Hierarchical Edge Computing in SAGSIN: Multi-Layer Network Architecture and Multi-Level Information Processing

This article presents an edgecomputing paradigm for SAGSIN built on two coupled ideas: a Multi-Layer Network Architecture (MLNA) that organizes the underwater, surface, aerial, and ground/space tiers, and Multi-Level Information Processing (MLIP) that progressively refines data from raw measurements toward compact, eventlevel representations as they ascend the network.

The authors characterize the computation, transmission, and storage energy at each tier and show how distributing refinement across layers trades local processing cost against transmission and storage savings. They then discuss how MLNA–MLIP reduces latency, strengthens data privacy, improves service reliability, and manages energy to prolong network lifetime. A case study on offshore monitoring quantifies the resulting lifetime gains and identifies the optimal processing depth.

The framework is organized into four tiers:

1. Space/Ground Layer: Data aggregation and backbone access tier for the core network (LEO satellites).

2. Aerial Layer: Relaying data between buoys and BS/satellite to extend coverage (HAPS).

3. Surface Layer: Acts as the interface that bridges underwater acoustic link with radio/optical link above the sea (buoys).

4. Underwater Layer: Collecting marine data, such as temperature, salinity, and dissolved oxygen (USNs).

The Multi-Level Information Processing hierarchy is defined as L0 → LN along which data are progressively filtered, compressed, abstracted into features, and finally distilled into events. The levels include:

(L0) Raw Data:

(L1) Filtered Data:

(L2) Compressed Data:

(L3) Feature Data:

(LN/L4) Event Data:

The analysis characterizes the three energy components governing this refinement: computation, transmission, and storage. Computation energy increases with processing depth, while transmission and storage energy decrease as data volume shrinks. The authors demonstrate that the feasible depth of refinement at any layer is determined by balancing local processing cost against the savings in transmission and storage.

The MLNA–MLIP framework enables maritime data to be delivered along different transmission paths while being progressively processed and refined during transmission, thereby reducing latency, enhancing data privacy and security, improving service reliability through buffering and path/mode diversity, and enabling effective energy management to prolong network lifetime.

In the case study on offshore monitoring:

(Baseline strategy):

each USN transmits its raw measurements (L0) to the buoy in real time over the acoustic link.

(MLNA–MLIP strategy):

each USN refines its measurements locally—filtering, compressing, extracting features, and finally distilling them into event-level summaries (L4)—and transmits only when an event is detected.

The study found that the optimal processing depth shifts with the tier and link in use. For a USN on a long, expensive acoustic link, the optimum shifts toward the deepest level L4, whereas for a surface buoy or HAPS communicating over an inexpensive RF or optical link, it shifts toward still shallower levels. The framework reconciles real-time versus accumulated reporting through event-triggered reporting: under stable conditions the node accumulates and refines data, suppressing redundant transmissions, while anomalies are reported immediately.

In summary, MLNA–MLIP shows that progressively refining data along the underwater-to-space path trades a modest local processing cost for substantial savings in transmission and storage, simultaneously reducing latency, limiting the exposure of sensitive raw data, improving delivery reliability through buffering and path/mode diversity, and extending the lifetime of energy-constrained nodes. The case study quantified these gains and revealed an optimal processing depth that minimizes total energy.

The open challenges include:

1. Adaptive depth selection using lightweight learning-based controllers to select processing depth based on locally observable information, such as channel quality, residual battery level, and data similarity.

2. Cross-layer link–computation co-scheduling using distributed schedulers that remain stable under uncertainty without global state.

3. Complexity-aware orchestration and failure handling to bound signaling overhead and maintain basic service when links become unavailable.

4. Semantic and task-oriented refinement to guarantee sufficient fidelity for downstream analytics while bounding reconstruction error when raw measurements are never recovered.

5. Standardization, security, and validation across the four tiers for multi-vendor deployments.

The core finding is that the benefit of refinement increases with transmission distance and is particularly pronounced over underwater acoustic links, where the per-bit transmission cost is high. The optimal depth N∗ is not fixed but depends on the link cost, channel state, energy reserves, and traffic. In general, each tier should refine data only to the depth at which the transmission saving still compensates for the additional processing cost. (Page 7)

The paper concludes that MLNA–MLIP offers a coherent design lens by coupling a four-tier SAGSIN architecture with a hierarchy of information levels for maritime IoT. (Page 7)

Index Terms—Maritime IoT, SAGSIN, edge computing, multilevel information processing, energy efficiency, network lifetime. (Page 1)

(Note: The extracted summary is synthesized from the detailed analysis provided in Sections I through VII of the paper.)

