback,” IEEE Wireless Communications Letters, vol. 7, pp. 748–751, Oct 2018.
[13] E. Nachmani, Y. Be’ery, and D. Burshtein, “Learning to decode linear codes using deep learning,” in 2016 54th Annual Allerton Conference on Communication, Control, and Computing (Allerton), pp. 341–346, Sep. 2016.
[14] L. Lugosch and W. J. Gross, “Neural offset min-sum decoding,” in 2017 IEEE Interna- tional Symposium on Information Theory (ISIT), pp. 1361–1365, June 2017.
[15] F. Liang, C. Shen, and F. Wu, “An iterative bp-cnn architecture for channel decoding,” IEEE Journal of Selected Topics in Signal Processing, vol. 12, pp. 144–159, Feb 2018.
[16] D. Xiong and B. Tian, “Deep learning method of polar codes under colored noise,” in 2019 IEEE/CIC International Conference on Communications in China (ICCC), pp. 677–682, Aug 2019.
[17] W. Xu, X. You, C. Zhang, and Y. Beery, “Polar decoding on sparse graphs with deep learning,” in 2018 52nd Asilomar Conference on Signals, Systems, and Computers, pp. 599–603, Oct 2018.
[18] H. Kim, Y. Jiang, S. Kannan, S. Oh, and P. Viswanath, “Deepcode: Feedback codes via deep learning,” CoRR, vol. abs/1807.00801, 2018.
[19] Y. Jiang, H. Kim, H. Asnani, S. Kannan, S. Oh, and P. Viswanath, “Learn codes: Inventing low-latency codes via recurrent neural networks,” in ICC 2019 - 2019 IEEE International Conference on Communications (ICC), pp. 1–7, May 2019.
[20] N. Farsad, M. Rao, and A. Goldsmith, “Deep learning for joint source-channel cod- ing of text,” in 2018 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pp. 2326–2330, April 2018.
[21] Y. M. Saidutta, A. Abdi, and F. Fekri, “Joint source-channel coding of gaussian sources over awgn channels via manifold variational autoencoders,” in 2019 57th Annual Aller- ton Conference on Communication, Control, and Computing (Allerton), pp. 514–520, Sep. 2019.
[22] D. B. Kurka and D. Gndz, “Deepjscc-f: Deep joint-source channel coding of images with feedback,” 2019.
[23] K. Choi, K. Tatwawadi, A. Grover, T. Weissman, and S. Ermon, “Neural joint source- channel coding,” 2018.
[24] E. Bourtsoulatze, D. Burth Kurka, and D. Gndz, “Deep joint source-channel coding for wireless image transmission,” IEEE Transactions on Cognitive Communications and Networking, vol. 5, pp. 567–579, Sep. 2019.
[25] J. Ball ́e, V. Laparra, and E. P. Simoncelli, “End-to-end optimized image compression,” in Int’l. Conf. on Learning Representations (ICLR2017), (Toulon, France), April 2017. Available at http://arxiv.org/abs/1611.01704.
[26] S. Bhatnagar, D. Ghosal, and M. H. Kolekar, “Classification of fashion article images using convolutional neural networks,” in 2017 Fourth International Conference on Image Information Processing (ICIIP), pp. 1–6, Dec 2017.
[27] X. Glorot and Y. Bengio, “Understanding the difficulty of training deep feedforward neural networks,” in Proceedings of the Thirteenth International Conference on Artificial Intelligence and Statistics (Y. W. Teh and M. Titterington, eds.), vol. 9 of Proceedings of Machine Learning Research, (Chia Laguna Resort, Sardinia, Italy), pp. 249–256, PMLR, 13–15 May 2010.
[28] H. Xiao, K. Rasul, and R. Vollgraf, “Fashion-mnist: a novel image dataset for bench- marking machine learning algorithms,” CoRR, vol. abs/1708.07747, 2017.
[29] D. P. Kingma and J. Ba, “Adam: A method for stochastic optimization,” 2014.