Fast Adaptive Beamforming for Weather Observations with Convolutional Neural Networks

Round 1

Reviewer 1 Report

This is a well-written manuscript that only needs to undergo a few minor changes 

1-Please focus the abstract on your study and your results. In particular the last two sentence are vague. I would prefer to see some data  from this study in the abstract, rather than a description of “where to go next”. More generally, I suggest to focus the manuscript on the scientific results

2-Both the motivations and contributions of this paper are not referred in this paper, which need to be justified

3-To improve the readability of the paper, I suggest dividing the analysis into several subsections

4- It is highly recommended to provide a Table to evaluate your research contributions by comparing with existing reviews or surveys

 

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 2 Report

The paper presents a digital beam forming (DBF) approach using a convolutional neural network (CNN). It addresses the computational challenge of DBF when the number of receivers/transmitters become large. Their adaptive beam forming using CNN (ABCNN) claims to be computationally more efficient than existing approaches. The work is interesting and discusses the important research area of DBF in phased array antennas. Some minor issues should be addressed before the paper can be accepted for publication. My detailed comments are listed below:

1. The manuscript does not provide a satisfactory literature review of existing DBF techniques. For example, evolutionary stochastic optimization algorithms have been widely used for beam forming and engineering antenna radiation patterns (including sidelobe level set and null position engineering). Algorithms like the genetic algorithm and particle swarm optimization have been a corner stone of antenna array DBF. The reference section of the paper is unfortunately missing papers related to evolutionary optimization-based approaches. Some citations of such papers should be included. Some examples of such papers include:

A) Mahmoud, Korany R., et al. "Analysis of uniform circular arrays for adaptive beamforming applications using particle swarm optimization algorithm." International Journal of RF and Microwave Computer‐Aided Engineering: Co‐sponsored by the Center for Advanced Manufacturing and Packaging of Microwave, Optical, and Digital Electronics (CAMPmode) at the University of Colorado at Boulder 18.1 (2008): 42-52.

B) Lu, Yilong, and Beng-Kiong Yeo. "Adaptive wide null steering for digital beamforming array with the complex coded genetic algorithm." Proceedings 2000 IEEE International Conference on Phased Array Systems and Technology (Cat. No. 00TH8510). IEEE, 2000.

C) Zaman, Mohammad Asif, and Md Abdul Matin. "Nonuniformly spaced linear antenna array design using firefly algorithm." International Journal of Microwave Science and Technology 2012 (2012).

 

2. In the equations, the vectors are represented as bold variable (as they should be). However, they are also in italic fonts. This is incorrect. The vectors should be bold but not italic. Also, the weight vector should be represented by “w”. It appears to have been represented by \omega. This can be confusing and \omega is often used to represent the angular frequency and is related to \lambda (the wavelength). Please fix the variables to address these issues.

 

3. The data of supercell storm cases and corresponding adaptive beam forming weights were taken as training data. Please comment when whether the type of the training data would affect the generality of the ABCNN engine. Could the developed ABCNN (after training) be applied to general conditions or would it be limited to the conditions of the training data?

 

 

The English is fine. 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

This manuscript investigates the application of deep learning with ABCNN on the adaptive DBF to reduce its computation time. In this framework, the experiment is well designed, and the preliminary results are interesting. Based on the results and comparison with Capon method and Fourier method, the authors claim that ABCNN can significantly reduce the computational burden of the adaptive DBF methods and mitigate the contamination from antenna sidelobes. 

 

I think the authors did a good job in setting up the experiment appropriately. The results in this manuscript should be shared in our remote sensing community to enhance our comprehension of the practical implementation of ABCNN deep learning on radar meteorology. Nonetheless, despite these strengths, there are issues with the manuscript. I recommend a minor revision before accepting it. A couple of specific concerns are outline below: 

 

1. 1. Uncertainty and errors in ABCNN application: The application using the CNN method needs a lot of work for catching uncertainty. Unfortunately, this manuscript falls short in addressing the uncertianty and errors associated with CNN, particularly in the context of its implications for the conclusions. It is imperative that these aspects are thoroughly discussed within the conclusion section to provide a comprehensive understanding of the method’s limitations and potential future directions.  

2.  

2. 2. Sample size consideration: Machine learning methodologies are notably sensitive to the size of the training dataset. Regrettably, the current manuscript employs a training set consisting of only 561 simulated IQ realizations derived from a single tornadic supercell storm case. The limited sample size could compromise the robustness of the results. It is recommended that the authors dedicate attention to elucidating the potential impact of this relatively small sample size on the final outcomes. Addressing this concern would significantly enhance the credibility of the study.

The English language usage is good and I didn't find obvious errors. 

Author Response

Please see the attachment.

Author Response File: Author Response.docx