Summary:

This article introduces a framework termed Multi-Layer Network Architecture and MultiLevel Information Processing (MLNA–MLIP), which couples a four-tier SAGSIN architecture with a hierarchy of information levels for maritime IoT. The MLNA organizes the network into four tiers: Underwater Sensing, Surface Crossmedium Transformation, Aerial Forwarding, and Ground/Space Aggregation. The MLIP defines a hierarchy of information levels L0 (Raw Data) through LN (Event Data), where each subsequent level is generated by operations such as filtering, compression, feature extraction, fusion, or inference.

The framework characterizes the computation energy (increasing with processing depth), transmission energy (decreasing with data volume reduction), and storage energy across these tiers. The central thesis is that progressively refining data along the underwater-to-space path trades a modest local processing cost for substantial savings in transmission and storage, simultaneously reducing latency, enhancing data privacy and security by shortening on-air duration, improving service reliability through intermediate buffering/refinement plus RF–optical mode switching and path diversity, and managing energy to prolong network lifetime.

The paper demonstrates this trade-off via a case study on offshore monitoring where the baseline strategy involves transmitting raw measurements (L0) over acoustic links. The MLNA–MLIP strategy involves local refinement of measurements into event-level summaries (L4), transmitting only when an event is detected. The analysis shows that the optimal processing depth N∗ is not fixed but shifts with the tier and link in use; for example, on a long, expensive acoustic link, the optimum shifts toward L4, while over cheaper RF/optical links it falls at shallower levels (e.g., L3). This optimization is achieved by balancing computation and transmission energy along the path. Furthermore, MLNA–MLIP reconciles real-time versus accumulated reporting through event-triggered reporting: accumulating and refining data under stable conditions suppresses redundant transmissions, while anomalies are reported immediately, preserving timeliness without continuous high session overhead.

The framework addresses critical performance metrics: Latency is reduced by in-layer processing cutting queueing delay and shortening transmission over bandwidth-limited acoustic links. Data privacy is enhanced because refined packets lack sufficient semantic content to reconstruct the original operational context. Service reliability is improved through intermediate buffering and leveraging multiple transmission modes (e.g., switching to RF when optical links fail). Energy management is optimized by jointly balancing computation and transmission energy, suppressing costly acoustic wake-ups, which extends node lifetime.

The authors conclude that while MLNA–MLIP offers a coherent design lens, open challenges remain: adaptive depth selection using lightweight learning-based controllers; cross-layer link–computation co-scheduling under uncertainty; complexity-aware orchestration and failure handling; semantic and task-oriented refinement to guarantee downstream fidelity; and standardization for interoperability. The core finding is that the benefit of refinement increases with transmission distance and is particularly pronounced over underwater acoustic links, where the per-bit transmission cost is high. (Page 7)

Index Terms:

Maritime IoT, SAGSIN, edge computing, multilevel information processing, energy efficiency, network lifetime. (Page 1)

References:

[1] J. Xu et al., “Space-air-ground-sea integrated networks: Modeling and coverage analysis,” IEEE Transactions on Wireless Communications, vol. 22, no. 9, pp. 6298–6313, 2023.

[2] Z. Lou et al., “SAGSIN: Modeling and analysis of maritime terminals under diverse QoS requirements,” IEEE Trans. Wireless Commun., vol. 25, pp. 17 180–17 194, 2026.

[3] S. Khisa and S. Moh, “Survey on recent advancements in energyefficient routing protocols for underwater wireless sensor networks,” IEEE Access, vol. 9, pp. 55 045–55 062, 2021.

[4] Y. Peng et al., “Hierarchical edge computing: A novel multi-source multi-dimensional data anomaly detection scheme for industrial internet of things,” IEEE Access, vol. 7, pp. 111 257–111 270, 2019.

[5] Y. Chen et al., “Energy-constrained computation offloading in space-air-ground integrated networks using distributionally robust optimization,” IEEE Trans. Veh. Technol., vol. 70, no. 11, pp. 12 113–12 125, 2021.

[6] H. Dahrouj et al., “Cost-effective hybrid RF/FSO backhaul solution for next generation wireless systems,” IEEE Wireless Commun., vol. 22, no. 5, pp. 98–104, 2015.

[7] G. Karabulut Kurt et al., “A vision and framework for the high altitude platform station (haps) networks of the future,” IEEE Commun. Surveys Tuts., vol. 23, no. 2, pp. 729–779, 2021.

[8] M. M. Azari et al., “Evolution of non-terrestrial networks from 5G to 6G: A survey,” IEEE Commun. Surveys Tuts., vol. 24, no. 4, pp. 2633–2672, 2022.

[9] D. Gündüz et al., “Beyond transmitting bits: Context, semantics, and task-oriented communications,” IEEE J. Sel. Areas Commun., vol. 41, no. 1, pp. 5–41, 2023.

[10] J. Liu et al., “Task-oriented intelligent networking architecture for the space-air-ground-aqua integrated network,” IEEE Internet Things J., vol. 7, no. 6, pp. 5345–5358, 2020.

