A Deep Learning Framework for Adaptive Beamforming in Massive MIMO Millimeter Wave 5G Multicellular Networks

Round 1

Reviewer 1 Report

Equation (1) is written straightforwardly, and the equation can have some initial steps.

In the Algorithm given in Table 1, the final output is not clearly defined.

Why is "Step 5 (updated)" next to the last step of the Algorithm?

It could be better to define Table 1 as the Algorithm rather than the Table.

How to contrast this contribution with the intelligent reflection surface? The authors should include it in the introduction section.

In the manuscript, sometimes Figure is used, and sometimes its abbreviated form Fig. is used; the authors must comply with the journal requirements.

Author Response

We would like to thank you for your time and effort to revise our manuscript, as well as for your useful comments. Please find our responses in the attached file.

Author Response File: Author Response.pdf

Reviewer 2 Report

The paper presents a deep learning framework for adaptive beamforming in 5G millimeter wave multicellular networks. Two neural networks are trained to optimize beamforming configuration based on spectral efficiency and energy efficiency. Through extensive simulations, the authors demonstrate that the proposed machine learning-based adaptive beamforming approach significantly improves energy efficiency compared to the standard non-ML framework, albeit with a slight increase in blocking probability and required radiating elements for high data rate services. The authors need to address the following comments to improve the paper's quality.

1- Provide more details about the dataset used for training the neural networks. Information on how the dataset was collected, its size, and its representativeness of real-world scenarios would strengthen the study's credibility.

2- Further clarification on the specific criteria used for MSE minimization during neural network training is essential. Understanding the trade-offs made in the optimization process would help readers comprehend the networks' decision-making process.

3- While the paper mentions a developed system-level simulator, more information on the simulator's underlying assumptions, accuracy, and validation against real-world measurements would enhance the study's validity.

4- The discussion on the increase in blocking probability and required radiating elements could be expanded to explore potential methods to mitigate these issues. Suggestions for optimizing the trade-offs between energy efficiency gains and additional resource requirements would be beneficial.

5- Provide a sensitivity analysis for the proposed approach concerning changes in network parameters, such as the number of users, transmission power, or antenna configurations, which would offer a deeper understanding of its robustness and generalizability.

6- In-depth analysis of the convergence behavior during neural network training would provide valuable insights into the networks' learning process and stability.

7- Including comparisons with other existing adaptive beamforming techniques and machine learning approaches would highlight the uniqueness and advantages of the proposed method over alternative solutions.

8- While the paper addresses spectral efficiency and energy efficiency, it could elaborate on other relevant performance metrics, such as latency and throughput, to provide a more comprehensive evaluation of the proposed approach.

9- A discussion of potential deployment challenges and practical considerations, including computational resource requirements and implementation complexities, would help researchers and practitioners gauge the feasibility of adopting the proposed framework in real-world 5G networks.

10- The implications of the proposed machine learning-based approach on hardware and software requirements, as well as scalability to large-scale networks, could be explored to assess its practical applicability for future 5G deployments.

Extensive editing

Author Response

We would like to thank you for your time and effort to revise our manuscript, as well as for your useful comments. Please find our responses in the attached file.

Author Response File: Author Response.pdf

Reviewer 3 Report

 

The manuscript presents a novel design of an antenna array configuration and a machine learning (ML) adaptive beamforming framework for millimeter-wave (mmWave) massive Multiple-Input Multiple-Output (m-MIMO) cellular networks. The proposed antenna configuration consists of a 21×21 array of rounded crossed bowtie radiating elements (REs). The antenna is designed to achieve a unidirectional radiation pattern with minimal energy waste. This is achieved through a rotation of ±45° for an adaptive dual polarized radiation pattern. The grounding of each antenna and the specific rotation used are unique aspects of the design that merit appreciation for their contribution to enhancing cellular network communications. Furthermore, the authors propose an ML adaptive beamforming framework that optimizes spectral efficiency (SE) and energy efficiency (EE) during model training. This approach allows for the accommodation of varying throughput and power variations in the network, which addresses the limitations in previous models. The idea of using ML to improve adaptive beamforming and to optimize SE and EE in m-MIMO networks is innovative and could pave the way for more intelligent network designs.

 

In conclusion, the manuscript offers valuable insights into the possibilities of combining advanced antenna design with ML for improved performance in m-MIMO cellular networks. The research is well-founded and the results obtained highlight significant potential improvements in EE and SE. However, a more detailed discussion about the ML model used would be beneficial to readers and researchers in the field.

 

However, there are several aspects the authors can seek to improve the paper quality. First, more elaboration is suggested for the selection of DL models, the type of optimizers (SGD, Adam, Adagrad, etc.), and more importantly, the cost function and loss function.   

 

Also, in the introductory part for massive MIMO and AI roles in it, ref. [1] is useful to give an overall view. In addition, [2], [3] provide some interesting and valuable ideas about hybrid beamforming system designs for mmWave. 

 

[1] Huo, Y.; Lin, X.; Di, B.; Zhang, H.; Hernando, F.J.L.; Tan, A.S.; Mumtaz, S.; Demir, Ö.T.; Chen-Hu, K. Technology Trends for Massive MIMO towards 6G. Sensors 2023, 23, 6062. https://doi.org/10.3390/s23136062

 

[2] Zhang, Y. et al. Hybrid beamforming design for mmWave OFDM distributed antenna systems. Sci. China Inf. Sci. 63, 192301 (2020). https://doi.org/10.1007/s11432-019-2799-y

 

[3] W. Son and D. S. Han, "Deep Learning Approach for Improving Spectral Efficiency in mmWave Hybrid Beamforming Systems," 2022 27th Asia Pacific Conference on Communications (APCC), Jeju Island, Korea, Republic of, 2022, pp. 66-69, doi: 10.1109/APCC55198.2022.9943726.

Author Response

We would like to thank you for your time and effort to revise our manuscript, as well as for your useful comments. Please find our responses in the attached file.

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

The authors addressed all the comments, paper can be accepted in its present form.

Just minor editing is required