The review of the referenced manuscript, CL2019-2001, entitled CNN-Based Automatic Modulation Classification for 5G Communications, is now complete. I regret to inform you that based on the enclosed reviews and my own reading of your manuscript, I am unable to recommend its publication in IEEE Communications Letters. I agree with the reviewers comments and the reasons for this decision are as follows: Although the proposed work is promising, however, the manuscript lacks details, clarifications, and comparisons with some latest works. In addition, the motivation for reducing the computational complexity of AMC for 5G networks is not very clear. The inputs, output and loss function used for the network are not mentioned. The information on how the dataset is prepared to simulate real-world condition is also missing.
You may revise and resubmit your manuscript to IEEE Communications Letters. When you do so, please include a cover letter that indicates the new submission is a revision of an earlier manuscript and the reference number of that prior manuscript. Also include as a supporting document a point-by-point response to the comments of the reviewers and the editor. The responses to comments file must be uploaded in PDF format in the same section in your submission as the body of your paper in ScholarOne Manuscripts, and not under the cover letter. Please also be aware that ALL submissions to IEEE Communications letters must complete the Electronic Copyright Process.
If you decide to resubmit your manuscript you should complete the resubmission through the Manuscript Central. Click on "Manuscripts with Decisions" under My Manuscripts on the left side of your Author Dashboard and then click on "Create a Resubmission" under the Actions Column next to your manuscript.
Your resubmission will be due within 75 days and is due on 05-Jan-2020. Please ensure that your revision is submitted in a timely manner as the web-based system will not allow a revision to enter the system after 75 days have elapsed. Please be aware that the time at which your revision permission will expire is 11:59 PM EST on the 75th day.
Note, that according to the IEEE COMML policy, the maximum number of permitted resubmissions after a Reject-Resubmission Allowed decision is one (1) and the maximum number of permitted Minor Revisions is (2).
Additional comments include:
The reviewers' comments are found at the end of this email.
Thank you for submitting your work to the IEEE Communications Letters.
Dr. Animesh Yadav
IEEE Communications Letters
Comments to the Author
Some interesting ideas and results have been presented. Some minor suggestions below,
• Since the authors propose the importance of time constraints in 5G applications. The problem is that the wireless devices supporting 5G will not have GPUs (one exception is the AIR-T SDR) in most cases hence the time provided may not be relevant to current 5G devices. Please elaborate on this. Can the computation be provided in terms of a metric that is independent of the processor used?
• The authors state ¡°However, most studies measured only the accuracy rate of the system but not the computing time.¡±. There do exist AMC work that discusses computation couple of examples below,
F. Meng et al., "Automatic Modulation Classification: A Deep Learning Enabled Approach," in IEEE Transactions on Vehicular Technology, vol. 67, no. 11, pp. 10760-10772, Nov. 2018.
J. Jagannath et al., "Artificial Neural Network Based Automatic Modulation Classification over a Software Defined Radio Testbed," 2018 IEEE International Conference on Communications (ICC), Kansas City, MO, 2018, pp. 1-6.
J. Jagannath, D. O¡¯Connor, N. Polosky, B. Sheaffer, L.N. Theagarajan, S. Foulke, P.K. Varshney, S.P. Reichhart, Design and evaluation of hierarchical hybrid automatic modulation classifier using software defined radios, in: Proc. of IEEE Annual Computing and Communication Workshop and Conference (CCWC), 2017. Las Vegas, NV, USA
The proposed solution seems to outperform them so it might be worth citing those work and mentioning the improvement achieved. Please make sure the comparison is fair since GPUs may not be the platform used by them.
Other points that were not entirely clear,
• Why are CNN and CLDNN from  used rather than the models used in  which are more recent?
• Similarly, why wasn¡¯t the latest data set from  used?
Comments to the Author
In this paper, the author solves a classification problem using CNN architectures. The novelty of the paper needs to be stated. Can the author include more papers that were using deep learning techniques for automatic modulation classification or 5-G communications? Is it the first paper which uses CNN to solve modulation classification problem?
Reject (Resubmission Allowed)