Collaborative Computation Offloading and Resource Management in Space–Air–Ground Integrated Networking: A Deep Reinforcement Learning Approach

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

 

In this paper, the authors proposed a deep-learning strategy for resource management in space-air-ground networks. To address the challenge of balancing computation offloading and resource management, the authors formulated a mixed-integer nonlinear program, achieving cost constraints in a heterogeneous environment. I suggest allowing this paper to have a major revision. To further improve the quality of this paper, the authors should address the comments below.

 

1.   For a compelling comparison, consider including data in your literature survey. This information should be structured to compare existing approaches with your proposed work. Specifically, it should highlight the weaknesses of previous methods and emphasize the novelty of your contribution. For instance, the author's perspective on the limitations of edge networks can be visualized through a table. This table would highlight how current edge networks cannot deliver computational services to all devices everywhere.

2.   Who (or what) are the "some scholars" mentioned on lines 81 and 105? Instead of  “some scholars”, use the phrase “authors of this paper”, “From the perspective of this paper's authors”, “In this paper, the authors” etc.

3.   I will suggest moving your Figure 1- “Architecture of Edge Computing in Space-Air-Ground Integrated Networking” to the introduction section and drawing Figure 2 in terms of offloading scenarios.

4.   SAGIN-like networks, with their vast and dynamic collection of hardware resources, introduce unique challenges. How authors will justify a particular problem of minimizing the error in offloading procedures when neighboring devices are also dynamic.

5.   Authors should discuss the signaling overhead per increasing node count.

6.   The simulation results section contains duplicate headings such as 6.1. parameter setting and 6.2. parameter setting. It is suggested to revise one of the duplicate headings. 'Simulation Environment' and 'Simulation Parameters' are potential alternative titles to distinguish these sections.

7.   How does the proposed DQN-based approach for computation offloading and resource management in SAGIN compare to conventional offloading strategies and scenarios where no offloading takes place?

8.   While the learning rates chosen for the DNN in Figure 6 (0.005, 0.01, 0.05, 0.1) cover a reasonable range, a more comprehensive analysis approach could be achieved by exploring values within the commonly suggested range of 0.001 to 0.1(i.e., 0.005, 0.03, 0.077,  0.099, 0.1). By using a broader range of learning rates, the authors could provide a deeper understanding of how this parameter impacts the performance of the DQN-based mechanism in SAGIN.

9.   For Figure 2, The authors should discuss the loss function and depict the results in the result section.

10. In Algorithm 1(Q-learning), for initialization, the state behavior value is

11.  Among the different batch sizes tested in Figure 8, a batch size of 32 achieves the best performance after 400 iterations, even compared to 64, 128, and 256.

12.  At line 352, it is Fig . 9 instead of Fig .8.

13.  Authors should discuss how their results provide evidence for the strengths of the proposed solution.

14.  For a stronger impact, the abstract and conclusion should emphasize the novelty and benefits of the proposed strategy. Quantify the results by including key performance metrics.

15.  The article contains several issues with conjunction and coherence. Revising these areas would improve the flow and clarity of the writing. Additionally, the similarity index of 30% is above the accepted threshold. Consider revising the text to ensure originality.

 

 

Comments on the Quality of English Language

The article contains several issues with conjunction and coherence. Revising these areas would improve the flow and clarity of the writing.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

This work deals with the problem of computation offloading and resource management in SAGIN. The authors propose a computation offloading architecture for SAGIN, where the terminal device’s computation task can be processed locally or offloaded to edge servers. Besides, the SDN’s controllers can make real-time offloading decisions and network resource allocation by using a deep reinforcement learning approach. They jointly optimize offloading decisions, spectrum allocation, computation, and storage resource management in SAGIN as a MINLP problem. They solve this problem using deep reinforcement learning. They also perform some numerical experiments to investigate the effect of different parameters.

Some advantages of this paper are:

(1) Research motivation is clear.

(2) Comments on the related references are appropriate and enough.

(3) Research methodology sounds workable.

 

There are still some shortcomings.

(1) The titles of subsections 6.1 and 6.1 are the same.

(2) Line 308, the parameters in this experiment are presented in Table II. I cannot find the Table II.

(3) The descriptions of parameter settings are not clear. How to decide the values of those parameters? By referring to other papers or?

(4) In Figure 6, the x-axis title is incorrect. It should be “iteration”. Such faults also happened in Figures 7, 8, and 9.

 

(5) This work deals with the problem of computation offloading and resource management in SAGIN. However, in the simulation, the authors only investigate the effect of different parameters. They do not address the system performance issues, such as throughput and delay. These indexes are supposed to be discussed in the simulation.  

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

Thank you for your comprehensive and accurate responses.

The clarity and logical coherence of the author’s explanations, coupled with their proactive approach to implementing essential revisions, have conclusively elevated the paper's overall quality. The feedback provided has been thoroughly addressed, particularly regarding:

1.      The explanations for specific objections related to the proposed system and simulation.

2.      Extended literature survey section.

3.      Abstract and conclusion section.

4.      Edited for originality.

Reviewer 2 Report

Comments and Suggestions for Authors

The revised version has fully adressed my concerns.