Next Article in Journal
Deterministic and Stochastic Nonlinear Model for Transmission Dynamics of COVID-19 with Vaccinations Following Bayesian-Type Procedure
Previous Article in Journal
Mathematical Logic Model for Analysing the Controllability of Mining Equipment
 
 
Article
Peer-Review Record

Convolutional Neural Networks for Local Component Number Estimation from Time–Frequency Distributions of Multicomponent Nonstationary Signals

Mathematics 2024, 12(11), 1661; https://doi.org/10.3390/math12111661
by Vedran Jurdana * and Sandi Baressi Å egota
Reviewer 1:
Reviewer 2: Anonymous
Mathematics 2024, 12(11), 1661; https://doi.org/10.3390/math12111661
Submission received: 29 April 2024 / Revised: 19 May 2024 / Accepted: 24 May 2024 / Published: 26 May 2024
(This article belongs to the Section Mathematics and Computer Science)

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

Dear authors,

The manuscript shows an acceptable level of novelty with promising results. However, for the sake of completeness, you ought to resolve the following issues or shortcomings with the research work:

1-       In the abstract, you need to mention specific limitations in the previously developed methods that you tackled in the present study and how the characteristics of the proposed method have resolved them.

2-       I think the introduction section needs to be embellished with a table representing the previous relevant works with their pros and cons. Furthermore, you have to enumerate the contributions of the present study at the end of this section.

3-       In the materials and methods section, you have to use multiple tables to summarize sections 2.1 and 2.3. The associated tables for these sections should be comprehensive and represent the works completely.

4-       In the results and discussion section, one expects to see a comprehensive comparison with other extensions of the lRE. Besides this important point, you have to compare it with the other methods developed and applied to real-world datasets.

5-       I think you could have trained an end-to-end CNN or any other deep network without taking a TF transform.

6-        you have not applied wavelet transforms on the signals instead of the ones used in this work, why?

7-       I think you need to have a computational complexity analysis of the proposed method and the other relevant methods.

8-       I wonder if one must do normalization before providing input signals to the TF transforms.

9-       You must have a block diagram to provide a big picture of the proposed method with great details.

10-  Please provide a table to represent the real-world datasets used in this work.

11-  The limitations of the present study should be thoroughly elaborated and put in a subsection before the conclusion. Now, you have a shallow view of this important topic.

12-  It makes the following sentence a bit vague by using an inappropriate verb: … enhances the sensitivity …. You enhanced insensitivity nor sensitivity if you asked me.

13- you would have the abbreviation table on the first page of the manuscript.

 

 

Yours faithfully,

Comments on the Quality of English Language

please look at my comment #12.

 

 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

The authors of this paper propose a convolutional neural network-based approach to address various limitations in determining the number of components in frequency-modulated signals using localized Rényi entropy. They constructed a dataset for training and conducted various comparative experiments between the proposed CNN method and the traditional LRE method. The experimental results indicate that their method performs positively in challenging synthetic signals and real-world signals.

In the paper, sufficient experimental data and charts can be seen, and there is a good analysis of the experimental results. The authors introduced some real-world signals to validate the effectiveness of their CNN method, which is very good. I think this paper has a high level of completion and can be accepted with just a few modifications. The following are the recommended parts for revision:

1.Some of the references cited in the paper are too outdated, and authors should pay attention to this issue.

2.On page 9, line 266, "All of the networks are trained with the batch size of 8, 16, and 32, for 5, 10, 25, and 50 epochs." Does this mean that the authors used different batch sizes and training epochs for different models, and what is the basis for adopting this training strategy?

3.On page 19, Table 2, the precision of the data after the decimal point is inconsistent, perhaps due to negligence, and the precision should be unified.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

Thanks for the revisions made to the manuscript.

Back to TopTop