Robustness of Deep Learning-Based Specific Emitter Identification under Adversarial Attacks
Round 1
Reviewer 1 Report
The Paper needs the following revisions and is subject for re-review, and after re-review the final decision for the paper will be done:
1. Add in the last lines of abstract in what %age and in what parameters the proposed methodology is better and what is the overall analysis observed at time of experimentation.
2. Add more information with regard to the scope and problem definition in Introduction Section.
3. Literature review is missing. Add min 10-15 papers which are taken as base for the draft of the proposed methodology.
4. Add more details to the proposed methodology in terms of Steps fo working, Algorithm and Flowchart.
5. Add Analysis section to the paper.
6. Addition of future scope is suggested to the paper.
Author Response
Please see the attachment
Reviewer 2 Report
Vulnerability of DL-based SEI is investigated from an adversarial perspective in this paper. Based on three adversarial example generation methods, the performance loss under adversarial attacks is systematically studied on real-world data and simulated data.
To improve the quality of the paper, a few suggestions are as follows.
1. The novelty of this paper is not obvious since the core work, DL-based SEI model already existed. Please try to give prominence to your own work.
2. The necessary of the work is missing and it should be detailed in the abstract and introduction parts.
3. The literature review presented here is highly insufficient and generalized. Please improve it using recent papers.
4. All the variables should be coincident. For example, is Am(f) in eqn(1) the same as A(f) in (2)? Moreover, what is bm,l in line 154? Please pay more attention to this kind of issue.
5. Few variables are not defined. Please correct it.
6. Few short forms have been used without giving full forms. Please cross-check throughout the paper properly.
7. Reorganize the total paper. It would be better to take part 1,2,and 3 into one part since it seems that this three parts are just the collections of existing work.
8. Elaborate discussions of results. Try to compare your methods with the state-of-art methods.
9. Describe the engineering applications part in more detail if you mentioned this area in your paper.
10. Some sentences should be rewritten. They include line165~170.
Author Response
Please see the attachment.
Reviewer 3 Report
Please elaborate on your specific emitter solution. What is it exactly? You've supplied general descriptions for assaults, signals, and so on. Figure 4 appears to be a redrawn version of a generic Resnet. How is the adversarial context understood here? Maybe provide a sample? Include both pseudo code and a UML-based activity diagram illustrating your approach as well as the network's entire mathematical apparatus (generic current formulae does not differ much from book like materials). Add hyper-parameter table.
Please specify if the experiments are an overview of known themes or something of your own creation. The experiments could be adequately contextualized (the reader should not be left to assume that they will get their own conclusions). The experiments should be reported in greater detail (e.g. set up and carry out process, results in raw format, etc.). Include a statistical study of dependability. Add computational performance in FLOPS. Add model size.
Finally, I fail to find the IoT context in the article. A dedicated section must be added, considering the promise in the abstract. What devices were analyzed, what networks? Lorra, GSM?
Author Response
Please see the attachment.
Author Response File: 
Author Response.pdf
Reviewer 4 Report
In this paper, the authors present a method to protect radio frequency fingerprinting of electronic devices when adversarial attacks are taking place. Although there are a few typos, the paper is well written and the structure is correct.
In my opinion, from the scientific point of view the paper is very good. I’m not an transmission-equations expert, but the problem statement, the experiments and the results are all clear. The paper has enough references and they are more or less recent.
I think that the paper may be improved by:
· Increasing context information in the introduction: clearly define the risks to mitigate with the presented technique (from a practical perspective) and provide a definition of an adversarial attack.
· Trying to avoid speaking about IoT. The paper is not an IoT paper but describes a technique that may be applied in a number of scenarios. IoT is briefly mentioned in the abstract and the introduction, but then the contents are not about IoT. In fact, the main dataset used in the experiments is radar data from aeroplanes. It looks like the first 4 lines have been added for convenience or just to have a couple of self citations. This is not needed.
· Checking the alignment of tables.
· Trying to increase the quality of confusion matrices and low-dimensional feature distribution graphs. The numbers in those graphs cannot be seen.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
The Revised Paper stands Accepted and now no further revisions are requirted.
Author Response
Thank you very much for your approval. Thanks again for your time and effort expended to improve our paper.
Reviewer 2 Report
The paper has been revised according to my comments. It can be accepted.
Author Response
Thank you very much for your approval. Thanks again for your time and effort expended to improve our paper.
Reviewer 3 Report
Thank you for corrections, however article still misses a clarification on the network, please highlight the modification and add pseudo code as was requested, proving that the approach is indeed novel and not just an application. Further, I would recommend enhancing statistical analysis, as the results are hard to tell how significant they are, also reliability is questionable, as well as the crossvalidation process.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Round 3
Reviewer 3 Report
Thank you, the paper can now be accepted for publications. I would still recommend explaining how crossvalidation was done (if at all). Article would also benefit from some "extremity" analysis (what happened in the "stand out" parts of the results).
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
