kumoh national institute of technology
Networked Systems Lab.

Heidy Indrayani, Rizki Rivai Ginanjar, Jae-Min Lee, Dong-Seong Kim, "Improved Deep learning-based CSI Matrix Reconstruction for Massive MIMO Communication System", IEEE Communications Letters (R)
By : Heidy Indrayani
Date : 2020-01-30
Views : 192

Dear Author(s):

The review of the referenced manuscript, CL2020-0005, entitled Improved Deep learning-based CSI Matrix Reconstruction for Massive MIMO Communication System, is now complete. I regret to inform you that based on the enclosed reviews and my own reading of your manuscript, I am unable to recommend its publication in IEEE Communications Letters.

Your paper may not be resubmitted for review. The reasons for this are as follows: All experienced reviewers point out the contribution is quite limited by simply adopting DNN to the massive MIMO system. The comparison and improvement between this submission and [7] are unclear. Finally, the complexity of the proposed method has not been well addressed.

Additional comments include:

The reviewers' comments are found at the end of this email.

Thank you for submitting your work to the IEEE Communications Letters.


Prof. Jiayi Zhang
Associate Editor
IEEE Communications Letters

Reviewer: 1

Comments to the Author
This paper proposes an improved deep learning architecture for channel state information (CSI) matrix reconstruction technique in the MIMO communication system. Here are my comments:

1. The authors claimed that this work addressed the challenge of the complexity of work [7] in the Section I. However, nor complexity analysis or comparison are mentioned in this paper, which make the motivation and advantages of this paper unconvinced.
2. The contribution of this paper is weak, since the dropout is a basic technique involved in deep learning. And the dropout technique seems to be the only difference compared with the work [7].
3. The format of the references should be corrected, such as the [7,9,10]

Reviewer: 2

Comments to the Author
This paper proposes to improve an existing neural network architecture known as CSINET by using the dropout technique for CSI matrix reconstruction for massive MIMO systems. The paper is well written.

This paper has a very limited contribution. Dropout is one of the very well known techniques for dealing with overfitting in neural networks. It is widely used while training neural networks by data scientists and researchers. I do not see any scientific contribution in adding dropout to a neural network. Therefore I cannot recommend this paper for publication in IEEE Communications Letters.

Reviewer: 3

Comments to the Author
This paper considers deel learning-based CSI feedback codebook design. However, the novelty is very limited and the contribution is hence questionable. Therefore, I recommand rejection. Specifically, major concerns are

1. The proposed deep learning framework is simply appling the exisitng well-known DNN to this scenario, including the dropout method which is also well-known.

2. The feedback codebook design has been investigated intensively in the literature. The paper does not give any comparison with existing works.

3. The codebook should be a discrete book. The authors only compress the CSI, which is not enough.

4. The paper is poorly written.