Predictive Wireless Channel Modeling of MmWave Bands Using Machine Learning

Round 1

Reviewer 1 Report

A data-driven method for estimating/predicting channel parameters is proposed. The manuscript needs major revisions/improvements before it can be considered for publication. The following are a few comments which may help in improving the manuscript,

• The abstract focuses on the general attributes rather than the key points of the proposed work/contributions. The key novel contributions of the proposed work should be explicitly enlisted in the last paragraph of the Introduction section.
• There exist various notable studies in the literature which discuss the potential of ML for channel modeling/characterization which are not discussed in the manuscript, see e.g., [A].
• High pathloss and dominant “reflections” phenomena in mmWave bands lead to sparsely structured MPC in delay, angular, and Doppler domains. The channel impulse response (CIR) discussed in eq. (12)-(14) does not consider/discuss the sparse structure of the channel vector in the delay domain. Also, when defining mmWave channels, it is important to discuss angular domain, as such channels are highly directional in nature.
• The manuscript contains several typos, grammatical mistakes, and presentation problems, e.g.,
  1. Recheck the word “Multilelayers” at line 12 of page 1.
  2. In lines 251-252 on page 9, it is stated that “seen in the below figure”, while there is no figure below this text.
  3. The source of data/samples plotted in Fig. 4 is unclear. Is the plotted pathloss based on some measurement campaign on ML-based prediction based on analytical expressions?
  4. Letter W in the word “Where” used after multiple equations (e.g., eq. 31) should be a small letter, etc.
• The presentation of mathematical expressions needs improvements, e.g.,
  1. In eq. (31), the equating operator = is missing.
  2. the scaling operations in the logarithmic scale can be transformed into shifting operations in eq. (7).
  3. In eq. (8), the asterisk symbol is used to represent the multiplication operation while in other equations the multiplication operation is represented with no operator. In eq. (12), the asterisk symbol is used to represent convolution.
  4. Size of brackets should match the size of input argument/text, see e.g., eq. (20), etc. 
  5. In eq (24), a symbol may be used to represent error instead of using the text. 
• MSE is defined in eq. (31), while no results for MSE are provided. Is the Loss presented in Fig. 7 and 8 related to MSE? 
• Considering the highly directional nature and high pathloss of mmWave channels, it is important to highlight the potential future applications of such mmWave channels, see e.g., as discussed in [B].

[A] R. He, B. Ai, A. F. Molisch, G. L. Stüber, Q. Li, Z. Zhong, and J. Yu "Clustering Enabled Wireless Channel Modeling Using Big Data Algorithms," in IEEE Communications Magazine, vol. 56, no. 5, pp. 177-183, May 2018, doi: 10.1109/MCOM.2018.1700701.

[B] F. Jameel, S. Wyne, S. J. Nawaz, and Z. Chang, "Propagation Channels for mmWave Vehicular Communications: State-of-the-art and Future Research Directions," in IEEE Wireless Communications, vol. 26, no. 1, pp. 144-150, February 2019, doi: 10.1109/MWC.2018.1800174.

Author Response

Dear Reviewer,

Upon the request by Electronics of entitled “Predictive Wireless Channel Modeling of MmWave Bands Using Machine Learning”, authors, would like to appreciate your comments. Moreover, replies to the comments is attached and the manuscript has been modified as well.

Author Response File: Author Response.pdf

Reviewer 2 Report

Generally, this manuscript looks like a haste work with many things tried to deliver but lack of the focus and good organization. 

The very first concern lies in the incompetence of the academic/technical writing, many expressions and sentences are not correct or informal. Background introduction is kind of causal, with many big concepts mentioned, but not fit them into the proper logic flows and organization. For example, how would quantum computing be used for the future wireless communication and how it is related to the investigation scope of this paper? Either you describe it in a very comprehensive way or better not to mention it using only one or two sentences. 

The second concern lies in the background of using ML for predictive PL models. More wordings and reviews are needed to express the importance of this technique and how other researcher have been doing and using it to improve the performance of wireless networks. Also, using ML is not free, it is costly and time-consuming, and why should you use it instead of traditional wireless piloting method to do the real-time channel estimation or even do the channel sounding?

