A High-Security Probabilistic Constellation Shaping Transmission Scheme Based on Recurrent Neural Networks
, Yaya Mao 1,2,3, Xiangyu Wu 1,2,3, Zeqian Guo 1,2,3, Xu Zhu 1,2,3, Zhongwen Ding 1,2,3, Mengjie Wu 1,2,3, Feng Wang 1,2,3, Rahat Ullah 1,2,3, Yongfeng Wu 1,2,3, Lilong Zhao 1,2,3 and Ying Li 1,2,3
Round 1
Reviewer 1 Report
The authors claim that they proposed and demonstrated a high-security probabilistic constellation-shaped QAM scheme based on a 4D biplane fractional-order chaos (BFC) system. The scheme utilizes BFC systems which are more secure than integer-order chaotic systems. Moreover, the proposed method applies a recurrent neural network (RNN) to constellation point probability distribution training to obtain high noise tolerance. The proposed method can provide a nice idea to increase secure and reliable communication. However, the following points need to be revised to make this manuscript suitable for publication.
1. Due to the application of RNN, people are concerned about overfitting. Please add detailed information on how to prevent overfitting during RNN training, such as regularizing loss functions.
2. In the conclusion, the author claims that the proposed scheme can achieve a net bit rate of 48.82Gb/s in a seven core fiber optic transmission system, and the author should explain how 6.97Gb/s is calculated. According to equation (5), the author should add a description of the calculation parameters, such as AWG_ Sampling_ Rate, IFFT_ Bin_ Length, GI, and GIP.
3. What is the difference between PCS constellation obtained through automatic encoder and using MB distribution? From Figures 3 (b) and (c), the author analyzes that the proposed scheme has lost a portion of GMI performance and achieved an improvement in transmission rate. The author should clarify how to measure the gains and losses between the two and highlight the novelty of the proposed optimization scheme.
4. Some important descriptions should be added to make it understood by non-experts in this field. For example, how to derive the key space from Fig. 10.
5. There are some spelling/grammar errors in the text, such as the upsampling frequency in Table 1. And the x label in Figure 4 (a) should be added. Please check carefully.
Author Response
Please see the attachment.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
In this paper, the authors propose the use of probabilistic constellation shaping together with a "biplane fractional-order chaotic system" for high security transmission. The proposed method can provide a good idea for secure communication.
1. The author proposes that fractional order chaotic systems have higher randomness and security compared to integer order chaotic systems, but there is a lack of comparison in the article.
2. As author demonstrated, “we tested the BER performance in the case of eavesdropping or brute force decryption by an illegal ONU at the receiving end without the correct private key at the receiving end performance”. A reasonable result is that data could not be recovered owing to the incorrect private key. It is direct. So, what does Security mean by?
3. What are the advantages of using recurrent neural networks? Can other neural networks be used instead?
4. Suggest the author to pay attention to the layout of the formula.
5. There are some grammar errors in the article, and it is recommended that the author carefully check them.
There are some grammar errors in the article, and it is recommended that the author carefully check them.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
