Signals Recognition by CNN Based on Attention Mechanism
Round 1
Reviewer 1 Report
The authors have presented a signals recognition approach based on CNN deep learning method. the main problem of the paper is the lack of contribution, which should be clarified and emphasized in the manuscript. The following modifications should be considered in the manuscript:
- Kindly clarify and emphasize the contributions and novelties of the proposed method in the abstract, and introduction section.
- The equations, which are not extracted by the authors should be cited in the manuscript.
- All of the figures should be mentioned and explained in the text. for example, all of subfigures in Fig.9 are not explained in the text.
- The similarity percentage of the manuscript should be decreased. Some parts are copied from the other papers.
- The reported accuracy numbers are related to training or test process?
- Training, test, and validation accuracy should be reported. for example, plots of training, test, and validation accuracy versus iterations can be added in the manuscript.
- How did the signal parameters are considered? For example, the frequency (fs) is 200 kHz, the center frequencies of digital and analog modulation types are 902 MHz and 100 MHz, and the signal sequence length is L=128. The effects of these parameters are not considered in the manuscript. For instance, the performance of the proposed method under different signal lengths can be added.
- In “Figure 8. Average recognition accuracy under different SNRs.” The proposed method should be indicated. If the proposed method is ResNet+CBAM it should be indicted in the figure legend and also explained in the text.
- There are some methods using CNN for modulation classification in wireless communications with better accuracy. For example, the proposed method in [R1] has achieved the accuracy of 0.9996. A comparison table should be added to compare the obtained accuracy for the proposed method and the state-of-the-art algorithms.
[R1] Zheng S, Qi P, Chen S, Yang X. Fusion methods for CNN-based automatic modulation classification. IEEE Access. 2019 May 22;7:66496-504.
- There are several typos and grammar errors in the manuscript, which should be corrected carefully.
Author Response
Please see the attachment.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
In this submission authors proposed the use of different deep learning topologies for modulation recognition for signals in complex electromagnetic environments. They propose a method of modulation recognition using ResNext topology and introduce somekind of modified attention layers.
The submission lacks of any comparison of similar approaches proposed in the recent literature. Many of the references used are prior to 2000 and are referred to widely known techniques. The results are based on signals created in the lab and not on real ones (or even on benchmarking data).
The whole paper seems as a first thesis approach to the application of DL topologies slightly modified to signal modulation without any sound scientific advantages.
In my opinion the whole approach should be considered and resubmitted with a different structure, possibly starting from the specific advantages compared with recent approaches.
Author Response
Please see my response in the attachment.
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
All of my comments are considered in the manuscript and the manuscript can be accepted after minor modifications.
- Kindly correct typos in the manuscript. For instance, write “SNR” instead of “snr” in Figure 11.
Author Response
Please see the attachment
Author Response File: Author Response.pdf
Reviewer 2 Report
The authors have answered to the majority of the points I've noted and I think that the paper can be accepted with minor corrections mainly editing and typos in the text.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