 

Third, the real-life environment, like dense urban, can be really complicated and different, city by city, street by street, building by building can the tolerance be good enough to deploy such ML predictive models?  

 

 

 

Author Response

Dear Reviewer,

Upon the request by Electronics of entitled “Predictive Wireless Channel Modeling of MmWave Bands Using Machine Learning”, authors, would like to appreciate your comments. Moreover, replies to the comments is attached and the manuscript has been modified as well.

Again, authors appreciate your comments and time.

Regards,

 

Author Response File: Author Response.pdf

Reviewer 3 Report

Major concerns:

Thanks for working on "Predictive Wireless Channel Modeling of MmWave Bands Using Machine Learning". after reading the whole manuscript I did not see the improvement results in comparison to the current state of the arts. Why such a complex technique is implemented as we have many easy approaches to go. One contribution of this work that needed to be considered as the enhancing the accuracy of supervised data learning, through the random forests by combining an unsupervised algorithm to the learning.  

Equation (1) to (8) are very well known equations and authors need to reference them at least from basic antenna book, like stutzman and balanis. Only writing final equation and referencing properly will work. every wireless and antenna engineer knows it.

There is nothing new in ML selection. Model is trained using deep learning technique which is very common and output predict the path loss. what about a very complex scenario. there will be high delay and this model will no more be applicable. In fact authors have only studied simulations that is the reason. In my opinion, this is quite far from practiacility.

Channel impulse response section: Nothing new and only gaussian noise is considered. what if we have considerable ringing and its derivative of gaussian? Eq. (12) to (20) again taken from literature. Authors must have their own theoretical developments atleast.

considerable work needed to modify the direction and made the research in a new direction in comparison to current state of the art.

Other minor concerns: There is a lot of typos and grammatical mistakes as well. Line 25, extra 0. ORCID ID should not be mentioned in the way performed. Figure 1 is going out of the page. so on author need to check for spelling and structuring mistakes.

 

Author Response

Dear Reviewer,

Upon the request by Electronics of entitled “Predictive Wireless Channel Modeling of MmWave Bands Using Machine Learning”, authors, would like to appreciate your comments. Moreover, replies to the comments is attached and the manuscript has been modified as well.

Again, authors appreciate your comments and time.

Regards,

 

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

The authors have addressed all the concerns of the first review round. The manuscript may be accepted for publication. 

Author Response

Thank you for accepting our paper.

Reviewer 2 Report

Good improvement has been observed but still with some pending details missed out: 

1. Which machine learning framework and platform has been used to conduct this research?
2.  On-line, e.g. cloud, or off-line (GPU based) has been used for data training and generating inference models? Could you list the resources consumed and time spent on that? some reference for comparison is provided here: https://arxiv.org/pdf/2110.01848.pdf

Author Response

Dear Reviewer,

Authors, would like to appreciate your comments and your support of the previous version. Moreover, a reply to the comments is below and the manuscript has been modified as well.

Regards,

Author Response File: Author Response.pdf

Reviewer 3 Report

Some comments need to be considered very carefully. I am not satisfied with these two.

Thanks for working on "Predictive Wireless Channel Modeling of MmWave Bands Using Machine Learning". after reading the whole manuscript I did not see the improvement results in comparison to the current state of the arts. Why such a complex technique is implemented as we have many easy approaches to go. One contribution of this work that needed to be considered as the enhancing the accuracy of supervised data learning, through the random forests by combining an unsupervised algorithm to the learning.

 

There is nothing new in ML selection. Model is trained using deep learning
technique which is very common and output predict the path loss. what about a very complex scenario. there will be high delay and this model will no more be applicable. In fact authors have only studied simulations that is the reason. In my opinion, this is quite far from practiacility.

Author Response

Dear Reviewer,

Authors, would like to appreciate your comments and your support of the previous version. Moreover, a reply to the comments is below and the manuscript has been modified as well.

Regards,

Author Response File: Author Response.pdf

Round 3

Reviewer 2 Report

Thanks for the revision, I have no further comment.

Reviewer 3 Report

Thank your for your revision

Well, i am not fully convinced by the author's response anyhow it's time to finish the review process.

Thanks