Next Article in Journal
Corrections of Mesoscale Eddies and Kuroshio Extension Surface Velocities Derived from Satellite Altimeters
Previous Article in Journal
A Multiscale Spatiotemporal Fusion Network Based on an Attention Mechanism
 
 
Article
Peer-Review Record

CAE-CNN-Based DOA Estimation Method for Low-Elevation-Angle Target

Remote Sens. 2023, 15(1), 185; https://doi.org/10.3390/rs15010185
by Fangzheng Zhao 1,*, Guoping Hu 2, Hao Zhou 2 and Chenghong Zhan 1
Reviewer 1: Anonymous
Reviewer 2:
Remote Sens. 2023, 15(1), 185; https://doi.org/10.3390/rs15010185
Submission received: 22 November 2022 / Revised: 22 December 2022 / Accepted: 23 December 2022 / Published: 29 December 2022

Round 1

Reviewer 1 Report (Previous Reviewer 1)

I'm not questioning the originality/novelty of the proposed method. Indeed, it presents a new methodology of DOA estimation in the presence of the multipath effect. However, it seems that the proposed method only suits the situation where only one pair of targets (the target itself and its multipath) exists. It would be glad to see such work at a conference instead of a well-known journal. 

In addition, the training set seems to contain so many samples. These samples nearly cover all the solution space, which makes the solving process quite brutal. And utilizing the CNNs of multi-branches to estimate the angle is also brutal and inefficient. And I think this framework needs further to be studied, analyzed, or discussed, including aspects like the effect of the number of branches; the training process of multiple networks with different tasks. 

Compared with other deep-learning-based methods, the proposed method also needs to discretize the solution space, making it also suffer related problems. I can't find how the method handles it. Also, the cited references are not for DOA estimation with the existence of the multipath. Authors should review and compare the proposed method with more relevant research.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 2 Report (New Reviewer)

no further comments

Author Response

Thank you very much for your recommendation!

Reviewer 3 Report (New Reviewer)

- Acronyms appear in the text without any explanation about them. For example, in the title and abstract.

- The text has several highlighted passages as if it were a revised version.

- On page 4 the authors misspelled L'Hopital's law.

- The acronym CNN has not been explained anywhere in the text.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report (Previous Reviewer 1)

The confusion about novelties and contributions of the work has been resolved through the reply. The insights about the network design related to the second point are valuable for readers to comprehend the work. It's suggested not to be presented only in the reply but added in the manuscript, for example, in the discussion part.

 

Author Response

Please see the attachment.

Author Response File: Author Response.docx

This manuscript is a resubmission of an earlier submission. The following is a list of the peer review reports and author responses from that submission.


Round 1

Reviewer 1 Report

Comments can be found in the attachment.

Comments for author File: Comments.pdf

Reviewer 2 Report

The manuscript is about build a hybird CAE and ELM model for DOA problem. The novelty is not new and the framework of the design is kind of confusion and somehow not practical in either synthetic data generation or deep learning training. The details are listed as followsL

* line 155-159 on page 5 is very confusing. If the Gamma_d and Gammd_i do not exist in the received signal, you must come up with a way to either estimate it or calculate it. So in that sense, how to get the input R for neural network for real simulation. You cannot just directly say I simulate the R and get the input for CAE.

* Figure 6 is definitely is low-efficient and weird way to do, in order to calculate that, depends on how many arrays to own, you need to create the same amount of  local CNN to calculate the one angle.

* How can we make sure that the weights learned during the CAE is been co-trained or propagated to the bottom MLP? Is there a joint optimization loss function?

* Why is there a suddenly jump in Figure 12 for CAE-MUSIC and CAE-ESPRIT?

 

 

The authors are required to polish there grammar as well as correct their typos, some example like page 4 line 123, it is not L’hopita's law but L'Hospital's law

Back to TopTop