Voting-Based Deep Convolutional Neural Networks (VB-DCNNs) for M-QAM and M-PSK Signals Classification

Round 1

Reviewer 1 Report

The paper is about the application of voting-based deep convolutional neural networks (VB-DCNN) applied to the detection of the modulation format of M-QAM and M-PSK signals that carry a degree of noise. The authors demonstrate that their approach requires little to no manual preparation of the data prior to the application of the neural network and that their approach is capable of detecting a broad class of modulation types. It performs equally well or better than four other approaches taken from the literature.

The paper is well written, structured appropriately, introduction and references are adequate, the description of the method goes into sufficient detail.

Figures and tables are well prepared. 

There is the occasionally grammar or spelling oversight and in lines 176-177 the authors refer to figures while missing the figure numbers. 

After minor corrections I consider the paper to be ready for publication.

There is the occasionally grammar or spelling oversight and in lines 176-177 the authors refer to figures while missing the figure numbers. 

After minor corrections I consider the paper to be ready for publication.

Author Response

The response to the reviewer's comment is attached. 

Author Response File: Author Response.pdf

Reviewer 2 Report

The authors propose an approach capable of classifying modulated signals using a voting-based deep convolutional neural network in the context of automatic modulation classification. The idea is very interesting and the paper well writes, with some minor errors found and listed in sequence. The comments listed as follows are intended to improve paper quality and also readers' understanding.

 

Please provide more detail regarding the state of the art algorhtims compared. Why do you believe the authors achieved a superior performance compared to them? What are the main differences that may be responsible for the improvement in performance? The authors claim that the classification accuracy increases as the number of layers increases in the architecture. Would this be the case with the approaches compared? Does it mean that previous approaches have a lower number of layers?

 

 "It requires a significant amount of time for the training and careful selection of the parameters.":

 - time is not significant (some minutes, as reported by the paper) (you only need to run once in order to train the network, correct?)

 - isn't the parameter selection automatic?

 

 Please provide more future work.

 

 

More general comments and minor errors are listed as follows.

 

" [6]-[7]." -> " [6, 7]."

" in software-defined" -> " in a software-defined"

"approaches:-" -> "approaches:"

" Section 3, describes" -> " Section 3 describes"

" [11]-[12]." -> " [11, 12]."

"in [17], presented" -> "s in [17] presented"

" [24]-[25]" -> " [24, 25]"

"AWGN" -> please define the term at its first occurence

"expressed as:-" -> "expressed as:"

"follows:-" -> "follows:"

"where, m" -> "where m"

"modulation formats as shown in figure 2" -> "modulation formats as shown in figure 1"

"depicted in figure 2." -> "depicted in figure 1."

Algorithm 1 should be located after its reference in the text.

"The class with the most votes of votes " -> please rewrite

"[1,2,3,...,M], From" -> "[1,2,3,...,M]. From"

"The simulation parameters are shown in Table 4." -> "The simulation parameters are shown in Table 3."

"10000" -> "10,000"

"2500" -> "2,500"

"are demonstrated" -> "is demonstrated"

"shown in figure ??" -> please fix this

"from figure ??" -> please fix this

", 97.3% & 90.2%," -> ", 97.3% and 90.2%,"

"5 & 10 dB." -> "5 and 10 dB."

"76% & 80%," -> "76% and 80%,"

Author Response

The response to the reviewer's comments is attached. 

Author Response File: Author Response.pdf

Reviewer 3 Report

This paper proposes a voting-based deep convolutional neural network (VB-DCNN) to automatically classify the received modulated signals. The DCNN is designed using MATLAB Toolbox where the simulated experimental data are used to analyse the proposed network. Below are my comments and questions to be addressed before this manuscript can be published:

1. L.1: The authors should enhance the Abstract by clearly highlighting the novel aspects of their study. At present, I am unable to identify any statement that effectively conveys the value of this paper and why it should be considered for publication

2. L.18: The current discussion on the AMC topic is insufficiently detailed and requires further expansion to fully address the relevant concepts and considerations.

3. L.98: How do you take into account the influence of the input signal amplitude on the received signal? 

4. L.100: What is the justification for the signal length of 1024x1

5. L.108: Why the yi(m) is set to 25 layers?

6. L.151: It would be interesting to see the evaluation of the VB-DCNN for QAM modulated signal of 64QAM as this can be quite critical for the receiver signal accuracy

7. L.189: Have you tried the SNR at 25 dB to see the effect on the accuracy?

8. L.196: Where in the paper we can see the relationship between the training and testing accuracy?

9. L.212: One of the known challenges of VB-DCNN is the increased computational complexity, as multiple CNNs have to be trained and combined. I cannot see a demonstration or results on how you address these challenges in the paper

10. L.227: Need to further elaborate on the key findings on the VB-DCNN classification and efficiency. 

 

Final proofreading required

Author Response

The response to the reviewer's comments is attached. 

Author Response File: Author Response.pdf

Round 2

Reviewer 3 Report

Thank you for addressing my comments and suggestions. I can now recommend the paper to be published in Electronics.