Semantic-Aware Resource Allocation for Massive Payload Data Backhaul in Space-Ground TT&C Networks

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

1. It is recommended to further clarify the core differences from existing work in the introduction section, especially highlighting the shortcomings of "DRL methods" in terms of constraint satisfaction.
2. Many symbols appear throughout this paper and could be summarized and organized in the appendix.
3. The authors should further analyze the root causes of HMADRL failure in constraint satisfaction.
4. The task priority a_k,t is set as a fixed weight value in the experiment. Please explain the rationale behind this setting.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

The paper proposes a semantic-aware resource allocation framework for space-ground TT&C networks based on a hierarchical architecture that combines multi-agent reinforcement learning (HMAPPO) with convex optimization via KKT conditions. The objective is to optimize the communication–computation trade-off in highly congested LEO scenarios through adaptive model splitting. The manuscript presents a timely and interesting proposal; however, several improvements are still required to reach publication quality:

• The introduction is well structured and properly contextualizes SGIN, edge computing, and semantic communications. However, it lacks a clear critical positioning with respect to the most recent state of the art (particularly very recent works on NTN + semantic communications). In addition, some statements are generic and lack quantitative support or direct comparison. The combination of RL and convex optimization has already been explored in MEC literature. A possible improvement would be to include an explicit comparison table against state-of-the-art approaches (not only a narrative discussion).
• Although the model is comprehensive, reproducibility is limited. Key details are missing, such as the exact RL network architecture, buffer size, exploration strategy, and training details (e.g., random seeds, stability). While the KKT component is well explained, the RL environment is not fully reproducible. The authors should also clarify how certain components are modeled (e.g., fitting procedures, datasets, interpolation methods).
• The semantic accuracy model is insufficiently defined and remains overly abstract. It is described as “empirical,” but no explanation is provided regarding how it is obtained or how it generalizes. This is a critical issue, as it directly affects the objective function.
• The results are well organized (convergence, load, scalability, ablation), and the figures are generally clear with a thorough analysis. However, statistical evaluation is missing (e.g., variance, confidence intervals). Additionally, the baseline comparison is limited, as it only includes HMADRL and simple heuristics.
• Some figures (e.g., Figs. 5 and 6) are overcrowded and use scales that are not intuitive. Furthermore, some axes lack explicit units.
• How does the system behave when the semantic accuracy function A(\gamma, SNR) is misestimated or dynamically changes (e.g., due to a domain shift in payload data)? Does the framework remain optimal, or does performance degrade significantly?
• Although the problem is decomposed into two levels, can it be formally demonstrated that the hierarchical solution converges to a near-optimal solution of the original MINLP problem, or is there a risk of systematic suboptimality due to the discrete–continuous separation?

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

The proposed task-oriented joint resource allocation framework based on semantic communications is timely and highly relevant for future 6G space-ground integrated networks (SGIN) and intelligent telemetry, tracking, and command (TT&C) systems. However, several critical issues need to be addressed to improve the quality of the work.

1. Originality and Innovation

The paper’s main contribution is claimed to be the reconciliation of narrow-band TT&C links with massive payload data via adaptive DNN model splitting and a bi-level optimization solver. However, DNN partitioning (split computing) between edge and cloud is an established concept in terrestrial MEC. The authors do not sufficiently explain the fundamental theoretical novelties of the "adaptive semantic split" specifically for the TT&C protocol stack compared to standard feature-based compression in Earth observation satellites.

2. Evaluation of Dynamic Topology and Link Intermittency

Please explain the reasons for the lack of a detailed evaluation regarding the impact of high-speed satellite mobility on the semantic feature integrity. The paper mentions "harsh time-varying channel conditions" and assumes Rician fading , but the simulations (Section VI) primarily focus on task arrival rates and user numbers. There is no evaluation of how rapid Doppler shifts or handover-induced jitter affect the "crossing behavior" of the accuracy-SNR curves. The experiment lacks a comparison of the proposed HMAPPO’s robustness against imperfect Channel State Information (CSI), which is a common bottleneck in LEO constellations.

3. Mathematical Rigor and Optimization Assumptions

In equation (25) and the surrounding text, the author utilizes a "soft penalty" to handle the ground access congestion constraint. While the inner-layer KKT solver guarantees zero-violation of computing capacity , the relaxation of the ground access limit into a penalty term may lead to physically infeasible solutions in real-time deployments where the ground station hardware has hard RF chain limits. Ignoring the strict enforcement of this discrete global constraint in the "action masking" phase may result in overly optimistic system utility results.

4. Comparative Analysis of MARL Baselines

The literature review and simulations extensively compare HMAPPO with the HMADRL baseline , but they do not quantitatively compare the proposed HMAPPO with other non-learning-based optimization benchmarks such as Lyapunov-based online scheduling or Distributed ADMM, which are often used for MINLP problems in satellite networks. Quantitatively demonstrating the overhead-performance trade-off of the MARL approach against these traditional methods would strengthen the argument for deploying such a complex bi-level architecture on SWaP-constrained satellites.

5. Recommendation of Recent Related Studies

To provide a comprehensive background, it is recommended to discuss and compare related recent studies on satellite resource management and semantic communication. For instance: Zhou K, Li J, Zhou Q, et al. Modeling and Analysis of Terahertz Inter-Satellite Communication-Ranging System under Platform Vibrations[J]. IEEE Transactions on Communications, 2026. Huang C, Chen X, Chen G, et al. Deep reinforcement learning-based resource allocation for hybrid bit and generative semantic communications in space-air-ground integrated networks[J]. IEEE Journal on Selected Areas in Communications, 2025.

Comments on the Quality of English Language

The English is generally comprehensible, but minor editing is required to improve grammar and fluency.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

Comments and Suggestions for Authors

All my comments were addressed. No more comments. The contribution is clearer.

Reviewer 3 Report

Comments and Suggestions for Authors

All previous comments have been adequately addressed in the revised manuscript. I am satisfied with the current version and have no additional suggestions.

Comments on the Quality of English Language

The English is generally comprehensible, but minor editing is required to improve grammar and fluency.