Supervised Learning Spectrum Sensing Method via Geometric Power Feature
Round 1
Reviewer 1 Report
The manuscript is far from publishable and I would suggest the authors add substantially more data and discussions before they resubmit it to Electronics or another journal for review. My main concerns are listed below:
(1) The manuscript was written based on limited data. The training data set is very small (10,000 data points). As the authors claimed, the training data set was derived by superimposing artificial Gaussian noises to simulated PU data from an experimental device. Why not generate more data for model training and validation? As a result of the small training data set, the acquired simulation results are quite limited, with literally no discussions -- the whole Results and Discussion section is less than 300 words.
(2) The experimental design is flawed. It is essential to compare the proposed ML methods against a basal method (or methods) so that the pros and cons of the newly-proposed methods can be better demonstrated. No uncertainty or sensitivity analysis (e.g., the impact of noise levels on SVM and KNN methods) was conducted. No cross-validation or validation (with additional data sets) was performed, either.
The manuscript is overall readable. However, there are still many typos and grammatical errors. I would suggest the authors pay attention to verb tenses. My specific comments can be found in the attached pdf file.
Comments for author File: 
Comments.pdf
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 2 Report
For physical reasons, what can be said when comparing power, energy, and differential entropy at a low signal-to-noise ratio? In comparison with energy, the use of power means another averaging over a time interval. But with this averaging, the signal-to-noise ratio improves. Therefore, it is quite an expected result. I recommend publishing the article in the presented form with possible correction of small details.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 3 Report
The manuscript by Hu et al. is devoted to developing methods of using machine learning to improve the spectrum sensing performance under low signal noise ratio. Potentially, it can be interesting; however, I suppose that this work requires at least two important points. (1) Aims of the work should be clearly described. Place of these aims in the modern investigation in this field should be clarified. (2) Results should be discussed; their novelty and importance should be described in detail. Now, this work seems to be too intricate for understanding and, probably, will be difficult for potential readers.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
The authors' revisions substantially improves the manuscript's readability and quality. But I don't agree with the authors on their excuse for the absence of cross validation. Also, a thorough check of the format and language is needed before acceptance for publication.
Author Response
Dear reviewer, I am sorry that my answer did not satisfy you, due to my misunderstanding, I was unable to answer on point. Regarding cross validation, the data used in the experiment includes 1024 datasets, each containing 10000 points. During the simulation experiment, the datasets were divided, including 820 datasets for training and 204 datasets for testing. I have also made changes in the corresponding places in the article. Thank you for your reminder.
Reviewer 3 Report
Authors considered my comments. I have not additional remarks or questions.
Author Response
Thank you for your suggestion.