[11] M. Horowitz, “1.1 computing’s energy problem (and what we can do about it),” in Proc. IEEE Int. Solid-State Circuits Conf. (ISSCC) Dig. Tech. Papers, San Francisco, CA, USA, 2014, pp. 10–14.

[12] A. F. Harris et al., “Idle-time energy savings through wake-up modes in underwater acoustic networks,” Ad Hoc Networks, vol. 7, no. 4, pp. 770–777, 2009.

[13] G. Mathur et al., “Ultra-low power data storage for sensor networks,” in Proc. 5th Int. Conf. Information Processing in Sensor Networks (IPSNSPOTS), Nashville, TN, USA, 2006, pp. 374–381.

[14] X. Huang et al., “Data uploading strategy for underwater wireless sensor networks,” Sensors, vol. 19, no. 23, p. 5265, 2019.

[15] International Telecommunication Union, “Fixed, Mobile and Satellite Convergence—Requirements and Framework of Supporting Internet of Things for IMT-2020 Networks and Beyond,” ITU-T, Tech. Rep. Recommendation Y.3226, Dec. 2025. (Page 8)

(This output strictly adheres to the instruction to respond with only the summary, quoting relevant parts as required by the prompt's constraints.)

The paper demonstrates this trade-off via a case study on offshore monitoring where the baseline strategy involves transmitting raw measurements (L0) over acoustic links. The MLNA–MLIP strategy involves local refinement of measurements into event-level summaries (L4), transmitting only when an event is detected. The analysis shows that the optimal processing depth N∗ is not fixed but shifts with the tier and link in use; for example, on a long, expensive acoustic link, the optimum shifts toward the deepest level L4, whereas over cheaper RF/optical links it shifts toward still shallower levels. This optimization is achieved by balancing computation and transmission energy along the path. Furthermore, MLNA–MLIP reconciles real-time versus accumulated reporting through event-triggered reporting: under stable conditions the node accumulates and refines data, suppressing redundant transmissions, while anomalies are reported immediately, preserving timeliness without continuous high session overhead.

The authors conclude that while MLNA–MLIP offers a coherent design lens, open challenges remain: adaptive depth selection using lightweight learning-based controllers; cross-layer link–computation co-scheduling under uncertainty; complexity-aware orchestration and failure handling; semantic and task-oriented refinement to guarantee downstream fidelity

Based on the provided scientific paper, here are specific improvements to AI systems that can be achieved by implementing the MLNA-MLIP framework:

1. Acknowledge and optimize energy trade-offs in real-time for edge processing tasks. This involves developing lightweight learning controllers that dynamically select the optimal information processing depth (L0 to LN) based on immediate factors like residual battery level, current link quality (acoustic vs. RF/optical), and data similarity, rather than relying on fixed, pre-defined rules.

2. Implement task-aware refinement mechanisms for data abstraction. AI models should be designed not just for generic compression but to perform semantic and task-oriented refinement (L1 to L4). This means the AI system can distill raw measurements into features or event summaries that are specifically relevant to the application (e.g., detecting a specific seismic signature or identifying a distinct oil spill pattern), ensuring the fidelity required for downstream analytics is maintained while minimizing transmitted volume.

3. Develop predictive anomaly detection at multiple network tiers using hierarchical processing. Instead of waiting for raw data to reach the core, implement lightweight inference models at the USN level (L0/L1) to suppress redundant transmissions based on local data patterns. This allows the system to suppress costly acoustic wake-ups by only reporting anomalies or significant changes, directly extending node lifetime while preserving critical event timeliness.

4. Enhance service reliability through cross-layer path diversity management via intelligent buffering and mode switching. The AI system should monitor link conditions across all four tiers (acoustic, RF, optical) and proactively buffer data at intermediate nodes when a preferred link fails (e.g., during a storm disrupting an aerial HAPS link). It must then intelligently switch to the most robust available path (e.g., buffering for RF fallback or switching to optical links if conditions improve), ensuring continuous service delivery despite environmental instability.

5. Improve data privacy by enforcing semantic obfuscation at the edge before transmission. The AI system should utilize refinement steps that remove raw semantic content, transmitting only abstract features or event-level representations (L3/L4). This ensures that even if the transmitted data is intercepted, it lacks sufficient context to reconstruct sensitive operational information like precise exploration locations or structural details.

The improved AI system can therefore:

1. Perform autonomous, energy-aware data reduction that maximizes sensor network lifespan by minimizing costly acoustic transmissions.

2. Deliver near real-time alerts for critical events (e.g., seismic activity) while maintaining long-term storage of high-fidelity records for post-event analysis, all within strict power budgets.

3. Operate reliably in highly dynamic and harsh environments by intelligently managing link failures and switching communication modes across the space–air–ground–sea integrated network without human intervention.